rekuest-next 0.5.7

rpc and node backbone

rekuest

self-documenting asynchronous scalable RPC based on provisionable untrusted actors

Idea

rekuest is the python client for the rekuest server, it provides a simple interface both register and provide nodes (functionality) and call other nodes in async and synchronous manner. Contrary to most RPC services, Rekuest is focussed on providing functionaly on the Client, and is especially tailored for scanarios where apps can be developed to perform tasks on users behalves, therefore requiring fine grained access control.

Prerequesits

Requires a running instance of a rekuest server (e.g in an arkitekt deployment).

Install

pip install rekuest-next

rekuest is relying heavily on asyncio patters and therefore only supports python 3.7 and above. Its api provides sync and async interfaces through koil.

If you are working in image analysis checkout the arkitekt platform that also provides data structures for image analysis (composed in the arkitekt platform)

Get started

from rekuest import Rekuest
from herre import Herre, ClientCredentialsGrant

auth = Herre(grant=ClientCredentialsGrant(client_id="ssss", client_secret="osinsoisnoein"))

client = Rekuest(auth=auth)

@client.register()
def rpc_function(x: int, name: str) -> str
    """
    A rpc function that we can
    simple call from anywhere

    ""
    print(str)


with client:
  client.run()

Run example:

arkitekt dev

This node is now register under the application and signed in user and can be provisioned and called to by other apps. By default users are only able to asign to their own apps. This can be changed on the rekuest server.

Calling

from rekuest import Rekuest, find
from herre import Herre, ClientCredentialsGrant

auth = Herre(grant=ClientCredentialsGrant(client_id="ssss", client_secret="osinsoisnoein"))

client = Rekuest(auth=auth)

with client:
  node = find(interface="rpc_function") #

  with node:
    node.assign(4, "johannes")

Usage with complex Datastructures

Rekuest takes care of serialization and documentation of standard python datastructures

• str
• bool
• int
• float
• Enum
• Dict
• List

To increase performance and latency it is not possible to serialize complex python objects like numpy arrays into the messages. These are best transformed into immutable objects on a centrally accessible storage and then only the reference is passed. Rekuest does not impose any rules on how you handle this storage (see mikro for ideas), it provides however a simple api.

class ComplexStructure:
    id: str # A reference for this structure on central storage

    async def shrink(self):
        return self.id

    @classmethod
    async def expand(cls, value):
        return cls.load_from_server(value)

by providing two functions:

• shrink
• expand

You can now use this Structure with simple typehints and arkitekt will automaticall shrink (serialize) and expand (deserialize) the structure on calling.

def complex_call(x: ComplexStrucuture) -> int:
    return x.max()

Terminology

Node A concept (documentatio) of a function that is enabled on the platform.

App: A provider of functions, that negotiates access right to data and other Apps through Oauth2

Template: An Implementation of a Node by an App.

Agent: An active instance of this App, its the host of actors. Agents connect and disconnect.

Actor: A stateless instance of a function that was provisioned

Provision: A contract between arkitekt and a Agent for the usage of a specific function. As long as the provision is active the connected agent will be required to provide the resources of the function (think Deployment in Kubernetes)

Reservation: A contract between a User and arkitekt that wants to use one or mulitple instances of functions (Actors). The platform tries to autocorrect and failure correct. Calls to the function are always address to the reservation no the provision (think: Exchange in RabbitMQ)