You are viewing a plain text version of this content. The canonical link for it is here.
Posted to issues@spark.apache.org by "zhengchenyu (Jira)" <ji...@apache.org> on 2022/11/11 09:37:00 UTC
[jira] [Commented] (SPARK-41073) Spark ThriftServer generate huge amounts of DelegationToken
[ https://issues.apache.org/jira/browse/SPARK-41073?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=17632209#comment-17632209 ]
zhengchenyu commented on SPARK-41073:
-------------------------------------
We must add a valid credentials to jobConf.
For now, I think sql in thriftserver or local mode can't get a valid credentials.
I have two proposal:
proposal A: set a global credentials for hadoop.
proposal B: extract HadoopDelegationTokenManager from CoarseGrainedSchedulerBackend. (Note:I think local spark also wanna global credentials)
I prefer B.
But A is simple, I have submit SPARK-41073.proposal.A.draft.001.patch, I solve the problem, but not graceful.
[~vanzin] [~xkrogen] Can you give me some suggesstion?
> Spark ThriftServer generate huge amounts of DelegationToken
> -----------------------------------------------------------
>
> Key: SPARK-41073
> URL: https://issues.apache.org/jira/browse/SPARK-41073
> Project: Spark
> Issue Type: Bug
> Components: SQL
> Affects Versions: 3.0.1
> Reporter: zhengchenyu
> Priority: Major
> Attachments: SPARK-41073.proposal.A.draft.001.patch
>
>
> In our cluster, zookeeper nearly crashed. I found the znodes of /zkdtsm/ZKDTSMRoot/ZKDTSMTokensRoot increased quickly.
> After some research, I found some sql running on spark-thriftserver obtain huge amounts of DelegationToken.
> The reason is that in these spark-sql, every hive parition acquire a different delegation token.
> And HadoopRDDs in thriftserver can't share credentials from CoarseGrainedSchedulerBackend::delegationTokens, we must share it.
--
This message was sent by Atlassian Jira
(v8.20.10#820010)
---------------------------------------------------------------------
To unsubscribe, e-mail: issues-unsubscribe@spark.apache.org
For additional commands, e-mail: issues-help@spark.apache.org