[jboss-jira] [JBoss JIRA] (DROOLS-4805) Memory leak in STREAM mode (# PointInTimeTriggers growing)

[Jesse White (Jira)]

    [ https://issues.jboss.org/browse/DROOLS-4805?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=13818318#comment-13818318 ] 

Jesse White commented on DROOLS-4805:
-------------------------------------

I think the problem may be related to the fact that our events are not strictly ordered sequentially in time.

I have a test case that reproduces the problem here: https://gist.github.com/j-white/7adee3e5c628f012ea83cd69695b93f9

(if the value of delta == 0, then the events are effectively ordered, otherwise they are random)

With out-of-order events {{org.drools.core.time.impl.TrackableTimeJobFactoryManager#addTimerJobInstance}} gets called many times and {{#removeTimerJobInstance}} is never called, leading to growth in the {{timerInstances}} map.

With ordered events, {{#addTimerJobInstance}} is only called once.

> Memory leak in STREAM mode (# PointInTimeTriggers growing)
> ----------------------------------------------------------
>
>                 Key: DROOLS-4805
>                 URL: https://issues.jboss.org/browse/DROOLS-4805
>             Project: Drools
>          Issue Type: Bug
>    Affects Versions: 7.29.0.Final
>            Reporter: Jesse White
>            Assignee: Mario Fusco
>            Priority: Major
>         Attachments: image-2019-11-25-10-31-07-281.png
>
>
> When running our engine in STREAM mode, we notice a number of Drools internal object references increasing time even though the fact count stays relatively constant.
> jmap -histo <pid> shows:
> {noformat}
>  num     #instances         #bytes  class name
> ----------------------------------------------
>    1:      17478466     1258449552  java.util.concurrent.ScheduledThreadPoolExecutor$ScheduledFutureTask
>    2:      17475174      699006960  org.drools.core.time.impl.JDKTimerService$JDKJobHandle
>    3:      17762331      568394592  java.util.concurrent.ConcurrentHashMap$Node
>    4:      17475174      559205568  org.drools.core.rule.SlidingTimeWindow$BehaviorJobContext
>    5:      17475174      559205568  org.drools.core.time.impl.DefaultTimerJobInstance
>    6:       8596776      551411680  [C
>    7:      18898675      453568200  java.lang.Long
>    8:      17475174      419404176  org.drools.core.time.SelfRemovalJobContext
>    9:      17475174      279602784  org.drools.core.time.SelfRemovalJob
>   10:      17475174      279602784  org.drools.core.time.impl.PointInTimeTrigger
>   11:       9263698      222328752  java.lang.String
>   12:       4584671      146709472  java.util.HashMap$Node
>   13:        246658      138804176  [B
> {noformat}
> These number continue to increase over time until the system OOMs.
> Graph for # of facts in the session looks like:
>  !image-2019-11-25-10-31-07-281.png|thumbnail! 
> The engine is fairly busy - we're constantly adding and removing facts.
> Our rules are available here:
> https://github.com/OpenNMS/opennms/blob/opennms-25.0.0-1/opennms-base-assembly/src/main/filtered/etc/alarmd/drools-rules.d/alarmd.drl#L234



--
This message was sent by Atlassian Jira
(v7.13.8#713008)