You are viewing a plain text version of this content. The canonical link for it is here.
Posted to hdfs-issues@hadoop.apache.org by "ASF GitHub Bot (Jira)" <ji...@apache.org> on 2025/02/17 05:20:00 UTC
[jira] [Commented] (HDFS-17670) HDFS Client's direct memory leaks with erasure coding enabled
[ https://issues.apache.org/jira/browse/HDFS-17670?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=17927608#comment-17927608 ]
ASF GitHub Bot commented on HDFS-17670:
---------------------------------------
EungsopYoo commented on PR #7168:
URL: https://github.com/apache/hadoop/pull/7168#issuecomment-2662083978
https://issues.apache.org/jira/browse/HDFS-17670
I have updated the results after applying this PR.
> HDFS Client's direct memory leaks with erasure coding enabled
> -------------------------------------------------------------
>
> Key: HDFS-17670
> URL: https://issues.apache.org/jira/browse/HDFS-17670
> Project: Hadoop HDFS
> Issue Type: Bug
> Components: erasure-coding
> Reporter: Eungsop Yoo
> Priority: Major
> Labels: pull-request-available
> Attachments: image-2024-11-18-13-53-23-690.png, image-2025-02-17-13-59-03-004.png
>
>
> We are testing erasure coding with HBase. But we faced that the size of direct memory of HBase RegionServer keep growing.
> !image-2024-11-18-13-53-23-690.png!
> It is caused by that the direct memory allocated by ElasticByteBufferPool of HDFS client is never released.
> https://issues.apache.org/jira/browse/HADOOP-18105
> We found that we already have the solution, WeakReferencedElasticByteBufferPool.
> Just replace ElasticByteBufferPool with WeakReferencedElasticByteBufferPool.
>
>
> PS. I applied this patch to my production HBase cluster using EC. It worked but the result was somewhat different from my expectations.
>
> [https://github.com/apache/hadoop/blob/trunk/hadoop-common-project/hadoop-common/src/main/java/org/apache/hadoop/io/erasurecode/rawcoder/NativeRSRawEncoder.java#L65-L68]
> Each store file allocates 6MB direct memory by NativeRSRawEncoder class, so if there are 1000 store files in a single RS, the RS occupies about 6GB of direct memory. We have to be careful to configure sufficient RAM to serve store files for EC enabled HBase clusters.
> !image-2025-02-17-13-59-03-004.png!
--
This message was sent by Atlassian Jira
(v8.20.10#820010)
---------------------------------------------------------------------
To unsubscribe, e-mail: hdfs-issues-unsubscribe@hadoop.apache.org
For additional commands, e-mail: hdfs-issues-help@hadoop.apache.org