[
https://issues.apache.org/jira/browse/HDFS-821?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=12789982#action_12789982
]
Todd Lipcon commented on HDFS-821:
----------------------------------
Two options:
# A time based heuristic as Mike suggested in the comment - 48 hours of "tmp"
should be safe to remove, since as far as I recall, client writes actually go
into the rbw/ directory now.
# I believe replicas in progress actually track the writing thread now, as
well. So, we can see if that thread is still running, and if it's not, remove
it.
> Garbage collect datanode tmp dirs
> ---------------------------------
>
> Key: HDFS-821
> URL: https://issues.apache.org/jira/browse/HDFS-821
> Project: Hadoop HDFS
> Issue Type: Improvement
> Components: data-node
> Affects Versions: 0.22.0
> Reporter: Todd Lipcon
>
> I've seen in practice (and it's been reported on the list) cases where the
> datanode's tmp dir can become quite full with abandoned blocks. There's an
> ancient comment from April 07:
> {code}
> // REMIND - mjc - eventually we should have a timeout system
> // in place to clean up block files left by abandoned clients.
> // We should have some timer in place, so that if a blockfile
> // is created but non-valid, and has been idle for >48 hours,
> // we can GC it safely.
> {code}
> Well, we can consider ourselves reminded, so let's do it!
--
This message is automatically generated by JIRA.
-
You can reply to this email to add a comment to the issue online.
