Setting up Custom Nodes in REST API

[rtest2344]

Hello everyone,

Apologies if this has been asked before, but I couldn't an answer in the Issues section. I am trying to add an custome node, an HTML parser, and Issue #1576 outlined a great example of the code on how to do this. Issue #955 covered my issue, but it has been some time since that was brought up and I was wondering the process of including a custom node for the your-pipeline-here.haystack-pipeline.yaml pipeline for the docker container. Any advice would be greatly appreciated, thank you!

Edit:
For an additional way of asking, I would like to create a custom node and include it in a pipeline.yaml for the REST API. If I create this custome node in its own .py file, where should I copy this over to in the docker container to utilize it for the pipeline.yaml?

[vblagoje]

@rtest2344 , apologies for missing your question. Have you made any progress already?

[rtest2344]

HI @vblagoje , I have not because I've been waiting on a reply, no worries on the time though. I noticed at https://github.com/deepset-ai/haystack/tree/main/rest_api/rest_api/pipeline there is a custom_component.py file, would I just construct the class in that file and upload it to the same location I upload the yaml to in the docker container to add a custom component? For example an HTML parser or something?

[Bryson14]

Is there any update on this? The Yaml file definition being the only way to define the REST API seems hardly enough to use in a production use. Using more complex ingestion, other Vector Stores, or even having a way to bulk upload documents to a shared location instead of always going through the API during a initial loading would be awesome.

[anakin87]

Hello, @Bryson14. One can consider the Haystack REST API as a convenience REST API and customize/rewrite it for different use cases.

In any case, I'm interested in your opinions: which technical/design improvements do you suggest?

[lambda-science]

Still no fix for this ? Even temporary / dirty ? I'm also trying to add a custom component to my YAML pipeline served in REST API and I feel dumb / helpless ahah @rtest2344

[warichet]

Hello,

A not very clean solution where you must re-generate image when you add new custom files.

To integrate a custom node, we first create a 'custom' directory within the Docker directory.
Then, we place the files associated with the custom node into this directory.
Finally, we update the Dockerfile to copy these files into /opt/haystack prior to the creation of the virtual environment.

# Shallow clone Haystack repo, we'll install from the local sources
RUN git clone --depth=1 --branch=${haystack_version} https://github.com/deepset-ai/haystack.git /opt/haystack
WORKDIR /opt/haystack

COPY ./custom_files/custom_pdf.py /opt/haystack/haystack/nodes/file_converter/custom_pdf.py
COPY ./custom_files/custom_preprocessor.py /opt/haystack/haystack/nodes/preprocessor/custom_preprocessor.py

# Use a virtualenv we can copy over the next build stage
RUN python3 -m venv --system-site-packages /opt/venv
ENV PATH="/opt/venv/bin:$PATH"

RUN pip install --upgrade pip && \
    pip install --no-cache-dir .${haystack_extras} && \
    pip install --no-cache-dir ./rest_api

[lambda-science]

Hey, thanks for the fix. But I would prefer a way through docker compose because this method requiere you to re-build the image at every single change of your component.