TraditionalWorkQueue.java revision ea1068c292e9b341af6d6b563cd8988a96be20a9
/*
* CDDL HEADER START
*
* The contents of this file are subject to the terms of the
* Common Development and Distribution License, Version 1.0 only
* (the "License"). You may not use this file except in compliance
* with the License.
*
* You can obtain a copy of the license at legal-notices/CDDLv1_0.txt
* or http://forgerock.org/license/CDDLv1.0.html.
* See the License for the specific language governing permissions
* and limitations under the License.
*
* When distributing Covered Code, include this CDDL HEADER in each
* file and include the License file at legal-notices/CDDLv1_0.txt.
* If applicable, add the following below this CDDL HEADER, with the
* fields enclosed by brackets "[]" replaced with your own identifying
* information:
* Portions Copyright [yyyy] [name of copyright owner]
*
* CDDL HEADER END
*
*
* Copyright 2006-2010 Sun Microsystems, Inc.
* Portions Copyright 2013-2015 ForgeRock AS.
*/
package org.opends.server.extensions;
import static org.opends.messages.ConfigMessages.*;
import static org.opends.messages.CoreMessages.*;
import java.util.ArrayList;
import java.util.List;
import java.util.concurrent.LinkedBlockingQueue;
import java.util.concurrent.TimeUnit;
import java.util.concurrent.atomic.AtomicLong;
import java.util.concurrent.locks.ReentrantReadWriteLock;
import java.util.concurrent.locks.ReentrantReadWriteLock.ReadLock;
import java.util.concurrent.locks.ReentrantReadWriteLock.WriteLock;
import org.forgerock.i18n.LocalizableMessage;
import org.forgerock.i18n.slf4j.LocalizedLogger;
import org.forgerock.opendj.config.server.ConfigException;
import org.forgerock.opendj.ldap.ResultCode;
import org.opends.server.admin.server.ConfigurationChangeListener;
import org.opends.server.admin.std.server.TraditionalWorkQueueCfg;
import org.opends.server.api.WorkQueue;
import org.opends.server.core.DirectoryServer;
import org.opends.server.monitors.TraditionalWorkQueueMonitor;
import org.opends.server.types.CancelRequest;
import org.forgerock.opendj.config.server.ConfigChangeResult;
import org.opends.server.types.DirectoryException;
import org.opends.server.types.InitializationException;
import org.opends.server.types.Operation;
/**
* This class defines a data structure for storing and interacting with the
* Directory Server work queue.
*/
public class TraditionalWorkQueue extends WorkQueue<TraditionalWorkQueueCfg>
implements ConfigurationChangeListener<TraditionalWorkQueueCfg>
{
private static final LocalizedLogger logger = LocalizedLogger.getLoggerForThisClass();
/**
* The maximum number of times to retry getting the next operation from the
* queue if an unexpected failure occurs.
*/
private static final int MAX_RETRY_COUNT = 5;
/** The set of worker threads that will be used to process this work queue. */
private final ArrayList<TraditionalWorkerThread> workerThreads =
new ArrayList<TraditionalWorkerThread>();
/**
* The number of operations that have been submitted to the work queue for
* processing.
*/
private AtomicLong opsSubmitted;
/**
* The number of times that an attempt to submit a new request has been
* rejected because the work queue is already at its maximum capacity.
*/
private AtomicLong queueFullRejects;
/**
* Indicates whether one or more of the worker threads needs to be killed at
* the next convenient opportunity.
*/
private boolean killThreads;
/** Indicates whether the Directory Server is shutting down. */
private boolean shutdownRequested;
/** The thread number used for the last worker thread that was created. */
private int lastThreadNumber;
/**
* The maximum number of pending requests that this work queue will allow
* before it will start rejecting them.
*/
private int maxCapacity;
/**
* The number of worker threads that should be active (or will be shortly if a
* configuration change has not been completely applied).
*/
private int numWorkerThreads;
/**
* The queue overflow policy: true indicates that operations will be blocked
* until the queue has available capacity, otherwise operations will be
* rejected.
* <p>
* This is hard-coded to true for now because a reject on full policy does not
* seem to have a valid use case.
* </p>
*/
private final boolean isBlocking = true;
/** The queue that will be used to actually hold the pending operations. */
private LinkedBlockingQueue<Operation> opQueue;
/**
* The lock used to provide threadsafe access for the queue, used for
* non-config changes.
*/
private final ReadLock queueReadLock;
/**
* The lock used to provide threadsafe access for the queue, used for config
* changes.
*/
private final WriteLock queueWriteLock;
{
ReentrantReadWriteLock lock = new ReentrantReadWriteLock();
queueReadLock = lock.readLock();
queueWriteLock = lock.writeLock();
}
/**
* Creates a new instance of this work queue. All initialization should be
* performed in the <CODE>initializeWorkQueue</CODE> method.
*/
public TraditionalWorkQueue()
{
// No implementation should be performed here.
}
/**
* {@inheritDoc}
*/
@Override
public void initializeWorkQueue(TraditionalWorkQueueCfg configuration)
throws ConfigException, InitializationException
{
queueWriteLock.lock();
try
{
shutdownRequested = false;
killThreads = false;
opsSubmitted = new AtomicLong(0);
queueFullRejects = new AtomicLong(0);
// Register to be notified of any configuration changes.
configuration.addTraditionalChangeListener(this);
// Get the necessary configuration from the provided entry.
numWorkerThreads =
computeNumWorkerThreads(configuration.getNumWorkerThreads());
maxCapacity = configuration.getMaxWorkQueueCapacity();
// Create the actual work queue.
if (maxCapacity > 0)
{
opQueue = new LinkedBlockingQueue<Operation>(maxCapacity);
}
else
{
// This will never be the case, since the configuration definition
// ensures that the capacity is always finite.
opQueue = new LinkedBlockingQueue<Operation>();
}
// Create the set of worker threads that should be used to service the
// work queue.
for (lastThreadNumber = 0; lastThreadNumber < numWorkerThreads;
lastThreadNumber++)
{
TraditionalWorkerThread t = new TraditionalWorkerThread(this,
lastThreadNumber);
t.start();
workerThreads.add(t);
}
// Create and register a monitor provider for the work queue.
try
{
TraditionalWorkQueueMonitor monitor = new TraditionalWorkQueueMonitor(
this);
monitor.initializeMonitorProvider(null);
DirectoryServer.registerMonitorProvider(monitor);
}
catch (Exception e)
{
logger.traceException(e);
logger.error(ERR_CONFIG_WORK_QUEUE_CANNOT_CREATE_MONITOR, TraditionalWorkQueueMonitor.class, e);
}
}
finally
{
queueWriteLock.unlock();
}
}
/**
* {@inheritDoc}
*/
@Override
public void finalizeWorkQueue(LocalizableMessage reason)
{
queueWriteLock.lock();
try
{
shutdownRequested = true;
}
finally
{
queueWriteLock.unlock();
}
// From now on no more operations can be enqueued or dequeued.
// Send responses to any operations in the pending queue to indicate that
// they won't be processed because the server is shutting down.
CancelRequest cancelRequest = new CancelRequest(true, reason);
ArrayList<Operation> pendingOperations = new ArrayList<Operation>();
opQueue.removeAll(pendingOperations);
for (Operation o : pendingOperations)
{
try
{
// The operation has no chance of responding to the cancel
// request so avoid waiting for a cancel response.
if (o.getCancelResult() == null)
{
o.abort(cancelRequest);
}
}
catch (Exception e)
{
logger.traceException(e);
logger.warn(WARN_QUEUE_UNABLE_TO_CANCEL, o, e);
}
}
// Notify all the worker threads of the shutdown.
for (TraditionalWorkerThread t : workerThreads)
{
try
{
t.shutDown();
}
catch (Exception e)
{
logger.traceException(e);
logger.warn(WARN_QUEUE_UNABLE_TO_NOTIFY_THREAD, t.getName(), e);
}
}
}
/**
* Indicates whether this work queue has received a request to shut down.
*
* @return <CODE>true</CODE> if the work queue has recieved a request to shut
* down, or <CODE>false</CODE> if not.
*/
public boolean shutdownRequested()
{
queueReadLock.lock();
try
{
return shutdownRequested;
}
finally
{
queueReadLock.unlock();
}
}
/**
* Submits an operation to be processed by one of the worker threads
* associated with this work queue.
*
* @param operation
* The operation to be processed.
* @throws DirectoryException
* If the provided operation is not accepted for some reason (e.g.,
* if the server is shutting down or the pending operation queue is
* already at its maximum capacity).
*/
@Override
public void submitOperation(Operation operation) throws DirectoryException
{
submitOperation(operation, isBlocking);
}
/** {@inheritDoc} */
@Override
public boolean trySubmitOperation(Operation operation)
throws DirectoryException
{
try
{
submitOperation(operation, false);
return true;
}
catch (DirectoryException e)
{
if (ResultCode.BUSY == e.getResultCode())
{
return false;
}
throw e;
}
}
private void submitOperation(Operation operation,
boolean blockEnqueuingWhenFull) throws DirectoryException
{
queueReadLock.lock();
try
{
if (shutdownRequested)
{
LocalizableMessage message = WARN_OP_REJECTED_BY_SHUTDOWN.get();
throw new DirectoryException(ResultCode.UNAVAILABLE, message);
}
if (blockEnqueuingWhenFull)
{
try
{
// If the queue is full and there is an administrative change taking
// place then starvation could arise: this thread will hold the read
// lock, the admin thread will be waiting on the write lock, and the
// worker threads may be queued behind the admin thread. Since the
// worker threads cannot run, the queue will never empty and allow
// this thread to proceed. To help things out we can periodically
// yield the read lock when the queue is full.
while (!opQueue.offer(operation, 1, TimeUnit.SECONDS))
{
queueReadLock.unlock();
Thread.yield();
queueReadLock.lock();
if (shutdownRequested)
{
LocalizableMessage message = WARN_OP_REJECTED_BY_SHUTDOWN.get();
throw new DirectoryException(ResultCode.UNAVAILABLE, message);
}
}
}
catch (InterruptedException e)
{
// We cannot handle the interruption here. Reject the request and
// re-interrupt this thread.
Thread.currentThread().interrupt();
queueFullRejects.incrementAndGet();
LocalizableMessage message = WARN_OP_REJECTED_BY_QUEUE_INTERRUPT.get();
throw new DirectoryException(ResultCode.BUSY, message);
}
}
else
{
if (!opQueue.offer(operation))
{
queueFullRejects.incrementAndGet();
LocalizableMessage message = WARN_OP_REJECTED_BY_QUEUE_FULL.get(maxCapacity);
throw new DirectoryException(ResultCode.BUSY, message);
}
}
opsSubmitted.incrementAndGet();
}
finally
{
queueReadLock.unlock();
}
}
/**
* Retrieves the next operation that should be processed by one of the worker
* threads, blocking if necessary until a new request arrives. This method
* should only be called by a worker thread associated with this work queue.
*
* @param workerThread
* The worker thread that is requesting the operation.
* @return The next operation that should be processed, or <CODE>null</CODE>
* if the server is shutting down and no more operations will be
* processed.
*/
public Operation nextOperation(TraditionalWorkerThread workerThread)
{
return retryNextOperation(workerThread, 0);
}
/**
* Retrieves the next operation that should be processed by one of the worker
* threads following a previous failure attempt. A maximum of five consecutive
* failures will be allowed before returning <CODE>null</CODE>, which will
* cause the associated thread to exit.
*
* @param workerThread
* The worker thread that is requesting the operation.
* @param numFailures
* The number of consecutive failures that the worker thread has
* experienced so far. If this gets too high, then this method will
* return <CODE>null</CODE> rather than retrying.
* @return The next operation that should be processed, or <CODE>null</CODE>
* if the server is shutting down and no more operations will be
* processed, or if there have been too many consecutive failures.
*/
private Operation retryNextOperation(TraditionalWorkerThread workerThread,
int numFailures)
{
// See if we should kill off this thread. This could be necessary if the
// number of worker threads has been decreased with the server online. If
// so, then return null and the thread will exit.
queueReadLock.lock();
try
{
if (shutdownRequested)
{
return null;
}
if (killThreads && tryKillThisWorkerThread(workerThread))
{
return null;
}
if (numFailures > MAX_RETRY_COUNT)
{
logger.error(ERR_CONFIG_WORK_QUEUE_TOO_MANY_FAILURES, Thread
.currentThread().getName(), numFailures, MAX_RETRY_COUNT);
return null;
}
while (true)
{
Operation nextOperation = opQueue.poll(5, TimeUnit.SECONDS);
if (nextOperation != null)
{
return nextOperation;
}
// There was no work to do in the specified length of time. Release the
// read lock allowing shutdown or config changes to proceed and then see
// if we should give up or check again.
queueReadLock.unlock();
Thread.yield();
queueReadLock.lock();
if (shutdownRequested)
{
return null;
}
if (killThreads && tryKillThisWorkerThread(workerThread))
{
return null;
}
}
}
catch (InterruptedException ie)
{
// This is somewhat expected so don't log.
// assert debugException(CLASS_NAME, "retryNextOperation", ie);
// If this occurs, then the worker thread must have been interrupted for
// some reason. This could be because the Directory Server is shutting
// down, in which case we should return null.
if (shutdownRequested)
{
return null;
}
// If we've gotten here, then the worker thread was interrupted for some
// other reason. This should not happen, and we need to log a message.
logger.warn(WARN_WORKER_INTERRUPTED_WITHOUT_SHUTDOWN, Thread.currentThread().getName(), ie);
}
catch (Exception e)
{
logger.traceException(e);
// This should not happen. The only recourse we have is to log a message
// and try again.
logger.warn(WARN_WORKER_WAITING_UNCAUGHT_EXCEPTION, Thread.currentThread().getName(), e);
}
finally
{
queueReadLock.unlock();
}
// An exception has occurred - retry.
return retryNextOperation(workerThread, numFailures + 1);
}
/**
* Kills this worker thread if needed. This method assumes that the read lock
* is already taken and ensure that it is taken on exit.
*
* @param workerThread
* The worker thread associated with this thread.
* @return {@code true} if this thread was killed or is about to be killed as
* a result of shutdown.
*/
private boolean tryKillThisWorkerThread(TraditionalWorkerThread workerThread)
{
queueReadLock.unlock();
queueWriteLock.lock();
try
{
if (shutdownRequested)
{
// Shutdown may have been requested between unlock/lock. This thread is
// about to shutdown anyway, so return true.
return true;
}
int currentThreads = workerThreads.size();
if (currentThreads > numWorkerThreads)
{
if (workerThreads.remove(Thread.currentThread()))
{
currentThreads--;
}
if (currentThreads <= numWorkerThreads)
{
killThreads = false;
}
workerThread.setStoppedByReducedThreadNumber();
return true;
}
}
finally
{
queueWriteLock.unlock();
queueReadLock.lock();
if (shutdownRequested)
{
// Shutdown may have been requested between unlock/lock. This thread is
// about to shutdown anyway, so return true.
return true;
}
}
return false;
}
/**
* Retrieves the total number of operations that have been successfully
* submitted to this work queue for processing since server startup. This does
* not include operations that have been rejected for some reason like the
* queue already at its maximum capacity.
*
* @return The total number of operations that have been successfully
* submitted to this work queue since startup.
*/
public long getOpsSubmitted()
{
return opsSubmitted.longValue();
}
/**
* Retrieves the total number of operations that have been rejected because
* the work queue was already at its maximum capacity.
*
* @return The total number of operations that have been rejected because the
* work queue was already at its maximum capacity.
*/
public long getOpsRejectedDueToQueueFull()
{
return queueFullRejects.longValue();
}
/**
* Retrieves the number of pending operations in the queue that have not yet
* been picked up for processing. Note that this method is not a constant-time
* operation and can be relatively inefficient, so it should be used
* sparingly.
*
* @return The number of pending operations in the queue that have not yet
* been picked up for processing.
*/
public int size()
{
queueReadLock.lock();
try
{
return opQueue.size();
}
finally
{
queueReadLock.unlock();
}
}
/**
* {@inheritDoc}
*/
@Override
public boolean isConfigurationChangeAcceptable(
TraditionalWorkQueueCfg configuration, List<LocalizableMessage> unacceptableReasons)
{
return true;
}
/**
* {@inheritDoc}
*/
@Override
public ConfigChangeResult applyConfigurationChange(
TraditionalWorkQueueCfg configuration)
{
int newNumThreads =
computeNumWorkerThreads(configuration.getNumWorkerThreads());
int newMaxCapacity = configuration.getMaxWorkQueueCapacity();
// Apply a change to the number of worker threads if appropriate.
int currentThreads = workerThreads.size();
if (newNumThreads != currentThreads)
{
queueWriteLock.lock();
try
{
int threadsToAdd = newNumThreads - currentThreads;
if (threadsToAdd > 0)
{
for (int i = 0; i < threadsToAdd; i++)
{
TraditionalWorkerThread t = new TraditionalWorkerThread(this,
lastThreadNumber++);
workerThreads.add(t);
t.start();
}
killThreads = false;
}
else
{
killThreads = true;
}
numWorkerThreads = newNumThreads;
}
catch (Exception e)
{
logger.traceException(e);
}
finally
{
queueWriteLock.unlock();
}
}
// Apply a change to the maximum capacity if appropriate. Since we can't
// change capacity on the fly, then we'll have to create a new queue and
// transfer any remaining items into it. Any thread that is waiting on the
// original queue will time out after at most a few seconds and further
// checks will be against the new queue.
if (newMaxCapacity != maxCapacity)
{
// First switch the queue with the exclusive lock.
queueWriteLock.lock();
LinkedBlockingQueue<Operation> oldOpQueue;
try
{
LinkedBlockingQueue<Operation> newOpQueue = null;
if (newMaxCapacity > 0)
{
newOpQueue = new LinkedBlockingQueue<Operation>(
newMaxCapacity);
}
else
{
newOpQueue = new LinkedBlockingQueue<Operation>();
}
oldOpQueue = opQueue;
opQueue = newOpQueue;
maxCapacity = newMaxCapacity;
}
finally
{
queueWriteLock.unlock();
}
// Now resubmit any pending requests - we'll need the shared lock.
Operation pendingOperation = null;
queueReadLock.lock();
try
{
// We have to be careful when adding any existing pending operations
// because the new capacity could be less than what was already
// backlogged in the previous queue. If that happens, we may have to
// loop a few times to get everything in there.
while ((pendingOperation = oldOpQueue.poll()) != null)
{
opQueue.put(pendingOperation);
}
}
catch (InterruptedException e)
{
// We cannot handle the interruption here. Cancel pending requests and
// re-interrupt this thread.
Thread.currentThread().interrupt();
LocalizableMessage message = WARN_OP_REJECTED_BY_QUEUE_INTERRUPT.get();
CancelRequest cancelRequest = new CancelRequest(true, message);
if (pendingOperation != null)
{
pendingOperation.abort(cancelRequest);
}
while ((pendingOperation = oldOpQueue.poll()) != null)
{
pendingOperation.abort(cancelRequest);
}
}
finally
{
queueReadLock.unlock();
}
}
return new ConfigChangeResult();
}
/**
* {@inheritDoc}
*/
@Override
public boolean isIdle()
{
queueReadLock.lock();
try
{
if (opQueue.size() > 0)
{
return false;
}
for (TraditionalWorkerThread t : workerThreads)
{
if (t.isActive())
{
return false;
}
}
return true;
}
finally
{
queueReadLock.unlock();
}
}
/**
* Return the number of worker threads used by this WorkQueue.
*
* @return the number of worker threads used by this WorkQueue
*/
@Override
public int getNumWorkerThreads()
{
return this.numWorkerThreads;
}
}