[jira] [Commented] (MAPREDUCE-6059) Speed up history server startup time

["Hudson (JIRA)" <ji...@apache.org>]

    [ https://issues.apache.org/jira/browse/MAPREDUCE-6059?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=14306424#comment-14306424 ] 

Hudson commented on MAPREDUCE-6059:
-----------------------------------

SUCCESS: Integrated in Hadoop-trunk-Commit #7013 (See [https://builds.apache.org/job/Hadoop-trunk-Commit/7013/])
MAPREDUCE-6059. Speed up history server startup time (Siqi Li via aw) (aw: rev fd57ab2002f97dcc83d455a5e0c770c8efde77a4)
* hadoop-mapreduce-project/CHANGES.txt
* hadoop-mapreduce-project/hadoop-mapreduce-client/hadoop-mapreduce-client-hs/src/main/java/org/apache/hadoop/mapreduce/v2/hs/HistoryFileManager.java


> Speed up history server startup time
> ------------------------------------
>
>                 Key: MAPREDUCE-6059
>                 URL: https://issues.apache.org/jira/browse/MAPREDUCE-6059
>             Project: Hadoop Map/Reduce
>          Issue Type: Improvement
>    Affects Versions: 2.4.0
>            Reporter: Siqi Li
>            Assignee: Siqi Li
>             Fix For: 3.0.0
>
>         Attachments: YARN-2366.v1.patch
>
>
> When history server starts up, It scans every history directories and put all history files into a cache, whereas this cache only stores 20K recent history files. Therefore, it is wasting a large portion of time loading old history files into the cache, and the startup time will keep increasing if we don't trim the number of history files. For example, when history server starts up with 2.5M history files in HDFS, it took ~5 minutes.



--
This message was sent by Atlassian JIRA
(v6.3.4#6332)