Hi @petitoff

Thanks for your contribution! Have you looked at using any libraries for implementing the retry logic?

For example, ADK already uses tenacity for retry

https://github.com/google/adk-python/blob/main/src/google/adk/sessions/vertex_ai_session_service.py#L121

@pwwpche hi,

I haven't tried using libraries like tenacity before, but I could explore that option. However, when creating this PR, I wanted a simple, fast, and working solution that could be implement immediately (like a hot fix)

@pwwpche hi,

I haven't tried using libraries like tenacity before, but I could explore that option. However, when creating this PR, I wanted a simple, fast, and working solution that could be implement immediately (like a hot fix)

I would highly recommend refactor this change using tenacity then. With ~10 lines of code, it could achieve the same outcome as the hundreds of lines of code we have in the lite_llm.py. Since this implementation is in the core of ADK, let's prioritize a bit more on longer term maintainability.

Something like

from tenacity import retry
from tenacity import retry_if_result
from tenacity import RetryError
from tenacity import stop_after_attempt
from tenacity import wait_exponential

# ...

    retry_wrapped_acompletion = retry(
          stop=stop_after_attempt(self.stop_attempts),
          wait=wait_exponential(multiplier=1, min=1, max=3),
          retry=retry_if_result(lambda response: not response),
          reraise=True,
      )(acompletion)

  return await retry_wrapped_acompletion(...)

Thanks!