osd: Calculate degraded and misplaced more accurately

[dzafman]

Calculate num_object_copies based on the larger of pool size,
up set size and acting set size.

Calculate num_objects_degraded as the difference between num_object_copies
and all copies found on acting set and backfilling up set OSDs.

Calculate num_objects_misplaced as all copies on acting set OSDs not in up set
less copies that have been backfilled to up set OSDs.

Fixes: http://tracker.ceph.com/issues/18619

[liewegas]

YES!

I'm confused about one thing, though.. when is acting.size() or up.size() larger than the pool size?

[dzafman]

@liewegas I was trying to handle the case that pool size is larger than both up.size() and acting.size(). Maybe more to you question, if you reduce pool size your acting.size() can be larger than pool size. Or maybe it won't do that because it won't need to remap? Does that matter in terms of calculations?

[dzafman]

Rados run: http://pulpito.ceph.com/dzafman-2017-01-20_12:26:00-rados-wip-calc-stats-distro-basic-smithi/

2 DEAD: Infrastructure

1 FAIL:
-81> 2017-01-20 22:51:50.166044 7fa5f4702700 10 log_client log_queue is 8 last_log 378 sent 369 num 8 unsent 9 sending 9
0> 2017-01-20 22:51:50.166949 7fa5f4702700 -1 /build/ceph-11.1.0-6744-g8423bc4/src/common/LogClient.cc: In function 'Message* LogClient::_get_mon_log_message()' thread 7fa5f4702700 time 2017-01-20 22:51:50.166046
/build/ceph-11.1.0-6744-g8423bc4/src/common/LogClient.cc: 310: FAILED assert(num_unsent <= log_queue.size())

[liewegas]

I think that if the pool size decreases, acting will also shrink. We certainly have to handle the case where it's smaller than the pool size, though. It's just a little confusing as written, but fine!