get_many returns shared ContentStream for documents with identical content

[absoludity]

Bug

MinioDocumentRepository.get_many() fetches unique content hashes in batch (step 3), then splices each document's metadata with its content stream (step 4). When two or more documents share the same content_multihash, they are assigned the same ContentStream object from content_results.

The first consumer that calls .read() on the stream exhausts it. All subsequent consumers get b"".

Reproduction

Upload the same file twice (producing two documents with identical content). Call get_many with both document IDs. The second document will have empty content_bytes after conversion.

This was observed in RBA when two spec sheet uploads of the same PDF produced two products with identical content_multihash. The first product listed had correct content; the second had content_bytes: "" and all parsed fields (id, name, description) as null.

Root cause

In get_many (around step 3-4):

# Step 3: one ContentStream per unique hash
content_results = self.get_many_binary_objects(...)

# Step 4: multiple documents can share the same stream object
content_stream = content_results[content_multihash]  # shared reference

Fix

Each document should receive its own independent stream. Options:

• Read the binary data once per hash, then wrap in a fresh BytesIO/ContentStream for each document that references it
• Or seek to 0 before returning each stream (works if the underlying stream is seekable)