You are viewing a plain text version of this content. The canonical link for it is here.
Posted to commits@cassandra.apache.org by be...@apache.org on 2015/07/07 17:46:06 UTC
[01/12] cassandra git commit: Scrub (recover) sstables even when
-Index.db is missing
Repository: cassandra
Updated Branches:
refs/heads/cassandra-2.0 62714a9fe -> 452d6a445
refs/heads/cassandra-2.1 4de943fd9 -> 4c94ef20d
refs/heads/cassandra-2.2 7d31068da -> ebe18bb2f
refs/heads/trunk db68d1cfd -> a8bb75a7e
Scrub (recover) sstables even when -Index.db is missing
patch by mck; reviewed by stefania for CASSANDRA-9591
Project: http://git-wip-us.apache.org/repos/asf/cassandra/repo
Commit: http://git-wip-us.apache.org/repos/asf/cassandra/commit/452d6a44
Tree: http://git-wip-us.apache.org/repos/asf/cassandra/tree/452d6a44
Diff: http://git-wip-us.apache.org/repos/asf/cassandra/diff/452d6a44
Branch: refs/heads/cassandra-2.0
Commit: 452d6a445a6935d3e7d0e0fdf59e87d2a2ff95e7
Parents: 62714a9
Author: Stefania Alborghetti <st...@datastax.com>
Authored: Mon Jun 15 16:49:03 2015 +0800
Committer: Benedict Elliott Smith <be...@apache.org>
Committed: Tue Jul 7 14:27:25 2015 +0100
----------------------------------------------------------------------
CHANGES.txt | 4 ++
.../cassandra/db/compaction/Scrubber.java | 36 +++++++++++----
.../cassandra/io/sstable/SSTableReader.java | 47 ++++++++++++++++----
.../cassandra/tools/StandaloneScrubber.java | 2 +-
.../unit/org/apache/cassandra/db/ScrubTest.java | 25 +++++++++++
5 files changed, 96 insertions(+), 18 deletions(-)
----------------------------------------------------------------------
http://git-wip-us.apache.org/repos/asf/cassandra/blob/452d6a44/CHANGES.txt
----------------------------------------------------------------------
diff --git a/CHANGES.txt b/CHANGES.txt
index ca4d4b5..bd1db92 100644
--- a/CHANGES.txt
+++ b/CHANGES.txt
@@ -1,3 +1,7 @@
+2.0.18
+* Scrub (recover) sstables even when -Index.db is missing, (CASSANDRA-9591)
+
+
2.0.17
* Avoid NPE in AuthSuccess#decode (CASSANDRA-9727)
* Add listen_address to system.local (CASSANDRA-9603)
http://git-wip-us.apache.org/repos/asf/cassandra/blob/452d6a44/src/java/org/apache/cassandra/db/compaction/Scrubber.java
----------------------------------------------------------------------
diff --git a/src/java/org/apache/cassandra/db/compaction/Scrubber.java b/src/java/org/apache/cassandra/db/compaction/Scrubber.java
index ea10855..dc60efa 100644
--- a/src/java/org/apache/cassandra/db/compaction/Scrubber.java
+++ b/src/java/org/apache/cassandra/db/compaction/Scrubber.java
@@ -96,7 +96,16 @@ public class Scrubber implements Closeable
? new ScrubController(cfs)
: new CompactionController(cfs, Collections.singleton(sstable), CompactionManager.getDefaultGcBefore(cfs));
this.isCommutative = cfs.metadata.getDefaultValidator().isCommutative();
- this.expectedBloomFilterSize = Math.max(cfs.metadata.getIndexInterval(), (int)(SSTableReader.getApproximateKeyCount(toScrub,cfs.metadata)));
+
+ boolean hasIndexFile = (new File(sstable.descriptor.filenameFor(Component.PRIMARY_INDEX))).exists();
+ if (!hasIndexFile)
+ {
+ // if there's any corruption in the -Data.db then rows can't be skipped over. but it's worth a shot.
+ outputHandler.warn("Missing component: " + sstable.descriptor.filenameFor(Component.PRIMARY_INDEX));
+ }
+
+ this.expectedBloomFilterSize = Math.max(cfs.metadata.getIndexInterval(),
+ hasIndexFile ? (int)(SSTableReader.getApproximateKeyCount(toScrub,cfs.metadata)) : 0);
// loop through each row, deserializing to check for damage.
// we'll also loop through the index at the same time, using the position from the index to recover if the
@@ -105,7 +114,11 @@ public class Scrubber implements Closeable
this.dataFile = isOffline
? sstable.openDataReader()
: sstable.openDataReader(CompactionManager.instance.getRateLimiter());
- this.indexFile = RandomAccessReader.open(new File(sstable.descriptor.filenameFor(Component.PRIMARY_INDEX)));
+
+ this.indexFile = hasIndexFile
+ ? RandomAccessReader.open(new File(sstable.descriptor.filenameFor(Component.PRIMARY_INDEX)))
+ : null;
+
this.scrubInfo = new ScrubInfo(dataFile, sstable);
this.currentRowPositionFromIndex = 0;
@@ -117,7 +130,8 @@ public class Scrubber implements Closeable
outputHandler.output(String.format("Scrubbing %s (%s bytes)", sstable, dataFile.length()));
try
{
- nextIndexKey = ByteBufferUtil.readWithShortLength(indexFile);
+ nextIndexKey = indexAvailable() ? ByteBufferUtil.readWithShortLength(indexFile) : null;
+ if (indexAvailable())
{
// throw away variable so we don't have a side effect in the assert
long firstRowPositionFromIndex = RowIndexEntry.serializer.deserialize(indexFile, sstable.descriptor.version).position;
@@ -181,7 +195,7 @@ public class Scrubber implements Closeable
outputHandler.debug(String.format("Index doublecheck: row %s is %s bytes", ByteBufferUtil.bytesToHex(currentIndexKey), dataSizeFromIndex));
}
- assert currentIndexKey != null || indexFile.isEOF();
+ assert currentIndexKey != null || !indexAvailable();
writer.mark();
try
@@ -198,10 +212,10 @@ public class Scrubber implements Closeable
if (dataSize > dataFile.length())
throw new IOError(new IOException("Impossible row size (greater than file length): " + dataSize));
- if (dataStart != dataStartFromIndex)
+ if (indexFile != null && dataStart != dataStartFromIndex)
outputHandler.warn(String.format("Data file row position %d differs from index file row position %d", dataStart, dataStartFromIndex));
- if (dataSize != dataSizeFromIndex)
+ if (indexFile != null && dataSize != dataSizeFromIndex)
outputHandler.warn(String.format("Data file row size %d differs from index file row size %d", dataSize, dataSizeFromIndex));
SSTableIdentityIterator atoms = new SSTableIdentityIterator(sstable, dataFile, key, dataSize, validateColumns);
@@ -317,8 +331,9 @@ public class Scrubber implements Closeable
currentRowPositionFromIndex = nextRowPositionFromIndex;
try
{
- nextIndexKey = indexFile.isEOF() ? null : ByteBufferUtil.readWithShortLength(indexFile);
- nextRowPositionFromIndex = indexFile.isEOF()
+ nextIndexKey = !indexAvailable() ? null : ByteBufferUtil.readWithShortLength(indexFile);
+
+ nextRowPositionFromIndex = !indexAvailable()
? dataFile.length()
: RowIndexEntry.serializer.deserialize(indexFile, sstable.descriptor.version).position;
}
@@ -330,6 +345,11 @@ public class Scrubber implements Closeable
}
}
+ private boolean indexAvailable()
+ {
+ return indexFile != null && !indexFile.isEOF();
+ }
+
private void seekToNextRow()
{
while(nextRowPositionFromIndex < dataFile.length())
http://git-wip-us.apache.org/repos/asf/cassandra/blob/452d6a44/src/java/org/apache/cassandra/io/sstable/SSTableReader.java
----------------------------------------------------------------------
diff --git a/src/java/org/apache/cassandra/io/sstable/SSTableReader.java b/src/java/org/apache/cassandra/io/sstable/SSTableReader.java
index 15808e8..8919a09 100644
--- a/src/java/org/apache/cassandra/io/sstable/SSTableReader.java
+++ b/src/java/org/apache/cassandra/io/sstable/SSTableReader.java
@@ -157,7 +157,7 @@ public class SSTableReader extends SSTable implements Closeable
public static SSTableReader openForBatch(Descriptor descriptor, Set<Component> components, CFMetaData metadata, IPartitioner partitioner) throws IOException
{
- SSTableMetadata sstableMetadata = openMetadata(descriptor, components, partitioner);
+ SSTableMetadata sstableMetadata = openMetadata(descriptor, components, partitioner, true);
SSTableReader sstable = new SSTableReader(descriptor,
components,
metadata,
@@ -191,7 +191,7 @@ public class SSTableReader extends SSTable implements Closeable
boolean validate) throws IOException
{
long start = System.nanoTime();
- SSTableMetadata sstableMetadata = openMetadata(descriptor, components, partitioner);
+ SSTableMetadata sstableMetadata = openMetadata(descriptor, components, partitioner, validate);
SSTableReader sstable = new SSTableReader(descriptor,
components,
@@ -213,12 +213,17 @@ public class SSTableReader extends SSTable implements Closeable
return sstable;
}
- private static SSTableMetadata openMetadata(Descriptor descriptor, Set<Component> components, IPartitioner partitioner) throws IOException
+ private static SSTableMetadata openMetadata(Descriptor descriptor,
+ Set<Component> components,
+ IPartitioner partitioner,
+ boolean primaryIndexRequired) throws IOException
{
assert partitioner != null;
// Minimum components without which we can't do anything
assert components.contains(Component.DATA) : "Data component is missing for sstable" + descriptor;
- assert components.contains(Component.PRIMARY_INDEX) : "Primary index component is missing for sstable " + descriptor;
+
+ assert !primaryIndexRequired || components.contains(Component.PRIMARY_INDEX)
+ : "Primary index component is missing for sstable " + descriptor;
logger.info("Opening {} ({} bytes)", descriptor, new File(descriptor.filenameFor(COMPONENT_DATA)).length());
@@ -382,11 +387,17 @@ public class SSTableReader extends SSTable implements Closeable
readMeterSyncFuture.cancel(false);
// Force finalizing mmapping if necessary
- ifile.cleanup();
+
+ if (null != ifile)
+ ifile.cleanup();
+
dfile.cleanup();
// close the BF so it can be opened later.
- bf.close();
- indexSummary.close();
+ if (null != bf)
+ bf.close();
+
+ if (null != indexSummary)
+ indexSummary.close();
}
public void setTrackedBy(DataTracker tracker)
@@ -406,6 +417,12 @@ public class SSTableReader extends SSTable implements Closeable
load(false, true);
bf = FilterFactory.AlwaysPresent;
}
+ else if (!components.contains(Component.PRIMARY_INDEX))
+ {
+ // avoid any reading of the missing primary index component.
+ // this should only happen during StandaloneScrubber
+ load(false, false);
+ }
else if (!components.contains(Component.FILTER))
{
// bf is enabled, but filter component is missing.
@@ -454,7 +471,9 @@ public class SSTableReader extends SSTable implements Closeable
if (recreateBloomFilter || !summaryLoaded)
buildSummary(recreateBloomFilter, ibuilder, dbuilder, summaryLoaded);
- ifile = ibuilder.complete(descriptor.filenameFor(Component.PRIMARY_INDEX));
+ if (components.contains(Component.PRIMARY_INDEX))
+ ifile = ibuilder.complete(descriptor.filenameFor(Component.PRIMARY_INDEX));
+
dfile = dbuilder.complete(descriptor.filenameFor(Component.DATA));
if (saveSummaryIfCreated && (recreateBloomFilter || !summaryLoaded)) // save summary information to disk
saveSummary(this, ibuilder, dbuilder);
@@ -462,6 +481,9 @@ public class SSTableReader extends SSTable implements Closeable
private void buildSummary(boolean recreateBloomFilter, SegmentedFile.Builder ibuilder, SegmentedFile.Builder dbuilder, boolean summaryLoaded) throws IOException
{
+ if (!components.contains(Component.PRIMARY_INDEX))
+ return;
+
// we read the positions in a BRAF so we don't have to worry about an entry spanning a mmap boundary.
RandomAccessReader primaryIndex = RandomAccessReader.open(new File(descriptor.filenameFor(Component.PRIMARY_INDEX)));
@@ -965,6 +987,9 @@ public class SSTableReader extends SSTable implements Closeable
}
}
+ if (ifile == null)
+ return null;
+
// scan the on-disk index, starting at the nearest sampled position.
// The check against IndexInterval is to be exit the loop in the EQ case when the key looked for is not present
// (bloom filter false positive). But note that for non-EQ cases, we might need to check the first key of the
@@ -1063,6 +1088,9 @@ public class SSTableReader extends SSTable implements Closeable
if (sampledPosition == -1)
sampledPosition = 0;
+ if (ifile == null)
+ return null;
+
Iterator<FileDataInput> segments = ifile.iterator(sampledPosition);
while (segments.hasNext())
{
@@ -1480,7 +1508,8 @@ public class SSTableReader extends SSTable implements Closeable
private void dropPageCache()
{
dropPageCache(dfile.path);
- dropPageCache(ifile.path);
+ if (null != ifile)
+ dropPageCache(ifile.path);
}
private void dropPageCache(String filePath)
http://git-wip-us.apache.org/repos/asf/cassandra/blob/452d6a44/src/java/org/apache/cassandra/tools/StandaloneScrubber.java
----------------------------------------------------------------------
diff --git a/src/java/org/apache/cassandra/tools/StandaloneScrubber.java b/src/java/org/apache/cassandra/tools/StandaloneScrubber.java
index 26768f3..8ace228 100644
--- a/src/java/org/apache/cassandra/tools/StandaloneScrubber.java
+++ b/src/java/org/apache/cassandra/tools/StandaloneScrubber.java
@@ -79,7 +79,7 @@ public class StandaloneScrubber
for (Map.Entry<Descriptor, Set<Component>> entry : lister.list().entrySet())
{
Set<Component> components = entry.getValue();
- if (!components.contains(Component.DATA) || !components.contains(Component.PRIMARY_INDEX))
+ if (!components.contains(Component.DATA))
continue;
try
http://git-wip-us.apache.org/repos/asf/cassandra/blob/452d6a44/test/unit/org/apache/cassandra/db/ScrubTest.java
----------------------------------------------------------------------
diff --git a/test/unit/org/apache/cassandra/db/ScrubTest.java b/test/unit/org/apache/cassandra/db/ScrubTest.java
index 94c7e34..978b3e8 100644
--- a/test/unit/org/apache/cassandra/db/ScrubTest.java
+++ b/test/unit/org/apache/cassandra/db/ScrubTest.java
@@ -270,6 +270,31 @@ public class ScrubTest extends SchemaLoader
}
@Test
+ public void testScrubNoIndex() throws IOException, ExecutionException, InterruptedException, ConfigurationException
+ {
+ CompactionManager.instance.disableAutoCompaction();
+ Keyspace keyspace = Keyspace.open(KEYSPACE);
+ ColumnFamilyStore cfs = keyspace.getColumnFamilyStore(CF);
+ cfs.clearUnsafe();
+
+ List<Row> rows;
+
+ // insert data and verify we get it back w/ range query
+ fillCF(cfs, 10);
+ rows = cfs.getRangeSlice(Util.range("", ""), null, new IdentityQueryFilter(), 1000);
+ assertEquals(10, rows.size());
+
+ for (SSTableReader sstable : cfs.getSSTables())
+ new File(sstable.descriptor.filenameFor(Component.PRIMARY_INDEX)).delete();
+
+ CompactionManager.instance.performScrub(cfs, false, true);
+
+ // check data is still there
+ rows = cfs.getRangeSlice(Util.range("", ""), null, new IdentityQueryFilter(), 1000);
+ assertEquals(10, rows.size());
+ }
+
+ @Test
public void testScrubOutOfOrder() throws Exception
{
CompactionManager.instance.disableAutoCompaction();
[05/12] cassandra git commit: Merge branch 'cassandra-2.0' into
cassandra-2.1
Posted by be...@apache.org.
Merge branch 'cassandra-2.0' into cassandra-2.1
Conflicts:
CHANGES.txt
src/java/org/apache/cassandra/db/compaction/Scrubber.java
src/java/org/apache/cassandra/io/sstable/SSTableReader.java
Project: http://git-wip-us.apache.org/repos/asf/cassandra/repo
Commit: http://git-wip-us.apache.org/repos/asf/cassandra/commit/4c94ef20
Tree: http://git-wip-us.apache.org/repos/asf/cassandra/tree/4c94ef20
Diff: http://git-wip-us.apache.org/repos/asf/cassandra/diff/4c94ef20
Branch: refs/heads/cassandra-2.2
Commit: 4c94ef20d3562ab8f0a922945d78464d6c475d98
Parents: 4de943f 452d6a4
Author: Benedict Elliott Smith <be...@apache.org>
Authored: Tue Jul 7 16:27:58 2015 +0100
Committer: Benedict Elliott Smith <be...@apache.org>
Committed: Tue Jul 7 16:27:58 2015 +0100
----------------------------------------------------------------------
CHANGES.txt | 5 ++
.../cassandra/db/compaction/Scrubber.java | 37 +++++++++++---
.../cassandra/io/sstable/SSTableReader.java | 52 ++++++++++++++------
.../cassandra/tools/StandaloneScrubber.java | 2 +-
.../unit/org/apache/cassandra/db/ScrubTest.java | 25 ++++++++++
5 files changed, 97 insertions(+), 24 deletions(-)
----------------------------------------------------------------------
http://git-wip-us.apache.org/repos/asf/cassandra/blob/4c94ef20/CHANGES.txt
----------------------------------------------------------------------
diff --cc CHANGES.txt
index 95dc8be,bd1db92..2cbc7c4
--- a/CHANGES.txt
+++ b/CHANGES.txt
@@@ -1,18 -1,8 +1,23 @@@
-2.0.18
-* Scrub (recover) sstables even when -Index.db is missing, (CASSANDRA-9591)
-
-
-2.0.17
++2.1.9
++Merged from 2.0:
++ * Scrub (recover) sstables even when -Index.db is missing, (CASSANDRA-9591)
++
++
+2.1.8
+ * (cqlsh) Fix bad check for CQL compatibility when DESCRIBE'ing
+ COMPACT STORAGE tables with no clustering columns
+ * Warn when an extra-large partition is compacted (CASSANDRA-9643)
+ * Eliminate strong self-reference chains in sstable ref tidiers (CASSANDRA-9656)
+ * Ensure StreamSession uses canonical sstable reader instances (CASSANDRA-9700)
+ * Ensure memtable book keeping is not corrupted in the event we shrink usage (CASSANDRA-9681)
+ * Update internal python driver for cqlsh (CASSANDRA-9064)
+ * Fix IndexOutOfBoundsException when inserting tuple with too many
+ elements using the string literal notation (CASSANDRA-9559)
+ * Allow JMX over SSL directly from nodetool (CASSANDRA-9090)
+ * Fix incorrect result for IN queries where column not found (CASSANDRA-9540)
+ * Enable describe on indices (CASSANDRA-7814)
+ * ColumnFamilyStore.selectAndReference may block during compaction (CASSANDRA-9637)
+Merged from 2.0:
* Avoid NPE in AuthSuccess#decode (CASSANDRA-9727)
* Add listen_address to system.local (CASSANDRA-9603)
* Bug fixes to resultset metadata construction (CASSANDRA-9636)
http://git-wip-us.apache.org/repos/asf/cassandra/blob/4c94ef20/src/java/org/apache/cassandra/db/compaction/Scrubber.java
----------------------------------------------------------------------
diff --cc src/java/org/apache/cassandra/db/compaction/Scrubber.java
index ce98a13,dc60efa..b1c12e0
--- a/src/java/org/apache/cassandra/db/compaction/Scrubber.java
+++ b/src/java/org/apache/cassandra/db/compaction/Scrubber.java
@@@ -100,8 -95,17 +100,18 @@@ public class Scrubber implements Closea
this.controller = isOffline
? new ScrubController(cfs)
: new CompactionController(cfs, Collections.singleton(sstable), CompactionManager.getDefaultGcBefore(cfs));
- this.isCommutative = cfs.metadata.getDefaultValidator().isCommutative();
+ this.isCommutative = cfs.metadata.isCounter();
- this.expectedBloomFilterSize = Math.max(cfs.metadata.getMinIndexInterval(), (int)(SSTableReader.getApproximateKeyCount(toScrub)));
+
+ boolean hasIndexFile = (new File(sstable.descriptor.filenameFor(Component.PRIMARY_INDEX))).exists();
+ if (!hasIndexFile)
+ {
+ // if there's any corruption in the -Data.db then rows can't be skipped over. but it's worth a shot.
+ outputHandler.warn("Missing component: " + sstable.descriptor.filenameFor(Component.PRIMARY_INDEX));
+ }
+
- this.expectedBloomFilterSize = Math.max(cfs.metadata.getIndexInterval(),
- hasIndexFile ? (int)(SSTableReader.getApproximateKeyCount(toScrub,cfs.metadata)) : 0);
++ this.expectedBloomFilterSize = Math.max(
++ cfs.metadata.getMinIndexInterval(),
++ hasIndexFile ? (int)(SSTableReader.getApproximateKeyCount(toScrub)) : 0);
// loop through each row, deserializing to check for damage.
// we'll also loop through the index at the same time, using the position from the index to recover if the
@@@ -120,14 -128,13 +134,15 @@@
public void scrub()
{
outputHandler.output(String.format("Scrubbing %s (%s bytes)", sstable, dataFile.length()));
+ Set<SSTableReader> oldSSTable = Sets.newHashSet(sstable);
+ SSTableRewriter writer = new SSTableRewriter(cfs, oldSSTable, sstable.maxDataAge, isOffline);
try
{
- nextIndexKey = ByteBufferUtil.readWithShortLength(indexFile);
+ nextIndexKey = indexAvailable() ? ByteBufferUtil.readWithShortLength(indexFile) : null;
+ if (indexAvailable())
{
// throw away variable so we don't have a side effect in the assert
- long firstRowPositionFromIndex = RowIndexEntry.serializer.deserialize(indexFile, sstable.descriptor.version).position;
+ long firstRowPositionFromIndex = sstable.metadata.comparator.rowIndexEntrySerializer().deserialize(indexFile, sstable.descriptor.version).position;
assert firstRowPositionFromIndex == 0 : firstRowPositionFromIndex;
}
@@@ -167,13 -179,25 +182,13 @@@
dataSizeFromIndex = nextRowPositionFromIndex - dataStartFromIndex;
}
- if (!sstable.descriptor.version.hasRowSizeAndColumnCount)
- {
- dataSize = dataSizeFromIndex;
- // avoid an NPE if key is null
- String keyName = key == null ? "(unreadable key)" : ByteBufferUtil.bytesToHex(key.key);
- outputHandler.debug(String.format("row %s is %s bytes", keyName, dataSize));
- }
- else
- {
- if (currentIndexKey != null)
- outputHandler.debug(String.format("Index doublecheck: row %s is %s bytes", ByteBufferUtil.bytesToHex(currentIndexKey), dataSizeFromIndex));
- }
+ dataSize = dataSizeFromIndex;
+ // avoid an NPE if key is null
+ String keyName = key == null ? "(unreadable key)" : ByteBufferUtil.bytesToHex(key.getKey());
+ outputHandler.debug(String.format("row %s is %s bytes", keyName, dataSize));
- assert currentIndexKey != null || indexFile.isEOF();
+ assert currentIndexKey != null || !indexAvailable();
- writer.mark();
try
{
if (key == null)
@@@ -188,11 -212,11 +203,11 @@@
if (dataSize > dataFile.length())
throw new IOError(new IOException("Impossible row size (greater than file length): " + dataSize));
- if (dataStart != dataStartFromIndex)
+ if (indexFile != null && dataStart != dataStartFromIndex)
outputHandler.warn(String.format("Data file row position %d differs from index file row position %d", dataStart, dataStartFromIndex));
- if (dataSize != dataSizeFromIndex)
+ if (indexFile != null && dataSize != dataSizeFromIndex)
- outputHandler.warn(String.format("Data file row size %d differs from index file row size %d", dataSize, dataSizeFromIndex));
+ outputHandler.warn(String.format("Data file row size %d different from index file row size %d", dataSize, dataSizeFromIndex));
SSTableIdentityIterator atoms = new SSTableIdentityIterator(sstable, dataFile, key, dataSize, validateColumns);
if (prevKey != null && prevKey.compareTo(key) > 0)
@@@ -312,10 -331,11 +327,11 @@@
currentRowPositionFromIndex = nextRowPositionFromIndex;
try
{
- nextIndexKey = indexFile.isEOF() ? null : ByteBufferUtil.readWithShortLength(indexFile);
- nextRowPositionFromIndex = indexFile.isEOF()
+ nextIndexKey = !indexAvailable() ? null : ByteBufferUtil.readWithShortLength(indexFile);
+
+ nextRowPositionFromIndex = !indexAvailable()
? dataFile.length()
- : RowIndexEntry.serializer.deserialize(indexFile, sstable.descriptor.version).position;
+ : sstable.metadata.comparator.rowIndexEntrySerializer().deserialize(indexFile, sstable.descriptor.version).position;
}
catch (Throwable th)
{
http://git-wip-us.apache.org/repos/asf/cassandra/blob/4c94ef20/src/java/org/apache/cassandra/io/sstable/SSTableReader.java
----------------------------------------------------------------------
diff --cc src/java/org/apache/cassandra/io/sstable/SSTableReader.java
index 7551d46,8919a09..6879834
--- a/src/java/org/apache/cassandra/io/sstable/SSTableReader.java
+++ b/src/java/org/apache/cassandra/io/sstable/SSTableReader.java
@@@ -448,27 -190,9 +448,27 @@@ public class SSTableReader extends SSTa
IPartitioner partitioner,
boolean validate) throws IOException
{
- long start = System.nanoTime();
- SSTableMetadata sstableMetadata = openMetadata(descriptor, components, partitioner, validate);
+ // Minimum components without which we can't do anything
+ assert components.contains(Component.DATA) : "Data component is missing for sstable" + descriptor;
- assert components.contains(Component.PRIMARY_INDEX) : "Primary index component is missing for sstable " + descriptor;
++ assert !validate || components.contains(Component.PRIMARY_INDEX) : "Primary index component is missing for sstable " + descriptor;
+
+ Map<MetadataType, MetadataComponent> sstableMetadata = descriptor.getMetadataSerializer().deserialize(descriptor,
- EnumSet.of(MetadataType.VALIDATION, MetadataType.STATS));
++ EnumSet.of(MetadataType.VALIDATION, MetadataType.STATS));
+ ValidationMetadata validationMetadata = (ValidationMetadata) sstableMetadata.get(MetadataType.VALIDATION);
+ StatsMetadata statsMetadata = (StatsMetadata) sstableMetadata.get(MetadataType.STATS);
+
+ // Check if sstable is created using same partitioner.
+ // Partitioner can be null, which indicates older version of sstable or no stats available.
+ // In that case, we skip the check.
+ String partitionerName = partitioner.getClass().getCanonicalName();
+ if (validationMetadata != null && !partitionerName.equals(validationMetadata.partitioner))
+ {
+ logger.error(String.format("Cannot open %s; partitioner %s does not match system partitioner %s. Note that the default partitioner starting with Cassandra 1.2 is Murmur3Partitioner, so you will need to edit that to match your old partitioner if upgrading.",
+ descriptor, validationMetadata.partitioner, partitionerName));
+ System.exit(1);
+ }
+ logger.info("Opening {} ({} bytes)", descriptor, new File(descriptor.filenameFor(Component.DATA)).length());
SSTableReader sstable = new SSTableReader(descriptor,
components,
metadata,
@@@ -603,48 -374,35 +603,43 @@@
this.dfile = dfile;
this.indexSummary = indexSummary;
this.bf = bloomFilter;
+ this.setup(false);
}
- /**
- * Clean up all opened resources.
- *
- * @throws IOException
- */
- public void close() throws IOException
+ public static long getTotalBytes(Iterable<SSTableReader> sstables)
{
- if (readMeterSyncFuture != null)
- readMeterSyncFuture.cancel(false);
+ long sum = 0;
+ for (SSTableReader sstable : sstables)
+ sum += sstable.onDiskLength();
+ return sum;
+ }
- // Force finalizing mmapping if necessary
+ public static long getTotalUncompressedBytes(Iterable<SSTableReader> sstables)
+ {
+ long sum = 0;
+ for (SSTableReader sstable : sstables)
+ sum += sstable.uncompressedLength();
- if (null != ifile)
- ifile.cleanup();
+ return sum;
+ }
+
+ public boolean equals(Object that)
+ {
+ return that instanceof SSTableReader && ((SSTableReader) that).descriptor.equals(this.descriptor);
+ }
- dfile.cleanup();
- // close the BF so it can be opened later.
- if (null != bf)
- bf.close();
+ public int hashCode()
+ {
+ return this.descriptor.hashCode();
+ }
- if (null != indexSummary)
- indexSummary.close();
+ public String getFilename()
+ {
+ return dfile.path;
}
- public String getIndexFilename()
- {
- return ifile.path;
- }
-
- public void setTrackedBy(DataTracker tracker)
+ public void setupKeyCache()
{
- deletingTask.setTracker(tracker);
// under normal operation we can do this at any time, but SSTR is also used outside C* proper,
// e.g. by BulkLoader, which does not initialize the cache. As a kludge, we set up the cache
// here when we know we're being wired into the rest of the server infrastructure.
@@@ -659,7 -417,13 +654,13 @@@
load(false, true);
bf = FilterFactory.AlwaysPresent;
}
+ else if (!components.contains(Component.PRIMARY_INDEX))
+ {
+ // avoid any reading of the missing primary index component.
- // this should only happen during StandaloneScrubber
++ // this should only happen for standalone tools
+ load(false, false);
+ }
- else if (!components.contains(Component.FILTER))
+ else if (!components.contains(Component.FILTER) || validation == null)
{
// bf is enabled, but filter component is missing.
load(true, true);
@@@ -708,418 -467,26 +709,427 @@@
? SegmentedFile.getCompressedBuilder()
: SegmentedFile.getBuilder(DatabaseDescriptor.getDiskAccessMode());
- boolean summaryLoaded = loadSummary(this, ibuilder, dbuilder, metadata);
+ boolean summaryLoaded = loadSummary(ibuilder, dbuilder);
+ boolean builtSummary = false;
if (recreateBloomFilter || !summaryLoaded)
- buildSummary(recreateBloomFilter, ibuilder, dbuilder, summaryLoaded);
+ {
+ buildSummary(recreateBloomFilter, ibuilder, dbuilder, summaryLoaded, Downsampling.BASE_SAMPLING_LEVEL);
+ builtSummary = true;
+ }
- ifile = ibuilder.complete(descriptor.filenameFor(Component.PRIMARY_INDEX));
+ if (components.contains(Component.PRIMARY_INDEX))
+ ifile = ibuilder.complete(descriptor.filenameFor(Component.PRIMARY_INDEX));
+
dfile = dbuilder.complete(descriptor.filenameFor(Component.DATA));
- if (saveSummaryIfCreated && (recreateBloomFilter || !summaryLoaded)) // save summary information to disk
- saveSummary(this, ibuilder, dbuilder);
+
+ // Check for an index summary that was downsampled even though the serialization format doesn't support
+ // that. If it was downsampled, rebuild it. See CASSANDRA-8993 for details.
- if (!descriptor.version.hasSamplingLevel && !builtSummary && !validateSummarySamplingLevel())
++ if (!descriptor.version.hasSamplingLevel && !builtSummary && !validateSummarySamplingLevel() && ifile != null)
+ {
+ indexSummary.close();
+ ifile.close();
+ dfile.close();
+
+ logger.info("Detected erroneously downsampled index summary; will rebuild summary at full sampling");
+ FileUtils.deleteWithConfirm(new File(descriptor.filenameFor(Component.SUMMARY)));
+ ibuilder = SegmentedFile.getBuilder(DatabaseDescriptor.getIndexAccessMode());
+ dbuilder = compression
+ ? SegmentedFile.getCompressedBuilder()
+ : SegmentedFile.getBuilder(DatabaseDescriptor.getDiskAccessMode());
+ buildSummary(false, ibuilder, dbuilder, false, Downsampling.BASE_SAMPLING_LEVEL);
+ ifile = ibuilder.complete(descriptor.filenameFor(Component.PRIMARY_INDEX));
+ dfile = dbuilder.complete(descriptor.filenameFor(Component.DATA));
+ saveSummary(ibuilder, dbuilder);
+ }
+ else if (saveSummaryIfCreated && builtSummary)
+ {
+ saveSummary(ibuilder, dbuilder);
+ }
}
- private void buildSummary(boolean recreateBloomFilter, SegmentedFile.Builder ibuilder, SegmentedFile.Builder dbuilder, boolean summaryLoaded) throws IOException
- {
+ /**
+ * Build index summary(and optionally bloom filter) by reading through Index.db file.
+ *
+ * @param recreateBloomFilter true if recreate bloom filter
+ * @param ibuilder
+ * @param dbuilder
+ * @param summaryLoaded true if index summary is already loaded and not need to build again
+ * @throws IOException
+ */
+ private void buildSummary(boolean recreateBloomFilter, SegmentedFile.Builder ibuilder, SegmentedFile.Builder dbuilder, boolean summaryLoaded, int samplingLevel) throws IOException
+ {
+ if (!components.contains(Component.PRIMARY_INDEX))
+ return;
+
// we read the positions in a BRAF so we don't have to worry about an entry spanning a mmap boundary.
RandomAccessReader primaryIndex = RandomAccessReader.open(new File(descriptor.filenameFor(Component.PRIMARY_INDEX)));
-
+
+ try
+ {
+ long indexSize = primaryIndex.length();
+ long histogramCount = sstableMetadata.estimatedRowSize.count();
+ long estimatedKeys = histogramCount > 0 && !sstableMetadata.estimatedRowSize.isOverflowed()
+ ? histogramCount
+ : estimateRowsFromIndex(primaryIndex); // statistics is supposed to be optional
+
+ try(IndexSummaryBuilder summaryBuilder = summaryLoaded ? null : new IndexSummaryBuilder(estimatedKeys, metadata.getMinIndexInterval(), samplingLevel))
+ {
+
+ if (recreateBloomFilter)
+ bf = FilterFactory.getFilter(estimatedKeys, metadata.getBloomFilterFpChance(), true);
+
+ long indexPosition;
+ while ((indexPosition = primaryIndex.getFilePointer()) != indexSize)
+ {
+ ByteBuffer key = ByteBufferUtil.readWithShortLength(primaryIndex);
+ RowIndexEntry indexEntry = metadata.comparator.rowIndexEntrySerializer().deserialize(primaryIndex, descriptor.version);
+ DecoratedKey decoratedKey = partitioner.decorateKey(key);
+ if (first == null)
+ first = decoratedKey;
+ last = decoratedKey;
+
+ if (recreateBloomFilter)
+ bf.add(decoratedKey.getKey());
+
+ // if summary was already read from disk we don't want to re-populate it using primary index
+ if (!summaryLoaded)
+ {
+ summaryBuilder.maybeAddEntry(decoratedKey, indexPosition);
+ ibuilder.addPotentialBoundary(indexPosition);
+ dbuilder.addPotentialBoundary(indexEntry.position);
+ }
+ }
+
+ if (!summaryLoaded)
+ indexSummary = summaryBuilder.build(partitioner);
+ }
+ }
+ finally
+ {
+ FileUtils.closeQuietly(primaryIndex);
+ }
+
+ first = getMinimalKey(first);
+ last = getMinimalKey(last);
+ }
+
+ /**
+ * Load index summary from Summary.db file if it exists.
+ *
+ * if loaded index summary has different index interval from current value stored in schema,
+ * then Summary.db file will be deleted and this returns false to rebuild summary.
+ *
+ * @param ibuilder
+ * @param dbuilder
+ * @return true if index summary is loaded successfully from Summary.db file.
+ */
+ public boolean loadSummary(SegmentedFile.Builder ibuilder, SegmentedFile.Builder dbuilder)
+ {
+ File summariesFile = new File(descriptor.filenameFor(Component.SUMMARY));
+ if (!summariesFile.exists())
+ return false;
+
+ DataInputStream iStream = null;
+ try
+ {
+ iStream = new DataInputStream(new FileInputStream(summariesFile));
+ indexSummary = IndexSummary.serializer.deserialize(
+ iStream, partitioner, descriptor.version.hasSamplingLevel,
+ metadata.getMinIndexInterval(), metadata.getMaxIndexInterval());
+ first = partitioner.decorateKey(ByteBufferUtil.readWithLength(iStream));
+ last = partitioner.decorateKey(ByteBufferUtil.readWithLength(iStream));
+ ibuilder.deserializeBounds(iStream);
+ dbuilder.deserializeBounds(iStream);
+ }
+ catch (IOException e)
+ {
+ if (indexSummary != null)
+ indexSummary.close();
+ logger.debug("Cannot deserialize SSTable Summary File {}: {}", summariesFile.getPath(), e.getMessage());
+ // corrupted; delete it and fall back to creating a new summary
+ FileUtils.closeQuietly(iStream);
+ // delete it and fall back to creating a new summary
+ FileUtils.deleteWithConfirm(summariesFile);
+ return false;
+ }
+ finally
+ {
+ FileUtils.closeQuietly(iStream);
+ }
+
+ return true;
+ }
+
+ /**
+ * Validates that an index summary has full sampling, as expected when the serialization format does not support
+ * persisting the sampling level.
+ * @return true if the summary has full sampling, false otherwise
+ */
+ private boolean validateSummarySamplingLevel()
+ {
+ // We need to check index summary entries against the index to verify that none of them were dropped due to
+ // downsampling. Downsampling can drop any of the first BASE_SAMPLING_LEVEL entries (repeating that drop pattern
+ // for the remainder of the summary). Unfortunately, the first entry to be dropped is the entry at
+ // index (BASE_SAMPLING_LEVEL - 1), so we need to check a full set of BASE_SAMPLING_LEVEL entries.
++ if (ifile == null)
++ return false;
++
+ Iterator<FileDataInput> segments = ifile.iterator(0);
+ int i = 0;
+ int summaryEntriesChecked = 0;
+ int expectedIndexInterval = getMinIndexInterval();
+ while (segments.hasNext())
+ {
+ FileDataInput in = segments.next();
+ try
+ {
+ while (!in.isEOF())
+ {
+ ByteBuffer indexKey = ByteBufferUtil.readWithShortLength(in);
+ if (i % expectedIndexInterval == 0)
+ {
+ ByteBuffer summaryKey = ByteBuffer.wrap(indexSummary.getKey(i / expectedIndexInterval));
+ if (!summaryKey.equals(indexKey))
+ return false;
+ summaryEntriesChecked++;
+
+ if (summaryEntriesChecked == Downsampling.BASE_SAMPLING_LEVEL)
+ return true;
+ }
+ RowIndexEntry.Serializer.skip(in);
+ i++;
+ }
+ }
+ catch (IOException e)
+ {
+ markSuspect();
+ throw new CorruptSSTableException(e, in.getPath());
+ }
+ finally
+ {
+ FileUtils.closeQuietly(in);
+ }
+ }
+
+ return true;
+ }
+
+ /**
+ * Save index summary to Summary.db file.
+ *
+ * @param ibuilder
+ * @param dbuilder
+ */
+ public void saveSummary(SegmentedFile.Builder ibuilder, SegmentedFile.Builder dbuilder)
+ {
+ saveSummary(ibuilder, dbuilder, indexSummary);
+ }
+
+ private void saveSummary(SegmentedFile.Builder ibuilder, SegmentedFile.Builder dbuilder, IndexSummary summary)
+ {
+ File summariesFile = new File(descriptor.filenameFor(Component.SUMMARY));
+ if (summariesFile.exists())
+ FileUtils.deleteWithConfirm(summariesFile);
+
+ DataOutputStreamAndChannel oStream = null;
+ try
+ {
+ oStream = new DataOutputStreamAndChannel(new FileOutputStream(summariesFile));
+ IndexSummary.serializer.serialize(summary, oStream, descriptor.version.hasSamplingLevel);
+ ByteBufferUtil.writeWithLength(first.getKey(), oStream);
+ ByteBufferUtil.writeWithLength(last.getKey(), oStream);
+ ibuilder.serializeBounds(oStream);
+ dbuilder.serializeBounds(oStream);
+ }
+ catch (IOException e)
+ {
+ logger.debug("Cannot save SSTable Summary: ", e);
+
+ // corrupted hence delete it and let it load it now.
+ if (summariesFile.exists())
+ FileUtils.deleteWithConfirm(summariesFile);
+ }
+ finally
+ {
+ FileUtils.closeQuietly(oStream);
+ }
+ }
+
+ public void setReplacedBy(SSTableReader replacement)
+ {
+ synchronized (tidy.global)
+ {
+ assert replacement != null;
+ assert !tidy.isReplaced;
+ tidy.isReplaced = true;
+ }
+ }
+
+ public boolean isReplaced()
+ {
+ return tidy.isReplaced;
+ }
+
+ /**
+ * Clone this reader with the provided start and open reason, and set the clone as replacement.
+ *
+ * @param newFirst the first key for the replacement (which can be different from the original due to the pre-emptive
+ * opening of compaction results).
+ * @param reason the {@code OpenReason} for the replacement.
+ *
+ * @return the cloned reader. That reader is set as a replacement by the method.
+ */
+ private SSTableReader cloneAndReplace(DecoratedKey newFirst, OpenReason reason)
+ {
+ return cloneAndReplace(newFirst, reason, indexSummary.sharedCopy());
+ }
+
+ /**
+ * Clone this reader with the new values and set the clone as replacement.
+ *
+ * @param newFirst the first key for the replacement (which can be different from the original due to the pre-emptive
+ * opening of compaction results).
+ * @param reason the {@code OpenReason} for the replacement.
+ * @param newSummary the index summary for the replacement.
+ *
+ * @return the cloned reader. That reader is set as a replacement by the method.
+ */
+ private SSTableReader cloneAndReplace(DecoratedKey newFirst, OpenReason reason, IndexSummary newSummary)
+ {
+ SSTableReader replacement = internalOpen(descriptor,
+ components,
+ metadata,
+ partitioner,
- ifile.sharedCopy(),
++ ifile != null ? ifile.sharedCopy() : null,
+ dfile.sharedCopy(),
+ newSummary,
+ bf.sharedCopy(),
+ maxDataAge,
+ sstableMetadata,
+ reason);
+ replacement.first = newFirst;
+ replacement.last = last;
+ replacement.isSuspect.set(isSuspect.get());
+ setReplacedBy(replacement);
+ return replacement;
+ }
+
+ // runOnClose must NOT be an anonymous or non-static inner class, nor must it retain a reference chain to this reader
+ public SSTableReader cloneWithNewStart(DecoratedKey newStart, final Runnable runOnClose)
+ {
+ synchronized (tidy.global)
+ {
+ assert openReason != OpenReason.EARLY;
+ // TODO: merge with caller's firstKeyBeyond() work,to save time
+ if (newStart.compareTo(first) > 0)
+ {
+ final long dataStart = getPosition(newStart, Operator.EQ).position;
+ final long indexStart = getIndexScanPosition(newStart);
+ this.tidy.runOnClose = new DropPageCache(dfile, dataStart, ifile, indexStart, runOnClose);
+ }
+
+ return cloneAndReplace(newStart, OpenReason.MOVED_START);
+ }
+ }
+
+ // runOnClose must NOT be an anonymous or non-static inner class, nor must it retain a reference chain to this reader
+ public SSTableReader cloneAsShadowed(final Runnable runOnClose)
+ {
+ synchronized (tidy.global)
+ {
+ assert openReason != OpenReason.EARLY;
+ this.tidy.runOnClose = new DropPageCache(dfile, 0, ifile, 0, runOnClose);
+ return cloneAndReplace(first, OpenReason.SHADOWED);
+ }
+ }
+
+ private static class DropPageCache implements Runnable
+ {
+ final SegmentedFile dfile;
+ final long dfilePosition;
+ final SegmentedFile ifile;
+ final long ifilePosition;
+ final Runnable andThen;
+
+ private DropPageCache(SegmentedFile dfile, long dfilePosition, SegmentedFile ifile, long ifilePosition, Runnable andThen)
+ {
+ this.dfile = dfile;
+ this.dfilePosition = dfilePosition;
+ this.ifile = ifile;
+ this.ifilePosition = ifilePosition;
+ this.andThen = andThen;
+ }
+
+ public void run()
+ {
+ dfile.dropPageCache(dfilePosition);
- ifile.dropPageCache(ifilePosition);
++ if (ifile != null)
++ ifile.dropPageCache(ifilePosition);
+ andThen.run();
+ }
+ }
+
+ /**
+ * Returns a new SSTableReader with the same properties as this SSTableReader except that a new IndexSummary will
+ * be built at the target samplingLevel. This (original) SSTableReader instance will be marked as replaced, have
+ * its DeletingTask removed, and have its periodic read-meter sync task cancelled.
+ * @param samplingLevel the desired sampling level for the index summary on the new SSTableReader
+ * @return a new SSTableReader
+ * @throws IOException
+ */
+ public SSTableReader cloneWithNewSummarySamplingLevel(ColumnFamilyStore parent, int samplingLevel) throws IOException
+ {
+ assert descriptor.version.hasSamplingLevel;
+
+ synchronized (tidy.global)
+ {
+ assert openReason != OpenReason.EARLY;
+
+ int minIndexInterval = metadata.getMinIndexInterval();
+ int maxIndexInterval = metadata.getMaxIndexInterval();
+ double effectiveInterval = indexSummary.getEffectiveIndexInterval();
+
+ IndexSummary newSummary;
+ long oldSize = bytesOnDisk();
+
+ // We have to rebuild the summary from the on-disk primary index in three cases:
+ // 1. The sampling level went up, so we need to read more entries off disk
+ // 2. The min_index_interval changed (in either direction); this changes what entries would be in the summary
+ // at full sampling (and consequently at any other sampling level)
+ // 3. The max_index_interval was lowered, forcing us to raise the sampling level
+ if (samplingLevel > indexSummary.getSamplingLevel() || indexSummary.getMinIndexInterval() != minIndexInterval || effectiveInterval > maxIndexInterval)
+ {
+ newSummary = buildSummaryAtLevel(samplingLevel);
+ }
+ else if (samplingLevel < indexSummary.getSamplingLevel())
+ {
+ // we can use the existing index summary to make a smaller one
+ newSummary = IndexSummaryBuilder.downsample(indexSummary, samplingLevel, minIndexInterval, partitioner);
+
+ SegmentedFile.Builder ibuilder = SegmentedFile.getBuilder(DatabaseDescriptor.getIndexAccessMode());
+ SegmentedFile.Builder dbuilder = compression
+ ? SegmentedFile.getCompressedBuilder()
+ : SegmentedFile.getBuilder(DatabaseDescriptor.getDiskAccessMode());
+ saveSummary(ibuilder, dbuilder, newSummary);
+ }
+ else
+ {
+ throw new AssertionError("Attempted to clone SSTableReader with the same index summary sampling level and " +
+ "no adjustments to min/max_index_interval");
+ }
+
+ long newSize = bytesOnDisk();
+ StorageMetrics.load.inc(newSize - oldSize);
+ parent.metric.liveDiskSpaceUsed.inc(newSize - oldSize);
+
+ return cloneAndReplace(first, OpenReason.METADATA_CHANGE, newSummary);
+ }
+ }
+
+ private IndexSummary buildSummaryAtLevel(int newSamplingLevel) throws IOException
+ {
+ // we read the positions in a BRAF so we don't have to worry about an entry spanning a mmap boundary.
+ RandomAccessReader primaryIndex = RandomAccessReader.open(new File(descriptor.filenameFor(Component.PRIMARY_INDEX)));
try
{
long indexSize = primaryIndex.length();
@@@ -1566,16 -974,22 +1576,19 @@@
{
if (op == Operator.EQ && updateCacheAndStats)
bloomFilterTracker.addFalsePositive();
- // we matched the -1th position: if the operator might match forward, we'll start at the first
- // position. We however need to return the correct index entry for that first position.
- if (op.apply(1) >= 0)
- {
- sampledPosition = 0;
- }
- else
- {
- Tracing.trace("Partition summary allows skipping sstable {}", descriptor.generation);
- return null;
- }
+ Tracing.trace("Check against min and max keys allows skipping sstable {}", descriptor.generation);
+ return null;
}
+ int binarySearchResult = indexSummary.binarySearch(key);
+ long sampledPosition = getIndexScanPositionFromBinarySearchResult(binarySearchResult, indexSummary);
+ int sampledIndex = getIndexSummaryIndexFromBinarySearchResult(binarySearchResult);
+
+ int effectiveInterval = indexSummary.getEffectiveIndexIntervalAfterIndex(sampledIndex);
+
+ if (ifile == null)
+ return null;
+
// scan the on-disk index, starting at the nearest sampled position.
// The check against IndexInterval is to be exit the loop in the EQ case when the key looked for is not present
// (bloom filter false positive). But note that for non-EQ cases, we might need to check the first key of the
@@@ -1670,11 -1084,13 +1683,14 @@@
*/
public DecoratedKey firstKeyBeyond(RowPosition token)
{
+ if (token.compareTo(first) < 0)
+ return first;
+
long sampledPosition = getIndexScanPosition(token);
- if (sampledPosition == -1)
- sampledPosition = 0;
+ if (ifile == null)
+ return null;
+
Iterator<FileDataInput> segments = ifile.iterator(sampledPosition);
while (segments.hasNext())
{
@@@ -1948,8 -1391,7 +1964,10 @@@
{
try
{
- return SSTableMetadata.serializer.deserialize(descriptor).right;
+ CompactionMetadata compactionMetadata = (CompactionMetadata) descriptor.getMetadataSerializer().deserialize(descriptor, MetadataType.COMPACTION);
- return compactionMetadata.ancestors;
++ if (compactionMetadata != null)
++ return compactionMetadata.ancestors;
++ return Collections.emptySet();
}
catch (IOException e)
{
@@@ -1995,7 -1441,7 +2013,9 @@@
public RandomAccessReader openIndexReader()
{
- return ifile.createReader();
- return RandomAccessReader.open(new File(getIndexFilename()));
++ if (ifile != null)
++ return ifile.createReader();
++ return null;
}
/**
@@@ -2024,152 -1470,73 +2044,154 @@@
}
/**
- * @param sstables
- * @return true if all desired references were acquired. Otherwise, it will unreference any partial acquisition, and return false.
+ * Increment the total row read count and read rate for this SSTable. This should not be incremented for range
+ * slice queries, row cache hits, or non-query reads, like compaction.
*/
- public static boolean acquireReferences(Iterable<SSTableReader> sstables)
+ public void incrementReadCount()
{
- SSTableReader failed = null;
- for (SSTableReader sstable : sstables)
+ if (readMeter != null)
+ readMeter.mark();
+ }
+
+ public static class SizeComparator implements Comparator<SSTableReader>
+ {
+ public int compare(SSTableReader o1, SSTableReader o2)
{
- if (!sstable.acquireReference())
- {
- failed = sstable;
- break;
- }
+ return Longs.compare(o1.onDiskLength(), o2.onDiskLength());
}
+ }
- if (failed == null)
- return true;
+ public Ref<SSTableReader> tryRef()
+ {
+ return selfRef.tryRef();
+ }
- for (SSTableReader sstable : sstables)
- {
- if (sstable == failed)
- break;
- sstable.releaseReference();
- }
- return false;
+ public Ref<SSTableReader> selfRef()
+ {
+ return selfRef;
}
- public static void releaseReferences(Iterable<SSTableReader> sstables)
+ public Ref<SSTableReader> ref()
{
- for (SSTableReader sstable : sstables)
- {
- sstable.releaseReference();
- }
+ return selfRef.ref();
}
- private void dropPageCache()
+ void setup(boolean isOffline)
{
- dropPageCache(dfile.path);
- if (null != ifile)
- dropPageCache(ifile.path);
+ tidy.setup(this, isOffline);
+ this.readMeter = tidy.global.readMeter;
}
- private void dropPageCache(String filePath)
+ @VisibleForTesting
+ public void overrideReadMeter(RestorableMeter readMeter)
{
- RandomAccessFile file = null;
+ this.readMeter = tidy.global.readMeter = readMeter;
+ }
- try
+ /**
+ * One instance per SSTableReader we create. This references the type-shared tidy, which in turn references
+ * the globally shared tidy, i.e.
+ *
+ * InstanceTidier => DescriptorTypeTitdy => GlobalTidy
+ *
+ * We can create many InstanceTidiers (one for every time we reopen an sstable with MOVED_START for example), but there can only be
+ * two DescriptorTypeTidy (FINAL and TEMPLINK) and only one GlobalTidy for one single logical sstable.
+ *
+ * When the InstanceTidier cleansup, it releases its reference to its DescriptorTypeTidy; when all InstanceTidiers
+ * for that type have run, the DescriptorTypeTidy cleansup. DescriptorTypeTidy behaves in the same way towards GlobalTidy.
+ *
+ * For ease, we stash a direct reference to both our type-shared and global tidier
+ */
+ private static final class InstanceTidier implements Tidy
+ {
+ private final Descriptor descriptor;
+ private final CFMetaData metadata;
+ private IFilter bf;
+ private IndexSummary summary;
+
+ private SegmentedFile dfile;
+ private SegmentedFile ifile;
+ private Runnable runOnClose;
+ private boolean isReplaced = false;
+
+ // a reference to our shared per-Descriptor.Type tidy instance, that
+ // we will release when we are ourselves released
+ private Ref<DescriptorTypeTidy> typeRef;
+
+ // a convenience stashing of the shared per-descriptor-type tidy instance itself
+ // and the per-logical-sstable globally shared state that it is linked to
+ private DescriptorTypeTidy type;
+ private GlobalTidy global;
+
+ private boolean setup;
+
+ void setup(SSTableReader reader, boolean isOffline)
+ {
+ this.setup = true;
+ this.bf = reader.bf;
+ this.summary = reader.indexSummary;
+ this.dfile = reader.dfile;
+ this.ifile = reader.ifile;
+ // get a new reference to the shared descriptor-type tidy
+ this.typeRef = DescriptorTypeTidy.get(reader);
+ this.type = typeRef.get();
+ this.global = type.globalRef.get();
+ if (!isOffline)
+ global.ensureReadMeter();
+ }
+
+ InstanceTidier(Descriptor descriptor, CFMetaData metadata)
{
- file = new RandomAccessFile(filePath, "r");
+ this.descriptor = descriptor;
+ this.metadata = metadata;
+ }
- int fd = CLibrary.getfd(file.getFD());
+ public void tidy()
+ {
+ // don't try to cleanup if the sstablereader was never fully constructed
+ if (!setup)
+ return;
- if (fd > 0)
+ final ColumnFamilyStore cfs = Schema.instance.getColumnFamilyStoreInstance(metadata.cfId);
+ final OpOrder.Barrier barrier;
+ if (cfs != null)
{
- if (logger.isDebugEnabled())
- logger.debug(String.format("Dropping page cache of file %s.", filePath));
-
- CLibrary.trySkipCache(fd, 0, 0);
+ barrier = cfs.readOrdering.newBarrier();
+ barrier.issue();
}
+ else
+ barrier = null;
+
+ ScheduledExecutors.nonPeriodicTasks.execute(new Runnable()
+ {
+ public void run()
+ {
+ if (barrier != null)
+ barrier.await();
- bf.close();
++ if (bf != null)
++ bf.close();
+ if (summary != null)
+ summary.close();
+ if (runOnClose != null)
+ runOnClose.run();
+ dfile.close();
- ifile.close();
++ if (ifile != null)
++ ifile.close();
+ typeRef.release();
+ }
+ });
}
- catch (IOException e)
+
+ public String name()
{
- // we don't care if cache cleanup fails
+ return descriptor.toString();
}
- finally
+
+ void releaseSummary()
{
- FileUtils.closeQuietly(file);
+ summary.close();
+ assert summary.isCleanedUp();
+ summary = null;
}
}
http://git-wip-us.apache.org/repos/asf/cassandra/blob/4c94ef20/src/java/org/apache/cassandra/tools/StandaloneScrubber.java
----------------------------------------------------------------------
http://git-wip-us.apache.org/repos/asf/cassandra/blob/4c94ef20/test/unit/org/apache/cassandra/db/ScrubTest.java
----------------------------------------------------------------------
[06/12] cassandra git commit: Merge branch 'cassandra-2.0' into
cassandra-2.1
Posted by be...@apache.org.
Merge branch 'cassandra-2.0' into cassandra-2.1
Conflicts:
CHANGES.txt
src/java/org/apache/cassandra/db/compaction/Scrubber.java
src/java/org/apache/cassandra/io/sstable/SSTableReader.java
Project: http://git-wip-us.apache.org/repos/asf/cassandra/repo
Commit: http://git-wip-us.apache.org/repos/asf/cassandra/commit/4c94ef20
Tree: http://git-wip-us.apache.org/repos/asf/cassandra/tree/4c94ef20
Diff: http://git-wip-us.apache.org/repos/asf/cassandra/diff/4c94ef20
Branch: refs/heads/cassandra-2.1
Commit: 4c94ef20d3562ab8f0a922945d78464d6c475d98
Parents: 4de943f 452d6a4
Author: Benedict Elliott Smith <be...@apache.org>
Authored: Tue Jul 7 16:27:58 2015 +0100
Committer: Benedict Elliott Smith <be...@apache.org>
Committed: Tue Jul 7 16:27:58 2015 +0100
----------------------------------------------------------------------
CHANGES.txt | 5 ++
.../cassandra/db/compaction/Scrubber.java | 37 +++++++++++---
.../cassandra/io/sstable/SSTableReader.java | 52 ++++++++++++++------
.../cassandra/tools/StandaloneScrubber.java | 2 +-
.../unit/org/apache/cassandra/db/ScrubTest.java | 25 ++++++++++
5 files changed, 97 insertions(+), 24 deletions(-)
----------------------------------------------------------------------
http://git-wip-us.apache.org/repos/asf/cassandra/blob/4c94ef20/CHANGES.txt
----------------------------------------------------------------------
diff --cc CHANGES.txt
index 95dc8be,bd1db92..2cbc7c4
--- a/CHANGES.txt
+++ b/CHANGES.txt
@@@ -1,18 -1,8 +1,23 @@@
-2.0.18
-* Scrub (recover) sstables even when -Index.db is missing, (CASSANDRA-9591)
-
-
-2.0.17
++2.1.9
++Merged from 2.0:
++ * Scrub (recover) sstables even when -Index.db is missing, (CASSANDRA-9591)
++
++
+2.1.8
+ * (cqlsh) Fix bad check for CQL compatibility when DESCRIBE'ing
+ COMPACT STORAGE tables with no clustering columns
+ * Warn when an extra-large partition is compacted (CASSANDRA-9643)
+ * Eliminate strong self-reference chains in sstable ref tidiers (CASSANDRA-9656)
+ * Ensure StreamSession uses canonical sstable reader instances (CASSANDRA-9700)
+ * Ensure memtable book keeping is not corrupted in the event we shrink usage (CASSANDRA-9681)
+ * Update internal python driver for cqlsh (CASSANDRA-9064)
+ * Fix IndexOutOfBoundsException when inserting tuple with too many
+ elements using the string literal notation (CASSANDRA-9559)
+ * Allow JMX over SSL directly from nodetool (CASSANDRA-9090)
+ * Fix incorrect result for IN queries where column not found (CASSANDRA-9540)
+ * Enable describe on indices (CASSANDRA-7814)
+ * ColumnFamilyStore.selectAndReference may block during compaction (CASSANDRA-9637)
+Merged from 2.0:
* Avoid NPE in AuthSuccess#decode (CASSANDRA-9727)
* Add listen_address to system.local (CASSANDRA-9603)
* Bug fixes to resultset metadata construction (CASSANDRA-9636)
http://git-wip-us.apache.org/repos/asf/cassandra/blob/4c94ef20/src/java/org/apache/cassandra/db/compaction/Scrubber.java
----------------------------------------------------------------------
diff --cc src/java/org/apache/cassandra/db/compaction/Scrubber.java
index ce98a13,dc60efa..b1c12e0
--- a/src/java/org/apache/cassandra/db/compaction/Scrubber.java
+++ b/src/java/org/apache/cassandra/db/compaction/Scrubber.java
@@@ -100,8 -95,17 +100,18 @@@ public class Scrubber implements Closea
this.controller = isOffline
? new ScrubController(cfs)
: new CompactionController(cfs, Collections.singleton(sstable), CompactionManager.getDefaultGcBefore(cfs));
- this.isCommutative = cfs.metadata.getDefaultValidator().isCommutative();
+ this.isCommutative = cfs.metadata.isCounter();
- this.expectedBloomFilterSize = Math.max(cfs.metadata.getMinIndexInterval(), (int)(SSTableReader.getApproximateKeyCount(toScrub)));
+
+ boolean hasIndexFile = (new File(sstable.descriptor.filenameFor(Component.PRIMARY_INDEX))).exists();
+ if (!hasIndexFile)
+ {
+ // if there's any corruption in the -Data.db then rows can't be skipped over. but it's worth a shot.
+ outputHandler.warn("Missing component: " + sstable.descriptor.filenameFor(Component.PRIMARY_INDEX));
+ }
+
- this.expectedBloomFilterSize = Math.max(cfs.metadata.getIndexInterval(),
- hasIndexFile ? (int)(SSTableReader.getApproximateKeyCount(toScrub,cfs.metadata)) : 0);
++ this.expectedBloomFilterSize = Math.max(
++ cfs.metadata.getMinIndexInterval(),
++ hasIndexFile ? (int)(SSTableReader.getApproximateKeyCount(toScrub)) : 0);
// loop through each row, deserializing to check for damage.
// we'll also loop through the index at the same time, using the position from the index to recover if the
@@@ -120,14 -128,13 +134,15 @@@
public void scrub()
{
outputHandler.output(String.format("Scrubbing %s (%s bytes)", sstable, dataFile.length()));
+ Set<SSTableReader> oldSSTable = Sets.newHashSet(sstable);
+ SSTableRewriter writer = new SSTableRewriter(cfs, oldSSTable, sstable.maxDataAge, isOffline);
try
{
- nextIndexKey = ByteBufferUtil.readWithShortLength(indexFile);
+ nextIndexKey = indexAvailable() ? ByteBufferUtil.readWithShortLength(indexFile) : null;
+ if (indexAvailable())
{
// throw away variable so we don't have a side effect in the assert
- long firstRowPositionFromIndex = RowIndexEntry.serializer.deserialize(indexFile, sstable.descriptor.version).position;
+ long firstRowPositionFromIndex = sstable.metadata.comparator.rowIndexEntrySerializer().deserialize(indexFile, sstable.descriptor.version).position;
assert firstRowPositionFromIndex == 0 : firstRowPositionFromIndex;
}
@@@ -167,13 -179,25 +182,13 @@@
dataSizeFromIndex = nextRowPositionFromIndex - dataStartFromIndex;
}
- if (!sstable.descriptor.version.hasRowSizeAndColumnCount)
- {
- dataSize = dataSizeFromIndex;
- // avoid an NPE if key is null
- String keyName = key == null ? "(unreadable key)" : ByteBufferUtil.bytesToHex(key.key);
- outputHandler.debug(String.format("row %s is %s bytes", keyName, dataSize));
- }
- else
- {
- if (currentIndexKey != null)
- outputHandler.debug(String.format("Index doublecheck: row %s is %s bytes", ByteBufferUtil.bytesToHex(currentIndexKey), dataSizeFromIndex));
- }
+ dataSize = dataSizeFromIndex;
+ // avoid an NPE if key is null
+ String keyName = key == null ? "(unreadable key)" : ByteBufferUtil.bytesToHex(key.getKey());
+ outputHandler.debug(String.format("row %s is %s bytes", keyName, dataSize));
- assert currentIndexKey != null || indexFile.isEOF();
+ assert currentIndexKey != null || !indexAvailable();
- writer.mark();
try
{
if (key == null)
@@@ -188,11 -212,11 +203,11 @@@
if (dataSize > dataFile.length())
throw new IOError(new IOException("Impossible row size (greater than file length): " + dataSize));
- if (dataStart != dataStartFromIndex)
+ if (indexFile != null && dataStart != dataStartFromIndex)
outputHandler.warn(String.format("Data file row position %d differs from index file row position %d", dataStart, dataStartFromIndex));
- if (dataSize != dataSizeFromIndex)
+ if (indexFile != null && dataSize != dataSizeFromIndex)
- outputHandler.warn(String.format("Data file row size %d differs from index file row size %d", dataSize, dataSizeFromIndex));
+ outputHandler.warn(String.format("Data file row size %d different from index file row size %d", dataSize, dataSizeFromIndex));
SSTableIdentityIterator atoms = new SSTableIdentityIterator(sstable, dataFile, key, dataSize, validateColumns);
if (prevKey != null && prevKey.compareTo(key) > 0)
@@@ -312,10 -331,11 +327,11 @@@
currentRowPositionFromIndex = nextRowPositionFromIndex;
try
{
- nextIndexKey = indexFile.isEOF() ? null : ByteBufferUtil.readWithShortLength(indexFile);
- nextRowPositionFromIndex = indexFile.isEOF()
+ nextIndexKey = !indexAvailable() ? null : ByteBufferUtil.readWithShortLength(indexFile);
+
+ nextRowPositionFromIndex = !indexAvailable()
? dataFile.length()
- : RowIndexEntry.serializer.deserialize(indexFile, sstable.descriptor.version).position;
+ : sstable.metadata.comparator.rowIndexEntrySerializer().deserialize(indexFile, sstable.descriptor.version).position;
}
catch (Throwable th)
{
http://git-wip-us.apache.org/repos/asf/cassandra/blob/4c94ef20/src/java/org/apache/cassandra/io/sstable/SSTableReader.java
----------------------------------------------------------------------
diff --cc src/java/org/apache/cassandra/io/sstable/SSTableReader.java
index 7551d46,8919a09..6879834
--- a/src/java/org/apache/cassandra/io/sstable/SSTableReader.java
+++ b/src/java/org/apache/cassandra/io/sstable/SSTableReader.java
@@@ -448,27 -190,9 +448,27 @@@ public class SSTableReader extends SSTa
IPartitioner partitioner,
boolean validate) throws IOException
{
- long start = System.nanoTime();
- SSTableMetadata sstableMetadata = openMetadata(descriptor, components, partitioner, validate);
+ // Minimum components without which we can't do anything
+ assert components.contains(Component.DATA) : "Data component is missing for sstable" + descriptor;
- assert components.contains(Component.PRIMARY_INDEX) : "Primary index component is missing for sstable " + descriptor;
++ assert !validate || components.contains(Component.PRIMARY_INDEX) : "Primary index component is missing for sstable " + descriptor;
+
+ Map<MetadataType, MetadataComponent> sstableMetadata = descriptor.getMetadataSerializer().deserialize(descriptor,
- EnumSet.of(MetadataType.VALIDATION, MetadataType.STATS));
++ EnumSet.of(MetadataType.VALIDATION, MetadataType.STATS));
+ ValidationMetadata validationMetadata = (ValidationMetadata) sstableMetadata.get(MetadataType.VALIDATION);
+ StatsMetadata statsMetadata = (StatsMetadata) sstableMetadata.get(MetadataType.STATS);
+
+ // Check if sstable is created using same partitioner.
+ // Partitioner can be null, which indicates older version of sstable or no stats available.
+ // In that case, we skip the check.
+ String partitionerName = partitioner.getClass().getCanonicalName();
+ if (validationMetadata != null && !partitionerName.equals(validationMetadata.partitioner))
+ {
+ logger.error(String.format("Cannot open %s; partitioner %s does not match system partitioner %s. Note that the default partitioner starting with Cassandra 1.2 is Murmur3Partitioner, so you will need to edit that to match your old partitioner if upgrading.",
+ descriptor, validationMetadata.partitioner, partitionerName));
+ System.exit(1);
+ }
+ logger.info("Opening {} ({} bytes)", descriptor, new File(descriptor.filenameFor(Component.DATA)).length());
SSTableReader sstable = new SSTableReader(descriptor,
components,
metadata,
@@@ -603,48 -374,35 +603,43 @@@
this.dfile = dfile;
this.indexSummary = indexSummary;
this.bf = bloomFilter;
+ this.setup(false);
}
- /**
- * Clean up all opened resources.
- *
- * @throws IOException
- */
- public void close() throws IOException
+ public static long getTotalBytes(Iterable<SSTableReader> sstables)
{
- if (readMeterSyncFuture != null)
- readMeterSyncFuture.cancel(false);
+ long sum = 0;
+ for (SSTableReader sstable : sstables)
+ sum += sstable.onDiskLength();
+ return sum;
+ }
- // Force finalizing mmapping if necessary
+ public static long getTotalUncompressedBytes(Iterable<SSTableReader> sstables)
+ {
+ long sum = 0;
+ for (SSTableReader sstable : sstables)
+ sum += sstable.uncompressedLength();
- if (null != ifile)
- ifile.cleanup();
+ return sum;
+ }
+
+ public boolean equals(Object that)
+ {
+ return that instanceof SSTableReader && ((SSTableReader) that).descriptor.equals(this.descriptor);
+ }
- dfile.cleanup();
- // close the BF so it can be opened later.
- if (null != bf)
- bf.close();
+ public int hashCode()
+ {
+ return this.descriptor.hashCode();
+ }
- if (null != indexSummary)
- indexSummary.close();
+ public String getFilename()
+ {
+ return dfile.path;
}
- public String getIndexFilename()
- {
- return ifile.path;
- }
-
- public void setTrackedBy(DataTracker tracker)
+ public void setupKeyCache()
{
- deletingTask.setTracker(tracker);
// under normal operation we can do this at any time, but SSTR is also used outside C* proper,
// e.g. by BulkLoader, which does not initialize the cache. As a kludge, we set up the cache
// here when we know we're being wired into the rest of the server infrastructure.
@@@ -659,7 -417,13 +654,13 @@@
load(false, true);
bf = FilterFactory.AlwaysPresent;
}
+ else if (!components.contains(Component.PRIMARY_INDEX))
+ {
+ // avoid any reading of the missing primary index component.
- // this should only happen during StandaloneScrubber
++ // this should only happen for standalone tools
+ load(false, false);
+ }
- else if (!components.contains(Component.FILTER))
+ else if (!components.contains(Component.FILTER) || validation == null)
{
// bf is enabled, but filter component is missing.
load(true, true);
@@@ -708,418 -467,26 +709,427 @@@
? SegmentedFile.getCompressedBuilder()
: SegmentedFile.getBuilder(DatabaseDescriptor.getDiskAccessMode());
- boolean summaryLoaded = loadSummary(this, ibuilder, dbuilder, metadata);
+ boolean summaryLoaded = loadSummary(ibuilder, dbuilder);
+ boolean builtSummary = false;
if (recreateBloomFilter || !summaryLoaded)
- buildSummary(recreateBloomFilter, ibuilder, dbuilder, summaryLoaded);
+ {
+ buildSummary(recreateBloomFilter, ibuilder, dbuilder, summaryLoaded, Downsampling.BASE_SAMPLING_LEVEL);
+ builtSummary = true;
+ }
- ifile = ibuilder.complete(descriptor.filenameFor(Component.PRIMARY_INDEX));
+ if (components.contains(Component.PRIMARY_INDEX))
+ ifile = ibuilder.complete(descriptor.filenameFor(Component.PRIMARY_INDEX));
+
dfile = dbuilder.complete(descriptor.filenameFor(Component.DATA));
- if (saveSummaryIfCreated && (recreateBloomFilter || !summaryLoaded)) // save summary information to disk
- saveSummary(this, ibuilder, dbuilder);
+
+ // Check for an index summary that was downsampled even though the serialization format doesn't support
+ // that. If it was downsampled, rebuild it. See CASSANDRA-8993 for details.
- if (!descriptor.version.hasSamplingLevel && !builtSummary && !validateSummarySamplingLevel())
++ if (!descriptor.version.hasSamplingLevel && !builtSummary && !validateSummarySamplingLevel() && ifile != null)
+ {
+ indexSummary.close();
+ ifile.close();
+ dfile.close();
+
+ logger.info("Detected erroneously downsampled index summary; will rebuild summary at full sampling");
+ FileUtils.deleteWithConfirm(new File(descriptor.filenameFor(Component.SUMMARY)));
+ ibuilder = SegmentedFile.getBuilder(DatabaseDescriptor.getIndexAccessMode());
+ dbuilder = compression
+ ? SegmentedFile.getCompressedBuilder()
+ : SegmentedFile.getBuilder(DatabaseDescriptor.getDiskAccessMode());
+ buildSummary(false, ibuilder, dbuilder, false, Downsampling.BASE_SAMPLING_LEVEL);
+ ifile = ibuilder.complete(descriptor.filenameFor(Component.PRIMARY_INDEX));
+ dfile = dbuilder.complete(descriptor.filenameFor(Component.DATA));
+ saveSummary(ibuilder, dbuilder);
+ }
+ else if (saveSummaryIfCreated && builtSummary)
+ {
+ saveSummary(ibuilder, dbuilder);
+ }
}
- private void buildSummary(boolean recreateBloomFilter, SegmentedFile.Builder ibuilder, SegmentedFile.Builder dbuilder, boolean summaryLoaded) throws IOException
- {
+ /**
+ * Build index summary(and optionally bloom filter) by reading through Index.db file.
+ *
+ * @param recreateBloomFilter true if recreate bloom filter
+ * @param ibuilder
+ * @param dbuilder
+ * @param summaryLoaded true if index summary is already loaded and not need to build again
+ * @throws IOException
+ */
+ private void buildSummary(boolean recreateBloomFilter, SegmentedFile.Builder ibuilder, SegmentedFile.Builder dbuilder, boolean summaryLoaded, int samplingLevel) throws IOException
+ {
+ if (!components.contains(Component.PRIMARY_INDEX))
+ return;
+
// we read the positions in a BRAF so we don't have to worry about an entry spanning a mmap boundary.
RandomAccessReader primaryIndex = RandomAccessReader.open(new File(descriptor.filenameFor(Component.PRIMARY_INDEX)));
-
+
+ try
+ {
+ long indexSize = primaryIndex.length();
+ long histogramCount = sstableMetadata.estimatedRowSize.count();
+ long estimatedKeys = histogramCount > 0 && !sstableMetadata.estimatedRowSize.isOverflowed()
+ ? histogramCount
+ : estimateRowsFromIndex(primaryIndex); // statistics is supposed to be optional
+
+ try(IndexSummaryBuilder summaryBuilder = summaryLoaded ? null : new IndexSummaryBuilder(estimatedKeys, metadata.getMinIndexInterval(), samplingLevel))
+ {
+
+ if (recreateBloomFilter)
+ bf = FilterFactory.getFilter(estimatedKeys, metadata.getBloomFilterFpChance(), true);
+
+ long indexPosition;
+ while ((indexPosition = primaryIndex.getFilePointer()) != indexSize)
+ {
+ ByteBuffer key = ByteBufferUtil.readWithShortLength(primaryIndex);
+ RowIndexEntry indexEntry = metadata.comparator.rowIndexEntrySerializer().deserialize(primaryIndex, descriptor.version);
+ DecoratedKey decoratedKey = partitioner.decorateKey(key);
+ if (first == null)
+ first = decoratedKey;
+ last = decoratedKey;
+
+ if (recreateBloomFilter)
+ bf.add(decoratedKey.getKey());
+
+ // if summary was already read from disk we don't want to re-populate it using primary index
+ if (!summaryLoaded)
+ {
+ summaryBuilder.maybeAddEntry(decoratedKey, indexPosition);
+ ibuilder.addPotentialBoundary(indexPosition);
+ dbuilder.addPotentialBoundary(indexEntry.position);
+ }
+ }
+
+ if (!summaryLoaded)
+ indexSummary = summaryBuilder.build(partitioner);
+ }
+ }
+ finally
+ {
+ FileUtils.closeQuietly(primaryIndex);
+ }
+
+ first = getMinimalKey(first);
+ last = getMinimalKey(last);
+ }
+
+ /**
+ * Load index summary from Summary.db file if it exists.
+ *
+ * if loaded index summary has different index interval from current value stored in schema,
+ * then Summary.db file will be deleted and this returns false to rebuild summary.
+ *
+ * @param ibuilder
+ * @param dbuilder
+ * @return true if index summary is loaded successfully from Summary.db file.
+ */
+ public boolean loadSummary(SegmentedFile.Builder ibuilder, SegmentedFile.Builder dbuilder)
+ {
+ File summariesFile = new File(descriptor.filenameFor(Component.SUMMARY));
+ if (!summariesFile.exists())
+ return false;
+
+ DataInputStream iStream = null;
+ try
+ {
+ iStream = new DataInputStream(new FileInputStream(summariesFile));
+ indexSummary = IndexSummary.serializer.deserialize(
+ iStream, partitioner, descriptor.version.hasSamplingLevel,
+ metadata.getMinIndexInterval(), metadata.getMaxIndexInterval());
+ first = partitioner.decorateKey(ByteBufferUtil.readWithLength(iStream));
+ last = partitioner.decorateKey(ByteBufferUtil.readWithLength(iStream));
+ ibuilder.deserializeBounds(iStream);
+ dbuilder.deserializeBounds(iStream);
+ }
+ catch (IOException e)
+ {
+ if (indexSummary != null)
+ indexSummary.close();
+ logger.debug("Cannot deserialize SSTable Summary File {}: {}", summariesFile.getPath(), e.getMessage());
+ // corrupted; delete it and fall back to creating a new summary
+ FileUtils.closeQuietly(iStream);
+ // delete it and fall back to creating a new summary
+ FileUtils.deleteWithConfirm(summariesFile);
+ return false;
+ }
+ finally
+ {
+ FileUtils.closeQuietly(iStream);
+ }
+
+ return true;
+ }
+
+ /**
+ * Validates that an index summary has full sampling, as expected when the serialization format does not support
+ * persisting the sampling level.
+ * @return true if the summary has full sampling, false otherwise
+ */
+ private boolean validateSummarySamplingLevel()
+ {
+ // We need to check index summary entries against the index to verify that none of them were dropped due to
+ // downsampling. Downsampling can drop any of the first BASE_SAMPLING_LEVEL entries (repeating that drop pattern
+ // for the remainder of the summary). Unfortunately, the first entry to be dropped is the entry at
+ // index (BASE_SAMPLING_LEVEL - 1), so we need to check a full set of BASE_SAMPLING_LEVEL entries.
++ if (ifile == null)
++ return false;
++
+ Iterator<FileDataInput> segments = ifile.iterator(0);
+ int i = 0;
+ int summaryEntriesChecked = 0;
+ int expectedIndexInterval = getMinIndexInterval();
+ while (segments.hasNext())
+ {
+ FileDataInput in = segments.next();
+ try
+ {
+ while (!in.isEOF())
+ {
+ ByteBuffer indexKey = ByteBufferUtil.readWithShortLength(in);
+ if (i % expectedIndexInterval == 0)
+ {
+ ByteBuffer summaryKey = ByteBuffer.wrap(indexSummary.getKey(i / expectedIndexInterval));
+ if (!summaryKey.equals(indexKey))
+ return false;
+ summaryEntriesChecked++;
+
+ if (summaryEntriesChecked == Downsampling.BASE_SAMPLING_LEVEL)
+ return true;
+ }
+ RowIndexEntry.Serializer.skip(in);
+ i++;
+ }
+ }
+ catch (IOException e)
+ {
+ markSuspect();
+ throw new CorruptSSTableException(e, in.getPath());
+ }
+ finally
+ {
+ FileUtils.closeQuietly(in);
+ }
+ }
+
+ return true;
+ }
+
+ /**
+ * Save index summary to Summary.db file.
+ *
+ * @param ibuilder
+ * @param dbuilder
+ */
+ public void saveSummary(SegmentedFile.Builder ibuilder, SegmentedFile.Builder dbuilder)
+ {
+ saveSummary(ibuilder, dbuilder, indexSummary);
+ }
+
+ private void saveSummary(SegmentedFile.Builder ibuilder, SegmentedFile.Builder dbuilder, IndexSummary summary)
+ {
+ File summariesFile = new File(descriptor.filenameFor(Component.SUMMARY));
+ if (summariesFile.exists())
+ FileUtils.deleteWithConfirm(summariesFile);
+
+ DataOutputStreamAndChannel oStream = null;
+ try
+ {
+ oStream = new DataOutputStreamAndChannel(new FileOutputStream(summariesFile));
+ IndexSummary.serializer.serialize(summary, oStream, descriptor.version.hasSamplingLevel);
+ ByteBufferUtil.writeWithLength(first.getKey(), oStream);
+ ByteBufferUtil.writeWithLength(last.getKey(), oStream);
+ ibuilder.serializeBounds(oStream);
+ dbuilder.serializeBounds(oStream);
+ }
+ catch (IOException e)
+ {
+ logger.debug("Cannot save SSTable Summary: ", e);
+
+ // corrupted hence delete it and let it load it now.
+ if (summariesFile.exists())
+ FileUtils.deleteWithConfirm(summariesFile);
+ }
+ finally
+ {
+ FileUtils.closeQuietly(oStream);
+ }
+ }
+
+ public void setReplacedBy(SSTableReader replacement)
+ {
+ synchronized (tidy.global)
+ {
+ assert replacement != null;
+ assert !tidy.isReplaced;
+ tidy.isReplaced = true;
+ }
+ }
+
+ public boolean isReplaced()
+ {
+ return tidy.isReplaced;
+ }
+
+ /**
+ * Clone this reader with the provided start and open reason, and set the clone as replacement.
+ *
+ * @param newFirst the first key for the replacement (which can be different from the original due to the pre-emptive
+ * opening of compaction results).
+ * @param reason the {@code OpenReason} for the replacement.
+ *
+ * @return the cloned reader. That reader is set as a replacement by the method.
+ */
+ private SSTableReader cloneAndReplace(DecoratedKey newFirst, OpenReason reason)
+ {
+ return cloneAndReplace(newFirst, reason, indexSummary.sharedCopy());
+ }
+
+ /**
+ * Clone this reader with the new values and set the clone as replacement.
+ *
+ * @param newFirst the first key for the replacement (which can be different from the original due to the pre-emptive
+ * opening of compaction results).
+ * @param reason the {@code OpenReason} for the replacement.
+ * @param newSummary the index summary for the replacement.
+ *
+ * @return the cloned reader. That reader is set as a replacement by the method.
+ */
+ private SSTableReader cloneAndReplace(DecoratedKey newFirst, OpenReason reason, IndexSummary newSummary)
+ {
+ SSTableReader replacement = internalOpen(descriptor,
+ components,
+ metadata,
+ partitioner,
- ifile.sharedCopy(),
++ ifile != null ? ifile.sharedCopy() : null,
+ dfile.sharedCopy(),
+ newSummary,
+ bf.sharedCopy(),
+ maxDataAge,
+ sstableMetadata,
+ reason);
+ replacement.first = newFirst;
+ replacement.last = last;
+ replacement.isSuspect.set(isSuspect.get());
+ setReplacedBy(replacement);
+ return replacement;
+ }
+
+ // runOnClose must NOT be an anonymous or non-static inner class, nor must it retain a reference chain to this reader
+ public SSTableReader cloneWithNewStart(DecoratedKey newStart, final Runnable runOnClose)
+ {
+ synchronized (tidy.global)
+ {
+ assert openReason != OpenReason.EARLY;
+ // TODO: merge with caller's firstKeyBeyond() work,to save time
+ if (newStart.compareTo(first) > 0)
+ {
+ final long dataStart = getPosition(newStart, Operator.EQ).position;
+ final long indexStart = getIndexScanPosition(newStart);
+ this.tidy.runOnClose = new DropPageCache(dfile, dataStart, ifile, indexStart, runOnClose);
+ }
+
+ return cloneAndReplace(newStart, OpenReason.MOVED_START);
+ }
+ }
+
+ // runOnClose must NOT be an anonymous or non-static inner class, nor must it retain a reference chain to this reader
+ public SSTableReader cloneAsShadowed(final Runnable runOnClose)
+ {
+ synchronized (tidy.global)
+ {
+ assert openReason != OpenReason.EARLY;
+ this.tidy.runOnClose = new DropPageCache(dfile, 0, ifile, 0, runOnClose);
+ return cloneAndReplace(first, OpenReason.SHADOWED);
+ }
+ }
+
+ private static class DropPageCache implements Runnable
+ {
+ final SegmentedFile dfile;
+ final long dfilePosition;
+ final SegmentedFile ifile;
+ final long ifilePosition;
+ final Runnable andThen;
+
+ private DropPageCache(SegmentedFile dfile, long dfilePosition, SegmentedFile ifile, long ifilePosition, Runnable andThen)
+ {
+ this.dfile = dfile;
+ this.dfilePosition = dfilePosition;
+ this.ifile = ifile;
+ this.ifilePosition = ifilePosition;
+ this.andThen = andThen;
+ }
+
+ public void run()
+ {
+ dfile.dropPageCache(dfilePosition);
- ifile.dropPageCache(ifilePosition);
++ if (ifile != null)
++ ifile.dropPageCache(ifilePosition);
+ andThen.run();
+ }
+ }
+
+ /**
+ * Returns a new SSTableReader with the same properties as this SSTableReader except that a new IndexSummary will
+ * be built at the target samplingLevel. This (original) SSTableReader instance will be marked as replaced, have
+ * its DeletingTask removed, and have its periodic read-meter sync task cancelled.
+ * @param samplingLevel the desired sampling level for the index summary on the new SSTableReader
+ * @return a new SSTableReader
+ * @throws IOException
+ */
+ public SSTableReader cloneWithNewSummarySamplingLevel(ColumnFamilyStore parent, int samplingLevel) throws IOException
+ {
+ assert descriptor.version.hasSamplingLevel;
+
+ synchronized (tidy.global)
+ {
+ assert openReason != OpenReason.EARLY;
+
+ int minIndexInterval = metadata.getMinIndexInterval();
+ int maxIndexInterval = metadata.getMaxIndexInterval();
+ double effectiveInterval = indexSummary.getEffectiveIndexInterval();
+
+ IndexSummary newSummary;
+ long oldSize = bytesOnDisk();
+
+ // We have to rebuild the summary from the on-disk primary index in three cases:
+ // 1. The sampling level went up, so we need to read more entries off disk
+ // 2. The min_index_interval changed (in either direction); this changes what entries would be in the summary
+ // at full sampling (and consequently at any other sampling level)
+ // 3. The max_index_interval was lowered, forcing us to raise the sampling level
+ if (samplingLevel > indexSummary.getSamplingLevel() || indexSummary.getMinIndexInterval() != minIndexInterval || effectiveInterval > maxIndexInterval)
+ {
+ newSummary = buildSummaryAtLevel(samplingLevel);
+ }
+ else if (samplingLevel < indexSummary.getSamplingLevel())
+ {
+ // we can use the existing index summary to make a smaller one
+ newSummary = IndexSummaryBuilder.downsample(indexSummary, samplingLevel, minIndexInterval, partitioner);
+
+ SegmentedFile.Builder ibuilder = SegmentedFile.getBuilder(DatabaseDescriptor.getIndexAccessMode());
+ SegmentedFile.Builder dbuilder = compression
+ ? SegmentedFile.getCompressedBuilder()
+ : SegmentedFile.getBuilder(DatabaseDescriptor.getDiskAccessMode());
+ saveSummary(ibuilder, dbuilder, newSummary);
+ }
+ else
+ {
+ throw new AssertionError("Attempted to clone SSTableReader with the same index summary sampling level and " +
+ "no adjustments to min/max_index_interval");
+ }
+
+ long newSize = bytesOnDisk();
+ StorageMetrics.load.inc(newSize - oldSize);
+ parent.metric.liveDiskSpaceUsed.inc(newSize - oldSize);
+
+ return cloneAndReplace(first, OpenReason.METADATA_CHANGE, newSummary);
+ }
+ }
+
+ private IndexSummary buildSummaryAtLevel(int newSamplingLevel) throws IOException
+ {
+ // we read the positions in a BRAF so we don't have to worry about an entry spanning a mmap boundary.
+ RandomAccessReader primaryIndex = RandomAccessReader.open(new File(descriptor.filenameFor(Component.PRIMARY_INDEX)));
try
{
long indexSize = primaryIndex.length();
@@@ -1566,16 -974,22 +1576,19 @@@
{
if (op == Operator.EQ && updateCacheAndStats)
bloomFilterTracker.addFalsePositive();
- // we matched the -1th position: if the operator might match forward, we'll start at the first
- // position. We however need to return the correct index entry for that first position.
- if (op.apply(1) >= 0)
- {
- sampledPosition = 0;
- }
- else
- {
- Tracing.trace("Partition summary allows skipping sstable {}", descriptor.generation);
- return null;
- }
+ Tracing.trace("Check against min and max keys allows skipping sstable {}", descriptor.generation);
+ return null;
}
+ int binarySearchResult = indexSummary.binarySearch(key);
+ long sampledPosition = getIndexScanPositionFromBinarySearchResult(binarySearchResult, indexSummary);
+ int sampledIndex = getIndexSummaryIndexFromBinarySearchResult(binarySearchResult);
+
+ int effectiveInterval = indexSummary.getEffectiveIndexIntervalAfterIndex(sampledIndex);
+
+ if (ifile == null)
+ return null;
+
// scan the on-disk index, starting at the nearest sampled position.
// The check against IndexInterval is to be exit the loop in the EQ case when the key looked for is not present
// (bloom filter false positive). But note that for non-EQ cases, we might need to check the first key of the
@@@ -1670,11 -1084,13 +1683,14 @@@
*/
public DecoratedKey firstKeyBeyond(RowPosition token)
{
+ if (token.compareTo(first) < 0)
+ return first;
+
long sampledPosition = getIndexScanPosition(token);
- if (sampledPosition == -1)
- sampledPosition = 0;
+ if (ifile == null)
+ return null;
+
Iterator<FileDataInput> segments = ifile.iterator(sampledPosition);
while (segments.hasNext())
{
@@@ -1948,8 -1391,7 +1964,10 @@@
{
try
{
- return SSTableMetadata.serializer.deserialize(descriptor).right;
+ CompactionMetadata compactionMetadata = (CompactionMetadata) descriptor.getMetadataSerializer().deserialize(descriptor, MetadataType.COMPACTION);
- return compactionMetadata.ancestors;
++ if (compactionMetadata != null)
++ return compactionMetadata.ancestors;
++ return Collections.emptySet();
}
catch (IOException e)
{
@@@ -1995,7 -1441,7 +2013,9 @@@
public RandomAccessReader openIndexReader()
{
- return ifile.createReader();
- return RandomAccessReader.open(new File(getIndexFilename()));
++ if (ifile != null)
++ return ifile.createReader();
++ return null;
}
/**
@@@ -2024,152 -1470,73 +2044,154 @@@
}
/**
- * @param sstables
- * @return true if all desired references were acquired. Otherwise, it will unreference any partial acquisition, and return false.
+ * Increment the total row read count and read rate for this SSTable. This should not be incremented for range
+ * slice queries, row cache hits, or non-query reads, like compaction.
*/
- public static boolean acquireReferences(Iterable<SSTableReader> sstables)
+ public void incrementReadCount()
{
- SSTableReader failed = null;
- for (SSTableReader sstable : sstables)
+ if (readMeter != null)
+ readMeter.mark();
+ }
+
+ public static class SizeComparator implements Comparator<SSTableReader>
+ {
+ public int compare(SSTableReader o1, SSTableReader o2)
{
- if (!sstable.acquireReference())
- {
- failed = sstable;
- break;
- }
+ return Longs.compare(o1.onDiskLength(), o2.onDiskLength());
}
+ }
- if (failed == null)
- return true;
+ public Ref<SSTableReader> tryRef()
+ {
+ return selfRef.tryRef();
+ }
- for (SSTableReader sstable : sstables)
- {
- if (sstable == failed)
- break;
- sstable.releaseReference();
- }
- return false;
+ public Ref<SSTableReader> selfRef()
+ {
+ return selfRef;
}
- public static void releaseReferences(Iterable<SSTableReader> sstables)
+ public Ref<SSTableReader> ref()
{
- for (SSTableReader sstable : sstables)
- {
- sstable.releaseReference();
- }
+ return selfRef.ref();
}
- private void dropPageCache()
+ void setup(boolean isOffline)
{
- dropPageCache(dfile.path);
- if (null != ifile)
- dropPageCache(ifile.path);
+ tidy.setup(this, isOffline);
+ this.readMeter = tidy.global.readMeter;
}
- private void dropPageCache(String filePath)
+ @VisibleForTesting
+ public void overrideReadMeter(RestorableMeter readMeter)
{
- RandomAccessFile file = null;
+ this.readMeter = tidy.global.readMeter = readMeter;
+ }
- try
+ /**
+ * One instance per SSTableReader we create. This references the type-shared tidy, which in turn references
+ * the globally shared tidy, i.e.
+ *
+ * InstanceTidier => DescriptorTypeTitdy => GlobalTidy
+ *
+ * We can create many InstanceTidiers (one for every time we reopen an sstable with MOVED_START for example), but there can only be
+ * two DescriptorTypeTidy (FINAL and TEMPLINK) and only one GlobalTidy for one single logical sstable.
+ *
+ * When the InstanceTidier cleansup, it releases its reference to its DescriptorTypeTidy; when all InstanceTidiers
+ * for that type have run, the DescriptorTypeTidy cleansup. DescriptorTypeTidy behaves in the same way towards GlobalTidy.
+ *
+ * For ease, we stash a direct reference to both our type-shared and global tidier
+ */
+ private static final class InstanceTidier implements Tidy
+ {
+ private final Descriptor descriptor;
+ private final CFMetaData metadata;
+ private IFilter bf;
+ private IndexSummary summary;
+
+ private SegmentedFile dfile;
+ private SegmentedFile ifile;
+ private Runnable runOnClose;
+ private boolean isReplaced = false;
+
+ // a reference to our shared per-Descriptor.Type tidy instance, that
+ // we will release when we are ourselves released
+ private Ref<DescriptorTypeTidy> typeRef;
+
+ // a convenience stashing of the shared per-descriptor-type tidy instance itself
+ // and the per-logical-sstable globally shared state that it is linked to
+ private DescriptorTypeTidy type;
+ private GlobalTidy global;
+
+ private boolean setup;
+
+ void setup(SSTableReader reader, boolean isOffline)
+ {
+ this.setup = true;
+ this.bf = reader.bf;
+ this.summary = reader.indexSummary;
+ this.dfile = reader.dfile;
+ this.ifile = reader.ifile;
+ // get a new reference to the shared descriptor-type tidy
+ this.typeRef = DescriptorTypeTidy.get(reader);
+ this.type = typeRef.get();
+ this.global = type.globalRef.get();
+ if (!isOffline)
+ global.ensureReadMeter();
+ }
+
+ InstanceTidier(Descriptor descriptor, CFMetaData metadata)
{
- file = new RandomAccessFile(filePath, "r");
+ this.descriptor = descriptor;
+ this.metadata = metadata;
+ }
- int fd = CLibrary.getfd(file.getFD());
+ public void tidy()
+ {
+ // don't try to cleanup if the sstablereader was never fully constructed
+ if (!setup)
+ return;
- if (fd > 0)
+ final ColumnFamilyStore cfs = Schema.instance.getColumnFamilyStoreInstance(metadata.cfId);
+ final OpOrder.Barrier barrier;
+ if (cfs != null)
{
- if (logger.isDebugEnabled())
- logger.debug(String.format("Dropping page cache of file %s.", filePath));
-
- CLibrary.trySkipCache(fd, 0, 0);
+ barrier = cfs.readOrdering.newBarrier();
+ barrier.issue();
}
+ else
+ barrier = null;
+
+ ScheduledExecutors.nonPeriodicTasks.execute(new Runnable()
+ {
+ public void run()
+ {
+ if (barrier != null)
+ barrier.await();
- bf.close();
++ if (bf != null)
++ bf.close();
+ if (summary != null)
+ summary.close();
+ if (runOnClose != null)
+ runOnClose.run();
+ dfile.close();
- ifile.close();
++ if (ifile != null)
++ ifile.close();
+ typeRef.release();
+ }
+ });
}
- catch (IOException e)
+
+ public String name()
{
- // we don't care if cache cleanup fails
+ return descriptor.toString();
}
- finally
+
+ void releaseSummary()
{
- FileUtils.closeQuietly(file);
+ summary.close();
+ assert summary.isCleanedUp();
+ summary = null;
}
}
http://git-wip-us.apache.org/repos/asf/cassandra/blob/4c94ef20/src/java/org/apache/cassandra/tools/StandaloneScrubber.java
----------------------------------------------------------------------
http://git-wip-us.apache.org/repos/asf/cassandra/blob/4c94ef20/test/unit/org/apache/cassandra/db/ScrubTest.java
----------------------------------------------------------------------
[03/12] cassandra git commit: Scrub (recover) sstables even when
-Index.db is missing
Posted by be...@apache.org.
Scrub (recover) sstables even when -Index.db is missing
patch by mck; reviewed by stefania for CASSANDRA-9591
Project: http://git-wip-us.apache.org/repos/asf/cassandra/repo
Commit: http://git-wip-us.apache.org/repos/asf/cassandra/commit/452d6a44
Tree: http://git-wip-us.apache.org/repos/asf/cassandra/tree/452d6a44
Diff: http://git-wip-us.apache.org/repos/asf/cassandra/diff/452d6a44
Branch: refs/heads/cassandra-2.2
Commit: 452d6a445a6935d3e7d0e0fdf59e87d2a2ff95e7
Parents: 62714a9
Author: Stefania Alborghetti <st...@datastax.com>
Authored: Mon Jun 15 16:49:03 2015 +0800
Committer: Benedict Elliott Smith <be...@apache.org>
Committed: Tue Jul 7 14:27:25 2015 +0100
----------------------------------------------------------------------
CHANGES.txt | 4 ++
.../cassandra/db/compaction/Scrubber.java | 36 +++++++++++----
.../cassandra/io/sstable/SSTableReader.java | 47 ++++++++++++++++----
.../cassandra/tools/StandaloneScrubber.java | 2 +-
.../unit/org/apache/cassandra/db/ScrubTest.java | 25 +++++++++++
5 files changed, 96 insertions(+), 18 deletions(-)
----------------------------------------------------------------------
http://git-wip-us.apache.org/repos/asf/cassandra/blob/452d6a44/CHANGES.txt
----------------------------------------------------------------------
diff --git a/CHANGES.txt b/CHANGES.txt
index ca4d4b5..bd1db92 100644
--- a/CHANGES.txt
+++ b/CHANGES.txt
@@ -1,3 +1,7 @@
+2.0.18
+* Scrub (recover) sstables even when -Index.db is missing, (CASSANDRA-9591)
+
+
2.0.17
* Avoid NPE in AuthSuccess#decode (CASSANDRA-9727)
* Add listen_address to system.local (CASSANDRA-9603)
http://git-wip-us.apache.org/repos/asf/cassandra/blob/452d6a44/src/java/org/apache/cassandra/db/compaction/Scrubber.java
----------------------------------------------------------------------
diff --git a/src/java/org/apache/cassandra/db/compaction/Scrubber.java b/src/java/org/apache/cassandra/db/compaction/Scrubber.java
index ea10855..dc60efa 100644
--- a/src/java/org/apache/cassandra/db/compaction/Scrubber.java
+++ b/src/java/org/apache/cassandra/db/compaction/Scrubber.java
@@ -96,7 +96,16 @@ public class Scrubber implements Closeable
? new ScrubController(cfs)
: new CompactionController(cfs, Collections.singleton(sstable), CompactionManager.getDefaultGcBefore(cfs));
this.isCommutative = cfs.metadata.getDefaultValidator().isCommutative();
- this.expectedBloomFilterSize = Math.max(cfs.metadata.getIndexInterval(), (int)(SSTableReader.getApproximateKeyCount(toScrub,cfs.metadata)));
+
+ boolean hasIndexFile = (new File(sstable.descriptor.filenameFor(Component.PRIMARY_INDEX))).exists();
+ if (!hasIndexFile)
+ {
+ // if there's any corruption in the -Data.db then rows can't be skipped over. but it's worth a shot.
+ outputHandler.warn("Missing component: " + sstable.descriptor.filenameFor(Component.PRIMARY_INDEX));
+ }
+
+ this.expectedBloomFilterSize = Math.max(cfs.metadata.getIndexInterval(),
+ hasIndexFile ? (int)(SSTableReader.getApproximateKeyCount(toScrub,cfs.metadata)) : 0);
// loop through each row, deserializing to check for damage.
// we'll also loop through the index at the same time, using the position from the index to recover if the
@@ -105,7 +114,11 @@ public class Scrubber implements Closeable
this.dataFile = isOffline
? sstable.openDataReader()
: sstable.openDataReader(CompactionManager.instance.getRateLimiter());
- this.indexFile = RandomAccessReader.open(new File(sstable.descriptor.filenameFor(Component.PRIMARY_INDEX)));
+
+ this.indexFile = hasIndexFile
+ ? RandomAccessReader.open(new File(sstable.descriptor.filenameFor(Component.PRIMARY_INDEX)))
+ : null;
+
this.scrubInfo = new ScrubInfo(dataFile, sstable);
this.currentRowPositionFromIndex = 0;
@@ -117,7 +130,8 @@ public class Scrubber implements Closeable
outputHandler.output(String.format("Scrubbing %s (%s bytes)", sstable, dataFile.length()));
try
{
- nextIndexKey = ByteBufferUtil.readWithShortLength(indexFile);
+ nextIndexKey = indexAvailable() ? ByteBufferUtil.readWithShortLength(indexFile) : null;
+ if (indexAvailable())
{
// throw away variable so we don't have a side effect in the assert
long firstRowPositionFromIndex = RowIndexEntry.serializer.deserialize(indexFile, sstable.descriptor.version).position;
@@ -181,7 +195,7 @@ public class Scrubber implements Closeable
outputHandler.debug(String.format("Index doublecheck: row %s is %s bytes", ByteBufferUtil.bytesToHex(currentIndexKey), dataSizeFromIndex));
}
- assert currentIndexKey != null || indexFile.isEOF();
+ assert currentIndexKey != null || !indexAvailable();
writer.mark();
try
@@ -198,10 +212,10 @@ public class Scrubber implements Closeable
if (dataSize > dataFile.length())
throw new IOError(new IOException("Impossible row size (greater than file length): " + dataSize));
- if (dataStart != dataStartFromIndex)
+ if (indexFile != null && dataStart != dataStartFromIndex)
outputHandler.warn(String.format("Data file row position %d differs from index file row position %d", dataStart, dataStartFromIndex));
- if (dataSize != dataSizeFromIndex)
+ if (indexFile != null && dataSize != dataSizeFromIndex)
outputHandler.warn(String.format("Data file row size %d differs from index file row size %d", dataSize, dataSizeFromIndex));
SSTableIdentityIterator atoms = new SSTableIdentityIterator(sstable, dataFile, key, dataSize, validateColumns);
@@ -317,8 +331,9 @@ public class Scrubber implements Closeable
currentRowPositionFromIndex = nextRowPositionFromIndex;
try
{
- nextIndexKey = indexFile.isEOF() ? null : ByteBufferUtil.readWithShortLength(indexFile);
- nextRowPositionFromIndex = indexFile.isEOF()
+ nextIndexKey = !indexAvailable() ? null : ByteBufferUtil.readWithShortLength(indexFile);
+
+ nextRowPositionFromIndex = !indexAvailable()
? dataFile.length()
: RowIndexEntry.serializer.deserialize(indexFile, sstable.descriptor.version).position;
}
@@ -330,6 +345,11 @@ public class Scrubber implements Closeable
}
}
+ private boolean indexAvailable()
+ {
+ return indexFile != null && !indexFile.isEOF();
+ }
+
private void seekToNextRow()
{
while(nextRowPositionFromIndex < dataFile.length())
http://git-wip-us.apache.org/repos/asf/cassandra/blob/452d6a44/src/java/org/apache/cassandra/io/sstable/SSTableReader.java
----------------------------------------------------------------------
diff --git a/src/java/org/apache/cassandra/io/sstable/SSTableReader.java b/src/java/org/apache/cassandra/io/sstable/SSTableReader.java
index 15808e8..8919a09 100644
--- a/src/java/org/apache/cassandra/io/sstable/SSTableReader.java
+++ b/src/java/org/apache/cassandra/io/sstable/SSTableReader.java
@@ -157,7 +157,7 @@ public class SSTableReader extends SSTable implements Closeable
public static SSTableReader openForBatch(Descriptor descriptor, Set<Component> components, CFMetaData metadata, IPartitioner partitioner) throws IOException
{
- SSTableMetadata sstableMetadata = openMetadata(descriptor, components, partitioner);
+ SSTableMetadata sstableMetadata = openMetadata(descriptor, components, partitioner, true);
SSTableReader sstable = new SSTableReader(descriptor,
components,
metadata,
@@ -191,7 +191,7 @@ public class SSTableReader extends SSTable implements Closeable
boolean validate) throws IOException
{
long start = System.nanoTime();
- SSTableMetadata sstableMetadata = openMetadata(descriptor, components, partitioner);
+ SSTableMetadata sstableMetadata = openMetadata(descriptor, components, partitioner, validate);
SSTableReader sstable = new SSTableReader(descriptor,
components,
@@ -213,12 +213,17 @@ public class SSTableReader extends SSTable implements Closeable
return sstable;
}
- private static SSTableMetadata openMetadata(Descriptor descriptor, Set<Component> components, IPartitioner partitioner) throws IOException
+ private static SSTableMetadata openMetadata(Descriptor descriptor,
+ Set<Component> components,
+ IPartitioner partitioner,
+ boolean primaryIndexRequired) throws IOException
{
assert partitioner != null;
// Minimum components without which we can't do anything
assert components.contains(Component.DATA) : "Data component is missing for sstable" + descriptor;
- assert components.contains(Component.PRIMARY_INDEX) : "Primary index component is missing for sstable " + descriptor;
+
+ assert !primaryIndexRequired || components.contains(Component.PRIMARY_INDEX)
+ : "Primary index component is missing for sstable " + descriptor;
logger.info("Opening {} ({} bytes)", descriptor, new File(descriptor.filenameFor(COMPONENT_DATA)).length());
@@ -382,11 +387,17 @@ public class SSTableReader extends SSTable implements Closeable
readMeterSyncFuture.cancel(false);
// Force finalizing mmapping if necessary
- ifile.cleanup();
+
+ if (null != ifile)
+ ifile.cleanup();
+
dfile.cleanup();
// close the BF so it can be opened later.
- bf.close();
- indexSummary.close();
+ if (null != bf)
+ bf.close();
+
+ if (null != indexSummary)
+ indexSummary.close();
}
public void setTrackedBy(DataTracker tracker)
@@ -406,6 +417,12 @@ public class SSTableReader extends SSTable implements Closeable
load(false, true);
bf = FilterFactory.AlwaysPresent;
}
+ else if (!components.contains(Component.PRIMARY_INDEX))
+ {
+ // avoid any reading of the missing primary index component.
+ // this should only happen during StandaloneScrubber
+ load(false, false);
+ }
else if (!components.contains(Component.FILTER))
{
// bf is enabled, but filter component is missing.
@@ -454,7 +471,9 @@ public class SSTableReader extends SSTable implements Closeable
if (recreateBloomFilter || !summaryLoaded)
buildSummary(recreateBloomFilter, ibuilder, dbuilder, summaryLoaded);
- ifile = ibuilder.complete(descriptor.filenameFor(Component.PRIMARY_INDEX));
+ if (components.contains(Component.PRIMARY_INDEX))
+ ifile = ibuilder.complete(descriptor.filenameFor(Component.PRIMARY_INDEX));
+
dfile = dbuilder.complete(descriptor.filenameFor(Component.DATA));
if (saveSummaryIfCreated && (recreateBloomFilter || !summaryLoaded)) // save summary information to disk
saveSummary(this, ibuilder, dbuilder);
@@ -462,6 +481,9 @@ public class SSTableReader extends SSTable implements Closeable
private void buildSummary(boolean recreateBloomFilter, SegmentedFile.Builder ibuilder, SegmentedFile.Builder dbuilder, boolean summaryLoaded) throws IOException
{
+ if (!components.contains(Component.PRIMARY_INDEX))
+ return;
+
// we read the positions in a BRAF so we don't have to worry about an entry spanning a mmap boundary.
RandomAccessReader primaryIndex = RandomAccessReader.open(new File(descriptor.filenameFor(Component.PRIMARY_INDEX)));
@@ -965,6 +987,9 @@ public class SSTableReader extends SSTable implements Closeable
}
}
+ if (ifile == null)
+ return null;
+
// scan the on-disk index, starting at the nearest sampled position.
// The check against IndexInterval is to be exit the loop in the EQ case when the key looked for is not present
// (bloom filter false positive). But note that for non-EQ cases, we might need to check the first key of the
@@ -1063,6 +1088,9 @@ public class SSTableReader extends SSTable implements Closeable
if (sampledPosition == -1)
sampledPosition = 0;
+ if (ifile == null)
+ return null;
+
Iterator<FileDataInput> segments = ifile.iterator(sampledPosition);
while (segments.hasNext())
{
@@ -1480,7 +1508,8 @@ public class SSTableReader extends SSTable implements Closeable
private void dropPageCache()
{
dropPageCache(dfile.path);
- dropPageCache(ifile.path);
+ if (null != ifile)
+ dropPageCache(ifile.path);
}
private void dropPageCache(String filePath)
http://git-wip-us.apache.org/repos/asf/cassandra/blob/452d6a44/src/java/org/apache/cassandra/tools/StandaloneScrubber.java
----------------------------------------------------------------------
diff --git a/src/java/org/apache/cassandra/tools/StandaloneScrubber.java b/src/java/org/apache/cassandra/tools/StandaloneScrubber.java
index 26768f3..8ace228 100644
--- a/src/java/org/apache/cassandra/tools/StandaloneScrubber.java
+++ b/src/java/org/apache/cassandra/tools/StandaloneScrubber.java
@@ -79,7 +79,7 @@ public class StandaloneScrubber
for (Map.Entry<Descriptor, Set<Component>> entry : lister.list().entrySet())
{
Set<Component> components = entry.getValue();
- if (!components.contains(Component.DATA) || !components.contains(Component.PRIMARY_INDEX))
+ if (!components.contains(Component.DATA))
continue;
try
http://git-wip-us.apache.org/repos/asf/cassandra/blob/452d6a44/test/unit/org/apache/cassandra/db/ScrubTest.java
----------------------------------------------------------------------
diff --git a/test/unit/org/apache/cassandra/db/ScrubTest.java b/test/unit/org/apache/cassandra/db/ScrubTest.java
index 94c7e34..978b3e8 100644
--- a/test/unit/org/apache/cassandra/db/ScrubTest.java
+++ b/test/unit/org/apache/cassandra/db/ScrubTest.java
@@ -270,6 +270,31 @@ public class ScrubTest extends SchemaLoader
}
@Test
+ public void testScrubNoIndex() throws IOException, ExecutionException, InterruptedException, ConfigurationException
+ {
+ CompactionManager.instance.disableAutoCompaction();
+ Keyspace keyspace = Keyspace.open(KEYSPACE);
+ ColumnFamilyStore cfs = keyspace.getColumnFamilyStore(CF);
+ cfs.clearUnsafe();
+
+ List<Row> rows;
+
+ // insert data and verify we get it back w/ range query
+ fillCF(cfs, 10);
+ rows = cfs.getRangeSlice(Util.range("", ""), null, new IdentityQueryFilter(), 1000);
+ assertEquals(10, rows.size());
+
+ for (SSTableReader sstable : cfs.getSSTables())
+ new File(sstable.descriptor.filenameFor(Component.PRIMARY_INDEX)).delete();
+
+ CompactionManager.instance.performScrub(cfs, false, true);
+
+ // check data is still there
+ rows = cfs.getRangeSlice(Util.range("", ""), null, new IdentityQueryFilter(), 1000);
+ assertEquals(10, rows.size());
+ }
+
+ @Test
public void testScrubOutOfOrder() throws Exception
{
CompactionManager.instance.disableAutoCompaction();
[09/12] cassandra git commit: Merge branch 'cassandra-2.1' into
cassandra-2.2
Posted by be...@apache.org.
Merge branch 'cassandra-2.1' into cassandra-2.2
Conflicts:
CHANGES.txt
src/java/org/apache/cassandra/db/compaction/Scrubber.java
src/java/org/apache/cassandra/io/sstable/format/SSTableReader.java
Project: http://git-wip-us.apache.org/repos/asf/cassandra/repo
Commit: http://git-wip-us.apache.org/repos/asf/cassandra/commit/ebe18bb2
Tree: http://git-wip-us.apache.org/repos/asf/cassandra/tree/ebe18bb2
Diff: http://git-wip-us.apache.org/repos/asf/cassandra/diff/ebe18bb2
Branch: refs/heads/cassandra-2.2
Commit: ebe18bb2ff62602fd5f55b969ecf665d2d3e5ace
Parents: 7d31068 4c94ef2
Author: Benedict Elliott Smith <be...@apache.org>
Authored: Tue Jul 7 16:28:19 2015 +0100
Committer: Benedict Elliott Smith <be...@apache.org>
Committed: Tue Jul 7 16:28:19 2015 +0100
----------------------------------------------------------------------
CHANGES.txt | 5 ++
.../cassandra/db/compaction/Scrubber.java | 38 +++++++++++----
.../io/sstable/format/SSTableReader.java | 50 ++++++++++++++------
.../cassandra/tools/StandaloneScrubber.java | 2 +-
.../unit/org/apache/cassandra/db/ScrubTest.java | 25 ++++++++++
5 files changed, 95 insertions(+), 25 deletions(-)
----------------------------------------------------------------------
http://git-wip-us.apache.org/repos/asf/cassandra/blob/ebe18bb2/CHANGES.txt
----------------------------------------------------------------------
diff --cc CHANGES.txt
index 7c6b4ad,2cbc7c4..a863ad8
--- a/CHANGES.txt
+++ b/CHANGES.txt
@@@ -1,28 -1,12 +1,33 @@@
-2.1.9
++2.2.0-rc3
+ Merged from 2.0:
- * Scrub (recover) sstables even when -Index.db is missing, (CASSANDRA-9591)
++ * Scrub (recover) sstables even when -Index.db is missing (CASSANDRA-9591)
+
+
-2.1.8
+2.2.0-rc2
+ * Re-enable memory-mapped I/O on Windows (CASSANDRA-9658)
+ * Warn when an extra-large partition is compacted (CASSANDRA-9643)
+ * (cqlsh) Allow setting the initial connection timeout (CASSANDRA-9601)
+ * BulkLoader has --transport-factory option but does not use it (CASSANDRA-9675)
+ * Allow JMX over SSL directly from nodetool (CASSANDRA-9090)
+ * Update cqlsh for UDFs (CASSANDRA-7556)
+ * Change Windows kernel default timer resolution (CASSANDRA-9634)
+ * Deprected sstable2json and json2sstable (CASSANDRA-9618)
+ * Allow native functions in user-defined aggregates (CASSANDRA-9542)
+ * Don't repair system_distributed by default (CASSANDRA-9621)
+ * Fix mixing min, max, and count aggregates for blob type (CASSANRA-9622)
+ * Rename class for DATE type in Java driver (CASSANDRA-9563)
+ * Duplicate compilation of UDFs on coordinator (CASSANDRA-9475)
+ * Fix connection leak in CqlRecordWriter (CASSANDRA-9576)
+ * Mlockall before opening system sstables & remove boot_without_jna option (CASSANDRA-9573)
+ * Add functions to convert timeuuid to date or time, deprecate dateOf and unixTimestampOf (CASSANDRA-9229)
+ * Make sure we cancel non-compacting sstables from LifecycleTransaction (CASSANDRA-9566)
+ * Fix deprecated repair JMX API (CASSANDRA-9570)
+ * Add logback metrics (CASSANDRA-9378)
+ * Update and refactor ant test/test-compression to run the tests in parallel (CASSANDRA-9583)
+ * Fix upgrading to new directory for secondary index (CASSANDRA-9687)
+Merged from 2.1:
* (cqlsh) Fix bad check for CQL compatibility when DESCRIBE'ing
COMPACT STORAGE tables with no clustering columns
- * Warn when an extra-large partition is compacted (CASSANDRA-9643)
* Eliminate strong self-reference chains in sstable ref tidiers (CASSANDRA-9656)
* Ensure StreamSession uses canonical sstable reader instances (CASSANDRA-9700)
* Ensure memtable book keeping is not corrupted in the event we shrink usage (CASSANDRA-9681)
http://git-wip-us.apache.org/repos/asf/cassandra/blob/ebe18bb2/src/java/org/apache/cassandra/db/compaction/Scrubber.java
----------------------------------------------------------------------
diff --cc src/java/org/apache/cassandra/db/compaction/Scrubber.java
index 10952e7,b1c12e0..5a0b354
--- a/src/java/org/apache/cassandra/db/compaction/Scrubber.java
+++ b/src/java/org/apache/cassandra/db/compaction/Scrubber.java
@@@ -109,9 -101,17 +109,18 @@@ public class Scrubber implements Closea
? new ScrubController(cfs)
: new CompactionController(cfs, Collections.singleton(sstable), CompactionManager.getDefaultGcBefore(cfs));
this.isCommutative = cfs.metadata.isCounter();
+
+ boolean hasIndexFile = (new File(sstable.descriptor.filenameFor(Component.PRIMARY_INDEX))).exists();
+ this.isIndex = cfs.isIndex();
+ if (!hasIndexFile)
+ {
+ // if there's any corruption in the -Data.db then rows can't be skipped over. but it's worth a shot.
+ outputHandler.warn("Missing component: " + sstable.descriptor.filenameFor(Component.PRIMARY_INDEX));
+ }
-
+ this.checkData = checkData && !this.isIndex; //LocalByPartitionerType does not support validation
- this.expectedBloomFilterSize = Math.max(cfs.metadata.getMinIndexInterval(), (int)(SSTableReader.getApproximateKeyCount(toScrub)));
+ this.expectedBloomFilterSize = Math.max(
+ cfs.metadata.getMinIndexInterval(),
+ hasIndexFile ? (int)(SSTableReader.getApproximateKeyCount(toScrub)) : 0);
// loop through each row, deserializing to check for damage.
// we'll also loop through the index at the same time, using the position from the index to recover if the
@@@ -130,12 -134,15 +143,13 @@@
public void scrub()
{
outputHandler.output(String.format("Scrubbing %s (%s bytes)", sstable, dataFile.length()));
- try (SSTableRewriter writer = new SSTableRewriter(cfs, transaction, sstable.maxDataAge, isOffline);)
- Set<SSTableReader> oldSSTable = Sets.newHashSet(sstable);
- SSTableRewriter writer = new SSTableRewriter(cfs, oldSSTable, sstable.maxDataAge, isOffline);
- try
++ try (SSTableRewriter writer = new SSTableRewriter(cfs, transaction, sstable.maxDataAge, isOffline))
{
- nextIndexKey = ByteBufferUtil.readWithShortLength(indexFile);
+ nextIndexKey = indexAvailable() ? ByteBufferUtil.readWithShortLength(indexFile) : null;
+ if (indexAvailable())
{
// throw away variable so we don't have a side effect in the assert
- long firstRowPositionFromIndex = sstable.metadata.comparator.rowIndexEntrySerializer().deserialize(indexFile, sstable.descriptor.version).position;
+ long firstRowPositionFromIndex = rowIndexEntrySerializer.deserialize(indexFile, sstable.descriptor.version).position;
assert firstRowPositionFromIndex == 0 : firstRowPositionFromIndex;
}
@@@ -174,11 -182,12 +188,11 @@@
dataSizeFromIndex = nextRowPositionFromIndex - dataStartFromIndex;
}
- dataSize = dataSizeFromIndex;
// avoid an NPE if key is null
String keyName = key == null ? "(unreadable key)" : ByteBufferUtil.bytesToHex(key.getKey());
- outputHandler.debug(String.format("row %s is %s bytes", keyName, dataSize));
+ outputHandler.debug(String.format("row %s is %s bytes", keyName, dataSizeFromIndex));
- assert currentIndexKey != null || indexFile.isEOF();
+ assert currentIndexKey != null || !indexAvailable();
try
{
@@@ -191,14 -200,16 +205,14 @@@
ByteBufferUtil.bytesToHex(key.getKey()), ByteBufferUtil.bytesToHex(currentIndexKey))));
}
- if (dataSizeFromIndex > dataFile.length())
- if (dataSize > dataFile.length())
- throw new IOError(new IOException("Impossible row size (greater than file length): " + dataSize));
++ if (indexFile != null && dataSizeFromIndex > dataFile.length())
+ throw new IOError(new IOException("Impossible row size (greater than file length): " + dataSizeFromIndex));
- if (dataStart != dataStartFromIndex)
+ if (indexFile != null && dataStart != dataStartFromIndex)
outputHandler.warn(String.format("Data file row position %d differs from index file row position %d", dataStart, dataStartFromIndex));
- if (indexFile != null && dataSize != dataSizeFromIndex)
- outputHandler.warn(String.format("Data file row size %d different from index file row size %d", dataSize, dataSizeFromIndex));
+ SSTableIdentityIterator atoms = new SSTableIdentityIterator(sstable, dataFile, key, checkData);
- SSTableIdentityIterator atoms = new SSTableIdentityIterator(sstable, dataFile, key, dataSize, validateColumns);
if (prevKey != null && prevKey.compareTo(key) > 0)
{
saveOutOfOrderRow(prevKey, key, atoms);
@@@ -316,10 -327,11 +330,11 @@@
currentRowPositionFromIndex = nextRowPositionFromIndex;
try
{
- nextIndexKey = indexFile.isEOF() ? null : ByteBufferUtil.readWithShortLength(indexFile);
- nextRowPositionFromIndex = indexFile.isEOF()
+ nextIndexKey = !indexAvailable() ? null : ByteBufferUtil.readWithShortLength(indexFile);
+
+ nextRowPositionFromIndex = !indexAvailable()
? dataFile.length()
- : sstable.metadata.comparator.rowIndexEntrySerializer().deserialize(indexFile, sstable.descriptor.version).position;
+ : rowIndexEntrySerializer.deserialize(indexFile, sstable.descriptor.version).position;
}
catch (Throwable th)
{
[12/12] cassandra git commit: Merge branch 'cassandra-2.2' into trunk
Posted by be...@apache.org.
Merge branch 'cassandra-2.2' into trunk
Conflicts:
CHANGES.txt
src/java/org/apache/cassandra/db/compaction/Scrubber.java
src/java/org/apache/cassandra/io/sstable/format/SSTableReader.java
Project: http://git-wip-us.apache.org/repos/asf/cassandra/repo
Commit: http://git-wip-us.apache.org/repos/asf/cassandra/commit/a8bb75a7
Tree: http://git-wip-us.apache.org/repos/asf/cassandra/tree/a8bb75a7
Diff: http://git-wip-us.apache.org/repos/asf/cassandra/diff/a8bb75a7
Branch: refs/heads/trunk
Commit: a8bb75a7e1a09ca05ceb8d566f2c9a88d3122c27
Parents: db68d1c ebe18bb
Author: Benedict Elliott Smith <be...@apache.org>
Authored: Tue Jul 7 16:45:02 2015 +0100
Committer: Benedict Elliott Smith <be...@apache.org>
Committed: Tue Jul 7 16:45:02 2015 +0100
----------------------------------------------------------------------
CHANGES.txt | 3 ++
.../db/compaction/CompactionManager.java | 15 ++++--
.../cassandra/db/compaction/Scrubber.java | 42 ++++++++++++-----
.../io/sstable/format/SSTableReader.java | 48 ++++++++++++++------
.../io/sstable/format/big/BigTableReader.java | 3 ++
.../cassandra/tools/StandaloneScrubber.java | 7 +--
.../unit/org/apache/cassandra/db/ScrubTest.java | 22 +++++++++
7 files changed, 106 insertions(+), 34 deletions(-)
----------------------------------------------------------------------
http://git-wip-us.apache.org/repos/asf/cassandra/blob/a8bb75a7/CHANGES.txt
----------------------------------------------------------------------
diff --cc CHANGES.txt
index 53beb26,a863ad8..9dee57d
--- a/CHANGES.txt
+++ b/CHANGES.txt
@@@ -1,19 -1,7 +1,22 @@@
-2.2.0-rc3
+3.0
+ * Storage engine refactor (CASSANDRA-8099, 9743)
+ * Update Guava to 18.0 (CASSANDRA-9653)
+ * Bloom filter false positive ratio is not honoured (CASSANDRA-8413)
+ * New option for cassandra-stress to leave a ratio of columns null (CASSANDRA-9522)
+ * Change hinted_handoff_enabled yaml setting, JMX (CASSANDRA-9035)
+ * Add algorithmic token allocation (CASSANDRA-7032)
+ * Add nodetool command to replay batchlog (CASSANDRA-9547)
+ * Make file buffer cache independent of paths being read (CASSANDRA-8897)
+ * Remove deprecated legacy Hadoop code (CASSANDRA-9353)
+ * Decommissioned nodes will not rejoin the cluster (CASSANDRA-8801)
+ * Change gossip stabilization to use endpoit size (CASSANDRA-9401)
+ * Change default garbage collector to G1 (CASSANDRA-7486)
+ * Populate TokenMetadata early during startup (CASSANDRA-9317)
+ * undeprecate cache recentHitRate (CASSANDRA-6591)
+ * Add support for selectively varint encoding fields (CASSANDRA-9499)
+ Merged from 2.0:
+ * Scrub (recover) sstables even when -Index.db is missing (CASSANDRA-9591)
+
2.2.0-rc2
* Re-enable memory-mapped I/O on Windows (CASSANDRA-9658)
http://git-wip-us.apache.org/repos/asf/cassandra/blob/a8bb75a7/src/java/org/apache/cassandra/db/compaction/CompactionManager.java
----------------------------------------------------------------------
diff --cc src/java/org/apache/cassandra/db/compaction/CompactionManager.java
index a6c3d8c,4c94fa0..e3e9b03
--- a/src/java/org/apache/cassandra/db/compaction/CompactionManager.java
+++ b/src/java/org/apache/cassandra/db/compaction/CompactionManager.java
@@@ -331,7 -332,7 +331,14 @@@ public class CompactionManager implemen
}
}
-- public AllSSTableOpStatus performScrub(final ColumnFamilyStore cfs, final boolean skipCorrupted, final boolean checkData) throws InterruptedException, ExecutionException
++ public AllSSTableOpStatus performScrub(final ColumnFamilyStore cfs, final boolean skipCorrupted, final boolean checkData)
++ throws InterruptedException, ExecutionException
++ {
++ return performScrub(cfs, skipCorrupted, checkData, false);
++ }
++
++ public AllSSTableOpStatus performScrub(final ColumnFamilyStore cfs, final boolean skipCorrupted, final boolean checkData, final boolean offline)
++ throws InterruptedException, ExecutionException
{
return parallelAllSSTableOperation(cfs, new OneSSTableOperation()
{
@@@ -344,7 -345,7 +351,7 @@@
@Override
public void execute(LifecycleTransaction input) throws IOException
{
-- scrubOne(cfs, input, skipCorrupted, checkData);
++ scrubOne(cfs, input, skipCorrupted, checkData, offline);
}
}, OperationType.SCRUB);
}
@@@ -691,11 -691,11 +698,11 @@@
}
}
-- private void scrubOne(ColumnFamilyStore cfs, LifecycleTransaction modifier, boolean skipCorrupted, boolean checkData) throws IOException
++ private void scrubOne(ColumnFamilyStore cfs, LifecycleTransaction modifier, boolean skipCorrupted, boolean checkData, boolean offline) throws IOException
{
CompactionInfo.Holder scrubInfo = null;
-- try (Scrubber scrubber = new Scrubber(cfs, modifier, skipCorrupted, false, checkData))
++ try (Scrubber scrubber = new Scrubber(cfs, modifier, skipCorrupted, offline, checkData))
{
scrubInfo = scrubber.getScrubInfo();
metrics.beginCompaction(scrubInfo);
http://git-wip-us.apache.org/repos/asf/cassandra/blob/a8bb75a7/src/java/org/apache/cassandra/db/compaction/Scrubber.java
----------------------------------------------------------------------
diff --cc src/java/org/apache/cassandra/db/compaction/Scrubber.java
index 562d681,5a0b354..94f3af7
--- a/src/java/org/apache/cassandra/db/compaction/Scrubber.java
+++ b/src/java/org/apache/cassandra/db/compaction/Scrubber.java
@@@ -108,10 -104,23 +108,19 @@@ public class Scrubber implements Closea
if (destination == null)
throw new IOException("disk full");
- // If we run scrub offline, we should never purge tombstone, as we cannot know if other sstable have data that the tombstone deletes.
- this.controller = isOffline
- ? new ScrubController(cfs)
- : new CompactionController(cfs, Collections.singleton(sstable), CompactionManager.getDefaultGcBefore(cfs));
this.isCommutative = cfs.metadata.isCounter();
+
+ boolean hasIndexFile = (new File(sstable.descriptor.filenameFor(Component.PRIMARY_INDEX))).exists();
this.isIndex = cfs.isIndex();
+ if (!hasIndexFile)
+ {
+ // if there's any corruption in the -Data.db then rows can't be skipped over. but it's worth a shot.
+ outputHandler.warn("Missing component: " + sstable.descriptor.filenameFor(Component.PRIMARY_INDEX));
+ }
this.checkData = checkData && !this.isIndex; //LocalByPartitionerType does not support validation
- this.expectedBloomFilterSize = Math.max(cfs.metadata.getMinIndexInterval(), (int)(SSTableReader.getApproximateKeyCount(toScrub)));
+ this.expectedBloomFilterSize = Math.max(
+ cfs.metadata.getMinIndexInterval(),
+ hasIndexFile ? (int)(SSTableReader.getApproximateKeyCount(toScrub)) : 0);
// loop through each row, deserializing to check for damage.
// we'll also loop through the index at the same time, using the position from the index to recover if the
@@@ -135,13 -143,13 +148,14 @@@
public void scrub()
{
outputHandler.output(String.format("Scrubbing %s (%s bytes)", sstable, dataFile.length()));
+ int nowInSec = FBUtilities.nowInSeconds();
- try (SSTableRewriter writer = new SSTableRewriter(cfs, transaction, sstable.maxDataAge, isOffline);)
+ try (SSTableRewriter writer = new SSTableRewriter(cfs, transaction, sstable.maxDataAge, isOffline))
{
- nextIndexKey = ByteBufferUtil.readWithShortLength(indexFile);
+ nextIndexKey = indexAvailable() ? ByteBufferUtil.readWithShortLength(indexFile) : null;
+ if (indexAvailable())
{
// throw away variable so we don't have a side effect in the assert
- long firstRowPositionFromIndex = rowIndexEntrySerializer.deserialize(indexFile, sstable.descriptor.version).position;
+ long firstRowPositionFromIndex = rowIndexEntrySerializer.deserialize(indexFile).position;
assert firstRowPositionFromIndex == 0 : firstRowPositionFromIndex;
}
@@@ -194,29 -202,29 +208,29 @@@
if (currentIndexKey != null && !key.getKey().equals(currentIndexKey))
{
throw new IOError(new IOException(String.format("Key from data file (%s) does not match key from index file (%s)",
- ByteBufferUtil.bytesToHex(key.getKey()), ByteBufferUtil.bytesToHex(currentIndexKey))));
+ //ByteBufferUtil.bytesToHex(key.getKey()), ByteBufferUtil.bytesToHex(currentIndexKey))));
+ "_too big_", ByteBufferUtil.bytesToHex(currentIndexKey))));
}
- if (dataSizeFromIndex > dataFile.length())
+ if (indexFile != null && dataSizeFromIndex > dataFile.length())
throw new IOError(new IOException("Impossible row size (greater than file length): " + dataSizeFromIndex));
- if (dataStart != dataStartFromIndex)
+ if (indexFile != null && dataStart != dataStartFromIndex)
outputHandler.warn(String.format("Data file row position %d differs from index file row position %d", dataStart, dataStartFromIndex));
- SSTableIdentityIterator atoms = new SSTableIdentityIterator(sstable, dataFile, key, checkData);
-
- if (prevKey != null && prevKey.compareTo(key) > 0)
+ try (UnfilteredRowIterator iterator = withValidation(new SSTableIdentityIterator(sstable, dataFile, key), dataFile.getPath()))
{
- saveOutOfOrderRow(prevKey, key, atoms);
- continue;
- }
+ if (prevKey != null && prevKey.compareTo(key) > 0)
+ {
+ saveOutOfOrderRow(prevKey, key, iterator);
+ continue;
+ }
- @SuppressWarnings("resource")
- AbstractCompactedRow compactedRow = new LazilyCompactedRow(controller, Collections.singletonList(atoms));
- if (writer.tryAppend(compactedRow) == null)
- emptyRows++;
- else
- goodRows++;
+ if (writer.tryAppend(iterator) == null)
+ emptyRows++;
+ else
+ goodRows++;
+ }
prevKey = key;
}
@@@ -318,10 -330,11 +332,11 @@@
currentRowPositionFromIndex = nextRowPositionFromIndex;
try
{
- nextIndexKey = indexFile.isEOF() ? null : ByteBufferUtil.readWithShortLength(indexFile);
- nextRowPositionFromIndex = indexFile.isEOF()
- ? dataFile.length()
- : rowIndexEntrySerializer.deserialize(indexFile).position;
+ nextIndexKey = !indexAvailable() ? null : ByteBufferUtil.readWithShortLength(indexFile);
+
+ nextRowPositionFromIndex = !indexAvailable()
+ ? dataFile.length()
- : rowIndexEntrySerializer.deserialize(indexFile, sstable.descriptor.version).position;
++ : rowIndexEntrySerializer.deserialize(indexFile).position;
}
catch (Throwable th)
{
http://git-wip-us.apache.org/repos/asf/cassandra/blob/a8bb75a7/src/java/org/apache/cassandra/io/sstable/format/SSTableReader.java
----------------------------------------------------------------------
diff --cc src/java/org/apache/cassandra/io/sstable/format/SSTableReader.java
index ab357a5,247d181..c326881
--- a/src/java/org/apache/cassandra/io/sstable/format/SSTableReader.java
+++ b/src/java/org/apache/cassandra/io/sstable/format/SSTableReader.java
@@@ -430,17 -421,12 +430,17 @@@ public abstract class SSTableReader ext
{
// Minimum components without which we can't do anything
assert components.contains(Component.DATA) : "Data component is missing for sstable " + descriptor;
- assert components.contains(Component.PRIMARY_INDEX) : "Primary index component is missing for sstable " + descriptor;
+ assert !validate || components.contains(Component.PRIMARY_INDEX) : "Primary index component is missing for sstable " + descriptor;
- Map<MetadataType, MetadataComponent> sstableMetadata = descriptor.getMetadataSerializer().deserialize(descriptor,
- EnumSet.of(MetadataType.VALIDATION, MetadataType.STATS));
+ // For the 3.0+ sstable format, the (misnomed) stats component hold the serialization header which we need to deserialize the sstable content
+ assert !descriptor.version.storeRows() || components.contains(Component.STATS) : "Stats component is missing for sstable " + descriptor;
+
+ EnumSet<MetadataType> types = EnumSet.of(MetadataType.VALIDATION, MetadataType.STATS, MetadataType.HEADER);
+ Map<MetadataType, MetadataComponent> sstableMetadata = descriptor.getMetadataSerializer().deserialize(descriptor, types);
ValidationMetadata validationMetadata = (ValidationMetadata) sstableMetadata.get(MetadataType.VALIDATION);
StatsMetadata statsMetadata = (StatsMetadata) sstableMetadata.get(MetadataType.STATS);
+ SerializationHeader.Component header = (SerializationHeader.Component) sstableMetadata.get(MetadataType.HEADER);
+ assert !descriptor.version.storeRows() || header != null;
// Check if sstable is created using same partitioner.
// Partitioner can be null, which indicates older version of sstable or no stats available.
http://git-wip-us.apache.org/repos/asf/cassandra/blob/a8bb75a7/src/java/org/apache/cassandra/io/sstable/format/big/BigTableReader.java
----------------------------------------------------------------------
diff --cc src/java/org/apache/cassandra/io/sstable/format/big/BigTableReader.java
index 7a7b913,3f375e7..b539c79
--- a/src/java/org/apache/cassandra/io/sstable/format/big/BigTableReader.java
+++ b/src/java/org/apache/cassandra/io/sstable/format/big/BigTableReader.java
@@@ -168,6 -167,6 +168,9 @@@ public class BigTableReader extends SST
int effectiveInterval = indexSummary.getEffectiveIndexIntervalAfterIndex(sampledIndex);
++ if (ifile == null)
++ return null;
++
// scan the on-disk index, starting at the nearest sampled position.
// The check against IndexInterval is to be exit the loop in the EQ case when the key looked for is not present
// (bloom filter false positive). But note that for non-EQ cases, we might need to check the first key of the
http://git-wip-us.apache.org/repos/asf/cassandra/blob/a8bb75a7/src/java/org/apache/cassandra/tools/StandaloneScrubber.java
----------------------------------------------------------------------
diff --cc src/java/org/apache/cassandra/tools/StandaloneScrubber.java
index c71caab,8982e2f..be73cf7
--- a/src/java/org/apache/cassandra/tools/StandaloneScrubber.java
+++ b/src/java/org/apache/cassandra/tools/StandaloneScrubber.java
@@@ -119,6 -119,6 +119,7 @@@ public class StandaloneScrubbe
{
try (LifecycleTransaction txn = LifecycleTransaction.offline(OperationType.SCRUB, sstable))
{
++ txn.obsoleteOriginals(); // make sure originals are deleted and avoid NPE if index is missing, CASSANDRA-9591
try (Scrubber scrubber = new Scrubber(cfs, txn, options.skipCorrupted, handler, true, !options.noValidate))
{
scrubber.scrub();
@@@ -131,10 -131,10 +132,6 @@@
throw t;
}
}
--
-- // Remove the sstable (it's been copied by scrub and snapshotted)
-- sstable.markObsolete(null);
-- sstable.selfRef().release();
}
catch (Exception e)
{
http://git-wip-us.apache.org/repos/asf/cassandra/blob/a8bb75a7/test/unit/org/apache/cassandra/db/ScrubTest.java
----------------------------------------------------------------------
diff --cc test/unit/org/apache/cassandra/db/ScrubTest.java
index c728f0d,b233e1f..2574847
--- a/test/unit/org/apache/cassandra/db/ScrubTest.java
+++ b/test/unit/org/apache/cassandra/db/ScrubTest.java
@@@ -29,15 -31,15 +29,16 @@@ import java.io.IOError
import java.io.IOException;
import java.io.RandomAccessFile;
-import org.apache.cassandra.config.KSMetaData;
+import org.apache.cassandra.Util;
+import org.apache.cassandra.UpdateBuilder;
import org.apache.cassandra.cql3.QueryProcessor;
-import org.apache.cassandra.db.composites.CellNameType;
+import org.apache.cassandra.db.partitions.*;
import org.apache.cassandra.db.marshal.BytesType;
-import org.apache.cassandra.db.marshal.CounterColumnType;
import org.apache.cassandra.db.marshal.UUIDType;
-import org.apache.cassandra.exceptions.ConfigurationException;
import org.apache.cassandra.db.compaction.OperationType;
import org.apache.cassandra.db.lifecycle.LifecycleTransaction;
++import org.apache.cassandra.db.rows.Row;
+import org.apache.cassandra.exceptions.ConfigurationException;
import org.apache.cassandra.exceptions.RequestExecutionException;
import org.apache.cassandra.io.compress.CompressionMetadata;
import org.apache.cassandra.io.sstable.format.SSTableFormat;
@@@ -280,6 -328,31 +281,27 @@@ public class ScrubTes
}
@Test
+ public void testScrubNoIndex() throws IOException, ExecutionException, InterruptedException, ConfigurationException
+ {
+ CompactionManager.instance.disableAutoCompaction();
+ Keyspace keyspace = Keyspace.open(KEYSPACE);
+ ColumnFamilyStore cfs = keyspace.getColumnFamilyStore(CF);
+ cfs.clearUnsafe();
+
- List<Row> rows;
-
+ // insert data and verify we get it back w/ range query
+ fillCF(cfs, 10);
- rows = cfs.getRangeSlice(Util.range("", ""), null, new IdentityQueryFilter(), 1000);
- assertEquals(10, rows.size());
++ assertOrderedAll(cfs, 10);
+
+ for (SSTableReader sstable : cfs.getSSTables())
+ new File(sstable.descriptor.filenameFor(Component.PRIMARY_INDEX)).delete();
+
- CompactionManager.instance.performScrub(cfs, false, true);
++ CompactionManager.instance.performScrub(cfs, false, true, true);
+
+ // check data is still there
- rows = cfs.getRangeSlice(Util.range("", ""), null, new IdentityQueryFilter(), 1000);
- assertEquals(10, rows.size());
++ assertOrderedAll(cfs, 10);
+ }
+
+ @Test
public void testScrubOutOfOrder() throws Exception
{
CompactionManager.instance.disableAutoCompaction();
[10/12] cassandra git commit: Merge branch 'cassandra-2.1' into
cassandra-2.2
Posted by be...@apache.org.
http://git-wip-us.apache.org/repos/asf/cassandra/blob/ebe18bb2/src/java/org/apache/cassandra/io/sstable/format/SSTableReader.java
----------------------------------------------------------------------
diff --cc src/java/org/apache/cassandra/io/sstable/format/SSTableReader.java
index 0cc425a,0000000..247d181
mode 100644,000000..100644
--- a/src/java/org/apache/cassandra/io/sstable/format/SSTableReader.java
+++ b/src/java/org/apache/cassandra/io/sstable/format/SSTableReader.java
@@@ -1,2218 -1,0 +1,2238 @@@
+/*
+ * Licensed to the Apache Software Foundation (ASF) under one
+ * or more contributor license agreements. See the NOTICE file
+ * distributed with this work for additional information
+ * regarding copyright ownership. The ASF licenses this file
+ * to you under the Apache License, Version 2.0 (the
+ * "License"); you may not use this file except in compliance
+ * with the License. You may obtain a copy of the License at
+ *
+ * http://www.apache.org/licenses/LICENSE-2.0
+ *
+ * Unless required by applicable law or agreed to in writing, software
+ * distributed under the License is distributed on an "AS IS" BASIS,
+ * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ * See the License for the specific language governing permissions and
+ * limitations under the License.
+ */
+package org.apache.cassandra.io.sstable.format;
+
+import java.io.*;
+import java.nio.ByteBuffer;
+import java.util.*;
+import java.util.concurrent.*;
+import java.util.concurrent.atomic.AtomicBoolean;
+import java.util.concurrent.atomic.AtomicLong;
+
+import com.google.common.annotations.VisibleForTesting;
+import com.google.common.base.Predicate;
+import com.google.common.collect.Iterables;
+import com.google.common.collect.Iterators;
+import com.google.common.collect.Ordering;
+import com.google.common.primitives.Longs;
+import com.google.common.util.concurrent.RateLimiter;
+
+import com.clearspring.analytics.stream.cardinality.CardinalityMergeException;
+import com.clearspring.analytics.stream.cardinality.HyperLogLogPlus;
+import com.clearspring.analytics.stream.cardinality.ICardinality;
+import com.codahale.metrics.Counter;
+import org.apache.cassandra.cache.CachingOptions;
+import org.apache.cassandra.cache.InstrumentingCache;
+import org.apache.cassandra.cache.KeyCacheKey;
+import org.apache.cassandra.concurrent.DebuggableThreadPoolExecutor;
+import org.apache.cassandra.concurrent.ScheduledExecutors;
+import org.apache.cassandra.config.*;
+import org.apache.cassandra.db.*;
+import org.apache.cassandra.db.columniterator.OnDiskAtomIterator;
+import org.apache.cassandra.db.commitlog.ReplayPosition;
+import org.apache.cassandra.db.composites.CellName;
+import org.apache.cassandra.db.filter.ColumnSlice;
+import org.apache.cassandra.db.index.SecondaryIndex;
+import org.apache.cassandra.db.lifecycle.Tracker;
+import org.apache.cassandra.dht.*;
+import org.apache.cassandra.io.compress.CompressionMetadata;
+import org.apache.cassandra.io.sstable.*;
+import org.apache.cassandra.io.sstable.metadata.*;
+import org.apache.cassandra.io.util.*;
+import org.apache.cassandra.metrics.RestorableMeter;
+import org.apache.cassandra.metrics.StorageMetrics;
+import org.apache.cassandra.service.ActiveRepairService;
+import org.apache.cassandra.service.CacheService;
+import org.apache.cassandra.service.StorageService;
+import org.apache.cassandra.utils.*;
+import org.apache.cassandra.utils.concurrent.OpOrder;
+import org.slf4j.Logger;
+import org.slf4j.LoggerFactory;
+import org.apache.cassandra.utils.concurrent.Ref;
+import org.apache.cassandra.utils.concurrent.SelfRefCounted;
+
+import static org.apache.cassandra.db.Directories.SECONDARY_INDEX_NAME_SEPARATOR;
+
+/**
+ * An SSTableReader can be constructed in a number of places, but typically is either
+ * read from disk at startup, or constructed from a flushed memtable, or after compaction
+ * to replace some existing sstables. However once created, an sstablereader may also be modified.
+ *
+ * A reader's OpenReason describes its current stage in its lifecycle, as follows:
+ *
+ *
+ * <pre> {@code
+ * NORMAL
+ * From: None => Reader has been read from disk, either at startup or from a flushed memtable
+ * EARLY => Reader is the final result of a compaction
+ * MOVED_START => Reader WAS being compacted, but this failed and it has been restored to NORMAL status
+ *
+ * EARLY
+ * From: None => Reader is a compaction replacement that is either incomplete and has been opened
+ * to represent its partial result status, or has been finished but the compaction
+ * it is a part of has not yet completed fully
+ * EARLY => Same as from None, only it is not the first time it has been
+ *
+ * MOVED_START
+ * From: NORMAL => Reader is being compacted. This compaction has not finished, but the compaction result
+ * is either partially or fully opened, to either partially or fully replace this reader.
+ * This reader's start key has been updated to represent this, so that reads only hit
+ * one or the other reader.
+ *
+ * METADATA_CHANGE
+ * From: NORMAL => Reader has seen low traffic and the amount of memory available for index summaries is
+ * constrained, so its index summary has been downsampled.
+ * METADATA_CHANGE => Same
+ * } </pre>
+ *
+ * Note that in parallel to this, there are two different Descriptor types; TMPLINK and FINAL; the latter corresponds
+ * to NORMAL state readers and all readers that replace a NORMAL one. TMPLINK is used for EARLY state readers and
+ * no others.
+ *
+ * When a reader is being compacted, if the result is large its replacement may be opened as EARLY before compaction
+ * completes in order to present the result to consumers earlier. In this case the reader will itself be changed to
+ * a MOVED_START state, where its start no longer represents its on-disk minimum key. This is to permit reads to be
+ * directed to only one reader when the two represent the same data. The EARLY file can represent a compaction result
+ * that is either partially complete and still in-progress, or a complete and immutable sstable that is part of a larger
+ * macro compaction action that has not yet fully completed.
+ *
+ * Currently ALL compaction results at least briefly go through an EARLY open state prior to completion, regardless
+ * of if early opening is enabled.
+ *
+ * Since a reader can be created multiple times over the same shared underlying resources, and the exact resources
+ * it shares between each instance differ subtly, we track the lifetime of any underlying resource with its own
+ * reference count, which each instance takes a Ref to. Each instance then tracks references to itself, and once these
+ * all expire it releases its Refs to these underlying resources.
+ *
+ * There is some shared cleanup behaviour needed only once all sstablereaders in a certain stage of their lifecycle
+ * (i.e. EARLY or NORMAL opening), and some that must only occur once all readers of any kind over a single logical
+ * sstable have expired. These are managed by the TypeTidy and GlobalTidy classes at the bottom, and are effectively
+ * managed as another resource each instance tracks its own Ref instance to, to ensure all of these resources are
+ * cleaned up safely and can be debugged otherwise.
+ *
+ * TODO: fill in details about Tracker and lifecycle interactions for tools, and for compaction strategies
+ */
+public abstract class SSTableReader extends SSTable implements SelfRefCounted<SSTableReader>
+{
+ private static final Logger logger = LoggerFactory.getLogger(SSTableReader.class);
+
+ private static final ScheduledThreadPoolExecutor syncExecutor = new ScheduledThreadPoolExecutor(1);
+ private static final RateLimiter meterSyncThrottle = RateLimiter.create(100.0);
+
+ public static final Comparator<SSTableReader> maxTimestampComparator = new Comparator<SSTableReader>()
+ {
+ public int compare(SSTableReader o1, SSTableReader o2)
+ {
+ long ts1 = o1.getMaxTimestamp();
+ long ts2 = o2.getMaxTimestamp();
+ return (ts1 > ts2 ? -1 : (ts1 == ts2 ? 0 : 1));
+ }
+ };
+
+ // it's just an object, which we use regular Object equality on; we introduce a special class just for easy recognition
+ public static final class UniqueIdentifier {}
+
+ public static final Comparator<SSTableReader> sstableComparator = new Comparator<SSTableReader>()
+ {
+ public int compare(SSTableReader o1, SSTableReader o2)
+ {
+ return o1.first.compareTo(o2.first);
+ }
+ };
+
+ public static final Ordering<SSTableReader> sstableOrdering = Ordering.from(sstableComparator);
+
+ /**
+ * maxDataAge is a timestamp in local server time (e.g. System.currentTimeMilli) which represents an upper bound
+ * to the newest piece of data stored in the sstable. In other words, this sstable does not contain items created
+ * later than maxDataAge.
+ *
+ * The field is not serialized to disk, so relying on it for more than what truncate does is not advised.
+ *
+ * When a new sstable is flushed, maxDataAge is set to the time of creation.
+ * When a sstable is created from compaction, maxDataAge is set to max of all merged sstables.
+ *
+ * The age is in milliseconds since epoc and is local to this host.
+ */
+ public final long maxDataAge;
+
+ public enum OpenReason
+ {
+ NORMAL,
+ EARLY,
+ METADATA_CHANGE,
+ MOVED_START
+ }
+
+ public final OpenReason openReason;
+ public final UniqueIdentifier instanceId = new UniqueIdentifier();
+
+ // indexfile and datafile: might be null before a call to load()
+ protected SegmentedFile ifile;
+ protected SegmentedFile dfile;
+ protected IndexSummary indexSummary;
+ protected IFilter bf;
+
+ protected final RowIndexEntry.IndexSerializer rowIndexEntrySerializer;
+
+ protected InstrumentingCache<KeyCacheKey, RowIndexEntry> keyCache;
+
+ protected final BloomFilterTracker bloomFilterTracker = new BloomFilterTracker();
+
+ // technically isCompacted is not necessary since it should never be unreferenced unless it is also compacted,
+ // but it seems like a good extra layer of protection against reference counting bugs to not delete data based on that alone
+ protected final AtomicBoolean isSuspect = new AtomicBoolean(false);
+
+ // not final since we need to be able to change level on a file.
+ protected volatile StatsMetadata sstableMetadata;
+
+ protected final AtomicLong keyCacheHit = new AtomicLong(0);
+ protected final AtomicLong keyCacheRequest = new AtomicLong(0);
+
+ private final InstanceTidier tidy = new InstanceTidier(descriptor, metadata);
+ private final Ref<SSTableReader> selfRef = new Ref<>(this, tidy);
+
+ private RestorableMeter readMeter;
+
+ /**
+ * Calculate approximate key count.
+ * If cardinality estimator is available on all given sstables, then this method use them to estimate
+ * key count.
+ * If not, then this uses index summaries.
+ *
+ * @param sstables SSTables to calculate key count
+ * @return estimated key count
+ */
+ public static long getApproximateKeyCount(Collection<SSTableReader> sstables)
+ {
+ long count = -1;
+
+ // check if cardinality estimator is available for all SSTables
+ boolean cardinalityAvailable = !sstables.isEmpty() && Iterators.all(sstables.iterator(), new Predicate<SSTableReader>()
+ {
+ public boolean apply(SSTableReader sstable)
+ {
+ return sstable.descriptor.version.hasNewStatsFile();
+ }
+ });
+
+ // if it is, load them to estimate key count
+ if (cardinalityAvailable)
+ {
+ boolean failed = false;
+ ICardinality cardinality = null;
+ for (SSTableReader sstable : sstables)
+ {
+ if (sstable.openReason == OpenReason.EARLY)
+ continue;
+
+ try
+ {
+ CompactionMetadata metadata = (CompactionMetadata) sstable.descriptor.getMetadataSerializer().deserialize(sstable.descriptor, MetadataType.COMPACTION);
+ assert metadata != null : sstable.getFilename();
+ if (cardinality == null)
+ cardinality = metadata.cardinalityEstimator;
+ else
+ cardinality = cardinality.merge(metadata.cardinalityEstimator);
+ }
+ catch (IOException e)
+ {
+ logger.warn("Reading cardinality from Statistics.db failed.", e);
+ failed = true;
+ break;
+ }
+ catch (CardinalityMergeException e)
+ {
+ logger.warn("Cardinality merge failed.", e);
+ failed = true;
+ break;
+ }
+ }
+ if (cardinality != null && !failed)
+ count = cardinality.cardinality();
+ }
+
+ // if something went wrong above or cardinality is not available, calculate using index summary
+ if (count < 0)
+ {
+ for (SSTableReader sstable : sstables)
+ count += sstable.estimatedKeys();
+ }
+ return count;
+ }
+
+ /**
+ * Estimates how much of the keys we would keep if the sstables were compacted together
+ */
+ public static double estimateCompactionGain(Set<SSTableReader> overlapping)
+ {
+ Set<ICardinality> cardinalities = new HashSet<>(overlapping.size());
+ for (SSTableReader sstable : overlapping)
+ {
+ try
+ {
+ ICardinality cardinality = ((CompactionMetadata) sstable.descriptor.getMetadataSerializer().deserialize(sstable.descriptor, MetadataType.COMPACTION)).cardinalityEstimator;
+ if (cardinality != null)
+ cardinalities.add(cardinality);
+ else
+ logger.debug("Got a null cardinality estimator in: {}", sstable.getFilename());
+ }
+ catch (IOException e)
+ {
+ logger.warn("Could not read up compaction metadata for {}", sstable, e);
+ }
+ }
+ long totalKeyCountBefore = 0;
+ for (ICardinality cardinality : cardinalities)
+ {
+ totalKeyCountBefore += cardinality.cardinality();
+ }
+ if (totalKeyCountBefore == 0)
+ return 1;
+
+ long totalKeyCountAfter = mergeCardinalities(cardinalities).cardinality();
+ logger.debug("Estimated compaction gain: {}/{}={}", totalKeyCountAfter, totalKeyCountBefore, ((double)totalKeyCountAfter)/totalKeyCountBefore);
+ return ((double)totalKeyCountAfter)/totalKeyCountBefore;
+ }
+
+ private static ICardinality mergeCardinalities(Collection<ICardinality> cardinalities)
+ {
+ ICardinality base = new HyperLogLogPlus(13, 25); // see MetadataCollector.cardinality
+ try
+ {
+ base = base.merge(cardinalities.toArray(new ICardinality[cardinalities.size()]));
+ }
+ catch (CardinalityMergeException e)
+ {
+ logger.warn("Could not merge cardinalities", e);
+ }
+ return base;
+ }
+
+ public static SSTableReader open(Descriptor descriptor) throws IOException
+ {
+ CFMetaData metadata;
+ if (descriptor.cfname.contains(SECONDARY_INDEX_NAME_SEPARATOR))
+ {
+ int i = descriptor.cfname.indexOf(SECONDARY_INDEX_NAME_SEPARATOR);
+ String parentName = descriptor.cfname.substring(0, i);
+ CFMetaData parent = Schema.instance.getCFMetaData(descriptor.ksname, parentName);
+ ColumnDefinition def = parent.getColumnDefinitionForIndex(descriptor.cfname.substring(i + 1));
+ metadata = CFMetaData.newIndexMetadata(parent, def, SecondaryIndex.getIndexComparator(parent, def));
+ }
+ else
+ {
+ metadata = Schema.instance.getCFMetaData(descriptor.ksname, descriptor.cfname);
+ }
+ return open(descriptor, metadata);
+ }
+
+ public static SSTableReader open(Descriptor desc, CFMetaData metadata) throws IOException
+ {
+ IPartitioner p = desc.cfname.contains(SECONDARY_INDEX_NAME_SEPARATOR)
+ ? new LocalPartitioner(metadata.getKeyValidator())
+ : StorageService.getPartitioner();
+ return open(desc, componentsFor(desc), metadata, p);
+ }
+
+ public static SSTableReader open(Descriptor descriptor, Set<Component> components, CFMetaData metadata, IPartitioner partitioner) throws IOException
+ {
+ return open(descriptor, components, metadata, partitioner, true);
+ }
+
+ // use only for offline or "Standalone" operations
+ public static SSTableReader openNoValidation(Descriptor descriptor, Set<Component> components, ColumnFamilyStore cfs) throws IOException
+ {
+ return open(descriptor, components, cfs.metadata, cfs.partitioner, false);
+ }
+
+ /**
+ * Open SSTable reader to be used in batch mode(such as sstableloader).
+ *
+ * @param descriptor
+ * @param components
+ * @param metadata
+ * @param partitioner
+ * @return opened SSTableReader
+ * @throws IOException
+ */
+ public static SSTableReader openForBatch(Descriptor descriptor, Set<Component> components, CFMetaData metadata, IPartitioner partitioner) throws IOException
+ {
+ // Minimum components without which we can't do anything
+ assert components.contains(Component.DATA) : "Data component is missing for sstable " + descriptor;
+ assert components.contains(Component.PRIMARY_INDEX) : "Primary index component is missing for sstable " + descriptor;
+
+ Map<MetadataType, MetadataComponent> sstableMetadata = descriptor.getMetadataSerializer().deserialize(descriptor,
+ EnumSet.of(MetadataType.VALIDATION, MetadataType.STATS));
+ ValidationMetadata validationMetadata = (ValidationMetadata) sstableMetadata.get(MetadataType.VALIDATION);
+ StatsMetadata statsMetadata = (StatsMetadata) sstableMetadata.get(MetadataType.STATS);
+
+ // Check if sstable is created using same partitioner.
+ // Partitioner can be null, which indicates older version of sstable or no stats available.
+ // In that case, we skip the check.
+ String partitionerName = partitioner.getClass().getCanonicalName();
+ if (validationMetadata != null && !partitionerName.equals(validationMetadata.partitioner))
+ {
+ logger.error(String.format("Cannot open %s; partitioner %s does not match system partitioner %s. Note that the default partitioner starting with Cassandra 1.2 is Murmur3Partitioner, so you will need to edit that to match your old partitioner if upgrading.",
+ descriptor, validationMetadata.partitioner, partitionerName));
+ System.exit(1);
+ }
+
+ logger.info("Opening {} ({} bytes)", descriptor, new File(descriptor.filenameFor(Component.DATA)).length());
+ SSTableReader sstable = internalOpen(descriptor, components, metadata, partitioner, System.currentTimeMillis(),
+ statsMetadata, OpenReason.NORMAL);
+
+ // special implementation of load to use non-pooled SegmentedFile builders
+ try(SegmentedFile.Builder ibuilder = new BufferedSegmentedFile.Builder();
+ SegmentedFile.Builder dbuilder = sstable.compression
+ ? new CompressedSegmentedFile.Builder(null)
+ : new BufferedSegmentedFile.Builder())
+ {
+ if (!sstable.loadSummary(ibuilder, dbuilder))
+ sstable.buildSummary(false, ibuilder, dbuilder, false, Downsampling.BASE_SAMPLING_LEVEL);
+ sstable.ifile = ibuilder.complete(sstable.descriptor.filenameFor(Component.PRIMARY_INDEX));
+ sstable.dfile = dbuilder.complete(sstable.descriptor.filenameFor(Component.DATA));
+ sstable.bf = FilterFactory.AlwaysPresent;
+ sstable.setup(true);
+ return sstable;
+ }
+ }
+
+ private static SSTableReader open(Descriptor descriptor,
+ Set<Component> components,
+ CFMetaData metadata,
+ IPartitioner partitioner,
+ boolean validate) throws IOException
+ {
+ // Minimum components without which we can't do anything
+ assert components.contains(Component.DATA) : "Data component is missing for sstable " + descriptor;
- assert components.contains(Component.PRIMARY_INDEX) : "Primary index component is missing for sstable " + descriptor;
++ assert !validate || components.contains(Component.PRIMARY_INDEX) : "Primary index component is missing for sstable " + descriptor;
+
+ Map<MetadataType, MetadataComponent> sstableMetadata = descriptor.getMetadataSerializer().deserialize(descriptor,
- EnumSet.of(MetadataType.VALIDATION, MetadataType.STATS));
++ EnumSet.of(MetadataType.VALIDATION, MetadataType.STATS));
+ ValidationMetadata validationMetadata = (ValidationMetadata) sstableMetadata.get(MetadataType.VALIDATION);
+ StatsMetadata statsMetadata = (StatsMetadata) sstableMetadata.get(MetadataType.STATS);
+
+ // Check if sstable is created using same partitioner.
+ // Partitioner can be null, which indicates older version of sstable or no stats available.
+ // In that case, we skip the check.
+ String partitionerName = partitioner.getClass().getCanonicalName();
+ if (validationMetadata != null && !partitionerName.equals(validationMetadata.partitioner))
+ {
+ logger.error(String.format("Cannot open %s; partitioner %s does not match system partitioner %s. Note that the default partitioner starting with Cassandra 1.2 is Murmur3Partitioner, so you will need to edit that to match your old partitioner if upgrading.",
+ descriptor, validationMetadata.partitioner, partitionerName));
+ System.exit(1);
+ }
+
+ logger.info("Opening {} ({} bytes)", descriptor, new File(descriptor.filenameFor(Component.DATA)).length());
+ SSTableReader sstable = internalOpen(descriptor, components, metadata, partitioner, System.currentTimeMillis(),
+ statsMetadata, OpenReason.NORMAL);
+
+ // load index and filter
+ long start = System.nanoTime();
+ sstable.load(validationMetadata);
+ logger.debug("INDEX LOAD TIME for {}: {} ms.", descriptor, TimeUnit.NANOSECONDS.toMillis(System.nanoTime() - start));
+
+ sstable.setup(!validate);
+ if (validate)
+ sstable.validate();
+
+ if (sstable.getKeyCache() != null)
+ logger.debug("key cache contains {}/{} keys", sstable.getKeyCache().size(), sstable.getKeyCache().getCapacity());
+
+ return sstable;
+ }
+
+ public static void logOpenException(Descriptor descriptor, IOException e)
+ {
+ if (e instanceof FileNotFoundException)
+ logger.error("Missing sstable component in {}; skipped because of {}", descriptor, e.getMessage());
+ else
+ logger.error("Corrupt sstable {}; skipped", descriptor, e);
+ }
+
+ public static Collection<SSTableReader> openAll(Set<Map.Entry<Descriptor, Set<Component>>> entries,
+ final CFMetaData metadata,
+ final IPartitioner partitioner)
+ {
+ final Collection<SSTableReader> sstables = new LinkedBlockingQueue<>();
+
+ ExecutorService executor = DebuggableThreadPoolExecutor.createWithFixedPoolSize("SSTableBatchOpen", FBUtilities.getAvailableProcessors());
+ for (final Map.Entry<Descriptor, Set<Component>> entry : entries)
+ {
+ Runnable runnable = new Runnable()
+ {
+ public void run()
+ {
+ SSTableReader sstable;
+ try
+ {
+ sstable = open(entry.getKey(), entry.getValue(), metadata, partitioner);
+ }
+ catch (IOException ex)
+ {
+ logger.error("Corrupt sstable {}; skipped", entry, ex);
+ return;
+ }
+ sstables.add(sstable);
+ }
+ };
+ executor.submit(runnable);
+ }
+
+ executor.shutdown();
+ try
+ {
+ executor.awaitTermination(7, TimeUnit.DAYS);
+ }
+ catch (InterruptedException e)
+ {
+ throw new AssertionError(e);
+ }
+
+ return sstables;
+
+ }
+
+ /**
+ * Open a RowIndexedReader which already has its state initialized (by SSTableWriter).
+ */
+ public static SSTableReader internalOpen(Descriptor desc,
+ Set<Component> components,
+ CFMetaData metadata,
+ IPartitioner partitioner,
+ SegmentedFile ifile,
+ SegmentedFile dfile,
+ IndexSummary isummary,
+ IFilter bf,
+ long maxDataAge,
+ StatsMetadata sstableMetadata,
+ OpenReason openReason)
+ {
+ assert desc != null && partitioner != null && ifile != null && dfile != null && isummary != null && bf != null && sstableMetadata != null;
+
+ SSTableReader reader = internalOpen(desc, components, metadata, partitioner, maxDataAge, sstableMetadata, openReason);
+
+ reader.bf = bf;
+ reader.ifile = ifile;
+ reader.dfile = dfile;
+ reader.indexSummary = isummary;
+ reader.setup(false);
+
+ return reader;
+ }
+
+
+ private static SSTableReader internalOpen(final Descriptor descriptor,
+ Set<Component> components,
+ CFMetaData metadata,
+ IPartitioner partitioner,
+ Long maxDataAge,
+ StatsMetadata sstableMetadata,
+ OpenReason openReason)
+ {
+ Factory readerFactory = descriptor.getFormat().getReaderFactory();
+
+ return readerFactory.open(descriptor, components, metadata, partitioner, maxDataAge, sstableMetadata, openReason);
+ }
+
+ protected SSTableReader(final Descriptor desc,
+ Set<Component> components,
+ CFMetaData metadata,
+ IPartitioner partitioner,
+ long maxDataAge,
+ StatsMetadata sstableMetadata,
+ OpenReason openReason)
+ {
+ super(desc, components, metadata, partitioner);
+ this.sstableMetadata = sstableMetadata;
+ this.maxDataAge = maxDataAge;
+ this.openReason = openReason;
+ this.rowIndexEntrySerializer = descriptor.version.getSSTableFormat().getIndexSerializer(metadata);
+ }
+
+ public static long getTotalBytes(Iterable<SSTableReader> sstables)
+ {
+ long sum = 0;
+ for (SSTableReader sstable : sstables)
+ sum += sstable.onDiskLength();
+ return sum;
+ }
+
+ public static long getTotalUncompressedBytes(Iterable<SSTableReader> sstables)
+ {
+ long sum = 0;
+ for (SSTableReader sstable : sstables)
+ sum += sstable.uncompressedLength();
+
+ return sum;
+ }
+
+ public boolean equals(Object that)
+ {
+ return that instanceof SSTableReader && ((SSTableReader) that).descriptor.equals(this.descriptor);
+ }
+
+ public int hashCode()
+ {
+ return this.descriptor.hashCode();
+ }
+
+ public String getFilename()
+ {
+ return dfile.path();
+ }
+
- public String getIndexFilename()
- {
- return ifile.path();
- }
-
+ public void setupKeyCache()
+ {
+ // under normal operation we can do this at any time, but SSTR is also used outside C* proper,
+ // e.g. by BulkLoader, which does not initialize the cache. As a kludge, we set up the cache
+ // here when we know we're being wired into the rest of the server infrastructure.
+ keyCache = CacheService.instance.keyCache;
+ }
+
+ private void load(ValidationMetadata validation) throws IOException
+ {
+ if (metadata.getBloomFilterFpChance() == 1.0)
+ {
+ // bf is disabled.
+ load(false, true);
+ bf = FilterFactory.AlwaysPresent;
+ }
++ else if (!components.contains(Component.PRIMARY_INDEX))
++ {
++ // avoid any reading of the missing primary index component.
++ // this should only happen during StandaloneScrubber
++ load(false, false);
++ }
+ else if (!components.contains(Component.FILTER) || validation == null)
+ {
+ // bf is enabled, but filter component is missing.
+ load(true, true);
+ }
+ else if (validation.bloomFilterFPChance != metadata.getBloomFilterFpChance())
+ {
+ // bf fp chance in sstable metadata and it has changed since compaction.
+ load(true, true);
+ }
+ else
+ {
+ // bf is enabled and fp chance matches the currently configured value.
+ load(false, true);
+ loadBloomFilter();
+ }
+ }
+
+ /**
+ * Load bloom filter from Filter.db file.
+ *
+ * @throws IOException
+ */
+ private void loadBloomFilter() throws IOException
+ {
+ try (DataInputStream stream = new DataInputStream(new BufferedInputStream(new FileInputStream(descriptor.filenameFor(Component.FILTER)))))
+ {
+ bf = FilterFactory.deserialize(stream, true);
+ }
+ }
+
+ /**
+ * Loads ifile, dfile and indexSummary, and optionally recreates the bloom filter.
+ * @param saveSummaryIfCreated for bulk loading purposes, if the summary was absent and needed to be built, you can
+ * avoid persisting it to disk by setting this to false
+ */
+ private void load(boolean recreateBloomFilter, boolean saveSummaryIfCreated) throws IOException
+ {
+ try(SegmentedFile.Builder ibuilder = SegmentedFile.getBuilder(DatabaseDescriptor.getIndexAccessMode(), false);
+ SegmentedFile.Builder dbuilder = SegmentedFile.getBuilder(DatabaseDescriptor.getDiskAccessMode(), compression))
+ {
+ boolean summaryLoaded = loadSummary(ibuilder, dbuilder);
+ boolean builtSummary = false;
+ if (recreateBloomFilter || !summaryLoaded)
+ {
+ buildSummary(recreateBloomFilter, ibuilder, dbuilder, summaryLoaded, Downsampling.BASE_SAMPLING_LEVEL);
+ builtSummary = true;
+ }
+
- ifile = ibuilder.complete(descriptor.filenameFor(Component.PRIMARY_INDEX));
++ if (components.contains(Component.PRIMARY_INDEX))
++ ifile = ibuilder.complete(descriptor.filenameFor(Component.PRIMARY_INDEX));
++
+ dfile = dbuilder.complete(descriptor.filenameFor(Component.DATA));
+
+ // Check for an index summary that was downsampled even though the serialization format doesn't support
+ // that. If it was downsampled, rebuild it. See CASSANDRA-8993 for details.
- if (!descriptor.version.hasSamplingLevel() && !builtSummary && !validateSummarySamplingLevel())
++ if (!descriptor.version.hasSamplingLevel() && !builtSummary && !validateSummarySamplingLevel() && ifile != null)
+ {
+ indexSummary.close();
+ ifile.close();
+ dfile.close();
+
+ logger.info("Detected erroneously downsampled index summary; will rebuild summary at full sampling");
+ FileUtils.deleteWithConfirm(new File(descriptor.filenameFor(Component.SUMMARY)));
+
+ try(SegmentedFile.Builder ibuilderRebuild = SegmentedFile.getBuilder(DatabaseDescriptor.getIndexAccessMode(), false);
+ SegmentedFile.Builder dbuilderRebuild = SegmentedFile.getBuilder(DatabaseDescriptor.getDiskAccessMode(), compression))
+ {
+ buildSummary(false, ibuilderRebuild, dbuilderRebuild, false, Downsampling.BASE_SAMPLING_LEVEL);
+ ifile = ibuilderRebuild.complete(descriptor.filenameFor(Component.PRIMARY_INDEX));
+ dfile = dbuilderRebuild.complete(descriptor.filenameFor(Component.DATA));
+ saveSummary(ibuilderRebuild, dbuilderRebuild);
+ }
+ }
+ else if (saveSummaryIfCreated && builtSummary)
+ {
+ saveSummary(ibuilder, dbuilder);
+ }
+ }
+ }
+
+ /**
+ * Build index summary(and optionally bloom filter) by reading through Index.db file.
+ *
+ * @param recreateBloomFilter true if recreate bloom filter
+ * @param ibuilder
+ * @param dbuilder
+ * @param summaryLoaded true if index summary is already loaded and not need to build again
+ * @throws IOException
+ */
+ private void buildSummary(boolean recreateBloomFilter, SegmentedFile.Builder ibuilder, SegmentedFile.Builder dbuilder, boolean summaryLoaded, int samplingLevel) throws IOException
+ {
++ if (!components.contains(Component.PRIMARY_INDEX))
++ return;
++
+ // we read the positions in a BRAF so we don't have to worry about an entry spanning a mmap boundary.
+ try (RandomAccessReader primaryIndex = RandomAccessReader.open(new File(descriptor.filenameFor(Component.PRIMARY_INDEX))))
+ {
+ long indexSize = primaryIndex.length();
+ long histogramCount = sstableMetadata.estimatedRowSize.count();
+ long estimatedKeys = histogramCount > 0 && !sstableMetadata.estimatedRowSize.isOverflowed()
+ ? histogramCount
+ : estimateRowsFromIndex(primaryIndex); // statistics is supposed to be optional
+
+ if (recreateBloomFilter)
+ bf = FilterFactory.getFilter(estimatedKeys, metadata.getBloomFilterFpChance(), true);
+
+ try (IndexSummaryBuilder summaryBuilder = summaryLoaded ? null : new IndexSummaryBuilder(estimatedKeys, metadata.getMinIndexInterval(), samplingLevel))
+ {
+ long indexPosition;
+ RowIndexEntry.IndexSerializer rowIndexSerializer = descriptor.getFormat().getIndexSerializer(metadata);
+
+ while ((indexPosition = primaryIndex.getFilePointer()) != indexSize)
+ {
+ ByteBuffer key = ByteBufferUtil.readWithShortLength(primaryIndex);
+ RowIndexEntry indexEntry = rowIndexSerializer.deserialize(primaryIndex, descriptor.version);
+ DecoratedKey decoratedKey = partitioner.decorateKey(key);
+ if (first == null)
+ first = decoratedKey;
+ last = decoratedKey;
+
+ if (recreateBloomFilter)
+ bf.add(decoratedKey);
+
+ // if summary was already read from disk we don't want to re-populate it using primary index
+ if (!summaryLoaded)
+ {
+ summaryBuilder.maybeAddEntry(decoratedKey, indexPosition);
+ ibuilder.addPotentialBoundary(indexPosition);
+ dbuilder.addPotentialBoundary(indexEntry.position);
+ }
+ }
+
+ if (!summaryLoaded)
+ indexSummary = summaryBuilder.build(partitioner);
+ }
+ }
+
+ first = getMinimalKey(first);
+ last = getMinimalKey(last);
+ }
+
+ /**
+ * Load index summary from Summary.db file if it exists.
+ *
+ * if loaded index summary has different index interval from current value stored in schema,
+ * then Summary.db file will be deleted and this returns false to rebuild summary.
+ *
+ * @param ibuilder
+ * @param dbuilder
+ * @return true if index summary is loaded successfully from Summary.db file.
+ */
+ @SuppressWarnings("resource")
+ public boolean loadSummary(SegmentedFile.Builder ibuilder, SegmentedFile.Builder dbuilder)
+ {
+ File summariesFile = new File(descriptor.filenameFor(Component.SUMMARY));
+ if (!summariesFile.exists())
+ return false;
+
+ DataInputStream iStream = null;
+ try
+ {
+ iStream = new DataInputStream(new FileInputStream(summariesFile));
+ indexSummary = IndexSummary.serializer.deserialize(
+ iStream, partitioner, descriptor.version.hasSamplingLevel(),
+ metadata.getMinIndexInterval(), metadata.getMaxIndexInterval());
+ first = partitioner.decorateKey(ByteBufferUtil.readWithLength(iStream));
+ last = partitioner.decorateKey(ByteBufferUtil.readWithLength(iStream));
+ ibuilder.deserializeBounds(iStream);
+ dbuilder.deserializeBounds(iStream);
+ }
+ catch (IOException e)
+ {
+ if (indexSummary != null)
+ indexSummary.close();
+ logger.debug("Cannot deserialize SSTable Summary File {}: {}", summariesFile.getPath(), e.getMessage());
+ // corrupted; delete it and fall back to creating a new summary
+ FileUtils.closeQuietly(iStream);
+ // delete it and fall back to creating a new summary
+ FileUtils.deleteWithConfirm(summariesFile);
+ return false;
+ }
+ finally
+ {
+ FileUtils.closeQuietly(iStream);
+ }
+
+ return true;
+ }
+
+ /**
+ * Validates that an index summary has full sampling, as expected when the serialization format does not support
+ * persisting the sampling level.
+ * @return true if the summary has full sampling, false otherwise
+ */
+ private boolean validateSummarySamplingLevel()
+ {
+ // We need to check index summary entries against the index to verify that none of them were dropped due to
+ // downsampling. Downsampling can drop any of the first BASE_SAMPLING_LEVEL entries (repeating that drop pattern
+ // for the remainder of the summary). Unfortunately, the first entry to be dropped is the entry at
+ // index (BASE_SAMPLING_LEVEL - 1), so we need to check a full set of BASE_SAMPLING_LEVEL entries.
++ if (ifile == null)
++ return false;
++
+ Iterator<FileDataInput> segments = ifile.iterator(0);
+ int i = 0;
+ int summaryEntriesChecked = 0;
+ int expectedIndexInterval = getMinIndexInterval();
+ while (segments.hasNext())
+ {
+ String path = null;
+ try (FileDataInput in = segments.next())
+ {
+ path = in.getPath();
+ while (!in.isEOF())
+ {
+ ByteBuffer indexKey = ByteBufferUtil.readWithShortLength(in);
+ if (i % expectedIndexInterval == 0)
+ {
+ ByteBuffer summaryKey = ByteBuffer.wrap(indexSummary.getKey(i / expectedIndexInterval));
+ if (!summaryKey.equals(indexKey))
+ return false;
+ summaryEntriesChecked++;
+
+ if (summaryEntriesChecked == Downsampling.BASE_SAMPLING_LEVEL)
+ return true;
+ }
+ RowIndexEntry.Serializer.skip(in);
+ i++;
+ }
+ }
+ catch (IOException e)
+ {
+ markSuspect();
+ throw new CorruptSSTableException(e, path);
+ }
+ }
+
+ return true;
+ }
+
+ /**
+ * Save index summary to Summary.db file.
+ *
+ * @param ibuilder
+ * @param dbuilder
+ */
+
+ public void saveSummary(SegmentedFile.Builder ibuilder, SegmentedFile.Builder dbuilder)
+ {
+ saveSummary(this.descriptor, this.first, this.last, ibuilder, dbuilder, indexSummary);
+ }
+
+ private void saveSummary(SegmentedFile.Builder ibuilder, SegmentedFile.Builder dbuilder, IndexSummary newSummary)
+ {
+ saveSummary(this.descriptor, this.first, this.last, ibuilder, dbuilder, newSummary);
+ }
+ /**
+ * Save index summary to Summary.db file.
+ */
+ public static void saveSummary(Descriptor descriptor, DecoratedKey first, DecoratedKey last,
+ SegmentedFile.Builder ibuilder, SegmentedFile.Builder dbuilder, IndexSummary summary)
+ {
+ File summariesFile = new File(descriptor.filenameFor(Component.SUMMARY));
+ if (summariesFile.exists())
+ FileUtils.deleteWithConfirm(summariesFile);
+
+ try (DataOutputStreamPlus oStream = new BufferedDataOutputStreamPlus(new FileOutputStream(summariesFile));)
+ {
+ IndexSummary.serializer.serialize(summary, oStream, descriptor.version.hasSamplingLevel());
+ ByteBufferUtil.writeWithLength(first.getKey(), oStream);
+ ByteBufferUtil.writeWithLength(last.getKey(), oStream);
+ ibuilder.serializeBounds(oStream);
+ dbuilder.serializeBounds(oStream);
+ }
+ catch (IOException e)
+ {
+ logger.debug("Cannot save SSTable Summary: ", e);
+
+ // corrupted hence delete it and let it load it now.
+ if (summariesFile.exists())
+ FileUtils.deleteWithConfirm(summariesFile);
+ }
+ }
+
+ public void setReplaced()
+ {
+ synchronized (tidy.global)
+ {
+ assert !tidy.isReplaced;
+ tidy.isReplaced = true;
+ }
+ }
+
+ public boolean isReplaced()
+ {
+ synchronized (tidy.global)
+ {
+ return tidy.isReplaced;
+ }
+ }
+
+ // runOnClose must NOT be an anonymous or non-static inner class, nor must it retain a reference chain to this reader
+ public void runOnClose(final Runnable runOnClose)
+ {
+ synchronized (tidy.global)
+ {
+ final Runnable existing = tidy.runOnClose;
+ tidy.runOnClose = AndThen.get(existing, runOnClose);
+ }
+ }
+
+ private static class AndThen implements Runnable
+ {
+ final Runnable runFirst;
+ final Runnable runSecond;
+
+ private AndThen(Runnable runFirst, Runnable runSecond)
+ {
+ this.runFirst = runFirst;
+ this.runSecond = runSecond;
+ }
+
+ public void run()
+ {
+ runFirst.run();
+ runSecond.run();
+ }
+
+ static Runnable get(Runnable runFirst, Runnable runSecond)
+ {
+ if (runFirst == null)
+ return runSecond;
+ return new AndThen(runFirst, runSecond);
+ }
+ }
+
+ /**
+ * Clone this reader with the provided start and open reason, and set the clone as replacement.
+ *
+ * @param newFirst the first key for the replacement (which can be different from the original due to the pre-emptive
+ * opening of compaction results).
+ * @param reason the {@code OpenReason} for the replacement.
+ *
+ * @return the cloned reader. That reader is set as a replacement by the method.
+ */
+ private SSTableReader cloneAndReplace(DecoratedKey newFirst, OpenReason reason)
+ {
+ return cloneAndReplace(newFirst, reason, indexSummary.sharedCopy());
+ }
+
+ /**
+ * Clone this reader with the new values and set the clone as replacement.
+ *
+ * @param newFirst the first key for the replacement (which can be different from the original due to the pre-emptive
+ * opening of compaction results).
+ * @param reason the {@code OpenReason} for the replacement.
+ * @param newSummary the index summary for the replacement.
+ *
+ * @return the cloned reader. That reader is set as a replacement by the method.
+ */
+ private SSTableReader cloneAndReplace(DecoratedKey newFirst, OpenReason reason, IndexSummary newSummary)
+ {
+ SSTableReader replacement = internalOpen(descriptor,
+ components,
+ metadata,
+ partitioner,
- ifile.sharedCopy(),
++ ifile != null ? ifile.sharedCopy() : null,
+ dfile.sharedCopy(),
+ newSummary,
+ bf.sharedCopy(),
+ maxDataAge,
+ sstableMetadata,
+ reason);
+ replacement.first = newFirst;
+ replacement.last = last;
+ replacement.isSuspect.set(isSuspect.get());
+ return replacement;
+ }
+
+ // runOnClose must NOT be an anonymous or non-static inner class, nor must it retain a reference chain to this reader
+ public SSTableReader cloneWithNewStart(DecoratedKey newStart, final Runnable runOnClose)
+ {
+ synchronized (tidy.global)
+ {
+ assert openReason != OpenReason.EARLY;
+ // TODO: merge with caller's firstKeyBeyond() work,to save time
+ if (newStart.compareTo(first) > 0)
+ {
+ final long dataStart = getPosition(newStart, Operator.EQ).position;
+ final long indexStart = getIndexScanPosition(newStart);
+ this.tidy.runOnClose = new DropPageCache(dfile, dataStart, ifile, indexStart, runOnClose);
+ }
+
+ return cloneAndReplace(newStart, OpenReason.MOVED_START);
+ }
+ }
+
+ private static class DropPageCache implements Runnable
+ {
+ final SegmentedFile dfile;
+ final long dfilePosition;
+ final SegmentedFile ifile;
+ final long ifilePosition;
+ final Runnable andThen;
+
+ private DropPageCache(SegmentedFile dfile, long dfilePosition, SegmentedFile ifile, long ifilePosition, Runnable andThen)
+ {
+ this.dfile = dfile;
+ this.dfilePosition = dfilePosition;
+ this.ifile = ifile;
+ this.ifilePosition = ifilePosition;
+ this.andThen = andThen;
+ }
+
+ public void run()
+ {
+ dfile.dropPageCache(dfilePosition);
- ifile.dropPageCache(ifilePosition);
++
++ if (ifile != null)
++ ifile.dropPageCache(ifilePosition);
+ andThen.run();
+ }
+ }
+
+ /**
+ * Returns a new SSTableReader with the same properties as this SSTableReader except that a new IndexSummary will
+ * be built at the target samplingLevel. This (original) SSTableReader instance will be marked as replaced, have
+ * its DeletingTask removed, and have its periodic read-meter sync task cancelled.
+ * @param samplingLevel the desired sampling level for the index summary on the new SSTableReader
+ * @return a new SSTableReader
+ * @throws IOException
+ */
+ @SuppressWarnings("resource")
+ public SSTableReader cloneWithNewSummarySamplingLevel(ColumnFamilyStore parent, int samplingLevel) throws IOException
+ {
+ assert descriptor.version.hasSamplingLevel();
+
+ synchronized (tidy.global)
+ {
+ assert openReason != OpenReason.EARLY;
+
+ int minIndexInterval = metadata.getMinIndexInterval();
+ int maxIndexInterval = metadata.getMaxIndexInterval();
+ double effectiveInterval = indexSummary.getEffectiveIndexInterval();
+
+ IndexSummary newSummary;
+ long oldSize = bytesOnDisk();
+
+ // We have to rebuild the summary from the on-disk primary index in three cases:
+ // 1. The sampling level went up, so we need to read more entries off disk
+ // 2. The min_index_interval changed (in either direction); this changes what entries would be in the summary
+ // at full sampling (and consequently at any other sampling level)
+ // 3. The max_index_interval was lowered, forcing us to raise the sampling level
+ if (samplingLevel > indexSummary.getSamplingLevel() || indexSummary.getMinIndexInterval() != minIndexInterval || effectiveInterval > maxIndexInterval)
+ {
+ newSummary = buildSummaryAtLevel(samplingLevel);
+ }
+ else if (samplingLevel < indexSummary.getSamplingLevel())
+ {
+ // we can use the existing index summary to make a smaller one
+ newSummary = IndexSummaryBuilder.downsample(indexSummary, samplingLevel, minIndexInterval, partitioner);
+
+ try(SegmentedFile.Builder ibuilder = SegmentedFile.getBuilder(DatabaseDescriptor.getIndexAccessMode(), false);
+ SegmentedFile.Builder dbuilder = SegmentedFile.getBuilder(DatabaseDescriptor.getDiskAccessMode(), compression))
+ {
+ saveSummary(ibuilder, dbuilder, newSummary);
+ }
+ }
+ else
+ {
+ throw new AssertionError("Attempted to clone SSTableReader with the same index summary sampling level and " +
+ "no adjustments to min/max_index_interval");
+ }
+
+ long newSize = bytesOnDisk();
+ StorageMetrics.load.inc(newSize - oldSize);
+ parent.metric.liveDiskSpaceUsed.inc(newSize - oldSize);
+
+ return cloneAndReplace(first, OpenReason.METADATA_CHANGE, newSummary);
+ }
+ }
+
+ private IndexSummary buildSummaryAtLevel(int newSamplingLevel) throws IOException
+ {
+ // we read the positions in a BRAF so we don't have to worry about an entry spanning a mmap boundary.
+ RandomAccessReader primaryIndex = RandomAccessReader.open(new File(descriptor.filenameFor(Component.PRIMARY_INDEX)));
+ try
+ {
+ long indexSize = primaryIndex.length();
+ try (IndexSummaryBuilder summaryBuilder = new IndexSummaryBuilder(estimatedKeys(), metadata.getMinIndexInterval(), newSamplingLevel))
+ {
+ long indexPosition;
+ while ((indexPosition = primaryIndex.getFilePointer()) != indexSize)
+ {
+ summaryBuilder.maybeAddEntry(partitioner.decorateKey(ByteBufferUtil.readWithShortLength(primaryIndex)), indexPosition);
+ RowIndexEntry.Serializer.skip(primaryIndex);
+ }
+
+ return summaryBuilder.build(partitioner);
+ }
+ }
+ finally
+ {
+ FileUtils.closeQuietly(primaryIndex);
+ }
+ }
+
+ public RestorableMeter getReadMeter()
+ {
+ return readMeter;
+ }
+
+ public int getIndexSummarySamplingLevel()
+ {
+ return indexSummary.getSamplingLevel();
+ }
+
+ public long getIndexSummaryOffHeapSize()
+ {
+ return indexSummary.getOffHeapSize();
+ }
+
+ public int getMinIndexInterval()
+ {
+ return indexSummary.getMinIndexInterval();
+ }
+
+ public double getEffectiveIndexInterval()
+ {
+ return indexSummary.getEffectiveIndexInterval();
+ }
+
+ public void releaseSummary()
+ {
+ tidy.releaseSummary();
+ indexSummary = null;
+ }
+
+ private void validate()
+ {
+ if (this.first.compareTo(this.last) > 0)
+ {
+ selfRef().release();
+ throw new IllegalStateException(String.format("SSTable first key %s > last key %s", this.first, this.last));
+ }
+ }
+
+ /**
+ * Gets the position in the index file to start scanning to find the given key (at most indexInterval keys away,
+ * modulo downsampling of the index summary). Always returns a value >= 0
+ */
+ public long getIndexScanPosition(RowPosition key)
+ {
+ if (openReason == OpenReason.MOVED_START && key.compareTo(first) < 0)
+ key = first;
+
+ return getIndexScanPositionFromBinarySearchResult(indexSummary.binarySearch(key), indexSummary);
+ }
+
+ @VisibleForTesting
+ public static long getIndexScanPositionFromBinarySearchResult(int binarySearchResult, IndexSummary referencedIndexSummary)
+ {
+ if (binarySearchResult == -1)
+ return 0;
+ else
+ return referencedIndexSummary.getPosition(getIndexSummaryIndexFromBinarySearchResult(binarySearchResult));
+ }
+
+ public static int getIndexSummaryIndexFromBinarySearchResult(int binarySearchResult)
+ {
+ if (binarySearchResult < 0)
+ {
+ // binary search gives us the first index _greater_ than the key searched for,
+ // i.e., its insertion position
+ int greaterThan = (binarySearchResult + 1) * -1;
+ if (greaterThan == 0)
+ return -1;
+ return greaterThan - 1;
+ }
+ else
+ {
+ return binarySearchResult;
+ }
+ }
+
+ /**
+ * Returns the compression metadata for this sstable.
+ * @throws IllegalStateException if the sstable is not compressed
+ */
+ public CompressionMetadata getCompressionMetadata()
+ {
+ if (!compression)
+ throw new IllegalStateException(this + " is not compressed");
+
+ CompressionMetadata cmd = ((ICompressedFile) dfile).getMetadata();
+
+ //We need the parent cf metadata
+ String cfName = metadata.isSecondaryIndex() ? metadata.getParentColumnFamilyName() : metadata.cfName;
+ cmd.parameters.setLiveMetadata(Schema.instance.getCFMetaData(metadata.ksName, cfName));
+
+ return cmd;
+ }
+
+ /**
+ * Returns the amount of memory in bytes used off heap by the compression meta-data.
+ * @return the amount of memory in bytes used off heap by the compression meta-data
+ */
+ public long getCompressionMetadataOffHeapSize()
+ {
+ if (!compression)
+ return 0;
+
+ return getCompressionMetadata().offHeapSize();
+ }
+
+ /**
+ * For testing purposes only.
+ */
+ public void forceFilterFailures()
+ {
+ bf = FilterFactory.AlwaysPresent;
+ }
+
+ public IFilter getBloomFilter()
+ {
+ return bf;
+ }
+
+ public long getBloomFilterSerializedSize()
+ {
+ return bf.serializedSize();
+ }
+
+ /**
+ * Returns the amount of memory in bytes used off heap by the bloom filter.
+ * @return the amount of memory in bytes used off heap by the bloom filter
+ */
+ public long getBloomFilterOffHeapSize()
+ {
+ return bf.offHeapSize();
+ }
+
+ /**
+ * @return An estimate of the number of keys in this SSTable based on the index summary.
+ */
+ public long estimatedKeys()
+ {
+ return indexSummary.getEstimatedKeyCount();
+ }
+
+ /**
+ * @param ranges
+ * @return An estimate of the number of keys for given ranges in this SSTable.
+ */
+ public long estimatedKeysForRanges(Collection<Range<Token>> ranges)
+ {
+ long sampleKeyCount = 0;
+ List<Pair<Integer, Integer>> sampleIndexes = getSampleIndexesForRanges(indexSummary, ranges);
+ for (Pair<Integer, Integer> sampleIndexRange : sampleIndexes)
+ sampleKeyCount += (sampleIndexRange.right - sampleIndexRange.left + 1);
+
+ // adjust for the current sampling level: (BSL / SL) * index_interval_at_full_sampling
+ long estimatedKeys = sampleKeyCount * ((long) Downsampling.BASE_SAMPLING_LEVEL * indexSummary.getMinIndexInterval()) / indexSummary.getSamplingLevel();
+ return Math.max(1, estimatedKeys);
+ }
+
+ /**
+ * Returns the number of entries in the IndexSummary. At full sampling, this is approximately 1/INDEX_INTERVALth of
+ * the keys in this SSTable.
+ */
+ public int getIndexSummarySize()
+ {
+ return indexSummary.size();
+ }
+
+ /**
+ * Returns the approximate number of entries the IndexSummary would contain if it were at full sampling.
+ */
+ public int getMaxIndexSummarySize()
+ {
+ return indexSummary.getMaxNumberOfEntries();
+ }
+
+ /**
+ * Returns the key for the index summary entry at `index`.
+ */
+ public byte[] getIndexSummaryKey(int index)
+ {
+ return indexSummary.getKey(index);
+ }
+
+ private static List<Pair<Integer,Integer>> getSampleIndexesForRanges(IndexSummary summary, Collection<Range<Token>> ranges)
+ {
+ // use the index to determine a minimal section for each range
+ List<Pair<Integer,Integer>> positions = new ArrayList<>();
+
+ for (Range<Token> range : Range.normalize(ranges))
+ {
+ RowPosition leftPosition = range.left.maxKeyBound();
+ RowPosition rightPosition = range.right.maxKeyBound();
+
+ int left = summary.binarySearch(leftPosition);
+ if (left < 0)
+ left = (left + 1) * -1;
+ else
+ // left range are start exclusive
+ left = left + 1;
+ if (left == summary.size())
+ // left is past the end of the sampling
+ continue;
+
+ int right = Range.isWrapAround(range.left, range.right)
+ ? summary.size() - 1
+ : summary.binarySearch(rightPosition);
+ if (right < 0)
+ {
+ // range are end inclusive so we use the previous index from what binarySearch give us
+ // since that will be the last index we will return
+ right = (right + 1) * -1;
+ if (right == 0)
+ // Means the first key is already stricly greater that the right bound
+ continue;
+ right--;
+ }
+
+ if (left > right)
+ // empty range
+ continue;
+ positions.add(Pair.create(left, right));
+ }
+ return positions;
+ }
+
+ public Iterable<DecoratedKey> getKeySamples(final Range<Token> range)
+ {
+ final List<Pair<Integer, Integer>> indexRanges = getSampleIndexesForRanges(indexSummary, Collections.singletonList(range));
+
+ if (indexRanges.isEmpty())
+ return Collections.emptyList();
+
+ return new Iterable<DecoratedKey>()
+ {
+ public Iterator<DecoratedKey> iterator()
+ {
+ return new Iterator<DecoratedKey>()
+ {
+ private Iterator<Pair<Integer, Integer>> rangeIter = indexRanges.iterator();
+ private Pair<Integer, Integer> current;
+ private int idx;
+
+ public boolean hasNext()
+ {
+ if (current == null || idx > current.right)
+ {
+ if (rangeIter.hasNext())
+ {
+ current = rangeIter.next();
+ idx = current.left;
+ return true;
+ }
+ return false;
+ }
+
+ return true;
+ }
+
+ public DecoratedKey next()
+ {
+ byte[] bytes = indexSummary.getKey(idx++);
+ return partitioner.decorateKey(ByteBuffer.wrap(bytes));
+ }
+
+ public void remove()
+ {
+ throw new UnsupportedOperationException();
+ }
+ };
+ }
+ };
+ }
+
+ /**
+ * Determine the minimal set of sections that can be extracted from this SSTable to cover the given ranges.
+ * @return A sorted list of (offset,end) pairs that cover the given ranges in the datafile for this SSTable.
+ */
+ public List<Pair<Long,Long>> getPositionsForRanges(Collection<Range<Token>> ranges)
+ {
+ // use the index to determine a minimal section for each range
+ List<Pair<Long,Long>> positions = new ArrayList<>();
+ for (Range<Token> range : Range.normalize(ranges))
+ {
+ assert !range.isWrapAround() || range.right.isMinimum();
+ // truncate the range so it at most covers the sstable
+ AbstractBounds<RowPosition> bounds = Range.makeRowRange(range);
+ RowPosition leftBound = bounds.left.compareTo(first) > 0 ? bounds.left : first.getToken().minKeyBound();
+ RowPosition rightBound = bounds.right.isMinimum() ? last.getToken().maxKeyBound() : bounds.right;
+
+ if (leftBound.compareTo(last) > 0 || rightBound.compareTo(first) < 0)
+ continue;
+
+ long left = getPosition(leftBound, Operator.GT).position;
+ long right = (rightBound.compareTo(last) > 0)
+ ? uncompressedLength()
+ : getPosition(rightBound, Operator.GT).position;
+
+ if (left == right)
+ // empty range
+ continue;
+
+ assert left < right : String.format("Range=%s openReason=%s first=%s last=%s left=%d right=%d", range, openReason, first, last, left, right);
+ positions.add(Pair.create(left, right));
+ }
+ return positions;
+ }
+
+ public KeyCacheKey getCacheKey(DecoratedKey key)
+ {
+ return new KeyCacheKey(metadata.cfId, descriptor, key.getKey());
+ }
+
+ public void cacheKey(DecoratedKey key, RowIndexEntry info)
+ {
+ CachingOptions caching = metadata.getCaching();
+
+ if (!caching.keyCache.isEnabled()
+ || keyCache == null
+ || keyCache.getCapacity() == 0)
+ {
+ return;
+ }
+
+ KeyCacheKey cacheKey = new KeyCacheKey(metadata.cfId, descriptor, key.getKey());
+ logger.trace("Adding cache entry for {} -> {}", cacheKey, info);
+ keyCache.put(cacheKey, info);
+ }
+
+ public RowIndexEntry getCachedPosition(DecoratedKey key, boolean updateStats)
+ {
+ return getCachedPosition(new KeyCacheKey(metadata.cfId, descriptor, key.getKey()), updateStats);
+ }
+
+ protected RowIndexEntry getCachedPosition(KeyCacheKey unifiedKey, boolean updateStats)
+ {
+ if (keyCache != null && keyCache.getCapacity() > 0) {
+ if (updateStats)
+ {
+ RowIndexEntry cachedEntry = keyCache.get(unifiedKey);
+ keyCacheRequest.incrementAndGet();
+ if (cachedEntry != null)
+ {
+ keyCacheHit.incrementAndGet();
+ bloomFilterTracker.addTruePositive();
+ }
+ return cachedEntry;
+ }
+ else
+ {
+ return keyCache.getInternal(unifiedKey);
+ }
+ }
+ return null;
+ }
+
+ /**
+ * Get position updating key cache and stats.
+ * @see #getPosition(org.apache.cassandra.db.RowPosition, SSTableReader.Operator, boolean)
+ */
+ public RowIndexEntry getPosition(RowPosition key, Operator op)
+ {
+ return getPosition(key, op, true, false);
+ }
+
+ public RowIndexEntry getPosition(RowPosition key, Operator op, boolean updateCacheAndStats)
+ {
+ return getPosition(key, op, updateCacheAndStats, false);
+ }
+ /**
+ * @param key The key to apply as the rhs to the given Operator. A 'fake' key is allowed to
+ * allow key selection by token bounds but only if op != * EQ
+ * @param op The Operator defining matching keys: the nearest key to the target matching the operator wins.
+ * @param updateCacheAndStats true if updating stats and cache
+ * @return The index entry corresponding to the key, or null if the key is not present
+ */
+ protected abstract RowIndexEntry getPosition(RowPosition key, Operator op, boolean updateCacheAndStats, boolean permitMatchPastLast);
+
+ //Corresponds to a name column
+ public abstract OnDiskAtomIterator iterator(DecoratedKey key, SortedSet<CellName> columns);
+ public abstract OnDiskAtomIterator iterator(FileDataInput file, DecoratedKey key, SortedSet<CellName> columns, RowIndexEntry indexEntry);
+
+ //Corresponds to a slice query
+ public abstract OnDiskAtomIterator iterator(DecoratedKey key, ColumnSlice[] slices, boolean reverse);
+ public abstract OnDiskAtomIterator iterator(FileDataInput file, DecoratedKey key, ColumnSlice[] slices, boolean reversed, RowIndexEntry indexEntry);
+
+ /**
+ * Finds and returns the first key beyond a given token in this SSTable or null if no such key exists.
+ */
+ public DecoratedKey firstKeyBeyond(RowPosition token)
+ {
+ if (token.compareTo(first) < 0)
+ return first;
+
+ long sampledPosition = getIndexScanPosition(token);
+
++ if (ifile == null)
++ return null;
++
+ Iterator<FileDataInput> segments = ifile.iterator(sampledPosition);
+ while (segments.hasNext())
+ {
+ String path = null;
+ try (FileDataInput in = segments.next();)
+ {
+ path = in.getPath();
+ while (!in.isEOF())
+ {
+ ByteBuffer indexKey = ByteBufferUtil.readWithShortLength(in);
+ DecoratedKey indexDecoratedKey = partitioner.decorateKey(indexKey);
+ if (indexDecoratedKey.compareTo(token) > 0)
+ return indexDecoratedKey;
+
+ RowIndexEntry.Serializer.skip(in);
+ }
+ }
+ catch (IOException e)
+ {
+ markSuspect();
+ throw new CorruptSSTableException(e, path);
+ }
+ }
+
+ return null;
+ }
+
+ /**
+ * @return The length in bytes of the data for this SSTable. For
+ * compressed files, this is not the same thing as the on disk size (see
+ * onDiskLength())
+ */
+ public long uncompressedLength()
+ {
+ return dfile.length;
+ }
+
+ /**
+ * @return The length in bytes of the on disk size for this SSTable. For
+ * compressed files, this is not the same thing as the data length (see
+ * length())
+ */
+ public long onDiskLength()
+ {
+ return dfile.onDiskLength;
+ }
+
+ /**
+ * Mark the sstable as obsolete, i.e., compacted into newer sstables.
+ *
+ * When calling this function, the caller must ensure that the SSTableReader is not referenced anywhere
+ * except for threads holding a reference.
+ *
+ * @return true if the this is the first time the file was marked obsolete. Calling this
+ * multiple times is usually buggy (see exceptions in Tracker.unmarkCompacting and removeOldSSTablesSize).
+ */
+ public boolean markObsolete(Tracker tracker)
+ {
+ if (logger.isDebugEnabled())
+ logger.debug("Marking {} compacted", getFilename());
+
+ synchronized (tidy.global)
+ {
+ assert !tidy.isReplaced;
+ }
+ if (!tidy.global.isCompacted.getAndSet(true))
+ {
+ tidy.type.markObsolete(this, tracker);
+ return true;
+ }
+ return false;
+ }
+
+ public boolean isMarkedCompacted()
+ {
+ return tidy.global.isCompacted.get();
+ }
+
+ public void markSuspect()
+ {
+ if (logger.isDebugEnabled())
+ logger.debug("Marking {} as a suspect for blacklisting.", getFilename());
+
+ isSuspect.getAndSet(true);
+ }
+
+ public boolean isMarkedSuspect()
+ {
+ return isSuspect.get();
+ }
+
+
+ /**
+ * I/O SSTableScanner
+ * @return A Scanner for seeking over the rows of the SSTable.
+ */
+ public ISSTableScanner getScanner()
+ {
+ return getScanner((RateLimiter) null);
+ }
+
+ public ISSTableScanner getScanner(RateLimiter limiter)
+ {
+ return getScanner(DataRange.allData(partitioner), limiter);
+ }
+
+ /**
+ *
+ * @param dataRange filter to use when reading the columns
+ * @return A Scanner for seeking over the rows of the SSTable.
+ */
+ public ISSTableScanner getScanner(DataRange dataRange)
+ {
+ return getScanner(dataRange, null);
+ }
+
+ /**
+ * Direct I/O SSTableScanner over a defined range of tokens.
+ *
+ * @param range the range of keys to cover
+ * @return A Scanner for seeking over the rows of the SSTable.
+ */
+ public ISSTableScanner getScanner(Range<Token> range, RateLimiter limiter)
+ {
+ if (range == null)
+ return getScanner(limiter);
+ return getScanner(Collections.singletonList(range), limiter);
+ }
+
+ /**
+ * Direct I/O SSTableScanner over a defined collection of ranges of tokens.
+ *
+ * @param ranges the range of keys to cover
+ * @return A Scanner for seeking over the rows of the SSTable.
+ */
+ public abstract ISSTableScanner getScanner(Collection<Range<Token>> ranges, RateLimiter limiter);
+
+ /**
+ *
+ * @param dataRange filter to use when reading the columns
+ * @return A Scanner for seeking over the rows of the SSTable.
+ */
+ public abstract ISSTableScanner getScanner(DataRange dataRange, RateLimiter limiter);
+
+
+
+ public FileDataInput getFileDataInput(long position)
+ {
+ return dfile.getSegment(position);
+ }
+
+ /**
+ * Tests if the sstable contains data newer than the given age param (in localhost currentMilli time).
+ * This works in conjunction with maxDataAge which is an upper bound on the create of data in this sstable.
+ * @param age The age to compare the maxDataAre of this sstable. Measured in millisec since epoc on this host
+ * @return True iff this sstable contains data that's newer than the given age parameter.
+ */
+ public boolean newSince(long age)
+ {
+ return maxDataAge > age;
+ }
+
+ public void createLinks(String snapshotDirectoryPath)
+ {
+ for (Component component : components)
+ {
+ File sourceFile = new File(descriptor.filenameFor(component));
+ File targetLink = new File(snapshotDirectoryPath, sourceFile.getName());
+ FileUtils.createHardLink(sourceFile, targetLink);
+ }
+ }
+
+ public boolean isRepaired()
+ {
+ return sstableMetadata.repairedAt != ActiveRepairService.UNREPAIRED_SSTABLE;
+ }
+
+ /**
+ * TODO: Move someplace reusable
+ */
+ public abstract static class Operator
+ {
+ public static final Operator EQ = new Equals();
+ public static final Operator GE = new GreaterThanOrEqualTo();
+ public static final Operator GT = new GreaterThan();
+
+ /**
+ * @param comparison The result of a call to compare/compareTo, with the desired field on the rhs.
+ * @return less than 0 if the operator cannot match forward, 0 if it matches, greater than 0 if it might match forward.
+ */
+ public abstract int apply(int comparison);
+
+ final static class Equals extends Operator
+ {
+ public int apply(int comparison) { return -comparison; }
+ }
+
+ final static class GreaterThanOrEqualTo extends Operator
+ {
+ public int apply(int comparison) { return comparison >= 0 ? 0 : 1; }
+ }
+
+ final static class GreaterThan extends Operator
+ {
+ public int apply(int comparison) { return comparison > 0 ? 0 : 1; }
+ }
+ }
+
+ public long getBloomFilterFalsePositiveCount()
+ {
+ return bloomFilterTracker.getFalsePositiveCount();
+ }
+
+ public long getRecentBloomFilterFalsePositiveCount()
+ {
+ return bloomFilterTracker.getRecentFalsePositiveCount();
+ }
+
+ public long getBloomFilterTruePositiveCount()
+ {
+ return bloomFilterTracker.getTruePositiveCount();
+ }
+
+ public long getRecentBloomFilterTruePositiveCount()
+ {
+ return bloomFilterTracker.getRecentTruePositiveCount();
+ }
+
+ public InstrumentingCache<KeyCacheKey, RowIndexEntry> getKeyCache()
+ {
+ return keyCache;
+ }
+
+ public EstimatedHistogram getEstimatedRowSize()
+ {
+ return sstableMetadata.estimatedRowSize;
+ }
+
+ public EstimatedHistogram getEstimatedColumnCount()
+ {
+ return sstableMetadata.estimatedColumnCount;
+ }
+
+ public double getEstimatedDroppableTombstoneRatio(int gcBefore)
+ {
+ return sstableMetadata.getEstimatedDroppableTombstoneRatio(gcBefore);
+ }
+
+ public double getDroppableTombstonesBefore(int gcBefore)
+ {
+ return sstableMetadata.getDroppableTombstonesBefore(gcBefore);
+ }
+
+ public double getCompressionRatio()
+ {
+ return sstableMetadata.compressionRatio;
+ }
+
+ public ReplayPosition getReplayPosition()
+ {
+ return sstableMetadata.replayPosition;
+ }
+
+ public long getMinTimestamp()
+ {
+ return sstableMetadata.minTimestamp;
+ }
+
+ public long getMaxTimestamp()
+ {
+ return sstableMetadata.maxTimestamp;
+ }
+
+ public Set<Integer> getAncestors()
+ {
+ try
+ {
+ CompactionMetadata compactionMetadata = (CompactionMetadata) descriptor.getMetadataSerializer().deserialize(descriptor, MetadataType.COMPACTION);
- return compactionMetadata.ancestors;
++ if (compactionMetadata != null)
++ return compactionMetadata.ancestors;
++ return Collections.emptySet();
+ }
+ catch (IOException e)
+ {
+ SSTableReader.logOpenException(descriptor, e);
+ return Collections.emptySet();
+ }
+ }
+
+ public int getSSTableLevel()
+ {
+ return sstableMetadata.sstableLevel;
+ }
+
+ /**
+ * Reloads the sstable metadata from disk.
+ *
+ * Called after level is changed on sstable, for example if the sstable is dropped to L0
+ *
+ * Might be possible to remove in future versions
+ *
+ * @throws IOException
+ */
+ public void reloadSSTableMetadata() throws IOException
+ {
+ this.sstableMetadata = (StatsMetadata) descriptor.getMetadataSerializer().deserialize(descriptor, MetadataType.STATS);
+ }
+
+ public StatsMetadata getSSTableMetadata()
+ {
+ return sstableMetadata;
+ }
+
+ public RandomAccessReader openDataReader(RateLimiter limiter)
+ {
+ assert limiter != null;
+ return dfile.createThrottledReader(limiter);
+ }
+
+ public RandomAccessReader openDataReader()
+ {
+ return dfile.createReader();
+ }
+
+ public RandomAccessReader openIndexReader()
+ {
- return ifile.createReader();
++ if (ifile != null)
++ return ifile.createReader();
++ return null;
+ }
+
+ /**
+ * @param component component to get timestamp.
+ * @return last modified time for given component. 0 if given component does not exist or IO error occurs.
+ */
+ public long getCreationTimeFor(Component component)
+ {
+ return new File(descriptor.filenameFor(component)).lastModified();
+ }
+
+ /**
+ * @return Number of key cache hit
+ */
+ public long getKeyCacheHit()
+ {
+ return keyCacheHit.get();
+ }
+
+ /**
+ * @return Number of key cache request
+ */
+ public long getKeyCacheRequest()
+ {
+ return keyCacheRequest.get();
+ }
+
+ /**
+ * Increment the total row read count and read rate for this SSTable. This should not be incremented for range
+ * slice queries, row cache hits, or non-query reads, like compaction.
+ */
+ public void incrementReadCount()
+ {
+ if (readMeter != null)
+ readMeter.mark();
+ }
+
+ public static class SizeComparator implements Comparator<SSTableReader>
+ {
+ public int compare(SSTableReader o1, SSTableReader o2)
+ {
+ return Longs.compare(o1.onDiskLength(), o2.onDiskLength());
+ }
+ }
+
+ public Ref<SSTableReader> tryRef()
+ {
+ return selfRef.tryRef();
+ }
+
+ public Ref<SSTableReader> selfRef()
+ {
+ return selfRef;
+ }
+
+ public Ref<SSTableReader> ref()
+ {
+ return selfRef.ref();
+ }
+
+ void setup(boolean isOffline)
+ {
+ tidy.setup(this, isOffline);
+ this.readMeter = tidy.global.readMeter;
+ }
+
+ @VisibleForTesting
+ public void overrideReadMeter(RestorableMeter readMeter)
+ {
+ this.readMeter = tidy.global.readMeter = readMeter;
+ }
+
+ /**
+ * One instance per SSTableReader we create. This references the type-shared tidy, which in turn references
+ * the globally shared tidy, i.e.
+ *
+ * InstanceTidier => DescriptorTypeTitdy => GlobalTidy
+ *
+ * We can create many InstanceTidiers (one for every time we reopen an sstable with MOVED_START for example), but there can only be
+ * two DescriptorTypeTidy (FINAL and TEMPLINK) and only one GlobalTidy for one single logical sstable.
+ *
+ * When the InstanceTidier cleansup, it releases its reference to its DescriptorTypeTidy; when all InstanceTidiers
+ * for that type have run, the DescriptorTypeTidy cleansup. DescriptorTypeTidy behaves in the same way towards GlobalTidy.
+ *
+ * For ease, we stash a direct reference to both our type-shared and global tidier
+ */
+ private static final class InstanceTidier implements Tidy
+ {
+ private final Descriptor descriptor;
+ private final CFMetaData metadata;
+ private IFilter bf;
+ private IndexSummary summary;
+
+ private SegmentedFile dfile;
+ private SegmentedFile ifile;
+ private Runnable runOnClose;
+ private boolean isReplaced = false;
+
+ // a reference to our shared per-Descriptor.Type tidy instance, that
+ // we will release when we are ourselves released
+ private Ref<DescriptorTypeTidy> typeRef;
+
+ // a convenience stashing of the shared per-descriptor-type tidy instance itself
+ // and the per-logical-sstable globally shared state that it is linked to
+ private DescriptorTypeTidy type;
+ private GlobalTidy global;
+
+ private boolean setup;
+
+ void setup(SSTableReader reader, boolean isOffline)
+ {
+ this.setup = true;
+ this.bf = reader.bf;
+ this.summary = reader.indexSummary;
+ this.dfile = reader.dfile;
+ this.ifile = reader.ifile;
+ // get a new reference to the shared descriptor-type tidy
+ this.typeRef = DescriptorTypeTidy.get(reader);
+ this.type = typeRef.get();
+ this.global = type.globalRef.get();
+ if (!isOffline)
+ global.ensureReadMeter();
+ }
+
+ InstanceTidier(Descriptor descriptor, CFMetaData metadata)
+ {
+ this.descriptor = descriptor;
+ this.metadata = metadata;
+ }
+
+ public void tidy()
+ {
+ // don't try to cleanup if the sstablereader was never fully constructed
+ if (!setup)
+ return;
+
+ final ColumnFamilyStore cfs = Schema.instance.getColumnFamilyStoreInstance(metadata.cfId);
+ final OpOrder.Barrier barrier;
+ if (cfs != null)
+ {
+ barrier = cfs.readOrdering.newBarrier();
+ barrier.issue();
+ }
+ else
+ barrier = null;
+
+ ScheduledExecutors.nonPeriodicTasks.execute(new Runnable()
+ {
+ public void run()
+ {
+ if (barrier != null)
+ barrier.await();
- bf.close();
++ if (bf != null)
++ bf.close();
+ if (summary != null)
+ summary.close();
+ if (runOnClose != null)
+ runOnClose.run();
+ dfile.close();
- ifile.close();
++ if (ifile != null)
++ ifile.close();
+ typeRef.release();
+ }
+ });
+ }
+
+ public String name()
+ {
+ return descriptor.toString();
+ }
+
+ void releaseSummary()
+ {
+ summary.close();
+ assert summary.isCleanedUp();
+ summary = null;
+ }
+ }
+
+ /**
+ * One shared between all instances of a given Descriptor.Type.
+ * Performs only two things: the deletion of the sstables for the type,
+ * if necessary; and the shared reference to the globally shared state.
+ *
+ * All InstanceTidiers, on setup(), ask the static get() method for their shared state,
+ * and stash a reference to it to be released when they are. Once all such references are
+ * released, the shared tidy will be performed.
+ */
+ static final class DescriptorTypeTidy implements Tidy
+ {
+ // keyed by REAL descriptor (TMPLINK/FINAL), mapping to the shared DescriptorTypeTidy for that descriptor
+ static final ConcurrentMap<Descriptor, Ref<DescriptorTypeTidy>> lookup = new ConcurrentHashMap<>();
+
+ private final Descriptor desc;
+ private final Ref<GlobalTidy> globalRef;
+ private final Set<Component> components;
+ private long sizeOnDelete;
+ private Counter totalDiskSpaceUsed;
+
+ DescriptorTypeTidy(Descriptor desc, SSTableReader sstable)
+ {
+ this.desc = desc;
+ // get a new reference to the shared global tidy
+ this.globalRef = GlobalTidy.get(sstable);
+ this.components = sstable.components;
+ }
+
+ void markObsolete(SSTableReader instance, Tracker tracker)
+ {
+ // the tracker is used only to notify listeners of deletion of the sstable;
+ // since deletion of a non-final file is not really deletion of the sstable,
+ // we don't want to notify the listeners in this event
+ if (tracker != null && tracker.cfstore != null && desc.type == Descriptor.Type.FINAL)
+ {
+ sizeOnDelete = instance.bytesOnDisk();
+ totalDiskSpaceUsed = tracker.cfstore.metric.totalDiskSpaceUsed;
+ tracker.notifyDeleting(instance);
+ }
+ }
+
+ public void tidy()
+ {
+ lookup.remove(desc);
+ boolean isCompacted = globalRef.get().isCompacted.get();
+ globalRef.release();
+ switch (desc.type)
+ {
+ case FINAL:
+ if (isCompacted)
+ new SSTableDeletingTask(desc, components, totalDiskSpaceUsed, sizeOnDelete).run();
+ break;
+ case TEMPLINK:
+ new SSTableDeletingTask(desc, components, null, 0).run();
+ break;
+ default:
+ throw new IllegalStateException();
+ }
+ }
+
+ public String name()
+ {
+ return desc.toString();
+ }
+
+ // get a new reference to the shared DescriptorTypeTidy for this sstable
+ @SuppressWarnings("resource")
+ public static Ref<DescriptorTypeTidy> get(SSTableReader sstable)
+ {
+ Descriptor desc = sstable.descriptor;
+ if (sstable.openReason == OpenReason.EARLY)
+ desc = desc.asType(Descriptor.Type.TEMPLINK);
+ Ref<DescriptorTypeTidy> refc = lookup.get(desc);
+ if (refc != null)
+ return refc.ref();
+ final DescriptorTypeTidy tidy = new DescriptorTypeTidy(desc, sstable);
+ refc = new Ref<>(tidy, tidy);
+ Ref<?> ex = lookup.putIfAbsent(desc, refc);
+ if (ex != null)
+ {
+ refc.close();
+ throw new AssertionError();
+ }
+ return refc;
+ }
+ }
+
+ /**
+ * One instance per logical sstable. This both tracks shared cleanup and some shared state related
+ * to the sstable's lifecycle. All DescriptorTypeTidy instances, on construction, obtain a reference to us
+ * via our static get(). There should only ever be at most two such references extant at any one time,
+ * since only TMPLINK and FINAL type descriptors should be open as readers. When all files of both
+ * kinds have been released, this shared tidy will be performed.
+ */
+ static final class GlobalTidy implements Tidy
+ {
+ // keyed by FINAL descriptor, mapping to the shared GlobalTidy for that descriptor
+ static final ConcurrentMap<Descriptor, Ref<GlobalTidy>> lookup = new ConcurrentHashMap<>();
+
+ private final Descriptor desc;
+ // the readMeter that is shared between all instances of the sstable, and can be overridden in all of them
+ // at once also, for testing purposes
+ private RestorableMeter readMeter;
+ // the scheduled persistence of the readMeter, that we will cancel once all instances of this logical
+ /
<TRUNCATED>
http://git-wip-us.apache.org/repos/asf/cassandra/blob/ebe18bb2/src/java/org/apache/cassandra/tools/StandaloneScrubber.java
----------------------------------------------------------------------
http://git-wip-us.apache.org/repos/asf/cassandra/blob/ebe18bb2/test/unit/org/apache/cassandra/db/ScrubTest.java
----------------------------------------------------------------------
[11/12] cassandra git commit: Merge branch 'cassandra-2.1' into
cassandra-2.2
Posted by be...@apache.org.
Merge branch 'cassandra-2.1' into cassandra-2.2
Conflicts:
CHANGES.txt
src/java/org/apache/cassandra/db/compaction/Scrubber.java
src/java/org/apache/cassandra/io/sstable/format/SSTableReader.java
Project: http://git-wip-us.apache.org/repos/asf/cassandra/repo
Commit: http://git-wip-us.apache.org/repos/asf/cassandra/commit/ebe18bb2
Tree: http://git-wip-us.apache.org/repos/asf/cassandra/tree/ebe18bb2
Diff: http://git-wip-us.apache.org/repos/asf/cassandra/diff/ebe18bb2
Branch: refs/heads/trunk
Commit: ebe18bb2ff62602fd5f55b969ecf665d2d3e5ace
Parents: 7d31068 4c94ef2
Author: Benedict Elliott Smith <be...@apache.org>
Authored: Tue Jul 7 16:28:19 2015 +0100
Committer: Benedict Elliott Smith <be...@apache.org>
Committed: Tue Jul 7 16:28:19 2015 +0100
----------------------------------------------------------------------
CHANGES.txt | 5 ++
.../cassandra/db/compaction/Scrubber.java | 38 +++++++++++----
.../io/sstable/format/SSTableReader.java | 50 ++++++++++++++------
.../cassandra/tools/StandaloneScrubber.java | 2 +-
.../unit/org/apache/cassandra/db/ScrubTest.java | 25 ++++++++++
5 files changed, 95 insertions(+), 25 deletions(-)
----------------------------------------------------------------------
http://git-wip-us.apache.org/repos/asf/cassandra/blob/ebe18bb2/CHANGES.txt
----------------------------------------------------------------------
diff --cc CHANGES.txt
index 7c6b4ad,2cbc7c4..a863ad8
--- a/CHANGES.txt
+++ b/CHANGES.txt
@@@ -1,28 -1,12 +1,33 @@@
-2.1.9
++2.2.0-rc3
+ Merged from 2.0:
- * Scrub (recover) sstables even when -Index.db is missing, (CASSANDRA-9591)
++ * Scrub (recover) sstables even when -Index.db is missing (CASSANDRA-9591)
+
+
-2.1.8
+2.2.0-rc2
+ * Re-enable memory-mapped I/O on Windows (CASSANDRA-9658)
+ * Warn when an extra-large partition is compacted (CASSANDRA-9643)
+ * (cqlsh) Allow setting the initial connection timeout (CASSANDRA-9601)
+ * BulkLoader has --transport-factory option but does not use it (CASSANDRA-9675)
+ * Allow JMX over SSL directly from nodetool (CASSANDRA-9090)
+ * Update cqlsh for UDFs (CASSANDRA-7556)
+ * Change Windows kernel default timer resolution (CASSANDRA-9634)
+ * Deprected sstable2json and json2sstable (CASSANDRA-9618)
+ * Allow native functions in user-defined aggregates (CASSANDRA-9542)
+ * Don't repair system_distributed by default (CASSANDRA-9621)
+ * Fix mixing min, max, and count aggregates for blob type (CASSANRA-9622)
+ * Rename class for DATE type in Java driver (CASSANDRA-9563)
+ * Duplicate compilation of UDFs on coordinator (CASSANDRA-9475)
+ * Fix connection leak in CqlRecordWriter (CASSANDRA-9576)
+ * Mlockall before opening system sstables & remove boot_without_jna option (CASSANDRA-9573)
+ * Add functions to convert timeuuid to date or time, deprecate dateOf and unixTimestampOf (CASSANDRA-9229)
+ * Make sure we cancel non-compacting sstables from LifecycleTransaction (CASSANDRA-9566)
+ * Fix deprecated repair JMX API (CASSANDRA-9570)
+ * Add logback metrics (CASSANDRA-9378)
+ * Update and refactor ant test/test-compression to run the tests in parallel (CASSANDRA-9583)
+ * Fix upgrading to new directory for secondary index (CASSANDRA-9687)
+Merged from 2.1:
* (cqlsh) Fix bad check for CQL compatibility when DESCRIBE'ing
COMPACT STORAGE tables with no clustering columns
- * Warn when an extra-large partition is compacted (CASSANDRA-9643)
* Eliminate strong self-reference chains in sstable ref tidiers (CASSANDRA-9656)
* Ensure StreamSession uses canonical sstable reader instances (CASSANDRA-9700)
* Ensure memtable book keeping is not corrupted in the event we shrink usage (CASSANDRA-9681)
http://git-wip-us.apache.org/repos/asf/cassandra/blob/ebe18bb2/src/java/org/apache/cassandra/db/compaction/Scrubber.java
----------------------------------------------------------------------
diff --cc src/java/org/apache/cassandra/db/compaction/Scrubber.java
index 10952e7,b1c12e0..5a0b354
--- a/src/java/org/apache/cassandra/db/compaction/Scrubber.java
+++ b/src/java/org/apache/cassandra/db/compaction/Scrubber.java
@@@ -109,9 -101,17 +109,18 @@@ public class Scrubber implements Closea
? new ScrubController(cfs)
: new CompactionController(cfs, Collections.singleton(sstable), CompactionManager.getDefaultGcBefore(cfs));
this.isCommutative = cfs.metadata.isCounter();
+
+ boolean hasIndexFile = (new File(sstable.descriptor.filenameFor(Component.PRIMARY_INDEX))).exists();
+ this.isIndex = cfs.isIndex();
+ if (!hasIndexFile)
+ {
+ // if there's any corruption in the -Data.db then rows can't be skipped over. but it's worth a shot.
+ outputHandler.warn("Missing component: " + sstable.descriptor.filenameFor(Component.PRIMARY_INDEX));
+ }
-
+ this.checkData = checkData && !this.isIndex; //LocalByPartitionerType does not support validation
- this.expectedBloomFilterSize = Math.max(cfs.metadata.getMinIndexInterval(), (int)(SSTableReader.getApproximateKeyCount(toScrub)));
+ this.expectedBloomFilterSize = Math.max(
+ cfs.metadata.getMinIndexInterval(),
+ hasIndexFile ? (int)(SSTableReader.getApproximateKeyCount(toScrub)) : 0);
// loop through each row, deserializing to check for damage.
// we'll also loop through the index at the same time, using the position from the index to recover if the
@@@ -130,12 -134,15 +143,13 @@@
public void scrub()
{
outputHandler.output(String.format("Scrubbing %s (%s bytes)", sstable, dataFile.length()));
- try (SSTableRewriter writer = new SSTableRewriter(cfs, transaction, sstable.maxDataAge, isOffline);)
- Set<SSTableReader> oldSSTable = Sets.newHashSet(sstable);
- SSTableRewriter writer = new SSTableRewriter(cfs, oldSSTable, sstable.maxDataAge, isOffline);
- try
++ try (SSTableRewriter writer = new SSTableRewriter(cfs, transaction, sstable.maxDataAge, isOffline))
{
- nextIndexKey = ByteBufferUtil.readWithShortLength(indexFile);
+ nextIndexKey = indexAvailable() ? ByteBufferUtil.readWithShortLength(indexFile) : null;
+ if (indexAvailable())
{
// throw away variable so we don't have a side effect in the assert
- long firstRowPositionFromIndex = sstable.metadata.comparator.rowIndexEntrySerializer().deserialize(indexFile, sstable.descriptor.version).position;
+ long firstRowPositionFromIndex = rowIndexEntrySerializer.deserialize(indexFile, sstable.descriptor.version).position;
assert firstRowPositionFromIndex == 0 : firstRowPositionFromIndex;
}
@@@ -174,11 -182,12 +188,11 @@@
dataSizeFromIndex = nextRowPositionFromIndex - dataStartFromIndex;
}
- dataSize = dataSizeFromIndex;
// avoid an NPE if key is null
String keyName = key == null ? "(unreadable key)" : ByteBufferUtil.bytesToHex(key.getKey());
- outputHandler.debug(String.format("row %s is %s bytes", keyName, dataSize));
+ outputHandler.debug(String.format("row %s is %s bytes", keyName, dataSizeFromIndex));
- assert currentIndexKey != null || indexFile.isEOF();
+ assert currentIndexKey != null || !indexAvailable();
try
{
@@@ -191,14 -200,16 +205,14 @@@
ByteBufferUtil.bytesToHex(key.getKey()), ByteBufferUtil.bytesToHex(currentIndexKey))));
}
- if (dataSizeFromIndex > dataFile.length())
- if (dataSize > dataFile.length())
- throw new IOError(new IOException("Impossible row size (greater than file length): " + dataSize));
++ if (indexFile != null && dataSizeFromIndex > dataFile.length())
+ throw new IOError(new IOException("Impossible row size (greater than file length): " + dataSizeFromIndex));
- if (dataStart != dataStartFromIndex)
+ if (indexFile != null && dataStart != dataStartFromIndex)
outputHandler.warn(String.format("Data file row position %d differs from index file row position %d", dataStart, dataStartFromIndex));
- if (indexFile != null && dataSize != dataSizeFromIndex)
- outputHandler.warn(String.format("Data file row size %d different from index file row size %d", dataSize, dataSizeFromIndex));
+ SSTableIdentityIterator atoms = new SSTableIdentityIterator(sstable, dataFile, key, checkData);
- SSTableIdentityIterator atoms = new SSTableIdentityIterator(sstable, dataFile, key, dataSize, validateColumns);
if (prevKey != null && prevKey.compareTo(key) > 0)
{
saveOutOfOrderRow(prevKey, key, atoms);
@@@ -316,10 -327,11 +330,11 @@@
currentRowPositionFromIndex = nextRowPositionFromIndex;
try
{
- nextIndexKey = indexFile.isEOF() ? null : ByteBufferUtil.readWithShortLength(indexFile);
- nextRowPositionFromIndex = indexFile.isEOF()
+ nextIndexKey = !indexAvailable() ? null : ByteBufferUtil.readWithShortLength(indexFile);
+
+ nextRowPositionFromIndex = !indexAvailable()
? dataFile.length()
- : sstable.metadata.comparator.rowIndexEntrySerializer().deserialize(indexFile, sstable.descriptor.version).position;
+ : rowIndexEntrySerializer.deserialize(indexFile, sstable.descriptor.version).position;
}
catch (Throwable th)
{
[08/12] cassandra git commit: Merge branch 'cassandra-2.1' into
cassandra-2.2
Posted by be...@apache.org.
http://git-wip-us.apache.org/repos/asf/cassandra/blob/ebe18bb2/src/java/org/apache/cassandra/io/sstable/format/SSTableReader.java
----------------------------------------------------------------------
diff --cc src/java/org/apache/cassandra/io/sstable/format/SSTableReader.java
index 0cc425a,0000000..247d181
mode 100644,000000..100644
--- a/src/java/org/apache/cassandra/io/sstable/format/SSTableReader.java
+++ b/src/java/org/apache/cassandra/io/sstable/format/SSTableReader.java
@@@ -1,2218 -1,0 +1,2238 @@@
+/*
+ * Licensed to the Apache Software Foundation (ASF) under one
+ * or more contributor license agreements. See the NOTICE file
+ * distributed with this work for additional information
+ * regarding copyright ownership. The ASF licenses this file
+ * to you under the Apache License, Version 2.0 (the
+ * "License"); you may not use this file except in compliance
+ * with the License. You may obtain a copy of the License at
+ *
+ * http://www.apache.org/licenses/LICENSE-2.0
+ *
+ * Unless required by applicable law or agreed to in writing, software
+ * distributed under the License is distributed on an "AS IS" BASIS,
+ * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ * See the License for the specific language governing permissions and
+ * limitations under the License.
+ */
+package org.apache.cassandra.io.sstable.format;
+
+import java.io.*;
+import java.nio.ByteBuffer;
+import java.util.*;
+import java.util.concurrent.*;
+import java.util.concurrent.atomic.AtomicBoolean;
+import java.util.concurrent.atomic.AtomicLong;
+
+import com.google.common.annotations.VisibleForTesting;
+import com.google.common.base.Predicate;
+import com.google.common.collect.Iterables;
+import com.google.common.collect.Iterators;
+import com.google.common.collect.Ordering;
+import com.google.common.primitives.Longs;
+import com.google.common.util.concurrent.RateLimiter;
+
+import com.clearspring.analytics.stream.cardinality.CardinalityMergeException;
+import com.clearspring.analytics.stream.cardinality.HyperLogLogPlus;
+import com.clearspring.analytics.stream.cardinality.ICardinality;
+import com.codahale.metrics.Counter;
+import org.apache.cassandra.cache.CachingOptions;
+import org.apache.cassandra.cache.InstrumentingCache;
+import org.apache.cassandra.cache.KeyCacheKey;
+import org.apache.cassandra.concurrent.DebuggableThreadPoolExecutor;
+import org.apache.cassandra.concurrent.ScheduledExecutors;
+import org.apache.cassandra.config.*;
+import org.apache.cassandra.db.*;
+import org.apache.cassandra.db.columniterator.OnDiskAtomIterator;
+import org.apache.cassandra.db.commitlog.ReplayPosition;
+import org.apache.cassandra.db.composites.CellName;
+import org.apache.cassandra.db.filter.ColumnSlice;
+import org.apache.cassandra.db.index.SecondaryIndex;
+import org.apache.cassandra.db.lifecycle.Tracker;
+import org.apache.cassandra.dht.*;
+import org.apache.cassandra.io.compress.CompressionMetadata;
+import org.apache.cassandra.io.sstable.*;
+import org.apache.cassandra.io.sstable.metadata.*;
+import org.apache.cassandra.io.util.*;
+import org.apache.cassandra.metrics.RestorableMeter;
+import org.apache.cassandra.metrics.StorageMetrics;
+import org.apache.cassandra.service.ActiveRepairService;
+import org.apache.cassandra.service.CacheService;
+import org.apache.cassandra.service.StorageService;
+import org.apache.cassandra.utils.*;
+import org.apache.cassandra.utils.concurrent.OpOrder;
+import org.slf4j.Logger;
+import org.slf4j.LoggerFactory;
+import org.apache.cassandra.utils.concurrent.Ref;
+import org.apache.cassandra.utils.concurrent.SelfRefCounted;
+
+import static org.apache.cassandra.db.Directories.SECONDARY_INDEX_NAME_SEPARATOR;
+
+/**
+ * An SSTableReader can be constructed in a number of places, but typically is either
+ * read from disk at startup, or constructed from a flushed memtable, or after compaction
+ * to replace some existing sstables. However once created, an sstablereader may also be modified.
+ *
+ * A reader's OpenReason describes its current stage in its lifecycle, as follows:
+ *
+ *
+ * <pre> {@code
+ * NORMAL
+ * From: None => Reader has been read from disk, either at startup or from a flushed memtable
+ * EARLY => Reader is the final result of a compaction
+ * MOVED_START => Reader WAS being compacted, but this failed and it has been restored to NORMAL status
+ *
+ * EARLY
+ * From: None => Reader is a compaction replacement that is either incomplete and has been opened
+ * to represent its partial result status, or has been finished but the compaction
+ * it is a part of has not yet completed fully
+ * EARLY => Same as from None, only it is not the first time it has been
+ *
+ * MOVED_START
+ * From: NORMAL => Reader is being compacted. This compaction has not finished, but the compaction result
+ * is either partially or fully opened, to either partially or fully replace this reader.
+ * This reader's start key has been updated to represent this, so that reads only hit
+ * one or the other reader.
+ *
+ * METADATA_CHANGE
+ * From: NORMAL => Reader has seen low traffic and the amount of memory available for index summaries is
+ * constrained, so its index summary has been downsampled.
+ * METADATA_CHANGE => Same
+ * } </pre>
+ *
+ * Note that in parallel to this, there are two different Descriptor types; TMPLINK and FINAL; the latter corresponds
+ * to NORMAL state readers and all readers that replace a NORMAL one. TMPLINK is used for EARLY state readers and
+ * no others.
+ *
+ * When a reader is being compacted, if the result is large its replacement may be opened as EARLY before compaction
+ * completes in order to present the result to consumers earlier. In this case the reader will itself be changed to
+ * a MOVED_START state, where its start no longer represents its on-disk minimum key. This is to permit reads to be
+ * directed to only one reader when the two represent the same data. The EARLY file can represent a compaction result
+ * that is either partially complete and still in-progress, or a complete and immutable sstable that is part of a larger
+ * macro compaction action that has not yet fully completed.
+ *
+ * Currently ALL compaction results at least briefly go through an EARLY open state prior to completion, regardless
+ * of if early opening is enabled.
+ *
+ * Since a reader can be created multiple times over the same shared underlying resources, and the exact resources
+ * it shares between each instance differ subtly, we track the lifetime of any underlying resource with its own
+ * reference count, which each instance takes a Ref to. Each instance then tracks references to itself, and once these
+ * all expire it releases its Refs to these underlying resources.
+ *
+ * There is some shared cleanup behaviour needed only once all sstablereaders in a certain stage of their lifecycle
+ * (i.e. EARLY or NORMAL opening), and some that must only occur once all readers of any kind over a single logical
+ * sstable have expired. These are managed by the TypeTidy and GlobalTidy classes at the bottom, and are effectively
+ * managed as another resource each instance tracks its own Ref instance to, to ensure all of these resources are
+ * cleaned up safely and can be debugged otherwise.
+ *
+ * TODO: fill in details about Tracker and lifecycle interactions for tools, and for compaction strategies
+ */
+public abstract class SSTableReader extends SSTable implements SelfRefCounted<SSTableReader>
+{
+ private static final Logger logger = LoggerFactory.getLogger(SSTableReader.class);
+
+ private static final ScheduledThreadPoolExecutor syncExecutor = new ScheduledThreadPoolExecutor(1);
+ private static final RateLimiter meterSyncThrottle = RateLimiter.create(100.0);
+
+ public static final Comparator<SSTableReader> maxTimestampComparator = new Comparator<SSTableReader>()
+ {
+ public int compare(SSTableReader o1, SSTableReader o2)
+ {
+ long ts1 = o1.getMaxTimestamp();
+ long ts2 = o2.getMaxTimestamp();
+ return (ts1 > ts2 ? -1 : (ts1 == ts2 ? 0 : 1));
+ }
+ };
+
+ // it's just an object, which we use regular Object equality on; we introduce a special class just for easy recognition
+ public static final class UniqueIdentifier {}
+
+ public static final Comparator<SSTableReader> sstableComparator = new Comparator<SSTableReader>()
+ {
+ public int compare(SSTableReader o1, SSTableReader o2)
+ {
+ return o1.first.compareTo(o2.first);
+ }
+ };
+
+ public static final Ordering<SSTableReader> sstableOrdering = Ordering.from(sstableComparator);
+
+ /**
+ * maxDataAge is a timestamp in local server time (e.g. System.currentTimeMilli) which represents an upper bound
+ * to the newest piece of data stored in the sstable. In other words, this sstable does not contain items created
+ * later than maxDataAge.
+ *
+ * The field is not serialized to disk, so relying on it for more than what truncate does is not advised.
+ *
+ * When a new sstable is flushed, maxDataAge is set to the time of creation.
+ * When a sstable is created from compaction, maxDataAge is set to max of all merged sstables.
+ *
+ * The age is in milliseconds since epoc and is local to this host.
+ */
+ public final long maxDataAge;
+
+ public enum OpenReason
+ {
+ NORMAL,
+ EARLY,
+ METADATA_CHANGE,
+ MOVED_START
+ }
+
+ public final OpenReason openReason;
+ public final UniqueIdentifier instanceId = new UniqueIdentifier();
+
+ // indexfile and datafile: might be null before a call to load()
+ protected SegmentedFile ifile;
+ protected SegmentedFile dfile;
+ protected IndexSummary indexSummary;
+ protected IFilter bf;
+
+ protected final RowIndexEntry.IndexSerializer rowIndexEntrySerializer;
+
+ protected InstrumentingCache<KeyCacheKey, RowIndexEntry> keyCache;
+
+ protected final BloomFilterTracker bloomFilterTracker = new BloomFilterTracker();
+
+ // technically isCompacted is not necessary since it should never be unreferenced unless it is also compacted,
+ // but it seems like a good extra layer of protection against reference counting bugs to not delete data based on that alone
+ protected final AtomicBoolean isSuspect = new AtomicBoolean(false);
+
+ // not final since we need to be able to change level on a file.
+ protected volatile StatsMetadata sstableMetadata;
+
+ protected final AtomicLong keyCacheHit = new AtomicLong(0);
+ protected final AtomicLong keyCacheRequest = new AtomicLong(0);
+
+ private final InstanceTidier tidy = new InstanceTidier(descriptor, metadata);
+ private final Ref<SSTableReader> selfRef = new Ref<>(this, tidy);
+
+ private RestorableMeter readMeter;
+
+ /**
+ * Calculate approximate key count.
+ * If cardinality estimator is available on all given sstables, then this method use them to estimate
+ * key count.
+ * If not, then this uses index summaries.
+ *
+ * @param sstables SSTables to calculate key count
+ * @return estimated key count
+ */
+ public static long getApproximateKeyCount(Collection<SSTableReader> sstables)
+ {
+ long count = -1;
+
+ // check if cardinality estimator is available for all SSTables
+ boolean cardinalityAvailable = !sstables.isEmpty() && Iterators.all(sstables.iterator(), new Predicate<SSTableReader>()
+ {
+ public boolean apply(SSTableReader sstable)
+ {
+ return sstable.descriptor.version.hasNewStatsFile();
+ }
+ });
+
+ // if it is, load them to estimate key count
+ if (cardinalityAvailable)
+ {
+ boolean failed = false;
+ ICardinality cardinality = null;
+ for (SSTableReader sstable : sstables)
+ {
+ if (sstable.openReason == OpenReason.EARLY)
+ continue;
+
+ try
+ {
+ CompactionMetadata metadata = (CompactionMetadata) sstable.descriptor.getMetadataSerializer().deserialize(sstable.descriptor, MetadataType.COMPACTION);
+ assert metadata != null : sstable.getFilename();
+ if (cardinality == null)
+ cardinality = metadata.cardinalityEstimator;
+ else
+ cardinality = cardinality.merge(metadata.cardinalityEstimator);
+ }
+ catch (IOException e)
+ {
+ logger.warn("Reading cardinality from Statistics.db failed.", e);
+ failed = true;
+ break;
+ }
+ catch (CardinalityMergeException e)
+ {
+ logger.warn("Cardinality merge failed.", e);
+ failed = true;
+ break;
+ }
+ }
+ if (cardinality != null && !failed)
+ count = cardinality.cardinality();
+ }
+
+ // if something went wrong above or cardinality is not available, calculate using index summary
+ if (count < 0)
+ {
+ for (SSTableReader sstable : sstables)
+ count += sstable.estimatedKeys();
+ }
+ return count;
+ }
+
+ /**
+ * Estimates how much of the keys we would keep if the sstables were compacted together
+ */
+ public static double estimateCompactionGain(Set<SSTableReader> overlapping)
+ {
+ Set<ICardinality> cardinalities = new HashSet<>(overlapping.size());
+ for (SSTableReader sstable : overlapping)
+ {
+ try
+ {
+ ICardinality cardinality = ((CompactionMetadata) sstable.descriptor.getMetadataSerializer().deserialize(sstable.descriptor, MetadataType.COMPACTION)).cardinalityEstimator;
+ if (cardinality != null)
+ cardinalities.add(cardinality);
+ else
+ logger.debug("Got a null cardinality estimator in: {}", sstable.getFilename());
+ }
+ catch (IOException e)
+ {
+ logger.warn("Could not read up compaction metadata for {}", sstable, e);
+ }
+ }
+ long totalKeyCountBefore = 0;
+ for (ICardinality cardinality : cardinalities)
+ {
+ totalKeyCountBefore += cardinality.cardinality();
+ }
+ if (totalKeyCountBefore == 0)
+ return 1;
+
+ long totalKeyCountAfter = mergeCardinalities(cardinalities).cardinality();
+ logger.debug("Estimated compaction gain: {}/{}={}", totalKeyCountAfter, totalKeyCountBefore, ((double)totalKeyCountAfter)/totalKeyCountBefore);
+ return ((double)totalKeyCountAfter)/totalKeyCountBefore;
+ }
+
+ private static ICardinality mergeCardinalities(Collection<ICardinality> cardinalities)
+ {
+ ICardinality base = new HyperLogLogPlus(13, 25); // see MetadataCollector.cardinality
+ try
+ {
+ base = base.merge(cardinalities.toArray(new ICardinality[cardinalities.size()]));
+ }
+ catch (CardinalityMergeException e)
+ {
+ logger.warn("Could not merge cardinalities", e);
+ }
+ return base;
+ }
+
+ public static SSTableReader open(Descriptor descriptor) throws IOException
+ {
+ CFMetaData metadata;
+ if (descriptor.cfname.contains(SECONDARY_INDEX_NAME_SEPARATOR))
+ {
+ int i = descriptor.cfname.indexOf(SECONDARY_INDEX_NAME_SEPARATOR);
+ String parentName = descriptor.cfname.substring(0, i);
+ CFMetaData parent = Schema.instance.getCFMetaData(descriptor.ksname, parentName);
+ ColumnDefinition def = parent.getColumnDefinitionForIndex(descriptor.cfname.substring(i + 1));
+ metadata = CFMetaData.newIndexMetadata(parent, def, SecondaryIndex.getIndexComparator(parent, def));
+ }
+ else
+ {
+ metadata = Schema.instance.getCFMetaData(descriptor.ksname, descriptor.cfname);
+ }
+ return open(descriptor, metadata);
+ }
+
+ public static SSTableReader open(Descriptor desc, CFMetaData metadata) throws IOException
+ {
+ IPartitioner p = desc.cfname.contains(SECONDARY_INDEX_NAME_SEPARATOR)
+ ? new LocalPartitioner(metadata.getKeyValidator())
+ : StorageService.getPartitioner();
+ return open(desc, componentsFor(desc), metadata, p);
+ }
+
+ public static SSTableReader open(Descriptor descriptor, Set<Component> components, CFMetaData metadata, IPartitioner partitioner) throws IOException
+ {
+ return open(descriptor, components, metadata, partitioner, true);
+ }
+
+ // use only for offline or "Standalone" operations
+ public static SSTableReader openNoValidation(Descriptor descriptor, Set<Component> components, ColumnFamilyStore cfs) throws IOException
+ {
+ return open(descriptor, components, cfs.metadata, cfs.partitioner, false);
+ }
+
+ /**
+ * Open SSTable reader to be used in batch mode(such as sstableloader).
+ *
+ * @param descriptor
+ * @param components
+ * @param metadata
+ * @param partitioner
+ * @return opened SSTableReader
+ * @throws IOException
+ */
+ public static SSTableReader openForBatch(Descriptor descriptor, Set<Component> components, CFMetaData metadata, IPartitioner partitioner) throws IOException
+ {
+ // Minimum components without which we can't do anything
+ assert components.contains(Component.DATA) : "Data component is missing for sstable " + descriptor;
+ assert components.contains(Component.PRIMARY_INDEX) : "Primary index component is missing for sstable " + descriptor;
+
+ Map<MetadataType, MetadataComponent> sstableMetadata = descriptor.getMetadataSerializer().deserialize(descriptor,
+ EnumSet.of(MetadataType.VALIDATION, MetadataType.STATS));
+ ValidationMetadata validationMetadata = (ValidationMetadata) sstableMetadata.get(MetadataType.VALIDATION);
+ StatsMetadata statsMetadata = (StatsMetadata) sstableMetadata.get(MetadataType.STATS);
+
+ // Check if sstable is created using same partitioner.
+ // Partitioner can be null, which indicates older version of sstable or no stats available.
+ // In that case, we skip the check.
+ String partitionerName = partitioner.getClass().getCanonicalName();
+ if (validationMetadata != null && !partitionerName.equals(validationMetadata.partitioner))
+ {
+ logger.error(String.format("Cannot open %s; partitioner %s does not match system partitioner %s. Note that the default partitioner starting with Cassandra 1.2 is Murmur3Partitioner, so you will need to edit that to match your old partitioner if upgrading.",
+ descriptor, validationMetadata.partitioner, partitionerName));
+ System.exit(1);
+ }
+
+ logger.info("Opening {} ({} bytes)", descriptor, new File(descriptor.filenameFor(Component.DATA)).length());
+ SSTableReader sstable = internalOpen(descriptor, components, metadata, partitioner, System.currentTimeMillis(),
+ statsMetadata, OpenReason.NORMAL);
+
+ // special implementation of load to use non-pooled SegmentedFile builders
+ try(SegmentedFile.Builder ibuilder = new BufferedSegmentedFile.Builder();
+ SegmentedFile.Builder dbuilder = sstable.compression
+ ? new CompressedSegmentedFile.Builder(null)
+ : new BufferedSegmentedFile.Builder())
+ {
+ if (!sstable.loadSummary(ibuilder, dbuilder))
+ sstable.buildSummary(false, ibuilder, dbuilder, false, Downsampling.BASE_SAMPLING_LEVEL);
+ sstable.ifile = ibuilder.complete(sstable.descriptor.filenameFor(Component.PRIMARY_INDEX));
+ sstable.dfile = dbuilder.complete(sstable.descriptor.filenameFor(Component.DATA));
+ sstable.bf = FilterFactory.AlwaysPresent;
+ sstable.setup(true);
+ return sstable;
+ }
+ }
+
+ private static SSTableReader open(Descriptor descriptor,
+ Set<Component> components,
+ CFMetaData metadata,
+ IPartitioner partitioner,
+ boolean validate) throws IOException
+ {
+ // Minimum components without which we can't do anything
+ assert components.contains(Component.DATA) : "Data component is missing for sstable " + descriptor;
- assert components.contains(Component.PRIMARY_INDEX) : "Primary index component is missing for sstable " + descriptor;
++ assert !validate || components.contains(Component.PRIMARY_INDEX) : "Primary index component is missing for sstable " + descriptor;
+
+ Map<MetadataType, MetadataComponent> sstableMetadata = descriptor.getMetadataSerializer().deserialize(descriptor,
- EnumSet.of(MetadataType.VALIDATION, MetadataType.STATS));
++ EnumSet.of(MetadataType.VALIDATION, MetadataType.STATS));
+ ValidationMetadata validationMetadata = (ValidationMetadata) sstableMetadata.get(MetadataType.VALIDATION);
+ StatsMetadata statsMetadata = (StatsMetadata) sstableMetadata.get(MetadataType.STATS);
+
+ // Check if sstable is created using same partitioner.
+ // Partitioner can be null, which indicates older version of sstable or no stats available.
+ // In that case, we skip the check.
+ String partitionerName = partitioner.getClass().getCanonicalName();
+ if (validationMetadata != null && !partitionerName.equals(validationMetadata.partitioner))
+ {
+ logger.error(String.format("Cannot open %s; partitioner %s does not match system partitioner %s. Note that the default partitioner starting with Cassandra 1.2 is Murmur3Partitioner, so you will need to edit that to match your old partitioner if upgrading.",
+ descriptor, validationMetadata.partitioner, partitionerName));
+ System.exit(1);
+ }
+
+ logger.info("Opening {} ({} bytes)", descriptor, new File(descriptor.filenameFor(Component.DATA)).length());
+ SSTableReader sstable = internalOpen(descriptor, components, metadata, partitioner, System.currentTimeMillis(),
+ statsMetadata, OpenReason.NORMAL);
+
+ // load index and filter
+ long start = System.nanoTime();
+ sstable.load(validationMetadata);
+ logger.debug("INDEX LOAD TIME for {}: {} ms.", descriptor, TimeUnit.NANOSECONDS.toMillis(System.nanoTime() - start));
+
+ sstable.setup(!validate);
+ if (validate)
+ sstable.validate();
+
+ if (sstable.getKeyCache() != null)
+ logger.debug("key cache contains {}/{} keys", sstable.getKeyCache().size(), sstable.getKeyCache().getCapacity());
+
+ return sstable;
+ }
+
+ public static void logOpenException(Descriptor descriptor, IOException e)
+ {
+ if (e instanceof FileNotFoundException)
+ logger.error("Missing sstable component in {}; skipped because of {}", descriptor, e.getMessage());
+ else
+ logger.error("Corrupt sstable {}; skipped", descriptor, e);
+ }
+
+ public static Collection<SSTableReader> openAll(Set<Map.Entry<Descriptor, Set<Component>>> entries,
+ final CFMetaData metadata,
+ final IPartitioner partitioner)
+ {
+ final Collection<SSTableReader> sstables = new LinkedBlockingQueue<>();
+
+ ExecutorService executor = DebuggableThreadPoolExecutor.createWithFixedPoolSize("SSTableBatchOpen", FBUtilities.getAvailableProcessors());
+ for (final Map.Entry<Descriptor, Set<Component>> entry : entries)
+ {
+ Runnable runnable = new Runnable()
+ {
+ public void run()
+ {
+ SSTableReader sstable;
+ try
+ {
+ sstable = open(entry.getKey(), entry.getValue(), metadata, partitioner);
+ }
+ catch (IOException ex)
+ {
+ logger.error("Corrupt sstable {}; skipped", entry, ex);
+ return;
+ }
+ sstables.add(sstable);
+ }
+ };
+ executor.submit(runnable);
+ }
+
+ executor.shutdown();
+ try
+ {
+ executor.awaitTermination(7, TimeUnit.DAYS);
+ }
+ catch (InterruptedException e)
+ {
+ throw new AssertionError(e);
+ }
+
+ return sstables;
+
+ }
+
+ /**
+ * Open a RowIndexedReader which already has its state initialized (by SSTableWriter).
+ */
+ public static SSTableReader internalOpen(Descriptor desc,
+ Set<Component> components,
+ CFMetaData metadata,
+ IPartitioner partitioner,
+ SegmentedFile ifile,
+ SegmentedFile dfile,
+ IndexSummary isummary,
+ IFilter bf,
+ long maxDataAge,
+ StatsMetadata sstableMetadata,
+ OpenReason openReason)
+ {
+ assert desc != null && partitioner != null && ifile != null && dfile != null && isummary != null && bf != null && sstableMetadata != null;
+
+ SSTableReader reader = internalOpen(desc, components, metadata, partitioner, maxDataAge, sstableMetadata, openReason);
+
+ reader.bf = bf;
+ reader.ifile = ifile;
+ reader.dfile = dfile;
+ reader.indexSummary = isummary;
+ reader.setup(false);
+
+ return reader;
+ }
+
+
+ private static SSTableReader internalOpen(final Descriptor descriptor,
+ Set<Component> components,
+ CFMetaData metadata,
+ IPartitioner partitioner,
+ Long maxDataAge,
+ StatsMetadata sstableMetadata,
+ OpenReason openReason)
+ {
+ Factory readerFactory = descriptor.getFormat().getReaderFactory();
+
+ return readerFactory.open(descriptor, components, metadata, partitioner, maxDataAge, sstableMetadata, openReason);
+ }
+
+ protected SSTableReader(final Descriptor desc,
+ Set<Component> components,
+ CFMetaData metadata,
+ IPartitioner partitioner,
+ long maxDataAge,
+ StatsMetadata sstableMetadata,
+ OpenReason openReason)
+ {
+ super(desc, components, metadata, partitioner);
+ this.sstableMetadata = sstableMetadata;
+ this.maxDataAge = maxDataAge;
+ this.openReason = openReason;
+ this.rowIndexEntrySerializer = descriptor.version.getSSTableFormat().getIndexSerializer(metadata);
+ }
+
+ public static long getTotalBytes(Iterable<SSTableReader> sstables)
+ {
+ long sum = 0;
+ for (SSTableReader sstable : sstables)
+ sum += sstable.onDiskLength();
+ return sum;
+ }
+
+ public static long getTotalUncompressedBytes(Iterable<SSTableReader> sstables)
+ {
+ long sum = 0;
+ for (SSTableReader sstable : sstables)
+ sum += sstable.uncompressedLength();
+
+ return sum;
+ }
+
+ public boolean equals(Object that)
+ {
+ return that instanceof SSTableReader && ((SSTableReader) that).descriptor.equals(this.descriptor);
+ }
+
+ public int hashCode()
+ {
+ return this.descriptor.hashCode();
+ }
+
+ public String getFilename()
+ {
+ return dfile.path();
+ }
+
- public String getIndexFilename()
- {
- return ifile.path();
- }
-
+ public void setupKeyCache()
+ {
+ // under normal operation we can do this at any time, but SSTR is also used outside C* proper,
+ // e.g. by BulkLoader, which does not initialize the cache. As a kludge, we set up the cache
+ // here when we know we're being wired into the rest of the server infrastructure.
+ keyCache = CacheService.instance.keyCache;
+ }
+
+ private void load(ValidationMetadata validation) throws IOException
+ {
+ if (metadata.getBloomFilterFpChance() == 1.0)
+ {
+ // bf is disabled.
+ load(false, true);
+ bf = FilterFactory.AlwaysPresent;
+ }
++ else if (!components.contains(Component.PRIMARY_INDEX))
++ {
++ // avoid any reading of the missing primary index component.
++ // this should only happen during StandaloneScrubber
++ load(false, false);
++ }
+ else if (!components.contains(Component.FILTER) || validation == null)
+ {
+ // bf is enabled, but filter component is missing.
+ load(true, true);
+ }
+ else if (validation.bloomFilterFPChance != metadata.getBloomFilterFpChance())
+ {
+ // bf fp chance in sstable metadata and it has changed since compaction.
+ load(true, true);
+ }
+ else
+ {
+ // bf is enabled and fp chance matches the currently configured value.
+ load(false, true);
+ loadBloomFilter();
+ }
+ }
+
+ /**
+ * Load bloom filter from Filter.db file.
+ *
+ * @throws IOException
+ */
+ private void loadBloomFilter() throws IOException
+ {
+ try (DataInputStream stream = new DataInputStream(new BufferedInputStream(new FileInputStream(descriptor.filenameFor(Component.FILTER)))))
+ {
+ bf = FilterFactory.deserialize(stream, true);
+ }
+ }
+
+ /**
+ * Loads ifile, dfile and indexSummary, and optionally recreates the bloom filter.
+ * @param saveSummaryIfCreated for bulk loading purposes, if the summary was absent and needed to be built, you can
+ * avoid persisting it to disk by setting this to false
+ */
+ private void load(boolean recreateBloomFilter, boolean saveSummaryIfCreated) throws IOException
+ {
+ try(SegmentedFile.Builder ibuilder = SegmentedFile.getBuilder(DatabaseDescriptor.getIndexAccessMode(), false);
+ SegmentedFile.Builder dbuilder = SegmentedFile.getBuilder(DatabaseDescriptor.getDiskAccessMode(), compression))
+ {
+ boolean summaryLoaded = loadSummary(ibuilder, dbuilder);
+ boolean builtSummary = false;
+ if (recreateBloomFilter || !summaryLoaded)
+ {
+ buildSummary(recreateBloomFilter, ibuilder, dbuilder, summaryLoaded, Downsampling.BASE_SAMPLING_LEVEL);
+ builtSummary = true;
+ }
+
- ifile = ibuilder.complete(descriptor.filenameFor(Component.PRIMARY_INDEX));
++ if (components.contains(Component.PRIMARY_INDEX))
++ ifile = ibuilder.complete(descriptor.filenameFor(Component.PRIMARY_INDEX));
++
+ dfile = dbuilder.complete(descriptor.filenameFor(Component.DATA));
+
+ // Check for an index summary that was downsampled even though the serialization format doesn't support
+ // that. If it was downsampled, rebuild it. See CASSANDRA-8993 for details.
- if (!descriptor.version.hasSamplingLevel() && !builtSummary && !validateSummarySamplingLevel())
++ if (!descriptor.version.hasSamplingLevel() && !builtSummary && !validateSummarySamplingLevel() && ifile != null)
+ {
+ indexSummary.close();
+ ifile.close();
+ dfile.close();
+
+ logger.info("Detected erroneously downsampled index summary; will rebuild summary at full sampling");
+ FileUtils.deleteWithConfirm(new File(descriptor.filenameFor(Component.SUMMARY)));
+
+ try(SegmentedFile.Builder ibuilderRebuild = SegmentedFile.getBuilder(DatabaseDescriptor.getIndexAccessMode(), false);
+ SegmentedFile.Builder dbuilderRebuild = SegmentedFile.getBuilder(DatabaseDescriptor.getDiskAccessMode(), compression))
+ {
+ buildSummary(false, ibuilderRebuild, dbuilderRebuild, false, Downsampling.BASE_SAMPLING_LEVEL);
+ ifile = ibuilderRebuild.complete(descriptor.filenameFor(Component.PRIMARY_INDEX));
+ dfile = dbuilderRebuild.complete(descriptor.filenameFor(Component.DATA));
+ saveSummary(ibuilderRebuild, dbuilderRebuild);
+ }
+ }
+ else if (saveSummaryIfCreated && builtSummary)
+ {
+ saveSummary(ibuilder, dbuilder);
+ }
+ }
+ }
+
+ /**
+ * Build index summary(and optionally bloom filter) by reading through Index.db file.
+ *
+ * @param recreateBloomFilter true if recreate bloom filter
+ * @param ibuilder
+ * @param dbuilder
+ * @param summaryLoaded true if index summary is already loaded and not need to build again
+ * @throws IOException
+ */
+ private void buildSummary(boolean recreateBloomFilter, SegmentedFile.Builder ibuilder, SegmentedFile.Builder dbuilder, boolean summaryLoaded, int samplingLevel) throws IOException
+ {
++ if (!components.contains(Component.PRIMARY_INDEX))
++ return;
++
+ // we read the positions in a BRAF so we don't have to worry about an entry spanning a mmap boundary.
+ try (RandomAccessReader primaryIndex = RandomAccessReader.open(new File(descriptor.filenameFor(Component.PRIMARY_INDEX))))
+ {
+ long indexSize = primaryIndex.length();
+ long histogramCount = sstableMetadata.estimatedRowSize.count();
+ long estimatedKeys = histogramCount > 0 && !sstableMetadata.estimatedRowSize.isOverflowed()
+ ? histogramCount
+ : estimateRowsFromIndex(primaryIndex); // statistics is supposed to be optional
+
+ if (recreateBloomFilter)
+ bf = FilterFactory.getFilter(estimatedKeys, metadata.getBloomFilterFpChance(), true);
+
+ try (IndexSummaryBuilder summaryBuilder = summaryLoaded ? null : new IndexSummaryBuilder(estimatedKeys, metadata.getMinIndexInterval(), samplingLevel))
+ {
+ long indexPosition;
+ RowIndexEntry.IndexSerializer rowIndexSerializer = descriptor.getFormat().getIndexSerializer(metadata);
+
+ while ((indexPosition = primaryIndex.getFilePointer()) != indexSize)
+ {
+ ByteBuffer key = ByteBufferUtil.readWithShortLength(primaryIndex);
+ RowIndexEntry indexEntry = rowIndexSerializer.deserialize(primaryIndex, descriptor.version);
+ DecoratedKey decoratedKey = partitioner.decorateKey(key);
+ if (first == null)
+ first = decoratedKey;
+ last = decoratedKey;
+
+ if (recreateBloomFilter)
+ bf.add(decoratedKey);
+
+ // if summary was already read from disk we don't want to re-populate it using primary index
+ if (!summaryLoaded)
+ {
+ summaryBuilder.maybeAddEntry(decoratedKey, indexPosition);
+ ibuilder.addPotentialBoundary(indexPosition);
+ dbuilder.addPotentialBoundary(indexEntry.position);
+ }
+ }
+
+ if (!summaryLoaded)
+ indexSummary = summaryBuilder.build(partitioner);
+ }
+ }
+
+ first = getMinimalKey(first);
+ last = getMinimalKey(last);
+ }
+
+ /**
+ * Load index summary from Summary.db file if it exists.
+ *
+ * if loaded index summary has different index interval from current value stored in schema,
+ * then Summary.db file will be deleted and this returns false to rebuild summary.
+ *
+ * @param ibuilder
+ * @param dbuilder
+ * @return true if index summary is loaded successfully from Summary.db file.
+ */
+ @SuppressWarnings("resource")
+ public boolean loadSummary(SegmentedFile.Builder ibuilder, SegmentedFile.Builder dbuilder)
+ {
+ File summariesFile = new File(descriptor.filenameFor(Component.SUMMARY));
+ if (!summariesFile.exists())
+ return false;
+
+ DataInputStream iStream = null;
+ try
+ {
+ iStream = new DataInputStream(new FileInputStream(summariesFile));
+ indexSummary = IndexSummary.serializer.deserialize(
+ iStream, partitioner, descriptor.version.hasSamplingLevel(),
+ metadata.getMinIndexInterval(), metadata.getMaxIndexInterval());
+ first = partitioner.decorateKey(ByteBufferUtil.readWithLength(iStream));
+ last = partitioner.decorateKey(ByteBufferUtil.readWithLength(iStream));
+ ibuilder.deserializeBounds(iStream);
+ dbuilder.deserializeBounds(iStream);
+ }
+ catch (IOException e)
+ {
+ if (indexSummary != null)
+ indexSummary.close();
+ logger.debug("Cannot deserialize SSTable Summary File {}: {}", summariesFile.getPath(), e.getMessage());
+ // corrupted; delete it and fall back to creating a new summary
+ FileUtils.closeQuietly(iStream);
+ // delete it and fall back to creating a new summary
+ FileUtils.deleteWithConfirm(summariesFile);
+ return false;
+ }
+ finally
+ {
+ FileUtils.closeQuietly(iStream);
+ }
+
+ return true;
+ }
+
+ /**
+ * Validates that an index summary has full sampling, as expected when the serialization format does not support
+ * persisting the sampling level.
+ * @return true if the summary has full sampling, false otherwise
+ */
+ private boolean validateSummarySamplingLevel()
+ {
+ // We need to check index summary entries against the index to verify that none of them were dropped due to
+ // downsampling. Downsampling can drop any of the first BASE_SAMPLING_LEVEL entries (repeating that drop pattern
+ // for the remainder of the summary). Unfortunately, the first entry to be dropped is the entry at
+ // index (BASE_SAMPLING_LEVEL - 1), so we need to check a full set of BASE_SAMPLING_LEVEL entries.
++ if (ifile == null)
++ return false;
++
+ Iterator<FileDataInput> segments = ifile.iterator(0);
+ int i = 0;
+ int summaryEntriesChecked = 0;
+ int expectedIndexInterval = getMinIndexInterval();
+ while (segments.hasNext())
+ {
+ String path = null;
+ try (FileDataInput in = segments.next())
+ {
+ path = in.getPath();
+ while (!in.isEOF())
+ {
+ ByteBuffer indexKey = ByteBufferUtil.readWithShortLength(in);
+ if (i % expectedIndexInterval == 0)
+ {
+ ByteBuffer summaryKey = ByteBuffer.wrap(indexSummary.getKey(i / expectedIndexInterval));
+ if (!summaryKey.equals(indexKey))
+ return false;
+ summaryEntriesChecked++;
+
+ if (summaryEntriesChecked == Downsampling.BASE_SAMPLING_LEVEL)
+ return true;
+ }
+ RowIndexEntry.Serializer.skip(in);
+ i++;
+ }
+ }
+ catch (IOException e)
+ {
+ markSuspect();
+ throw new CorruptSSTableException(e, path);
+ }
+ }
+
+ return true;
+ }
+
+ /**
+ * Save index summary to Summary.db file.
+ *
+ * @param ibuilder
+ * @param dbuilder
+ */
+
+ public void saveSummary(SegmentedFile.Builder ibuilder, SegmentedFile.Builder dbuilder)
+ {
+ saveSummary(this.descriptor, this.first, this.last, ibuilder, dbuilder, indexSummary);
+ }
+
+ private void saveSummary(SegmentedFile.Builder ibuilder, SegmentedFile.Builder dbuilder, IndexSummary newSummary)
+ {
+ saveSummary(this.descriptor, this.first, this.last, ibuilder, dbuilder, newSummary);
+ }
+ /**
+ * Save index summary to Summary.db file.
+ */
+ public static void saveSummary(Descriptor descriptor, DecoratedKey first, DecoratedKey last,
+ SegmentedFile.Builder ibuilder, SegmentedFile.Builder dbuilder, IndexSummary summary)
+ {
+ File summariesFile = new File(descriptor.filenameFor(Component.SUMMARY));
+ if (summariesFile.exists())
+ FileUtils.deleteWithConfirm(summariesFile);
+
+ try (DataOutputStreamPlus oStream = new BufferedDataOutputStreamPlus(new FileOutputStream(summariesFile));)
+ {
+ IndexSummary.serializer.serialize(summary, oStream, descriptor.version.hasSamplingLevel());
+ ByteBufferUtil.writeWithLength(first.getKey(), oStream);
+ ByteBufferUtil.writeWithLength(last.getKey(), oStream);
+ ibuilder.serializeBounds(oStream);
+ dbuilder.serializeBounds(oStream);
+ }
+ catch (IOException e)
+ {
+ logger.debug("Cannot save SSTable Summary: ", e);
+
+ // corrupted hence delete it and let it load it now.
+ if (summariesFile.exists())
+ FileUtils.deleteWithConfirm(summariesFile);
+ }
+ }
+
+ public void setReplaced()
+ {
+ synchronized (tidy.global)
+ {
+ assert !tidy.isReplaced;
+ tidy.isReplaced = true;
+ }
+ }
+
+ public boolean isReplaced()
+ {
+ synchronized (tidy.global)
+ {
+ return tidy.isReplaced;
+ }
+ }
+
+ // runOnClose must NOT be an anonymous or non-static inner class, nor must it retain a reference chain to this reader
+ public void runOnClose(final Runnable runOnClose)
+ {
+ synchronized (tidy.global)
+ {
+ final Runnable existing = tidy.runOnClose;
+ tidy.runOnClose = AndThen.get(existing, runOnClose);
+ }
+ }
+
+ private static class AndThen implements Runnable
+ {
+ final Runnable runFirst;
+ final Runnable runSecond;
+
+ private AndThen(Runnable runFirst, Runnable runSecond)
+ {
+ this.runFirst = runFirst;
+ this.runSecond = runSecond;
+ }
+
+ public void run()
+ {
+ runFirst.run();
+ runSecond.run();
+ }
+
+ static Runnable get(Runnable runFirst, Runnable runSecond)
+ {
+ if (runFirst == null)
+ return runSecond;
+ return new AndThen(runFirst, runSecond);
+ }
+ }
+
+ /**
+ * Clone this reader with the provided start and open reason, and set the clone as replacement.
+ *
+ * @param newFirst the first key for the replacement (which can be different from the original due to the pre-emptive
+ * opening of compaction results).
+ * @param reason the {@code OpenReason} for the replacement.
+ *
+ * @return the cloned reader. That reader is set as a replacement by the method.
+ */
+ private SSTableReader cloneAndReplace(DecoratedKey newFirst, OpenReason reason)
+ {
+ return cloneAndReplace(newFirst, reason, indexSummary.sharedCopy());
+ }
+
+ /**
+ * Clone this reader with the new values and set the clone as replacement.
+ *
+ * @param newFirst the first key for the replacement (which can be different from the original due to the pre-emptive
+ * opening of compaction results).
+ * @param reason the {@code OpenReason} for the replacement.
+ * @param newSummary the index summary for the replacement.
+ *
+ * @return the cloned reader. That reader is set as a replacement by the method.
+ */
+ private SSTableReader cloneAndReplace(DecoratedKey newFirst, OpenReason reason, IndexSummary newSummary)
+ {
+ SSTableReader replacement = internalOpen(descriptor,
+ components,
+ metadata,
+ partitioner,
- ifile.sharedCopy(),
++ ifile != null ? ifile.sharedCopy() : null,
+ dfile.sharedCopy(),
+ newSummary,
+ bf.sharedCopy(),
+ maxDataAge,
+ sstableMetadata,
+ reason);
+ replacement.first = newFirst;
+ replacement.last = last;
+ replacement.isSuspect.set(isSuspect.get());
+ return replacement;
+ }
+
+ // runOnClose must NOT be an anonymous or non-static inner class, nor must it retain a reference chain to this reader
+ public SSTableReader cloneWithNewStart(DecoratedKey newStart, final Runnable runOnClose)
+ {
+ synchronized (tidy.global)
+ {
+ assert openReason != OpenReason.EARLY;
+ // TODO: merge with caller's firstKeyBeyond() work,to save time
+ if (newStart.compareTo(first) > 0)
+ {
+ final long dataStart = getPosition(newStart, Operator.EQ).position;
+ final long indexStart = getIndexScanPosition(newStart);
+ this.tidy.runOnClose = new DropPageCache(dfile, dataStart, ifile, indexStart, runOnClose);
+ }
+
+ return cloneAndReplace(newStart, OpenReason.MOVED_START);
+ }
+ }
+
+ private static class DropPageCache implements Runnable
+ {
+ final SegmentedFile dfile;
+ final long dfilePosition;
+ final SegmentedFile ifile;
+ final long ifilePosition;
+ final Runnable andThen;
+
+ private DropPageCache(SegmentedFile dfile, long dfilePosition, SegmentedFile ifile, long ifilePosition, Runnable andThen)
+ {
+ this.dfile = dfile;
+ this.dfilePosition = dfilePosition;
+ this.ifile = ifile;
+ this.ifilePosition = ifilePosition;
+ this.andThen = andThen;
+ }
+
+ public void run()
+ {
+ dfile.dropPageCache(dfilePosition);
- ifile.dropPageCache(ifilePosition);
++
++ if (ifile != null)
++ ifile.dropPageCache(ifilePosition);
+ andThen.run();
+ }
+ }
+
+ /**
+ * Returns a new SSTableReader with the same properties as this SSTableReader except that a new IndexSummary will
+ * be built at the target samplingLevel. This (original) SSTableReader instance will be marked as replaced, have
+ * its DeletingTask removed, and have its periodic read-meter sync task cancelled.
+ * @param samplingLevel the desired sampling level for the index summary on the new SSTableReader
+ * @return a new SSTableReader
+ * @throws IOException
+ */
+ @SuppressWarnings("resource")
+ public SSTableReader cloneWithNewSummarySamplingLevel(ColumnFamilyStore parent, int samplingLevel) throws IOException
+ {
+ assert descriptor.version.hasSamplingLevel();
+
+ synchronized (tidy.global)
+ {
+ assert openReason != OpenReason.EARLY;
+
+ int minIndexInterval = metadata.getMinIndexInterval();
+ int maxIndexInterval = metadata.getMaxIndexInterval();
+ double effectiveInterval = indexSummary.getEffectiveIndexInterval();
+
+ IndexSummary newSummary;
+ long oldSize = bytesOnDisk();
+
+ // We have to rebuild the summary from the on-disk primary index in three cases:
+ // 1. The sampling level went up, so we need to read more entries off disk
+ // 2. The min_index_interval changed (in either direction); this changes what entries would be in the summary
+ // at full sampling (and consequently at any other sampling level)
+ // 3. The max_index_interval was lowered, forcing us to raise the sampling level
+ if (samplingLevel > indexSummary.getSamplingLevel() || indexSummary.getMinIndexInterval() != minIndexInterval || effectiveInterval > maxIndexInterval)
+ {
+ newSummary = buildSummaryAtLevel(samplingLevel);
+ }
+ else if (samplingLevel < indexSummary.getSamplingLevel())
+ {
+ // we can use the existing index summary to make a smaller one
+ newSummary = IndexSummaryBuilder.downsample(indexSummary, samplingLevel, minIndexInterval, partitioner);
+
+ try(SegmentedFile.Builder ibuilder = SegmentedFile.getBuilder(DatabaseDescriptor.getIndexAccessMode(), false);
+ SegmentedFile.Builder dbuilder = SegmentedFile.getBuilder(DatabaseDescriptor.getDiskAccessMode(), compression))
+ {
+ saveSummary(ibuilder, dbuilder, newSummary);
+ }
+ }
+ else
+ {
+ throw new AssertionError("Attempted to clone SSTableReader with the same index summary sampling level and " +
+ "no adjustments to min/max_index_interval");
+ }
+
+ long newSize = bytesOnDisk();
+ StorageMetrics.load.inc(newSize - oldSize);
+ parent.metric.liveDiskSpaceUsed.inc(newSize - oldSize);
+
+ return cloneAndReplace(first, OpenReason.METADATA_CHANGE, newSummary);
+ }
+ }
+
+ private IndexSummary buildSummaryAtLevel(int newSamplingLevel) throws IOException
+ {
+ // we read the positions in a BRAF so we don't have to worry about an entry spanning a mmap boundary.
+ RandomAccessReader primaryIndex = RandomAccessReader.open(new File(descriptor.filenameFor(Component.PRIMARY_INDEX)));
+ try
+ {
+ long indexSize = primaryIndex.length();
+ try (IndexSummaryBuilder summaryBuilder = new IndexSummaryBuilder(estimatedKeys(), metadata.getMinIndexInterval(), newSamplingLevel))
+ {
+ long indexPosition;
+ while ((indexPosition = primaryIndex.getFilePointer()) != indexSize)
+ {
+ summaryBuilder.maybeAddEntry(partitioner.decorateKey(ByteBufferUtil.readWithShortLength(primaryIndex)), indexPosition);
+ RowIndexEntry.Serializer.skip(primaryIndex);
+ }
+
+ return summaryBuilder.build(partitioner);
+ }
+ }
+ finally
+ {
+ FileUtils.closeQuietly(primaryIndex);
+ }
+ }
+
+ public RestorableMeter getReadMeter()
+ {
+ return readMeter;
+ }
+
+ public int getIndexSummarySamplingLevel()
+ {
+ return indexSummary.getSamplingLevel();
+ }
+
+ public long getIndexSummaryOffHeapSize()
+ {
+ return indexSummary.getOffHeapSize();
+ }
+
+ public int getMinIndexInterval()
+ {
+ return indexSummary.getMinIndexInterval();
+ }
+
+ public double getEffectiveIndexInterval()
+ {
+ return indexSummary.getEffectiveIndexInterval();
+ }
+
+ public void releaseSummary()
+ {
+ tidy.releaseSummary();
+ indexSummary = null;
+ }
+
+ private void validate()
+ {
+ if (this.first.compareTo(this.last) > 0)
+ {
+ selfRef().release();
+ throw new IllegalStateException(String.format("SSTable first key %s > last key %s", this.first, this.last));
+ }
+ }
+
+ /**
+ * Gets the position in the index file to start scanning to find the given key (at most indexInterval keys away,
+ * modulo downsampling of the index summary). Always returns a value >= 0
+ */
+ public long getIndexScanPosition(RowPosition key)
+ {
+ if (openReason == OpenReason.MOVED_START && key.compareTo(first) < 0)
+ key = first;
+
+ return getIndexScanPositionFromBinarySearchResult(indexSummary.binarySearch(key), indexSummary);
+ }
+
+ @VisibleForTesting
+ public static long getIndexScanPositionFromBinarySearchResult(int binarySearchResult, IndexSummary referencedIndexSummary)
+ {
+ if (binarySearchResult == -1)
+ return 0;
+ else
+ return referencedIndexSummary.getPosition(getIndexSummaryIndexFromBinarySearchResult(binarySearchResult));
+ }
+
+ public static int getIndexSummaryIndexFromBinarySearchResult(int binarySearchResult)
+ {
+ if (binarySearchResult < 0)
+ {
+ // binary search gives us the first index _greater_ than the key searched for,
+ // i.e., its insertion position
+ int greaterThan = (binarySearchResult + 1) * -1;
+ if (greaterThan == 0)
+ return -1;
+ return greaterThan - 1;
+ }
+ else
+ {
+ return binarySearchResult;
+ }
+ }
+
+ /**
+ * Returns the compression metadata for this sstable.
+ * @throws IllegalStateException if the sstable is not compressed
+ */
+ public CompressionMetadata getCompressionMetadata()
+ {
+ if (!compression)
+ throw new IllegalStateException(this + " is not compressed");
+
+ CompressionMetadata cmd = ((ICompressedFile) dfile).getMetadata();
+
+ //We need the parent cf metadata
+ String cfName = metadata.isSecondaryIndex() ? metadata.getParentColumnFamilyName() : metadata.cfName;
+ cmd.parameters.setLiveMetadata(Schema.instance.getCFMetaData(metadata.ksName, cfName));
+
+ return cmd;
+ }
+
+ /**
+ * Returns the amount of memory in bytes used off heap by the compression meta-data.
+ * @return the amount of memory in bytes used off heap by the compression meta-data
+ */
+ public long getCompressionMetadataOffHeapSize()
+ {
+ if (!compression)
+ return 0;
+
+ return getCompressionMetadata().offHeapSize();
+ }
+
+ /**
+ * For testing purposes only.
+ */
+ public void forceFilterFailures()
+ {
+ bf = FilterFactory.AlwaysPresent;
+ }
+
+ public IFilter getBloomFilter()
+ {
+ return bf;
+ }
+
+ public long getBloomFilterSerializedSize()
+ {
+ return bf.serializedSize();
+ }
+
+ /**
+ * Returns the amount of memory in bytes used off heap by the bloom filter.
+ * @return the amount of memory in bytes used off heap by the bloom filter
+ */
+ public long getBloomFilterOffHeapSize()
+ {
+ return bf.offHeapSize();
+ }
+
+ /**
+ * @return An estimate of the number of keys in this SSTable based on the index summary.
+ */
+ public long estimatedKeys()
+ {
+ return indexSummary.getEstimatedKeyCount();
+ }
+
+ /**
+ * @param ranges
+ * @return An estimate of the number of keys for given ranges in this SSTable.
+ */
+ public long estimatedKeysForRanges(Collection<Range<Token>> ranges)
+ {
+ long sampleKeyCount = 0;
+ List<Pair<Integer, Integer>> sampleIndexes = getSampleIndexesForRanges(indexSummary, ranges);
+ for (Pair<Integer, Integer> sampleIndexRange : sampleIndexes)
+ sampleKeyCount += (sampleIndexRange.right - sampleIndexRange.left + 1);
+
+ // adjust for the current sampling level: (BSL / SL) * index_interval_at_full_sampling
+ long estimatedKeys = sampleKeyCount * ((long) Downsampling.BASE_SAMPLING_LEVEL * indexSummary.getMinIndexInterval()) / indexSummary.getSamplingLevel();
+ return Math.max(1, estimatedKeys);
+ }
+
+ /**
+ * Returns the number of entries in the IndexSummary. At full sampling, this is approximately 1/INDEX_INTERVALth of
+ * the keys in this SSTable.
+ */
+ public int getIndexSummarySize()
+ {
+ return indexSummary.size();
+ }
+
+ /**
+ * Returns the approximate number of entries the IndexSummary would contain if it were at full sampling.
+ */
+ public int getMaxIndexSummarySize()
+ {
+ return indexSummary.getMaxNumberOfEntries();
+ }
+
+ /**
+ * Returns the key for the index summary entry at `index`.
+ */
+ public byte[] getIndexSummaryKey(int index)
+ {
+ return indexSummary.getKey(index);
+ }
+
+ private static List<Pair<Integer,Integer>> getSampleIndexesForRanges(IndexSummary summary, Collection<Range<Token>> ranges)
+ {
+ // use the index to determine a minimal section for each range
+ List<Pair<Integer,Integer>> positions = new ArrayList<>();
+
+ for (Range<Token> range : Range.normalize(ranges))
+ {
+ RowPosition leftPosition = range.left.maxKeyBound();
+ RowPosition rightPosition = range.right.maxKeyBound();
+
+ int left = summary.binarySearch(leftPosition);
+ if (left < 0)
+ left = (left + 1) * -1;
+ else
+ // left range are start exclusive
+ left = left + 1;
+ if (left == summary.size())
+ // left is past the end of the sampling
+ continue;
+
+ int right = Range.isWrapAround(range.left, range.right)
+ ? summary.size() - 1
+ : summary.binarySearch(rightPosition);
+ if (right < 0)
+ {
+ // range are end inclusive so we use the previous index from what binarySearch give us
+ // since that will be the last index we will return
+ right = (right + 1) * -1;
+ if (right == 0)
+ // Means the first key is already stricly greater that the right bound
+ continue;
+ right--;
+ }
+
+ if (left > right)
+ // empty range
+ continue;
+ positions.add(Pair.create(left, right));
+ }
+ return positions;
+ }
+
+ public Iterable<DecoratedKey> getKeySamples(final Range<Token> range)
+ {
+ final List<Pair<Integer, Integer>> indexRanges = getSampleIndexesForRanges(indexSummary, Collections.singletonList(range));
+
+ if (indexRanges.isEmpty())
+ return Collections.emptyList();
+
+ return new Iterable<DecoratedKey>()
+ {
+ public Iterator<DecoratedKey> iterator()
+ {
+ return new Iterator<DecoratedKey>()
+ {
+ private Iterator<Pair<Integer, Integer>> rangeIter = indexRanges.iterator();
+ private Pair<Integer, Integer> current;
+ private int idx;
+
+ public boolean hasNext()
+ {
+ if (current == null || idx > current.right)
+ {
+ if (rangeIter.hasNext())
+ {
+ current = rangeIter.next();
+ idx = current.left;
+ return true;
+ }
+ return false;
+ }
+
+ return true;
+ }
+
+ public DecoratedKey next()
+ {
+ byte[] bytes = indexSummary.getKey(idx++);
+ return partitioner.decorateKey(ByteBuffer.wrap(bytes));
+ }
+
+ public void remove()
+ {
+ throw new UnsupportedOperationException();
+ }
+ };
+ }
+ };
+ }
+
+ /**
+ * Determine the minimal set of sections that can be extracted from this SSTable to cover the given ranges.
+ * @return A sorted list of (offset,end) pairs that cover the given ranges in the datafile for this SSTable.
+ */
+ public List<Pair<Long,Long>> getPositionsForRanges(Collection<Range<Token>> ranges)
+ {
+ // use the index to determine a minimal section for each range
+ List<Pair<Long,Long>> positions = new ArrayList<>();
+ for (Range<Token> range : Range.normalize(ranges))
+ {
+ assert !range.isWrapAround() || range.right.isMinimum();
+ // truncate the range so it at most covers the sstable
+ AbstractBounds<RowPosition> bounds = Range.makeRowRange(range);
+ RowPosition leftBound = bounds.left.compareTo(first) > 0 ? bounds.left : first.getToken().minKeyBound();
+ RowPosition rightBound = bounds.right.isMinimum() ? last.getToken().maxKeyBound() : bounds.right;
+
+ if (leftBound.compareTo(last) > 0 || rightBound.compareTo(first) < 0)
+ continue;
+
+ long left = getPosition(leftBound, Operator.GT).position;
+ long right = (rightBound.compareTo(last) > 0)
+ ? uncompressedLength()
+ : getPosition(rightBound, Operator.GT).position;
+
+ if (left == right)
+ // empty range
+ continue;
+
+ assert left < right : String.format("Range=%s openReason=%s first=%s last=%s left=%d right=%d", range, openReason, first, last, left, right);
+ positions.add(Pair.create(left, right));
+ }
+ return positions;
+ }
+
+ public KeyCacheKey getCacheKey(DecoratedKey key)
+ {
+ return new KeyCacheKey(metadata.cfId, descriptor, key.getKey());
+ }
+
+ public void cacheKey(DecoratedKey key, RowIndexEntry info)
+ {
+ CachingOptions caching = metadata.getCaching();
+
+ if (!caching.keyCache.isEnabled()
+ || keyCache == null
+ || keyCache.getCapacity() == 0)
+ {
+ return;
+ }
+
+ KeyCacheKey cacheKey = new KeyCacheKey(metadata.cfId, descriptor, key.getKey());
+ logger.trace("Adding cache entry for {} -> {}", cacheKey, info);
+ keyCache.put(cacheKey, info);
+ }
+
+ public RowIndexEntry getCachedPosition(DecoratedKey key, boolean updateStats)
+ {
+ return getCachedPosition(new KeyCacheKey(metadata.cfId, descriptor, key.getKey()), updateStats);
+ }
+
+ protected RowIndexEntry getCachedPosition(KeyCacheKey unifiedKey, boolean updateStats)
+ {
+ if (keyCache != null && keyCache.getCapacity() > 0) {
+ if (updateStats)
+ {
+ RowIndexEntry cachedEntry = keyCache.get(unifiedKey);
+ keyCacheRequest.incrementAndGet();
+ if (cachedEntry != null)
+ {
+ keyCacheHit.incrementAndGet();
+ bloomFilterTracker.addTruePositive();
+ }
+ return cachedEntry;
+ }
+ else
+ {
+ return keyCache.getInternal(unifiedKey);
+ }
+ }
+ return null;
+ }
+
+ /**
+ * Get position updating key cache and stats.
+ * @see #getPosition(org.apache.cassandra.db.RowPosition, SSTableReader.Operator, boolean)
+ */
+ public RowIndexEntry getPosition(RowPosition key, Operator op)
+ {
+ return getPosition(key, op, true, false);
+ }
+
+ public RowIndexEntry getPosition(RowPosition key, Operator op, boolean updateCacheAndStats)
+ {
+ return getPosition(key, op, updateCacheAndStats, false);
+ }
+ /**
+ * @param key The key to apply as the rhs to the given Operator. A 'fake' key is allowed to
+ * allow key selection by token bounds but only if op != * EQ
+ * @param op The Operator defining matching keys: the nearest key to the target matching the operator wins.
+ * @param updateCacheAndStats true if updating stats and cache
+ * @return The index entry corresponding to the key, or null if the key is not present
+ */
+ protected abstract RowIndexEntry getPosition(RowPosition key, Operator op, boolean updateCacheAndStats, boolean permitMatchPastLast);
+
+ //Corresponds to a name column
+ public abstract OnDiskAtomIterator iterator(DecoratedKey key, SortedSet<CellName> columns);
+ public abstract OnDiskAtomIterator iterator(FileDataInput file, DecoratedKey key, SortedSet<CellName> columns, RowIndexEntry indexEntry);
+
+ //Corresponds to a slice query
+ public abstract OnDiskAtomIterator iterator(DecoratedKey key, ColumnSlice[] slices, boolean reverse);
+ public abstract OnDiskAtomIterator iterator(FileDataInput file, DecoratedKey key, ColumnSlice[] slices, boolean reversed, RowIndexEntry indexEntry);
+
+ /**
+ * Finds and returns the first key beyond a given token in this SSTable or null if no such key exists.
+ */
+ public DecoratedKey firstKeyBeyond(RowPosition token)
+ {
+ if (token.compareTo(first) < 0)
+ return first;
+
+ long sampledPosition = getIndexScanPosition(token);
+
++ if (ifile == null)
++ return null;
++
+ Iterator<FileDataInput> segments = ifile.iterator(sampledPosition);
+ while (segments.hasNext())
+ {
+ String path = null;
+ try (FileDataInput in = segments.next();)
+ {
+ path = in.getPath();
+ while (!in.isEOF())
+ {
+ ByteBuffer indexKey = ByteBufferUtil.readWithShortLength(in);
+ DecoratedKey indexDecoratedKey = partitioner.decorateKey(indexKey);
+ if (indexDecoratedKey.compareTo(token) > 0)
+ return indexDecoratedKey;
+
+ RowIndexEntry.Serializer.skip(in);
+ }
+ }
+ catch (IOException e)
+ {
+ markSuspect();
+ throw new CorruptSSTableException(e, path);
+ }
+ }
+
+ return null;
+ }
+
+ /**
+ * @return The length in bytes of the data for this SSTable. For
+ * compressed files, this is not the same thing as the on disk size (see
+ * onDiskLength())
+ */
+ public long uncompressedLength()
+ {
+ return dfile.length;
+ }
+
+ /**
+ * @return The length in bytes of the on disk size for this SSTable. For
+ * compressed files, this is not the same thing as the data length (see
+ * length())
+ */
+ public long onDiskLength()
+ {
+ return dfile.onDiskLength;
+ }
+
+ /**
+ * Mark the sstable as obsolete, i.e., compacted into newer sstables.
+ *
+ * When calling this function, the caller must ensure that the SSTableReader is not referenced anywhere
+ * except for threads holding a reference.
+ *
+ * @return true if the this is the first time the file was marked obsolete. Calling this
+ * multiple times is usually buggy (see exceptions in Tracker.unmarkCompacting and removeOldSSTablesSize).
+ */
+ public boolean markObsolete(Tracker tracker)
+ {
+ if (logger.isDebugEnabled())
+ logger.debug("Marking {} compacted", getFilename());
+
+ synchronized (tidy.global)
+ {
+ assert !tidy.isReplaced;
+ }
+ if (!tidy.global.isCompacted.getAndSet(true))
+ {
+ tidy.type.markObsolete(this, tracker);
+ return true;
+ }
+ return false;
+ }
+
+ public boolean isMarkedCompacted()
+ {
+ return tidy.global.isCompacted.get();
+ }
+
+ public void markSuspect()
+ {
+ if (logger.isDebugEnabled())
+ logger.debug("Marking {} as a suspect for blacklisting.", getFilename());
+
+ isSuspect.getAndSet(true);
+ }
+
+ public boolean isMarkedSuspect()
+ {
+ return isSuspect.get();
+ }
+
+
+ /**
+ * I/O SSTableScanner
+ * @return A Scanner for seeking over the rows of the SSTable.
+ */
+ public ISSTableScanner getScanner()
+ {
+ return getScanner((RateLimiter) null);
+ }
+
+ public ISSTableScanner getScanner(RateLimiter limiter)
+ {
+ return getScanner(DataRange.allData(partitioner), limiter);
+ }
+
+ /**
+ *
+ * @param dataRange filter to use when reading the columns
+ * @return A Scanner for seeking over the rows of the SSTable.
+ */
+ public ISSTableScanner getScanner(DataRange dataRange)
+ {
+ return getScanner(dataRange, null);
+ }
+
+ /**
+ * Direct I/O SSTableScanner over a defined range of tokens.
+ *
+ * @param range the range of keys to cover
+ * @return A Scanner for seeking over the rows of the SSTable.
+ */
+ public ISSTableScanner getScanner(Range<Token> range, RateLimiter limiter)
+ {
+ if (range == null)
+ return getScanner(limiter);
+ return getScanner(Collections.singletonList(range), limiter);
+ }
+
+ /**
+ * Direct I/O SSTableScanner over a defined collection of ranges of tokens.
+ *
+ * @param ranges the range of keys to cover
+ * @return A Scanner for seeking over the rows of the SSTable.
+ */
+ public abstract ISSTableScanner getScanner(Collection<Range<Token>> ranges, RateLimiter limiter);
+
+ /**
+ *
+ * @param dataRange filter to use when reading the columns
+ * @return A Scanner for seeking over the rows of the SSTable.
+ */
+ public abstract ISSTableScanner getScanner(DataRange dataRange, RateLimiter limiter);
+
+
+
+ public FileDataInput getFileDataInput(long position)
+ {
+ return dfile.getSegment(position);
+ }
+
+ /**
+ * Tests if the sstable contains data newer than the given age param (in localhost currentMilli time).
+ * This works in conjunction with maxDataAge which is an upper bound on the create of data in this sstable.
+ * @param age The age to compare the maxDataAre of this sstable. Measured in millisec since epoc on this host
+ * @return True iff this sstable contains data that's newer than the given age parameter.
+ */
+ public boolean newSince(long age)
+ {
+ return maxDataAge > age;
+ }
+
+ public void createLinks(String snapshotDirectoryPath)
+ {
+ for (Component component : components)
+ {
+ File sourceFile = new File(descriptor.filenameFor(component));
+ File targetLink = new File(snapshotDirectoryPath, sourceFile.getName());
+ FileUtils.createHardLink(sourceFile, targetLink);
+ }
+ }
+
+ public boolean isRepaired()
+ {
+ return sstableMetadata.repairedAt != ActiveRepairService.UNREPAIRED_SSTABLE;
+ }
+
+ /**
+ * TODO: Move someplace reusable
+ */
+ public abstract static class Operator
+ {
+ public static final Operator EQ = new Equals();
+ public static final Operator GE = new GreaterThanOrEqualTo();
+ public static final Operator GT = new GreaterThan();
+
+ /**
+ * @param comparison The result of a call to compare/compareTo, with the desired field on the rhs.
+ * @return less than 0 if the operator cannot match forward, 0 if it matches, greater than 0 if it might match forward.
+ */
+ public abstract int apply(int comparison);
+
+ final static class Equals extends Operator
+ {
+ public int apply(int comparison) { return -comparison; }
+ }
+
+ final static class GreaterThanOrEqualTo extends Operator
+ {
+ public int apply(int comparison) { return comparison >= 0 ? 0 : 1; }
+ }
+
+ final static class GreaterThan extends Operator
+ {
+ public int apply(int comparison) { return comparison > 0 ? 0 : 1; }
+ }
+ }
+
+ public long getBloomFilterFalsePositiveCount()
+ {
+ return bloomFilterTracker.getFalsePositiveCount();
+ }
+
+ public long getRecentBloomFilterFalsePositiveCount()
+ {
+ return bloomFilterTracker.getRecentFalsePositiveCount();
+ }
+
+ public long getBloomFilterTruePositiveCount()
+ {
+ return bloomFilterTracker.getTruePositiveCount();
+ }
+
+ public long getRecentBloomFilterTruePositiveCount()
+ {
+ return bloomFilterTracker.getRecentTruePositiveCount();
+ }
+
+ public InstrumentingCache<KeyCacheKey, RowIndexEntry> getKeyCache()
+ {
+ return keyCache;
+ }
+
+ public EstimatedHistogram getEstimatedRowSize()
+ {
+ return sstableMetadata.estimatedRowSize;
+ }
+
+ public EstimatedHistogram getEstimatedColumnCount()
+ {
+ return sstableMetadata.estimatedColumnCount;
+ }
+
+ public double getEstimatedDroppableTombstoneRatio(int gcBefore)
+ {
+ return sstableMetadata.getEstimatedDroppableTombstoneRatio(gcBefore);
+ }
+
+ public double getDroppableTombstonesBefore(int gcBefore)
+ {
+ return sstableMetadata.getDroppableTombstonesBefore(gcBefore);
+ }
+
+ public double getCompressionRatio()
+ {
+ return sstableMetadata.compressionRatio;
+ }
+
+ public ReplayPosition getReplayPosition()
+ {
+ return sstableMetadata.replayPosition;
+ }
+
+ public long getMinTimestamp()
+ {
+ return sstableMetadata.minTimestamp;
+ }
+
+ public long getMaxTimestamp()
+ {
+ return sstableMetadata.maxTimestamp;
+ }
+
+ public Set<Integer> getAncestors()
+ {
+ try
+ {
+ CompactionMetadata compactionMetadata = (CompactionMetadata) descriptor.getMetadataSerializer().deserialize(descriptor, MetadataType.COMPACTION);
- return compactionMetadata.ancestors;
++ if (compactionMetadata != null)
++ return compactionMetadata.ancestors;
++ return Collections.emptySet();
+ }
+ catch (IOException e)
+ {
+ SSTableReader.logOpenException(descriptor, e);
+ return Collections.emptySet();
+ }
+ }
+
+ public int getSSTableLevel()
+ {
+ return sstableMetadata.sstableLevel;
+ }
+
+ /**
+ * Reloads the sstable metadata from disk.
+ *
+ * Called after level is changed on sstable, for example if the sstable is dropped to L0
+ *
+ * Might be possible to remove in future versions
+ *
+ * @throws IOException
+ */
+ public void reloadSSTableMetadata() throws IOException
+ {
+ this.sstableMetadata = (StatsMetadata) descriptor.getMetadataSerializer().deserialize(descriptor, MetadataType.STATS);
+ }
+
+ public StatsMetadata getSSTableMetadata()
+ {
+ return sstableMetadata;
+ }
+
+ public RandomAccessReader openDataReader(RateLimiter limiter)
+ {
+ assert limiter != null;
+ return dfile.createThrottledReader(limiter);
+ }
+
+ public RandomAccessReader openDataReader()
+ {
+ return dfile.createReader();
+ }
+
+ public RandomAccessReader openIndexReader()
+ {
- return ifile.createReader();
++ if (ifile != null)
++ return ifile.createReader();
++ return null;
+ }
+
+ /**
+ * @param component component to get timestamp.
+ * @return last modified time for given component. 0 if given component does not exist or IO error occurs.
+ */
+ public long getCreationTimeFor(Component component)
+ {
+ return new File(descriptor.filenameFor(component)).lastModified();
+ }
+
+ /**
+ * @return Number of key cache hit
+ */
+ public long getKeyCacheHit()
+ {
+ return keyCacheHit.get();
+ }
+
+ /**
+ * @return Number of key cache request
+ */
+ public long getKeyCacheRequest()
+ {
+ return keyCacheRequest.get();
+ }
+
+ /**
+ * Increment the total row read count and read rate for this SSTable. This should not be incremented for range
+ * slice queries, row cache hits, or non-query reads, like compaction.
+ */
+ public void incrementReadCount()
+ {
+ if (readMeter != null)
+ readMeter.mark();
+ }
+
+ public static class SizeComparator implements Comparator<SSTableReader>
+ {
+ public int compare(SSTableReader o1, SSTableReader o2)
+ {
+ return Longs.compare(o1.onDiskLength(), o2.onDiskLength());
+ }
+ }
+
+ public Ref<SSTableReader> tryRef()
+ {
+ return selfRef.tryRef();
+ }
+
+ public Ref<SSTableReader> selfRef()
+ {
+ return selfRef;
+ }
+
+ public Ref<SSTableReader> ref()
+ {
+ return selfRef.ref();
+ }
+
+ void setup(boolean isOffline)
+ {
+ tidy.setup(this, isOffline);
+ this.readMeter = tidy.global.readMeter;
+ }
+
+ @VisibleForTesting
+ public void overrideReadMeter(RestorableMeter readMeter)
+ {
+ this.readMeter = tidy.global.readMeter = readMeter;
+ }
+
+ /**
+ * One instance per SSTableReader we create. This references the type-shared tidy, which in turn references
+ * the globally shared tidy, i.e.
+ *
+ * InstanceTidier => DescriptorTypeTitdy => GlobalTidy
+ *
+ * We can create many InstanceTidiers (one for every time we reopen an sstable with MOVED_START for example), but there can only be
+ * two DescriptorTypeTidy (FINAL and TEMPLINK) and only one GlobalTidy for one single logical sstable.
+ *
+ * When the InstanceTidier cleansup, it releases its reference to its DescriptorTypeTidy; when all InstanceTidiers
+ * for that type have run, the DescriptorTypeTidy cleansup. DescriptorTypeTidy behaves in the same way towards GlobalTidy.
+ *
+ * For ease, we stash a direct reference to both our type-shared and global tidier
+ */
+ private static final class InstanceTidier implements Tidy
+ {
+ private final Descriptor descriptor;
+ private final CFMetaData metadata;
+ private IFilter bf;
+ private IndexSummary summary;
+
+ private SegmentedFile dfile;
+ private SegmentedFile ifile;
+ private Runnable runOnClose;
+ private boolean isReplaced = false;
+
+ // a reference to our shared per-Descriptor.Type tidy instance, that
+ // we will release when we are ourselves released
+ private Ref<DescriptorTypeTidy> typeRef;
+
+ // a convenience stashing of the shared per-descriptor-type tidy instance itself
+ // and the per-logical-sstable globally shared state that it is linked to
+ private DescriptorTypeTidy type;
+ private GlobalTidy global;
+
+ private boolean setup;
+
+ void setup(SSTableReader reader, boolean isOffline)
+ {
+ this.setup = true;
+ this.bf = reader.bf;
+ this.summary = reader.indexSummary;
+ this.dfile = reader.dfile;
+ this.ifile = reader.ifile;
+ // get a new reference to the shared descriptor-type tidy
+ this.typeRef = DescriptorTypeTidy.get(reader);
+ this.type = typeRef.get();
+ this.global = type.globalRef.get();
+ if (!isOffline)
+ global.ensureReadMeter();
+ }
+
+ InstanceTidier(Descriptor descriptor, CFMetaData metadata)
+ {
+ this.descriptor = descriptor;
+ this.metadata = metadata;
+ }
+
+ public void tidy()
+ {
+ // don't try to cleanup if the sstablereader was never fully constructed
+ if (!setup)
+ return;
+
+ final ColumnFamilyStore cfs = Schema.instance.getColumnFamilyStoreInstance(metadata.cfId);
+ final OpOrder.Barrier barrier;
+ if (cfs != null)
+ {
+ barrier = cfs.readOrdering.newBarrier();
+ barrier.issue();
+ }
+ else
+ barrier = null;
+
+ ScheduledExecutors.nonPeriodicTasks.execute(new Runnable()
+ {
+ public void run()
+ {
+ if (barrier != null)
+ barrier.await();
- bf.close();
++ if (bf != null)
++ bf.close();
+ if (summary != null)
+ summary.close();
+ if (runOnClose != null)
+ runOnClose.run();
+ dfile.close();
- ifile.close();
++ if (ifile != null)
++ ifile.close();
+ typeRef.release();
+ }
+ });
+ }
+
+ public String name()
+ {
+ return descriptor.toString();
+ }
+
+ void releaseSummary()
+ {
+ summary.close();
+ assert summary.isCleanedUp();
+ summary = null;
+ }
+ }
+
+ /**
+ * One shared between all instances of a given Descriptor.Type.
+ * Performs only two things: the deletion of the sstables for the type,
+ * if necessary; and the shared reference to the globally shared state.
+ *
+ * All InstanceTidiers, on setup(), ask the static get() method for their shared state,
+ * and stash a reference to it to be released when they are. Once all such references are
+ * released, the shared tidy will be performed.
+ */
+ static final class DescriptorTypeTidy implements Tidy
+ {
+ // keyed by REAL descriptor (TMPLINK/FINAL), mapping to the shared DescriptorTypeTidy for that descriptor
+ static final ConcurrentMap<Descriptor, Ref<DescriptorTypeTidy>> lookup = new ConcurrentHashMap<>();
+
+ private final Descriptor desc;
+ private final Ref<GlobalTidy> globalRef;
+ private final Set<Component> components;
+ private long sizeOnDelete;
+ private Counter totalDiskSpaceUsed;
+
+ DescriptorTypeTidy(Descriptor desc, SSTableReader sstable)
+ {
+ this.desc = desc;
+ // get a new reference to the shared global tidy
+ this.globalRef = GlobalTidy.get(sstable);
+ this.components = sstable.components;
+ }
+
+ void markObsolete(SSTableReader instance, Tracker tracker)
+ {
+ // the tracker is used only to notify listeners of deletion of the sstable;
+ // since deletion of a non-final file is not really deletion of the sstable,
+ // we don't want to notify the listeners in this event
+ if (tracker != null && tracker.cfstore != null && desc.type == Descriptor.Type.FINAL)
+ {
+ sizeOnDelete = instance.bytesOnDisk();
+ totalDiskSpaceUsed = tracker.cfstore.metric.totalDiskSpaceUsed;
+ tracker.notifyDeleting(instance);
+ }
+ }
+
+ public void tidy()
+ {
+ lookup.remove(desc);
+ boolean isCompacted = globalRef.get().isCompacted.get();
+ globalRef.release();
+ switch (desc.type)
+ {
+ case FINAL:
+ if (isCompacted)
+ new SSTableDeletingTask(desc, components, totalDiskSpaceUsed, sizeOnDelete).run();
+ break;
+ case TEMPLINK:
+ new SSTableDeletingTask(desc, components, null, 0).run();
+ break;
+ default:
+ throw new IllegalStateException();
+ }
+ }
+
+ public String name()
+ {
+ return desc.toString();
+ }
+
+ // get a new reference to the shared DescriptorTypeTidy for this sstable
+ @SuppressWarnings("resource")
+ public static Ref<DescriptorTypeTidy> get(SSTableReader sstable)
+ {
+ Descriptor desc = sstable.descriptor;
+ if (sstable.openReason == OpenReason.EARLY)
+ desc = desc.asType(Descriptor.Type.TEMPLINK);
+ Ref<DescriptorTypeTidy> refc = lookup.get(desc);
+ if (refc != null)
+ return refc.ref();
+ final DescriptorTypeTidy tidy = new DescriptorTypeTidy(desc, sstable);
+ refc = new Ref<>(tidy, tidy);
+ Ref<?> ex = lookup.putIfAbsent(desc, refc);
+ if (ex != null)
+ {
+ refc.close();
+ throw new AssertionError();
+ }
+ return refc;
+ }
+ }
+
+ /**
+ * One instance per logical sstable. This both tracks shared cleanup and some shared state related
+ * to the sstable's lifecycle. All DescriptorTypeTidy instances, on construction, obtain a reference to us
+ * via our static get(). There should only ever be at most two such references extant at any one time,
+ * since only TMPLINK and FINAL type descriptors should be open as readers. When all files of both
+ * kinds have been released, this shared tidy will be performed.
+ */
+ static final class GlobalTidy implements Tidy
+ {
+ // keyed by FINAL descriptor, mapping to the shared GlobalTidy for that descriptor
+ static final ConcurrentMap<Descriptor, Ref<GlobalTidy>> lookup = new ConcurrentHashMap<>();
+
+ private final Descriptor desc;
+ // the readMeter that is shared between all instances of the sstable, and can be overridden in all of them
+ // at once also, for testing purposes
+ private RestorableMeter readMeter;
+ // the scheduled persistence of the readMeter, that we will cancel once all instances of this logical
+ /
<TRUNCATED>
http://git-wip-us.apache.org/repos/asf/cassandra/blob/ebe18bb2/src/java/org/apache/cassandra/tools/StandaloneScrubber.java
----------------------------------------------------------------------
http://git-wip-us.apache.org/repos/asf/cassandra/blob/ebe18bb2/test/unit/org/apache/cassandra/db/ScrubTest.java
----------------------------------------------------------------------
[02/12] cassandra git commit: Scrub (recover) sstables even when
-Index.db is missing
Posted by be...@apache.org.
Scrub (recover) sstables even when -Index.db is missing
patch by mck; reviewed by stefania for CASSANDRA-9591
Project: http://git-wip-us.apache.org/repos/asf/cassandra/repo
Commit: http://git-wip-us.apache.org/repos/asf/cassandra/commit/452d6a44
Tree: http://git-wip-us.apache.org/repos/asf/cassandra/tree/452d6a44
Diff: http://git-wip-us.apache.org/repos/asf/cassandra/diff/452d6a44
Branch: refs/heads/cassandra-2.1
Commit: 452d6a445a6935d3e7d0e0fdf59e87d2a2ff95e7
Parents: 62714a9
Author: Stefania Alborghetti <st...@datastax.com>
Authored: Mon Jun 15 16:49:03 2015 +0800
Committer: Benedict Elliott Smith <be...@apache.org>
Committed: Tue Jul 7 14:27:25 2015 +0100
----------------------------------------------------------------------
CHANGES.txt | 4 ++
.../cassandra/db/compaction/Scrubber.java | 36 +++++++++++----
.../cassandra/io/sstable/SSTableReader.java | 47 ++++++++++++++++----
.../cassandra/tools/StandaloneScrubber.java | 2 +-
.../unit/org/apache/cassandra/db/ScrubTest.java | 25 +++++++++++
5 files changed, 96 insertions(+), 18 deletions(-)
----------------------------------------------------------------------
http://git-wip-us.apache.org/repos/asf/cassandra/blob/452d6a44/CHANGES.txt
----------------------------------------------------------------------
diff --git a/CHANGES.txt b/CHANGES.txt
index ca4d4b5..bd1db92 100644
--- a/CHANGES.txt
+++ b/CHANGES.txt
@@ -1,3 +1,7 @@
+2.0.18
+* Scrub (recover) sstables even when -Index.db is missing, (CASSANDRA-9591)
+
+
2.0.17
* Avoid NPE in AuthSuccess#decode (CASSANDRA-9727)
* Add listen_address to system.local (CASSANDRA-9603)
http://git-wip-us.apache.org/repos/asf/cassandra/blob/452d6a44/src/java/org/apache/cassandra/db/compaction/Scrubber.java
----------------------------------------------------------------------
diff --git a/src/java/org/apache/cassandra/db/compaction/Scrubber.java b/src/java/org/apache/cassandra/db/compaction/Scrubber.java
index ea10855..dc60efa 100644
--- a/src/java/org/apache/cassandra/db/compaction/Scrubber.java
+++ b/src/java/org/apache/cassandra/db/compaction/Scrubber.java
@@ -96,7 +96,16 @@ public class Scrubber implements Closeable
? new ScrubController(cfs)
: new CompactionController(cfs, Collections.singleton(sstable), CompactionManager.getDefaultGcBefore(cfs));
this.isCommutative = cfs.metadata.getDefaultValidator().isCommutative();
- this.expectedBloomFilterSize = Math.max(cfs.metadata.getIndexInterval(), (int)(SSTableReader.getApproximateKeyCount(toScrub,cfs.metadata)));
+
+ boolean hasIndexFile = (new File(sstable.descriptor.filenameFor(Component.PRIMARY_INDEX))).exists();
+ if (!hasIndexFile)
+ {
+ // if there's any corruption in the -Data.db then rows can't be skipped over. but it's worth a shot.
+ outputHandler.warn("Missing component: " + sstable.descriptor.filenameFor(Component.PRIMARY_INDEX));
+ }
+
+ this.expectedBloomFilterSize = Math.max(cfs.metadata.getIndexInterval(),
+ hasIndexFile ? (int)(SSTableReader.getApproximateKeyCount(toScrub,cfs.metadata)) : 0);
// loop through each row, deserializing to check for damage.
// we'll also loop through the index at the same time, using the position from the index to recover if the
@@ -105,7 +114,11 @@ public class Scrubber implements Closeable
this.dataFile = isOffline
? sstable.openDataReader()
: sstable.openDataReader(CompactionManager.instance.getRateLimiter());
- this.indexFile = RandomAccessReader.open(new File(sstable.descriptor.filenameFor(Component.PRIMARY_INDEX)));
+
+ this.indexFile = hasIndexFile
+ ? RandomAccessReader.open(new File(sstable.descriptor.filenameFor(Component.PRIMARY_INDEX)))
+ : null;
+
this.scrubInfo = new ScrubInfo(dataFile, sstable);
this.currentRowPositionFromIndex = 0;
@@ -117,7 +130,8 @@ public class Scrubber implements Closeable
outputHandler.output(String.format("Scrubbing %s (%s bytes)", sstable, dataFile.length()));
try
{
- nextIndexKey = ByteBufferUtil.readWithShortLength(indexFile);
+ nextIndexKey = indexAvailable() ? ByteBufferUtil.readWithShortLength(indexFile) : null;
+ if (indexAvailable())
{
// throw away variable so we don't have a side effect in the assert
long firstRowPositionFromIndex = RowIndexEntry.serializer.deserialize(indexFile, sstable.descriptor.version).position;
@@ -181,7 +195,7 @@ public class Scrubber implements Closeable
outputHandler.debug(String.format("Index doublecheck: row %s is %s bytes", ByteBufferUtil.bytesToHex(currentIndexKey), dataSizeFromIndex));
}
- assert currentIndexKey != null || indexFile.isEOF();
+ assert currentIndexKey != null || !indexAvailable();
writer.mark();
try
@@ -198,10 +212,10 @@ public class Scrubber implements Closeable
if (dataSize > dataFile.length())
throw new IOError(new IOException("Impossible row size (greater than file length): " + dataSize));
- if (dataStart != dataStartFromIndex)
+ if (indexFile != null && dataStart != dataStartFromIndex)
outputHandler.warn(String.format("Data file row position %d differs from index file row position %d", dataStart, dataStartFromIndex));
- if (dataSize != dataSizeFromIndex)
+ if (indexFile != null && dataSize != dataSizeFromIndex)
outputHandler.warn(String.format("Data file row size %d differs from index file row size %d", dataSize, dataSizeFromIndex));
SSTableIdentityIterator atoms = new SSTableIdentityIterator(sstable, dataFile, key, dataSize, validateColumns);
@@ -317,8 +331,9 @@ public class Scrubber implements Closeable
currentRowPositionFromIndex = nextRowPositionFromIndex;
try
{
- nextIndexKey = indexFile.isEOF() ? null : ByteBufferUtil.readWithShortLength(indexFile);
- nextRowPositionFromIndex = indexFile.isEOF()
+ nextIndexKey = !indexAvailable() ? null : ByteBufferUtil.readWithShortLength(indexFile);
+
+ nextRowPositionFromIndex = !indexAvailable()
? dataFile.length()
: RowIndexEntry.serializer.deserialize(indexFile, sstable.descriptor.version).position;
}
@@ -330,6 +345,11 @@ public class Scrubber implements Closeable
}
}
+ private boolean indexAvailable()
+ {
+ return indexFile != null && !indexFile.isEOF();
+ }
+
private void seekToNextRow()
{
while(nextRowPositionFromIndex < dataFile.length())
http://git-wip-us.apache.org/repos/asf/cassandra/blob/452d6a44/src/java/org/apache/cassandra/io/sstable/SSTableReader.java
----------------------------------------------------------------------
diff --git a/src/java/org/apache/cassandra/io/sstable/SSTableReader.java b/src/java/org/apache/cassandra/io/sstable/SSTableReader.java
index 15808e8..8919a09 100644
--- a/src/java/org/apache/cassandra/io/sstable/SSTableReader.java
+++ b/src/java/org/apache/cassandra/io/sstable/SSTableReader.java
@@ -157,7 +157,7 @@ public class SSTableReader extends SSTable implements Closeable
public static SSTableReader openForBatch(Descriptor descriptor, Set<Component> components, CFMetaData metadata, IPartitioner partitioner) throws IOException
{
- SSTableMetadata sstableMetadata = openMetadata(descriptor, components, partitioner);
+ SSTableMetadata sstableMetadata = openMetadata(descriptor, components, partitioner, true);
SSTableReader sstable = new SSTableReader(descriptor,
components,
metadata,
@@ -191,7 +191,7 @@ public class SSTableReader extends SSTable implements Closeable
boolean validate) throws IOException
{
long start = System.nanoTime();
- SSTableMetadata sstableMetadata = openMetadata(descriptor, components, partitioner);
+ SSTableMetadata sstableMetadata = openMetadata(descriptor, components, partitioner, validate);
SSTableReader sstable = new SSTableReader(descriptor,
components,
@@ -213,12 +213,17 @@ public class SSTableReader extends SSTable implements Closeable
return sstable;
}
- private static SSTableMetadata openMetadata(Descriptor descriptor, Set<Component> components, IPartitioner partitioner) throws IOException
+ private static SSTableMetadata openMetadata(Descriptor descriptor,
+ Set<Component> components,
+ IPartitioner partitioner,
+ boolean primaryIndexRequired) throws IOException
{
assert partitioner != null;
// Minimum components without which we can't do anything
assert components.contains(Component.DATA) : "Data component is missing for sstable" + descriptor;
- assert components.contains(Component.PRIMARY_INDEX) : "Primary index component is missing for sstable " + descriptor;
+
+ assert !primaryIndexRequired || components.contains(Component.PRIMARY_INDEX)
+ : "Primary index component is missing for sstable " + descriptor;
logger.info("Opening {} ({} bytes)", descriptor, new File(descriptor.filenameFor(COMPONENT_DATA)).length());
@@ -382,11 +387,17 @@ public class SSTableReader extends SSTable implements Closeable
readMeterSyncFuture.cancel(false);
// Force finalizing mmapping if necessary
- ifile.cleanup();
+
+ if (null != ifile)
+ ifile.cleanup();
+
dfile.cleanup();
// close the BF so it can be opened later.
- bf.close();
- indexSummary.close();
+ if (null != bf)
+ bf.close();
+
+ if (null != indexSummary)
+ indexSummary.close();
}
public void setTrackedBy(DataTracker tracker)
@@ -406,6 +417,12 @@ public class SSTableReader extends SSTable implements Closeable
load(false, true);
bf = FilterFactory.AlwaysPresent;
}
+ else if (!components.contains(Component.PRIMARY_INDEX))
+ {
+ // avoid any reading of the missing primary index component.
+ // this should only happen during StandaloneScrubber
+ load(false, false);
+ }
else if (!components.contains(Component.FILTER))
{
// bf is enabled, but filter component is missing.
@@ -454,7 +471,9 @@ public class SSTableReader extends SSTable implements Closeable
if (recreateBloomFilter || !summaryLoaded)
buildSummary(recreateBloomFilter, ibuilder, dbuilder, summaryLoaded);
- ifile = ibuilder.complete(descriptor.filenameFor(Component.PRIMARY_INDEX));
+ if (components.contains(Component.PRIMARY_INDEX))
+ ifile = ibuilder.complete(descriptor.filenameFor(Component.PRIMARY_INDEX));
+
dfile = dbuilder.complete(descriptor.filenameFor(Component.DATA));
if (saveSummaryIfCreated && (recreateBloomFilter || !summaryLoaded)) // save summary information to disk
saveSummary(this, ibuilder, dbuilder);
@@ -462,6 +481,9 @@ public class SSTableReader extends SSTable implements Closeable
private void buildSummary(boolean recreateBloomFilter, SegmentedFile.Builder ibuilder, SegmentedFile.Builder dbuilder, boolean summaryLoaded) throws IOException
{
+ if (!components.contains(Component.PRIMARY_INDEX))
+ return;
+
// we read the positions in a BRAF so we don't have to worry about an entry spanning a mmap boundary.
RandomAccessReader primaryIndex = RandomAccessReader.open(new File(descriptor.filenameFor(Component.PRIMARY_INDEX)));
@@ -965,6 +987,9 @@ public class SSTableReader extends SSTable implements Closeable
}
}
+ if (ifile == null)
+ return null;
+
// scan the on-disk index, starting at the nearest sampled position.
// The check against IndexInterval is to be exit the loop in the EQ case when the key looked for is not present
// (bloom filter false positive). But note that for non-EQ cases, we might need to check the first key of the
@@ -1063,6 +1088,9 @@ public class SSTableReader extends SSTable implements Closeable
if (sampledPosition == -1)
sampledPosition = 0;
+ if (ifile == null)
+ return null;
+
Iterator<FileDataInput> segments = ifile.iterator(sampledPosition);
while (segments.hasNext())
{
@@ -1480,7 +1508,8 @@ public class SSTableReader extends SSTable implements Closeable
private void dropPageCache()
{
dropPageCache(dfile.path);
- dropPageCache(ifile.path);
+ if (null != ifile)
+ dropPageCache(ifile.path);
}
private void dropPageCache(String filePath)
http://git-wip-us.apache.org/repos/asf/cassandra/blob/452d6a44/src/java/org/apache/cassandra/tools/StandaloneScrubber.java
----------------------------------------------------------------------
diff --git a/src/java/org/apache/cassandra/tools/StandaloneScrubber.java b/src/java/org/apache/cassandra/tools/StandaloneScrubber.java
index 26768f3..8ace228 100644
--- a/src/java/org/apache/cassandra/tools/StandaloneScrubber.java
+++ b/src/java/org/apache/cassandra/tools/StandaloneScrubber.java
@@ -79,7 +79,7 @@ public class StandaloneScrubber
for (Map.Entry<Descriptor, Set<Component>> entry : lister.list().entrySet())
{
Set<Component> components = entry.getValue();
- if (!components.contains(Component.DATA) || !components.contains(Component.PRIMARY_INDEX))
+ if (!components.contains(Component.DATA))
continue;
try
http://git-wip-us.apache.org/repos/asf/cassandra/blob/452d6a44/test/unit/org/apache/cassandra/db/ScrubTest.java
----------------------------------------------------------------------
diff --git a/test/unit/org/apache/cassandra/db/ScrubTest.java b/test/unit/org/apache/cassandra/db/ScrubTest.java
index 94c7e34..978b3e8 100644
--- a/test/unit/org/apache/cassandra/db/ScrubTest.java
+++ b/test/unit/org/apache/cassandra/db/ScrubTest.java
@@ -270,6 +270,31 @@ public class ScrubTest extends SchemaLoader
}
@Test
+ public void testScrubNoIndex() throws IOException, ExecutionException, InterruptedException, ConfigurationException
+ {
+ CompactionManager.instance.disableAutoCompaction();
+ Keyspace keyspace = Keyspace.open(KEYSPACE);
+ ColumnFamilyStore cfs = keyspace.getColumnFamilyStore(CF);
+ cfs.clearUnsafe();
+
+ List<Row> rows;
+
+ // insert data and verify we get it back w/ range query
+ fillCF(cfs, 10);
+ rows = cfs.getRangeSlice(Util.range("", ""), null, new IdentityQueryFilter(), 1000);
+ assertEquals(10, rows.size());
+
+ for (SSTableReader sstable : cfs.getSSTables())
+ new File(sstable.descriptor.filenameFor(Component.PRIMARY_INDEX)).delete();
+
+ CompactionManager.instance.performScrub(cfs, false, true);
+
+ // check data is still there
+ rows = cfs.getRangeSlice(Util.range("", ""), null, new IdentityQueryFilter(), 1000);
+ assertEquals(10, rows.size());
+ }
+
+ @Test
public void testScrubOutOfOrder() throws Exception
{
CompactionManager.instance.disableAutoCompaction();
[04/12] cassandra git commit: Scrub (recover) sstables even when
-Index.db is missing
Posted by be...@apache.org.
Scrub (recover) sstables even when -Index.db is missing
patch by mck; reviewed by stefania for CASSANDRA-9591
Project: http://git-wip-us.apache.org/repos/asf/cassandra/repo
Commit: http://git-wip-us.apache.org/repos/asf/cassandra/commit/452d6a44
Tree: http://git-wip-us.apache.org/repos/asf/cassandra/tree/452d6a44
Diff: http://git-wip-us.apache.org/repos/asf/cassandra/diff/452d6a44
Branch: refs/heads/trunk
Commit: 452d6a445a6935d3e7d0e0fdf59e87d2a2ff95e7
Parents: 62714a9
Author: Stefania Alborghetti <st...@datastax.com>
Authored: Mon Jun 15 16:49:03 2015 +0800
Committer: Benedict Elliott Smith <be...@apache.org>
Committed: Tue Jul 7 14:27:25 2015 +0100
----------------------------------------------------------------------
CHANGES.txt | 4 ++
.../cassandra/db/compaction/Scrubber.java | 36 +++++++++++----
.../cassandra/io/sstable/SSTableReader.java | 47 ++++++++++++++++----
.../cassandra/tools/StandaloneScrubber.java | 2 +-
.../unit/org/apache/cassandra/db/ScrubTest.java | 25 +++++++++++
5 files changed, 96 insertions(+), 18 deletions(-)
----------------------------------------------------------------------
http://git-wip-us.apache.org/repos/asf/cassandra/blob/452d6a44/CHANGES.txt
----------------------------------------------------------------------
diff --git a/CHANGES.txt b/CHANGES.txt
index ca4d4b5..bd1db92 100644
--- a/CHANGES.txt
+++ b/CHANGES.txt
@@ -1,3 +1,7 @@
+2.0.18
+* Scrub (recover) sstables even when -Index.db is missing, (CASSANDRA-9591)
+
+
2.0.17
* Avoid NPE in AuthSuccess#decode (CASSANDRA-9727)
* Add listen_address to system.local (CASSANDRA-9603)
http://git-wip-us.apache.org/repos/asf/cassandra/blob/452d6a44/src/java/org/apache/cassandra/db/compaction/Scrubber.java
----------------------------------------------------------------------
diff --git a/src/java/org/apache/cassandra/db/compaction/Scrubber.java b/src/java/org/apache/cassandra/db/compaction/Scrubber.java
index ea10855..dc60efa 100644
--- a/src/java/org/apache/cassandra/db/compaction/Scrubber.java
+++ b/src/java/org/apache/cassandra/db/compaction/Scrubber.java
@@ -96,7 +96,16 @@ public class Scrubber implements Closeable
? new ScrubController(cfs)
: new CompactionController(cfs, Collections.singleton(sstable), CompactionManager.getDefaultGcBefore(cfs));
this.isCommutative = cfs.metadata.getDefaultValidator().isCommutative();
- this.expectedBloomFilterSize = Math.max(cfs.metadata.getIndexInterval(), (int)(SSTableReader.getApproximateKeyCount(toScrub,cfs.metadata)));
+
+ boolean hasIndexFile = (new File(sstable.descriptor.filenameFor(Component.PRIMARY_INDEX))).exists();
+ if (!hasIndexFile)
+ {
+ // if there's any corruption in the -Data.db then rows can't be skipped over. but it's worth a shot.
+ outputHandler.warn("Missing component: " + sstable.descriptor.filenameFor(Component.PRIMARY_INDEX));
+ }
+
+ this.expectedBloomFilterSize = Math.max(cfs.metadata.getIndexInterval(),
+ hasIndexFile ? (int)(SSTableReader.getApproximateKeyCount(toScrub,cfs.metadata)) : 0);
// loop through each row, deserializing to check for damage.
// we'll also loop through the index at the same time, using the position from the index to recover if the
@@ -105,7 +114,11 @@ public class Scrubber implements Closeable
this.dataFile = isOffline
? sstable.openDataReader()
: sstable.openDataReader(CompactionManager.instance.getRateLimiter());
- this.indexFile = RandomAccessReader.open(new File(sstable.descriptor.filenameFor(Component.PRIMARY_INDEX)));
+
+ this.indexFile = hasIndexFile
+ ? RandomAccessReader.open(new File(sstable.descriptor.filenameFor(Component.PRIMARY_INDEX)))
+ : null;
+
this.scrubInfo = new ScrubInfo(dataFile, sstable);
this.currentRowPositionFromIndex = 0;
@@ -117,7 +130,8 @@ public class Scrubber implements Closeable
outputHandler.output(String.format("Scrubbing %s (%s bytes)", sstable, dataFile.length()));
try
{
- nextIndexKey = ByteBufferUtil.readWithShortLength(indexFile);
+ nextIndexKey = indexAvailable() ? ByteBufferUtil.readWithShortLength(indexFile) : null;
+ if (indexAvailable())
{
// throw away variable so we don't have a side effect in the assert
long firstRowPositionFromIndex = RowIndexEntry.serializer.deserialize(indexFile, sstable.descriptor.version).position;
@@ -181,7 +195,7 @@ public class Scrubber implements Closeable
outputHandler.debug(String.format("Index doublecheck: row %s is %s bytes", ByteBufferUtil.bytesToHex(currentIndexKey), dataSizeFromIndex));
}
- assert currentIndexKey != null || indexFile.isEOF();
+ assert currentIndexKey != null || !indexAvailable();
writer.mark();
try
@@ -198,10 +212,10 @@ public class Scrubber implements Closeable
if (dataSize > dataFile.length())
throw new IOError(new IOException("Impossible row size (greater than file length): " + dataSize));
- if (dataStart != dataStartFromIndex)
+ if (indexFile != null && dataStart != dataStartFromIndex)
outputHandler.warn(String.format("Data file row position %d differs from index file row position %d", dataStart, dataStartFromIndex));
- if (dataSize != dataSizeFromIndex)
+ if (indexFile != null && dataSize != dataSizeFromIndex)
outputHandler.warn(String.format("Data file row size %d differs from index file row size %d", dataSize, dataSizeFromIndex));
SSTableIdentityIterator atoms = new SSTableIdentityIterator(sstable, dataFile, key, dataSize, validateColumns);
@@ -317,8 +331,9 @@ public class Scrubber implements Closeable
currentRowPositionFromIndex = nextRowPositionFromIndex;
try
{
- nextIndexKey = indexFile.isEOF() ? null : ByteBufferUtil.readWithShortLength(indexFile);
- nextRowPositionFromIndex = indexFile.isEOF()
+ nextIndexKey = !indexAvailable() ? null : ByteBufferUtil.readWithShortLength(indexFile);
+
+ nextRowPositionFromIndex = !indexAvailable()
? dataFile.length()
: RowIndexEntry.serializer.deserialize(indexFile, sstable.descriptor.version).position;
}
@@ -330,6 +345,11 @@ public class Scrubber implements Closeable
}
}
+ private boolean indexAvailable()
+ {
+ return indexFile != null && !indexFile.isEOF();
+ }
+
private void seekToNextRow()
{
while(nextRowPositionFromIndex < dataFile.length())
http://git-wip-us.apache.org/repos/asf/cassandra/blob/452d6a44/src/java/org/apache/cassandra/io/sstable/SSTableReader.java
----------------------------------------------------------------------
diff --git a/src/java/org/apache/cassandra/io/sstable/SSTableReader.java b/src/java/org/apache/cassandra/io/sstable/SSTableReader.java
index 15808e8..8919a09 100644
--- a/src/java/org/apache/cassandra/io/sstable/SSTableReader.java
+++ b/src/java/org/apache/cassandra/io/sstable/SSTableReader.java
@@ -157,7 +157,7 @@ public class SSTableReader extends SSTable implements Closeable
public static SSTableReader openForBatch(Descriptor descriptor, Set<Component> components, CFMetaData metadata, IPartitioner partitioner) throws IOException
{
- SSTableMetadata sstableMetadata = openMetadata(descriptor, components, partitioner);
+ SSTableMetadata sstableMetadata = openMetadata(descriptor, components, partitioner, true);
SSTableReader sstable = new SSTableReader(descriptor,
components,
metadata,
@@ -191,7 +191,7 @@ public class SSTableReader extends SSTable implements Closeable
boolean validate) throws IOException
{
long start = System.nanoTime();
- SSTableMetadata sstableMetadata = openMetadata(descriptor, components, partitioner);
+ SSTableMetadata sstableMetadata = openMetadata(descriptor, components, partitioner, validate);
SSTableReader sstable = new SSTableReader(descriptor,
components,
@@ -213,12 +213,17 @@ public class SSTableReader extends SSTable implements Closeable
return sstable;
}
- private static SSTableMetadata openMetadata(Descriptor descriptor, Set<Component> components, IPartitioner partitioner) throws IOException
+ private static SSTableMetadata openMetadata(Descriptor descriptor,
+ Set<Component> components,
+ IPartitioner partitioner,
+ boolean primaryIndexRequired) throws IOException
{
assert partitioner != null;
// Minimum components without which we can't do anything
assert components.contains(Component.DATA) : "Data component is missing for sstable" + descriptor;
- assert components.contains(Component.PRIMARY_INDEX) : "Primary index component is missing for sstable " + descriptor;
+
+ assert !primaryIndexRequired || components.contains(Component.PRIMARY_INDEX)
+ : "Primary index component is missing for sstable " + descriptor;
logger.info("Opening {} ({} bytes)", descriptor, new File(descriptor.filenameFor(COMPONENT_DATA)).length());
@@ -382,11 +387,17 @@ public class SSTableReader extends SSTable implements Closeable
readMeterSyncFuture.cancel(false);
// Force finalizing mmapping if necessary
- ifile.cleanup();
+
+ if (null != ifile)
+ ifile.cleanup();
+
dfile.cleanup();
// close the BF so it can be opened later.
- bf.close();
- indexSummary.close();
+ if (null != bf)
+ bf.close();
+
+ if (null != indexSummary)
+ indexSummary.close();
}
public void setTrackedBy(DataTracker tracker)
@@ -406,6 +417,12 @@ public class SSTableReader extends SSTable implements Closeable
load(false, true);
bf = FilterFactory.AlwaysPresent;
}
+ else if (!components.contains(Component.PRIMARY_INDEX))
+ {
+ // avoid any reading of the missing primary index component.
+ // this should only happen during StandaloneScrubber
+ load(false, false);
+ }
else if (!components.contains(Component.FILTER))
{
// bf is enabled, but filter component is missing.
@@ -454,7 +471,9 @@ public class SSTableReader extends SSTable implements Closeable
if (recreateBloomFilter || !summaryLoaded)
buildSummary(recreateBloomFilter, ibuilder, dbuilder, summaryLoaded);
- ifile = ibuilder.complete(descriptor.filenameFor(Component.PRIMARY_INDEX));
+ if (components.contains(Component.PRIMARY_INDEX))
+ ifile = ibuilder.complete(descriptor.filenameFor(Component.PRIMARY_INDEX));
+
dfile = dbuilder.complete(descriptor.filenameFor(Component.DATA));
if (saveSummaryIfCreated && (recreateBloomFilter || !summaryLoaded)) // save summary information to disk
saveSummary(this, ibuilder, dbuilder);
@@ -462,6 +481,9 @@ public class SSTableReader extends SSTable implements Closeable
private void buildSummary(boolean recreateBloomFilter, SegmentedFile.Builder ibuilder, SegmentedFile.Builder dbuilder, boolean summaryLoaded) throws IOException
{
+ if (!components.contains(Component.PRIMARY_INDEX))
+ return;
+
// we read the positions in a BRAF so we don't have to worry about an entry spanning a mmap boundary.
RandomAccessReader primaryIndex = RandomAccessReader.open(new File(descriptor.filenameFor(Component.PRIMARY_INDEX)));
@@ -965,6 +987,9 @@ public class SSTableReader extends SSTable implements Closeable
}
}
+ if (ifile == null)
+ return null;
+
// scan the on-disk index, starting at the nearest sampled position.
// The check against IndexInterval is to be exit the loop in the EQ case when the key looked for is not present
// (bloom filter false positive). But note that for non-EQ cases, we might need to check the first key of the
@@ -1063,6 +1088,9 @@ public class SSTableReader extends SSTable implements Closeable
if (sampledPosition == -1)
sampledPosition = 0;
+ if (ifile == null)
+ return null;
+
Iterator<FileDataInput> segments = ifile.iterator(sampledPosition);
while (segments.hasNext())
{
@@ -1480,7 +1508,8 @@ public class SSTableReader extends SSTable implements Closeable
private void dropPageCache()
{
dropPageCache(dfile.path);
- dropPageCache(ifile.path);
+ if (null != ifile)
+ dropPageCache(ifile.path);
}
private void dropPageCache(String filePath)
http://git-wip-us.apache.org/repos/asf/cassandra/blob/452d6a44/src/java/org/apache/cassandra/tools/StandaloneScrubber.java
----------------------------------------------------------------------
diff --git a/src/java/org/apache/cassandra/tools/StandaloneScrubber.java b/src/java/org/apache/cassandra/tools/StandaloneScrubber.java
index 26768f3..8ace228 100644
--- a/src/java/org/apache/cassandra/tools/StandaloneScrubber.java
+++ b/src/java/org/apache/cassandra/tools/StandaloneScrubber.java
@@ -79,7 +79,7 @@ public class StandaloneScrubber
for (Map.Entry<Descriptor, Set<Component>> entry : lister.list().entrySet())
{
Set<Component> components = entry.getValue();
- if (!components.contains(Component.DATA) || !components.contains(Component.PRIMARY_INDEX))
+ if (!components.contains(Component.DATA))
continue;
try
http://git-wip-us.apache.org/repos/asf/cassandra/blob/452d6a44/test/unit/org/apache/cassandra/db/ScrubTest.java
----------------------------------------------------------------------
diff --git a/test/unit/org/apache/cassandra/db/ScrubTest.java b/test/unit/org/apache/cassandra/db/ScrubTest.java
index 94c7e34..978b3e8 100644
--- a/test/unit/org/apache/cassandra/db/ScrubTest.java
+++ b/test/unit/org/apache/cassandra/db/ScrubTest.java
@@ -270,6 +270,31 @@ public class ScrubTest extends SchemaLoader
}
@Test
+ public void testScrubNoIndex() throws IOException, ExecutionException, InterruptedException, ConfigurationException
+ {
+ CompactionManager.instance.disableAutoCompaction();
+ Keyspace keyspace = Keyspace.open(KEYSPACE);
+ ColumnFamilyStore cfs = keyspace.getColumnFamilyStore(CF);
+ cfs.clearUnsafe();
+
+ List<Row> rows;
+
+ // insert data and verify we get it back w/ range query
+ fillCF(cfs, 10);
+ rows = cfs.getRangeSlice(Util.range("", ""), null, new IdentityQueryFilter(), 1000);
+ assertEquals(10, rows.size());
+
+ for (SSTableReader sstable : cfs.getSSTables())
+ new File(sstable.descriptor.filenameFor(Component.PRIMARY_INDEX)).delete();
+
+ CompactionManager.instance.performScrub(cfs, false, true);
+
+ // check data is still there
+ rows = cfs.getRangeSlice(Util.range("", ""), null, new IdentityQueryFilter(), 1000);
+ assertEquals(10, rows.size());
+ }
+
+ @Test
public void testScrubOutOfOrder() throws Exception
{
CompactionManager.instance.disableAutoCompaction();
[07/12] cassandra git commit: Merge branch 'cassandra-2.0' into
cassandra-2.1
Posted by be...@apache.org.
Merge branch 'cassandra-2.0' into cassandra-2.1
Conflicts:
CHANGES.txt
src/java/org/apache/cassandra/db/compaction/Scrubber.java
src/java/org/apache/cassandra/io/sstable/SSTableReader.java
Project: http://git-wip-us.apache.org/repos/asf/cassandra/repo
Commit: http://git-wip-us.apache.org/repos/asf/cassandra/commit/4c94ef20
Tree: http://git-wip-us.apache.org/repos/asf/cassandra/tree/4c94ef20
Diff: http://git-wip-us.apache.org/repos/asf/cassandra/diff/4c94ef20
Branch: refs/heads/trunk
Commit: 4c94ef20d3562ab8f0a922945d78464d6c475d98
Parents: 4de943f 452d6a4
Author: Benedict Elliott Smith <be...@apache.org>
Authored: Tue Jul 7 16:27:58 2015 +0100
Committer: Benedict Elliott Smith <be...@apache.org>
Committed: Tue Jul 7 16:27:58 2015 +0100
----------------------------------------------------------------------
CHANGES.txt | 5 ++
.../cassandra/db/compaction/Scrubber.java | 37 +++++++++++---
.../cassandra/io/sstable/SSTableReader.java | 52 ++++++++++++++------
.../cassandra/tools/StandaloneScrubber.java | 2 +-
.../unit/org/apache/cassandra/db/ScrubTest.java | 25 ++++++++++
5 files changed, 97 insertions(+), 24 deletions(-)
----------------------------------------------------------------------
http://git-wip-us.apache.org/repos/asf/cassandra/blob/4c94ef20/CHANGES.txt
----------------------------------------------------------------------
diff --cc CHANGES.txt
index 95dc8be,bd1db92..2cbc7c4
--- a/CHANGES.txt
+++ b/CHANGES.txt
@@@ -1,18 -1,8 +1,23 @@@
-2.0.18
-* Scrub (recover) sstables even when -Index.db is missing, (CASSANDRA-9591)
-
-
-2.0.17
++2.1.9
++Merged from 2.0:
++ * Scrub (recover) sstables even when -Index.db is missing, (CASSANDRA-9591)
++
++
+2.1.8
+ * (cqlsh) Fix bad check for CQL compatibility when DESCRIBE'ing
+ COMPACT STORAGE tables with no clustering columns
+ * Warn when an extra-large partition is compacted (CASSANDRA-9643)
+ * Eliminate strong self-reference chains in sstable ref tidiers (CASSANDRA-9656)
+ * Ensure StreamSession uses canonical sstable reader instances (CASSANDRA-9700)
+ * Ensure memtable book keeping is not corrupted in the event we shrink usage (CASSANDRA-9681)
+ * Update internal python driver for cqlsh (CASSANDRA-9064)
+ * Fix IndexOutOfBoundsException when inserting tuple with too many
+ elements using the string literal notation (CASSANDRA-9559)
+ * Allow JMX over SSL directly from nodetool (CASSANDRA-9090)
+ * Fix incorrect result for IN queries where column not found (CASSANDRA-9540)
+ * Enable describe on indices (CASSANDRA-7814)
+ * ColumnFamilyStore.selectAndReference may block during compaction (CASSANDRA-9637)
+Merged from 2.0:
* Avoid NPE in AuthSuccess#decode (CASSANDRA-9727)
* Add listen_address to system.local (CASSANDRA-9603)
* Bug fixes to resultset metadata construction (CASSANDRA-9636)
http://git-wip-us.apache.org/repos/asf/cassandra/blob/4c94ef20/src/java/org/apache/cassandra/db/compaction/Scrubber.java
----------------------------------------------------------------------
diff --cc src/java/org/apache/cassandra/db/compaction/Scrubber.java
index ce98a13,dc60efa..b1c12e0
--- a/src/java/org/apache/cassandra/db/compaction/Scrubber.java
+++ b/src/java/org/apache/cassandra/db/compaction/Scrubber.java
@@@ -100,8 -95,17 +100,18 @@@ public class Scrubber implements Closea
this.controller = isOffline
? new ScrubController(cfs)
: new CompactionController(cfs, Collections.singleton(sstable), CompactionManager.getDefaultGcBefore(cfs));
- this.isCommutative = cfs.metadata.getDefaultValidator().isCommutative();
+ this.isCommutative = cfs.metadata.isCounter();
- this.expectedBloomFilterSize = Math.max(cfs.metadata.getMinIndexInterval(), (int)(SSTableReader.getApproximateKeyCount(toScrub)));
+
+ boolean hasIndexFile = (new File(sstable.descriptor.filenameFor(Component.PRIMARY_INDEX))).exists();
+ if (!hasIndexFile)
+ {
+ // if there's any corruption in the -Data.db then rows can't be skipped over. but it's worth a shot.
+ outputHandler.warn("Missing component: " + sstable.descriptor.filenameFor(Component.PRIMARY_INDEX));
+ }
+
- this.expectedBloomFilterSize = Math.max(cfs.metadata.getIndexInterval(),
- hasIndexFile ? (int)(SSTableReader.getApproximateKeyCount(toScrub,cfs.metadata)) : 0);
++ this.expectedBloomFilterSize = Math.max(
++ cfs.metadata.getMinIndexInterval(),
++ hasIndexFile ? (int)(SSTableReader.getApproximateKeyCount(toScrub)) : 0);
// loop through each row, deserializing to check for damage.
// we'll also loop through the index at the same time, using the position from the index to recover if the
@@@ -120,14 -128,13 +134,15 @@@
public void scrub()
{
outputHandler.output(String.format("Scrubbing %s (%s bytes)", sstable, dataFile.length()));
+ Set<SSTableReader> oldSSTable = Sets.newHashSet(sstable);
+ SSTableRewriter writer = new SSTableRewriter(cfs, oldSSTable, sstable.maxDataAge, isOffline);
try
{
- nextIndexKey = ByteBufferUtil.readWithShortLength(indexFile);
+ nextIndexKey = indexAvailable() ? ByteBufferUtil.readWithShortLength(indexFile) : null;
+ if (indexAvailable())
{
// throw away variable so we don't have a side effect in the assert
- long firstRowPositionFromIndex = RowIndexEntry.serializer.deserialize(indexFile, sstable.descriptor.version).position;
+ long firstRowPositionFromIndex = sstable.metadata.comparator.rowIndexEntrySerializer().deserialize(indexFile, sstable.descriptor.version).position;
assert firstRowPositionFromIndex == 0 : firstRowPositionFromIndex;
}
@@@ -167,13 -179,25 +182,13 @@@
dataSizeFromIndex = nextRowPositionFromIndex - dataStartFromIndex;
}
- if (!sstable.descriptor.version.hasRowSizeAndColumnCount)
- {
- dataSize = dataSizeFromIndex;
- // avoid an NPE if key is null
- String keyName = key == null ? "(unreadable key)" : ByteBufferUtil.bytesToHex(key.key);
- outputHandler.debug(String.format("row %s is %s bytes", keyName, dataSize));
- }
- else
- {
- if (currentIndexKey != null)
- outputHandler.debug(String.format("Index doublecheck: row %s is %s bytes", ByteBufferUtil.bytesToHex(currentIndexKey), dataSizeFromIndex));
- }
+ dataSize = dataSizeFromIndex;
+ // avoid an NPE if key is null
+ String keyName = key == null ? "(unreadable key)" : ByteBufferUtil.bytesToHex(key.getKey());
+ outputHandler.debug(String.format("row %s is %s bytes", keyName, dataSize));
- assert currentIndexKey != null || indexFile.isEOF();
+ assert currentIndexKey != null || !indexAvailable();
- writer.mark();
try
{
if (key == null)
@@@ -188,11 -212,11 +203,11 @@@
if (dataSize > dataFile.length())
throw new IOError(new IOException("Impossible row size (greater than file length): " + dataSize));
- if (dataStart != dataStartFromIndex)
+ if (indexFile != null && dataStart != dataStartFromIndex)
outputHandler.warn(String.format("Data file row position %d differs from index file row position %d", dataStart, dataStartFromIndex));
- if (dataSize != dataSizeFromIndex)
+ if (indexFile != null && dataSize != dataSizeFromIndex)
- outputHandler.warn(String.format("Data file row size %d differs from index file row size %d", dataSize, dataSizeFromIndex));
+ outputHandler.warn(String.format("Data file row size %d different from index file row size %d", dataSize, dataSizeFromIndex));
SSTableIdentityIterator atoms = new SSTableIdentityIterator(sstable, dataFile, key, dataSize, validateColumns);
if (prevKey != null && prevKey.compareTo(key) > 0)
@@@ -312,10 -331,11 +327,11 @@@
currentRowPositionFromIndex = nextRowPositionFromIndex;
try
{
- nextIndexKey = indexFile.isEOF() ? null : ByteBufferUtil.readWithShortLength(indexFile);
- nextRowPositionFromIndex = indexFile.isEOF()
+ nextIndexKey = !indexAvailable() ? null : ByteBufferUtil.readWithShortLength(indexFile);
+
+ nextRowPositionFromIndex = !indexAvailable()
? dataFile.length()
- : RowIndexEntry.serializer.deserialize(indexFile, sstable.descriptor.version).position;
+ : sstable.metadata.comparator.rowIndexEntrySerializer().deserialize(indexFile, sstable.descriptor.version).position;
}
catch (Throwable th)
{
http://git-wip-us.apache.org/repos/asf/cassandra/blob/4c94ef20/src/java/org/apache/cassandra/io/sstable/SSTableReader.java
----------------------------------------------------------------------
diff --cc src/java/org/apache/cassandra/io/sstable/SSTableReader.java
index 7551d46,8919a09..6879834
--- a/src/java/org/apache/cassandra/io/sstable/SSTableReader.java
+++ b/src/java/org/apache/cassandra/io/sstable/SSTableReader.java
@@@ -448,27 -190,9 +448,27 @@@ public class SSTableReader extends SSTa
IPartitioner partitioner,
boolean validate) throws IOException
{
- long start = System.nanoTime();
- SSTableMetadata sstableMetadata = openMetadata(descriptor, components, partitioner, validate);
+ // Minimum components without which we can't do anything
+ assert components.contains(Component.DATA) : "Data component is missing for sstable" + descriptor;
- assert components.contains(Component.PRIMARY_INDEX) : "Primary index component is missing for sstable " + descriptor;
++ assert !validate || components.contains(Component.PRIMARY_INDEX) : "Primary index component is missing for sstable " + descriptor;
+
+ Map<MetadataType, MetadataComponent> sstableMetadata = descriptor.getMetadataSerializer().deserialize(descriptor,
- EnumSet.of(MetadataType.VALIDATION, MetadataType.STATS));
++ EnumSet.of(MetadataType.VALIDATION, MetadataType.STATS));
+ ValidationMetadata validationMetadata = (ValidationMetadata) sstableMetadata.get(MetadataType.VALIDATION);
+ StatsMetadata statsMetadata = (StatsMetadata) sstableMetadata.get(MetadataType.STATS);
+
+ // Check if sstable is created using same partitioner.
+ // Partitioner can be null, which indicates older version of sstable or no stats available.
+ // In that case, we skip the check.
+ String partitionerName = partitioner.getClass().getCanonicalName();
+ if (validationMetadata != null && !partitionerName.equals(validationMetadata.partitioner))
+ {
+ logger.error(String.format("Cannot open %s; partitioner %s does not match system partitioner %s. Note that the default partitioner starting with Cassandra 1.2 is Murmur3Partitioner, so you will need to edit that to match your old partitioner if upgrading.",
+ descriptor, validationMetadata.partitioner, partitionerName));
+ System.exit(1);
+ }
+ logger.info("Opening {} ({} bytes)", descriptor, new File(descriptor.filenameFor(Component.DATA)).length());
SSTableReader sstable = new SSTableReader(descriptor,
components,
metadata,
@@@ -603,48 -374,35 +603,43 @@@
this.dfile = dfile;
this.indexSummary = indexSummary;
this.bf = bloomFilter;
+ this.setup(false);
}
- /**
- * Clean up all opened resources.
- *
- * @throws IOException
- */
- public void close() throws IOException
+ public static long getTotalBytes(Iterable<SSTableReader> sstables)
{
- if (readMeterSyncFuture != null)
- readMeterSyncFuture.cancel(false);
+ long sum = 0;
+ for (SSTableReader sstable : sstables)
+ sum += sstable.onDiskLength();
+ return sum;
+ }
- // Force finalizing mmapping if necessary
+ public static long getTotalUncompressedBytes(Iterable<SSTableReader> sstables)
+ {
+ long sum = 0;
+ for (SSTableReader sstable : sstables)
+ sum += sstable.uncompressedLength();
- if (null != ifile)
- ifile.cleanup();
+ return sum;
+ }
+
+ public boolean equals(Object that)
+ {
+ return that instanceof SSTableReader && ((SSTableReader) that).descriptor.equals(this.descriptor);
+ }
- dfile.cleanup();
- // close the BF so it can be opened later.
- if (null != bf)
- bf.close();
+ public int hashCode()
+ {
+ return this.descriptor.hashCode();
+ }
- if (null != indexSummary)
- indexSummary.close();
+ public String getFilename()
+ {
+ return dfile.path;
}
- public String getIndexFilename()
- {
- return ifile.path;
- }
-
- public void setTrackedBy(DataTracker tracker)
+ public void setupKeyCache()
{
- deletingTask.setTracker(tracker);
// under normal operation we can do this at any time, but SSTR is also used outside C* proper,
// e.g. by BulkLoader, which does not initialize the cache. As a kludge, we set up the cache
// here when we know we're being wired into the rest of the server infrastructure.
@@@ -659,7 -417,13 +654,13 @@@
load(false, true);
bf = FilterFactory.AlwaysPresent;
}
+ else if (!components.contains(Component.PRIMARY_INDEX))
+ {
+ // avoid any reading of the missing primary index component.
- // this should only happen during StandaloneScrubber
++ // this should only happen for standalone tools
+ load(false, false);
+ }
- else if (!components.contains(Component.FILTER))
+ else if (!components.contains(Component.FILTER) || validation == null)
{
// bf is enabled, but filter component is missing.
load(true, true);
@@@ -708,418 -467,26 +709,427 @@@
? SegmentedFile.getCompressedBuilder()
: SegmentedFile.getBuilder(DatabaseDescriptor.getDiskAccessMode());
- boolean summaryLoaded = loadSummary(this, ibuilder, dbuilder, metadata);
+ boolean summaryLoaded = loadSummary(ibuilder, dbuilder);
+ boolean builtSummary = false;
if (recreateBloomFilter || !summaryLoaded)
- buildSummary(recreateBloomFilter, ibuilder, dbuilder, summaryLoaded);
+ {
+ buildSummary(recreateBloomFilter, ibuilder, dbuilder, summaryLoaded, Downsampling.BASE_SAMPLING_LEVEL);
+ builtSummary = true;
+ }
- ifile = ibuilder.complete(descriptor.filenameFor(Component.PRIMARY_INDEX));
+ if (components.contains(Component.PRIMARY_INDEX))
+ ifile = ibuilder.complete(descriptor.filenameFor(Component.PRIMARY_INDEX));
+
dfile = dbuilder.complete(descriptor.filenameFor(Component.DATA));
- if (saveSummaryIfCreated && (recreateBloomFilter || !summaryLoaded)) // save summary information to disk
- saveSummary(this, ibuilder, dbuilder);
+
+ // Check for an index summary that was downsampled even though the serialization format doesn't support
+ // that. If it was downsampled, rebuild it. See CASSANDRA-8993 for details.
- if (!descriptor.version.hasSamplingLevel && !builtSummary && !validateSummarySamplingLevel())
++ if (!descriptor.version.hasSamplingLevel && !builtSummary && !validateSummarySamplingLevel() && ifile != null)
+ {
+ indexSummary.close();
+ ifile.close();
+ dfile.close();
+
+ logger.info("Detected erroneously downsampled index summary; will rebuild summary at full sampling");
+ FileUtils.deleteWithConfirm(new File(descriptor.filenameFor(Component.SUMMARY)));
+ ibuilder = SegmentedFile.getBuilder(DatabaseDescriptor.getIndexAccessMode());
+ dbuilder = compression
+ ? SegmentedFile.getCompressedBuilder()
+ : SegmentedFile.getBuilder(DatabaseDescriptor.getDiskAccessMode());
+ buildSummary(false, ibuilder, dbuilder, false, Downsampling.BASE_SAMPLING_LEVEL);
+ ifile = ibuilder.complete(descriptor.filenameFor(Component.PRIMARY_INDEX));
+ dfile = dbuilder.complete(descriptor.filenameFor(Component.DATA));
+ saveSummary(ibuilder, dbuilder);
+ }
+ else if (saveSummaryIfCreated && builtSummary)
+ {
+ saveSummary(ibuilder, dbuilder);
+ }
}
- private void buildSummary(boolean recreateBloomFilter, SegmentedFile.Builder ibuilder, SegmentedFile.Builder dbuilder, boolean summaryLoaded) throws IOException
- {
+ /**
+ * Build index summary(and optionally bloom filter) by reading through Index.db file.
+ *
+ * @param recreateBloomFilter true if recreate bloom filter
+ * @param ibuilder
+ * @param dbuilder
+ * @param summaryLoaded true if index summary is already loaded and not need to build again
+ * @throws IOException
+ */
+ private void buildSummary(boolean recreateBloomFilter, SegmentedFile.Builder ibuilder, SegmentedFile.Builder dbuilder, boolean summaryLoaded, int samplingLevel) throws IOException
+ {
+ if (!components.contains(Component.PRIMARY_INDEX))
+ return;
+
// we read the positions in a BRAF so we don't have to worry about an entry spanning a mmap boundary.
RandomAccessReader primaryIndex = RandomAccessReader.open(new File(descriptor.filenameFor(Component.PRIMARY_INDEX)));
-
+
+ try
+ {
+ long indexSize = primaryIndex.length();
+ long histogramCount = sstableMetadata.estimatedRowSize.count();
+ long estimatedKeys = histogramCount > 0 && !sstableMetadata.estimatedRowSize.isOverflowed()
+ ? histogramCount
+ : estimateRowsFromIndex(primaryIndex); // statistics is supposed to be optional
+
+ try(IndexSummaryBuilder summaryBuilder = summaryLoaded ? null : new IndexSummaryBuilder(estimatedKeys, metadata.getMinIndexInterval(), samplingLevel))
+ {
+
+ if (recreateBloomFilter)
+ bf = FilterFactory.getFilter(estimatedKeys, metadata.getBloomFilterFpChance(), true);
+
+ long indexPosition;
+ while ((indexPosition = primaryIndex.getFilePointer()) != indexSize)
+ {
+ ByteBuffer key = ByteBufferUtil.readWithShortLength(primaryIndex);
+ RowIndexEntry indexEntry = metadata.comparator.rowIndexEntrySerializer().deserialize(primaryIndex, descriptor.version);
+ DecoratedKey decoratedKey = partitioner.decorateKey(key);
+ if (first == null)
+ first = decoratedKey;
+ last = decoratedKey;
+
+ if (recreateBloomFilter)
+ bf.add(decoratedKey.getKey());
+
+ // if summary was already read from disk we don't want to re-populate it using primary index
+ if (!summaryLoaded)
+ {
+ summaryBuilder.maybeAddEntry(decoratedKey, indexPosition);
+ ibuilder.addPotentialBoundary(indexPosition);
+ dbuilder.addPotentialBoundary(indexEntry.position);
+ }
+ }
+
+ if (!summaryLoaded)
+ indexSummary = summaryBuilder.build(partitioner);
+ }
+ }
+ finally
+ {
+ FileUtils.closeQuietly(primaryIndex);
+ }
+
+ first = getMinimalKey(first);
+ last = getMinimalKey(last);
+ }
+
+ /**
+ * Load index summary from Summary.db file if it exists.
+ *
+ * if loaded index summary has different index interval from current value stored in schema,
+ * then Summary.db file will be deleted and this returns false to rebuild summary.
+ *
+ * @param ibuilder
+ * @param dbuilder
+ * @return true if index summary is loaded successfully from Summary.db file.
+ */
+ public boolean loadSummary(SegmentedFile.Builder ibuilder, SegmentedFile.Builder dbuilder)
+ {
+ File summariesFile = new File(descriptor.filenameFor(Component.SUMMARY));
+ if (!summariesFile.exists())
+ return false;
+
+ DataInputStream iStream = null;
+ try
+ {
+ iStream = new DataInputStream(new FileInputStream(summariesFile));
+ indexSummary = IndexSummary.serializer.deserialize(
+ iStream, partitioner, descriptor.version.hasSamplingLevel,
+ metadata.getMinIndexInterval(), metadata.getMaxIndexInterval());
+ first = partitioner.decorateKey(ByteBufferUtil.readWithLength(iStream));
+ last = partitioner.decorateKey(ByteBufferUtil.readWithLength(iStream));
+ ibuilder.deserializeBounds(iStream);
+ dbuilder.deserializeBounds(iStream);
+ }
+ catch (IOException e)
+ {
+ if (indexSummary != null)
+ indexSummary.close();
+ logger.debug("Cannot deserialize SSTable Summary File {}: {}", summariesFile.getPath(), e.getMessage());
+ // corrupted; delete it and fall back to creating a new summary
+ FileUtils.closeQuietly(iStream);
+ // delete it and fall back to creating a new summary
+ FileUtils.deleteWithConfirm(summariesFile);
+ return false;
+ }
+ finally
+ {
+ FileUtils.closeQuietly(iStream);
+ }
+
+ return true;
+ }
+
+ /**
+ * Validates that an index summary has full sampling, as expected when the serialization format does not support
+ * persisting the sampling level.
+ * @return true if the summary has full sampling, false otherwise
+ */
+ private boolean validateSummarySamplingLevel()
+ {
+ // We need to check index summary entries against the index to verify that none of them were dropped due to
+ // downsampling. Downsampling can drop any of the first BASE_SAMPLING_LEVEL entries (repeating that drop pattern
+ // for the remainder of the summary). Unfortunately, the first entry to be dropped is the entry at
+ // index (BASE_SAMPLING_LEVEL - 1), so we need to check a full set of BASE_SAMPLING_LEVEL entries.
++ if (ifile == null)
++ return false;
++
+ Iterator<FileDataInput> segments = ifile.iterator(0);
+ int i = 0;
+ int summaryEntriesChecked = 0;
+ int expectedIndexInterval = getMinIndexInterval();
+ while (segments.hasNext())
+ {
+ FileDataInput in = segments.next();
+ try
+ {
+ while (!in.isEOF())
+ {
+ ByteBuffer indexKey = ByteBufferUtil.readWithShortLength(in);
+ if (i % expectedIndexInterval == 0)
+ {
+ ByteBuffer summaryKey = ByteBuffer.wrap(indexSummary.getKey(i / expectedIndexInterval));
+ if (!summaryKey.equals(indexKey))
+ return false;
+ summaryEntriesChecked++;
+
+ if (summaryEntriesChecked == Downsampling.BASE_SAMPLING_LEVEL)
+ return true;
+ }
+ RowIndexEntry.Serializer.skip(in);
+ i++;
+ }
+ }
+ catch (IOException e)
+ {
+ markSuspect();
+ throw new CorruptSSTableException(e, in.getPath());
+ }
+ finally
+ {
+ FileUtils.closeQuietly(in);
+ }
+ }
+
+ return true;
+ }
+
+ /**
+ * Save index summary to Summary.db file.
+ *
+ * @param ibuilder
+ * @param dbuilder
+ */
+ public void saveSummary(SegmentedFile.Builder ibuilder, SegmentedFile.Builder dbuilder)
+ {
+ saveSummary(ibuilder, dbuilder, indexSummary);
+ }
+
+ private void saveSummary(SegmentedFile.Builder ibuilder, SegmentedFile.Builder dbuilder, IndexSummary summary)
+ {
+ File summariesFile = new File(descriptor.filenameFor(Component.SUMMARY));
+ if (summariesFile.exists())
+ FileUtils.deleteWithConfirm(summariesFile);
+
+ DataOutputStreamAndChannel oStream = null;
+ try
+ {
+ oStream = new DataOutputStreamAndChannel(new FileOutputStream(summariesFile));
+ IndexSummary.serializer.serialize(summary, oStream, descriptor.version.hasSamplingLevel);
+ ByteBufferUtil.writeWithLength(first.getKey(), oStream);
+ ByteBufferUtil.writeWithLength(last.getKey(), oStream);
+ ibuilder.serializeBounds(oStream);
+ dbuilder.serializeBounds(oStream);
+ }
+ catch (IOException e)
+ {
+ logger.debug("Cannot save SSTable Summary: ", e);
+
+ // corrupted hence delete it and let it load it now.
+ if (summariesFile.exists())
+ FileUtils.deleteWithConfirm(summariesFile);
+ }
+ finally
+ {
+ FileUtils.closeQuietly(oStream);
+ }
+ }
+
+ public void setReplacedBy(SSTableReader replacement)
+ {
+ synchronized (tidy.global)
+ {
+ assert replacement != null;
+ assert !tidy.isReplaced;
+ tidy.isReplaced = true;
+ }
+ }
+
+ public boolean isReplaced()
+ {
+ return tidy.isReplaced;
+ }
+
+ /**
+ * Clone this reader with the provided start and open reason, and set the clone as replacement.
+ *
+ * @param newFirst the first key for the replacement (which can be different from the original due to the pre-emptive
+ * opening of compaction results).
+ * @param reason the {@code OpenReason} for the replacement.
+ *
+ * @return the cloned reader. That reader is set as a replacement by the method.
+ */
+ private SSTableReader cloneAndReplace(DecoratedKey newFirst, OpenReason reason)
+ {
+ return cloneAndReplace(newFirst, reason, indexSummary.sharedCopy());
+ }
+
+ /**
+ * Clone this reader with the new values and set the clone as replacement.
+ *
+ * @param newFirst the first key for the replacement (which can be different from the original due to the pre-emptive
+ * opening of compaction results).
+ * @param reason the {@code OpenReason} for the replacement.
+ * @param newSummary the index summary for the replacement.
+ *
+ * @return the cloned reader. That reader is set as a replacement by the method.
+ */
+ private SSTableReader cloneAndReplace(DecoratedKey newFirst, OpenReason reason, IndexSummary newSummary)
+ {
+ SSTableReader replacement = internalOpen(descriptor,
+ components,
+ metadata,
+ partitioner,
- ifile.sharedCopy(),
++ ifile != null ? ifile.sharedCopy() : null,
+ dfile.sharedCopy(),
+ newSummary,
+ bf.sharedCopy(),
+ maxDataAge,
+ sstableMetadata,
+ reason);
+ replacement.first = newFirst;
+ replacement.last = last;
+ replacement.isSuspect.set(isSuspect.get());
+ setReplacedBy(replacement);
+ return replacement;
+ }
+
+ // runOnClose must NOT be an anonymous or non-static inner class, nor must it retain a reference chain to this reader
+ public SSTableReader cloneWithNewStart(DecoratedKey newStart, final Runnable runOnClose)
+ {
+ synchronized (tidy.global)
+ {
+ assert openReason != OpenReason.EARLY;
+ // TODO: merge with caller's firstKeyBeyond() work,to save time
+ if (newStart.compareTo(first) > 0)
+ {
+ final long dataStart = getPosition(newStart, Operator.EQ).position;
+ final long indexStart = getIndexScanPosition(newStart);
+ this.tidy.runOnClose = new DropPageCache(dfile, dataStart, ifile, indexStart, runOnClose);
+ }
+
+ return cloneAndReplace(newStart, OpenReason.MOVED_START);
+ }
+ }
+
+ // runOnClose must NOT be an anonymous or non-static inner class, nor must it retain a reference chain to this reader
+ public SSTableReader cloneAsShadowed(final Runnable runOnClose)
+ {
+ synchronized (tidy.global)
+ {
+ assert openReason != OpenReason.EARLY;
+ this.tidy.runOnClose = new DropPageCache(dfile, 0, ifile, 0, runOnClose);
+ return cloneAndReplace(first, OpenReason.SHADOWED);
+ }
+ }
+
+ private static class DropPageCache implements Runnable
+ {
+ final SegmentedFile dfile;
+ final long dfilePosition;
+ final SegmentedFile ifile;
+ final long ifilePosition;
+ final Runnable andThen;
+
+ private DropPageCache(SegmentedFile dfile, long dfilePosition, SegmentedFile ifile, long ifilePosition, Runnable andThen)
+ {
+ this.dfile = dfile;
+ this.dfilePosition = dfilePosition;
+ this.ifile = ifile;
+ this.ifilePosition = ifilePosition;
+ this.andThen = andThen;
+ }
+
+ public void run()
+ {
+ dfile.dropPageCache(dfilePosition);
- ifile.dropPageCache(ifilePosition);
++ if (ifile != null)
++ ifile.dropPageCache(ifilePosition);
+ andThen.run();
+ }
+ }
+
+ /**
+ * Returns a new SSTableReader with the same properties as this SSTableReader except that a new IndexSummary will
+ * be built at the target samplingLevel. This (original) SSTableReader instance will be marked as replaced, have
+ * its DeletingTask removed, and have its periodic read-meter sync task cancelled.
+ * @param samplingLevel the desired sampling level for the index summary on the new SSTableReader
+ * @return a new SSTableReader
+ * @throws IOException
+ */
+ public SSTableReader cloneWithNewSummarySamplingLevel(ColumnFamilyStore parent, int samplingLevel) throws IOException
+ {
+ assert descriptor.version.hasSamplingLevel;
+
+ synchronized (tidy.global)
+ {
+ assert openReason != OpenReason.EARLY;
+
+ int minIndexInterval = metadata.getMinIndexInterval();
+ int maxIndexInterval = metadata.getMaxIndexInterval();
+ double effectiveInterval = indexSummary.getEffectiveIndexInterval();
+
+ IndexSummary newSummary;
+ long oldSize = bytesOnDisk();
+
+ // We have to rebuild the summary from the on-disk primary index in three cases:
+ // 1. The sampling level went up, so we need to read more entries off disk
+ // 2. The min_index_interval changed (in either direction); this changes what entries would be in the summary
+ // at full sampling (and consequently at any other sampling level)
+ // 3. The max_index_interval was lowered, forcing us to raise the sampling level
+ if (samplingLevel > indexSummary.getSamplingLevel() || indexSummary.getMinIndexInterval() != minIndexInterval || effectiveInterval > maxIndexInterval)
+ {
+ newSummary = buildSummaryAtLevel(samplingLevel);
+ }
+ else if (samplingLevel < indexSummary.getSamplingLevel())
+ {
+ // we can use the existing index summary to make a smaller one
+ newSummary = IndexSummaryBuilder.downsample(indexSummary, samplingLevel, minIndexInterval, partitioner);
+
+ SegmentedFile.Builder ibuilder = SegmentedFile.getBuilder(DatabaseDescriptor.getIndexAccessMode());
+ SegmentedFile.Builder dbuilder = compression
+ ? SegmentedFile.getCompressedBuilder()
+ : SegmentedFile.getBuilder(DatabaseDescriptor.getDiskAccessMode());
+ saveSummary(ibuilder, dbuilder, newSummary);
+ }
+ else
+ {
+ throw new AssertionError("Attempted to clone SSTableReader with the same index summary sampling level and " +
+ "no adjustments to min/max_index_interval");
+ }
+
+ long newSize = bytesOnDisk();
+ StorageMetrics.load.inc(newSize - oldSize);
+ parent.metric.liveDiskSpaceUsed.inc(newSize - oldSize);
+
+ return cloneAndReplace(first, OpenReason.METADATA_CHANGE, newSummary);
+ }
+ }
+
+ private IndexSummary buildSummaryAtLevel(int newSamplingLevel) throws IOException
+ {
+ // we read the positions in a BRAF so we don't have to worry about an entry spanning a mmap boundary.
+ RandomAccessReader primaryIndex = RandomAccessReader.open(new File(descriptor.filenameFor(Component.PRIMARY_INDEX)));
try
{
long indexSize = primaryIndex.length();
@@@ -1566,16 -974,22 +1576,19 @@@
{
if (op == Operator.EQ && updateCacheAndStats)
bloomFilterTracker.addFalsePositive();
- // we matched the -1th position: if the operator might match forward, we'll start at the first
- // position. We however need to return the correct index entry for that first position.
- if (op.apply(1) >= 0)
- {
- sampledPosition = 0;
- }
- else
- {
- Tracing.trace("Partition summary allows skipping sstable {}", descriptor.generation);
- return null;
- }
+ Tracing.trace("Check against min and max keys allows skipping sstable {}", descriptor.generation);
+ return null;
}
+ int binarySearchResult = indexSummary.binarySearch(key);
+ long sampledPosition = getIndexScanPositionFromBinarySearchResult(binarySearchResult, indexSummary);
+ int sampledIndex = getIndexSummaryIndexFromBinarySearchResult(binarySearchResult);
+
+ int effectiveInterval = indexSummary.getEffectiveIndexIntervalAfterIndex(sampledIndex);
+
+ if (ifile == null)
+ return null;
+
// scan the on-disk index, starting at the nearest sampled position.
// The check against IndexInterval is to be exit the loop in the EQ case when the key looked for is not present
// (bloom filter false positive). But note that for non-EQ cases, we might need to check the first key of the
@@@ -1670,11 -1084,13 +1683,14 @@@
*/
public DecoratedKey firstKeyBeyond(RowPosition token)
{
+ if (token.compareTo(first) < 0)
+ return first;
+
long sampledPosition = getIndexScanPosition(token);
- if (sampledPosition == -1)
- sampledPosition = 0;
+ if (ifile == null)
+ return null;
+
Iterator<FileDataInput> segments = ifile.iterator(sampledPosition);
while (segments.hasNext())
{
@@@ -1948,8 -1391,7 +1964,10 @@@
{
try
{
- return SSTableMetadata.serializer.deserialize(descriptor).right;
+ CompactionMetadata compactionMetadata = (CompactionMetadata) descriptor.getMetadataSerializer().deserialize(descriptor, MetadataType.COMPACTION);
- return compactionMetadata.ancestors;
++ if (compactionMetadata != null)
++ return compactionMetadata.ancestors;
++ return Collections.emptySet();
}
catch (IOException e)
{
@@@ -1995,7 -1441,7 +2013,9 @@@
public RandomAccessReader openIndexReader()
{
- return ifile.createReader();
- return RandomAccessReader.open(new File(getIndexFilename()));
++ if (ifile != null)
++ return ifile.createReader();
++ return null;
}
/**
@@@ -2024,152 -1470,73 +2044,154 @@@
}
/**
- * @param sstables
- * @return true if all desired references were acquired. Otherwise, it will unreference any partial acquisition, and return false.
+ * Increment the total row read count and read rate for this SSTable. This should not be incremented for range
+ * slice queries, row cache hits, or non-query reads, like compaction.
*/
- public static boolean acquireReferences(Iterable<SSTableReader> sstables)
+ public void incrementReadCount()
{
- SSTableReader failed = null;
- for (SSTableReader sstable : sstables)
+ if (readMeter != null)
+ readMeter.mark();
+ }
+
+ public static class SizeComparator implements Comparator<SSTableReader>
+ {
+ public int compare(SSTableReader o1, SSTableReader o2)
{
- if (!sstable.acquireReference())
- {
- failed = sstable;
- break;
- }
+ return Longs.compare(o1.onDiskLength(), o2.onDiskLength());
}
+ }
- if (failed == null)
- return true;
+ public Ref<SSTableReader> tryRef()
+ {
+ return selfRef.tryRef();
+ }
- for (SSTableReader sstable : sstables)
- {
- if (sstable == failed)
- break;
- sstable.releaseReference();
- }
- return false;
+ public Ref<SSTableReader> selfRef()
+ {
+ return selfRef;
}
- public static void releaseReferences(Iterable<SSTableReader> sstables)
+ public Ref<SSTableReader> ref()
{
- for (SSTableReader sstable : sstables)
- {
- sstable.releaseReference();
- }
+ return selfRef.ref();
}
- private void dropPageCache()
+ void setup(boolean isOffline)
{
- dropPageCache(dfile.path);
- if (null != ifile)
- dropPageCache(ifile.path);
+ tidy.setup(this, isOffline);
+ this.readMeter = tidy.global.readMeter;
}
- private void dropPageCache(String filePath)
+ @VisibleForTesting
+ public void overrideReadMeter(RestorableMeter readMeter)
{
- RandomAccessFile file = null;
+ this.readMeter = tidy.global.readMeter = readMeter;
+ }
- try
+ /**
+ * One instance per SSTableReader we create. This references the type-shared tidy, which in turn references
+ * the globally shared tidy, i.e.
+ *
+ * InstanceTidier => DescriptorTypeTitdy => GlobalTidy
+ *
+ * We can create many InstanceTidiers (one for every time we reopen an sstable with MOVED_START for example), but there can only be
+ * two DescriptorTypeTidy (FINAL and TEMPLINK) and only one GlobalTidy for one single logical sstable.
+ *
+ * When the InstanceTidier cleansup, it releases its reference to its DescriptorTypeTidy; when all InstanceTidiers
+ * for that type have run, the DescriptorTypeTidy cleansup. DescriptorTypeTidy behaves in the same way towards GlobalTidy.
+ *
+ * For ease, we stash a direct reference to both our type-shared and global tidier
+ */
+ private static final class InstanceTidier implements Tidy
+ {
+ private final Descriptor descriptor;
+ private final CFMetaData metadata;
+ private IFilter bf;
+ private IndexSummary summary;
+
+ private SegmentedFile dfile;
+ private SegmentedFile ifile;
+ private Runnable runOnClose;
+ private boolean isReplaced = false;
+
+ // a reference to our shared per-Descriptor.Type tidy instance, that
+ // we will release when we are ourselves released
+ private Ref<DescriptorTypeTidy> typeRef;
+
+ // a convenience stashing of the shared per-descriptor-type tidy instance itself
+ // and the per-logical-sstable globally shared state that it is linked to
+ private DescriptorTypeTidy type;
+ private GlobalTidy global;
+
+ private boolean setup;
+
+ void setup(SSTableReader reader, boolean isOffline)
+ {
+ this.setup = true;
+ this.bf = reader.bf;
+ this.summary = reader.indexSummary;
+ this.dfile = reader.dfile;
+ this.ifile = reader.ifile;
+ // get a new reference to the shared descriptor-type tidy
+ this.typeRef = DescriptorTypeTidy.get(reader);
+ this.type = typeRef.get();
+ this.global = type.globalRef.get();
+ if (!isOffline)
+ global.ensureReadMeter();
+ }
+
+ InstanceTidier(Descriptor descriptor, CFMetaData metadata)
{
- file = new RandomAccessFile(filePath, "r");
+ this.descriptor = descriptor;
+ this.metadata = metadata;
+ }
- int fd = CLibrary.getfd(file.getFD());
+ public void tidy()
+ {
+ // don't try to cleanup if the sstablereader was never fully constructed
+ if (!setup)
+ return;
- if (fd > 0)
+ final ColumnFamilyStore cfs = Schema.instance.getColumnFamilyStoreInstance(metadata.cfId);
+ final OpOrder.Barrier barrier;
+ if (cfs != null)
{
- if (logger.isDebugEnabled())
- logger.debug(String.format("Dropping page cache of file %s.", filePath));
-
- CLibrary.trySkipCache(fd, 0, 0);
+ barrier = cfs.readOrdering.newBarrier();
+ barrier.issue();
}
+ else
+ barrier = null;
+
+ ScheduledExecutors.nonPeriodicTasks.execute(new Runnable()
+ {
+ public void run()
+ {
+ if (barrier != null)
+ barrier.await();
- bf.close();
++ if (bf != null)
++ bf.close();
+ if (summary != null)
+ summary.close();
+ if (runOnClose != null)
+ runOnClose.run();
+ dfile.close();
- ifile.close();
++ if (ifile != null)
++ ifile.close();
+ typeRef.release();
+ }
+ });
}
- catch (IOException e)
+
+ public String name()
{
- // we don't care if cache cleanup fails
+ return descriptor.toString();
}
- finally
+
+ void releaseSummary()
{
- FileUtils.closeQuietly(file);
+ summary.close();
+ assert summary.isCleanedUp();
+ summary = null;
}
}
http://git-wip-us.apache.org/repos/asf/cassandra/blob/4c94ef20/src/java/org/apache/cassandra/tools/StandaloneScrubber.java
----------------------------------------------------------------------
http://git-wip-us.apache.org/repos/asf/cassandra/blob/4c94ef20/test/unit/org/apache/cassandra/db/ScrubTest.java
----------------------------------------------------------------------