Skip to content

Workermanager

What is the Workermanager?

The Workermanager is a lightweight aiohttp service that supervises all workers on a host. It:

  • Maintains worker metadata on disk (~/.funcnodes/workers/worker_<uuid>.json/.p/.runstate).
  • Spawns, stops, restarts, deletes, and lists workers.
  • Acts as a WebSocket hub so UIs can discover and control workers.
  • Optionally provisions per-worker virtualenvs and upgrades packages on activation.

By default it listens on localhost:9380 at / for WebSocket clients (no auth by default—see Security).

Message protocol (WebSocket)

Simple string commands:

  • ping → pong
  • identify → JSON { "class": "WorkerManager", "py": sys.executable }
  • worker_status → lists active/inactive workers
  • stop → stop the manager

JSON commands (selected):

  • { "type": "new_worker", "kwargs": {...} } → create worker (options: name, reference, copyLib, copyNS, in_venv toggle)
  • { "type": "set_active", "workerid": uuid } → activate (start) worker
  • { "type": "stop_worker", "workerid": uuid }
  • { "type": "restart_worker", "workerid": uuid }
  • { "type": "delete_worker", "workerid": uuid }

Responses/broadcasts include:

  • worker_status (active/inactive lists)
  • worker_created / worker_deleted
  • set_worker (full worker config once reachable)
  • progress (text + progress float) to drive UI HUDs

Files & liveness

Each worker has:

  • worker_<uuid>.json — config (host/port/env paths/nodespace path/flags)
  • worker_<uuid>.p — PID file written by the worker
  • worker_<uuid>.runstate — textual status during startup/running

Missing or stale files mark workers as inactive; status is refreshed every ~10 s.

Virtualenv & dependency management

  • New workers default to their own venv unless --not-in-venv was set.
  • On activation, optional update_on_startup flags can reinstall funcnodes, funcnodes-core, and unpinned dependencies.
  • CLI helpers funcnodes worker modules … run inside the worker env.

Auto-start behavior

Clients (e.g., funcnodes runserver) call assert_worker_manager_running: it pings/identifies the manager and, if unreachable, spawns a fresh instance via python -m funcnodes startworkermanager (optionally through subprocess_monitor).

Security

  • No built-in authentication; expose only behind an authenticated reverse proxy.
  • Keep the WS/HTTP ports private; block direct internet access.
  • Size limits: large message/store defaults come from worker settings (MAX_DATA_SIZE, message expiry).
  • TLS termination is not handled by the Workermanager itself; it serves plain HTTP/WebSocket. Terminate TLS at your proxy/load balancer if you need HTTPS/WSS.

Operations cheat sheet

  • Start manager: funcnodes startworkermanager --host 0.0.0.0 --port 9380
  • List workers: funcnodes worker list [--full]
  • New worker: funcnodes worker new --name demo
  • Start worker: funcnodes worker --name demo start
  • Delete worker: funcnodes worker --name demo delete