[Gluster-users] gluster heal entry reappears

[Ravishankar N]

On 05/28/2017 10:31 PM, Markus Stockhausen wrote:
> Hi,
>
> I'm fairly new to gluster and quite happy with it. We are using it in 
> an OVirt
> environment that stores its VM images in the gluster. Setup is as 
> follows and
> Clients mount the volume with gluster native fuse protocol.
>
> 3 storage nodes: Centos 7, Gluster 3.8.12 (managed by me), 2 bricks each
> 5 virtualization nodes: Centos 7, Gluster 3.8.12 (managed by OVirt engine)
>
> After todays reboot of one of the storage nodes the recovery did not 
> finish
> successfully. The state of one brick remained in:
>
> [root at cfiler301 dom_md]# gluster volume heal gluster1 info
> ...
> Brick cfilers201:/var/data/brick1/brick
> /b1de7818-020b-4f47-938f-f3ebb51836a3/dom_md/ids
> Status: Connected
> Number of entries: 1
> ...
>
> The above file is used by sanlock runing on the OVirt nodes to handle VM
> image locking. Issuing a manual heal with "gluster volume heal 
> gluster1" fixed
> the problem but the unsynced entry reappeared a few seconds later.
>
> My question: Should this situation be recovered automatically and if yes
> what might be the culprit?

We have had this observed by our QE guys while testing too, but in all 
cases, there was an intermittent disconnect from the fuse mount to the 
bricks, leading to the 'ids' file needing heal (and being healed on 
reconnect) again and again.
Perhaps you should check if and why the mount is getting disconnected 
from the bricks.

HTH,
Ravi
>
> Best regards.
>
> Markus
>
> P.S. I finally fixed the issue by remounting the filesystems (on Ovirt 
> nodes) and
> so sanlock was restarted too.
>
>
>
>
> _______________________________________________
> Gluster-users mailing list
> Gluster-users at gluster.org
> http://lists.gluster.org/mailman/listinfo/gluster-users


-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.gluster.org/pipermail/gluster-users/attachments/20170529/53a3bf9a/attachment.html>