Skip to content

GitHub Actions runners on Azure

backmesh runs each GitHub Actions job on its own VM in your Azure subscription, spot unless the label says on demand or the job reruns after a spot reclaim. The VM is deleted when the job ends, and Azure bills you for it.

A job’s VM is the first type in its label’s list that has capacity:

Label vCPU Memory Types, in order
backmesh-2 2 8 GiB Standard_D2ads_v5, Standard_D2as_v5, Standard_D2s_v5
backmesh-4 4 16 GiB Standard_D4ads_v5, Standard_D4as_v5, Standard_D4s_v5
backmesh-8 8 32 GiB Standard_D8ads_v5, Standard_D8as_v5, Standard_D8s_v5
backmesh-16 16 64 GiB Standard_D16ads_v5, Standard_D16as_v5, Standard_D16s_v5

The KVM labels use as_v5 and s_v5 only, since ads_v5 doesn’t nest. When Azure is the control plane’s cloud, its labels are plain, as above; otherwise they name it, such as backmesh-azure-4.

Runners use a fixed list of US regions, not discovered through your cloud CLI or chosen on the command line: every one whose catalog offers every type above — every US region qualifies — narrowed to the ones your account allows. Each job tries the cheapest region first, then another when Azure refuses for capacity or quota. More.

Install works through your az CLI, signed in with a subscription, and creates:

  • A role connection: a custom role, a resource group with a managed identity and its federated credential, and a role assignment. The identity trusts only your control plane, so no Azure keys are stored.
  • Cache buckets, through that connection: one per runner region, bm-cache-<id>-<region>, for the cache.
  • The control plane, when Azure is its cloud: a Standard_D2s_v5 VM and a Standard_B2s VM for the listener, with their disks and public addresses, and a checkpoint bucket for its backups.

Uninstall deletes the VMs and buckets. It asks before deleting the checkpoint bucket, and leaves the role connection’s setup for you to delete in the console.

Spot labels count against each region’s “Total Regional Low-priority vCPUs” quota, and on-demand labels against “Total Regional vCPUs”. Install reads both through az vm list-usage and prints them as how many backmesh-4 jobs fit at once. A region whose quota can’t fit one VM of a label gets none of that label’s runners until you raise it; install and backmesh status name the quota. The on-demand GPU label also counts against “Standard NCASv3_T4 Family vCPUs”, 0 on most new subscriptions; request at least 4 under Quotas in the portal. More.

These are catalog prices in westus2, spot in the lowest zone, for a backmesh-4 VM while its job runs:

Type Spot $/h On demand $/h
Standard_D4ads_v5 0.0381 0.206
Standard_D4as_v5 0.0318 0.172
Standard_D4s_v5 0.0355 0.192

Other sizes scale with their vCPUs. Each job also pays for its VM’s boot and its 100 GB disk. The control plane costs about $111 a month. At those prices, 150,000 minutes a month of backmesh-4 jobs cost at least $191–$206 with the control plane, before disks, boots and runners started ahead of jobs; see the costs. Spot prices change by the hour.