[Gluster-devel] Memory-leak in geo-replication?

[Mikhail T.]

Hello!

We are running geo-replication between data-centers and were just
alerted by the monitoring system about swap shortage on one of the
boxes. The investigation showed the glusterfs-process responsible for
geo-replication consuming 5.7Gb of memory... There was nothing obvious
in the logs. The replication log, in particular had the routine message:

    [2014-05-18 09:27:37.228969] I [master:669:crawl] _GMaster: completed 57 crawls, 0 turns

repeated over and over once per minute. The last non-informational
message logged was dated April 20 -- a month ago. I restarted the
geo-replication and the new process is using under 1Gb.

A similar picture is on the other geo-replicating system -- although its
memory usage has not triggered a monitor yet, the glusterfs process
responsible for geo-replication, which started on February 1st, is
currently consuming well over 4Gb of memory.

We are using version 3.4.1 of Gluster-built RPMs on RHEL6 -- is this,
perhaps, a known problem? Should we upgrade to have it fixed? Thank you!

    -mi

-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://supercolony.gluster.org/pipermail/gluster-devel/attachments/20140519/7435d998/attachment-0002.html>