Feature #56670
openceph orch upgrade should set noout, nodeep-scrub and noscrub and unset when upgrade will complete
0%
Description
- ceph orch upgrade should set noout, nodeep-scrub, and noscrub and unset when the upgrade will complete
- This was the case when we used to use ceph-ansible
- This is a kind of feature parity b/w ceph-ansible and cephadm
- This feature can be designed as optional with default as True so if some users/admins do not want then can set it to false.
Benefits:
1. Less load from scrubbing during the upgrade when we expect to have recovery in the cluster
2. If an OSD is taking longer to reboot -> boot due to different issues1 or slow boot
[1] For example, PG dups issue - https://tracker.ceph.com/issues/53729 - it takes approx 7 to 8 minutes for an NVMe OSD to boot with 50M dups and approx 12-15 minutes for hybrid HDD OSDs
and if an OSD takes more than 10 minutes the Monitor marks the down OSD out and we will have backfill/recovery in the cluster when the upgrade is running and we do not want that
There can be multiple examples hence running the upgrade with the following flags is recommended:
noscrub
nodeep-scrub
noout
Updated by Vikhyat Umrao almost 2 years ago
I think we should also have this for rook I have created this feature issue in rook - https://github.com/rook/rook/issues/10619.
Updated by Vikhyat Umrao almost 2 years ago
Downstream Bug - https://bugzilla.redhat.com/show_bug.cgi?id=2111224
Updated by Adam King about 1 year ago
- Status changed from New to In Progress
- Assignee set to Adam King
- Pull request ID set to 50508