Skip to main content

Deploy Verification

Featured

After each production deploy, compares error rate, latency and SLO burn against the pre-deploy baseline and last week. On a regression it can attribute to the change, it pages the owner and opens a revert PR.

How it works​

  1. Runs when: a deployment to production succeeds.
  2. Compares: the 30 minutes after the deploy with the 30 minutes before, for error rate, p99 latency per endpoint and SLO burn.
  3. Attributes carefully: a movement counts as the deploy's only if the same hour last week does not show it, it is confined to the service's own spans, and its dependencies are flat.
  4. You get: a verdict that names which checks passed, with evidence. On a real regression, the owning team is paged and a revert pull request waits for their approval.

What it needs​

  • Datadog: read metrics, traces and SLOs
  • PagerDuty: page the owning team
  • GitHub: read the deployed pull request, open a revert