Date: 2026-09-04 UTC
AutoCoder deployment run 2426 built and rolled out backend revision 2d8506cf3241, but deploy_cronjobs failed before its first workflow step. Gitea task 12539 reported that gitea/runner-images:ubuntu-latest existed, then Docker container creation failed with No such image.
The long-lived act-runner caches image availability while the independent storage-maintenance sidecar ran docker image prune --all --force --filter until=168h. Docker evaluates the age of the image itself, so the tagged 1.63 GB runner image was removed after a job even though the runner process still considered it cached. The next queued job skipped the pull and failed at container creation.
The live gitea-act-runner-storage-guard ConfigMap now uses dangling-only pruning:
docker image prune --force --filter until=168h
The three general runner deployments were confirmed idle and restarted sequentially to load the script and clear their in-memory image caches. Rust runners were not restarted.
The repository-managed correction was merged through AutoCoder-Engine PR #2981 at main revision c9719c72e09adf69db7c4df28001cd0ac7a11ec4, with requirements checkpoint 1775a91a and implementation 93e9b42f.
drg-k8s-worker-01.auth.home.lan.docker image prune --all is absent.ubuntu-latest jobs on generic runner 18; both completed successfully.gitea/runner-images:ubuntu-latest at image ID c4e8f9bef973 and size 1.63 GB.