You are viewing a plain text version of this content. The canonical link for it is here.
Posted to issues@spark.apache.org by "Steve Loughran (JIRA)" <ji...@apache.org> on 2019/02/14 21:52:00 UTC

[jira] [Commented] (SPARK-25766) AMCredentialRenewer can leak FS clients

    [ https://issues.apache.org/jira/browse/SPARK-25766?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=16768747#comment-16768747 ] 

Steve Loughran commented on SPARK-25766:
----------------------------------------

sounds good, though I should check. There's a FileSystem.closeAllForUGI() method which purges all cached entries for a user and closes them; can be used.

That said, staring at that code its doing cleanup in a sync block, so it has its own issues, especially if someone has been creating files for delete-on-exit. Created HADOOP-16111 to remind me.

> AMCredentialRenewer can leak FS clients
> ---------------------------------------
>
>                 Key: SPARK-25766
>                 URL: https://issues.apache.org/jira/browse/SPARK-25766
>             Project: Spark
>          Issue Type: Bug
>          Components: YARN
>    Affects Versions: 2.4.0
>            Reporter: Steve Loughran
>            Priority: Trivial
>
> AMCredentialRenewer's scheduled {{writeNewCredentialsToHDFS}} operation creates a new FS connector each time, so as to access the store with refreshed credentials
> but it doesn't close it after, so any resources used by the client are kept around. This is more expensive with the cloud store connectors which create thread pools. 
> It should call {{remoteFs.close()}} at the end of its work.



--
This message was sent by Atlassian JIRA
(v7.6.3#76005)

---------------------------------------------------------------------
To unsubscribe, e-mail: issues-unsubscribe@spark.apache.org
For additional commands, e-mail: issues-help@spark.apache.org