[rules-dev] Deadlock in the Drools core - Drools 5.0 - any suggestions for resolution?

Michael Neale michael.neale at gmail.com
Wed Nov 4 05:26:15 EST 2009 

ha - was just musing with someone the other day who uses "duration"
anymore ;) I guess its still useful to people !

I would say that the "duration" codebase is probably fairly "old" - in
the sense that it probably pre-dates the availability of
j.u.concurrent (which was java 5 I think? ) - so please try out that
patch, if it works, we can probably pull it in (hoping Edson can take
a look).

On Wed, Nov 4, 2009 at 4:43 PM, Greg Barton <greg_barton at yahoo.com> wrote:
> Well, I'm not sure how to avoid the deadlock without changing the drools codebase.  I was, however, able to change the type of AbstractWorkingMemory.actionQueue to java.util.concurrent.ConcurrentLinkedQueue and remove the synchronization over the queue with no apparent ill effects. (Two tests failed for drools-core, but they failed whether the change was made or not.)  Also I don't like the fact that the current code synchronizes on actionQueue, but then exposes it outside the class through the getActionQueue() method, where access can be unsynchronized.  Changing it to ConcurrentLinkedQueue makes it safe to expose externally. (Not to mention that the lock can be stolen externally with the current code.)
>
> diff attached.  If you can run drools compiled from trunk, apply the diff and see if it resolves the deadlock.  If it does it's up to the drools devs as to whether the change should be made.  I'm just hacking about. :P
>
> --- On Tue, 11/3/09, Edward Archibald <edward.archibald at continuent.com> wrote:
>
>> From: Edward Archibald <edward.archibald at continuent.com>
>> Subject: [rules-dev] Deadlock in the Drools core - Drools 5.0 - any suggestions for resolution?
>> To: "rules-dev at lists.jboss.org" <rules-dev at lists.jboss.org>
>> Date: Tuesday, November 3, 2009, 9:41 PM
>>
>> I found the following deadlock which is, apparently, due to
>> the concurrent execution
>> of a task for a 'delayed' rule with a concurrently
>> executing application thread attempting to get access to a
>> 'global'.  Any recommendations for avoiding this type
>> of deadlock besides not using rules with 'duration()' etc.
>> which cause asynchronous execution with respect to my main
>> application thread?
>>
>> This problem is somewhat difficult to reproduce on demand
>> but it does come up frequently when the 'delayed' rule
>> "DETECT MONITORING HAS STOPPED" is activated as a result of
>> the trigger conditions.
>>
>> ===================================================================================
>>
>> This thread, my application's EnterprisePolicyManager
>> thread, is attempting to get access to a global, policyMgr,
>> and is waiting for
>> the 'lock.lock' on RetooStatefulSession
>>
>> It owns the 'ReteooStatefulSession.actionQueue'
>> and is waiting for the ReteooStatefulSession.lock.lock
>>
>> owns: java.util.LinkedList<E>  (id=207)
>> waited by: Thread [pool-3-thread-1] (Suspended)
>> owns:
>> com.continuent.tungsten.cluster.manager.rules.engine.RulesEngine
>> (id=208)
>> sun.misc.Unsafe.park(boolean, long) line: not available
>> [native method] [local variables unavailable]
>> java.util.concurrent.locks.LockSupport.park() line: 118
>> [local variables unavailable]
>> java.util.concurrent.locks.AbstractQueuedSynchronizer.parkAndCheckInterrupt()
>> line: 681 [local variables unavailable]
>> java.util.concurrent.locks.ReentrantLock$NonfairSync(java.util.concurrent.locks.AbstractQueuedSynchronizer).acquireQueued(java.util.concurrent.locks.AbstractQueuedSynchronizer$Node,
>> int) line: 711
>> java.util.concurrent.locks.ReentrantLock$NonfairSync(java.util.concurrent.locks.AbstractQueuedSynchronizer).acquire(int)
>> line: 1041
>> java.util.concurrent.locks.ReentrantLock$NonfairSync.lock()
>> line: 184 [local variables unavailable]
>> java.util.concurrent.locks.ReentrantLock.lock() line: 256
>> [local variables unavailable]
>> org.drools.reteoo.ReteooStatefulSession(org.drools.common.AbstractWorkingMemory).getGlobal(java.lang.String)
>> line: 587
>> com.continuent.tungsten.cluster.manager.policy.Rule_IF_IN_MAINTENANCE_MODE__CONSUME_ALL_NOTIFICATIONS_0Eval0Invoker.evaluate(org.drools.spi.Tuple,
>> org.drools.rule.Declaration[], org.drools.WorkingMemory,
>> java.lang.Object) line: not available
>> org.drools.rule.EvalCondition.isAllowed(org.drools.spi.Tuple,
>> org.drools.WorkingMemory, java.lang.Object) line: 117
>> org.drools.reteoo.EvalConditionNode.assertLeftTuple(org.drools.reteoo.LeftTuple,
>> org.drools.spi.PropagationContext,
>> org.drools.common.InternalWorkingMemory) line: 180
>> org.drools.reteoo.SingleLeftTupleSinkAdapter.doPropagateAssertLeftTuple(org.drools.spi.PropagationContext,
>> org.drools.common.InternalWorkingMemory,
>> org.drools.reteoo.LeftTuple) line: 117
>> org.drools.reteoo.SingleLeftTupleSinkAdapter.propagateAssertLeftTuple(org.drools.reteoo.LeftTuple,
>> org.drools.reteoo.RightTuple,
>> org.drools.spi.PropagationContext,
>> org.drools.common.InternalWorkingMemory, boolean) line: 28
>> org.drools.reteoo.JoinNode.assertObject(org.drools.common.InternalFactHandle,
>> org.drools.spi.PropagationContext,
>> org.drools.common.InternalWorkingMemory) line: 175
>> org.drools.reteoo.SingleObjectSinkAdapter.propagateAssertObject(org.drools.common.InternalFactHandle,
>> org.drools.spi.PropagationContext,
>> org.drools.common.InternalWorkingMemory) line: 42
>> org.drools.reteoo.PropagationQueuingNode$AssertAction.execute(org.drools.reteoo.ObjectSinkPropagator,
>> org.drools.common.InternalWorkingMemory) line: 326
>> org.drools.reteoo.PropagationQueuingNode.propagateActions(org.drools.common.InternalWorkingMemory)
>> line: 221
>> org.drools.reteoo.PropagationQueuingNode$PropagateAction.execute(org.drools.common.InternalWorkingMemory)
>> line: 394
>> org.drools.reteoo.ReteooStatefulSession(org.drools.common.AbstractWorkingMemory).executeQueuedActions()
>> line: 1486
>> org.drools.common.NamedEntryPoint.insert(org.drools.common.InternalFactHandle,
>> java.lang.Object, org.drools.rule.Rule,
>> org.drools.spi.Activation) line: 158
>> org.drools.common.NamedEntryPoint.insert(java.lang.Object,
>> boolean, boolean, org.drools.rule.Rule,
>> org.drools.spi.Activation) line: 122
>> org.drools.common.NamedEntryPoint.insert(java.lang.Object)
>> line: 80
>> com.continuent.tungsten.cluster.manager.rules.engine.RulesEngine.insertFact(com.continuent.tungsten.commons.cluster.resource.notification.NotificationStreamID,
>> java.lang.Object, boolean) line: 162
>> com.continuent.tungsten.cluster.manager.policy.EnterprisePolicyManager.run()
>> line: 249
>> java.lang.Thread.run() line: 595
>>
>> The rule implicated in the above thread is:
>>
>> rule "IF IN MAINTENANCE MODE, CONSUME ALL NOTIFICATIONS"
>> salience 999
>>   when
>>     notification : ClusterResourceNotification()
>> from entry-point "MONITORING"
>>     eval(policyMgr.getMode() ==
>> ClusterPolicyManagerMode.MAINTENANCE)
>>   then
>>      statistics.increment("IF IN
>> MAINTENANCE MODE, CONSUME ALL NOTIFICATIONS");
>>     retract(notification);
>> end
>>
>>
>>
>> This other thread, apparently a scheduled thread for a rule
>> with a 10 second duration,
>> is attempting to insert a fact and owns the 'lock.lock' on
>> ReteooStatefulSession and
>> is waiting for the 'ReteooStatefulSession.actionQueue'.
>>
>> owns: org.drools.common.DefaultAgenda  (id=4046)
>> waiting for: java.util.LinkedList<E>  (id=207)
>> org.drools.reteoo.ReteooStatefulSession(org.drools.common.AbstractWorkingMemory).executeQueuedActions()
>> line: 1480
>> org.drools.reteoo.ReteooStatefulSession(org.drools.common.AbstractWorkingMemory).insert(org.drools.common.InternalFactHandle,
>> java.lang.Object, org.drools.rule.Rule,
>> org.drools.spi.Activation, org.drools.reteoo.ObjectTypeConf)
>> line: 1051
>> org.drools.reteoo.ReteooStatefulSession(org.drools.common.AbstractWorkingMemory).insert(java.lang.Object,
>> boolean, boolean, org.drools.rule.Rule,
>> org.drools.spi.Activation) line: 1001
>> org.drools.base.DefaultKnowledgeHelper.insert(java.lang.Object,
>> boolean) line: 114
>> org.drools.base.DefaultKnowledgeHelper.insert(java.lang.Object)
>> line: 108
>> com.continuent.tungsten.cluster.manager.policy.Rule_DETECT_MONITORING_HAS_STOPPED_0.consequence(org.drools.spi.KnowledgeHelper,
>> com.continuent.tungsten.commons.cluster.resource.notification.DataServerNotification,
>> org.drools.FactHandle, java.lang.String,
>> org.drools.FactHandle,
>> com.continuent.tungsten.cluster.manager.policy.EnterprisePolicyManager,
>> org.apache.log4j.Logger) line: not available
>> com.continuent.tungsten.cluster.manager.policy.Rule_DETECT_MONITORING_HAS_STOPPED_0ConsequenceInvoker.evaluate(org.drools.spi.KnowledgeHelper,
>> org.drools.WorkingMemory) line: not available
>> org.drools.common.DefaultAgenda.fireActivation(org.drools.spi.Activation)
>> line: 934
>> org.drools.common.Scheduler$DuractionJob.execute(org.drools.time.JobContext)
>> line: 70
>> org.drools.time.impl.JDKTimerService$JDKCallableJob.call()
>> line: 132
>> org.drools.time.impl.JDKTimerService$JDKCallableJob.call()
>> line: 110
>> java.util.concurrent.FutureTask$Sync.innerRun() line: 269
>> java.util.concurrent.ScheduledThreadPoolExecutor$ScheduledFutureTask<V>(java.util.concurrent.FutureTask<V>).run()
>> line: 123
>> java.util.concurrent.ScheduledThreadPoolExecutor$ScheduledFutureTask<V>.access$301(java.util.concurrent.ScheduledThreadPoolExecutor$ScheduledFutureTask)
>> line: 65
>> java.util.concurrent.ScheduledThreadPoolExecutor$ScheduledFutureTask<V>.run()
>> line: 168
>> java.util.concurrent.ThreadPoolExecutor$Worker.runTask(java.lang.Runnable)
>> line: 650
>> java.util.concurrent.ThreadPoolExecutor$Worker.run() line:
>> 675
>> java.lang.Thread.run() line: 595
>>
>> The rule for this task looks like:
>> rule "DETECT MONITORING HAS STOPPED"
>> duration(10s)
>> salience 1000
>>   when
>>     lastNotification :
>> DataServerNotification($resourceName : resourceName)
>>
>>   from entry-point "MONITORING"
>>
>>     not (DataServerNotification(resourceName ==
>> $resourceName,
>>
>>
>>    this after [10s] lastNotification)
>>
>>   from entry-point "MONITORING")
>>
>>     not (ManagerFailedAlarm(expired == false,
>>
>>    resourceName == $resourceName))
>>
>>     not (DataSource(name == $resourceName,
>>
>>    state == ResourceState.SHUNNED ||
>>
>>    state == ResourceState.FAILED))
>>
>>   then
>>     Object[] params = {$resourceName};
>>     if (policyMgr.getMode() !=
>> ClusterPolicyManagerMode.MAINTENANCE)
>>     {
>>
>> lastNotification.setResourceState(ResourceState.UNKNOWN);
>>       ManagerFailedAlarm alarm =
>>
>>         new
>> ManagerFailedAlarm(lastNotification, "rule detected monitor
>> stop",
>>
>>
>>    6, AlarmSeverity.FAULT);
>>       logger.info(alarm.toString());
>>       insert(alarm);
>>       update(lastNotification);
>>      }
>> end
>>
>>
>>
>>
>>
>>
>>
>> _______________________________________________
>> rules-dev mailing list
>> rules-dev at lists.jboss.org
>> https://lists.jboss.org/mailman/listinfo/rules-dev
>>
>
>
>
> _______________________________________________
> rules-dev mailing list
> rules-dev at lists.jboss.org
> https://lists.jboss.org/mailman/listinfo/rules-dev
>
>



-- 
Michael D Neale
home: www.michaelneale.net
blog: michaelneale.blogspot.com

More information about the rules-dev mailing list