Can PM2's reload command achieve zero-downtime migration for a small stateful Node.js application?

Direct Answer

No. PM2's reload command provides zero-downtime availability—the listening port stays open and requests continue to be served—but it cannot preserve in-memory session state across worker replacements. Each reloaded worker starts with a fresh JavaScript heap; any sessions stored in local memory are discarded when that worker exits.

What PM2 Reload Actually Does

• In cluster mode, pm2 reload restarts workers sequentially (one at a time by default).
• The master process keeps the socket open, so new connections are never refused during the transition.
• The replacement worker is spawned, brought online, and only then is the old worker sent a shutdown signal (SIGINT by default).
• If --wait-ready is used, PM2 waits for the new worker to call process.send('ready') before stopping the old one.

This sequence guarantees that the service stays up. It does not migrate JavaScript objects, closures, or module-level variables from the old worker to the new one.

Why In-Memory Sessions Are Lost

• Session data held in a Map, object, or array inside the worker process lives only in that process's heap.
• When the worker receives the shutdown signal, it can run cleanup code, but there is no built-in mechanism to hand that memory to the incoming worker.
• Even with sticky-session routing (e.g., PM2's sticky option or an external load balancer), the session disappears the moment its owning worker is reloaded.

What Would Be Required to Preserve Sessions

To keep sessions across reloads without a network-backed store (Redis, database), you must add application-level code that:

1. Serializes session state to a shared location (file system, shared memory segment, or SQLite file) before the worker exits.
2. Restores that state in the new worker during startup.
3. Handles concurrency, file locking, and crash recovery for that shared location.

This is a code and architecture change, not a PM2 configuration flag.

One Diagnostic Question

Is losing in-memory sessions during a reload acceptable for your use case? If yes, pm2 reload alone gives you zero-downtime availability. If no, you must implement session persistence as described above.