harmoney

Distributed Function Caller Framework for Python

Installation:

pip install harmoney

Dependencies:

websockets
fastapi
uvicorn
pydantic

Usage:

Requires 3 scripts: Client, Router and Runner

1. Router

The Router will mediate load balancing and connection handling between the function runners and callers, so this should start first. One port must to facilitate the communication.

This example starts the router on IP 192.168.0.110 and port 7732.


from harmoney.router import startRouter

startRouter("0.0.0.0", 7732)

2. Runner

Runner exposes the function calls and contains all the heavy compute logic. Below code connects one instance of runner to the router.


from harmoney.runner import startRunner

def customFunction(arg1: int, arg2: str) -> str:
    return arg2*arg1

funcs = {"custFn": customFunction}

startRunner(funcs, "192.168.0.110", 7732)

Functions must return objects that are pickleable. None or no return statement is valid.

3. Client

Client requests the function - argument combinations to run on runners.


from harmoney.client import Client

cli = Client("192.168.0.110", 7732)

retVal = cli.runSingle("custFn", arg1=10, arg2="arst") # run only this combination

print(retVal)

cli.addCall("custFn", arg1=123, arg2="qwf") # add this combination to queue
cli.addCall("custFn", arg1=321, arg2="ars") # add this combination to queue
res = cli.runAllCalls() # send queued combs at a time.
print(res) # Print all the results

Return values from runALlCalls are in the order of their function combination in the queue.

TODO:

Error catching, keeping the connection to the broker
Error info should return to the client
Remove dependency on fastapi and requests, move to completely to websockets

~ A Grammer Society Project.

1.9 KiB Raw Blame History