acoustic-carpenter-78188
05/11/2026, 7:24 PMImagePullBackOff waiting reason. Today, once the ImagePullBackoffGracePeriod elapses, DemystifyPending returns PhaseInfoRetryableFailureWithCleanup, which is classified as a USER error and consumes the user's retry budget.
429s from the registry are an infrastructure problem outside the user's control — typically transient throttling on the registry side. Misclassifying them as user errors means a transient registry rate-limit can exhaust a task's user retries even though the user did nothing wrong.
Example real-world pod waiting message:
Back-off pulling image "us-central1-docker.pkg.dev/.../internal-gnina:...": ErrImagePull: failed to pull and unpack image "...": failed to copy: httpReadSeeker: failed open: unexpected status code https://us-central1-docker.pkg.dev/v2/...: 429 Too Many Requests
What changes were proposed in this pull request?
• `pod_helper.go`: in classifyWaitingContainer, after the ImagePullBackOff grace period, detect the 429 Too Many Requests substring in the waiting message and return PhaseInfoSystemRetryableFailureWithCleanup instead of PhaseInfoRetryableFailureWithCleanup.
• Add small helper isRegistryRateLimited(message string) bool.
• All other ImagePullBackOff messages keep their existing user-error classification.
How was this patch tested?
• New unit test TestDemystifyPending/ImagePullBackOffOutsideGracePeriod_RegistryRateLimited exercises a realistic 429 waiting message and asserts core.ExecutionError_SYSTEM.
• Existing ImagePullBackOffOutsideGracePeriod (non-429) test continues to pass, confirming the default classification is unchanged.
• go test ./go/tasks/pluginmachinery/flytek8s/... passes locally.
Labels
• fixed
Setup process
N/A
Screenshots
N/A
Check all the applicable boxes
• I updated the documentation accordingly.
• All new and existing tests passed.
• All commits are signed-off.
Related PRs
N/A
Docs link
N/A
flyteorg/flyteacoustic-carpenter-78188
05/21/2026, 8:18 PM