Slow tests

A handful of slow tests can dominate wall-clock time long before they dominate the failure rate. Karva surfaces them in two complementary ways: a threshold-based SLOW status during the run, and a --durations ranking after it.

Flagging slow tests during a run

--slow-timeout=SECONDS flags every test that runs for longer than the given duration:

1

karva test --slow-timeout=2.0

1
2

[tool.karva.profile.default.test]
slow-timeout = 2.0

The threshold accepts fractional seconds (--slow-timeout=0.5). Slow tests get a dedicated SLOW status line and are counted in the run summary.

The SLOW line is gated behind --status-level=slow (or higher); the summary slow count appears once --final-status-level=slow is set:

1

karva test --slow-timeout=2.0 --status-level=slow --final-status-level=slow

Slow detection is purely informational — it does not fail the run or kill the test. To time-box a test instead, use @karva.tags.timeout.

Ranking the slowest tests

--durations=N prints the N slowest tests after the run completes, regardless of whether --slow-timeout is set:

1

karva test --durations=10

Use this when investigating a slow CI job: it answers "which tests are eating the budget?" without requiring you to guess at a sensible threshold first.