Unstructured-IO / unstructured-platform-plugins

Geek Repo:Geek Repo

Github PK Tool:Github PK Tool

Unstructured Platform Plugins

CI

Information about how to build custom plugins to integrate with Unstructured Platform.

Plugin Development

Any plugin must be published in a dedicated docker image with all required dependencies that when run, exposes an api on port 8000 with the required endpoints to interact with the Unstructured Platform product:

  • /invoke: A POST endpoint which gets all data to run the underlying logic in the request body and expects a json serializable response.
  • /schema: A GET endpoint which publishes a json schema formatted response with the schema of the input and output expected by the plugin.
  • /id: A GET endpoint which publishes a string unique identifier for this instance of the plugin. Will default to a hash of the schema response if one is not set explicitly.

Utility CLI

When installing this repo, it also installs the cli etl-uvicorn. This takes a pointer to any generic python function and wraps it in a FastApi application to conform to the patterns that are expected by the api hosting the plugin logic. This cli extends the existing uvicorn cli which takes in a pointer to a fastapi instance or factory but instead takes in a pointer to a python function/class which gets wrapped with a FastApi application.

Example usage

For all following commands, make sure you have the local repo in your PYTHONPATH:

export PYTHONPATH=.

Wrapping a basic function with a typed dict response

etl-uvicorn test.assets.typed_dict_response:sample_function

Wrapping a basic async function with a typed dict response

etl-uvicorn test.assets.async_typed_dict_response:async_sample_function

Wrapping a class. For this to work, the class must be self instantiating. When passing a class in, a method needs to be passed in as well, otherwise __call__ is used. The following example returns a pydantic BaseModel

etl-uvicorn test.assets.pydantic_response_class_method:SampleClass --method-name sample_method

Wrapping an instance of a class.

etl-uvicorn test.assets.pydantic_response_class_method:sample_class --method-name sample_method

The CLI does some validation on the wrapped function, which must have explicit inputs and outputs, meaning *args and **kwargs are not supported. These will cause the cli to fail fast.

etl-uvicorn test.assets.improper_function:sample_improper_function

/id requirements

All the the above examples caused the CLI to autogenerate the /id endpoint with a hash of the generated schema. However, you can also provide it a reference to use for the id value. This can be a reference to a concrete value (i.e. plugin_id="my_plugin_id") or a function in the same way that one was passed in to be wrapped above.

Will populate the response of /id with the static value of hash_value:

etl-uvicorn test.assets.typed_dict_response:sample_function --plugin-id test.assets.simple_hash_value:hash_value

Can populate it using a lambda:

etl-uvicorn test.assets.typed_dict_response:sample_function --plugin-id test.assets.simple_hash_lambda:hash_lambda_fn

Similar to the function being wrapped, can also use a class:

etl-uvicorn test.assets.typed_dict_response:sample_function --plugin-id test.assets.simple_hash_class:GetHash --plugin-id-method my_hash

Or the instantiated class:

etl-uvicorn test.assets.typed_dict_response:sample_function --plugin-id test.assets.simple_hash_class:get_hash_class_instance --plugin-id-method my_hash

About

License:Apache License 2.0


Languages

Language:Python 95.5%Language:Makefile 4.0%Language:Shell 0.6%