You almost certainly don't need this. One machine (ROLE=all, the current setup) runs everything and scales a long way. If it gets slow, move to a bigger VPS first. Only split the engine across machines when a single box genuinely can't keep up. There's no separate download: a worker runs this same codebase with ROLE=engine, and workers coordinate only through this database. They never talk to each other.

Before you start

• The worker machine must reach this box's PostgreSQL over a private network. The database is deliberately not exposed to the internet, and it must stay that way (never open port 5432 publicly).
• Each Broker connects its own price and execution session over FIX.

On this machine (once)

1. Set ROLE=web in .env and redeploy. It keeps serving the app and live quotes but stops running the engine locally. Do this only once at least one engine worker is already running, otherwise nothing gets liquidated. (While it's ROLE=all it owns every Server, so a worker would sit idle.)

On each worker machine

1. Install Docker, copy this codebase over, and docker build -t etrader .
2. Create .env with: ROLE=engine, DATABASE_URL=postgres://etrader:PASSWORD@<this-box-private-ip>:5432/etrader, MAX_SERVERS_PER_WORKER=<total Servers ÷ number of workers> (this is what splits the Servers evenly), any feed keys your Brokers rely on. The production boot check also requires NODE_ENV=production, JWT_SECRET (32+ chars), ADMIN_PASSWORD, CORS_ORIGIN and BACKUP_DIR=/backups. A worker doesn't use them, but it won't start without them.
3. docker run -d --restart always --env-file .env etrader (no port needed, a worker serves no HTTP).

The worker registers within a few seconds and appears in the table above, leasing its share. Stop one and its Servers are re-claimed by the others within about 20 seconds.