Distributed Pipelines with Ray

[PedroSantanaDev]

Hello,

Is there any working colab examples of working with the Ray pipeline on haystack? I am trying to build a Ray pipeline but I get different errors. See below. Thanks

`Traceback (most recent call last):
File "C:\Users\psantana\PycharmProjects\haystack_test\main.py", line 40, in
pipeline = RayPipeline.load_from_yaml(path="ray.pipeline-pipelines.yaml", pipeline_name="ray_query_pipeline")
File "C:\Python Environments\haystack_test\lib\site-packages\haystack\pipelines\ray.py", line 187, in load_from_yaml
return RayPipeline.load_from_config(
File "C:\Python Environments\haystack_test\lib\site-packages\haystack\pipelines\ray.py", line 89, in load_from_config
validate_config(pipeline_config, strict_version_check=strict_version_check, extras="ray")
File "C:\Python Environments\haystack_test\lib\site-packages\haystack\pipelines\config.py", line 241, in validate_config
validate_schema(pipeline_config=pipeline_config, strict_version_check=strict_version_check, extras=extras)
File "C:\Python Environments\haystack_test\lib\site-packages\haystack\pipelines\config.py", line 269, in validate_schema
if (not extras and extras_in_config) or (extras and extras_in_config not in extras):
TypeError: 'in ' requires string as left operand, not NoneType

`

[julian-risch]

@ArzelaAscoIi Do you maybe have any code snippets showing Ray pipelines in action?

[PedroSantanaDev]

This is the pipeline yaml.

components:
- name: FAISSDocumentStore
  params:
    embedding_dim: 128
  type: FAISSDocumentStore
- name: Retriever
  params:
    document_store: FAISSDocumentStore
    embed_title: false
    passage_embedding_model: vblagoje/dpr-ctx_encoder-single-lfqa-wiki
    query_embedding_model: vblagoje/dpr-question_encoder-single-lfqa-wiki
  type: DensePassageRetriever

pipelines:
  - name: ray_query_pipeline
    type: RayPipeline
    nodes:
    - inputs:
      - Query
      name: Retriever
version: 1.9.0a

Here is how I am trying to load it.
pipeline = RayPipeline.load_from_yaml(path="ray.pipeline-pipelines.yaml", pipeline_name="ray_query_pipeline")

[ZanSara]

Hey @PedroSantanaDev, unfortunately the RayPipeline docs misses one crucial detail. I submitted a change, the docs are going to be fixed soon. In the meantine, I believe you should fix the YAML this way:

components:
- name: FAISSDocumentStore
  params:
    embedding_dim: 128
  type: FAISSDocumentStore
- name: Retriever
  params:
    document_store: FAISSDocumentStore
    embed_title: false
    passage_embedding_model: vblagoje/dpr-ctx_encoder-single-lfqa-wiki
    query_embedding_model: vblagoje/dpr-question_encoder-single-lfqa-wiki
  type: DensePassageRetriever

pipelines:
  - name: ray_query_pipeline
    # Remove the 'type' key
    nodes:
    - inputs:
      - Query
      name: Retriever
version: 1.9.0  # Use an existing version that matches the one you have installed. If you're installing from GitHub, use 'ignore'
extras: ray  # Add this line

[ArzelaAscoIi]

Thanks @ZanSara

[ArzelaAscoIi]

Hi @PedroSantanaDev. From the logs you shared it seems like the validation is not happy with the provided yaml. Did you already validate that the yaml works without ray? If yes I would also have to manually check what might go wrong in ray.
Is it maybe possible that the version: 1.9.0a is an extra field that is not allowed ?