relife2 — your apps stay up. Period.

The PM2 you wish you had from day one — without the gotchas.

You run 5–10 apps on one VPS. PM2 was great — until pm2 update silently spawned a second daemon, pm2 save was forgotten after a reboot, and bun apps ran as node with extra overhead. relife2 is the fix.

Why teams switch from PM2

| PM2 pain you know | relife2 answer | |---|---| | pm2 update duplicates the God daemon → 2→4→8 daemons, OOM | Exactly one daemon, always. Socket bind + flock. Second daemon can't start. daemon-upgrade is explicit and locked. | | Reboot wipes everything if you forgot pm2 save | Autosave on by default. Every change hits snapshot.json + journal. Reboot → systemd + linger → all apps back. | | interpreter: bun actually runs as node | Bun is first-class. bun <script> directly, auto-detect via bun.lockb/bun.lock, bun --hot in dev. | | Log files grow forever (needs pm2-logrotate) | Built-in rotation: rotate --max-size --retain with gzip, no plugin. | | Unknown config fields silently ignored | Honest errors: warnings for unknown fields, errors for incompatible combos. | | Daemon is heavy (God ~60–90 MB) | Light: ~20–30 MB RSS daemon, no God overhead. |

One command to prove it:
relife2 doctor — checks snapshot, lock, orphans, transport, uptime. PM2 has no equivalent.

Who is it for

• Solo dev / small team with 5–15 apps on a single Oracle Linux 9 / Debian / Ubuntu VPS (the primary target — developed on Windows, but Linux-first)
• You use Caddy for TLS/reverse proxy and want real supervision for the upstreams
• You ship Bun + Node mix (Nuxt, Astro, bots, Java services, Rust binaries)
• You are tired of babysitting processes

If you run Kubernetes for everything, you don't need this. If you run a VPS and want systemd-grade reliability without writing unit files by hand — you do.

30-second demo

npm i -g relife2            # or: curl single binary → /usr/local/bin/relife2
relife2 start ecosystem.config.cjs
relife2 list
relife2 logs web -f         # follow
relife2 doctor              # 5 checks, exit 1 on error
relife2 startup             # systemd user unit + loginctl enable-linger
# reboot → everything is back. No pm2 save dance.

Migrate from PM2 in one line:

relife2 import-pm2 ~/.pm2/dump.pm2 && relife2 list   # reads PM2 dump, starts via relife2
relife2 from-caddy /etc/caddy/Caddyfile --dry-run    # Caddyfile → relife2.config.cjs

Find forgotten configs (the PM2 paper trail):

relife2 find                       # scans $HOME depth 6
relife2 find --root /opt --depth 8 --json | jq .
relife2 find start                 # starts every stopped config it found

Features that sell themselves

• Single binary deploy: bun build --compile → relife2-linux-x64 (~40–60 MB). No Node/Bun on the VPS. tar.gz / .rpm / .deb ready.
• Forgotten-path healing: moved from /home/debian to /opt? Configs with stale cwd/script auto-fix (tries ~/tail, sourceDir, basename search) and logs not found → auto-fixed instead of crash-looping.
• Caddy-native: from-caddy parses reverse_proxy/php_fastcgi loopback upstreams, generates one app per port with HOST/PORT/SITE_URL, finds entry (Nuxt .output/server/index.mjs etc.), --dry-run plan before writing.
• Dev that doesn't fight you: relife2 dev ./server.ts --watch ./src --hot → bun --hot for .ts, fs.watch otherwise. q to quit.
• Upgrade without drama: relife2 daemon-upgrade — save → shutdown old → spawn new → resurrect under lock, protocol version checked. No silent second daemon.
• Observability out of the box: list / describe / metrics (daemon + per-app RSS) / monit TUI / logs --lines -f.
• PM2-compatible config surface: ecosystem.config.{js,cjs,mjs,json} + relife2.config.ts, all 8 example/ configs work without edits (golden tests).

Installation

npm (any OS, Node ≥20):

npm i -g relife2
relife2 --version   # relife2 1.0.1 (node)
rl2 --version       # alias

Single binary (VPS — recommended, no runtime on target):

npm run build:single:linux        # → dist/relife2-linux-x64
scp dist/relife2-linux-x64 vps:/usr/local/bin/relife2
ssh vps 'chmod +x /usr/local/bin/relife2 && relife2 --version'
# or from GitHub Releases:
curl -L https://github.com/<org>/relife2/releases/latest/download/relife2-linux-x64 -o /usr/local/bin/relife2
chmod +x /usr/local/bin/relife2

tar.gz / rpm / deb:

npm run dist:tar   # → dist/relife2-1.0.1-linux-x64.tar.gz (binary + README + LICENSE)
npm run dist:rpm   # → dist/relife2-1.0.1-1.x86_64.rpm  (needs fpm)
npm run dist:deb   # → dist/relife2_1.0.1_amd64.deb     (uses dpkg-deb or fpm)
sudo dnf install ./dist/relife2-*.rpm        # Oracle Linux / RHEL / Fedora
sudo apt install ./dist/relife2_*_amd64.deb     # Debian / Ubuntu

All packages install /usr/bin/relife2 + /usr/bin/rl2; autostart is still relife2 startup (user unit) so it respects XDG_RUNTIME_DIR.

Autostart (Linux):

relife2 startup        # writes ~/.config/systemd/user/relife2.service + loginctl enable-linger
systemctl --user daemon-reload && systemctl --user enable --now relife2
systemctl --user status relife2
relife2 doctor

Quick Start

# 1. From a config (PM2-compatible)
relife2 start ecosystem.config.cjs
relife2 list
relife2 logs myapp --lines 100
relife2 doctor

# 2. Single script
relife2 start ./server.js --name myapp -- --port 3000

# 3. Auto-discover relife2.config.* / ecosystem.config.* in cwd
relife2 start

# 4. Zero-downtime reload / scale
relife2 reload myapp
relife2 start --instances max --exec-mode cluster --env production

# 5. Find & rescue forgotten configs
relife2 find --root /opt --depth 8
relife2 find start

# 6. Caddyfile → relife2
relife2 from-caddy /etc/caddy/Caddyfile --dry-run
relife2 from-caddy /etc/caddy/Caddyfile -o relife2.config.cjs && relife2 start relife2.config.cjs

More → docs/CLI.md · docs/CONFIG.md · docs/INSTALL.md

Config at a glance

// ecosystem.config.cjs — same file you use for PM2
module.exports = {
  apps: [{
    name: "web",
    script: "./dist/server/entry.mjs",
    cwd: __dirname,
    interpreter: "bun",          // "bun" | "node" | "none" | "/usr/bin/python"
    args: ["--port", "3000"],    // string or string[]
    instances: "max",            // number | "max" (= cpu count)
    exec_mode: "cluster",
    env: { NODE_ENV: "production", PORT: "3000" },
    env_production: { PORT: "4000" }, // merged with --env production
    env_file: "./.env",          // dotenv, does NOT override real env
    out_file: "./logs/web-out.log",
    error_file: "./logs/web-err.log",
    merge_logs: true,
    time: true,
    log_date_format: "YYYY-MM-DD HH:mm:ss Z",
    autorestart: true,
    max_restarts: 15,
    restart_delay: 3000,
    min_uptime: "10s",
    kill_timeout: 5000,
    max_memory_restart: "512M",
    cron_restart: "0 4 * * *",
    watch: false,                // true | false | ["./src", ".env"]
    watch_delay: 1000,
    wait_ready: false,           // wait for process.send('ready')
  }]
};

Full matrix → docs/CONFIG.md. CLI overrides: relife2 start --env production --instances max.

Bun, Caddy, Linux — first-class

• Bun: interpreter: "bun" spawns bun directly; auto-detect via bun.lockb/bun.lock or .ts script; relife2 dev ./app.ts uses bun --hot (HMR); bun relife2 ... works (CI matrix node+bun).
• Caddy: Caddy does TLS/static/proxy, relife2 does supervision. from-caddy + # relife2: dir=… name=… hints.
• Linux: XDG-first data ($XDG_DATA_HOME/relife2) + runtime ($XDG_RUNTIME_DIR/relife2), RELIFE2_DIR override for tests, /proc RSS, setsid process groups, systemd Restart=always + KillMode=process. SELinux enforcing = experimental (default contexts, no custom policy — author has it disabled).

Development & benchmark

npm install
npm run build        # → dist/cli.js (esbuild, single ESM, shebang)
npm run typecheck && npm run lint && npm test   # 67 tests (m7 integrity inside)
bun dist/cli.js --version
npm run bench        # startup latency + daemon RSS
npm run bench -- --pm2   # compare with PM2 (if installed)

See docs/BENCHMARK.md — numbers are host-dependent, run on your VPS for truth.

License

MIT — see LICENSE. Roadmap → TODO.md, decisions → docs/DECISIONS.md.