Disable "pg_autoscaler" failed.

[zhucan]

[BlaineEXE]

What version of Ceph is this?

[zhucan]

16.2.7

[travisn]

@zhucan Please include repro steps. Did you set in the cluster.yaml the pg_autoscaler to disabled: false? Or what did you set? And what version of rook?

[zhucan]

Only set in the cluster.yaml the pg_autoscaler to disabled: false. and kubectl apply -f cluter.yaml. the version of rook is 1.8.5. @travisn

[travisn]

@zhucan I was able to repro this as well. When running the disable command in the toolbox, there are some more details:

[rook@rook-ceph-tools-d6d7c985c-lqpl6 /]$ ceph mgr module disable pg_autoscaler   
Error EINVAL: module 'pg_autoscaler' cannot be disabled (always-on)

So it appears ceph does not anymore allow disabling the autoscaler.

[zhucan]

@travisn Thanks

[zhucan]

@travisn Is possible support to create pool with specail pg and configurable model of pg autoscale?

[travisn]

Yes, looks like the PGs can be controlled per-pool as seen in the Ceph PG docs.

The autoscaler is generally considered more helpful than negative, but if someone needs granular control then support can certainly be added to the pool CR.

[zhucan]

If we want to set to pool of rgw，we need to discuss it more deep

[travisn]

The pools in the object store use the same spec as the CephBlockPool, so if we make a change to the CephBlockPool, the pools in the CephObjectStore will inherit those settings.

[zhucan]

so we need to create a issue to releated it?

[travisn]

yes sounds good to create an issue

[zhucan]

There are some potential negative effects to be aware of when using Ceph PG autoscaling:

1. Increased latency: Autoscaling can result in increased latency if it takes time for Ceph to rebalance the data across the cluster.
2. Increased network traffic: Autoscaling can result in increased network traffic as data is rebalanced across the cluster, which can impact the performance of other applications on the same network.
3. Increased storage overhead: Autoscaling can result in increased storage overhead, as additional placement groups are created to store data, which can lead to increased storage costs.
4. Complexity: Autoscaling can add complexity to Ceph's architecture, as it requires the use of additional tools and services to monitor and manage the placement groups.
5. Management: Autoscaling can be difficult to manage in Ceph, especially if there are multiple clusters with different usage patterns.

[cosminapo203]

dose anyone have any idea about this fail:
[root@controller0 ceph(keystone_admin)]# ceph -v
ceph version 14.2.22 (ca74598065096e6fcbd8433c8779a2be0c889351) nautilus (stable)
[root@controller0 ceph(keystone_admin)]# ceph mgr module enable pg_autoscaler
Error ENOENT: all mgr daemons do not support module 'pg_autoscaler', pass --force to force enablement
[root@controller0 ceph(keystone_admin)]# ceph -s
cluster:
id: 5037a183-fac1-4f8d-9e3a-1223914b0d47
health: HEALTH_WARN
mon is allowing insecure global_id reclaim

services:
mon: 1 daemons, quorum controller0 (age 14m)
mgr: no daemons active
osd: 0 osds: 0 up, 0 in

data:
pools: 0 pools, 0 pgs
objects: 0 objects, 0 B
usage: 0 B used, 0 B / 0 B avail
pgs:

is there any guide on that

[zhucan]

@cosminapo203 please pasted the results of cmd "ceph mgr module ls"?

[zhucan]

and then run "ceph mgr module enable pg_autoscaler --force"