Bedrock Agent InvokeAgent API throttling despite Service Quota increase

Would suggest:

1. Enable Exponential Backoff with Jitter • AWS recommends implementing retries with exponential backoff and random jitter to avoid retry storms:

for attempt in range(retries):
    try:
        response = client.invoke_agent(...)
        return response
    except client.exceptions.ThrottlingException:
        wait_time = delay * (2 ** attempt) + random.uniform(0, 1)
        time.sleep(wait_time)

2. Use Cross-Region Inference Profiles • Bedrock maintains independent capacity pools per region. • You can route requests across regions (e.g., ap-northeast-1, us-west-2) to avoid localized throttling.
3. Contact AWS Support for Agent-Specific Limits • Some quotas (like InvokeAgent RPM) are not adjustable via Service Quotas UI. • Open a support case and request: o Specific InvokeAgent RPM and TPM limits for your agent. o Increase for those limits if needed.
4. Check for Agent Configuration Bottlenecks • If your agent uses complex action groups, knowledge bases, or trace enablement, it may consume more tokens per request. • Try disabling enableTrace or simplifying the agent flow.

Hey,

Hope you're keeping well.

The InvokeAgent API has its own internal throttling limits that aren’t exposed in the Service Quotas console, so increasing the “APIs per Agent” quota won’t remove those caps. Bedrock recommends implementing retries with exponential backoff and jitter, as even low request rates can hit per‑second or burst limits depending on agent configuration. Also check CloudWatch metrics for ThrottlingException under the AWS/Bedrock namespace to confirm actual request patterns. If the throttling persists at very low rates, open an AWS Support case with the agent ID and timestamps so they can verify backend limits specific to your account and region.

Thanks and regards,
Taz