scriptoriumadmin/litellm
1
0
Fork 0
mirror of https://github.com/BerriAI/litellm.git synced 2026-09-08 22:21:35 +00:00
Code Issues Projects Releases Packages Wiki Activity Actions 78

[Perf] - Cut memory leak in half (#17784)

Browse source 
Stop passing references to large objects to fire-and-forget tasks.
This commit is contained in:
Alexsander Hamir 2025-12-10 12:31:58 -08:00 • committed by GitHub
parent 5ee32167c0
commit ccb6cd880a
No known key found for this signature in database
GPG key ID: B5690EEEBB952194
2 changed files with 8 additions and 13 deletions
Show stats Download patch file Download diff file Expand all files Collapse all files

15
litellm/proxy/db/db_spend_update_writer.py
View file

@ -6,6 +6,7 @@ Module responsible for
"""
import asyncio
import copy
import json
import os
import random
@ -162,14 +163,14 @@ class DBSpendUpdateWriter:
asyncio.create_task(
self._update_tag_db(
response_cost=response_cost,
request_tags=payload.get("request_tags"),
request_tags=copy.deepcopy(payload.get("request_tags")),
prisma_client=prisma_client,
)
)
if disable_spend_logs is False:
await self._insert_spend_log_to_db(
payload=payload,
payload=copy.deepcopy(payload),
prisma_client=prisma_client,
)
else:
@ -179,34 +180,34 @@ class DBSpendUpdateWriter:
asyncio.create_task(
self.add_spend_log_transaction_to_daily_user_transaction(
payload=payload,
payload=copy.deepcopy(payload),
prisma_client=prisma_client,
)
)
asyncio.create_task(
self.add_spend_log_transaction_to_daily_end_user_transaction(
payload=payload,
payload=copy.deepcopy(payload),
prisma_client=prisma_client,
)
)
asyncio.create_task(
self.add_spend_log_transaction_to_daily_team_transaction(
payload=payload,
payload=copy.deepcopy(payload),
prisma_client=prisma_client,
)
)
asyncio.create_task(
self.add_spend_log_transaction_to_daily_org_transaction(
payload=payload,
payload=copy.deepcopy(payload),
org_id=org_id,
prisma_client=prisma_client,
)
)
asyncio.create_task(
self.add_spend_log_transaction_to_daily_tag_transaction(
payload=payload,
payload=copy.deepcopy(payload),
prisma_client=prisma_client,
)
)

6
litellm/proxy/utils.py
View file

@ -1,6 +1,5 @@
import asyncio
import copy
import gc
import hashlib
import json
import os
@ -3379,7 +3378,6 @@ class ProxyUpdateSpend:
headers={"Content-Type": "application/json"},
)
del json_data
gc.collect()
if response.status_code == 200:
prisma_client.spend_log_transactions = (
prisma_client.spend_log_transactions[
@ -3401,9 +3399,6 @@ class ProxyUpdateSpend:
)
# Explicitly clear batch memory
del batch, batch_with_dates
# Only run gc every 5 batches to reduce overhead
if j % (BATCH_SIZE * 5) == 0:
gc.collect()
prisma_client.spend_log_transactions = (
prisma_client.spend_log_transactions[len(logs_to_process) :]
@ -3429,7 +3424,6 @@ class ProxyUpdateSpend:
finally:
# Clean up logs_to_process after all processing is complete
del logs_to_process
gc.collect()
@staticmethod
def disable_spend_updates() -> bool: