11 Apr
2023
11 Apr
'23
4:24 a.m.
Hi,
If you remember, I hit bug https://tracker.ceph.com/issues/58489 so I
was very relieved when 17.2.6 was released and started to update
immediately.
But now I'm stuck again with my broken MDS. MDS won't get into up:active
without the update but the update waits for them to get into up:active
state. Seems like a deadlock / chicken-egg problem to me.
Since I'm still relatively new to Ceph, could you help me?
What I see when watching the update status:
{
"target_image":
"quay.io/ceph/ceph@sha256:1161e35e4e02cf377c93b913ce78773f8413f5a8d7c5eaee4b4773a4f9dd6635",
"in_progress": true,
"which": "Upgrading all daemon types on all hosts",
"services_complete": [
"crash",
"mgr",
"mon",
"osd"
],
"progress": "18/40 daemons upgraded",
"message": "Error: UPGRADE_OFFLINE_HOST: Upgrade: Failed to connect
to host ceph01 at addr (192.168.23.61)",
"is_paused": false
}
(The offline host was one host that broke during the upgrade. I fixed
that in the meantime and the update went on.)
And in the log:
2023-04-10T19:23:48.750129+0000 mgr.ceph04.qaexpv [INF] Upgrade: Waiting
for mds.mds01.ceph04.hcmvae to be up:active (currently up:replay)
2023-04-10T19:23:58.758141+0000 mgr.ceph04.qaexpv [WRN] Upgrade: No mds
is up; continuing upgrade procedure to poke things in the right direction
Please give me a hint what I can do.
Cheers,
Thomas
--
http://www.widhalm.or.at
GnuPG : 6265BAE6 , A84CB603
Threema: H7AV7D33
Telegram, Signal: widhalmt(a)widhalm.or.at
11 Apr
11 Apr
4:51 a.m.
You could try pausing the upgrade and manually "upgrading" the mds daemons
by redeploying them on the new image. Something like "ceph orch daemon
redeploy <mds-daemon-name> --image <17.2.6 image>" (daemon names should
match those in "ceph orch ps" output). If you do that for all of them and
then get them into an up state you should be able to resume the upgrade and
have it complete.
On Mon, Apr 10, 2023 at 3:25 PM Thomas Widhalm <widhalmt(a)widhalm.or.at>
wrote:
Hi,
If you remember, I hit bug https://tracker.ceph.com/issues/58489 so I
was very relieved when 17.2.6 was released and started to update
immediately.
But now I'm stuck again with my broken MDS. MDS won't get into up:active
without the update but the update waits for them to get into up:active
state. Seems like a deadlock / chicken-egg problem to me.
Since I'm still relatively new to Ceph, could you help me?
What I see when watching the update status:
{
"target_image":
"
quay.io/ceph/ceph@sha256:1161e35e4e02cf377c93b913ce78773f8413f5a8d7c5eaee4b4773a4f9dd6635
",
"in_progress": true,
"which": "Upgrading all daemon types on all hosts",
"services_complete": [
"crash",
"mgr",
"mon",
"osd"
],
"progress": "18/40 daemons upgraded",
"message": "Error: UPGRADE_OFFLINE_HOST: Upgrade: Failed to connect
to host ceph01 at addr (192.168.23.61)",
"is_paused": false
}
(The offline host was one host that broke during the upgrade. I fixed
that in the meantime and the update went on.)
And in the log:
2023-04-10T19:23:48.750129+0000 mgr.ceph04.qaexpv [INF] Upgrade: Waiting
for mds.mds01.ceph04.hcmvae to be up:active (currently up:replay)
2023-04-10T19:23:58.758141+0000 mgr.ceph04.qaexpv [WRN] Upgrade: No mds
is up; continuing upgrade procedure to poke things in the right direction
Please give me a hint what I can do.
Cheers,
Thomas
--
http://www.widhalm.or.at
GnuPG : 6265BAE6 , A84CB603
Threema: H7AV7D33
Telegram, Signal: widhalmt(a)widhalm.or.at
_______________________________________________
ceph-users mailing list -- ceph-users(a)ceph.io
To unsubscribe send an email to ceph-users-leave(a)ceph.io
4:53 a.m.
Will also note that the normal upgrade process scales down the mds service
to have only 1 mds per fs before upgrading it, so maybe something you'd
want to do as well if the upgrade didn't do it already. It does so by
setting the max_mds to 1 for the fs.
On Mon, Apr 10, 2023 at 3:51 PM Adam King <adking(a)redhat.com> wrote:
You could try pausing the upgrade and manually
"upgrading" the mds daemons
by redeploying them on the new image. Something like "ceph orch daemon
redeploy <mds-daemon-name> --image <17.2.6 image>" (daemon names should
match those in "ceph orch ps" output). If you do that for all of them and
then get them into an up state you should be able to resume the upgrade and
have it complete.
On Mon, Apr 10, 2023 at 3:25 PM Thomas Widhalm <widhalmt(a)widhalm.or.at>
wrote:
Hi,
If you remember, I hit bug https://tracker.ceph.com/issues/58489 so I
was very relieved when 17.2.6 was released and started to update
immediately.
But now I'm stuck again with my broken MDS. MDS won't get into up:active
without the update but the update waits for them to get into up:active
state. Seems like a deadlock / chicken-egg problem to me.
Since I'm still relatively new to Ceph, could you help me?
What I see when watching the update status:
{
"target_image":
"
quay.io/ceph/ceph@sha256:1161e35e4e02cf377c93b913ce78773f8413f5a8d7c5eaee4b4773a4f9dd6635
",
"in_progress": true,
"which": "Upgrading all daemon types on all hosts",
"services_complete": [
"crash",
"mgr",
"mon",
"osd"
],
"progress": "18/40 daemons upgraded",
"message": "Error: UPGRADE_OFFLINE_HOST: Upgrade: Failed to connect
to host ceph01 at addr (192.168.23.61)",
"is_paused": false
}
(The offline host was one host that broke during the upgrade. I fixed
that in the meantime and the update went on.)
And in the log:
2023-04-10T19:23:48.750129+0000 mgr.ceph04.qaexpv [INF] Upgrade: Waiting
for mds.mds01.ceph04.hcmvae to be up:active (currently up:replay)
2023-04-10T19:23:58.758141+0000 mgr.ceph04.qaexpv [WRN] Upgrade: No mds
is up; continuing upgrade procedure to poke things in the right direction
Please give me a hint what I can do.
Cheers,
Thomas
--
http://www.widhalm.or.at
GnuPG : 6265BAE6 , A84CB603
Threema: H7AV7D33
Telegram, Signal: widhalmt(a)widhalm.or.at
_______________________________________________
ceph-users mailing list -- ceph-users(a)ceph.io
To unsubscribe send an email to ceph-users-leave(a)ceph.io
5:25 a.m.
I did what you told me.
I also see in the log, that the command went through:
2023-04-10T19:58:46.522477+0000 mgr.ceph04.qaexpv [INF] Schedule
redeploy daemon mds.mds01.ceph06.rrxmks
2023-04-10T20:01:03.360559+0000 mgr.ceph04.qaexpv [INF] Schedule
redeploy daemon mds.mds01.ceph05.pqxmvt
2023-04-10T20:01:21.787635+0000 mgr.ceph04.qaexpv [INF] Schedule
redeploy daemon mds.mds01.ceph07.omdisd
But the MDS never start. They stay in error state. I tried to redeploy
and start them a few times. Even restarted one host where a MDS should run.
mds.mds01.ceph03.xqwdjy ceph03 error 32m ago
2M - - <unknown> <unknown> <unknown>
mds.mds01.ceph04.hcmvae ceph04 error 31m ago
2h - - <unknown> <unknown> <unknown>
mds.mds01.ceph05.pqxmvt ceph05 error 32m ago
9M - - <unknown> <unknown> <unknown>
mds.mds01.ceph06.rrxmks ceph06 error 32m ago
10w - - <unknown> <unknown> <unknown>
mds.mds01.ceph07.omdisd ceph07 error 32m ago
2M - - <unknown> <unknown> <unknown>
And other ideas? Or am I missing something.
Cheers,
Thomas
On 10.04.23 21:53, Adam King wrote:
Will also note that the normal upgrade process scales
down the mds
service to have only 1 mds per fs before upgrading it, so maybe
something you'd want to do as well if the upgrade didn't do it already.
It does so by setting the max_mds to 1 for the fs.
On Mon, Apr 10, 2023 at 3:51 PM Adam King <adking(a)redhat.com
<mailto:adking@redhat.com>> wrote:
You could try pausing the upgrade and manually "upgrading" the mds
daemons by redeploying them on the new image. Something like "ceph
orch daemon redeploy <mds-daemon-name> --image <17.2.6 image>"
(daemon names should match those in "ceph orch ps" output). If you
do that for all of them and then get them into an up state you
should be able to resume the upgrade and have it complete.
On Mon, Apr 10, 2023 at 3:25 PM Thomas Widhalm
<widhalmt(a)widhalm.or.at <mailto:widhalmt@widhalm.or.at>> wrote:
Hi,
If you remember, I hit bug https://tracker.ceph.com/issues/58489
<https://tracker.ceph.com/issues/58489> so I
was very relieved when 17.2.6 was released and started to update
immediately.
But now I'm stuck again with my broken MDS. MDS won't get into
up:active
without the update but the update waits for them to get into
up:active
state. Seems like a deadlock / chicken-egg problem to me.
Since I'm still relatively new to Ceph, could you help me?
What I see when watching the update status:
{
"target_image":
"quay.io/ceph/ceph@sha256:1161e35e4e02cf377c93b913ce78773f8413f5a8d7c5eaee4b4773a4f9dd6635
<http://quay.io/ceph/ceph@sha256:1161e35e4e02cf377c93b913ce78773f8413f5a8d7c5eaee4b4773a4f9dd6635>",
"in_progress": true,
"which": "Upgrading all daemon types on all hosts",
"services_complete": [
"crash",
"mgr",
"mon",
"osd"
],
"progress": "18/40 daemons upgraded",
"message": "Error: UPGRADE_OFFLINE_HOST: Upgrade: Failed
to connect
to host ceph01 at addr (192.168.23.61)",
"is_paused": false
}
(The offline host was one host that broke during the upgrade. I
fixed
that in the meantime and the update went on.)
And in the log:
2023-04-10T19:23:48.750129+0000 mgr.ceph04.qaexpv [INF] Upgrade:
Waiting
for mds.mds01.ceph04.hcmvae to be up:active (currently up:replay)
2023-04-10T19:23:58.758141+0000 mgr.ceph04.qaexpv [WRN] Upgrade:
No mds
is up; continuing upgrade procedure to poke things in the right
direction
Please give me a hint what I can do.
Cheers,
Thomas
--
http://www.widhalm.or.at <http://www.widhalm.or.at>
GnuPG : 6265BAE6 , A84CB603
Threema: H7AV7D33
Telegram, Signal: widhalmt(a)widhalm.or.at
<mailto:widhalmt@widhalm.or.at>
_______________________________________________
ceph-users mailing list -- ceph-users(a)ceph.io
<mailto:ceph-users@ceph.io>
To unsubscribe send an email to ceph-users-leave(a)ceph.io
<mailto:ceph-users-leave@ceph.io>
6:02 a.m.
It seems like it maybe didn't actually do the redeploy as it should log
something saying it's actually doing it on top of the line saying it
scheduled it. To confirm, the upgrade is paused ("ceph orch upgrade status"
reports is_paused as false)? If so, maybe try doing a mgr failover ("ceph
mgr fail") and then check "ceph orch ps" and "ceph orch device
ls" a few
minutes later and look at the REFRESHED column. If any of those are giving
amounts of time farther back then when you did the failover, there's
probably something going on on the host(s) where it says it hasn't
refreshed recently that's sticking things up (you'd have to go on that host
and look for hanging cephadm commands). Lastly, you could look at the
/var/lib/ceph/<fsid>/<mds-daemon-name>/unit.run file on the hosts where the
mds daemons are deployed. The (very long) last podman/docker run line in
that file should have the image name of the image the daemon is being
deployed with. So you could use that to confirm if cephadm ever actually
tried a redeploy of the mds with the new image. You could also check the
journal logs for the mds. Cephadm reports the sytemd unit name for the
daemon as part of "cephadm ls" output if you put a copy of the cephadm
binary, un "cephadm ls" with it, grab the systemd unit name for the mds
daemon form that output, you could use that to check the journal logs which
should tell the last restart time and why it's gone down.
On Mon, Apr 10, 2023 at 4:25 PM Thomas Widhalm <widhalmt(a)widhalm.or.at>
wrote:
I did what you told me.
I also see in the log, that the command went through:
2023-04-10T19:58:46.522477+0000 mgr.ceph04.qaexpv [INF] Schedule
redeploy daemon mds.mds01.ceph06.rrxmks
2023-04-10T20:01:03.360559+0000 mgr.ceph04.qaexpv [INF] Schedule
redeploy daemon mds.mds01.ceph05.pqxmvt
2023-04-10T20:01:21.787635+0000 mgr.ceph04.qaexpv [INF] Schedule
redeploy daemon mds.mds01.ceph07.omdisd
But the MDS never start. They stay in error state. I tried to redeploy
and start them a few times. Even restarted one host where a MDS should run.
mds.mds01.ceph03.xqwdjy ceph03 error 32m ago
2M - - <unknown> <unknown> <unknown>
mds.mds01.ceph04.hcmvae ceph04 error 31m ago
2h - - <unknown> <unknown> <unknown>
mds.mds01.ceph05.pqxmvt ceph05 error 32m ago
9M - - <unknown> <unknown> <unknown>
mds.mds01.ceph06.rrxmks ceph06 error 32m ago
10w - - <unknown> <unknown> <unknown>
mds.mds01.ceph07.omdisd ceph07 error 32m ago
2M - - <unknown> <unknown> <unknown>
And other ideas? Or am I missing something.
Cheers,
Thomas
On 10.04.23 21:53, Adam King wrote:
Will also note that the normal upgrade process
scales down the mds
service to have only 1 mds per fs before upgrading it, so maybe
something you'd want to do as well if the upgrade didn't do it already.
It does so by setting the max_mds to 1 for the fs.
On Mon, Apr 10, 2023 at 3:51 PM Adam King <adking(a)redhat.com
<mailto:adking@redhat.com>> wrote:
You could try pausing the upgrade and manually "upgrading" the mds
daemons by redeploying them on the new image. Something like "ceph
orch daemon redeploy <mds-daemon-name> --image <17.2.6 image>"
(daemon names should match those in "ceph orch ps" output). If you
do that for all of them and then get them into an up state you
should be able to resume the upgrade and have it complete.
On Mon, Apr 10, 2023 at 3:25 PM Thomas Widhalm
<widhalmt(a)widhalm.or.at <mailto:widhalmt@widhalm.or.at>> wrote:
Hi,
If you remember, I hit bug https://tracker.ceph.com/issues/58489
<https://tracker.ceph.com/issues/58489> so I
was very relieved when 17.2.6 was released and started to update
immediately.
But now I'm stuck again with my broken MDS. MDS won't get into
up:active
without the update but the update waits for them to get into
up:active
state. Seems like a deadlock / chicken-egg problem to me.
Since I'm still relatively new to Ceph, could you help me?
What I see when watching the update status:
{
"target_image":
"
quay.io/ceph/ceph@sha256:1161e35e4e02cf377c93b913ce78773f8413f5a8d7c5eaee4b4773a4f9dd6635
<
http://quay.io/ceph/ceph@sha256:1161e35e4e02cf377c93b913ce78773f8413f5a8d7c…
",
"in_progress": true,
"which": "Upgrading all daemon types on all hosts",
"services_complete": [
"crash",
"mgr",
"mon",
"osd"
],
"progress": "18/40 daemons upgraded",
"message": "Error: UPGRADE_OFFLINE_HOST: Upgrade: Failed
to connect
to host ceph01 at addr (192.168.23.61)",
"is_paused": false
}
(The offline host was one host that broke during the upgrade. I
fixed
that in the meantime and the update went on.)
And in the log:
2023-04-10T19:23:48.750129+0000 mgr.ceph04.qaexpv [INF] Upgrade:
Waiting
for mds.mds01.ceph04.hcmvae to be up:active (currently up:replay)
2023-04-10T19:23:58.758141+0000 mgr.ceph04.qaexpv [WRN] Upgrade:
No mds
is up; continuing upgrade procedure to poke things in the right
direction
Please give me a hint what I can do.
Cheers,
Thomas
--
http://www.widhalm.or.at <http://www.widhalm.or.at>
GnuPG : 6265BAE6 , A84CB603
Threema: H7AV7D33
Telegram, Signal: widhalmt(a)widhalm.or.at
<mailto:widhalmt@widhalm.or.at>
_______________________________________________
ceph-users mailing list -- ceph-users(a)ceph.io
<mailto:ceph-users@ceph.io>
To unsubscribe send an email to ceph-users-leave(a)ceph.io
<mailto:ceph-users-leave@ceph.io>
4:16 p.m.
On 4/11/23 03:24, Thomas Widhalm wrote:
Hi,
If you remember, I hit bug https://tracker.ceph.com/issues/58489 so I
was very relieved when 17.2.6 was released and started to update
immediately.
Please note, this fix is not in the v17.2.6 yet in upstream code.
Thanks
- Xiubo
But now I'm stuck again with my broken MDS. MDS
won't get into
up:active without the update but the update waits for them to get into
up:active state. Seems like a deadlock / chicken-egg problem to me.
Since I'm still relatively new to Ceph, could you help me?
What I see when watching the update status:
{
"target_image":
"quay.io/ceph/ceph@sha256:1161e35e4e02cf377c93b913ce78773f8413f5a8d7c5eaee4b4773a4f9dd6635",
"in_progress": true,
"which": "Upgrading all daemon types on all hosts",
"services_complete": [
"crash",
"mgr",
"mon",
"osd"
],
"progress": "18/40 daemons upgraded",
"message": "Error: UPGRADE_OFFLINE_HOST: Upgrade: Failed to
connect to host ceph01 at addr (192.168.23.61)",
"is_paused": false
}
(The offline host was one host that broke during the upgrade. I fixed
that in the meantime and the update went on.)
And in the log:
2023-04-10T19:23:48.750129+0000 mgr.ceph04.qaexpv [INF] Upgrade:
Waiting for mds.mds01.ceph04.hcmvae to be up:active (currently up:replay)
2023-04-10T19:23:58.758141+0000 mgr.ceph04.qaexpv [WRN] Upgrade: No
mds is up; continuing upgrade procedure to poke things in the right
direction
Please give me a hint what I can do.
Cheers,
Thomas
_______________________________________________
ceph-users mailing list -- ceph-users(a)ceph.io
To unsubscribe send an email to ceph-users-leave(a)ceph.io
4:59 p.m.
On 11.04.23 09:16, Xiubo Li wrote:
On 4/11/23 03:24, Thomas Widhalm wrote:
Thanks for the information. I misread the information in the tracker. Do
you have a predicted schedule for the backport? Or should I go for a
specific pre-release? I don't want to take chances but I'm desperate
because my production system is affected and offline for several weeks now.
Thanks,
Thomas
Hi,
If you remember, I hit bug https://tracker.ceph.com/issues/58489 so I
was very relieved when 17.2.6 was released and started to update
immediately.
Please note, this fix is not in the v17.2.6 yet in upstream code.
Thanks
- Xiubo
But now I'm stuck again with my broken MDS.
MDS won't get into
up:active without the update but the update waits for them to get into
up:active state. Seems like a deadlock / chicken-egg problem to me.
Since I'm still relatively new to Ceph, could you help me?
What I see when watching the update status:
{
"target_image":
"quay.io/ceph/ceph@sha256:1161e35e4e02cf377c93b913ce78773f8413f5a8d7c5eaee4b4773a4f9dd6635",
"in_progress": true,
"which": "Upgrading all daemon types on all hosts",
"services_complete": [
"crash",
"mgr",
"mon",
"osd"
],
"progress": "18/40 daemons upgraded",
"message": "Error: UPGRADE_OFFLINE_HOST: Upgrade: Failed to
connect to host ceph01 at addr (192.168.23.61)",
"is_paused": false
}
(The offline host was one host that broke during the upgrade. I fixed
that in the meantime and the update went on.)
And in the log:
2023-04-10T19:23:48.750129+0000 mgr.ceph04.qaexpv [INF] Upgrade:
Waiting for mds.mds01.ceph04.hcmvae to be up:active (currently up:replay)
2023-04-10T19:23:58.758141+0000 mgr.ceph04.qaexpv [WRN] Upgrade: No
mds is up; continuing upgrade procedure to poke things in the right
direction
Please give me a hint what I can do.
Cheers,
Thomas
_______________________________________________
ceph-users mailing list -- ceph-users(a)ceph.io
To unsubscribe send an email to ceph-users-leave(a)ceph.io
12 Apr
12 Apr
10:06 a.m.
On 4/11/23 15:59, Thomas Widhalm wrote:
On 11.04.23 09:16, Xiubo Li wrote:
The backport is already queued to review and test, but I am not very
sure when it can get merged. I am not very sure this can 100% resolve
your issue in case when you have other corruptions in you production.
Thanks
On 4/11/23 03:24, Thomas Widhalm wrote:
Thanks for the information. I misread the information in the tracker.
Do you have a predicted schedule for the backport? Or should I go for
a specific pre-release? I don't want to take chances but I'm desperate
because my production system is affected and offline for several weeks
now.
Hi,
If you remember, I hit bug https://tracker.ceph.com/issues/58489 so
I was very relieved when 17.2.6 was released and started to update
immediately.
Please note, this fix is not in the v17.2.6 yet in upstream code.
Thanks,
Thomas
> Thanks
>
> - Xiubo
>
>
>> But now I'm stuck again with my broken MDS. MDS won't get into
>> up:active without the update but the update waits for them to get
>> into up:active state. Seems like a deadlock / chicken-egg problem to
>> me.
>>
>> Since I'm still relatively new to Ceph, could you help me?
>>
>> What I see when watching the update status:
>>
>> {
>> "target_image":
>>
"quay.io/ceph/ceph@sha256:1161e35e4e02cf377c93b913ce78773f8413f5a8d7c5eaee4b4773a4f9dd6635",
>> "in_progress": true,
>> "which": "Upgrading all daemon types on all hosts",
>> "services_complete": [
>> "crash",
>> "mgr",
>> "mon",
>> "osd"
>> ],
>> "progress": "18/40 daemons upgraded",
>> "message": "Error: UPGRADE_OFFLINE_HOST: Upgrade: Failed to
>> connect to host ceph01 at addr (192.168.23.61)",
>> "is_paused": false
>> }
>>
>> (The offline host was one host that broke during the upgrade. I
>> fixed that in the meantime and the update went on.)
>>
>> And in the log:
>>
>> 2023-04-10T19:23:48.750129+0000 mgr.ceph04.qaexpv [INF] Upgrade:
>> Waiting for mds.mds01.ceph04.hcmvae to be up:active (currently
>> up:replay)
>> 2023-04-10T19:23:58.758141+0000 mgr.ceph04.qaexpv [WRN] Upgrade: No
>> mds is up; continuing upgrade procedure to poke things in the right
>> direction
>>
>>
>> Please give me a hint what I can do.
>>
>> Cheers,
>> Thomas
>>
>> _______________________________________________
>> ceph-users mailing list -- ceph-users(a)ceph.io
>> To unsubscribe send an email to ceph-users-leave(a)ceph.io
>
13 Apr
13 Apr
4:45 a.m.
Thanks for your detailed explanations! That helped a lot.
All MDS are still in status error. "ceph orch device ls" showed that
some hosts seem to not have enough space on devices. I wonder why I
didn't see that in monitoring. Anyway, I'll fix that and then try to
proceed.
When the backport is available I'll try to upgrade as fast as possible.
Hopefully that will suffice to restore the system.
Thank you all so far for your great work!
On 4/10/23 23:02, Adam King wrote:
> It seems like it maybe didn't actually do the redeploy as it should log
> something saying it's actually doing it on top of the line saying it
> scheduled it. To confirm, the upgrade is paused ("ceph orch upgrade
> status" reports is_paused as false)? If so, maybe try doing a mgr
> failover ("ceph mgr fail") and then check "ceph orch ps" and
"ceph orch
> device ls" a few minutes later and look at the REFRESHED column. If any
> of those are giving amounts of time farther back then when you did the
> failover, there's probably something going on on the host(s) where it
> says it hasn't refreshed recently that's sticking things up (you'd have
> to go on that host and look for hanging cephadm commands). Lastly, you
> could look at the /var/lib/ceph/<fsid>/<mds-daemon-name>/unit.run file
> on the hosts where the mds daemons are deployed. The (very long) last
> podman/docker run line in that file should have the image name of the
> image the daemon is being deployed with. So you could use that to
> confirm if cephadm ever actually tried a redeploy of the mds with the
> new image. You could also check the journal logs for the mds.
> Cephadm reports the sytemd unit name for the daemon as part of "cephadm
> ls" output if you put a copy of the cephadm binary, un "cephadm ls"
with
> it, grab the systemd unit name for the mds daemon form that output, you
> could use that to check the journal logs which should tell the last
> restart time and why it's gone down.
>
> On Mon, Apr 10, 2023 at 4:25 PM Thomas Widhalm <widhalmt(a)widhalm.or.at
> <mailto:widhalmt@widhalm.or.at>> wrote:
>
> I did what you told me.
>
> I also see in the log, that the command went through:
>
> 2023-04-10T19:58:46.522477+0000 mgr.ceph04.qaexpv [INF] Schedule
> redeploy daemon mds.mds01.ceph06.rrxmks
> 2023-04-10T20:01:03.360559+0000 mgr.ceph04.qaexpv [INF] Schedule
> redeploy daemon mds.mds01.ceph05.pqxmvt
> 2023-04-10T20:01:21.787635+0000 mgr.ceph04.qaexpv [INF] Schedule
> redeploy daemon mds.mds01.ceph07.omdisd
>
>
> But the MDS never start. They stay in error state. I tried to redeploy
> and start them a few times. Even restarted one host where a MDS
> should run.
>
> mds.mds01.ceph03.xqwdjy ceph03 error 32m ago
> 2M - - <unknown> <unknown> <unknown>
> mds.mds01.ceph04.hcmvae ceph04 error 31m ago
> 2h - - <unknown> <unknown> <unknown>
> mds.mds01.ceph05.pqxmvt ceph05 error 32m ago
> 9M - - <unknown> <unknown> <unknown>
> mds.mds01.ceph06.rrxmks ceph06 error 32m ago
> 10w - - <unknown> <unknown> <unknown>
> mds.mds01.ceph07.omdisd ceph07 error 32m ago
> 2M - - <unknown> <unknown> <unknown>
>
>
> And other ideas? Or am I missing something.
>
> Cheers,
> Thomas
>
> On 10.04.23 21:53, Adam King wrote:
> > Will also note that the normal upgrade process scales down the mds
> > service to have only 1 mds per fs before upgrading it, so maybe
> > something you'd want to do as well if the upgrade didn't do it
> already.
> > It does so by setting the max_mds to 1 for the fs.
> >
> > On Mon, Apr 10, 2023 at 3:51 PM Adam King <adking(a)redhat.com
> <mailto:adking@redhat.com>
> > <mailto:adking@redhat.com <mailto:adking@redhat.com>>>
wrote:
> >
> > You could try pausing the upgrade and manually "upgrading"
> the mds
> > daemons by redeploying them on the new image. Something like
> "ceph
> > orch daemon redeploy <mds-daemon-name> --image <17.2.6
image>"
> > (daemon names should match those in "ceph orch ps" output).
> If you
> > do that for all of them and then get them into an up state you
> > should be able to resume the upgrade and have it complete.
> >
> > On Mon, Apr 10, 2023 at 3:25 PM Thomas Widhalm
> > <widhalmt(a)widhalm.or.at <mailto:widhalmt@widhalm.or.at>
> <mailto:widhalmt@widhalm.or.at <mailto:widhalmt@widhalm.or.at>>>
wrote:
> >
> > Hi,
> >
> > If you remember, I hit bug
> https://tracker.ceph.com/issues/58489
> <https://tracker.ceph.com/issues/58489>
> > <https://tracker.ceph.com/issues/58489
> <https://tracker.ceph.com/issues/58489>> so I
> > was very relieved when 17.2.6 was released and started to
> update
> > immediately.
> >
> > But now I'm stuck again with my broken MDS. MDS won't get
> into
> > up:active
> > without the update but the update waits for them to get into
> > up:active
> > state. Seems like a deadlock / chicken-egg problem to me.
> >
> > Since I'm still relatively new to Ceph, could you help me?
> >
> > What I see when watching the update status:
> >
> > {
> > "target_image":
> >
>
"quay.io/ceph/ceph@sha256:1161e35e4e02cf377c93b913ce78773f8413f5a8d7c5eaee4b4773a4f9dd6635
<http://quay.io/ceph/ceph@sha256:1161e35e4e02cf377c93b913ce78773f8413f5a8d7c5eaee4b4773a4f9dd6635>
<http://quay.io/ceph/ceph@sha256:1161e35e4e02cf377c93b913ce78773f8413f5a8d7c5eaee4b4773a4f9dd6635
<http://quay.io/ceph/ceph@sha256:1161e35e4e02cf377c93b913ce78773f8413f5a8d7c5eaee4b4773a4f9dd6635>>",
> > "in_progress": true,
> > "which": "Upgrading all daemon types on all
hosts",
> > "services_complete": [
> > "crash",
> > "mgr",
> > "mon",
> > "osd"
> > ],
> > "progress": "18/40 daemons upgraded",
> > "message": "Error: UPGRADE_OFFLINE_HOST:
Upgrade:
> Failed
> > to connect
> > to host ceph01 at addr (192.168.23.61)",
> > "is_paused": false
> > }
> >
> > (The offline host was one host that broke during the
> upgrade. I
> > fixed
> > that in the meantime and the update went on.)
> >
> > And in the log:
> >
> > 2023-04-10T19:23:48.750129+0000 mgr.ceph04.qaexpv [INF]
> Upgrade:
> > Waiting
> > for mds.mds01.ceph04.hcmvae to be up:active (currently
> up:replay)
> > 2023-04-10T19:23:58.758141+0000 mgr.ceph04.qaexpv [WRN]
> Upgrade:
> > No mds
> > is up; continuing upgrade procedure to poke things in the
> right
> > direction
> >
> >
> > Please give me a hint what I can do.
> >
> > Cheers,
> > Thomas
> > --
> > http://www.widhalm.or.at <http://www.widhalm.or.at>
> <http://www.widhalm.or.at <http://www.widhalm.or.at>>
> > GnuPG : 6265BAE6 , A84CB603
> > Threema: H7AV7D33
> > Telegram, Signal: widhalmt(a)widhalm.or.at
> <mailto:widhalmt@widhalm.or.at>
> > <mailto:widhalmt@widhalm.or.at
> <mailto:widhalmt@widhalm.or.at>>
> > _______________________________________________
> > ceph-users mailing list -- ceph-users(a)ceph.io
> <mailto:ceph-users@ceph.io>
> > <mailto:ceph-users@ceph.io
<mailto:ceph-users@ceph.io>>
> > To unsubscribe send an email to ceph-users-leave(a)ceph.io
> <mailto:ceph-users-leave@ceph.io>
> > <mailto:ceph-users-leave@ceph.io
> <mailto:ceph-users-leave@ceph.io>>
> >
>
5:24 a.m.
Sorry - the info about the insufficient space seems like it referred to
why the devices are not available. So that's just as is should be.
All MDS are still in error state and were refreshed 2d ago. Even right
after a mgr failover. So it seems, there's something else going on.
One thing that might be essential to debugging is that I have one MDS
(mds.mds01) but two CephFS pools (cephfs and cephfs-insecure). I can't
remember why I only had the one and not two. I used the dashboard for
setting it up and maybe I didn't realize that I needed two. Or do I? I'm
just reading through documentation and other sources.
On 4/10/23 23:02, Adam King wrote:
> It seems like it maybe didn't actually do the redeploy as it should log
> something saying it's actually doing it on top of the line saying it
> scheduled it. To confirm, the upgrade is paused ("ceph orch upgrade
> status" reports is_paused as false)? If so, maybe try doing a mgr
> failover ("ceph mgr fail") and then check "ceph orch ps" and
"ceph orch
> device ls" a few minutes later and look at the REFRESHED column. If any
> of those are giving amounts of time farther back then when you did the
> failover, there's probably something going on on the host(s) where it
> says it hasn't refreshed recently that's sticking things up (you'd have
> to go on that host and look for hanging cephadm commands). Lastly, you
> could look at the /var/lib/ceph/<fsid>/<mds-daemon-name>/unit.run file
> on the hosts where the mds daemons are deployed. The (very long) last
> podman/docker run line in that file should have the image name of the
> image the daemon is being deployed with. So you could use that to
> confirm if cephadm ever actually tried a redeploy of the mds with the
> new image. You could also check the journal logs for the mds.
> Cephadm reports the sytemd unit name for the daemon as part of "cephadm
> ls" output if you put a copy of the cephadm binary, un "cephadm ls"
with
> it, grab the systemd unit name for the mds daemon form that output, you
> could use that to check the journal logs which should tell the last
> restart time and why it's gone down.
>
> On Mon, Apr 10, 2023 at 4:25 PM Thomas Widhalm <widhalmt(a)widhalm.or.at
> <mailto:widhalmt@widhalm.or.at>> wrote:
>
> I did what you told me.
>
> I also see in the log, that the command went through:
>
> 2023-04-10T19:58:46.522477+0000 mgr.ceph04.qaexpv [INF] Schedule
> redeploy daemon mds.mds01.ceph06.rrxmks
> 2023-04-10T20:01:03.360559+0000 mgr.ceph04.qaexpv [INF] Schedule
> redeploy daemon mds.mds01.ceph05.pqxmvt
> 2023-04-10T20:01:21.787635+0000 mgr.ceph04.qaexpv [INF] Schedule
> redeploy daemon mds.mds01.ceph07.omdisd
>
>
> But the MDS never start. They stay in error state. I tried to redeploy
> and start them a few times. Even restarted one host where a MDS
> should run.
>
> mds.mds01.ceph03.xqwdjy ceph03 error 32m ago
> 2M - - <unknown> <unknown> <unknown>
> mds.mds01.ceph04.hcmvae ceph04 error 31m ago
> 2h - - <unknown> <unknown> <unknown>
> mds.mds01.ceph05.pqxmvt ceph05 error 32m ago
> 9M - - <unknown> <unknown> <unknown>
> mds.mds01.ceph06.rrxmks ceph06 error 32m ago
> 10w - - <unknown> <unknown> <unknown>
> mds.mds01.ceph07.omdisd ceph07 error 32m ago
> 2M - - <unknown> <unknown> <unknown>
>
>
> And other ideas? Or am I missing something.
>
> Cheers,
> Thomas
>
> On 10.04.23 21:53, Adam King wrote:
> > Will also note that the normal upgrade process scales down the mds
> > service to have only 1 mds per fs before upgrading it, so maybe
> > something you'd want to do as well if the upgrade didn't do it
> already.
> > It does so by setting the max_mds to 1 for the fs.
> >
> > On Mon, Apr 10, 2023 at 3:51 PM Adam King <adking(a)redhat.com
> <mailto:adking@redhat.com>
> > <mailto:adking@redhat.com <mailto:adking@redhat.com>>>
wrote:
> >
> > You could try pausing the upgrade and manually "upgrading"
> the mds
> > daemons by redeploying them on the new image. Something like
> "ceph
> > orch daemon redeploy <mds-daemon-name> --image <17.2.6
image>"
> > (daemon names should match those in "ceph orch ps" output).
> If you
> > do that for all of them and then get them into an up state you
> > should be able to resume the upgrade and have it complete.
> >
> > On Mon, Apr 10, 2023 at 3:25 PM Thomas Widhalm
> > <widhalmt(a)widhalm.or.at <mailto:widhalmt@widhalm.or.at>
> <mailto:widhalmt@widhalm.or.at <mailto:widhalmt@widhalm.or.at>>>
wrote:
> >
> > Hi,
> >
> > If you remember, I hit bug
> https://tracker.ceph.com/issues/58489
> <https://tracker.ceph.com/issues/58489>
> > <https://tracker.ceph.com/issues/58489
> <https://tracker.ceph.com/issues/58489>> so I
> > was very relieved when 17.2.6 was released and started to
> update
> > immediately.
> >
> > But now I'm stuck again with my broken MDS. MDS won't get
> into
> > up:active
> > without the update but the update waits for them to get into
> > up:active
> > state. Seems like a deadlock / chicken-egg problem to me.
> >
> > Since I'm still relatively new to Ceph, could you help me?
> >
> > What I see when watching the update status:
> >
> > {
> > "target_image":
> >
>
"quay.io/ceph/ceph@sha256:1161e35e4e02cf377c93b913ce78773f8413f5a8d7c5eaee4b4773a4f9dd6635
<http://quay.io/ceph/ceph@sha256:1161e35e4e02cf377c93b913ce78773f8413f5a8d7c5eaee4b4773a4f9dd6635>
<http://quay.io/ceph/ceph@sha256:1161e35e4e02cf377c93b913ce78773f8413f5a8d7c5eaee4b4773a4f9dd6635
<http://quay.io/ceph/ceph@sha256:1161e35e4e02cf377c93b913ce78773f8413f5a8d7c5eaee4b4773a4f9dd6635>>",
> > "in_progress": true,
> > "which": "Upgrading all daemon types on all
hosts",
> > "services_complete": [
> > "crash",
> > "mgr",
> > "mon",
> > "osd"
> > ],
> > "progress": "18/40 daemons upgraded",
> > "message": "Error: UPGRADE_OFFLINE_HOST:
Upgrade:
> Failed
> > to connect
> > to host ceph01 at addr (192.168.23.61)",
> > "is_paused": false
> > }
> >
> > (The offline host was one host that broke during the
> upgrade. I
> > fixed
> > that in the meantime and the update went on.)
> >
> > And in the log:
> >
> > 2023-04-10T19:23:48.750129+0000 mgr.ceph04.qaexpv [INF]
> Upgrade:
> > Waiting
> > for mds.mds01.ceph04.hcmvae to be up:active (currently
> up:replay)
> > 2023-04-10T19:23:58.758141+0000 mgr.ceph04.qaexpv [WRN]
> Upgrade:
> > No mds
> > is up; continuing upgrade procedure to poke things in the
> right
> > direction
> >
> >
> > Please give me a hint what I can do.
> >
> > Cheers,
> > Thomas
> > --
> > http://www.widhalm.or.at <http://www.widhalm.or.at>
> <http://www.widhalm.or.at <http://www.widhalm.or.at>>
> > GnuPG : 6265BAE6 , A84CB603
> > Threema: H7AV7D33
> > Telegram, Signal: widhalmt(a)widhalm.or.at
> <mailto:widhalmt@widhalm.or.at>
> > <mailto:widhalmt@widhalm.or.at
> <mailto:widhalmt@widhalm.or.at>>
> > _______________________________________________
> > ceph-users mailing list -- ceph-users(a)ceph.io
> <mailto:ceph-users@ceph.io>
> > <mailto:ceph-users@ceph.io
<mailto:ceph-users@ceph.io>>
> > To unsubscribe send an email to ceph-users-leave(a)ceph.io
> <mailto:ceph-users-leave@ceph.io>
> > <mailto:ceph-users-leave@ceph.io
> <mailto:ceph-users-leave@ceph.io>>
> >
>
411
days inactive
413
days old
9 comments
3 participants
participants (3)
-
Adam King -
Thomas Widhalm -
Xiubo Li