[Python] Fan-out from a Node to Multiple Instances of a Node Function in LangChain's LangGraph?

I'm designing a LangGraph application and have a scenario where I need to process a document in parallel. My current conceptual flow, similar to the attached diagram involves:

Document Chunking: An initial node takes a document and splits it into multiple chunks.
Parallel Processing: Each of these chunks needs to be processed independently by the same document processing logic.
Post-processing: Once all chunks are processed, their results need to be aggregated in a final post-processing step.


Essentially, I want to "fan out" from a single "document chunking node" to multiple, concurrent executions of a "document processing node function," each handling a different chunk. After these parallel executions complete, I need to "fan in" to a "post-processing" node.

Is this kind of fan-out and fan-in pattern natively supported or easily achievable in LangGraph? If so, what are the recommended patterns or features to use? Specifically, I'm looking for guidance on:

How to define a graph structure that allows a single output from one node to trigger multiple instances of the next node.
How to pass different inputs (e.g., individual chunks) to each instance of the parallel processing node.
How to collect and aggregate the results from all parallel instances before proceeding to a final node.


Any code examples or conceptual explanations would be greatly appreciated!

Continue reading...