[
https://issues.apache.org/jira/browse/FLINK-1332?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=14251627#comment-14251627
]
Stephan Ewen commented on FLINK-1332:
-------------------------------------
Can you share the program, or the program plan? Does it involve iterations or
DAGs?
Broadcast variables can OOM the TaskManager, if they become too large. They are
essentially user-code variables. OOM errors are a bit non-deterministic in
Java, and depend on heap fragmentation, ...
We are pushing the new network code into 0.9-SNAPSHOT soon, which should solve
the deadlocks in principled way completely...
> Possible deadlock with two broadcast sets
> -----------------------------------------
>
> Key: FLINK-1332
> URL: https://issues.apache.org/jira/browse/FLINK-1332
> Project: Flink
> Issue Type: Bug
> Components: Distributed Runtime
> Affects Versions: 0.7.0-incubating
> Reporter: Sebastian Schelter
>
> We are writing some code for a student homework that reads two broadcast sets
> in the open method of a RichMapFunction.
> We are seeing inconsistent behavior, sometimes the code works, sometimes it
> deadlocks, sometimes it crashes with an OutOfMemoryError...
--
This message was sent by Atlassian JIRA
(v6.3.4#6332)
