DEV Community

Shell QA
Shell QA

Posted on

Taming Test Flakiness: A Practical Guide to Retries, Parallelism, and Stability Thresholds

Flaky tests degrade confidence in CI/CD pipelines, delay releases, and waste engineering cycles. Managing retries, test isolation, and worker parallelism strategically allows teams to control execution speed while ensuring test reliability.

Here is a practical guide to handling test flakiness, configuring retries, and establishing actionable stability thresholds for Playwright and automated test suites.

Core Recommendations for Test Stability

  • Retry Policy: Keep auto-retries low (1–2 attempts max). Use retries only for transient network blips and explicitly track retry attempts in execution logs.

  • Test Isolation: Ensure every test scenario is stateless. Clean up shared artifacts (database records, downloaded files, local storage) before and after each execution.

  • Stable Selectors: Prefer dedicated test attributes (e.g., data-test-id) over fragile layout-dependent CSS or XPath locators.

  • Controlled Parallelism: Start CI runs with limited workers (e.g., workers: 2) to avoid resource exhaustion on runner nodes. Scale parallelism up only after flakiness stays consistently low.

Flakiness Thresholds & Escalation Matrix

To keep test suites healthy, establish clear operational thresholds based on a rolling 7-day window:

  • Acceptable (< 1% intermittent failures): Standard operational state; monitor trends routinely.
  • Warning (1% – 5% intermittent failures): Trigger an immediate investigation and root-cause analysis sprint task.
  • Critical (> 5% intermittent failures): Block release deployments until infrastructure and test design issues are resolved.

Observability & Reporting

  • Emit Metrics: Include retry counts and flaky test flags directly inside generated test execution reports (e.g., Allure or JSON reporters).
  • Track Dashboards: Maintain a weekly flakiness dashboard tracking total tests executed against failure rates and retry counts to catch degrading tests early.

Top comments (0)