[
https://issues.apache.org/jira/browse/HBASE-16524?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=15456914#comment-15456914
]
Appy commented on HBASE-16524:
------------------------------
Attaching flame graph form a run of {{./bin/hbase
org.apache.hadoop.hbase.procedure2.store.wal.ProcedureWALPerformanceEvaluation
-procs 500000}} (captured 15sec).
As expected, bottleneck is subtraction of two bitmaps.
The lowest hanging fruit here is to not call removeInactiveLogs() on every sync.
Also, instead of waiting for a predetermined (and randomly) chosen time, it'll
be better to call the cleaner once every wal roll. Only the written wals (not
the active one) are used to decided the obsolete ones, therefore, calling
removeInactiveLogs() repeatedly until the active log is rolled won't mark any
new logs for removal.
> Clean procedure wal periodically instead of on every sync
> ---------------------------------------------------------
>
> Key: HBASE-16524
> URL: https://issues.apache.org/jira/browse/HBASE-16524
> Project: HBase
> Issue Type: Bug
> Reporter: Appy
> Assignee: Appy
> Priority: Minor
> Attachments: HBASE-16524.master.001.patch
>
>
--
This message was sent by Atlassian JIRA
(v6.3.4#6332)
