Ceph » mgr

Bug Description¶

Occasionally, after a ceph-mgr restart, some mgr modules may become temporarily unavailable, causing their commands to fail. For example:

# ceph mgr fail; ceph orch host ls
Error ENOTSUP: Warning: due to ceph-mgr restart, some PG states may not be up to date
Module 'orchestrator' is not enabled/loaded (required by command 'orch host ls'): use `ceph mgr module enable orchestrator` to enable it

Who is affected?¶

Users who run commands dependent on ceph-mgr modules may experience this after running `ceph mgr fail`. Users may also experience this bug when a mgr daemon restarts during an upgrade.

Workaround¶

Turning off the balancer module reduces the time it takes to load the rest of the mgr modules. Users experiencing the error above may disable the balancer module temporarily with `ceph balancer off` and rerun the commands.

Related issues¶

Tracker ticket #68657 addresses an obvious bottleneck in the balancer module that exacerbated the above issue. With this fix, the frequency of the error was reduced.

Tracker ticket #68648 addresses the error code that gets returned and modifies it to return a more appropriate EAGAIN message to convey that the command may be retried.

Related BZ: https://bugzilla.redhat.com/show_bug.cgi?id=2314146