GCP Scenario

Serverless Nightly Regression with Workflows

Cloud Scheduler kicks a Workflow that seeds data, runs API + UI automation in parallel Cloud Run Jobs, aggregates to BigQuery.

Architecture

Cloud Scheduler (0 2 * * *) ─► Workflows
   ├─► Cloud Function: seed Firestore fixtures
   ├─► Parallel
   │     ├─► Cloud Run Job: Newman (API tests)
   │     └─► Cloud Run Job: Playwright (UI tests, sharded)
   ├─► Cloud Function: merge JUnit → HTML + trend deltas
   ├─► GCS: upload report
   ├─► BigQuery: stream per-test metrics
   └─► Pub/Sub: emit qa.run.completed

Services used

Cloud Scheduler Workflows Cloud Functions Cloud Run Firestore Cloud Storage BigQuery Pub/Sub

Steps

1. Schedule
Cloud Scheduler fires HTTP target on the Workflow with the run correlation id.
2. Seed
Function clears + seeds Firestore collections used by tests.
3. Parallel automation
Workflows runs Cloud Run Jobs concurrently — Newman for APIs, sharded Playwright for UI.
4. Aggregate + trend
Function merges JUnit, computes duration deltas vs last 7 nights, builds HTML, uploads to GCS.
5. Metrics
Per-test rows streamed into BigQuery for flakiness + duration analytics.
6. Notify
Pub/Sub fans out to Slack + dashboards.

Takeaways

Zero idle infrastructure — pay only per nightly run.
BigQuery is the source of truth for QA telemetry and flaky-test detection.