Alert rule: CephFilesystemInsufficientStandby

Please consider opening a PR to improve this runbook if you gain new information about causes of the alert, or how to debug or resolve the alert. Click "Edit this Page" in the top right corner to create a PR directly on GitHub.

Overview

The minimum number of standby daemons required by standby_count_wanted is less than the current number of standby daemons. Adjust the standby count or increase the number of MDS daemons.

Steps for debugging

Check CephFS MDS pods

$ ceph_cluster_ns=syn-rook-ceph-cluster
$ kubectl -n "${ceph_cluster_ns}" get pods -lapp=rook-ceph-mds

Check CephFS status

$ ceph_cluster_ns=syn-rook-ceph-cluster
$ kubectl -n "${ceph_cluster_ns}" exec -it deploy/rook-ceph-tools -- ceph fs status
fspool - 0 clients (1)
======
RANK      STATE         MDS        ACTIVITY     DNS    INOS   DIRS   CAPS (2)
 0        active      fspool-a  Reqs:    0 /s  1237    308    307      0  (2)
0-s   standby-replay  fspool-b  Evts:    0 /s  1601    306    305      0  (2)
      POOL         TYPE     USED  AVAIL (3)
fspool-metadata  metadata  55.0M  83.8G (3)
  fspool-data0     data       0   83.8G (3)
MDS version: ceph version 16.2.6 (ee28fb57e47e9f88813e24bbf4c14496ca299d31) pacific (stable) (4)
1 Name of the filesystem and amount of connected clients
2 List of MDS replicas. See the Ceph documentation on MDS states for more information about the values shown in the STATE column.
3 Filesystem metadata and data usage and remaining capacity.
4 Metadata server Ceph version

If a Rook-managed Ceph upgrade goes wrong and leaves one of the MDS replicas in CrashLoopBackOff for too long the MDS cluster can go into state failed. In that case, you should be able to recover the MDS cluster with the following commands:

These commands may make the CephFS filesystem temporarily unavailable.

  1. First, get the filesystem into a state where you can promote one of the MDS replicas to active

    $ ceph_cluster_ns=syn-rook-ceph-cluster
$ cephfs_name=fspool (1)
$ kubectl -n "${ceph_cluster_ns}" exec -it deploy/rook-ceph-tools -- ceph fs dump (2)
[ ... truncated ... ]
Filesystem 'fspool' (1)
fs_name	fspool
[ ... truncated ... ]
max_mds	1 (2)
in	0
up	{0=23899}
[ ... truncated ... ]
$ kubectl -n "${ceph_cluster_ns}" exec -it deploy/rook-ceph-tools -- ceph fs set "${cephfs_name}" allow_standby_replay false (3)
$ kubectl -n "${ceph_cluster_ns}" exec -it deploy/rook-ceph-tools -- ceph fs set "${cephfs_name}" max_mds 1 (4)
    1 Change to the name of the filesystem you want to recover
    2 Show filesystem status, and make a note of max_mds for your target filesystem.
    3 Disable standby-replay MDS replicas. This allows safely restarting the MDS cluster.
    4 This is only required if dump showed max_mds > 1

  2. Check if one of the MDS replicas got promoted to active by the MONs.

    $ kubectl -n "${ceph_cluster_ns}" exec -it deploy/rook-ceph-tools -- ceph fs status

    If not, scale down all but one of the filesystem’s MDS replicas. You may have to temporarily disable the rook-ceph operator to be able to scale down MDS replicas. To disable the rook-ceph operator, you need to disable auto sync for the root and rook-ceph apps in ArgoCD and then scale the operator deployment to 0 replicas.

  3. Once the filesystem is back to state active, enable standby replicas again

    $ kubectl -n "${ceph_cluster_ns}" exec -it deploy/rook-ceph-tools -- ceph fs set "${cephfs_name}" allow_standby_replay true (1)
$ kubectl -n "${ceph_cluster_ns}" exec -it deploy/rook-ceph-tools -- ceph fs set "${cephfs_name}" max_mds <orig_max_mds> (2)
    1 Enable standby-replay MDS replicas for the filesystem
    2 If your filesystem originally had max_mds > 1, reconfigure it to have the original value of max_mds.

Upstream documentation

docs.ceph.com/en/latest/cephfs/health-messages/#mds-insufficient-standby