Infra: upgrade swarm act_runner fleet gitea/runner:2.2.0 → 3.0.2 #915

Closed
opened 2026-08-05 13:58:41 +00:00 by spikerj · 1 comment
Owner

The 7 swarm runners (4090 ×4, laptop-server, DREAMSTREAM5/6) run gitea/runner:2.2.0; the new Mac runners (m5-builder + m5-runner-1..4) run v3.0.2. Version skew is confirmed harmless today but the fleet should converge on 3.0.2:

  • gitea/runner:3.0.2 is published multi-arch (amd64 + arm64 — Jetsons covered; verified via docker manifest inspect).
  • Compatibility with the Gitea 1.27.0 server is empirically proven — the five v3.0.2 Mac runners already execute jobs against it.

Plan: bump the 7 image pins in spikersoft-infrastructure/gitea-act-runner/docker-stack.yml, update the RUNNER IMAGE comment block, fix the stale gitea/act_runner:nightly mirror line in docker/upstream-images.txt, canary laptop-server via docker service update, then full docker stack deploy as gitea-runners (.runner identity persists — no re-registration). Gated on the m5 arm-runner docker-build canary so only one variable changes at a time.

Post-upgrade smoke: one angular main.yml run (continue-on-error e2e-smoke aggregation) + one gpu-amd64 job.

The 7 swarm runners (4090 ×4, laptop-server, DREAMSTREAM5/6) run `gitea/runner:2.2.0`; the new Mac runners (m5-builder + m5-runner-1..4) run v3.0.2. Version skew is confirmed harmless today but the fleet should converge on 3.0.2: - `gitea/runner:3.0.2` is published multi-arch (amd64 + arm64 — Jetsons covered; verified via `docker manifest inspect`). - Compatibility with the Gitea 1.27.0 server is empirically proven — the five v3.0.2 Mac runners already execute jobs against it. Plan: bump the 7 image pins in `spikersoft-infrastructure/gitea-act-runner/docker-stack.yml`, update the RUNNER IMAGE comment block, fix the stale `gitea/act_runner:nightly` mirror line in `docker/upstream-images.txt`, canary `laptop-server` via `docker service update`, then full `docker stack deploy` as `gitea-runners` (`.runner` identity persists — no re-registration). Gated on the m5 arm-runner docker-build canary so only one variable changes at a time. Post-upgrade smoke: one angular `main.yml` run (continue-on-error e2e-smoke aggregation) + one `gpu-amd64` job.
Author
Owner

Done. Infra PR #157 merged; rolled out canary-first (laptop-server via docker service update, verified v3.0.2 ... declare successfully), then full docker stack deploy as gitea-runners. All 7 swarm runners now report v3.0.2 with correct labels; post-upgrade smokes green on the new version (angular run 19411 incl. SonarQube on 4090-3, android run 19425 on 4090-2). All 12 fleet runners online. /mnt/infrastructure self-updates on the next CI deploy (#584 behavior). Closing.

Done. Infra PR #157 merged; rolled out canary-first (laptop-server via `docker service update`, verified `v3.0.2 ... declare successfully`), then full `docker stack deploy` as `gitea-runners`. All 7 swarm runners now report v3.0.2 with correct labels; post-upgrade smokes green on the new version (angular run 19411 incl. SonarQube on 4090-3, android run 19425 on 4090-2). All 12 fleet runners online. `/mnt/infrastructure` self-updates on the next CI deploy (#584 behavior). Closing.
Sign in to join this conversation.