You are viewing a plain text version of this content. The canonical link for it is here.

Posted to commits@hudi.apache.org by "Zhaojing Yu (Jira)" <ji...@apache.org> on 2022/10/01 12:09:00 UTC

[jira] [Updated] (HUDI-4792) Speed up cleaning with metadata table enabled

     [ https://issues.apache.org/jira/browse/HUDI-4792?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel ]

Zhaojing Yu updated HUDI-4792:
------------------------------
    Fix Version/s: 0.13.0
                       (was: 0.12.1)

> Speed up cleaning with metadata table enabled 
> ----------------------------------------------
>
>                 Key: HUDI-4792
>                 URL: https://issues.apache.org/jira/browse/HUDI-4792
>             Project: Apache Hudi
>          Issue Type: Improvement
>          Components: cleaning
>            Reporter: nicolas paris
>            Assignee: nicolas paris
>            Priority: Critical
>              Labels: pull-request-available
>             Fix For: 0.13.0
>
>
> Currently fetching file group to be deleted is parallelized over each partition. As a result, in case of many partition, many calls are made on the metadata. While this is ok for file system view, this is highly inefficient with the metadata table view (MDT){*}.{*} Likely each call makes the MoR happens on the MDT and in the case of thousand of partitions the process is incredibly slow. 
> I benchmarked (non incremental) cleaning on the same table w/ and w/o MDT on a 40k partitionned hudi table :
>  * w/ MDT: 5 hours
>  * w/o MDT: 5 minutes
> This slowness makes the use of MDT not reasonable in the case of many partitions, because cleaning is a  must-have.



--
This message was sent by Atlassian Jira
(v8.20.10#820010)