You are viewing a plain text version of this content. The canonical link for it is here.
Posted to commits@cassandra.apache.org by jb...@apache.org on 2011/08/11 21:29:34 UTC
svn commit: r1156757 - in /cassandra/trunk: ./
src/java/org/apache/cassandra/db/ src/java/org/apache/cassandra/db/commitlog/
test/unit/org/apache/cassandra/db/
Author: jbellis
Date: Thu Aug 11 19:29:34 2011
New Revision: 1156757
URL: http://svn.apache.org/viewvc?rev=1156757&view=rev
Log:
make sure pre-truncate CL segments are discarded
Modified:
cassandra/trunk/CHANGES.txt
cassandra/trunk/src/java/org/apache/cassandra/db/ColumnFamilyStore.java
cassandra/trunk/src/java/org/apache/cassandra/db/SystemTable.java
cassandra/trunk/src/java/org/apache/cassandra/db/commitlog/CommitLog.java
cassandra/trunk/test/unit/org/apache/cassandra/db/RecoveryManagerTruncateTest.java
Modified: cassandra/trunk/CHANGES.txt
URL: http://svn.apache.org/viewvc/cassandra/trunk/CHANGES.txt?rev=1156757&r1=1156756&r2=1156757&view=diff
==============================================================================
--- cassandra/trunk/CHANGES.txt (original)
+++ cassandra/trunk/CHANGES.txt Thu Aug 11 19:29:34 2011
@@ -232,7 +232,7 @@
* Disable compaction throttling during bootstrap (CASSANDRA-2612)
* fix CQL treatment of > and < operators in range slices (CASSANDRA-2592)
* fix potential double-application of counter updates on commitlog replay
- (CASSANDRA-2419)
+ by moving replay position from header to sstable metadata (CASSANDRA-2419)
* JDBC CQL driver exposes getColumn for access to timestamp
* JDBC ResultSetMetadata properties added to AbstractType
* r/m clustertool (CASSANDRA-2607)
Modified: cassandra/trunk/src/java/org/apache/cassandra/db/ColumnFamilyStore.java
URL: http://svn.apache.org/viewvc/cassandra/trunk/src/java/org/apache/cassandra/db/ColumnFamilyStore.java?rev=1156757&r1=1156756&r2=1156757&view=diff
==============================================================================
--- cassandra/trunk/src/java/org/apache/cassandra/db/ColumnFamilyStore.java (original)
+++ cassandra/trunk/src/java/org/apache/cassandra/db/ColumnFamilyStore.java Thu Aug 11 19:29:34 2011
@@ -581,6 +581,7 @@ public class ColumnFamilyStore implement
assert getMemtableThreadSafe() == oldMemtable;
oldMemtable.freeze();
final ReplayPosition ctx = writeCommitLog ? CommitLog.instance.getContext() : ReplayPosition.NONE;
+ logger.debug("flush position is {}", ctx);
// submit the memtable for any indexed sub-cfses, and our own.
List<ColumnFamilyStore> icc = new ArrayList<ColumnFamilyStore>();
@@ -1532,6 +1533,37 @@ public class ColumnFamilyStore implement
}
/**
+ * Waits for flushes started BEFORE THIS METHOD IS CALLED to finish.
+ * Does NOT guarantee that no flush is active when it returns.
+ */
+ private void waitForActiveFlushes()
+ {
+ Future<?> future;
+ Table.switchLock.writeLock().lock();
+ try
+ {
+ future = postFlushExecutor.submit(new Runnable() { public void run() { } });
+ }
+ finally
+ {
+ Table.switchLock.writeLock().unlock();
+ }
+
+ try
+ {
+ future.get();
+ }
+ catch (InterruptedException e)
+ {
+ throw new AssertionError(e);
+ }
+ catch (ExecutionException e)
+ {
+ throw new AssertionError(e);
+ }
+ }
+
+ /**
* Truncate practically deletes the entire column family's data
* @return a Future to the delete operation. Call the future's get() to make
* sure the column family has been deleted
@@ -1544,14 +1576,33 @@ public class ColumnFamilyStore implement
// We accomplish this by first flushing manually, then snapshotting, and
// recording the timestamp IN BETWEEN those actions. Any sstables created
// with this timestamp or greater time, will not be marked for delete.
- try
- {
- forceBlockingFlush();
- }
- catch (Exception e)
- {
- throw new RuntimeException(e);
- }
+ //
+ // Bonus complication: since we store replay position in sstable metadata,
+ // truncating those sstables means we will replay any CL segments from the
+ // beginning if we restart before they are discarded for normal reasons
+ // post-truncate. So we need to (a) force a new segment so the currently
+ // active one can be discarded, and (b) flush *all* CFs so that unflushed
+ // data in others don't keep any pre-truncate CL segments alive.
+ //
+ // Bonus bonus: simply forceFlush of all the CF is not enough, because if
+ // for a given column family the memtable is clean, forceFlush will return
+ // immediately, even though there could be a memtable being flush at the same
+ // time. So to guarantee that all segments can be cleaned out, we need
+ // "waitForActiveFlushes" after the new segment has been created.
+ CommitLog.instance.forceNewSegment();
+ waitForActiveFlushes();
+ List<Future<?>> futures = new ArrayList<Future<?>>();
+ ReplayPosition position = CommitLog.instance.getContext();
+ for (ColumnFamilyStore cfs : ColumnFamilyStore.all())
+ {
+ Future<?> f = cfs.forceFlush();
+ if (f != null)
+ futures.add(f);
+ }
+ FBUtilities.waitOnFutures(futures);
+ // if everything was clean, flush won't have called discard
+ CommitLog.instance.discardCompletedSegments(metadata.cfId, position);
+
// sleep a little to make sure that our truncatedAt comes after any sstable
// that was part of the flushed we forced; otherwise on a tie, it won't get deleted.
try
Modified: cassandra/trunk/src/java/org/apache/cassandra/db/SystemTable.java
URL: http://svn.apache.org/viewvc/cassandra/trunk/src/java/org/apache/cassandra/db/SystemTable.java?rev=1156757&r1=1156756&r2=1156757&view=diff
==============================================================================
--- cassandra/trunk/src/java/org/apache/cassandra/db/SystemTable.java (original)
+++ cassandra/trunk/src/java/org/apache/cassandra/db/SystemTable.java Thu Aug 11 19:29:34 2011
@@ -81,9 +81,14 @@ public class SystemTable
ColumnFamily cf = table.getColumnFamilyStore(STATUS_CF).getColumnFamily(dotSeven);
if (cf == null)
{
- // upgrading from 0.6 to 0.7.
- logger.info("Upgrading to 0.7. Purging hints if there are any. Old hints will be snapshotted.");
- new Truncation(Table.SYSTEM_TABLE, HintedHandOffManager.HINTS_CF).apply();
+ // 0.7+ marker not found. Remove hints and add the marker.
+ ColumnFamilyStore hintsCfs = Table.open(Table.SYSTEM_TABLE).getColumnFamilyStore(HintedHandOffManager.HINTS_CF);
+ if (hintsCfs.getSSTables().size() > 0)
+ {
+ logger.info("Possible 0.6-format hints found. Snapshotting as 'old-hints' and purging");
+ hintsCfs.snapshot("old-hints");
+ hintsCfs.removeAllSSTables();
+ }
RowMutation rm = new RowMutation(Table.SYSTEM_TABLE, COOKIE_KEY);
rm.add(new QueryPath(STATUS_CF, null, hintsPurged6to7), ByteBufferUtil.bytes("oh yes, it they were purged."), System.currentTimeMillis());
rm.apply();
Modified: cassandra/trunk/src/java/org/apache/cassandra/db/commitlog/CommitLog.java
URL: http://svn.apache.org/viewvc/cassandra/trunk/src/java/org/apache/cassandra/db/commitlog/CommitLog.java?rev=1156757&r1=1156756&r2=1156757&view=diff
==============================================================================
--- cassandra/trunk/src/java/org/apache/cassandra/db/commitlog/CommitLog.java (original)
+++ cassandra/trunk/src/java/org/apache/cassandra/db/commitlog/CommitLog.java Thu Aug 11 19:29:34 2011
@@ -496,7 +496,6 @@ public class CommitLog implements Commit
}
}
-
void sync() throws IOException
{
currentSegment().sync();
@@ -532,6 +531,50 @@ public class CommitLog implements Commit
return getSize();
}
+ public void forceNewSegment()
+ {
+ Callable<?> task = new Callable()
+ {
+ public Object call() throws IOException
+ {
+ createNewSegment();
+ return null;
+ }
+ };
+
+ try
+ {
+ executor.submit(task).get();
+ }
+ catch (InterruptedException e)
+ {
+ throw new AssertionError(e);
+ }
+ catch (ExecutionException e)
+ {
+ throw new RuntimeException(e);
+ }
+ }
+
+ private void createNewSegment() throws IOException
+ {
+ sync();
+ segments.add(new CommitLogSegment());
+
+ // Maintain desired CL size cap
+ if (getSize() >= DatabaseDescriptor.getTotalCommitlogSpaceInMB() * 1024 * 1024)
+ {
+ // Force a flush on all CFs keeping the oldest segment from being removed
+ CommitLogSegment oldestSegment = segments.peek();
+ assert oldestSegment != null; // has to be at least the one we just added
+ for (Integer dirtyCFId : oldestSegment.cfLastWrite.keySet())
+ {
+ String keypace = CFMetaData.getCF(dirtyCFId).left;
+ Table.open(keypace).getColumnFamilyStore(dirtyCFId).forceFlush();
+ }
+ }
+ }
+
// TODO this should be a Runnable since it doesn't actually return anything, but it's difficult to do that
// without breaking the fragile CheaterFutureTask in BatchCLES.
class LogRecordAdder implements Callable, Runnable
@@ -550,23 +593,7 @@ public class CommitLog implements Commit
currentSegment().write(rowMutation);
// roll log if necessary
if (currentSegment().length() >= SEGMENT_SIZE)
- {
- sync();
- segments.add(new CommitLogSegment());
-
- // Maintain desired CL size cap
- if (getSize() >= DatabaseDescriptor.getTotalCommitlogSpaceInMB() * 1024 * 1024)
- {
- // Force a flush on all CFs keeping the oldest segment from being removed
- CommitLogSegment oldestSegment = segments.peek();
- assert oldestSegment != null; // has to be at least the one we just added
- for (Integer dirtyCFId : oldestSegment.cfLastWrite.keySet())
- {
- String keypace = CFMetaData.getCF(dirtyCFId).left;
- Table.open(keypace).getColumnFamilyStore(dirtyCFId).forceFlush();
- }
- }
- }
+ createNewSegment();
}
catch (IOException e)
{
Modified: cassandra/trunk/test/unit/org/apache/cassandra/db/RecoveryManagerTruncateTest.java
URL: http://svn.apache.org/viewvc/cassandra/trunk/test/unit/org/apache/cassandra/db/RecoveryManagerTruncateTest.java?rev=1156757&r1=1156756&r2=1156757&view=diff
==============================================================================
--- cassandra/trunk/test/unit/org/apache/cassandra/db/RecoveryManagerTruncateTest.java (original)
+++ cassandra/trunk/test/unit/org/apache/cassandra/db/RecoveryManagerTruncateTest.java Thu Aug 11 19:29:34 2011
@@ -73,6 +73,7 @@ public class RecoveryManagerTruncateTest
rm.apply();
cfs.forceBlockingFlush();
cfs.truncate().get();
+ CommitLog.instance.resetUnsafe();
CommitLog.recover();
assertNull(getFromTable(table, "Standard1", "keymulti", "col1"));
}