Merge pull request #4911 from guardian/jsh/add-dev-commands

Add a dev command to boot the e2e stack without running tests

Author: jonathonherbert

.gitignore

@@ -42,6 +42,3 @@ image-embedder-lambda/__tests__/embedder/test-data
 /e2e-tests/blob-report/
 /e2e-tests/playwright/.cache/
 /e2e-tests/playwright/.auth/
-
-# Testcontainers global-setup artifacts
-/e2e-tests/.grid-urls.json

e2e-tests/README.md

@@ -45,17 +45,37 @@ npm run test:ui         # open browser and test suite, run tests at your leisure
 Traces are captured `on-first-retry` (see [`playwright.config.ts`](playwright.config.ts)),
 so a failed test on CI leaves a trace you can open with `npx playwright show-trace`.
 
-## How they work
+## Running the stack without the tests
 
-Playwright's `globalSetup` ([`global-setup.ts`](global-setup.ts)) uses
-[Testcontainers](https://testcontainers.com/) to stand up everything the tests need:
+`npm run dev:e2e` boots exactly the same stack the tests use, prints the service URLs and
+holds it open until you press Ctrl-C, which tears it all down.
 
-1. a shared Docker network;
-2. **Elasticsearch** and **LocalStack** (the backing infrastructure);
-3. the **CloudFormation core stack** (`dev/cloudformation/grid-dev-core.yml`) and seeded
-   buckets, provisioned into LocalStack;
-4. per-service config, generated by `dev/script/generate-config/service-config.js`;
-5. the Grid Docker image (either `grid-e2e-ci` or `grid-e2e-dev`) — a single container running Grid's Play services.
+```bash
+npm run dev:e2e                  # uses your local dev-nginx for the https://*.media.<domain> domains
+GRID_PROXY=true npm run dev:e2e  # no dev-nginx? start the bundled Caddy proxy on :443 instead
+```
+
+The stack binds fixed host ports, so two stacks cannot run
+at once. Starting a second one fails immediately rather than timing out.
+
+### Running the tests against a stack you already started
+
+The test commands reuse a running stack instead of booting their own, which turns a
+multi-minute boot into a couple of seconds. Leave `npm run dev:e2e` running in one
+terminal, then use `npm test`, `npm run test:ui` or any of the others as normal — they
+attach automatically and leave the stack running when they finish.
+
+If only some services are up (usually because the stack is still booting), the run stops
+straight away and names the ports it is waiting on.
+
+| Variable | Effect |
+| --- | --- |
+| `GRID_RESEED=true` | Reload the Elasticsearch fixtures into the reused stack. |
+
+**Watch out for stale provisioning.** A reused stack picks up Scala changes (the repo is
+bind-mounted and services run under `sbt run`), but *not* changes to anything applied at
+boot: generated service config, the CloudFormation template, bucket contents, permissions
+or the Elasticsearch fixtures. After changing any of those, restart `dev:e2e`.
 
-Elasticsearch is then seeded with image fixtures, and the resolved Kahuna base URL is
-exposed to the tests. `globalTeardown` stops everything and cleans up.
+Reuse also means state carries over between runs. The current suite is read-only, so this
+is harmless today, but a test that uploads or edits an image will want a fresh stack.