#Flaky and slow scripts

Two options cover the cases that waste the most CI time: a test that fails once and passes on the retry, and a process that hangs until the job times out an hour later.

#A test that fails intermittently

'test:network': {
  command: 'vitest run tests/network',
  retries: 2,
  retryDelay: 'exponential',
}

Three attempts at most. 'exponential' waits 2^attempt seconds, so 2 s before the second attempt and 4 s before the third. A plain number is a fixed wait in milliseconds.

Retries happen only on failure, so a passing script runs once.

The part worth knowing: anything watching the script sees only the final attempt. The reported exit code, a group's failure handling and a success policy all read one result. A script that fails twice and then passes is a success, not three events.

#A process that can hang

e2e: {
  command: 'playwright test',
  timeout: 600000,
}

When the budget elapses, Rune terminates the whole process tree and exits 124, the same code GNU timeout uses:

error: `e2e` exceeded its 600000 ms timeout — process tree terminated

A script that finishes inside its budget is untouched. No early termination, no altered code.

#Both together

Each attempt gets a fresh budget, so this is four attempts of up to 30 seconds each, not 30 seconds in total. A timed-out attempt counts as a retryable failure.

'deploy:smoke': {
  command: 'node scripts/smoke.mjs',
  timeout: 30000,
  retries: 3,
}

#Letting a server clean up

killSignal and killTimeout describe how Rune terminates a script it decided to stop, whether from a timeout or from a failing sibling in a parallel group.

'dev:api': {
  command: 'node --watch server.js',
  killSignal: 'SIGINT',
  killTimeout: 2000,
}

A server that closes its connections on SIGINT gets two seconds to do it before Rune escalates to SIGKILL.

This does not apply to Ctrl+C from the terminal. That interrupt reaches the child directly and Rune simply waits for it. See Exit codes.

#Where these belong

On command and extends scripts, single or grouped. A group entry that declares one is a validation error, because a group is not a process. Put them on the members.

The full field list is in Timeouts and retries.