premerge/linux_runners_values.yaml

githubConfigUrl: "https://github.com/llvm"
githubConfigSecret: "github-token"

minRunners: 0
maxRunners: 16

runnerGroup: ${ runner_group_name }

template:
  metadata:
    annotations:
      cluster-autoscaler.kubernetes.io/safe-to-evict: "false"
  spec:
    serviceAccountName: linux-runners-ksa
    tolerations:
    - key: "premerge-platform"
      operator: "Equal"
      value: "linux"
      effect: "NoSchedule"
    nodeSelector:
      premerge-platform: linux
    containers:
    - name: runner
      image: ghcr.io/llvm/ci-ubuntu-24.04-agent:latest
      command: ["/home/gha/actions-runner/run.sh"]
      resources:
        # If we don't set the CPU request high-enough here, 2 runners might
        # be scheduled on the same pod, meaning 2 jobs, and they will starve
        # each other.
        #
        # This number should be:
        #  - greater than number_of_cores / 2:
        #    A value lower than that could allow the scheduler to put 2
        #    runners on the same node. Meaning 2 jobs sharing the resources of
        #    a single node.
        #  - lower than number_of_cores:
        #    Each pod has some basic services running (metrics for ex). Those
        #    already require some amount of CPU (~0.5). This means we don't
        #    exactly have N cores to allocate, but N - epsilon.
        #
        # We also need to request sufficient memory to not get OOM killed.
        requests:
          cpu: 55
          memory: "200Gi"
        limits:
          cpu: 64
          memory: "256Gi"
      env:
      # We pass the GCS bucket name in as an environment variable as it is
      # different per cluster. We do not directly pass this to sccache
      # through the SCCACHE_GCS_BUCKET variable so that we can control how
      # sccache is caching directly in the workflow.
      - name: CACHE_GCS_BUCKET
        value: ${ cache_gcs_bucket }