4:13 a.m.
I restored my engine to a gluster volume named :/engine on a three node hyperconverged
oVirt 4.3.3.1 cluster. Before restoring I was checking the status of the volumes. They
were clean. No heal entries. All peers connected. gluster volume status looked good. Then
I restored. This went well. The engine is up. But the engine gluster volume shows entries
on node02 and node03. The engine was installed to node01. I have to deploy the engine to
the other two hosts to reach full HA, but I bet maintenance is not possible until the
volume is healed.
I tried "gluster volume heal engine" also with added "full". The heal
entries will disappear for a few seconds and then /dom_md/ids will pop up again. The
__DIRECT_IO_TEST__ will join later. The split-brain info has no entries. Is this some kind
of hidden split brain? Maybe there is data on node01 brick which got not synced to the
other two nodes? I can only speculate. Gluster docs say: this should heal. But it
doesn't. I have two other volumes. Those are fine. One of them containing 3 VMs that
are running. I also tried to shut down the engine, so no-one was using the volume. Then
heal. Same effect. Those two files will always show up. But none other. Heal can always be
started successfully from any of the participating nodes.
Reset the volume bricks one by one and cross fingers?
[root@node03 ~]# gluster volume heal engine info
Brick node01.infra.solutions.work:/gluster_bricks/engine/engine
Status: Connected
Number of entries: 0
Brick node02.infra.solutions.work:/gluster_bricks/engine/engine
/9f4d5ae9-e01d-4b73-8b6d-e349279e9782/dom_md/ids
/__DIRECT_IO_TEST__
Status: Connected
Number of entries: 2
Brick node03.infra.solutions.work:/gluster_bricks/engine/engine
/9f4d5ae9-e01d-4b73-8b6d-e349279e9782/dom_md/ids
/__DIRECT_IO_TEST__
Status: Connected
Number of entries: 2
10:14 a.m.
New subject: Gluster volume heals and after 5 seconds has /dom_md/ids dirty again
I see this sometimes after rebooting a server, and it usually stops happening, generally
within a few hours, I’ve never tracked it down further. Don’t know for sure, but I assume
it’s related to healing and goes away once everything syncs up.
Occasionally it turns out to be a communications problem between servers (usually an
update to something screws up my firewall), so I always check my peer status when I see it
and make sure all servers are talking to each other.
10:21 a.m.
New subject: Gluster volume heals and after 5 seconds has /dom_md/ids dirty again
I've been having problems with my gluster Engine volume recently as well
after updating to latest stable 4.3.3. For the past few days I've seen a
random brick in the Engine volume go down and I have to force start it to
get it working again. Right now I'm seeing that there are unsynced entries
and one node is showing transport endpoint not connected, even though peer
status if fine and all other volumes are working normally.
On Mon, May 13, 2019 at 12:14 PM Darrell Budic <budic(a)onholyground.com>
wrote:
10:25 a.m.
New subject: Gluster volume heals and after 5 seconds has /dom_md/ids dirty again
Yes. After a reboot you could have a sync issue for up to a few hours. But this issue
persists now for 24 days. Additionally I see errors in the glustershd.log of the two hosts
that are having heal info for that volume. The first node shows as OK and has no errors in
its glustershd.log.
The errors are like this:
[2019-05-13 15:18:40.808945] W [MSGID: 114031]
[client-rpc-fops_v2.c:2634:client4_0_lookup_cbk] 0-engine-client-1: remote operation
failed. Path: <gfid:95ba9fb2-b0ae-436c-9c31-2779cf202235>
(95ba9fb2-b0ae-436c-9c31-2779cf202235) [No such file or directory]
[2019-05-13 15:18:40.809113] W [MSGID: 114031]
[client-rpc-fops_v2.c:2634:client4_0_lookup_cbk] 0-engine-client-2: remote operation
failed. Path: <gfid:95ba9fb2-b0ae-436c-9c31-2779cf202235>
(95ba9fb2-b0ae-436c-9c31-2779cf202235) [No such file or directory]
[root@node02 ~]#
Looks like the first node is sane and the other two are the masters but are not so sane.
:-/
10:37 a.m.
New subject: Gluster volume heals and after 5 seconds has /dom_md/ids dirty again
I encountered serious issues with 5.3-5.5 (crashing bricks, multiple brick processes for
the same brick causing disconnects and excessive heals). I had better luck with 5.6,
although it’s not clear to me if the duplicate brick process issue is still present in
that version. I finally jumped to 6 which has been more stable for me. I’d recommend
upgrading at least to 5.6 if not going right to 6.1.
10:47 a.m.
New subject: Gluster volume heals and after 5 seconds has /dom_md/ids dirty again
Please note that I am running a hyper-converged NodeNG setup. I understand that upgrading
single components is not really possible with a ovirt Node NG. And could probably break
the datacenter upgrade path.
Could you point out some reference for your suggestions? Docs, Bug reports or the sorts?
> I encountered serious issues with 5.3-5.5 (crashing bricks, multiple brick processes
for
> the same brick causing disconnects and excessive heals). I had better luck with 5.6,
> although it’s not clear to me if the duplicate brick process issue is still present
in
> that version. I finally jumped to 6 which has been more stable for me. I’d recommend
> upgrading at least to 5.6 if not going right to 6.1.
11:02 a.m.
New subject: Gluster volume heals and after 5 seconds has /dom_md/ids dirty again
Ovirt just pulls in the gluster5 repos, if you upgrade now you should get gluster 5.6 on
your nodes. If you’re running them on centos, you can install centos-release-gluster6 to
go to gluster6. Ovirt NodeNG is a different story, as you mention, but I believe you can
still run an update on it to get the latest gluster version?
Those recommendations are based on my personal experience, but see also:
https://bugzilla.redhat.com/show_bug.cgi?id=1683602
<https://bugzilla.redhat.com/show_bug.cgi?id=1683602>
https://bugzilla.redhat.com/show_bug.cgi?id=1677319
<https://bugzilla.redhat.com/show_bug.cgi?id=1677319>
11:29 a.m.
New subject: Gluster volume heals and after 5 seconds has /dom_md/ids dirty again
I use node NG as well, I just updated to 4.3.3 two days ago and I'm on
Gluster 5.5. Yum update on host node yields no updates available
On Mon, May 13, 2019 at 1:03 PM Darrell Budic <budic(a)onholyground.com>
wrote:
4:12 a.m.
New subject: Gluster volume heals and after 5 seconds has /dom_md/ids dirty again
There are plans for Node NG 4.3.5 to switch to Gluster V6.
So far I still have unsynced entries with gluster 5.5. I tried to reset every brick, but
that resolved to having the same heal info for that volume.
I'm currently re-installing a node that has heal info on the gluster volume. Let's
see what happens after that.
See:
https://lists.ovirt.org/archives/list/devel@ovirt.org/thread/IVTYCFKCQ5P2...
2163
days inactive
2164
days old
11 comments
3 participants
participants (3)
-
Andreas Elvers -
Darrell Budic -
Jayme