e2e-testing

E2E Testing

End-to-end tests are the last line of defense before production: they exercise real user journeys across the whole stack and catch integration failures unit tests structurally cannot. Keep the suite small, critical, and stable — a flaky E2E suite that nobody trusts is worse than none.

Define journeys by risk

Test journeys, not pages. A journey is a full user-visible flow (sign in → create → see it persist). Rank them and cover top-down; do not chase 100% of screens.

• HIGH — money, auth, data loss (payment, login, delete/restore)
• MEDIUM — core navigation, search, primary CRUD
• LOW — cosmetic/UI polish (usually not worth an E2E test)

For each journey cover the happy path first, then the one or two error paths that would actually hurt (declined payment, expired session).

Write for stability

• Semantic locators only. [data-testid="…"] > role/label > CSS > XPath. A test tied to DOM structure breaks on every refactor; one tied to intent does not.
• Wait for conditions, never for time. waitForResponse() / expect(...).toBeVisible(), never waitForTimeout(3000). Time-based waits are the single largest source of flake and they silently slow the suite.
• Isolate every test. No shared state, no order dependence; each test sets up and tears down its own data so it can run alone and in parallel.
• Assert at each key step, not just at the end, so a failure points at the step that broke.

Quarantine flake, don't ignore it

A flaky test that stays green-ish erodes trust in the whole suite. When a test flakes:

1. Confirm it with a repeat run (--repeat-each=10 or equivalent).
2. Quarantine it explicitly (test.fixme/test.skip) with a tracked issue reference in the reason — never a bare skip that disappears silently.
3. Fix the root cause (usually a time-wait, a race, or shared state), then un-quarantine. Quarantine is a holding cell, not a graveyard.

Capture artifacts on failure

Configure screenshots, video, and traces to record on failure/first-retry (not always — artifacts are expensive). On CI, upload them and put the artifact path in the failure report so a red run is debuggable without a local repro.