anyserve-runtime 0.4.0

Rust-backed Python package for AnyServe

anyserve Python Bindings

This package provides high-performance Python bindings for AnyServe. Its transport implementation comes from crates/anyserve-client; the public package adds role-specific facades and higher-level worker helpers on top.

Install

From PyPI:

pip install anyserve-runtime

The published distribution name is anyserve-runtime, while the import package remains:

import anyserve

From a local wheel:

pip install target/wheels/anyserve_runtime-*.whl

From local source:

pip install ./clients/python

Source installs build the native extension locally, so they require a working Rust toolchain on PATH. Inside this repository, mise exec -- ... provides the expected toolchains.

The top-level AnyserveClient exposes the generic control-plane surface and also offers role-specific views through submitter() and worker().

It also includes a high-level worker API for function-style handlers:

from anyserve import serve, worker


@worker(
    interface="demo.echo.v1",
    attributes={"runtime": "python"},
    capacity={"slot": 1},
    codec="bytes",
)
def echo(payload: bytes) -> bytes:
    return payload


serve(echo, endpoint="http://127.0.0.1:50052")

Examples use endpoint strings like http://127.0.0.1:50052 because the bindings connect through a gRPC channel URI. This is not a REST base URL.

API Surface

• submit_job
• watch_job
• list_jobs
• get_job
• cancel_job
• get_attempt
• list_attempts
• register_worker
• heartbeat_worker
• poll_lease
• renew_lease
• report_event
• complete_lease
• fail_lease
• open_stream
• get_stream
• list_streams
• close_stream
• push_frames
• pull_frames

watch_job(...) and pull_frames(...) return iterators so the Python API preserves streaming semantics instead of materializing the full stream into a list.

It also includes two convenience constructors:

• inline_object(...)
• uri_object(...)

And high-level worker helpers:

• worker(...)
• serve(...)
• get_handler_spec(...)

Local Development

Build and install from local source into the active environment:

pip install ./clients/python

Or install the extension directly into the active environment through maturin:

python3 -m pip install --user maturin
python3 -m maturin develop --manifest-path clients/python/Cargo.toml

Live end-to-end validation:

mise run python-sdk-e2e

PyPI Publishing

The repository is set up to publish the Python package through PyPI Trusted Publishing.

• PyPI project name: anyserve-runtime
• GitHub repository: anyserve/anyserve
• Workflow path: .github/workflows/python-publish.yml
• GitHub environment: pypi

For the first release, create a pending Trusted Publisher for anyserve-runtime on PyPI, then run the publish workflow or push a v* tag.

Example

High-level worker example:

from anyserve import serve, worker


@worker(
    interface="demo.echo.v1",
    attributes={"runtime": "python"},
    capacity={"slot": 1},
    codec="bytes",
)
def echo(payload: bytes) -> bytes:
    return payload


serve(echo, endpoint="http://127.0.0.1:50052")

Low-level submitter example:

from anyserve import AnyserveClient, FRAME_DATA

client = AnyserveClient("http://127.0.0.1:50052")
# Assumes the control plane and a compatible worker are already running.

job = client.submit_job(
    interface_name="demo.echo.v1",
    required_attributes={"runtime": "demo"},
    required_capacity={"slot": 1},
)

stream = client.open_stream(job["job_id"], "input.default")
client.push_frames(
    stream["stream_id"],
    [(FRAME_DATA, b"hello from python", {})],
)
client.close_stream(stream["stream_id"])

for event in client.watch_job(job["job_id"]):
    print(event["kind"], event["metadata"])