On 1/08/24 12:56, Mark Kirkwood wrote:
This 2nd post is about the next type of flapping osds we encountered
after upgrading. We started to see osds going down with this in 'ceph
-w':
2024-08-01 12:02:57.437135 mon.cat-hlz-stor001 [INF] osd.479 marked
down after no beacon for 902.637005 seconds
2024-08-01 12:02:57.468372 mon.cat-hlz-stor001 [WRN] Health check
failed: 1 osds down (OSD_DOWN)
We have the beacon interval set to 300. To fix this we tried:
- restarting osds
- restarting mons
- ntp tidyup
- restarting mgrs
However it is still happening. Poking around in the osd and mon logs
we did see some lines that hinted that the mon might be listening for
beacons using v1 - which could be broken (see part 1). Hence
restarting them again. This did not have any effect.
Apart from enabling the v2 msgr we have not altered our Luminous
config for Nautilus, are we missing something?
Regards
Mark
We noticed that collectd was running 'ceph perf dump' regularly on the
Ceph hosts and it appears we are seeing a version of this bug:
https://tracker.ceph.com/issues/25211
We've stopped collectd doing this and so far no flapping osds and no 'no
beacon for...' log messages.
Looking at the Nautilus branch code, it appears to have the fix for
issue 25211, but we'll dig a bit more.
Regards
Mark
_______________________________________________
ceph-users mailing list -- ceph-users@xxxxxxx
To unsubscribe send an email to ceph-users-leave@xxxxxxx
