accumulo-notifications mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From "Adam J Shook (JIRA)" <j...@apache.org>
Subject [jira] [Commented] (ACCUMULO-4428) GC does not delete WAL files belonging to dead tservers
Date Mon, 29 Aug 2016 16:12:22 GMT

    [ https://issues.apache.org/jira/browse/ACCUMULO-4428?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=15446304#comment-15446304
] 

Adam J Shook commented on ACCUMULO-4428:
----------------------------------------

Yeah, we are seeing this in action.  I enabled the trace logs and every period it re-discovers
the dead tservers.  Not sure how to write any test cases to cover it, but we're hoping to
get this into 1.7.3 and will manually clean up the WALs in the meantime.  Happy to work on
this and get you a PR -- I need to patch the GC anyway and run it manually to clean up the
files.  I was thinking of just making the firstSeenDead map static.  I don't see a change
like that having any adverse side effects.

> GC does not delete WAL files belonging to dead tservers
> -------------------------------------------------------
>
>                 Key: ACCUMULO-4428
>                 URL: https://issues.apache.org/jira/browse/ACCUMULO-4428
>             Project: Accumulo
>          Issue Type: Bug
>    Affects Versions: 1.7.2
>            Reporter: Adam J Shook
>            Assignee: Michael Wall
>            Priority: Blocker
>             Fix For: 1.7.3, 1.8.0
>
>
> The GarbageCollectWriteAheadLogs uses a Map<HostAndPort,Long> to track when it
had first seen a dead tserver, waiting an hour before deleting the files.  However, a new
instance of this class is re-created during each run of the SimpleGarbageCollector, causing
the state of the dead tservers to be lost.  All of the WAL files belonging to a dead tserver
will never be removed.



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

Mime
View raw message