Lee-W commented on code in PR #76999: URL: https://github.com/apache/airflow/pull/76999#discussion_r4615274286
########## providers/common/ai/docs/provider_fallback.rst: ########## @@ -140,9 +140,15 @@ vendor can answer the same prompt now; a bad prompt, an exhausted quota on every a permanent data error still has to be decided by the retry policy. Adding a chain changes what the retry layer sees. When every connection in the chain -fails, the exception the task raises is ``pydantic_ai.exceptions.FallbackExceptionGroup``, -not the last provider's own exception, so retry rules matched against a provider-specific -exception type stop matching. Before adding a chain to a connection that Dags already use, +fails, the exception the task raises is +:class:`~airflow.providers.common.ai.fallback_exceptions.FallbackChainExhaustedError`, a subclass of Review Comment: * the operators and `@task.llm` / `@task.agent` raise `FallbackChainExhaustedError` * a plain `@task` running a `PydanticAIHook` agent gets the bare `FallbackExceptionGroup`, without `member_errors` or the journal counts `provider_fallback.rst` has the full explanation and the other two point to it. Rules still match the group and its last member on both paths. -- This is an automated message from the Apache Git Service. To respond to the message, please log on to GitHub and use the URL above to go to the specific comment. To unsubscribe, e-mail: [email protected] For queries about this service, please contact Infrastructure at: [email protected]
