fix: retry on APIConnectionError in flex_llm.py
@zen-agentchecks…zen/fix-connection-retry-ucsrom → master1 files · +17 −1updated 1mo agoloading diff…
@zen-agentchecks…zen/fix-connection-retry-ucsrom → master1 files · +17 −1updated 1mo ago@zen-agentchecks…zen/fix-connection-retry-ucsrom → master1 files · +17 −1updated 1mo agoConnection errors (APIConnectionError) in flex_chat_completions_parse were falling through to the generic except Exception handler and failing immediately without retry, while timeouts (APITimeoutError) and rate limits (APIStatusError 429/408) were properly retried with exponential backoff.
This caused 33 permanent LLM call failures in a single ingestion pipeline cycle during a brief network instability period (up from 2 in previous runs). Each connection error failed the entire session analysis permanently instead of being retried.
The fix adds an explicit except APIConnectionError handler with the same retry+backoff logic used for timeouts.
ruff check and ruff format — no lint or formatting issues[ERROR] Unexpected error on attempt 1: Connection error.) back to flex_llm.py:161 generic exception handlerAPIConnectionError is a distinct exception type in the OpenAI SDK (not a subclass of APIStatusError or APITimeoutError)APITimeoutError patternOrigin: cron-5d55c321e47a / zen-cron-18502cb01ac3