Ceph » rgw

Hi,
I've been testing upgrade of our test cluster from jewel to luminous. It's a small (3 hosts, 6 osds) cluster with nothing fancy (no multisite).

Now that #21196 seems to be fixed, I've noticed wastly increased execution times for 'radosgw-admin usage show'.
This is from our production cluster (on Jewel):

$ time /usr/bin/radosgw-admin usage show --start-date="2017-11-04 06:00:00" --end-date="2017-11-04 07:00:00"

real 0m4.964s
user 0m0.102s
sys 0m0.041s

This is from test cluster (on Luminous).
$ time /usr/bin/radosgw-admin usage show --start-date="2017-11-04 06:00:00" --end-date="2017-11-04 07:00:00"

real 0m34.621s
user 0m1.702s
sys 0m1.122s

And this is after all usage data (should) have been trimmed with "radosgw-admin usage trim --yes-i-really-mean-it".
It executes fast if I add --uid to is, however it's hardly a solution for us, because we'd need to query several thousand users.

I've put some debug level and rerun radosgw-admin with it. Logs attached. If you need anything more, just let me know.