3:29 p.m.
Hi,
I am using Drools for some time an I recently increased the number of rules
and their complexity.
I have now about 200 rules, for 700 x 200 facts (rules on a cartesian
product on them).
Since then, the computation of rules explodes with the following exception :
java.lang.OutOfMemoryError: unable to create new native thread
at java.lang.Thread.$$YJP$$start0(Native Method) [na:1.6.0_30]
at java.lang.Thread.start0(Thread.java) [na:1.6.0_30]
at java.lang.Thread.start(Thread.java:640) [na:1.6.0_30]
at
java.util.concurrent.ThreadPoolExecutor.addIfUnderMaximumPoolSize(ThreadPoolExecutor.java:727)
~[na:1.6.0_30]
at
java.util.concurrent.ThreadPoolExecutor.execute(ThreadPoolExecutor.java:657)
~[na:1.6.0_30]
at
org.drools.rule.constraint.MvelConstraint.jitEvaluator(MvelConstraint.java:232)
~[org-drools-drools-core-5.5.0.Final.jar:5.5.0.Final]
at
org.drools.rule.constraint.MvelConstraint.evaluate(MvelConstraint.java:197)
~[org-drools-drools-core-5.5.0.Final.jar:5.5.0.Final]
at
org.drools.rule.constraint.MvelConstraint.isAllowed(MvelConstraint.java:157)
~[org-drools-drools-core-5.5.0.Final.jar:5.5.0.Final]
at org.drools.reteoo.AlphaNode.assertObject(AlphaNode.java:137)
~[org-drools-drools-core-5.5.0.Final.jar:5.5.0.Final]
at
org.drools.reteoo.SingleObjectSinkAdapter.propagateAssertObject(SingleObjectSinkAdapter.java:59)
~[org-drools-drools-core-5.5.0.Final.jar:5.5.0.Final]
at org.drools.reteoo.AlphaNode.assertObject(AlphaNode.java:141)
~[org-drools-drools-core-5.5.0.Final.jar:5.5.0.Final]
at
org.drools.reteoo.SingleObjectSinkAdapter.propagateAssertObject(SingleObjectSinkAdapter.java:59)
~[org-drools-drools-core-5.5.0.Final.jar:5.5.0.Final]
at org.drools.reteoo.AlphaNode.assertObject(AlphaNode.java:141)
~[org-drools-drools-core-5.5.0.Final.jar:5.5.0.Final]
at
org.drools.reteoo.SingleObjectSinkAdapter.propagateAssertObject(SingleObjectSinkAdapter.java:59)
~[org-drools-drools-core-5.5.0.Final.jar:5.5.0.Final]
at org.drools.reteoo.AlphaNode.assertObject(AlphaNode.java:141)
~[org-drools-drools-core-5.5.0.Final.jar:5.5.0.Final]
at
org.drools.reteoo.SingleObjectSinkAdapter.propagateAssertObject(SingleObjectSinkAdapter.java:59)
~[org-drools-drools-core-5.5.0.Final.jar:5.5.0.Final]
...
After Googling a bit, I saw that this is due to JIT compilation of my rules.
For some reason, each one seem to create a Thread and ends up with
thousands of them.
First of all, I find a bit risky that the JIT compilation is fired by
some *hardcoded
*threshold :
I was not expecting such a technical tricky optimization to randomly
activate depending of my data.
Anyway, I wanted to disable this JIT thing.
I saw that the disabling is only available since *5.5.0.-final* right ?
I have switched to that version but I can't seem to be able to disable it.
I have tried both :
*-Dmvel2.disable.jit=true* (which would not be a viable solution for
production anyway : We can't rely on some settings of the server to ensure
things work)
And also this :
*OptimizerFactory.setDefaultOptimizer("reflective"); *(found
here<http://mvel.codehaus.org/Disabling+the+JIT+Compiler>
)
Any help would be appreciated.
Brgds,
Raphael
1:10 p.m.
Any help ?
A simple hint to turn JIT off would help.
Thanks.
--
View this message in context:
http://drools.46999.n3.nabble.com/rules-users-Out-of-Mem-and-thousand-thr...
Sent from the Drools: User forum mailing list archive at Nabble.com.
2:11 p.m.
Raphael,
the option to disable jitting is in the class PermGenThresholdOption and the
property name is drools.permgenThreshold.
The idea there was that the main problem that jitting could cause is the
PermGen occupation, so we decided to provide an option to set the percentage
of PermGen space that can be used before it stops jitting. Of course if you
set that value to 0 it will cause to disable jitting completely.
As for the problem you're reporting it sounds a bit weird to me because
we're using a cached thread pool Executor there. Anyway I'll try to
reproduce the problem and check how I could prevent this.
Thank you for having reported it,
Mario
--
View this message in context:
http://drools.46999.n3.nabble.com/rules-users-Out-of-Mem-and-thousand-thr...
Sent from the Drools: User forum mailing list archive at Nabble.com.
2:29 p.m.
Hi,
Thanks for the hint.
I have looked around for a generic solution of OutOfMemory, native thread,
etc :
https://community.jboss.org/thread/164955
It seems actually that Drools uses a thread pool for Jitting and that in my
case the thread pool
thinks it can create an infinity of threads : and it does.
But at one point, the System blocks it (based on ulimit -a settings or
actual memory issues)
It seems to me that Jit compilation requires fine tuning for it to work,
and all those configuration magic numbers (JIT_THRESHOLD
and so on) should not be hidden to the user and properly documented.
In am running Drools in the context of a J2EE environment (JBoss in my
case). And thread pool settings is already a critical
topic. We definitely need more control (more visible ones) on those issues.
I will try to diable JIT in my case in the meantime, but I am waiting
forward
to be able to use it in the future :)
Thanks again,
Brgds,
Raphael
2:43 p.m.
Hi again.
I have looked at PermGenThresholdOption but do not find an exmaple of how
to set those option porgammatically:
KnowledgeBuilder#newKnowledgeBase() does not seem to take any configuration.
Thanks for your help.
Raphael
10:02 a.m.
Ok, I have figured it out :
I created a META-INF/drools.rulebase.conf with the correct setting.
It works fine now.
Thanks
2013/11/12 Raphael Jolivet <raphael.jolivet(a)gmail.com>
Hi again.
I have looked at PermGenThresholdOption but do not find an exmaple of how
to set those option porgammatically:
KnowledgeBuilder#newKnowledgeBase() does not seem to take any
configuration.
Thanks for your help.
Raphael
3825
days inactive
3830
days old
5 comments
3 participants
participants (3)
-
Mario Fusco -
Raphael -
Raphael Jolivet