Skip to main content

Cost Anomaly Triage

For each daily cost anomaly, finds the deploy, config change, scaling event or traffic change that caused it, names the owning team, and proposes a fix with its dollar impact.

How it works​

  1. Runs when: once a day, after the overnight cost refresh, for each open anomaly.
  2. Finds the step: the hour spend changed and the usage type that moved.
  3. Lines up four candidates: configuration changes, production deploys, scaling events and request rate. It names a cause only when timing and mechanism match and the other three are ruled out. Otherwise it reports the anomaly as unexplained and says what is missing.
  4. You get: the cause, the owning team, and the impact per day, per month and so far. When the resource is declared in code, the fix comes as a pull request, plus a guard if the cause was a hand edit.

What it needs​

  • Vantage: read anomalies and cost history
  • AWS: read Cost Explorer and CloudTrail
  • GitHub: read deploys, open pull requests
  • Slack: post the summary