flamepy.app Guide

Package Python code and run functions and classes as remote Flame services.

flamepy.app Guide

flamepy.app packages the current Python project, registers an application from the flmrun template, and exposes Python functions and classes as remote services. The former flamepy.runner module is no longer in the current SDK.

Prerequisites

Run a Flame cluster with a session manager, executor manager, object cache, and registered flmrun template. Install flamepy in the client environment and make application dependencies available to executors. Configure FLAME_ENDPOINT and FLAME_CACHE_ENDPOINT or ~/.flame/flame.yaml as shown in Getting Started.

Remote Functions

import flamepy.app as app

app.init("hello-app")


@app.service()
def hello(name: str) -> str:
    return f"hello, {name}"


try:
    print(hello.remote("flame").get())
finally:
    app.destroy()

Initialize the application before declaring services. Calling the decorated function directly executes it locally; .remote() submits a task and returns an ObjectFuture. Use .get() to fetch the result, .wait() to wait without fetching it, or app.get() to resolve a collection of futures. app.destroy() closes owned services and cleans up an application created by this process; an already existing application is reused and is not cleaned up by that process.

Remote Classes

import flamepy.app as app

app.init("counter-app")


@app.service(resreq="cpu=1,mem=1g")
class Counter:
    def __init__(self, start: int):
        self.value = start

    def increment(self, amount: int) -> int:
        self.value += amount
        return self.value


try:
    counter = Counter.remote(10)
    print(counter.increment(2).get())
finally:
    app.destroy()

A class proxy represents a service; executor processes retain its execution object across calls. Scheduling can use more than one executor, so do not assume all calls share a single in-memory object unless your session configuration ensures it. Set resource requirements and scheduling options on @app.service(). For grouped work, batch_size expresses the size of a gang; see the App setup guide for current details.

Packaging and Storage

App packages the current working directory into dist/<name>.tar.gz. Use .flameignore or .flmignore to exclude files. If package.storage is not configured, App uploads the package through the configured object cache. An explicit storage backend must be accessible to clients and executors.

Check flmctl list --application if the flmrun template cannot be found. See the upstream App setup guide for configuration, data-aware scheduling, and troubleshooting.