NVIDIA / nodewright / 41128597801

84%

Build Type

push

github

Committed by web-flow

Commit Message

fix(operator): report image pull failures with a condition (#867)

* fix(operator): report image pull failures with a condition

A package pod whose image cannot be pulled stays Pending with its container
Waiting on ErrImagePull/ImagePullBackOff, so no terminal-failure predicate
sees it and nothing surfaced the problem until the stage deadline.

The reconciler now sets an ImagePullFailed condition on the NodeWright, with
a Warning event on its first appearance, once a package pod has failed to
pull for two minutes. It is rebuilt from the pods on every pass and clears
when none is waiting on a failed pull. The clock starts at pod creation or
the last finished init container.

The pull is deliberately not recorded as a package or node failure: the
kubelet keeps retrying, and an erroring package entry is read elsewhere as
evidence the stage genuinely failed (Job failure classification, the config
update gate, UninstallFailed). The stage deadline still fails a pull that
never recovers.

Co-Authored-By: Claude Sonnet 5.5 <noreply@anthropic.com>
Signed-off-by: Steven Stolarski <sstolarski@nvidia.com>

* docs: document the ImagePullFailed condition

Co-Authored-By: Claude Sonnet 5.5 <noreply@anthropic.com>
Signed-off-by: Steven Stolarski <sstolarski@nvidia.com>

---------

Signed-off-by: Steven Stolarski <sstolarski@nvidia.com>
Co-authored-by: Claude Sonnet 5.5 <noreply@anthropic.com>

Coverage Stats

80 of 90 new or added lines in 2 files covered. (88.89%)

15 existing lines in 1 file now uncovered.

10292 of 12275 relevant lines covered (83.85%)

7.65 hits per line