[
https://issues.apache.org/jira/browse/HBASE-15100?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=15102192#comment-15102192
]
Matteo Bertozzi commented on HBASE-15100:
-----------------------------------------
still looking to see if I found other stuff, but I think the main problem was a
wal deleted too early. (see TestWALProcedureStore#testNoTrailerDoubleRestart).
The patch also resolves one of the pending TODOs where on replay of completed
procedures we avoid the conversion to the Procedure instance going directly to
the ProcedureInfo (used to track the result). This should also provide some
speedup on replay and allow downgrades (when a clean restart/shutdown is done).
there is another TODO about rewriting long waiting procs to the new WAL which
will be the one that will avoid a large amount of WALs kept around even in case
of bugs, but maybe I'll open another jira for that. since at the moment we
don't have procs with a long life span.
> Master WALProcs still never clean up
> ------------------------------------
>
> Key: HBASE-15100
> URL: https://issues.apache.org/jira/browse/HBASE-15100
> Project: HBase
> Issue Type: Bug
> Components: master, proc-v2
> Affects Versions: 1.2.0
> Reporter: Elliott Clark
> Assignee: Matteo Bertozzi
> Priority: Critical
> Attachments: HBASE-15100-v0.patch
>
>
> {code}
> bin/hdfs dfs -ls /hbase/MasterProcWALs | wc -l
> 218631
> {code}
--
This message was sent by Atlassian JIRA
(v6.3.4#6332)
