End-to-end tests against real SnapRAID, snapraid-daemon and the Docker image

[firsttris]

What

End-to-end tests (e2e/, Playwright with Chromium): the UI is driven through the browser against real SnapRAID and snapraid-daemon on throwaway arrays.

Test	Covers
array.spec.ts	Dashboard health and disks; sync with preview (new and deleted files); bit rot found by a full scrub and repaired byte for byte by Repair (fix -e); Undelete
schedules.spec.ts	Nightly routine (touch → sync → scrub) created in the form and started with Run now; scheduled sync skipped while a disk is not mounted, dashboard shows Disk not available
notifications.spec.ts	Webhook set up in the UI: the test message and job failed reach a local receiver
config.spec.ts	Data disk added in the visual editor is protected by the next sync
docker.spec.ts	Container is paused during the sync and resumed after it (docker events)
daemon.spec.ts	Daemon mode switched on in the UI, Test connection, then the sync runs as a snapraid-daemon task

• npm run setup builds the pinned versions into e2e/.tools: SnapRAID 14.10 (the image's version), SnapRAID 15, and snapraid-daemon at REVIEWED_DAEMON_VERSION.
• The global setup starts the daemon, the backend, the production frontend build, and a small proxy that routes /api and /ws the way nginx does.
• E2E_IMAGE=<image> runs the same 10 tests against the Docker image, started as in docker-compose.yml (data in /app/snapraid, disks at the same paths as on the host, Docker socket mounted).
  • The container writes as root, so in this mode the tests check files with docker exec inside the container.
  • At the end the container hands the test data back to the caller, so runners without root can read and delete it.
• CI: a new e2e matrix job with local and image. The image job builds the image from docker/Dockerfile. Tool builds are cached, and on failure traces and server logs are uploaded.

Run now: a play button on each schedule runs it once immediately: same steps, sync guard, disk checks, Docker pause and notifications. It also works on disabled schedules. Backend: POST /api/schedules/:id/run, returning 202, 404 if the schedule doesn't exist, 409 if a job is running.

Bug found by the E2E test: adding a data disk with a trailing slash in its path produced content /mnt/disk1//.snapraid.content. SnapRAID only excludes its content files when the path matches as written. So this file was synced as a data file, and every diff showed it as changed. Now uses join(), with a unit test.

Other changes

• The backend reads PORT (default 8080, as in the image's supervisord).
• e2e/ is added to .dockerignore.
• Docs: new End-to-end tests section in docs/development.md, Run now in docs/scheduling.md. The upstream release checklists also name e2e/scripts/setup-tools.sh.

Verified

• Backend: deno check, deno test (122 passed).
• Frontend: Biome, typecheck, Vitest (24).
• mkdocs build --strict.
• E2E local: 10/10, three runs in a row, about 1.3 min including the frontend build.
• E2E against the image: 10/10.
  • In CI (e2e (image)), with the image built from docker/Dockerfile.
  • Locally as a non-root user, the same setup as the GitHub runner.

🤖 Generated with Claude Code

https://claude.ai/code/session_01JuXnquCPdbG5sBbDyin7yF