I noticed that DRA `PrepareDynamicResources` / `NodePrepareResources` failures appear not to be propagated as pod sync errors in kubelet. As a result, the retry seems to happen on the regular pod resync interval instead of the pod sync error backoff path.
I would like to understand whether this is intentional, and if so, what the design rationale is.
code path
When PrepareDynamicResources fails,the function returns early, but it does not appear to mark the PodSyncResult as failed, for example via either:
This matters because if result.Error() == nil, the kubelet will re-process the pod after the default sync interval (60s–90s). Otherwise, the pod is placed into the backoff queue and will be retried sooner. In our experience, the first few PrepareDynamicResources calls sometimes fail, forcing us to wait through several default sync periods, which significantly increases pod startup latency.