Platform for stateful agents: AI with advanced memory that can learn and self-improve over time.
feat: Throttle pinecone uploads (#3373)
Matthew Zhoud0d67d8b9be7
diff --git a/letta/constants.py b/letta/constants.pyindex 05711ab3b..210319ab6 100644--- a/letta/constants.py+++ b/letta/constants.py@@ -378,3 +378,4 @@ PINECONE_MAX_RETRY_ATTEMPTS = 3PINECONE_RETRY_BASE_DELAY = 1.0 # secondsPINECONE_RETRY_MAX_DELAY = 60.0 # secondsPINECONE_RETRY_BACKOFF_FACTOR = 2.0+PINECONE_THROTTLE_DELAY = 0.75 # seconds base delay between batchesdiff --git a/letta/helpers/pinecone_utils.py b/letta/helpers/pinecone_utils.pyindex 7caa52808..e014061b6 100644--- a/letta/helpers/pinecone_utils.py+++ b/letta/helpers/pinecone_utils.py@@ -31,6 +31,7 @@ from letta.constants import (PINECONE_RETRY_BASE_DELAY,PINECONE_RETRY_MAX_DELAY,PINECONE_TEXT_FIELD_NAME,+ PINECONE_THROTTLE_DELAY,)from letta.log import get_loggerfrom letta.schemas.user import User@@ -256,6 +257,13 @@ async def upsert_records_to_pinecone_index(records: List[dict], actor: User):logger.debug(f"[Pinecone] Upserting batch {batch_num}/{total_batches} with {len(batch)} records")await dense_index.upsert_records(actor.organization_id, batch)+ # throttle between batches (except the last one)+ if batch_num < total_batches:+ jitter = random.uniform(0, PINECONE_THROTTLE_DELAY * 0.2) # ±20% jitter+ throttle_delay = PINECONE_THROTTLE_DELAY + jitter+ logger.debug(f"[Pinecone] Throttling for {throttle_delay:.3f}s before next batch")+ await asyncio.sleep(throttle_delay)+logger.info(f"[Pinecone] Successfully upserted all {len(records)} records in {total_batches} batches")