When using most vector store integrations, the docstore and index_store are not used, and everything goes into the vector store. There are some recent changes to the docstore that need to be translated to each vector store individually (i.e. index.ref_doc_info won't work when using a vector store integration yet)
In milvus, it's a slightly special case in that it's not able to store nodes directly (only text). This is what it looks like when inserting data for milvus
# Process that data we are going to insert
for result in embedding_results:
ids.append(result.id)
doc_ids.append(result.ref_doc_id)
texts.append(result.node.get_text())
embeddings.append(result.embedding)
try:
# Insert the data into milvus
self.collection.insert([ids, doc_ids, texts, embeddings])
logger.debug(
f"Successfully inserted embeddings into: {self.collection_name} "
f"Num Inserted: {len(ids)}"
)
https://github.com/jerryjliu/llama_index/blob/main/llama_index/vector_stores/milvus.py#L322