- Web and iOS end-to-end tests.
pnpm test:e2ebuilds the Worker and runs the web suite inapps/e2ewith e2e 0.23.1: sign up, verify, sign in; the seeded Member's session; create and delete a todo;/docs; the anonymous dashboard redirect; and the todo flow driven by agent goals, which skips without a model and replays fromapps/e2e/recordings/once recorded.pnpm test:e2e:mobilesigns the seeded Member in on the Expo development build and finds a seeded todo. Locators come from the apps' own English catalogs. (End-to-end tests, f5b4ef72, 993c4f49) - One stack per machine, and nothing left behind. The suites' app command starts the built Worker in local workerd on a temporary PostgreSQL under a lock: a second run fails at once naming the owner pid, a dead owner's lock is taken over and its cluster and workerd stopped, a killed runner's stack stops itself, and the mobile wrapper shuts down only a simulator it booted.
node scripts/e2e-cleanup-proof.mjsproves a normal run,kill -9mid-run and two runs at once. (End-to-end tests, f5b4ef72) - Killed test runs no longer leave servers or containers.
scripts/test-local-postgres.mjs --serverecords its dev server group, and the next run stops it with the orphaned cluster; a SIGKILLed run used to leave wrangler/workerd serving.test:rls-matrixstarts its containers with--rm, cleans up on SIGTERM and SIGHUP, and first removes, with their volumes, the containers of runs that are gone. (Testing, 6616cf79, 3d1f95cf) - CI. A new
e2eworkflow runs the web suite on same-repository pull requests and pushes tomainwith the runner's own PostgreSQL binaries and uploadsapps/e2e/.e2e/; it holds no deploy or database secret. (End-to-end tests, 475c4251)
End-to-end tests on one local stack that a killed run never leaves behind
apps/e2e runs web and iOS end-to-end tests with e2e on the built Worker and a locked temporary PostgreSQL; test-local-postgres and the RLS matrix sweep what a killed run left.
Edit on GitHubDocumentation is in English.