[Gluster-users] Conv distr-repl 2 to repl 3 arb 1 now 2 of 6 arb bricks won't get healed

a.schwibbe at gmx.net a.schwibbe at gmx.net
Sun May 30 12:12:32 UTC 2021


I am seeking help here after looking for solutions on the web for my distributed-replicated volume.
My volume is operated since v3.10 and I upgraded through to 7.9, replaced nodes, replaced bricks without a problem. I love it.
Finally I wanted to extend my 6x2 distributed replicated volume with arbiters for better split-brain protection.

So I add-brick with replication 3 arbiter 1 (as I had a 6x2 I obviously added 6 arb bricks) and it successfully converted to 6 x (2 +1) and self-heal immideately started. Looking good.


Version: 7.9

Number of Bricks: 6 x (2 + 1) = 18

cluster.max-op-version: 70200

Peers: 3 (node[0..2])

Layout

|node0 |node1 |node2
|brick0 |brick0 |arbit0

|arbit1 |brick1 |brick1

....


I then recognized that arbiter volumes on node0 & node1 have been healed successfully.
Unfortunately all arbiter volumes on node2 have not been healed!
I realized that the main dir on my arb mount point has been added (mount point /var/brick/arb_0 now contains dir "brick") however this dir on _all_ other bricks has numeric ID 33, but on this on it has 0). The brick dir on the faulty arb-volumes does contain ".glusterfs", however it has only very few entries. Other than that "brick" is empty.
At that point I changed brick dir owner with chown to 33:33 and hoped for self-heal to work. It did not.
I hoped a rebalance fix-layout would fix things. It did not.
I hoped for a glusterd restart on node2 (as this is happening to both arb volumes on this node exclusively) would help. It did not.

Active mount points via nfs-ganesha or fuse continue to work.
Existing clients cause errors in the arb-brick logs on node2 for missing files or dirs, but clients seem not affected. r/w operations work.

New clients are not able to fuse mount the volume for "authentication error".
heal statistics heal-count show several hundred files need healing, this count is rising. Watching df on the arb-brick mount point on node2 shows every now and then a few bytes written, but then removed immideately after that.

Any help/recommendation for you highly appreciated.
Thank you!

A.


More information about the Gluster-users mailing list