Feature #57180
option for pg_autoscaler to retain same state of existing pools when upgrading to versions >= pacific
0%
Description
Currently, any version of Ceph that is >= Pacific will have autoscaler enabled by default even for existing pools.
We should have an option for the autoscaler to just retain its state before the upgrade.
History
#1 Updated by Kamoltat (Junior) Sirivadhna about 1 year ago
Update:
Since all the upgrade suites in Pacific turn the autoscaler off by default. I had to write a new upgrade test to try and reproduce this particular issue, https://github.com/ceph/ceph/pull/47677.
The result from a run I scheduled in Teuthology, /a/ksirivad-2022-08-19_22:01:49-upgrade:nautilus-x:autoscaler-pacific-distro-default-smithi/6981635/, suggests that the autoscaler does indeed retain its state when upgrading from Nautilus to Pacific.
2022-08-19T22:18:29.453 INFO:tasks.workunit.client.0.smithi161.stdout:
Before the upgrade, the current version: Nautilus
POOL AUTOSCALE
test1 warn
test2 warn
rbd off
test3 off
2022-08-19T22:22:43.458 INFO:tasks.workunit.client.0.smithi161.stdout:
After the upgrade, the current version: Pacific
POOL AUTOSCALE
rbd off
test1 warn
test2 warn
test3 off
device_health_metrics off
#2 Updated by Vikhyat Umrao about 1 year ago
We discussed this one and the issue is in the following rolling upgrade playbook -
The following playbook tasks should check the pg autoscaler state before disabling them.
- name: disable pg autoscale on pools
- name: re-enable pg autoscale on pools
#3 Updated by Vikhyat Umrao about 1 year ago
- Status changed from New to Rejected
#4 Updated by Vikhyat Umrao about 1 year ago
Downstream ceph-ansible BZ - https://bugzilla.redhat.com/show_bug.cgi?id=2121097