You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
In our pod handlers we have code that checks if a pod is scheduled. If
the pod is not scheduled, then we do not add the pod to our retry
framework. However, there are times where we automatically enqueue all
pods on a node or across the cluster. This can be from when a new node
comes up (pre-IC) and need to add all pods on that node. Another use
case is when a new UDN/NAD is created and the controller spins up...we
add all pods then.
We shouldn't be queuing pods to the retry framework that are not
scheduled. It's a waste of operations. However, even if we do enqueue
them, we definitely should not be treating a non-scheduled resource as
an error and retrying it again later.
This commit changes the retry framework to detect the above case, and
log an error. It does not trigger retrying of the resource, which may
perpetually fail and then then cause OVNKubernetesResourceRetryFailure.
Signed-off-by: Tim Rozet <[email protected]>
0 commit comments