How to add LLM Fallback and Max retries in Agents

[sulakshana13]

Hello,

I'm new to ADK and I'm looking to implement fallback and max retry logic (with backoff) in case we hit rate limits when running an Agent.

I understand that LiteLLM supports retries and fallbacks and that ADK can work with LiteLLM, but I'm not sure how to integrate this functionality within an Agent.

Would appreciate some support on how to set up retries and fallbacks for an Agent in ADK.

[hangfei]

It's supported here: d4c63fc now.

[hangfei]

@joefernandez to see if we need to add documentation for this change. Thanks.

[adityapandey216]

Hi @hangfei . Is the fallback supported for Gemini native as well ?
model=Gemini(
model='gemini-2.0-flash'
retry_options=genai_types.HttpRetryOptions(
initial_delay=2,
max_delay=120,
attempts=7,
exp_base=2,
jitter=0.2,
http_status_codes=[429, 503]
)
),