Repository navigation
fix: use IfNotPresent pull policy for Connect content-init container - #177
stevenolen wants to merge 2 commits into
Conversation
The Connect init-runtime init container on content pods was hardcoded to ImagePullPolicy: Always, so every content start made a live registry call to ghcr.io. A registry hiccup could then fail content starts cluster-wide with ImagePullBackOff even though the image was already cached on every node. The content-init tag is pinned to the Connect version, so switch it to IfNotPresent and add a unit test asserting the policy.
|
Claude finished @stevenolen's task in 39s —— View job Review
The change is correct and well-motivated. One suggestion (inline): The 3-line comment in No other issues. The test is minimal but sufficient — it asserts the right container name and pull policy, which is exactly what changed. |
| // The content-init image tag is pinned to the Connect version, so a | ||
| // cached copy is always correct. PullAlways forced a registry call on | ||
| // every content start, failing starts cluster-wide on registry hiccups. | ||
| ImagePullPolicy: corev1.PullIfNotPresent, |
There was a problem hiding this comment.
The WHY here is worth keeping, but the codebase convention is a single-line comment. Consider condensing:
| // The content-init image tag is pinned to the Connect version, so a | |
| // cached copy is always correct. PullAlways forced a registry call on | |
| // every content start, failing starts cluster-wide on registry hiccups. | |
| ImagePullPolicy: corev1.PullIfNotPresent, | |
| // PullIfNotPresent: tag is pinned to Connect version, so a cached copy is always correct. | |
| ImagePullPolicy: corev1.PullIfNotPresent, |
|
@claude again |
|
Claude finished @stevenolen's task in 27s —— View job Re-review
The suggestion was applied cleanly. // PullIfNotPresent: tag is pinned to Connect version, so a cached copy is always correct.
ImagePullPolicy: corev1.PullIfNotPresent,That matches the codebase convention. The test at |
The Connect
init-runtimeinit container on content pods was hardcoded toImagePullPolicy: Always, so every content start made a live registry call to ghcr.io. A registry hiccup could then fail content starts cluster-wide with ImagePullBackOff even though the image was already cached on every node. The content-init tag is pinned to the Connect version, so this switches it toIfNotPresentand adds a unit test asserting the policy.