Feature #20606: mds: improve usability of cluster rank manipulation and setting cluster up/down
MDSMonitor: add new command to shrink the cluster in an automated way
Deprecate `ceph fs set <fs_name> max_mds <max_mds>`. New `ceph fs set <fs_name> ranks <max_mds>` sets max_mds and begins stopping ranks with the same logic as `ceph fs set <fs_name> down`. This will rely on a new internal MDSMap bool field "shrink_to_fit". (Probably setting max_mds through the deprected command unsets shrink_to_fit. Setting <ranks> will turn it on. The user cannot change this field explicitly.)
#1 Updated by Douglas Fuller over 6 years ago
I think we should still call this max_mds to avoid confusion between the filesystem itself (the data) and its MDSs. I also think we should just print a message and go ahead and shrink the cluster, ignoring the past behavior. This behavior is more likely what users would currently expect, and I think it's extremely unlikely any users are relying on the old behavior.
#2 Updated by Patrick Donnelly over 6 years ago
Dropping the old behavior for decreasing max_mds (i.e. do nothing but set it) is okay with me. BTW, there should be an cluster log INFO message when the cluster is shrinking. See new guidelines here: https://github.com/ceph/ceph/pull/16292
I would encourage you to think of a consistent command structure for this and the rest of the issues so it's clearer to end users. Keeping `set max_mds` as a deprecated command is not intrusive. OTOH, making the concept of ranks clearer in the commands is valuable for reducing user confusion. i.e. which commands operate on file system ranks vs. which operate on named MDS daemons.