https://issues.redhat.com/browse/OCPBUGS-15087 broke OKD

[cgwalters]

In https://issues.redhat.com/browse/OCPBUGS-15087 AKA openshift/machine-config-operator#3865 we did a giant rework of the systemd units; it is believed that this broke OKD.

[cgwalters]

Digging into this scos run we failed to provision control plane nodes at all.

Kubelet started up, but looks like it failed to register with the bootstrap? Logs are full of
Sun 2023-10-22 22:19:44 UTC ip-10-0-51-83 kubelet.service[2154]: E1022 22:19:44.582987 2154 kubelet_node_status.go:485] "Error getting the current node from lister" err="node \"ip-10-0-51-83\" not found"

[vrutkovs]

Tested this manually - seems on some nodes I need to do systemctl daemon-reload and restart kubelet service.

Also, in some installations nodes show up as ip-10-0-51-83 instead of ip-10-0-51-83.us-west-1.compute.internal - perhaps some hostname/afterburn service in not included in kubelet-dependencies.target?

[cgwalters]

Ah this may be related to https://issues.redhat.com/browse/OCPBUGS-20369 since it's AWS

[vrutkovs]

Right, that seems to be AWS-specific indeed - I don't see this happening on GCP and comments in OCPBUGS-20369 seem to be similar to what I'm seeing

[vrutkovs]

Yeah, the fix proposed for https://issues.redhat.com/browse/OCPBUGS-20369 seems to resolve it. Closing