10 Nov, 2026Shankar Narayanan
Acceptance Testing for CI/CD: How Monk CI Tests Its Runners
A build can finish successfully while the platform records the wrong outcome or fails to request runner cleanup. Acceptance testing checks whether the complete workflow meets its requirements, including what happens after execution.
How Monk CI tests its runners
Monk CI's Go acceptance suite uses Ginkgo and Gomega to dispatch real GitHub Actions jobs and verify:
1. The webhook is processed and the job is persisted.
2. A runner is assigned and executes the job.
3. GitHub and platform records agree on success or failure.
4. VM deletion is requested after the job finishes
.We exercise both passing and deliberately failing builds. A failed build must still produce the correct records and trigger cleanup. Each run captures step-level evidence in HTML and JUnit reports so engineers can locate the broken transition.
A separate scenario follows a failed build through the opt-in AI debugging workflow and checks that processing reaches a recorded outcome.
How acceptance testing supports security
Monk CI's webhook tests include invalid-signature cases.
Our service CI workflows also include unit and integration tests, Go race detection, and security scanning:
Gosec: identifies potentially unsafe Go code patterns
Trivy: scans for known vulnerabilities, secrets, and configuration issues.
Gitleaks: detects potential exposed secrets .Acceptance tests verify required behavior; scanners look for additional classes of risk. Neither a passing workflow nor a clean scan guarantees vulnerability-free software.
Which acceptance testing framework should your team use?
| Your testing need | Framework to consider |
| Go services and asynchronous infrastructure workflows | Ginkgo with Gomega |
| Browser-based customer journeys | Playwright |
| Shared, readable Given/When/Then scenarios | Cucumber |
Where should teams start?
Choose one critical customer journey and write explicit pass/fail criteria. Test success, expected failure, and rejection of invalid requests. Check the final records and cleanup behavior, and retain evidence for every failed step. This turns a broad claim that a feature works into a repeatable check.