[
https://issues.apache.org/jira/browse/FLINK-1326?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=14246587#comment-14246587
]
Stephan Ewen commented on FLINK-1326:
-------------------------------------
I have implemented the quickfix variant, for now.
> The output serializers may consume excessive memory
> ---------------------------------------------------
>
> Key: FLINK-1326
> URL: https://issues.apache.org/jira/browse/FLINK-1326
> Project: Flink
> Issue Type: Bug
> Components: Distributed Runtime
> Affects Versions: 0.8-incubating
> Reporter: Stephan Ewen
> Assignee: Stephan Ewen
>
> The SpanningRecordSerializer creates large byte arrays to buffer records
> during serialization. Since they never release the records, they may occupy a
> large amount of memory. Currently, we use one per output partition, which is
> due to the original architecture.
> There are two ways to fix this:
> 1. Use only one serializer per output, rather than one per output partition.
> This is the preferable way in the long run anyways.
> 2. Release the arrays in the serializer (quick fix).
--
This message was sent by Atlassian JIRA
(v6.3.4#6332)
