/*
Copyright (C) SYSTAP, LLC DBA Blazegraph 2006-2016. All rights reserved.
Contact:
SYSTAP, LLC DBA Blazegraph
2501 Calvert ST NW #106
Washington, DC 20008
licenses@blazegraph.com
This program is free software; you can redistribute it and/or modify
it under the terms of the GNU General Public License as published by
the Free Software Foundation; version 2 of the License.
This program is distributed in the hope that it will be useful,
but WITHOUT ANY WARRANTY; without even the implied warranty of
MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
GNU General Public License for more details.
You should have received a copy of the GNU General Public License
along with this program; if not, write to the Free Software
Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA 02111-1307 USA
*/
/*
* Created on Apr 16, 2009
*/
package com.bigdata.service.ndx.pipeline;
import java.lang.ref.WeakReference;
import java.util.Iterator;
import java.util.LinkedHashMap;
import java.util.Map;
import java.util.concurrent.atomic.AtomicInteger;
import java.util.concurrent.atomic.AtomicLong;
import com.bigdata.cache.ConcurrentWeakValueCache;
import com.bigdata.counters.CounterSet;
import com.bigdata.counters.Instrument;
import com.bigdata.relation.accesspath.BlockingBuffer;
import com.bigdata.resources.StaleLocatorException;
/**
* Abstract base class providing statistics for the {@link AbstractMasterTask}
* and a factory for the statistics for the subtasks.
* <p>
* Note: Since there are concurrent threads which need to write on the counters
* on this class the practice is to be <code>synchronized</code> on this
* reference before you update those counters. Without this, the counters might
* not update correctly as thing like <code>a += 2</code> may not produce the
* correct result.
* <p>
* Note: The use of {@link AtomicLong}s provides atomic visibility changes
* relied on by some unit tests.
*
* @param <L>
* The generic type of the key used to lookup subtasks in the
* internal map.
* @param <HS>
* The generic type of the subtask statistics.
*
* @author <a href="mailto:thompsonbry@users.sourceforge.net">Bryan Thompson</a>
* @version $Id$
*/
abstract public class AbstractMasterStats<L, HS extends AbstractSubtaskStats> {
/**
* The #of subtasks which have started.
*/
public final AtomicLong subtaskStartCount = new AtomicLong();
/**
* The #of subtasks which have finished (either the buffer has been closed
* and all buffered data has been flushed -or- the task was interrupted or
* otherwise threw an exception).
*/
public final AtomicLong subtaskEndCount = new AtomicLong();
/**
* The #of subtasks which were closed due to an idle timeout.
*/
public final AtomicLong subtaskIdleTimeoutCount = new AtomicLong();
/**
* The maximum #of distinct partitions for which the master has caused
* subtasks to be created at any given time.
*/
final private AtomicInteger maximumPartitionCount = new AtomicInteger();
/**
* The maximum #of distinct partitions for which the master has caused
* subtasks to be created at any given time.
*/
public int getMaximumPartitionCount() {
return maximumPartitionCount.get();
}
/**
* The #of active sinks.
*/
public int getActiveSinkCount() {
return sinkStats.size();
}
/**
* The #of redirects ({@link StaleLocatorException}s) that were handled.
*/
public final AtomicLong redirectCount = new AtomicLong();
/**
* Elapsed nanoseconds handling a redirect.
*/
public long elapsedRedirectNanos = 0L;
/**
* The #of chunks drained from the {@link BlockingBuffer} by the
* {@link AbstractMasterTask}.
*/
public final AtomicLong chunksIn = new AtomicLong();
/**
* The #of elements drained from the {@link BlockingBuffer} by the
* {@link AbstractMasterTask}.
*/
public final AtomicLong elementsIn = new AtomicLong();
/**
* The #of elements in the output chunks written onto the index partitions
* (not including any eliminated duplicates).
*/
public final AtomicLong elementsOut = new AtomicLong();
/**
* The #of elements on the output sink queues. This is incremented when a
* chunk of elements is transferred onto an output sink queue and
* decremented when a chunk of elements is drained from an output sink
* queue. It does not reflect the #of elements on the master queue, which
* can be approximated as {@link #elementsTransferred}/
* {@link #chunksTransferred} X the averageMasterQueueSize). Neither does it
* include the #of elements in a prepared or outstanding write on an index
* partition.
*/
public final AtomicLong elementsOnSinkQueues = new AtomicLong();
/**
* The #of chunks transferred from the master to the sinks. Where there is
* more than one index partition, there will be more than one sink and each
* chunk written on the master will be divided among the sinks based on the
* key-ranges of the tuples in the chunks. Therefore each chunk drained from
* the master will be registered as some larger #of chunks transferred to
* the sink(s) for that master.
*/
public final AtomicLong chunksTransferred = new AtomicLong();
/**
* The #of elements transferred from the master to the sinks. Where there is
* more than one index partition, there will be more than one sink and each
* chunk written on the master will be divided among the sinks based on the
* key-ranges of the tuples in the chunks. This reduces the average chunk
* size entering the sink accordingly.
*/
public final AtomicLong elementsTransferred = new AtomicLong();
/**
* The #of chunks written onto index partitions using RMI.
*/
public final AtomicLong chunksOut = new AtomicLong();
/**
* Elapsed nanoseconds the master spends offering a chunk for transfer to a
* sink.
*/
public long elapsedSinkOfferNanos = 0L;
/**
* Elapsed time across sinks waiting for another chunk to be ready so that
* it can be written onto the index partition.
*/
public long elapsedSinkChunkWaitingNanos = 0L;
/**
* Elapsed nanoseconds across sinks writing chunks on an index partition
* (RMI requests).
*/
public long elapsedSinkChunkWritingNanos = 0L;
/**
* Map for the per-index partition statistics. This ensures that we can
* report the aggregate statistics in detail. A weak value hash map is used
* so that the statistics for inactive index partitions can be discarded.
* The backing hard reference queue is not used since the
* {@link AbstractSubtask sink task}s are strongly held until they are
* closed so we don't really need a backing hard reference queue at all in
* this case.
*/
private final ConcurrentWeakValueCache<L, HS> sinkStats = new ConcurrentWeakValueCache<L, HS>(
0/* queueCapacity */);
/**
* Weak value hash map of the active masters.
*/
protected final ConcurrentWeakValueCache<Integer, AbstractMasterTask> masters = new ConcurrentWeakValueCache<Integer, AbstractMasterTask>(
0/* queueCapacity */);
/**
* The #of master tasks which have been created for the index whose
* asynchronous write statistics are reported on by this object.
*/
public final AtomicInteger masterCreateCount = new AtomicInteger();
/**
* The approximate #of active master tasks. This is based on a weak value
* hash map. The size of that map is reported.
*/
public int getMasterActiveCount() {
/*
* Note: Map entries for cleared weak references are removed from the
* map before the size of that map is reported.
*/
return masters.size();
}
/**
* A new master task declares itself to this statistics object using this
* method. This allows the statistics object to report on the #of master
* tasks, their queue sizes, and the sizes of their sink queues.
*
* @param master
*/
@SuppressWarnings("unchecked")
void addMaster(final AbstractMasterTask master) {
if (master == null)
throw new IllegalArgumentException();
if (masters.putIfAbsent(master.hashCode(), master) == null) {
masterCreateCount.incrementAndGet();
}
}
/**
* Return the statistics object for the specified index partition and never
* <code>null</code> (a new instance is created if none exists).
*
* @param locator
* The index partition.
*
* @return The statistics for that index partition.
*/
public HS getSubtaskStats(final L locator) {
synchronized (sinkStats) {
HS t = sinkStats.get(locator);
if (t == null) {
t = newSubtaskStats(locator);
sinkStats.put(locator, t);
// tmp.add(t);
maximumPartitionCount.set(Math.max(maximumPartitionCount.get(),
sinkStats.size()));
}
return t;
}
}
// private final List<HS> tmp = new LinkedList<HS>();
/**
* Factory for the subtask statistics.
*
* @param locator
* The subtask key.
*
* @return The statistics for the subtask.
*/
protected abstract HS newSubtaskStats(L locator);
/**
* Return a snapshot of the statistics for each index partition.
*/
public Map<L, HS> getSubtaskStats() {
final Map<L, HS> m = new LinkedHashMap<L, HS>(sinkStats
.size());
final Iterator<Map.Entry<L, WeakReference<HS>>> itr = sinkStats
.entryIterator();
while (itr.hasNext()) {
final Map.Entry<L, WeakReference<HS>> e = itr.next();
final HS subtaskStats = e.getValue().get();
if (subtaskStats == null) {
// weak reference was cleared.
continue;
}
m.put(e.getKey(), subtaskStats);
}
return m;
}
public AbstractMasterStats() {
}
/**
* Return a {@link CounterSet} which may be used to report the statistics on
* the index write operation. The {@link CounterSet} is NOT placed into any
* namespace.
*/
public CounterSet getCounterSet() {
final CounterSet t = new CounterSet();
t.addCounter("masterCreateCount", new Instrument<Integer>() {
@Override
protected void sample() {
setValue(masterCreateCount.get());
}
});
t.addCounter("masterActiveCount", new Instrument<Integer>() {
@Override
protected void sample() {
setValue(getMasterActiveCount());
}
});
t.addCounter("subtaskStartCount", new Instrument<Long>() {
@Override
protected void sample() {
setValue(subtaskStartCount.get());
}
});
t.addCounter("subtaskEndCount", new Instrument<Long>() {
@Override
protected void sample() {
setValue(subtaskEndCount.get());
}
});
t.addCounter("subtaskIdleTimeout", new Instrument<Long>() {
@Override
protected void sample() {
setValue(subtaskIdleTimeoutCount.get());
}
});
t.addCounter("maximumPartitionCount", new Instrument<Integer>() {
@Override
protected void sample() {
setValue(maximumPartitionCount.get());
}
});
t.addCounter("activePartitionCount", new Instrument<Integer>() {
@Override
protected void sample() {
setValue(sinkStats.size());
}
});
t.addCounter("redirectCount", new Instrument<Long>() {
@Override
protected void sample() {
setValue(redirectCount.get());
}
});
t.addCounter("elapsedRedirectNanos", new Instrument<Long>() {
@Override
protected void sample() {
setValue(elapsedRedirectNanos);
}
});
t.addCounter("chunksIn", new Instrument<Long>() {
@Override
protected void sample() {
setValue(chunksIn.get());
}
});
t.addCounter("elementsIn", new Instrument<Long>() {
@Override
protected void sample() {
setValue(elementsIn.get());
}
});
t.addCounter("chunksOut", new Instrument<Long>() {
@Override
protected void sample() {
setValue(chunksOut.get());
}
});
t.addCounter("elementsOut", new Instrument<Long>() {
@Override
protected void sample() {
setValue(elementsOut.get());
}
});
t.addCounter("chunksTransferred", new Instrument<Long>() {
@Override
protected void sample() {
setValue(chunksTransferred.get());
}
});
t.addCounter("elementsTransferred", new Instrument<Long>() {
@Override
protected void sample() {
setValue(elementsTransferred.get());
}
});
t.addCounter("elapsedSinkOfferNanos", new Instrument<Long>() {
@Override
protected void sample() {
setValue(elapsedSinkOfferNanos);
}
});
t.addCounter("elapsedChunkWaitingNanos", new Instrument<Long>() {
@Override
protected void sample() {
setValue(elapsedSinkChunkWaitingNanos);
}
});
t.addCounter("elapsedChunkWritingNanos", new Instrument<Long>() {
@Override
protected void sample() {
setValue(elapsedSinkChunkWritingNanos);
}
});
// t.addCounter("averageMillisPerWait", new Instrument<Long>() {
// @Override
// protected void sample() {
// setValue(TimeUnit.NANOSECONDS
// .toMillis((long) getAverageNanosPerWait()));
// }
// });
//
// t.addCounter("averageMillisPerWrite", new Instrument<Long>() {
// @Override
// protected void sample() {
// setValue(TimeUnit.NANOSECONDS
// .toMillis((long) getAverageNanosPerWrite()));
// }
// });
//
// t.addCounter("averageElementsPerWrite", new Instrument<Double>() {
// @Override
// protected void sample() {
// setValue(getAverageElementsPerWrite());
// }
// });
return t;
}
// /**
// * The average #of nanoseconds waiting for a chunk to become ready so that
// * it can be written on an output sink.
// */
// public double getAverageNanosPerWait() {
//
// return (chunksOut == 0L ? 0 : elapsedChunkWaitingNanos
// / (double) chunksOut);
//
// }
//
// /**
// * The average #of nanoseconds per chunk written on an output sink.
// */
// public double getAverageNanosPerWrite() {
//
// return (chunksOut == 0L ? 0 : elapsedChunkWritingNanos
// / (double) chunksOut);
//
// }
//
// /**
// * The average #of elements (tuples) per chunk written on an output sink.
// */
// public double getAverageElementsPerWrite() {
//
// return (chunksOut == 0L ? 0 : elementsOut / (double) chunksOut);
//
// }
public String toString() {
return getClass().getName()
+ "{subtaskStartCount="
+ subtaskStartCount
+ ", subtaskEndCount="
+ subtaskEndCount
+ ", subtaskIdleTimeout="
+ subtaskIdleTimeoutCount
+ ", partitionCount="
+ maximumPartitionCount
+ ", activePartitionCount="
+ getActiveSinkCount()
+ ", redirectCount="
+ redirectCount
+ ", chunkIn="
+ chunksIn
+ ", elementIn="
+ elementsIn
+ ", chunksOut="
+ chunksOut
+ ", elementsOut="
+ elementsOut
+ ", elementsOnSinkQueues="
+ elementsOnSinkQueues
+ ", chunksTransferred="
+ chunksTransferred
+ ", elementsTransferred="
+ elementsTransferred
+ ", elapsedChunkWaitingNanos="
+ elapsedSinkChunkWaitingNanos
+ ", elapsedChunkWritingNanos="
+ elapsedSinkChunkWritingNanos
// + ", averageMillisPerWait="
// + TimeUnit.NANOSECONDS
// .toMillis((long) getAverageNanosPerWait())
// + ", averageMillisPerWrite="
// + TimeUnit.NANOSECONDS
// .toMillis((long) getAverageNanosPerWrite())
// + ", averageElementsPerWrite=" + getAverageElementsPerWrite()
+ "}";
}
}