When a fact that is the right input of a join node is added and then immediately removed before the join node is evaluated it gets added to the delete staged set of the right memory of that node. If the join node never gets evaluated the right tuple will remain forever in that staged set. Iterating this process with a big numbers of these facts can ultimately lead to a OutOfMemory. The following test case reproduces the problem. It doesn't make any assertion so it doesn't fail. Anyway putting a breakpoint in the RightTupleSetsImpl.addDelete method it is possible to see that the set has 100 RightTuple in its staged delete set. @Test public void testStagedTupleLeak() throws Exception { String str = "rule R1 when\n" + " $i : Integer()\n" + "then\n" + " insertLogical( $i.toString() );\n" + "end\n" + "\n" + "rule R2 when\n" + " $i : Integer()\n" + "then\n" + " delete( $i );\n" + "end\n" + "\n" + "rule R3 when\n" + " $l : Long()\n" + " $s : String( this == $l.toString() )\n" + "then\n" + "end\n"; KnowledgeBase kbase = loadKnowledgeBaseFromString(str); StatefulKnowledgeSession ksession = kbase.newStatefulKnowledgeSession(); for (int i = 0; i < 100; i++) { ksession.insert(i); ksession.fireAllRules(); } }
This was found today and being a memory leak, we would like to include the fix in 6.0.0.CR2.
Fixed by https://github.com/droolsjbpm/drools/commit/453f0f68b
This fix has been cherry-picked to both 6.0.x and 6.0.0.GA.x branches.
Mario Fusco <mario.fusco> updated the status of jira DROOLS-411 to Resolved
The unit test to reproduce the issue has failed with CR2. Reopening. https://jenkins.mw.lab.eng.bos.redhat.com/hudson/view/BRMS/view/BRMS-6/view/BRMS-6-all/job/kie-unit-single/15/testReport/org.drools.compiler.integrationtests/Misc2Test/testStagedTupleLeak/
(In reply to Lukáš Petrovický from comment #6) > The unit test to reproduce the issue has failed with CR2. Reopening. The unit test DID NOT fail, mea culpa. This bug is indeed VERIFIED.