Ceph » CephFS

This is coming from the folks in the field - apparently it helps to know early on before the MDS starts throwing up cache and cap related warnings that (some) clients are hammering the MDS hard taking it close to its running capacity.

The ask is to provide a mechanism - in the form of either

- a mgr module that collects perf dumps and scans over the collected set to assess if some clients are spiking the load often
- have the mds warn on certain counters rising too fast (say, when it notices a sharp increase in caps issued to a client)

It might also help to dump/warn additional stats such as operation type that is contributing to the load spike, e.g., setxattr for recursive selinux relabels, unlink (for large directory removals), etc..

This is an umbrella ticket - need to create specific subtasks once the deliverable is finalised.