CEPH Nearfull Limit

[BORNXenon]

I've read that the Nearfull limit on CEPH is 80%, but where do I check that limit?
My rbd currently says it has 77.58% usage, but the Performance metric on the CEPH heath screen says it is at 56% usage.

Which one do I worry about?

 

[Alwin]

See 'Used (%)' on the OSD tab. The near-full limit counts per OSD.

 

[BORNXenon]

See 'Used (%)' on the OSD tab. The near-full limit counts per OSD.

Thanks Alwin, makes sense.
I have several disks which are near the 80% but most are well under that. Will CEPH rebalance automagically, or is this somehing I need to initiate?

 

[Alwin]

I have several disks which are near the 80% but most are well under that. Will CEPH rebalance automagically, or is this somehing I need to initiate?

You do not have enough PGs on the cluster. Either use the online pgcalc or enable the autoscaler.
https://docs.ceph.com/en/latest/rados/operations/placement-groups/

 

[BORNXenon]

You do not have enough PGs on the cluster. Either use the online pgcalc or enable the autoscaler.
https://docs.ceph.com/en/latest/rados/operations/placement-groups/

Thanks again Alwin.
I've turned on the autoscaler in 'warn' mode and it doesn't tell me I should have more PGs.
From what the pgcalc tells me, I should have a pg_num setting of 512, I've read on this forum that I should increment the PGs gradually (i.e. change it to 256 and let it rebalance, then increase it again, is this the recommendation?

 

[Alwin]

From what the pgcalc tells me, I should have a pg_num setting of 512, I've read on this forum that I should increment the PGs gradually (i.e. change it to 256 and let it rebalance, then increase it again, is this the recommendation?

Do it at once, the creation and peering, as the data movement stays the same. Just that people sometimes prefer to run it during off hours.