Merge pull request 'Batch Paperless chunk embeddings during indexing' (#73) from perf/batch-paperless-embeddings into main

blueprints/rag/logic.py

@@ -67,6 +67,9 @@ def _get_engine():
     return _get_engine._engine
 
 
+# Chunks per embedding request when indexing Paperless documents.
+EMBED_BATCH_SIZE = 32
+
 text_splitter = RecursiveCharacterTextSplitter(
     chunk_size=1000,  # chunk size (characters)
     chunk_overlap=200,  # chunk overlap (characters)
@@ -200,15 +203,25 @@ async def index_documents():
     splits = _sanitize_documents(splits)
     logger.info(f"Indexing {len(splits)} chunks from {len(documents)} documents")
     vector_store = _get_vector_store()
-    for i, split in enumerate(splits):
+    for start in range(0, len(splits), EMBED_BATCH_SIZE):
+        batch = splits[start : start + EMBED_BATCH_SIZE]
         try:
-            await vector_store.aadd_documents(documents=[split])
-        except Exception as e:
-            logger.error(
-                f"Failed to embed chunk {i} from {split.metadata.get('filename', 'unknown')}: {e}"
+            await vector_store.aadd_documents(documents=batch)
+        except Exception as batch_error:
+            # Embedding fails before anything is inserted; retry the batch one
+            # chunk at a time to name the chunk the embedding server rejects.
+            logger.warning(
+                f"Batch of {len(batch)} chunks failed ({batch_error}); retrying one at a time"
             )
-            logger.debug(f"Chunk content preview: {split.page_content[:200]!r}")
-            raise
+            for i, split in enumerate(batch, start):
+                try:
+                    await vector_store.aadd_documents(documents=[split])
+                except Exception as e:
+                    logger.error(
+                        f"Failed to embed chunk {i} from {split.metadata.get('filename', 'unknown')}: {e}"
+                    )
+                    logger.debug(f"Chunk content preview: {split.page_content[:200]!r}")
+                    raise
 
 
 async def fetch_obsidian_documents() -> list[Document]: