2026 Rent Mac Mini: launchd vs PM2 for 7×24 Long Tasks — Process Guardian Matrix, Plist Keys, PM2 Boot, and Log Rotation

Indie developers and small teams that run seven by twenty four jobs on a rented Mac Mini need a clear answer: should macOS launchd own the process, or should PM2 sit in front of your Node and Python workers?

This article gives a comparison table, a plist field checklist, PM2 boot wiring, rotation notes, and an FAQ. Jump to Home, the Blog, or Help Center when you are ready to connect. Related long-run guides include long-term task SLA FAQ, cron and watchdog, and Ollama keepalive steps.

Pain points before you pick a supervisor

1. Session confusion. Anything started only inside an SSH shell stops when the session ends unless launchd or PM2 owns it.
2. Restart storms. Aggressive respawn without ThrottleInterval or PM2 backoff can hide the real defect and fill disks with duplicate logs.
3. Ops visibility. You need predictable log paths, exit reasons, and a single place to inspect state after a provider maintenance window.

Selection decision matrix

Use this scan-first grid when you compare process guardians for Apple Silicon hosts. It focuses on fit, logging, crash behavior, and steady-state cost.

A common hybrid is PM2 under launchd: launchd guarantees login-level start, while PM2 manages app-level restarts. Pair this pattern with alerts from health checks and webhooks when you expose HTTP ports.

launchd plist key checklist

Place user agents in ~/Library/LaunchAgents, validate with plutil -lint, then load with launchctl bootstrap gui/$UID on modern macOS.

• Label unique string referenced by launchctl.
• ProgramArguments array with absolute paths; avoid brittle PATH assumptions.
• WorkingDirectory so relative assets resolve the same after reboot.
• RunAtLoad true when you expect start at login.
• KeepAlive dictionary with SuccessfulExit false when exits should trigger restart.
• ThrottleInterval integer seconds between restarts; start near ten for noisy jobs.
• StandardOutPath and StandardErrorPath on a volume with quota headroom.
• EnvironmentVariables for tokens instead of baking secrets into the plist when possible.
• Nice or LowPriorityIO when batch work must yield to interactive sessions.

After edits, launchctl bootout the old entry, then bootstrap again so process guardian state matches disk.

PM2 configuration and login boot

Treat PM2 as your app supervisor while macOS still owns session lifecycle. These steps mirror what most Mac Mini renters need after first SSH setup.

1. Install Node LTS, then npm install -g pm2 for a global CLI.
2. Author ecosystem.config.cjs with name, script, cwd, env, and max_memory_restart guardrails.
3. Run pm2 start ecosystem.config.cjs and confirm pm2 status shows online workers.
4. Persist the process list with pm2 save after you trust the graph.
5. Generate launchd glue via pm2 startup launchd, paste the printed command, reboot once, and verify pm2 resurrect behavior.
6. Document ports and health URLs beside credentials you store from Console.

Resources and log rotation

Long jobs on a rented Mac Mini fail quietly when APFS free space shrinks because uncompressed logs grew without a cap.

• Point launchd stdout files into a dedicated logs/ tree and add newsyslog or a weekly truncate script.
• Enable PM2 log rotation module or ship logs to remote storage when retention must exceed weeks.
• Watch unified memory pressure when you run both inference and many workers; align max_memory_restart with real unified RAM.

Citeable defaults: start ThrottleInterval at ten seconds for flaky workers. Expect a small PM2 resident set near fifty to ninety megabytes before counting app heaps. Rotate or compress logs before fifteen percent free space warnings trigger.

FAQ

Next steps. Wire your process guardian, validate reboot behavior once, then move automation to a dedicated node. Review Pricing, complete Purchase, and keep Blog runbooks beside your on-call notes.