[
https://issues.apache.org/jira/browse/FLINK-28691?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=17695973#comment-17695973
]
Shammon commented on FLINK-28691:
---------------------------------
Hi [~jark] We found the metaspace fullgc problem in codegen and [~FrankZou]
fixed it our internal branch. This may involve codegen and planner, what do you
think of it? I found that other users also reported the same problem such as
https://issues.apache.org/jira/browse/FLINK-31308
> Improve cache hit rate of generated class
> -----------------------------------------
>
> Key: FLINK-28691
> URL: https://issues.apache.org/jira/browse/FLINK-28691
> Project: Flink
> Issue Type: Sub-task
> Components: Table SQL / Runtime
> Reporter: Zou
> Priority: Major
>
> In OLAP scenarios, compiling generated classes is very frequent, it will
> consume a lot of CPU and large amount of generated classes will also takes up
> a lot of space in metaspace, which will lead to frequent Full GC.
> As we use a self-incrementing counter in CodeGenUtils#newName, it means we
> could not get the same generated class between two queries even when they are
> exactly the same. Maybe we could share the same generated class between
> different queries if they has the same logic, it will be good for job latency
> and resource consumption.
--
This message was sent by Atlassian Jira
(v8.20.10#820010)
