Ceph

The fix for tracker #5332 was not correct. Even though the monitor code passes a vector to PGMonitor::dump_stuck_pg_stats() that function only looks at the first element. The fix is either to support multiple elements or switch to a single string and set n=1 in the command interface. This would also be a usage change.

Keep in mind that the type of "StuckPG" is an enum. We could create a bit mask of PG_STATE_* values from the vector of stuck type names. If get_stuck_stats() were passed a mask it could have to check for all matching states and assuming any one was over the threshold count include it. When specifying multiple states, a pg returned doesn't tell you which of the 1 or more states are over the threshold.