[
https://issues.apache.org/jira/browse/MESOS-308?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=13860852#comment-13860852
]
Vinod Kone commented on MESOS-308:
----------------------------------
None of the 3 listed above have been fixed.
Regarding what 2nd improves, this would make it more granular when gc'ing old
executor directories after a slave restart. That said, with slave recovery this
is much less of an issue because slave directory is typically reused after a
restart.
Regarding 3rd, instead of using fs::usage() to calculate the gc time of an
executor directory the idea is to use "disk space used by slave working
directory / disk space allocated to slave". Not sure if there is a faster way
to calculate the slave directory size than calling 'du' on the directory which
could be really slow.
> Improve Garbage Collection
> --------------------------
>
> Key: MESOS-308
> URL: https://issues.apache.org/jira/browse/MESOS-308
> Project: Mesos
> Issue Type: Improvement
> Reporter: Vinod Kone
> Assignee: Vinod Kone
> Labels: twitter
>
> Improvements:
> --> Evaluate the 'usage' after every directory deletion.
> --> Schedule each of the old slave's executor directories for deletion
> instead of the whole slave directory.
> --> Also, base GC on the amount of disk allocated to slave, not the whole
> disk on the host.
--
This message was sent by Atlassian JIRA
(v6.1.5#6160)
