37Left-Shift & Right-Shift Testing

37.1 The two halves

37.2 Shift-left practices

37.3 Shift-right practices

37.4 Synthetic monitoring with Playwright

// tests/prod-smoke.spec.ts — runs every 5 minutes in production
test('prod login still works', async ({ page }) => {
  await page.goto('https://app.acme.com/login');
  await page.getByLabel('Email').fill(process.env.SYNTH_USER!);
  await page.getByLabel('Password').fill(process.env.SYNTH_PASS!);
  await page.getByRole('button', { name: 'Sign in' }).click();
  await expect(page.getByText('Dashboard')).toBeVisible({ timeout: 10_000 });
});

test('prod checkout API is up', async ({ request }) => {
  const r = await request.get('https://api.acme.com/health');
  expect(r.status()).toBe(200);
});

Schedule via GitHub Actions cron, Datadog Synthetics, or Checkly. Page the on-call if a check fails twice in a row.

37.5 Contract testing with Pact (1-min mental model)

1. Consumer (frontend) defines what it expects from the API.
2. Consumer's tests use a mock that records the expectation as a "Pact" file.
3. Pact file is published to a broker.
4. Producer (API) replays the contract in its own tests — must match.
5. If producer changes break the contract, producer's CI fails before deploy.

Result: frontend and backend evolve independently without integration surprises. Common in microservice orgs.

37.6 Feature flags + canaries

// Flag in code
if (await featureFlags.isEnabled('new-checkout')) {
  return <NewCheckout />;
} else {
  return <OldCheckout />;
}

// SDET checks both paths in CI
test('old checkout', async ({ page }) => {
  await page.context().addCookies([{ name: 'ff-new-checkout', value: 'off', url: '/' }]);
  /* ...assert old flow... */
});
test('new checkout', async ({ page }) => {
  await page.context().addCookies([{ name: 'ff-new-checkout', value: 'on', url: '/' }]);
  /* ...assert new flow... */
});

37.7 What "right-shift" looks like in practice

• You ship a feature behind a flag, off for everyone.
• Enable it for 1% of users.
• Monitor: error rate, conversion, latency on the 1%.
• If metrics OK, ramp to 10%, then 50%, then 100%.
• If error rate spikes, flag off — instant rollback without a deploy.
• SDET defines the "metrics-OK" thresholds and authors the dashboards.