Hayhooks

Table of Contents

• Hayhooks
  • Quick start
    • Install the package
    • Check Hayhooks status
    • Deploy a Haystack pipeline
    • Have a look at the API schema
    • Run the pipeline
    • Undeploy the pipeline
  • Docker setup
  • Next steps
  • License

Quick start

Install the package

Start by installing the package:

pip install hayhooks

The hayhooks package ships both the server and the client component, and the client is capable of starting the server. From a shell, start the server with:

$ hayhooks run
INFO:     Started server process [44782]
INFO:     Waiting for application startup.
INFO:     Application startup complete.
INFO:     Uvicorn running on http://localhost:1416 (Press CTRL+C to quit)

Check Hayhooks status

From a different shell, you can query the status of the server with:

$ hayhooks status
Hayhooks server is up and running.

Deploy a Haystack pipeline

Time to deploy a Haystack pipeline. The pipeline must be in Yaml format (the output of pipeline.dump()), if you don't have one at hand, you can use one from this repository. From the root of the repo:

$ hayhooks deploy tests/test_files/test_pipeline_01.yml
Pipeline successfully deployed with name: test_pipeline_01

Another call to status should confirm your pipeline is ready to serve requests:

$ hayhooks status
Hayhooks server is up and running.

Pipelines deployed:
- test_pipeline_01

Have a look at the API schema

Hayhooks will use introspection to set up the OpenAPI schema accordingly to the inputs and outputs of your pipeline, and to see how this works let's get the pipeline diagram with:

$ curl http://localhost:1416/draw/test_pipeline_01 --output test_pipeline_01.png

The downloaded image should look like this:

As you can see, the pipeline in order to start requires an input of type int named value, and optionally we can pass another input of type int named add. At the end of the run, the pipeline will return an output of type int named result.

If you open a browser at http://localhost:1416/docs#/ you should see two schemas, one for the Request, where we'll pass the pipeline inputs (note how add is optional):

Test_pipeline_01RunRequest
    first_addition
        value* integer
        add (integer | null)

And another one for the Response, where we'll receive the pipeline results:

Test_pipeline_01RunResponse
    double
        value* integer

Run the pipeline

At this point, knowing the schema we can run our pipeline with an HTTP client:

$ curl -X 'POST' \
  'http://localhost:1416/test_pipeline_01' \
  -H 'accept: application/json' \
  -H 'Content-Type: application/json' \
  -d '{
  "first_addition": {
    "value": 19
  }
}'

{"double":{"value":42}}%

Undeploy the pipeline

Hayhooks tries to to as much bookkeeping as possible without restarting the server. For example, to free up resources you can undeploy the pipeline directly from the client:

$ hayhooks undeploy test_pipeline_01
Pipeline successfully undeployed

Docker setup

[!TIP] If you have docker-compose installed and you want to quickly get up and running, just do docker-compose up -d

Instead of launching the server in a separate shell like we did in the Quick Start, you can run it in a Docker container :

$ docker run --rm -p 1416:1416 deepset/hayhooks:main
...

If you want to build the container yourself:

$ cd docker
$ docker buildx bake
...

There are 2 special folders in the container you can override using a mount:

1. A folder /opt/pipelines containing pipeline definitions that will be automatically deployed when the container starts
2. A folder /opt/custom-components containing custom components that Haystack will be able to import if part of a pipeline

For example, you can mount a local ./pipelines folder containing pipelines you want to run at start-up like this:

$ docker run --rm -p 1416:1416 -v $PWD/pipelines:/opt/pipelines "deepset/hayhooks:main"

Next steps

At this stage Hayhooks is just a prototype, the natural next steps would be:

• Improve server configuration management (easy)
• Manage pipeline dependencies: one way could be adding the required packages in the pipeline's metadata and let the server handle installation (complex)

License

hayhooks is distributed under the terms of the Apache-2.0 license.