You are viewing a plain text version of this content. The canonical link for it is here.
Posted to issues@flink.apache.org by "Chesnay Schepler (Jira)" <ji...@apache.org> on 2021/11/12 14:10:00 UTC
[jira] [Updated] (FLINK-24888) Run jobgraph deserialization in
separate thread
[ https://issues.apache.org/jira/browse/FLINK-24888?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel ]
Chesnay Schepler updated FLINK-24888:
-------------------------------------
Description:
The JobSubmitHandler deserializers submitted job graphs. During the deserialization pretty much anything can happen, including access to thread locals and such.
Since the deserializing thread is long-lived it may thus keep a reference to the user-code around for a long time.
We should guard against this by spawning a separate thread to deserialize the jobgraph; this would prevent it since the thread can terminate after serialization.
was:
The JobSubmitHandler deserializers submitted job graphs. During the deserialization pretty much anything can happen, including access to thread locals and such.
Since the deserializing thread is long-lived it may thus keep a reference to the user-code around for a long time.
We should guard against this by spawning a separate to deserialize the jobgraph; this would prevent it since the thread can terminate after serialization.
> Run jobgraph deserialization in separate thread
> -----------------------------------------------
>
> Key: FLINK-24888
> URL: https://issues.apache.org/jira/browse/FLINK-24888
> Project: Flink
> Issue Type: Improvement
> Components: Runtime / REST
> Reporter: Chesnay Schepler
> Assignee: Chesnay Schepler
> Priority: Major
> Fix For: 1.15.0
>
>
> The JobSubmitHandler deserializers submitted job graphs. During the deserialization pretty much anything can happen, including access to thread locals and such.
> Since the deserializing thread is long-lived it may thus keep a reference to the user-code around for a long time.
> We should guard against this by spawning a separate thread to deserialize the jobgraph; this would prevent it since the thread can terminate after serialization.
--
This message was sent by Atlassian Jira
(v8.20.1#820001)