Skip to content

Check-in — 2026-08-22 runner 2.336.0 rollout

TL;DR

The GitHub Actions fleet now uses runner 2.336.0 and image r5 across all six placements. General and Docker placements were drained, upgraded, and resumed without runner loss. Live image labels match the manifest version and SHA-256 pin.

Phase status

Item Status
general-a / general-b Upgraded and resumed
docker-a / docker-b Upgraded and resumed
lan-a / mgmt-a Pinned image verified
Late PR #317 runners Both general runners upgraded; Docker runner covered
GitHub App Client ID migration Done 2026-08-22

What shipped

  • General placement upgrades completed through fleet operations: general-b run 32568184667 and general-a run 32569604364.
  • Docker placements used the same guarded drain → upgrade → resume contract through the direct WSL-to-Windows SSH tunnel, avoiding repeated dispatcher startup overhead.
  • Every placement exposes image labels org.opencontainers.image.version=2.336.0 and app.github.runner.sha256=04cf0be1…505d5d.
  • PR #317 merged during the rollout. Its two general runners were handled as exact targets after the placement runs; its Docker runner was included in the Docker host pass.

Known gaps / drift

  • The dispatcher is safe but slow for placement rolls: each operation starts LAN validation, wakes the ephemeral management runner, and applies serially.
  • The WSL SSH tunnel config was missing gha-docker-a; the local operator config now includes 172.29.240.13.
  • GitHub App fleet workflows use client-id + RUNNER_FLEET_*_CLIENT_ID.

What remains

Who Work Priority
(done) GitHub App Client IDs + workflow client-id switch Done
Later Reduce safe dispatcher setup latency for placement operations P3 
flowchart LR
  manifest["fleet.yml<br/>2.336.0-r5"] --> drain[Graceful drain]
  drain --> general[General A/B]
  drain --> docker[Docker A/B]
  general --> verify[Version + SHA verification]
  docker --> verify
  lan[LAN] --> verify
  mgmt[Management] --> verify
  verify --> resume[All runners resumed]