hbase-issues mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From "Yu Li (JIRA)" <j...@apache.org>
Subject [jira] [Updated] (HBASE-16195) Should not add chunk into chunkQueue if not using chunk pool in HeapMemStoreLAB
Date Tue, 12 Jul 2016 07:09:20 GMT

     [ https://issues.apache.org/jira/browse/HBASE-16195?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel
]

Yu Li updated HBASE-16195:
--------------------------
    Attachment: HBASE-16195_v4.patch

bq. What if we limit and need to add one more entry into Q? It will be problematic as we will
not return back that chunk to pool when that MSLAB is closed right? 
The chunk pool maxsize is limited to a percent of {{globalMemStorePercent}}, and if there're
more elements in the Q than the pool maxsize, there'll be reference of un-reclaimable chunk
left in Q. I think such kind of reference of a useless chunk is dangerous and will potentially
cause memory leak. If the Q is full, I think we should directly abandon new entry rather than
adding into Q since the entry is not reclaimable anyway.

The new patch will set the Q length to pool maxsize and do necessary Q cleanup in {{MemStoreChunkPool#putbackChunks}},
will also cover this in the newly-added UT case. Please let me know whether this looks good
to you [~anoop.hbase], thanks.

> Should not add chunk into chunkQueue if not using chunk pool in HeapMemStoreLAB
> -------------------------------------------------------------------------------
>
>                 Key: HBASE-16195
>                 URL: https://issues.apache.org/jira/browse/HBASE-16195
>             Project: HBase
>          Issue Type: Sub-task
>            Reporter: Yu Li
>            Assignee: Yu Li
>         Attachments: HBASE-16195.patch, HBASE-16195_v2.patch, HBASE-16195_v3.patch, HBASE-16195_v4.patch
>
>
> Problem description and analysis please refer to HBASE-16193



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

Mime
View raw message