Carlos Florêncio
2.6 KiB
id, title, status, wave, depends_on, plan, spec
| id | title | status | wave | depends_on | plan | spec |
|---|---|---|---|---|---|---|
| 06-container-isolation | Isolate container E2E state | completed | 1 | plan.md | ../../specs/platform/requirements/e2e-duration-aware-sharding.md |
Task 06: Isolate container E2E state
Acceptance
- Storage-maintenance tests create and count only resources owned by the
current test or worker, using unique labels/identifiers rather than a global
kandev.managed=truecount. - Setup and teardown remove only resources created by that test and remain safe when another container E2E shard is running on the same daemon.
- The original first-attempt timeout path has a regression assertion or diagnostic that distinguishes resource leakage from genuine slow work.
Verification
cd apps/web
KANDEV_E2E_CONTAINERS=1 pnpm exec playwright test --config e2e/playwright.config.ts --project=containers e2e/tests/docker/storage-maintenance.spec.ts
Run the targeted test with at least two container processes when the local Docker daemon permits it, and verify that cleanup does not touch unrelated Kandev-labelled resources.
Files likely touched
apps/web/e2e/tests/docker/storage-maintenance.spec.tsapps/web/e2e/fixtures/docker-test-base.tsapps/web/e2e/fixtures/docker-probe.ts- related Docker/SSH fixture helpers only if ownership labels cross the shared boundary
Dependencies
None for implementation. Task 08's performance experiment depends on this task's scoped cleanup guarantee.
Parallelism
Parallel candidate with Tasks 01, 04, and 05. Coordinate only if a shared Docker fixture helper is also being changed by another repair.
Inputs
- The global cleanup and count behavior in the Docker fixture.
- The storage-maintenance test's
removeKandevContainers()call and comments. - The existing container project timeout and worker lifecycle.
Output contract
Report the ownership label/identifier scheme, cleanup scope, concurrent-shard test evidence, targeted result, and any daemon prerequisites. Do not solve the flake by raising the 180-second test timeout.
Implementation result
Docker ownership now propagates from the E2E process into backend-created containers and storage accounting. The storage-maintenance test passed once and in two concurrent container processes; no managed containers remained afterward. Backend lifecycle/storage tests passed in the focused Go packages. Review remediation now runs process-scoped container removal after every API reset. A serial regression proves the following test cannot observe a scoped container left by its predecessor; all three storage-maintenance tests passed.