AWS Builder Center is the official home for builders on AWS. Share and read what others are working on, follow people who inspire you, explore training and workshops, and find tools to support what you're building.
Service unavailable exception thrown for one agent, but not for multi-agent amazon bedrock
Hi, I keep receiving this error: "A service unavailable exception was thrown from Bedrock when invoking the model. Retry the request."
When I ask my bedrock agent a question that prompts it to invoke a function in my action group with Claude 3.5 Sonnet. When I ask the same question for another agent I've created with collaborator agents (with access to the same action group and knowledge base) using 3.5, I receive a response.
When I look at the trace for the agent receiving the error, it correctly executes the action group function, but fails to create/return an answer.
Does anyone know why this might happen? Is it more efficient to have multi-agent systems than singular agents or is this an issue caused by something else?
- Tags
- Amazon Bedrock Prompt ManagementAmazon Bedrock FlowsAnthropic Claude in Amazon BedrockAmazon Bedrock
- Language
- English
asked a year ago1K views
- Newest
- Most votes
- Most comments
Based on the information provided, it appears you're experiencing a "Service unavailable exception" with a single Bedrock agent using Claude 3.5 Sonnet, while a multi-agent system with the same model works correctly.
This issue could be related to several factors:
-
Action group naming conventions: One common issue with Bedrock agents is related to action group names. If your action group name contains hyphens ("-"), this can cause errors. Try replacing any hyphens with underscores ("_") in your action group name.
-
Permissions issues: The multi-agent system might have different or more comprehensive permissions set up. For multi-agent collaboration, specific IAM policies need to be attached to the service role, including permissions for
bedrock:GetAgentAliasandbedrock:InvokeAgent. -
Model timeout or resource constraints: The single agent might be hitting timeout limits or resource constraints that the multi-agent system avoids through its distributed approach.
-
State handling differences: Multi-agent systems handle state differently, which might be avoiding whatever condition is causing the error in the single agent.
To troubleshoot:
- Check your action group naming conventions
- Review the IAM permissions for both agent setups
- Look at the complete logs and traces to identify where exactly the failure occurs
- Try implementing more extensive logging in your Lambda function
- Consider restarting or redeploying the problematic agent
It's not necessarily that multi-agent systems are more "efficient" in general, but they do distribute workload differently, which might be avoiding the specific condition causing your error.
Sources
Issue with Bedrock Agents and Lambda Function – Model Timeout/Error | AWS re:Post
Bedrock agent "Internal Server Exception" | AWS re:Post
Create a service role for Amazon Bedrock Agents - Amazon Bedrock
answered a year ago
Relevant content
asked 8 months ago
asked 2 years ago
asked 10 months ago
- AWS OFFICIALUpdated a year ago
My action groups have no hyphens, besides the lambda function itself having hyphens. But they are configured to the same lambda function for both agents.