GCP Scenario

Massively Parallel Playwright on Cloud Run Jobs

Use Cloud Run Jobs task parallelism to run hundreds of Playwright shards concurrently, then aggregate via a Workflow.

Architecture

Cloud Build ─► image: playwright-runner ─► Artifact Registry
                                  │
       Workflow ─► gcloud run jobs execute e2e \
                       --tasks 100 --parallelism 100
                                  │
       Each task: npx playwright test --shard $TASK/$TOTAL
                                  │
                       JUnit + trace.zip ─► GCS
                                  │
       Workflow ─► Cloud Function: merge shards → HTML report

Services used

Cloud Run Cloud Build Artifact Registry Cloud Storage Workflows Cloud Functions

Steps

1. Build runner
Cloud Build produces a Playwright runner image (browsers preinstalled) pushed to Artifact Registry.
2. Fan out
Workflow calls `gcloud run jobs execute` with `--tasks N --parallelism N`; each task gets `CLOUD_RUN_TASK_INDEX`.
3. Execute shard
Each task runs `--shard=$((INDEX+1))/$TOTAL`, uploads JUnit + trace + video to GCS.
4. Merge
Function downloads shard outputs, merges into a unified HTML report and writes a `summary.json`.
5. Gate
Workflow checks summary; non-zero failures or duration-regression > 20% fails the release.

Takeaways

Cloud Run Jobs `--parallelism` removes the need to manage a grid.
Shard granularity is the main lever for wall-clock vs cost.
Centralizing traces in GCS makes every failure replayable later.