fix: Missing error handling in embedding service

app/services/embedding.py

@@ -1,11 +1,16 @@
 """Embedding service — wraps LiteLLM for provider-agnostic vector generation."""
+import asyncio
+import logging
 
 from __future__ import annotations
 
 from litellm import aembedding
 
 # Default embedding model — small, fast, cheap
 DEFAULT_EMBEDDING_MODEL = "text-embedding-3-small"
+# Retry configuration
+MAX_RETRIES = 3
+RETRY_DELAY = 1.0  # seconds
 DEFAULT_EMBEDDING_DIMENSIONS = 1536
 MAX_BATCH_SIZE = 128
 
@@ -36,9 +41,22 @@ async def embed_texts(
         batch = texts[i : i + MAX_BATCH_SIZE]
         kwargs: dict = {"model": model, "input": batch}
         if api_key:
-            kwargs["api_key"] = api_key
-
-        response = await aembedding(**kwargs)
+        # Retry logic for API failures
+        for attempt in range(MAX_RETRIES):
+            try:
+                response = await aembedding(**kwargs)
+                batch_embeddings = [item["embedding"] for item in response.data]
+                all_embeddings.extend(batch_embeddings)
+                break  # Success, exit retry loop
+            except Exception as e:
+                if attempt == MAX_RETRIES - 1:
+                    # Final attempt failed, re-raise the exception
+                    logging.error(f"Embedding API failed after {MAX_RETRIES} attempts: {e}")
+                    raise RuntimeError(f"Failed to generate embeddings after {MAX_RETRIES} attempts: {e}") from e
+                else:
+                    # Transient failure, wait and retry
+                    logging.warning(f"Embedding API attempt {attempt + 1} failed: {e}. Retrying...")
+                    await asyncio.sleep(RETRY_DELAY * (2 ** attempt))  # Exponential backoff
         batch_embeddings = [item["embedding"] for item in response.data]
         all_embeddings.extend(batch_embeddings)