Ceph » mgr

Modules should try not to be too sensitive to mgr daemon restarts -- remember that these will also happen during e.g. upgrades.

The current behaviour of restarting on module enable/disable is just for simplicity/robustness. If it does seem worthwhile to implement dynamic activation of modules without a restart, then I would recommend still to keep the restart behaviour when a module is *dis*abled -- I'd rather not be worrying about how to handle a buggy module that e.g. doesn't shut down promptly.

Thanks for the feedback, John. Paul Cuzner mentioned this behaviour today and I just wanted to have it recorded for any future action.

Note: there is an open pull request that will change the dashboard authentication scheme to using JWT tokens, which supposedly would resolve the "session loss at mgr restart" problem for the dashboard. But this is not fixing the root cause that the mgr performs a restart of all modules every time a module is enabled or disabled.

This affects #24651, as there is no way to properly restart just the dashboard module.

Relates to https://github.com/ceph/ceph/pull/32051: We cannot (And will likely never be able to) properly unload modules.