New subject: forcing an osd down

3 Sep 2019

      "ceph osd down" will mark an OSD down once, but not shut it down. Hence, it will continue to send heartbeats and request to be marked up again after a couple of seconds. To keep it down, there are 2 ways:

- either set "ceph osd set noup",
- or actually shut the OSD down.

The first version will allow the OSD to keep running so you can talk to the daemon while it is marked "down" . Be aware that the OSD will be marked "out" after a while. You might need to mark it "in" manually when you are done with maintenance.

I believe with nautilus it is possible to set the noup flag on a specific OSD, which is much safer.

Best regards,

=================
Frank Schilder
AIT Risø Campus
Bygning 109, rum S14

________________________________________
From: ceph-users <ceph-users-bounces@lists.ceph.com> on behalf of solarflow99 <solarflow99@gmail.com>
Sent: 03 September 2019 19:40:59
To: Ceph Users
Subject: [ceph-users] forcing an osd down

I noticed this has happened before, this time I can't get it to stay down at all, it just keeps coming back up:

# ceph osd down osd.48
marked down osd.48.

# ceph osd tree |grep osd.48
48   3.64000         osd.48         down        0          1.00000

# ceph osd tree |grep osd.48
48   3.64000         osd.48           up        0          1.00000

health HEALTH_WARN
            2 pgs backfilling
            1 pgs degraded
            2 pgs stuck unclean
            recovery 18/164089686 objects degraded (0.000%)
            recovery 1467405/164089686 objects misplaced (0.894%)
     monmap e1: 3 mons at {0=192.168.4.10:6789/0,1=192.168.4.11:6789/0,2=192.168.4.12:6789/0<http://192.168.4.10:6789/0,1=192.168.4.11:6789/0,2=192.168.4.12:6789/0>}
            election epoch 210, quorum 0,1,2 0,1,2
     mdsmap e166: 1/1/1 up {0=0=up:active}, 2 up:standby
     osdmap e25733: 45 osds: 45 up, 44 in; 2 remapped pgs

Re: forcing an osd down

Frank Schilder

solarflow99

tags

participants (2)