[jira] [Commented] (HIVE-15068) Run ClearDanglingScratchDir periodically inside HS2

["royal (Jira)" <ji...@apache.org>]

    [ https://issues.apache.org/jira/browse/HIVE-15068?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=17534321#comment-17534321 ] 

royal commented on HIVE-15068:
------------------------------

[~daijy]  [~leftyl]  [~hiveqa]  [~thejas]  [~h_o] can you help me solve this problem, Thank You

https://stackoverflow.com/questions/72186228/hive-fetch-result-from-hdfs-is-too-slow-because-of-too-many-the-map-only-task

> Run ClearDanglingScratchDir periodically inside HS2
> ---------------------------------------------------
>
>                 Key: HIVE-15068
>                 URL: https://issues.apache.org/jira/browse/HIVE-15068
>             Project: Hive
>          Issue Type: Improvement
>          Components: HiveServer2
>            Reporter: Daniel Dai
>            Assignee: Daniel Dai
>            Priority: Major
>              Labels: TODOC1.3, TODOC2.2
>             Fix For: 1.3.0, 2.2.0
>
>         Attachments: HIVE-15068.1.patch, HIVE-15068.2.patch, HIVE-15068.3.patch, HIVE-15068.4.branch-1.patch, HIVE-15068.4.patch
>
>
> In HIVE-13429, we introduce a tool which clear dangling scratch directory. In this ticket, we want to invoke the tool automatically on a Hive cluster. Options are:
> 1. cron job, which would involve manual cron job setup
> 2. As a metastore thread. However, it is possible we run metastore without hdfs in the future (eg, managing s3 files). ClearDanglingScratchDir needs support which only exists in hdfs, it won't work if the above scenario happens
> 3. As a HS2 thread. The downside is if no HS2 is running, the tool will not run automatically. But we expect HS2 will be a required component down the road
> Here I choose approach 3 in the implementation.



--
This message was sent by Atlassian Jira
(v8.20.7#820007)