Using EmbeddingRetriever as a node in the indexing pipeline

[andrewyazura]

Hi, @larsbaunwall and @simonclausen! First of all, I'd like to say thanks for the effort you've put in this project, it really helped me with figuring haystack out.

I'd like to ask a question regarding the EmbeddingRetriever. In the indexing pipelines, you only compute embeddings after storing the documents: indexing.py#L61-L64. Does this have any advantages over using the EmbeddingRetriever as a node in the pipeline, like I did in one of my recent projects (example below)?

pipe.add_node(
    component=EmbeddingRetriever(
        **embedding_settings,
    ),
    name="Retriever",
    inputs=[last_node],
)

pipe.add_node(
    component=PgvectorStore(**store_settings),
    name="DocumentStore",
    inputs=["Retriever"],
)

Which way should I use and what differences do they have?