We have a non-zero number of customers using VM scale sets, AKS, or spot instances and they complain that their workload gets stuck for some amount of time and then magically continues.  Almost every time this happens, it is due to a client getting deprovisioned without properly unmounting.  This is a problem on the end-user's part, but it would be nice to not have to walk logs to identify that evictions are occurring.  Alerts and other useful actions can be built around this stat changing then.

This work item tracks creation of a Lustre stat that increments every time a client eviction occurs.

Patch to be sent shortly.