Run timeout

--run-timeout puts a wall-clock limit on a test run. Its deadline starts before collection, so collection, partitioning, and worker startup consume the same budget as test execution. Use it as a CI tripwire for a runaway suite; use @karva.tags.timeout when one test needs its own limit.

Measure healthy CI runs first, choose a timeout that leaves room for normal variance, then pass it as --run-timeout=SECONDS, replacing SECONDS with that measured value.

The same limit can be stored as run-timeout under the profile's test configuration.

When the deadline expires, Karva stops remaining workers and fails the run. Completed test results are still reported. Configured report files are also written: native JSON/JSONL marks the run as failed, while JUnit contains the completed results. CI should therefore honor karva's process exit status rather than inferring timeout success from JUnit alone.

Worker shutdown

Karva first asks workers to terminate gracefully. Workers that remain alive after termination-grace-period are force-killed. Configure the grace period long enough for any process signal handlers that must finish.

The same shutdown sequence applies to Ctrl-C and fail-fast cancellation.