Scheduling & Self-heal

Pypen has two layers of "keep it running" logic:

  1. s6-overlay handles crash-restart at the process level. If a worker exits, s6 restarts it within seconds.
  2. app/cron.py handles scheduled restarts — the kind you want for long-running scripts that "go bad" after a while (memory leaks, stale connections, expired tokens).

Periodic Restarts — restart_on

Restart the worker every N hours. Useful for bots and scrapers whose process state drifts over time. 

[project.cron]
restart_on = "6"   # restart every 6 hours

Hard Redeploys — redeploy

When redeploy = "true", a scheduled restart also wipes and rebuilds the venv. Use it when a project's dependencies change often and you want each restart to pick up the latest pinned versions.

Idle Restarts — idle

idle = "10" auto-starts a worker after 10 minutes of inactivity (no log output). This is handy for jobs that are supposed to "always be talking" — if they go quiet, something is wrong, so bounce them.

Pull-on-restart — pull_commits

With pull_commits = "true" (the default), every restart is preceded by git pull. Combined with restart_on this gives you a poor-man's continuous deployment: push to main, wait at most N hours, the new code is live.

Self-heal Behaviour

Crash-loops are absorbed by s6-overlay. If a worker is crashing repeatedly, the dashboard surfaces it with a red status; you can stop or rebuild it from the UI without restarting the whole container.

Recommended starting point

For typical bots: restart_on = "12", redeploy = "false", idle = "", pull_commits = "true". Tighten from there if you observe drift.