Enquiry about PG count increase

[Sean Png]

Hi,

We have done upgrading of our px-sg1 cluster from version 5.x to 6.x and the ceph from luminous to nubilous yesterday. Our Pool PG count is 1152, so ceph is showing health warning now. Please see the warning below,

======
1 pool(s) have non-power-of-two pg_numpool 'rbd-vm' pg_num 1152 is not a power of two
======

We have calculated the PG count on the basis of our OSDs ( now is 67 ) and foud the suggested count as 2048.
May I know if there will be any implications for proceeding with the PG count increase from 1152 to 2048? If there will be any implication when making 2048, please let us know if we can reduce the issues like I/O delay after changing it to a next near value of 1152 which is a power of two rather than 2048?

Please help to resolve this issue.
Thank You.

 

[aaron]

Check out the Ceph documentation on PGs [0]. It explains quite well how the number of PGs affects a disaster situation where Ceph needs to recover data.

The TL;DR is to rather have more PGs than less.


[0] https://docs.ceph.com/docs/nautilus/rados/operations/placement-groups/#placement-groups-tradeoffs