- Shipped
- November 16, 2026 at 11:30 PM UTC
- Author
- Kamo
- Commit
- 509384c
BuildKit RUN steps (npm install / next build) landed in **************** outside cibuild.slice, so image builds on k1m1 stayed uncapped and starved etcd on sdc (rke2 restarted 193 times on 2026-11-16). - dockerd now runs in cibuild.slice (docker.service Slice= drop-in), so BuildKit's relative cgroups nest under the caps - the CI limits Job targets k3m1 (no etcd); the slice resolves the docker disk per host (@CI_DISK@); limits 20 cores, 24/30G, 80M + 400/200 IOPS - apply.sh enables and starts the runner on the build host - k1m1 runner decommissioned (registration moved to k3m1); docker prune now also runs on k3m1 from /etc/cron.d/ci-docker-prune
