Test isolation / setup and teardown

Test isolation means each test starts from a known state and leaves nothing behind, so no test can depend on another test having run first. Setup and teardown are the machinery: hooks like beforeEach and afterEach in Vitest, Jest, or Mocha, fixtures in pytest and Playwright, that build fresh state before the test and dismantle it after. The smell that tells you it is missing is famous: the test passes on its own, fails inside the suite, or fails only when the runner shuffles the order or shards across machines.

The state that leaks is rarely just database rows. Module-level variables and caches, a global store, localStorage, mocked timers, spies that were never restored, environment variables, the DOM from the last render, an open server or socket. Practical fixes, cheapest first: create data inside the test with unique values instead of relying on a shared seed, wrap each test in a database transaction and roll it back, reset mocks and modules in a global beforeEach, and let libraries do their own cleanup (Testing Library unmounts automatically when the runner exposes an afterEach).

Gotcha: prefer building clean state in setup over promising to clean it in teardown. Most runners do still run teardown after a test fails or times out, but nothing runs it when the process crashes, CI kills the job, or the runner itself falls over, and once the cleanup is skipped every later test inherits the mess. Second gotcha: beforeAll looks like a performance win but it shares one object across tests, so the first test that mutates it silently couples the rest. Use it only for genuinely read-only setup, like starting a container or compiling a schema.