You are viewing a plain text version of this content. The canonical link for it is here.
Posted to commits@hbase.apache.org by ns...@apache.org on 2011/10/11 04:05:32 UTC
svn commit: r1181401 - in
/hbase/branches/0.89/src/main/java/org/apache/hadoop/hbase/regionserver:
CompactSplitThread.java HRegion.java metrics/RegionServerMetrics.java
Author: nspiegelberg
Date: Tue Oct 11 02:05:31 2011
New Revision: 1181401
URL: http://svn.apache.org/viewvc?rev=1181401&view=rev
Log:
HBASE-3102: Add Compaction Time to JMX
Summary:
Added the ability to track compaction count + duration. Since compactions are
long running, the avg/opcount stats are should be reset longer than the polling
period (60 sec). Added the idea of a 'marathon', or a long duration after which
some metrics should expire. Compaction stats & all min/max stats expire after
the marathon (1 hr).
Test Plan:
mvn clean install
loaded on cluster
DiffCamp Revision: 167960
Reviewed By: kannan
CC: nspiegelberg, kannan
Revert Plan:
OK
Modified:
hbase/branches/0.89/src/main/java/org/apache/hadoop/hbase/regionserver/CompactSplitThread.java
hbase/branches/0.89/src/main/java/org/apache/hadoop/hbase/regionserver/HRegion.java
hbase/branches/0.89/src/main/java/org/apache/hadoop/hbase/regionserver/metrics/RegionServerMetrics.java
Modified: hbase/branches/0.89/src/main/java/org/apache/hadoop/hbase/regionserver/CompactSplitThread.java
URL: http://svn.apache.org/viewvc/hbase/branches/0.89/src/main/java/org/apache/hadoop/hbase/regionserver/CompactSplitThread.java?rev=1181401&r1=1181400&r2=1181401&view=diff
==============================================================================
--- hbase/branches/0.89/src/main/java/org/apache/hadoop/hbase/regionserver/CompactSplitThread.java (original)
+++ hbase/branches/0.89/src/main/java/org/apache/hadoop/hbase/regionserver/CompactSplitThread.java Tue Oct 11 02:05:31 2011
@@ -101,6 +101,10 @@ class CompactSplitThread extends Thread
if(!this.server.isStopRequested()) {
// Don't interrupt us while we are working
byte [] midKey = r.compactStores();
+ long lastCompactPeriod = r.getLastCompactPeriod();
+ if (lastCompactPeriod > 0) { // 0 == compaction aborted
+ this.server.getMetrics().addCompaction(lastCompactPeriod);
+ }
LOG.debug("Just finished a compaction. " +
" Current Compaction Queue Size: " +
getCompactionQueueSize());
Modified: hbase/branches/0.89/src/main/java/org/apache/hadoop/hbase/regionserver/HRegion.java
URL: http://svn.apache.org/viewvc/hbase/branches/0.89/src/main/java/org/apache/hadoop/hbase/regionserver/HRegion.java?rev=1181401&r1=1181400&r2=1181401&view=diff
==============================================================================
--- hbase/branches/0.89/src/main/java/org/apache/hadoop/hbase/regionserver/HRegion.java (original)
+++ hbase/branches/0.89/src/main/java/org/apache/hadoop/hbase/regionserver/HRegion.java Tue Oct 11 02:05:31 2011
@@ -177,6 +177,7 @@ public class HRegion implements HeapSize
* major compaction. Cleared each time through compaction code.
*/
private volatile boolean forceMajorCompaction = false;
+ private long lastCompactPeriod = 0;
/*
* Data structure of write state flags used coordinating flushes,
@@ -606,6 +607,11 @@ public class HRegion implements HeapSize
return this.fs;
}
+ /** @return how long the last compaction took */
+ public long getLastCompactPeriod() {
+ return this.lastCompactPeriod;
+ }
+
/** @return the last time the region was flushed */
public long getLastFlushTime() {
return this.lastFlushTime;
@@ -847,6 +853,7 @@ public class HRegion implements HeapSize
LOG.info(((completed) ? "completed" : "aborted")
+ " compaction on region " + this
+ " after " + StringUtils.formatTimeDiff(now, startTime));
+ this.lastCompactPeriod = (completed) ? (now - startTime) / 1000 : 0;
}
} finally {
synchronized (writestate) {
Modified: hbase/branches/0.89/src/main/java/org/apache/hadoop/hbase/regionserver/metrics/RegionServerMetrics.java
URL: http://svn.apache.org/viewvc/hbase/branches/0.89/src/main/java/org/apache/hadoop/hbase/regionserver/metrics/RegionServerMetrics.java?rev=1181401&r1=1181400&r2=1181401&view=diff
==============================================================================
--- hbase/branches/0.89/src/main/java/org/apache/hadoop/hbase/regionserver/metrics/RegionServerMetrics.java (original)
+++ hbase/branches/0.89/src/main/java/org/apache/hadoop/hbase/regionserver/metrics/RegionServerMetrics.java Tue Oct 11 02:05:31 2011
@@ -25,6 +25,7 @@ import org.apache.hadoop.hbase.io.hfile.
import org.apache.hadoop.hbase.metrics.MetricsRate;
import org.apache.hadoop.hbase.regionserver.wal.HLog;
import org.apache.hadoop.hbase.util.Strings;
+import org.apache.hadoop.metrics.ContextFactory;
import org.apache.hadoop.metrics.MetricsContext;
import org.apache.hadoop.metrics.MetricsRecord;
import org.apache.hadoop.metrics.MetricsUtil;
@@ -35,6 +36,7 @@ import org.apache.hadoop.metrics.util.Me
import org.apache.hadoop.metrics.util.MetricsRegistry;
import org.apache.hadoop.metrics.util.MetricsTimeVaryingRate;
+import java.io.IOException;
import java.lang.management.ManagementFactory;
import java.lang.management.MemoryUsage;
@@ -50,6 +52,8 @@ public class RegionServerMetrics impleme
private final Log LOG = LogFactory.getLog(this.getClass());
private final MetricsRecord metricsRecord;
private long lastUpdate = System.currentTimeMillis();
+ private long lastMarathon = System.currentTimeMillis();
+ private long marathonPeriod = 0;
private static final int MB = 1024*1024;
private MetricsRegistry registry = new MetricsRegistry();
private final RegionServerStatistics statistics;
@@ -134,6 +138,12 @@ public class RegionServerMetrics impleme
public final MetricsTimeVaryingRate fsSyncLatency =
new MetricsTimeVaryingRate("fsSyncLatency", registry);
+ /**
+ * time each scheduled compaction takes
+ */
+ public final MetricsTimeVaryingRate compaction =
+ new MetricsTimeVaryingRate("compaction", registry);
+
public RegionServerMetrics() {
MetricsContext context = MetricsUtil.getContext("hbase");
metricsRecord = MetricsUtil.createRecord(context, "regionserver");
@@ -146,6 +156,16 @@ public class RegionServerMetrics impleme
// export for JMX
statistics = new RegionServerStatistics(this.registry, name);
+ // get custom attributes
+ try {
+ Object m = ContextFactory.getFactory().getAttribute("hbase.marathon");
+ if (m instanceof String) {
+ this.marathonPeriod = Long.parseLong((String) m)*1000;
+ }
+ } catch (IOException ioe) {
+ LOG.info("Couldn't load ContextFactory for Metrics config info");
+ }
+
LOG.info("Initialized");
}
@@ -157,10 +177,20 @@ public class RegionServerMetrics impleme
/**
* Since this object is a registered updater, this method will be called
* periodically, e.g. every 5 seconds.
- * @param unused unused argument
+ * @param caller the metrics context that this responsible for calling us
*/
- public void doUpdates(MetricsContext unused) {
+ public void doUpdates(MetricsContext caller) {
synchronized (this) {
+ this.lastUpdate = System.currentTimeMillis();
+
+ // has the marathon period for long-living stats elapsed?
+ boolean marathonExpired = (this.marathonPeriod > 0) &&
+ (this.lastUpdate - this.lastMarathon >= this.marathonPeriod);
+ if (marathonExpired) {
+ this.lastMarathon = this.lastUpdate;
+ this.resetAllMinMax();
+ }
+
this.stores.pushMetric(this.metricsRecord);
this.storefiles.pushMetric(this.metricsRecord);
this.storefileIndexSizeMB.pushMetric(this.metricsRecord);
@@ -196,15 +226,39 @@ public class RegionServerMetrics impleme
this.fsReadLatency.pushMetric(this.metricsRecord);
this.fsWriteLatency.pushMetric(this.metricsRecord);
this.fsSyncLatency.pushMetric(this.metricsRecord);
+ this.compaction.pushMetric(this.metricsRecord);
+
+ if (!marathonExpired) {
+ maintainStats(this.compaction);
+ }
}
this.metricsRecord.update();
- this.lastUpdate = System.currentTimeMillis();
+ }
+
+ /* MetricsTimeVaryingRate will reset every time pushMetric() is called
+ * This is annoying for long-running stats that might not get a single
+ * operation in the polling period. This function ensures that values
+ * for those stat entries don't get reset.
+ */
+ protected void maintainStats(MetricsTimeVaryingRate stat) {
+ int curOps = stat.getPreviousIntervalNumOps();
+ if (curOps > 0) {
+ long curTime = stat.getPreviousIntervalAverageTime();
+ long totalTime = curTime * curOps;
+ if (totalTime / curTime == curOps) {
+ stat.inc(curOps, totalTime);
+ } else {
+ LOG.info("Stats for " + stat.getName() + " overflowed! resetting");
+ }
+ }
}
public void resetAllMinMax() {
this.atomicIncrementTime.resetMinMax();
this.fsReadLatency.resetMinMax();
this.fsWriteLatency.resetMinMax();
+ this.fsSyncLatency.resetMinMax();
+ this.compaction.resetMinMax();
}
/**
@@ -215,6 +269,15 @@ public class RegionServerMetrics impleme
}
/**
+ * @param periodSec seconds that last compaction took
+ */
+ public void addCompaction(final long periodSec) {
+ synchronized (this) {
+ this.compaction.inc(periodSec);
+ }
+ }
+
+ /**
* @param inc How much to add to requests.
*/
public void incrementRequests(final int inc) {