Patch release fixing a silent prod regression discovered while dogfooding v2 against a real customer (Trux).
The bug
SkaleData's helm chart for Airflow configures defaultAirflowTag: current + imagePullPolicy: Always. Pods always re-pull :current on rolling restart. But v2.0.0 (and v1 before it) only pushed :<sha> and :latest. So :current got set once on first deploy (via SkaleData's provision_app helm upgrade) and never updated again. Every subsequent customer build pushed new image SHAs into ECR but pods kept loading the original first-ever-deploy image. Result: silent invisible deploys.
The fix
action.yml now also tags :current and pushes it alongside :<sha> and :latest. One-line behavioral change, no input/output schema changes.
+ CURRENT="${ar_url}/${app-type}:current"
docker build \
-t "$IMAGE" \
-t "$LATEST" \
+ -t "$CURRENT" \
...
docker push "$IMAGE"
docker push "$LATEST"
+ docker push "$CURRENT"Upgrade
Pinned to @v2.0.1 or using @v2? No change needed — the floating v2 tag now points at this commit. Your next deploy will start updating :current automatically.
If you've been affected by this regression (any commit post your first deploy was invisible in prod), once you bump to v2.0.1, the next push will rebuild the image AND finally move :current to point at it.
Detected by
Trux team's tdp-airflow EKS migration. PR: #2.