Skip to content

About

The sandbox runtime for workspace methodologies: the Python interface, the harness and the runtime image (lake#655)

Resources

Code of conduct

Contributing

Security policy

Stars

0 stars

Watchers

0 watching

Forks

Repository files navigation

methodology-runtime

The sandbox runtime for a Lake workspace's own methodologies (lake#655). A workspace methodology is Python code in one of the workspace's repositories; this repository holds:

  • orama_methodology/ — the interface a methodology implements (fit, predict, score, optional explain), the Data object it receives, the manifest and lock-file checks, the sandbox checks, and the harness;
  • bin/methodology — the command the three run pods use: stage, run, collect; a model deployment's stage-model and serve (lake#690); plus validate-manifest for local use and install-locked for image builds;
  • Dockerfile — the image oramasearch/methodology-runtime (Python 3.12 pinned by digest, non-root uid 10001, standard library only);
  • templates/methodology.Dockerfile — the image Lake builds for each methodology, FROM this one;
  • examples/ — a working standard-library logistic regression and a broken example.

Read

  • docs/interface.md — what a methodology is: the manifest, the functions, the data, the lock file.
  • docs/sandbox.md — what the run container can and cannot see, and how data moves in and out only through Lake.
  • docs/cli.md — the commands, the HTTP calls stage and collect make, exit codes and image paths.

Build and test

python3 -m unittest                                # the whole suite, standard library only
./bin/methodology validate-manifest examples/logistic
docker build -t methodology-runtime:local .        # a local image; never pushed from here
tools/sandbox-smoke.sh                             # run it as the run pod does, no registry

A model a training run produced is served by the same image in serve mode (orama_methodology/serve.py, lake#690): Ship → Models in Lake asks the operator for a deployment, and predictions go through Lake.

Who runs it: the Orama operator (orama-operator#291) runs the stage, run and collect pods; Lake serves methodology_run_input and methodology_run_output and records the result in the workspace Zetta.

Released: v0.1.0 (all releases).

The image is oramasearch/methodology-runtime on Docker Hub. Private repository; the Docker Hub repository oramasearch/methodology-runtime is private.

License

MIT, see LICENSE. Third-party components and their licenses are in THIRD_PARTY_LICENSES. Contributions are welcome under the Contributor License Agreement; see CONTRIBUTING.md.

About

The sandbox runtime for workspace methodologies: the Python interface, the harness and the runtime image (lake#655)

Resources

Code of conduct

Contributing

Security policy

Stars

0 stars

Watchers

0 watching

Forks

Releases

Packages

Contributors

Languages