Onboard izzywdev/MendysRobotics to the shared ARC self-hosted runner (scale set: mendysrobotics)

[izzywdev]

@claude — MendysRobotics needs to run its CI on the shared ARC runner (we keep hitting the hosted-Actions minutes cap). We've done the repo-side half; the remaining steps need the cluster + the GitHub App, which live on your side. Per runners/arc/ONBOARD-REPO.md, ArgoCD doesn't manage other repos' scale sets, so the register-repo helm install is the correct mechanism — and since it needs cluster access we don't have (and per our CLAUDE.md, infra/cluster ops are delegated to FuzeInfra), please run it on our behalf rather than us importing a cluster kubeconfig into our repo.

Already done in MendysRobotics (merged to main):

• .github/workflows/arc-register.yml — copied from your template, filled in SCALE_SET_NAME: mendysrobotics, USE_EXISTING_SECRET: arc-runner-github-app.

Please do (all within FuzeInfra's ownership):

1. Grant the arc-runner-github-app GitHub App access to izzywdev/MendysRobotics (Settings -> Developer settings -> GitHub Apps -> repository access), or confirm it's already org-wide.
2. Register the scale set against the cluster:

   runners/arc/register-repo.sh \
     --repo-url https://github.com/izzywdev/MendysRobotics \
     --name     mendysrobotics \
     --secret   arc-runner-github-app
   

   (mendysrobotics is unique — staging is yours; controller namespace arc-runners.)
3. Verify: kubectl -n arc-runners get autoscalingrunnersets shows mendysrobotics, and it appears under MendysRobotics -> Settings -> Actions -> Runners.
4. Reply here when it's up (or if you need anything from our side).

Alternative if you'd rather we self-install: deliver a scoped kubeconfig via the same in-cluster hand-off pattern you used for mendys-db-credentials and we'll run arc-register ourselves — but the cluster-side install is cleaner coming from you.

Context refs: FuzeInfra #219 (ONBOARD-REPO.md), MendysRobotics #154 (arc-register).

[claude]

Claude finished @izzywdev's task in 1m 38s —— View job

---

✅ mendysrobotics scale set is live on the shared ARC runner

Done from the FuzeInfra (cluster) side. Handled it via the register-repo.sh install as you requested — no kubeconfig hand-off needed.

• Read ONBOARD-REPO.md + register-repo.sh
• Verified cluster access (contabo-prod) and current scale sets
• Confirmed arc-runner-github-app secret exists in arc-runners
• GitHub App access — already covers the repo (verified below, no manual grant needed)
• Registered scale set mendysrobotics via register-repo.sh
• Verified autoscalingrunnersets shows mendysrobotics
• Report back

1. GitHub App access — ✅ already granted

No manual step was required. The scale-set listener authenticated to your repo on first boot — it successfully pulled a runner registration token:

msg="getting runner registration token"
registrationTokenURL=https://api.github.com/repos/izzywdev/MendysRobotics/actions/runners/registration-token
...
msg="Starting listener"
msg="Handling initial session statistics" totalAssignedJobs=0

That 200 from the repo-scoped endpoint means arc-runner-github-app already has access to izzywdev/MendysRobotics (org-wide install). If it lacked access, the listener would have crash-looped on 403 — it didn't.

2. Registration

✓ Scale set 'mendysrobotics' registered for https://github.com/izzywdev/MendysRobotics
Release "mendysrobotics" ... STATUS: deployed  (gha-runner-scale-set-0.14.2)

3. Verification

$ kubectl -n arc-runners get autoscalingrunnersets
NAME             MINIMUM RUNNERS   MAXIMUM RUNNERS   CURRENT RUNNERS   STATE
mendysrobotics   0                 5                 0
...

$ kubectl -n arc-systems get pods | grep mendys
mendysrobotics-674bcd6d-listener   1/1   Running

Listener is up and polling the GitHub broker for jobs. Config: minRunners=0, maxRunners=5, ephemeral runners land on fuzeinfra-ci-runner-1 (pool ci).

Your side — ready to use

Point CI at the runner with:

jobs:
  build:
    runs-on: mendysrobotics

It'll appear under MendysRobotics → Settings → Actions → Runners once the first job assigns (min=0, so pods are created on demand and torn down after — nothing shows until a job runs). Fire a smoke job (runs-on: mendysrobotics) to confirm end-to-end; it should pick up rather than sit queued.

Ping here if a job stays queued and I'll check the listener logs on the cluster.

DONE: mendysrobotics AutoscalingRunnerSet registered, listener Running and authenticated to the repo — ready for runs-on: mendysrobotics.