shmpipeline

Links

• PyPI
• Read the Docs
• Issues

What It Is

shmpipeline builds local, process-based compute pipelines on top of named shared-memory streams from pyshmem.

It is a good fit when you want explicit dataflow, process isolation, and shared-memory throughput without giving up a small Python API, YAML-driven configuration, or interactive tooling.

Typical uses include:

• adaptive optics and other real-time sensor / control pipelines
• CPU or GPU processing graphs around shared-memory streams
• validating, inspecting, and running pipelines from Python or the CLI
• editing and supervising pipelines from a desktop GUI
• extending the runtime with custom kernels, sources, and sinks

Good starting points:

• Installation guide
• Quickstart
• Worked examples

Installation

Full docs: Installation guide

Choose the smallest install that matches your workflow:

• Base runtime and CLI: pip install shmpipeline
• GPU support: pip install "shmpipeline[gpu]"
• Desktop GUI: pip install "shmpipeline[gui]"
• Remote control service: pip install "shmpipeline[control]"
• Editable source install: pip install -e .
• Full local development environment: pip install -e ".[control,gpu,gui,test,docs]"

Quick smoke test against a checked-in example:

shmpipeline validate examples/affine_transformation/pipeline.yaml
shmpipeline describe examples/affine_transformation/pipeline.yaml --json

Open The GUI For The Observatory AO Example

Relevant docs:

• GUI guide
• Observatory AO system example

From the repository root:

pip install "shmpipeline[gui]"
shmpipeline-gui examples/observatory_ao_system/pipeline.yaml

The full GUI can auto-launch a local loopback control server when you press Build or Start.

If you want to launch the server yourself first:

shmpipeline serve examples/observatory_ao_system/pipeline.yaml --host 127.0.0.1 --port 8765
shmpipeline-gui examples/observatory_ao_system/pipeline.yaml

Quickstart For The Python API

Full docs:

• Quickstart
• Configuration guide

Minimal pipeline config:

shared_memory:
  - name: input_frame
    shape: [4]
    dtype: float32
    storage: cpu

  - name: scaled_frame
    shape: [4]
    dtype: float32
    storage: cpu

kernels:
  - name: scale_stage
    kind: cpu.scale
    input: input_frame
    output: scaled_frame
    parameters:
      factor: 2.0

Run it from Python:

import numpy as np

from shmpipeline import PipelineConfig, PipelineManager

config = PipelineConfig.from_yaml("pipeline.yaml")
manager = PipelineManager(config)
manager.build()
manager.start()

manager.get_stream("input_frame").write(np.array([1, 2, 3, 4], dtype=np.float32))
result = manager.get_stream("scaled_frame").read_new(timeout=2.0)
print(result)

manager.stop()
manager.shutdown()

If you prefer the CLI for the same pipeline:

shmpipeline validate pipeline.yaml
shmpipeline describe pipeline.yaml
shmpipeline run pipeline.yaml --duration 5.0

Popular Example Pages

• Affine transformation
• Basic AO system
• Observatory AO system
• Source and sink plugins

More Documentation

Start with these in roughly the order most users need them:

1. Configuration guide for the full YAML model, parameters, and shared-memory definitions.
2. Worked examples for complete CPU, GPU, custom-operation, AO, and plugin-backed pipelines.
3. CLI guide for validate, describe, run, and serve.
4. GUI guide for editing configs, validating locally, inspecting graphs, and launching viewers.
5. Runtime guide for lifecycle, metrics, worker health, and synthetic inputs.
6. Extensions guide for custom kernels, sources, and sinks.
7. Control plane guide for remote management, SSE events, and the Python client.
8. API reference, core API, and kernel catalog for the detailed reference surface.
9. Troubleshooting for common setup and runtime issues.