You are viewing a plain text version of this content. The canonical link for it is here.
Posted to issues@flink.apache.org by "zhijiang (JIRA)" <ji...@apache.org> on 2018/08/16 09:05:00 UTC
[jira] [Commented] (FLINK-10158) The DataOutputSerializer may
consume excessive memory
[ https://issues.apache.org/jira/browse/FLINK-10158?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=16582241#comment-16582241 ]
zhijiang commented on FLINK-10158:
----------------------------------
I already improved it in [FLINK-9913|https://issues.apache.org/jira/browse/FLINK-9913], and there will be only one serializer for all the subpartitions, so the memories of intermediate serialization buffers are decreased a lot.
> The DataOutputSerializer may consume excessive memory
> -----------------------------------------------------
>
> Key: FLINK-10158
> URL: https://issues.apache.org/jira/browse/FLINK-10158
> Project: Flink
> Issue Type: Improvement
> Components: Core
> Affects Versions: 1.4.2, 1.5.2, 1.6.0
> Reporter: aitozi
> Assignee: aitozi
> Priority: Major
>
> I found that the dataOutputSerializer clear the intermediate buffer when the buffer exceed the 5M (as a fixed configuration), But when we encountered the rebalance or keyByPartition and the downstream has a large parallel it will also consume a lot memory, we can do two things :
> 1. make this config configurable
> 2. like the https://issues.apache.org/jira/projects/FLINK/issues/FLINK-1326 mentioned, we can make the serializer one for the output
> What's your idea ? [~StephanEwen]
--
This message was sent by Atlassian JIRA
(v7.6.3#76005)