holetron-lab/fleet-memory
1
0
Fork 0
Code Releases Packages Activity Actions

fix 400 retries on llm

Browse source
This commit is contained in:
Nicolò Boschi 2025-12-12 17:15:35 +01:00
parent 44e9571572
commit 70983f5817
1 changed files with 8 additions and 8 deletions
Show stats Download patch file Download diff file Expand all files Collapse all files

16
hindsight-api/hindsight_api/engine/llm_wrapper.py
View file

@ -268,9 +268,9 @@ class LLMProvider:
raise raise
except APIStatusError as e: except APIStatusError as e:
# Fast fail on 4xx client errors (except 429 rate limit and 498 which is treated as server error) # Fast fail only on 401 (unauthorized) and 403 (forbidden) - these won't recover with retries
if 400 <= e.status_code < 500 and e.status_code not in (429, 498): if e.status_code in (401, 403):
logger.error(f"Client error (HTTP {e.status_code}), not retrying: {str(e)}") logger.error(f"Auth error (HTTP {e.status_code}), not retrying: {str(e)}")
raise raise
last_exception = e last_exception = e
@ -408,13 +408,13 @@ class LLMProvider:
raise raise
except genai_errors.APIError as e: except genai_errors.APIError as e:
# Fast fail on 4xx client errors (except 429 rate limit) # Fast fail only on 401 (unauthorized) and 403 (forbidden) - these won't recover with retries
if e.code and 400 <= e.code < 500 and e.code != 429: if e.code in (401, 403):
logger.error(f"Gemini client error (HTTP {e.code}), not retrying: {str(e)}") logger.error(f"Gemini auth error (HTTP {e.code}), not retrying: {str(e)}")
raise raise
# Retry on 429 and 5xx # Retry on retryable errors (rate limits, server errors, and other client errors like 400)
if e.code in (429, 500, 502, 503, 504): if e.code in (400, 429, 500, 502, 503, 504) or (e.code and e.code >= 500):
last_exception = e last_exception = e
if attempt < max_retries: if attempt < max_retries:
backoff = min(initial_backoff * (2 ** attempt), max_backoff) backoff = min(initial_backoff * (2 ** attempt), max_backoff)