Hi Venky,
The recently crashed daemon is likely the MDS which you mentioned in
your subsequent email.
The "recently crashed daemon" was the osd.51 daemon which was in the
metadata pool.
But yes, in the process of trying to get the system running, I probably
did a few steps that were unnecessary. The steps generally moved me in
the right direction until I got to MDS state "up:rejoin" where things
paused, then got much worse.
Now I'm certainly in the phase of monkeying around trying desperately to
get a heartbeat out of the system and probably doing more damage than
good. If only I could ask the system "what are you actually trying to
do?" Scrolling through the source code doesn't help too much. My next
step will be to insert some useful debugging messages in the vicinity of
the error to extract more information. Failing on an assert() has
advantages, but also massive disadvantages when it comes to debugging.
Cheers,
Eric
On 05/12/2023 06:10, Venky Shankar wrote:
This email was sent to you by someone outside the University.
You should only click on links or attachments if you are certain that the email is genuine and the content is safe.
Hi Eric,
On Mon, Nov 27, 2023 at 8:00 PM Eric Tittley <Eric.Tittley@xxxxxxxx> wrote:
Hi all,
For about a week our CephFS has experienced issues with its MDS.
Currently the MDS is stuck in "up:rejoin"
Issues become apparent when simple commands like "mv foo bar/" hung.
I assume the MDS was active at this point in time where the command
hung. Would that be correct?
I unmounted CephFS offline on the clients, evicted those remaining, and then issued
ceph config set mds.0 mds_wipe_sessions true
ceph config set mds.1 mds_wipe_sessions true
which allowed me to delete the hung requests.
Most likely, the above steps weren't really required. The hung command
is possibly a deadlock in the MDS (during rename).
I've lost the exact commands I used, but something like
rados -p cephfs_metadata ls | grep mds
rados rm -p cephfs_metadata mds0_openfiles.0
etc
This allowed the MDS to get to "up:rejoin" where it has been stuck ever since which is getting on five days.
# ceph mds stat
cephfs:1/1 {0=cephfs.ceph00.uvlkrw=up:rejoin} 2 up:standby
root@ceph00:/var/log/ceph/a614303a-5eb5-11ed-b492-011f01e12c9a# ceph -s
cluster:
id: a614303a-5eb5-11ed-b492-011f01e12c9a
health: HEALTH_WARN
1 filesystem is degraded
1 pgs not deep-scrubbed in time
2 pool(s) do not have an application enabled
1 daemons have recently crashed
services:
mon: 3 daemons, quorum ceph00,ceph01,ceph02 (age 57m)
mgr: ceph01.lvdgyr(active, since 2h), standbys: ceph00.gpwpgs
mds: 1/1 daemons up, 2 standby
osd: 91 osds: 90 up (since 78m), 90 in (since 112m)
data:
volumes: 0/1 healthy, 1 recovering
pools: 5 pools, 1539 pgs
objects: 138.83M objects, 485 TiB
usage: 971 TiB used, 348 TiB / 1.3 PiB avail
pgs: 1527 active+clean
12 active+clean+scrubbing+deep
io:
client: 3.1 MiB/s rd, 3.16k op/s rd, 0 op/s wr
# ceph --version
ceph version 17.2.6 (d7ff0d10654d2280e08f1ab989c7cdf3064446a5) quincy (stable)
I've tried failing the MDS so it switches. Rebooted a couple of times.
I've added more OSDs to the metadata pool and took one out as I thought it might be a bad metadata OSD (The "recently crashed" daemon).
This isn't really going to do any good btw.
The recently crashed daemon is likely the MDS which you mentioned in
your subsequent email.
The error logs are full of
(prefix to all are:
Nov 27 14:02:44 ceph00 bash[2145]: debug 2023-11-27T14:02:44.619+0000 7f74e845e700 1 -- [v2:192.168.1.128:6800/2157301677,v1:192.168.1.128:6801/2157301677] --> [v2:192.168.1.133:6896/4289132926,v1:192.168.1.133:6897/4289132926]
)
crc :-1 s=READY pgs=12 cs=0 l=1 rev1=1 crypto rx=0 tx=0 comp rx=0 tx=0).send_message enqueueing message m=0x559be00adc00 type=42 osd_op(mds.0.36244:8142873 3.ff 3:ff5b34d6:::1.00000000:head [getxattr parent in=6b] snapc 0=[] ondisk+read+known_if_redirected+full_force+supports_pool_eio e32465) v8
crc :-1 s=READY pgs=12 cs=0 l=1 rev1=1 crypto rx=0 tx=0 comp rx=0 tx=0).write_message sending message m=0x559be00adc00 seq=8142643 osd_op(mds.0.36244:8142873 3.ff 3:ff5b34d6:::1.00000000:head [getxattr parent in=6b] snapc 0=[] ondisk+read+known_if_redirected+full_force+supports_pool_eio e32465) v8
crc :-1 s=THROTTLE_DONE pgs=12 cs=0 l=1 rev1=1 crypto rx=0 tx=0 comp rx=0 tx=0).handle_message got 154 + 0 + 30 byte message. envelope type=43 src osd.89 off 0
crc :-1 s=READ_MESSAGE_COMPLETE pgs=12 cs=0 l=1 rev1=1 crypto rx=0 tx=0 comp rx=0 tx=0).handle_message received message m=0x559be01f4480 seq=8142643 from=osd.89 type=43 osd_op_reply(8142873 1.00000000 [getxattr (30) out=30b] v0'0 uv560123 ondisk = 0) v8
osd_op_reply(8142873 1.00000000 [getxattr (30) out=30b] v0'0 uv560123 ondisk = 0) v8 ==== 154+0+30 (crc 0 0 0) 0x559be01f4480 con 0x559be00ad800
osd_op(unknown.0.36244:8142874 3.ff 3:ff5b34d6:::1.00000000:head [getxattr parent in=6b] snapc 0=[] ondisk+read+known_if_redirected+full_force+supports_pool_eio e32465) v8 -- 0x559be2caec00 con 0x559be00ad800
Repeating multiple times a second (and filling /var)
Prior to taking one of the cephfs_metadata OSDs offline, these came from communications from ceph00 to the node hosting the suspected bad OSD.
Now they are between ceph00 and the host of the replacement metadata OSD.
Does anyone have any suggestion on how to get the MDS to switch from "up:rejoin" to "up:active"?
Is there any way to debug this, to determine what issue really is? I'm unable to interpret the debug log.
Cheers,
Eric
________________________________________________________
Dr Eric Tittley
Research Computing Officer www.roe.ac.uk/~ert<http://www.roe.ac.uk/~ert>
Institute for Astronomy Royal Observatory, Edinburgh
The University of Edinburgh is a charitable body, registered in Scotland, with registration number SC005336. Is e buidheann carthannais a th’ ann an Oilthigh Dhùn Èideann, clàraichte an Alba, àireamh clàraidh SC005336.
_______________________________________________
ceph-users mailing list -- ceph-users@xxxxxxx
To unsubscribe send an email to ceph-users-leave@xxxxxxx
--
Cheers,
Venky
_______________________________________________
ceph-users mailing list -- ceph-users@xxxxxxx
To unsubscribe send an email to ceph-users-leave@xxxxxxx
