Ceph » RADOS

Pool stats like num_objects, num_bytes, increased after PGs were merged after manual set pg_num.

Steps to reproduce (nautilus):

# run cluster with vstart.sh
./bin/ceph osd pool create test 8 8
./bin/rados bench -p test 2 write --no-cleanup
# wait until stats synced
./bin/ceph df
./bin/ceph osd pool set default.rgw.buckets.data pg_num 4
# wait until pgs merged
./bin/ceph df

Also you can compare pool stats and sum of pg stats from pg dump.

# test pool has id=5
$ ./bin/ceph pg dump --format=json | jq '(.pg_map.pool_stats[] | select(.poolid == 5) | .stat_sum.num_objects ) - ([.pg_map.pg_stats[] | select(.pgid | startswith("5.")) | .stat_sum.num_objects] | add)'

This stats inconsistency disappear after mgr restart.

Reproduced on 14.2.8 and master (with pg_autoscale_mode off).