Hi All,
We're using "gluster volume heal ${volname} statistics heal-count"
to
monitor our systems w.r.t. healing not happening.
The reason we're using statistics heal-count and not info is because
it's extremely fast in comparison with info.
After upgrading to glusterfs 6.1 (from 4.1) we noticed that in many
cases heal-count would report >0 values, and then upon running info,
this just goes away.
Upon closer investigation I've noticed that indices/xattrop there are a
few gfid linked files which correlates with the counts given by
heal-count, for example:
# gluster volume heal mail statistics heal-count
Gathering count of entries to be healed on volume mail has been successful
Brick host_a:/mnt/gluster/mail
Number of entries: 0
Brick host_b:/mnt/gluster/mail
Number of entries: 3
And then:
host_b /mnt/gluster/mail/.glusterfs/indices/xattrop # for i in
[a-f0-9]*; do if stat ../../${i:0:2}/${i:2:2}/${i} &>/dev/null; then
echo $i exists; else echo $i does not; fi ; done
12427a88-4a42-4cc1-bbd3-13e4cb8d7e6a does not
1a1e0425-acdb-4ed1-9c62-bb866f34b0c7 does not
ed2cefe8-3854-49e5-9433-7198f53ffec5 does not
Which to me is indicative that upon file removal these xattrop files are
left behind.
I'm not sure if this is by design, or a bug, or more likely due to a
misunderstanding of how these actually function.
Since gluster volume heal ... info can potentially take a long time
under the kind of conditions that we're mindful of we'd prefer to use
heal-count so that we can receive our alerts in a more timely manner.
Kind Regards,
Jaco