7:55 p.m.
Hi,
I'm attempting to track down an out of memory error.
I have ~400 rules that operate on ~10000 facts. One set of facts throws an
out of memory error while other sets process w/o an error. It is easy to
reproduce.
When the system runs out of memory there are about 22,000,000 ReteTuple
objects active.
I assume I have a bad rule producing a large cross product.
How do I find it?
Is there an audit tool that will let me tie ReteTuple back to a specific
rule?
Regards,
Jared Davis
--
View this message in context:
http://drools-java-rules-engine.46999.n3.nabble.com/Memory-usage-tied-to-...
Sent from the Drools - User mailing list archive at Nabble.com.
2:39 a.m.
Jared,
Which version of Drools are you using? Drools 5.1 massively
reduces memory consumption for long-running stateful sessions, when
there are lots of update/modify calls, when compared to Drools 5.0.
Also, in 5.1 there is an internal report you can run. I did not
exposed this in drools-api yet, because I am still not happy with the
overall API... so, maybe you can try using it and provide feedback on
how you would improve it and we may eventually publish it on the next
version.
Anyway, to use it, after you have your session loaded with some data:
SessionInspector inspector = new SessionInspector( ksession );
StatefulKnowledgeSessionInfo info = inspector.getSessionInfo();
SessionReporter.addNamedTemplate( "topten",
getClass().getResourceAsStream( "customreports.mvel" ) );
String report = SessionReporter.generateReport( "topten", info, null );
The file referenced above (customreports.mvel) can be found here:
http://anonsvn.jboss.org/repos/labs/labs/jbossrules/trunk/drools-compiler...
Just add it to your classpath. Feel free to change it as you see
fit, but as is it will give you the top ten left and right node
memories.
Edson
2010/9/7 Jared Davis <sunray(a)davisprogramming.com>:
Hi,
I'm attempting to track down an out of memory error.
I have ~400 rules that operate on ~10000 facts. One set of facts throws an
out of memory error while other sets process w/o an error. It is easy to
reproduce.
When the system runs out of memory there are about 22,000,000 ReteTuple
objects active.
I assume I have a bad rule producing a large cross product.
How do I find it?
Is there an audit tool that will let me tie ReteTuple back to a specific
rule?
Regards,
Jared Davis
--
View this message in context:
http://drools-java-rules-engine.46999.n3.nabble.com/Memory-usage-tied-to-...
Sent from the Drools - User mailing list archive at Nabble.com.
_______________________________________________
rules-users mailing list
rules-users(a)lists.jboss.org
https://lists.jboss.org/mailman/listinfo/rules-users
--
Edson Tirelli
JBoss Drools Core Development
JBoss by Red Hat @ www.jboss.com
9:24 a.m.
Edson,
Thanks for the report. The top ten on the left was just what I needed to
track down the badly written rules.
The 385 rules generate 823 documents using 20000 facts.
My peak heap usage is now only 250mb. Before it would fail with a 1.2GB
heap.
Do you think this report is a good starting point to generate a graphML file
showing the same info as the text report?
Jared
--
View this message in context:
http://drools-java-rules-engine.46999.n3.nabble.com/Memory-usage-tied-to-...
Sent from the Drools - User mailing list archive at Nabble.com.
2:29 p.m.
Jared,
Glad to hear.
Yes, the idea of the API is to extract info from the working
memory and RETE network and expose it to report generation. As you
saw, I am using an MVEL template to generate the report at the moment.
It is very crude, so any suggestions for improvements are most
welcome. Once we get something better, we can expose it as an official
API in Drools API.
Edson
2010/10/5 Jared Davis <sunray(a)davisprogramming.com>:
Edson,
Thanks for the report. The top ten on the left was just what I needed to
track down the badly written rules.
The 385 rules generate 823 documents using 20000 facts.
My peak heap usage is now only 250mb. Before it would fail with a 1.2GB
heap.
Do you think this report is a good starting point to generate a graphML file
showing the same info as the text report?
Jared
--
View this message in context:
http://drools-java-rules-engine.46999.n3.nabble.com/Memory-usage-tied-to-...
Sent from the Drools - User mailing list archive at Nabble.com.
_______________________________________________
rules-users mailing list
rules-users(a)lists.jboss.org
https://lists.jboss.org/mailman/listinfo/rules-users
--
Edson Tirelli
JBoss Drools Core Development
JBoss by Red Hat @ www.jboss.com
6:33 p.m.
Edson,
I'm using yEd for the graph viewer and tgf as the format for the graph. Here
is the mvel template to produce the tgf file. I left the topten name the
same for now. Did I miss any nodes or edges?
@declare{"topten"}
@code{
buffer = new StringBuilder();
for (ni : session.nodeInfos) {
buffer.append( ni.id);
buffer.append( " ");
buffer.append( ni.node);
buffer.append( "\r\n");
}
buffer.append( "#\r\n");
for (node : session.nodeInfos) {
for (child : node.sinkList) {
buffer.append( node.id);
buffer.append( " ");
buffer.append( child.id);
buffer.append( "\r\n");
}
}
}
@{buffer.toString()}
@end{}
I would like to change the toString representations for all of the classes
used in the graph.
I there a better structure than a bunch of instanceof tests? (Would they
even work?)
if (ni.node instanceof AlphaNode) {
buffer.append("A: ");
if (ni.node.constraint instanceof ...) {
Thanks again for your help.
--
View this message in context:
http://drools-java-rules-engine.46999.n3.nabble.com/Memory-usage-tied-to-...
Sent from the Drools - User mailing list archive at Nabble.com.
6:48 p.m.
Hi Jared,
This was really a first attempt at getting something useful to look
inside the chest of a session and extract useful info. Feel free to
look at the classes and propose changes/provide patches. Once it
reaches a state we consider good enough, we can publish it in Drools
API.
Regarding use of "instanceof", I always prefer to use polymorphism
instead. So, define a method to do what you need in a super
class/interface and we implement it in all subclasses. Besides making
the code cleaner, this also makes sure that new nodes/classes created
in the future will not break or be absent in the report.
Edson
2010/10/7 Jared Davis <sunray(a)davisprogramming.com>:
Edson,
I'm using yEd for the graph viewer and tgf as the format for the graph. Here
is the mvel template to produce the tgf file. I left the topten name the
same for now. Did I miss any nodes or edges?
@declare{"topten"}
@code{
buffer = new StringBuilder();
for (ni : session.nodeInfos) {
buffer.append( ni.id);
buffer.append( " ");
buffer.append( ni.node);
buffer.append( "\r\n");
}
buffer.append( "#\r\n");
for (node : session.nodeInfos) {
for (child : node.sinkList) {
buffer.append( node.id);
buffer.append( " ");
buffer.append( child.id);
buffer.append( "\r\n");
}
}
}
@{buffer.toString()}
@end{}
I would like to change the toString representations for all of the classes
used in the graph.
I there a better structure than a bunch of instanceof tests? (Would they
even work?)
if (ni.node instanceof AlphaNode) {
buffer.append("A: ");
if (ni.node.constraint instanceof ...) {
Thanks again for your help.
--
View this message in context:
http://drools-java-rules-engine.46999.n3.nabble.com/Memory-usage-tied-to-...
Sent from the Drools - User mailing list archive at Nabble.com.
_______________________________________________
rules-users mailing list
rules-users(a)lists.jboss.org
https://lists.jboss.org/mailman/listinfo/rules-users
--
Edson Tirelli
JBoss Drools Core Development
JBoss by Red Hat @ www.jboss.com
4949
days inactive
4979
days old
5 comments
2 participants
participants (2)
-
Edson Tirelli -
Jared Davis