fal-runner

---

name: fal-runner description: Verify fal app behavior end to end: deploy aliases, test public websocket endpoints, and inspect live runners. Use when working with fal deploy, fal run, fal runners exec, or websocket readiness.

Fal Runner

Quick Start

Use this skill when you need to:

• verify that a fal app is reachable from its public URL
• confirm a websocket endpoint returns the expected initial message
• inspect the live runner process tree and filesystem

Default assumptions from this repo:

• uv run fal ... is the correct invocation
• realtime health is best checked through the public websocket, not just container logs
• runner behavior must be verified from both the public endpoint and the live runner

Standard Workflow

1. Confirm the public endpoint behavior first.
2. Inspect the active runner IDs for the deployed alias.
3. Exec into the live runner and inspect:
   • process list
   • /app/src/... files
   • __pycache__
   • relevant temp directories
   • environment variables
4. If testing temporary deploys, clearly mark them with a name that identifies the user and model (e.g. <user>-<model>) and clean them up afterward.

Public Verification

Check alias runners:

uv run fal apps runners <alias> --env main --json

Check public websocket readiness:

uv run python - <<'PY'
import asyncio, websockets

URL = "wss://fal.run/<team>/<alias>/ws"

async def main():
    async with websockets.connect(URL) as ws:
        print(await asyncio.wait_for(ws.recv(), timeout=20))

asyncio.run(main())
PY

Expected success pattern for this workflow:

• connection succeeds
• first message is a ready-style payload such as {"type":"ready",...}

If the websocket fails:

• verify the app is actually public
• verify the outer fal websocket is forwarding to a healthy inner runner
• do not assume a generic forwarding error means the inner app is broken

Runner Inspection

Use the live runner ID from fal apps runners.

Useful checks:

uv run fal runners exec <runner-id> -- sh -lc 'ps -efww'

uv run fal runners exec <runner-id> -- sh -lc 'ls -la /app/src/scope/cloud'

uv run fal runners exec <runner-id> -- sh -lc 'ls -la /app/src/scope/cloud/__pycache__'

uv run fal runners exec <runner-id> -- sh -lc 'tr "\0" "\n" </proc/1/environ | sed -n "1,160p"'

uv run fal runners exec <runner-id> -- sh -lc 'tr "\0" " " </proc/1/cmdline'

For broader artifact inspection:

uv run fal runners exec <runner-id> -- sh -lc '/usr/bin/python3.12 - <<\"PY\"
import os
for root in ["/app", "/tmp", "/root", "/local", "/alloc"]:
    if not os.path.exists(root):
        continue
    for dirpath, dirnames, filenames in os.walk(root):
        for name in filenames:
            if "livepeer_fal_app" in name or "fal_app" in name:
                print(os.path.join(dirpath, name))
PY'

Interpretation Rules

• A working public websocket matters more than whether the entrypoint file is visible under /app/src.
• Always test the public websocket directly rather than relying only on container logs.

Discovery And Packaging Rules

fal deploy does not blindly execute the file path as the runtime authority. It:

• executes the file with runpy.run_path(...)
• finds a fal.App symbol in that module dict
• wraps and serializes the resulting callable
• sends both serialized function data and source text to the server

Repo-Specific Notes

• For cloud app testing here, use uv run fal ....
• The live inner runner may be started from /app/src/scope/cloud/livepeer_app.py.

Additional Resources

• For deeper notes and investigation patterns, see reference.md.