[
https://issues.apache.org/jira/browse/FLINK-5107?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=15683391#comment-15683391
]
ASF GitHub Bot commented on FLINK-5107:
---------------------------------------
Github user zentol commented on a diff in the pull request:
https://github.com/apache/flink/pull/2837#discussion_r88879433
--- Diff:
flink-runtime/src/main/java/org/apache/flink/runtime/executiongraph/ExecutionVertex.java
---
@@ -125,7 +142,7 @@ public ExecutionVertex(
this.inputEdges = new
ExecutionEdge[jobVertex.getJobVertex().getInputs().size()][];
- this.priorExecutions = new CopyOnWriteArrayList<Execution>();
+ this.priorExecutions = new
EvictingBoundedList<>(maxPriorExecutionHistoryLength);
--- End diff --
Similarly, the ArchivedExecutionVertex is now probably broken if an
execution was discarded since the size checks in `getPriorExecutionAttempt` are
not adjusted.
> Job Manager goes out of memory from long history of prior execution attempts
> ----------------------------------------------------------------------------
>
> Key: FLINK-5107
> URL: https://issues.apache.org/jira/browse/FLINK-5107
> Project: Flink
> Issue Type: Bug
> Components: JobManager
> Reporter: Stefan Richter
> Assignee: Stefan Richter
>
> We have observed that the job manager can run out of memory during long
> running jobs with many vertexes. Analysis of the heap dump shows, that the
> ever-growing history of prior execution attempts is the culprit for this
> problem.
> We should limit this history to a number of n most recent attempts.
--
This message was sent by Atlassian JIRA
(v6.3.4#6332)
