[Gluster-users] Impact on GlusterD with multi threaded e-poll support - Workaround for GlusterD crash

Atin Mukherjee amukherj at redhat.com
Sun Oct 11 06:08:34 UTC 2015


Folks,

We've been noticing few GlusterD crashes in 3,7.x releases due to
enabling of multi threaded e-poll support in GlusterD.

As a work around you could add an entry "option event-threads 1" in
/etc/glusterfs/glusterd.vol and restart GlusterD service. We plan to
have this configuration change in 3.7.6 and we will revert this once we
get to the RCA of these crashes.

Note : Setting 'event-threads' to 1 for glusterd's epoll subsystem
doesn't work well with ping-timer being enabled especially for
volume-snapshot commands. In this case if you happen to see volume
snapshot commands failing, you would need to update the ping-timeout
value to 0.

Thanks,
Atin


More information about the Gluster-users mailing list