[
https://issues.apache.org/jira/browse/SPARK-33790?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=17265691#comment-17265691
]
dzcxzl commented on SPARK-33790:
--------------------------------
This is indeed a performance regression problem.
The following is my case 2.x version EventLoggingListener.codecMap is of type
mutable.HashMap, which is not thread-safe and may hang.
3.x version changed to EventLogFileReader.codecMap changed to ConcurrentHashMap
type.
In the 2.x version, the history server may not work.
I tried to use the 3.x version, and found that a round of scan has slowed down
a lot, 7min rose to about 23min.
In addition, do I need to fix the thread safety issues in version 2.x?
[~kabhwan]
> Reduce the rpc call of getFileStatus in SingleFileEventLogFileReader
> --------------------------------------------------------------------
>
> Key: SPARK-33790
> URL: https://issues.apache.org/jira/browse/SPARK-33790
> Project: Spark
> Issue Type: Improvement
> Components: Spark Core
> Affects Versions: 3.0.1
> Reporter: dzcxzl
> Assignee: dzcxzl
> Priority: Critical
> Fix For: 3.2.0
>
>
> FsHistoryProvider#checkForLogs already has FileStatus when constructing
> SingleFileEventLogFileReader, and there is no need to get the FileStatus
> again when SingleFileEventLogFileReader#fileSizeForLastIndex.
> This can reduce a lot of rpc calls and improve the speed of the history
> server.
--
This message was sent by Atlassian Jira
(v8.3.4#803005)
---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]
