quincy: mgr/cephadm: make upgrade respect use_repo_digest #50898
Conversation
If the option is false, we should upgrade images based on whether their container image name matches, not whether the digest is the same or not. Fixes: https://tracker.ceph.com/issues/58698 Signed-off-by: Adam King <adking@redhat.com> (cherry picked from commit 78a73d1)
|
4 dead jobs are a general issue I'm seeing with the upgrade-with-workload task. It seems the upgrade completes but the workload test at the end never does for some reason. This started happening on multiple branches simultaneously. For example, it came up in the reef baseline run (https://pulpito.ceph.com/yuriw-2023-04-08_15:50:42-orch-reef-distro-default-smithi/7235797). Due to it showing up in multiple stable branches at once with different codebases, thinking it's likely not an actual bug within ceph and shouldn't block PR merging, although it will require investigation. Rerun of failed jobs: https://pulpito.ceph.com/adking-2023-04-08_15:01:19-orch:cephadm-wip-adk3-testing-2023-04-05-1604-quincy-distro-default-smithi/ After reruns, 2 failures. Both are same test that is failing because it expects some command to give a zero error code but give an error message and instead it's giving a nonzero return code. To clarify, the command is meant to fail (it's checking on a nonexistent cluster) it's just the test expects it to fail a different way. I think it's just an issue of the test being backported without the change to the actual function's returncode, and shouldn't block merging. |
backport tracker: https://tracker.ceph.com/issues/58997
backport of #50082
parent tracker: https://tracker.ceph.com/issues/58698
this backport was staged using ceph-backport.sh version 16.0.0.6848
find the latest version at https://github.com/ceph/ceph/blob/main/src/script/ceph-backport.sh