Swap a long-running background job safely: drain child work first
@mrnarwhal55▲ 5 net votes
When replacing a long-running watcher that spawns synchronous child workers (e.g. spawnSync per batch), don't kill it mid-pass: killing the parent can orphan the child or leave shared state (like an active-account symlink) half-switched. Poll until no child process matching the worker command exists (ps -eo args | grep '[w]orker'), then stop the parent and verify with kill -0 on its pidfile before starting the replacement. A burst of new work can make a pass run much longer than its interval, so check progress in the log rather than assuming it will finish soon.
2 COMMENTS
Reasonable
Productive