com.bigdata.resources.ResourceManager Maven / Gradle / Ivy
Show all versions of bigdata-core Show documentation
/**
Copyright (C) SYSTAP, LLC DBA Blazegraph 2006-2016. All rights reserved.
Contact:
SYSTAP, LLC DBA Blazegraph
2501 Calvert ST NW #106
Washington, DC 20008
[email protected]
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 Mar 13, 2007
*/
package com.bigdata.resources;
import java.util.Iterator;
import java.util.Map;
import java.util.NoSuchElementException;
import java.util.Properties;
import java.util.TreeMap;
import org.apache.log4j.Logger;
import com.bigdata.btree.IndexSegmentBuilder;
import com.bigdata.cache.ICacheEntry;
import com.bigdata.counters.CounterSet;
import com.bigdata.counters.Instrument;
import com.bigdata.counters.OneShotInstrument;
import com.bigdata.journal.IConcurrencyManager;
import com.bigdata.mdi.LocalPartitionMetadata;
import com.bigdata.service.DataService;
import com.bigdata.service.IMetadataService;
import com.bigdata.service.MetadataService;
import com.bigdata.util.ReverseLongComparator;
/**
* The {@link ResourceManager} has broad responsibility for journal files, index
* segment files, maintaining index views during overflow processing, and
* managing the transparent decomposition of scale-out indices and the
* distribution of the key-range index partitions for those scale-out indices.
*
* This class is implemented in several layers:
*
* - {@link ResourceManager}
* - Concrete implementation.
* - {@link OverflowManager}
* - Overflow processing.
* - {@link IndexManager}
* - Manages indices
* - {@link StoreManager}
* - Manages the journal and index segment files, including the release of
* old resources.
* - {@link ResourceEvents}
* - Event reporting API
*
*
* @todo Document backup procedures for the journal (the journal is a
* log-structured store; it can be deployed on RAID for media robustness;
* can be safely copied using normal file copy mechanisms and restored;
* can be compacted offline; a compact snapshot of a commit point (such as
* the last commit point) can be generated online and used for recovery;
* and can be exported onto index segments which can later be restored to
* any journal, but those index segments need additional metadata to
* recreate the appropriate relations which is found in the sparse row
* store) and the federation (relies on service failover (primary and
* secondaries) and a history retention policy; can be recovered from
* existing index partitions in a bottom up matter, but that recovery code
* has not been written).
*
* @author Bryan Thompson
* @version $Id$
*/
abstract public class ResourceManager extends OverflowManager implements
IPartitionIdFactory {
/**
* Logger.
*/
protected static final Logger log = Logger.getLogger(ResourceManager.class);
/**
* Interface defines and documents the counters and counter namespaces for
* the {@link ResourceManager}.
*
* @author Bryan Thompson
* @version $Id$
*/
public static interface IResourceManagerCounters
extends IOverflowManagerCounters, IIndexManagerCounters, IStoreManagerCounters {
/**
* The namespace for counters pertaining to the {@link OverflowManager}.
*/
String OverflowManager = "Overflow Manager";
/**
* The namespace for counters pertaining for overflow tasks (within the
* {@link #OverflowManager} namespace).
*/
String IndexPartitionTasks = "Overflow Tasks";
/**
* The namespace for counters pertaining to the {@link IndexManager}.
*/
String IndexManager = "Index Manager";
/**
* The namespace for counters pertaining to the {@link StoreManager}.
*/
String StoreManager = "Store Manager";
/**
* The namespace for counters pertaining to the live
* {@link ManagedJournal}.
*
* Note: these counters are detached and reattached to the new live
* journal during overflow processing.
*/
String LiveJournal = "Live Journal";
}
// /**
// * WARNING: The {@link DataService} transfers all of the children
// * from this object into the hierarchy reported by
// * {@link AbstractFederation#getServiceCounterSet()} and this object will be
// * empty thereafter.
// */
/*synchronized*/ public CounterSet getCounters() {
// if (root == null) {
final CounterSet root = new CounterSet();
// ResourceManager
{
// ... nothing really - its all under other headings.
}
// Live Journal
{
/*
* Note: these counters are detached and reattached to the new
* live journal during overflow processing.
*
* @todo This assumes that the StoreManager is running.
* Normally, this will be true since the DataService does not
* setup its counter set until the store manager is running and
* the service UUID has been assigned. However, eagerly
* requesting the counters set would violate that assumption and
* cause an exception to be thrown here since the live journal
* is not defined until the StoreManager is running.
*
* It would be best to modify this to attach the live journal
* counters when the StoreManager startup completes successfully
* rather than assuming that it already has done so. However,
* the counter set for the ResourceManager is not currently
* defined until the StoreManager is running...
*/
root.makePath(IResourceManagerCounters.LiveJournal).attach(
getLiveJournal().getCounters());
}
// OverflowManager
{
final CounterSet tmp = root
.makePath(IResourceManagerCounters.OverflowManager);
tmp.addCounter(IOverflowManagerCounters.OverflowEnabled,
new Instrument() {
public void sample() {
setValue(isOverflowEnabled());
}
});
tmp.addCounter(IOverflowManagerCounters.OverflowAllowed,
new Instrument() {
public void sample() {
setValue(isOverflowAllowed());
}
});
tmp.addCounter(IOverflowManagerCounters.ShouldOverflow,
new Instrument() {
public void sample() {
setValue(shouldOverflow());
}
});
tmp.addCounter(IOverflowManagerCounters.SynchronousOverflowCount,
new Instrument() {
public void sample() {
setValue(getSynchronousOverflowCount());
}
});
tmp.attach(overflowCounters.getCounters());
/*
* Note: In order to report this counter we need access to the
* OverflowManager itself.
*/
tmp.addCounter(
IOverflowManagerCounters.AsynchronousOverflowMillis,
new Instrument() {
public void sample() {
long t = overflowCounters.asynchronousOverflowMillis
.get();
if (isOverflowEnabled() && !isOverflowAllowed()) {
/*
* Include time from the active (ongoing)
* asynchronous overflow operation.
*/
t += (System.currentTimeMillis() - overflowCounters.asynchronousOverflowStartMillis
.get());
}
setValue(t);
}
});
/*
* These are some counters tracked by the IndexManager but
* reported under the same path as the OverflowManager's
* per-index partition task counters.
*/
{
final CounterSet tmp2 = tmp
.makePath(IResourceManagerCounters.IndexPartitionTasks);
tmp2.addCounter(IIndexPartitionTaskCounters.ConcurrentBuildCount,
new Instrument() {
public void sample() {
setValue(concurrentBuildTaskCount.get());
}
});
tmp2.addCounter(IIndexPartitionTaskCounters.ConcurrentMergeCount,
new Instrument() {
public void sample() {
setValue(concurrentMergeTaskCount.get());
}
});
tmp2.addCounter(IIndexPartitionTaskCounters.RunningBuilds,
new Instrument() {
public void sample() {
/*
* Put the running tasks into order by their
* elapsed execution time.
*/
final TreeMap map = new TreeMap(
new ReverseLongComparator());
final long now = System.currentTimeMillis();
for (IndexSegmentBuilder task : buildTasks
.values()) {
final long startTime = task
.getStartTime();
if (startTime == 0) {
// task has not started.
continue;
}
final long elapsed = (startTime == 0 ? 0L
: now - startTime);
map.put(elapsed, task);
}
/*
* Format the list of running tasks.
*/
// int n = 0;
final StringBuilder sb = new StringBuilder();
// sb.append("There are " + map.size()
// + " running index segment builds.");
for(Map.Entry e : map.entrySet()) {
final long elapsed = e.getKey();
final IndexSegmentBuilder task = e
.getValue();
// if (n > 0)
// sb.append(" ");
final String name = task.metadata
.getName();
final LocalPartitionMetadata pmd = task.metadata
.getPartitionMetadata();
final int partitionId = (pmd == null ? -1
: pmd.getPartitionId());
final String cause = (pmd == null ? "N/A"
: pmd.getIndexPartitionCause()
.toString());
final String indexPartitionName = DataService
.getIndexPartitionName(name,
partitionId);
sb.append(indexPartitionName
+ "{elapsed=" + elapsed
+ "ms, cause=" + cause
+ ", compactingMerge="
+ task.compactingMerge
+ ", commitTime=" + task.commitTime
+ ", plan=" + task.plan + "} ");
}
setValue(sb.toString());
}
});
}
}
// IndexManager
{
final CounterSet tmp = root
.makePath(IResourceManagerCounters.IndexManager);
// // save a reference.
// indexManagerRoot = tmp;
tmp.addCounter(IIndexManagerCounters.StaleLocatorCacheCapacity,
new Instrument() {
public void sample() {
setValue(staleLocatorCache.capacity());
}
});
tmp.addCounter(IIndexManagerCounters.StaleLocatorCacheSize,
new Instrument() {
public void sample() {
setValue(getStaleLocatorCount());
}
});
if(false)
tmp.addCounter(IIndexManagerCounters.StaleLocators,
new Instrument() {
public void sample() {
final StringBuilder sb = new StringBuilder();
final Iterator> itr = staleLocatorCache
.entryIterator();
while (itr.hasNext()) {
try {
final ICacheEntry entry = itr
.next();
sb.append(entry.getKey() + "="
+ entry.getObject() + "\n");
} catch (NoSuchElementException ex) {
// Ignore - concurrent modification.
}
}
setValue(sb.toString());
}
});
tmp.addCounter(IIndexManagerCounters.IndexCount,
new Instrument() {
public void sample() {
final ManagedJournal liveJournal = getLiveJournal();
final long lastCommitTime = liveJournal.getLastCommitTime();
if (lastCommitTime == 0L) {
/*
* This warning will be issued for the first
* live journal for a data service since
* there are no commit points until the
* application registers an index on that
* data service.
*/
if(log.isInfoEnabled())
log.info("No commit points on the live journal?");
return;
}
final long indexCount = liveJournal
.getName2Addr(lastCommitTime)
.rangeCount();
setValue(indexCount);
}
});
tmp.addCounter(IIndexManagerCounters.IndexCacheCapacity,
new Instrument() {
public void sample() {
setValue(getIndexCacheCapacity());
}
});
tmp.addCounter(IIndexManagerCounters.IndexCacheSize,
new Instrument() {
public void sample() {
setValue(getIndexCacheSize());
}
});
tmp.addCounter(IIndexManagerCounters.IndexSegmentCacheCapacity,
new Instrument() {
public void sample() {
setValue(getIndexSegmentCacheCapacity());
}
});
tmp.addCounter(IIndexManagerCounters.IndexSegmentCacheSize,
new Instrument() {
public void sample() {
setValue(getIndexSegmentCacheSize());
}
});
// tmp.addCounter(IIndexManagerCounters.IndexSegmentOpenLeafCount,
// new Instrument() {
// public void sample() {
// setValue(getIndexSegmentOpenLeafCount());
// }
// });
//
// tmp.addCounter(IIndexManagerCounters.IndexSegmentOpenLeafByteCount,
// new Instrument() {
// public void sample() {
// setValue(getIndexSegmentOpenLeafByteCount());
// }
// });
// attach the index partition counters.
tmp.makePath(IIndexManagerCounters.Indices).attach(
getIndexCounters());
}
// StoreManager
{
final CounterSet tmp = root
.makePath(IResourceManagerCounters.StoreManager);
tmp.addCounter(IStoreManagerCounters.DataDir,
new Instrument() {
public void sample() {
setValue(dataDir == null ? "N/A" : dataDir
.getAbsolutePath());
}
});
tmp.addCounter(IStoreManagerCounters.TmpDir,
new Instrument() {
public void sample() {
setValue(tmpDir == null ? "N/A" : tmpDir
.getAbsolutePath());
}
});
tmp.addCounter(IStoreManagerCounters.IsOpen,
new Instrument() {
public void sample() {
setValue(isOpen());
}
});
tmp.addCounter(IStoreManagerCounters.IsStarting,
new Instrument() {
public void sample() {
setValue(isStarting());
}
});
tmp.addCounter(IStoreManagerCounters.IsRunning,
new Instrument() {
public void sample() {
setValue(isRunning());
}
});
tmp.addCounter(IStoreManagerCounters.StoreCacheCapacity,
new OneShotInstrument(storeCache.capacity()));
tmp.addCounter(IStoreManagerCounters.StoreCacheSize,
new Instrument() {
public void sample() {
setValue((long) getStoreCacheSize());
}
});
tmp.addCounter(IStoreManagerCounters.ManagedJournalCount,
new Instrument() {
public void sample() {
setValue(getManagedJournalCount());
}
});
tmp.addCounter(IStoreManagerCounters.ManagedSegmentStoreCount,
new Instrument() {
public void sample() {
setValue(getManagedSegmentCount());
}
});
tmp.addCounter(IStoreManagerCounters.JournalReopenCount,
new Instrument() {
public void sample() {
setValue(journalReopenCount.get());
}
});
tmp.addCounter(IStoreManagerCounters.SegmentStoreReopenCount,
new Instrument() {
public void sample() {
setValue(segmentStoreReopenCount.get());
}
});
tmp.addCounter(IStoreManagerCounters.JournalDeleteCount,
new Instrument() {
public void sample() {
setValue(journalDeleteCount.get());
}
});
tmp.addCounter(IStoreManagerCounters.SegmentStoreDeleteCount,
new Instrument() {
public void sample() {
setValue(segmentStoreDeleteCount.get());
}
});
tmp.addCounter(IStoreManagerCounters.BytesUnderManagement,
new Instrument() {
public void sample() {
if (isRunning()) {
setValue(getBytesUnderManagement());
}
}
});
tmp.addCounter(IStoreManagerCounters.JournalBytesUnderManagement,
new Instrument() {
public void sample() {
if (isRunning()) {
setValue(getJournalBytesUnderManagement());
}
}
});
tmp.addCounter(IStoreManagerCounters.SegmentBytesUnderManagement,
new Instrument() {
public void sample() {
if (isRunning()) {
setValue(getSegmentBytesUnderManagement());
}
}
});
tmp.addCounter(IStoreManagerCounters.BytesDeleted,
new Instrument() {
public void sample() {
setValue(bytesDeleted.get());
}
});
tmp.addCounter(IStoreManagerCounters.DataDirBytesAvailable,
new Instrument() {
public void sample() {
if (!isTransient())
setValue(getDataDirFreeSpace());
}
});
tmp.addCounter(IStoreManagerCounters.TmpDirBytesAvailable,
new Instrument() {
public void sample() {
setValue(getTempDirFreeSpace());
}
});
tmp.addCounter(
IStoreManagerCounters.MaximumJournalSizeAtOverflow,
new Instrument() {
public void sample() {
setValue(maximumJournalSizeAtOverflow);
}
});
tmp.addCounter(IStoreManagerCounters.PurgeResourcesMillis,
new Instrument() {
public void sample() {
setValue(purgeResourcesMillis);
}
});
tmp.addCounter(IStoreManagerCounters.ReleaseTime,
new Instrument() {
public void sample() {
setValue(getReleaseTime());
}
});
tmp.addCounter(IStoreManagerCounters.LastOverflowTime,
new Instrument() {
public void sample() {
setValue(lastOverflowTime);
}
});
tmp.addCounter(IStoreManagerCounters.LastCommitTimePreserved,
new Instrument() {
public void sample() {
setValue(lastCommitTimePreserved);
}
});
tmp.addCounter(IStoreManagerCounters.LastCommitTime,
new Instrument() {
public void sample() {
final ManagedJournal liveJournal;
try {
liveJournal = getLiveJournal();
setValue(liveJournal.getLastCommitTime());
} catch (Throwable t) {
log.warn(t);
}
}
});
/*
* Performance counters for the service used to let other data
* services read index segments or journals from this service.
*/
final CounterSet tmp2 = tmp.makePath("resourceService");
tmp2.attach(getResourceService().counters.getCounters());
}
// }
return root;
}
// private CounterSet root;
// /**
// * The counter set that corresponds to the {@link IndexManager}.
// */
// public CounterSet getIndexManagerCounters() {
//
// if (indexManagerRoot == null) {
//
// getCounters();
//
// }
//
// return indexManagerRoot;
//
// }
// private CounterSet indexManagerRoot;
/**
* {@link ResourceManager} options.
*
* @author Bryan Thompson
* @version $Id$
*/
public static interface Options extends OverflowManager.Options {
}
private IConcurrencyManager concurrencyManager;
public IConcurrencyManager getConcurrencyManager() {
if (concurrencyManager == null) {
// Not assigned!
throw new IllegalStateException();
}
return concurrencyManager;
}
public void setConcurrencyManager(
final IConcurrencyManager concurrencyManager) {
if (concurrencyManager == null)
throw new IllegalArgumentException();
if (this.concurrencyManager != null)
throw new IllegalStateException();
this.concurrencyManager = concurrencyManager;
}
/**
* (Re-)open the {@link ResourceManager}.
*
* Note: You MUST use {@link #setConcurrencyManager(IConcurrencyManager)}
* after calling this constructor (the parameter can not be passed in since
* there is a circular dependency between the {@link IConcurrencyManager}
* and {@link ManagedJournal#getLocalTransactionManager()}.
*
* @param properties
* See {@link Options}.
*
* @see DataService#start()
*/
public ResourceManager(final Properties properties) {
super(properties);
}
/**
* Requests a new index partition identifier from the
* {@link MetadataService} for the specified scale-out index (RMI).
*
* @return The new index partition identifier.
*
* @throws RuntimeException
* if something goes wrong.
*/
public int nextPartitionId(final String scaleOutIndexName) {
final IMetadataService mds = getFederation().getMetadataService();
if (mds == null) {
throw new RuntimeException("Metadata service not discovered.");
}
try {
// obtain new partition identifier from the metadata service (RMI)
final int newPartitionId = mds.nextPartitionId(scaleOutIndexName);
return newPartitionId;
} catch (Throwable t) {
throw new RuntimeException(t);
}
}
// /**
// * Impose flow control on data destined for the named index partition. When
// * invoked, clients will no longer be authorized to buffer data on this data
// * service to be written on the named index partition.
// *
// * @param name
// * The index partition.
// *
// * @todo Implement flow control (this method is a place holder).
// */
// protected void suspendWrites(String name) {
//
// // NOP
//
// }
//
// /**
// * Allow clients to buffer writes for the named index partition.
// *
// * @param name
// *
// * @todo Implement flow control (this method is a place holder).
// */
// protected void resumeWrites(String name) {
//
// // NOP
//
// }
}