Attributing self-hosted runner costs to specific workflows — anyone solved this?

[aizenio]

🏷️ Discussion Type

Question

💬 Feature/Topic Area

Actions Runner

Discussion Details

Running ephemeral self-hosted runners on EC2 (scale-to-zero, one instance per job via webhook + Lambda).

The gap: GitHub's Usage API gives billable minutes, Cost Explorer gives instance dollars, and nothing joins them. I want to answer "what did the integration-tests workflow cost last month?" without manual work.

Has anyone built or bought something that does per-workflow or per-repo cost attribution for self-hosted runners? Happy to hear about tools, tags+CUR tricks, or "we gave up and here's what we do instead."

[iawaisahmd]

one approach i've seen work is tagging each ephemeral runner instance at launch with the github run_id, repo, workflow name, and job name, then carrying those tags into the aws cur. after that, you can join the cur rows back to the github actions run/job api and aggregate cost per workflow.

for scale-to-zero runners this is much cleaner than trying to infer cost from minutes alone, especially when instance types and startup overhead differ. the main thing to watch is whether the ec2 tags are available early enough in the billing data for every short-lived instance.

[VikulChoudhary]

Yes—this can be solved by tagging each ephemeral EC2 runner with GitHub metadata such as repo, workflow, run_id, and job_id when Lambda launches it, then using AWS Cost & Usage Report (CUR) with Athena to aggregate the actual EC2 spend by workflow/repository. I’d also keep a small DynamoDB “runner ledger” recording instance ID, start/end time, workflow, and job ID so you can precisely correlate short-lived instances with GitHub jobs and reconcile the result with CUR. The architecture would look like this:

GitHub Actions
      │
      │ workflow_job webhook
      ▼
API Gateway → Lambda
      │
      │ launch EC2 + tags
      ▼
┌─────────────────────────┐
│ Ephemeral EC2 Runner     │
│ Repo=org/app             │
│ Workflow=integration     │
│ RunID=123                │
└────────────┬────────────┘
             │
        Job completes
             │
             ▼
        Terminate EC2
             │
      ┌──────┴──────┐
      ▼             ▼
DynamoDB Ledger     AWS CUR
      │             │
      └──────┬──────┘
             ▼
        Athena / Dashboard
             │
             ▼
integration-tests = $184.27

This gives you both exact job-level attribution from the runner ledger and authoritative AWS billing from CUR, without manually reconciling GitHub minutes with EC2 costs.