Mentioned in SAL (#wikimedia-cloud) [2021-04-27T10:48:55Z] <dcaro> ceph.eqiad: Tweaked the target_size_ratio of all the pools, enabling autoscaler (it will increase cinder pool only) (T273783)

Current autoscale setup (notice the target ratios and the new pg_num):

dcaro@cloudcephmon1001:~$ sudo ceph osd pool  autoscale-status
POOL                   SIZE TARGET SIZE RATE RAW CAPACITY  RATIO TARGET RATIO EFFECTIVE RATIO BIAS PG_NUM NEW PG_NUM AUTOSCALE
eqiad1-cinder         3329G              3.0       209.5T 0.0466       0.2000          0.2000  1.0    128       1024 on
eqiad1-glance-images 209.9G              3.0       209.5T 0.0029       0.1000          0.1000  1.0   1024            on
eqiad1-compute       46157G              3.0       209.5T 0.6454       0.7000          0.7000  1.0   4096            on

Mentioned in SAL (#wikimedia-cloud) [2021-04-27T10:51:42Z] <dcaro> ceph.eqiad: cinder pool got it's pg_num increased to 1024, re-shuffle started (T273783)

Done, everything rebalanced, now the balancer is able to do a better job, and the score (ceph balancer eval) went down
from 0.008 to 0.0047, moving the busiest osd from 77% to 70.83% (10% of space reclaimed! \o/).

It should have some positive effect also on the performance of ceph volumes.