I'm looking at full automated testing after deployment. Do you use Argo CD PostSync hooks, Argo Events, or something else? And do you run tests in static or ephemeral environments?
How do you run automated tests after an Argo deployment?
Answers
For a static staging environment, a small smoke suite in a PostSync hook is a straightforward option. Check that the service starts and can reach its database, then leave the longer end-to-end suite out of the sync path.
Reply to this answer
I'd separate the quick deployment check from the full suite. A PostSync hook can trigger a workflow that runs the longer tests in an ephemeral environment, rather than keeping the Argo CD sync open for the whole run.
The tradeoff is that triggering the workflow is not the same as waiting for its result. Decide separately how a failed test run should be reported or affect promotion. Ephemeral environments can also avoid some of the interference that comes with shared staging.
Reply to this answer
For tests involving multiple Argo CD applications, app-of-apps with sync waves is one option. Order the child Application resources so dependencies are healthy before the next wave, then put the test trigger last.
Watch the health assessment for the Application resource, though. Argo CD removed its built-in health check in version 1.8, so you may need a custom health assessment for waves to wait on the child applications' actual health.
9h+socketfeather_326:
That helps with the coordination question. The child application being created isn't necessarily the same as its workloads being ready.
Reply to this answer
Why wait until after deployment to run the full suite? Run as much as possible before deployment and keep the post-deploy checks focused. A long test run tied to sync can slow down deployments and make failures harder to interpret.
_q49+orbitmixer.8:
Pre-deploy testing still won't catch every environment difference. A small post-deploy check can find those problems before users do.
Reply to this answer
Add your answer
Answers are reviewed before publication.
_a47!brindlewinder.5:
That works neatly for one application. Tests spanning several applications need a way to know when all of them are ready.