InvokeAgentCommand timeout (Bedrock)

[abarone-btf]

Checkboxes for prior research

• I've gone through Developer Guide and API reference
• I've checked AWS Forums and StackOverflow.
• I've searched for previous similar issues and didn't find any solution.

Describe the bug

Hi all, I'm invoking a custom agent created in Bedrock using the InvokeAgentCommand. However, when a model call (in this case, Claude Sonnet 4) takes longer than 60 seconds, it fails with an InternalServerException. The exception details are all undefined, so I'm unsure whether this is due to a timeout quota on the invoke command (which isn’t mentioned in the documentation) or something else.

InternalServerException: Internal Server Exception
    at de_InternalServerExceptionRes (file:///var/task/bundle.mjs:1118501:21)
    at de_InternalServerException_event (file:///var/task/bundle.mjs:1118686:10)
    at process.processTicksAndRejections (node:internal/process/task_queues:105:5)
    at async file:///var/task/bundle.mjs:1118591:34
    at async Object.deserializer (file:///var/task/bundle.mjs:1117256:37)
    at async SmithyMessageDecoderStream.asyncIterator (file:///var/task/bundle.mjs:1117151:28)
    at async BedrockClientAgentRuntimeService.invokeCallback (file:///var/task/bundle.mjs:1119921:24)
    at async DatadogBedrockAgentService.invokeWrapper (file:///var/task/bundle.mjs:1120387:20)
    at async BedrockClientAgentRuntimeService.invokeAgent (file:///var/task/bundle.mjs:1119882:53)
    at async BedrockAgentsServiceImpl.invokeAgent (file:///var/task/bundle.mjs:1132609:34) {
  '$fault': 'server',
  '$metadata': {
    httpStatusCode: undefined,
    requestId: undefined,
    extendedRequestId: undefined,
    cfId: undefined
  },
  reason: undefined

My trace call is:

{
    "level": "INFO",
    "time": "2025-07-21T19:13:52.867Z",
    "pid": 5,
    "hostname": "169.254.51.131",
    "callbackWaitsForEmptyEventLoop": true,
    "functionVersion": "$LATEST",
    "functionName": "dev-lambda",
    "memoryLimitInMB": "2048",
    "logGroupName": "/aws/lambda/dev-lambda",
    "logStreamName": "2025/07/21/dev-lambda[$LATEST]traceid",
    "invokedFunctionArn": "arn:aws:lambda:us-east-1:ACCOUNT:function:dev-lambda",
    "awsRequestId": "e74a12e1-14a4-5570-8c2c-7219857bdd23",
    "service": "dev-btf-main-agents",
    "orchestrationTrace": {
        "modelInvocationInput": {
            "foundationModel": "anthropic.claude-sonnet-4-20250514-v1:0",
            "inferenceConfiguration": {
                "stopSequences": [
                    "</answer>"
                ],
                "temperature": 1
            },
            "text": "{\"system\":"..."}",
            "traceId": "c010b976-465c-4568-a587-a82e61bed5f5-1",
            "type": "ORCHESTRATION"
        }
    }
}

Do you know what may be happening here?

Thanks!

Regression Issue

• Select this option if this issue appears to be a regression.

SDK version number

@aws-sdk/client-bedrock-agent-runtime@3.848.0

Which JavaScript Runtime is this issue in?

Node.js

Details of the browser/Node.js/ReactNative version

Node 22

Reproduction Steps

Custom agente that calls a model that takes longer than 60 seconds

Observed Behavior

The agent throws an InternalServerException

Expected Behavior

To finish the invoke agent call without problem

Possible Solution

No response

Additional Information/Context

No response

[aBurmeseDev]

Hi @abarone-btf - thanks for reaching out. While I look more into it internally, can you share your code snippet? Does this only fail when a call takes longer than 60s and is it consistent? Also are you running this on Lambda?

[abarone-btf]

Thank you @aBurmeseDev ! 🙏 . Of course, this is a code snippet for the invoke agent call:

const input: InvokeAgentRequest = {
      agentId: agentRequest.agentId,
      agentAliasId: agentRequest.agentVersion,
      sessionId: agentRequest.conversationId,
      inputText: agentRequest.message,
      enableTrace: true,
      sessionState: {
        promptSessionAttributes: agentRequest.metadata ? { ...agentRequest.metadata } : {},
        sessionAttributes,
        knowledgeBaseConfigurations,
      },
    };

const command = new InvokeAgentCommand(input);
const output = await this.bedrockAgentRuntimeClient.send(command) // bedrockAgentRuntimeClient uses de default setup

Yes, we are running this inside a lambda. And yes the error is only shown when a LLM model is called (sonnet 4 in this case) and, on the ~60sec, it throws the error, the problem is that the message its really vague, so we dont know what may be happening.

Thanks for your help!

[abarone-btf]

Hi!, any news? 🙏 @aBurmeseDev

[aBurmeseDev]

Hi @abarone-btf - thanks for your patience.

I attempted a few times to reproduce this error on my end but was unable to. I've also reached out to service team for their insights and awaiting response. Meanwhile, here're a few suggestions:

• Try a different model claude-3.5-sonnet or claude-3.7-sonnet. If that works, this issue seems to be related to the temporary unavailability of the model.
• As this error comes from service side and according to their docs, they recommend retrying the request. Here's modified code with proper timeout and retry on client:

const bedrockAgentRuntimeClient = new BedrockAgentRuntimeClient({
  region: "us-west-1", // your region
  requestHandler: {
    requestTimeout: 300000, // 5 minutes
    connectionTimeout: 30000, // 30 seconds
  },
  maxAttempts: 4, // Avoid retries for long operations
});

• Since this's running on Lambda, I would also check on your Lambda Timeout but this's unlikely to be the root cause.

I'll check back in when I hear back from service team. Apologies for the inconvenience.

[abarone-btf]

Hi @aBurmeseDev thank you for your response.

• In your attempts the agent does call an LLM which last more than 60secs?, or you could not reproduce this behavior at all?.

• We tried with claude 3.7 and we encounter the same error on the same timecap (60 secs), claude 3.5 does work but it does because its faster (and the results are not what we want), so this solution does not work in our case.

• We already try the request timeout and the connection timeout settings but it does not work (besides, the default values are 0 which kind of means "infinite"), the agent call does not timeout but we are suspicious that the internal call from the bedrock agent to the LLM model does timeout, but we cant control that (as we know of).

• We indeed have retries, but because it does fail constantly we tend to exhaust them (when it does work is because we have luck and the call was less than 60 secs).

We'd greatly appreciate your help escalating this issue internally 🙏. Our team is committed to building on Bedrock and relies heavily on Claude Sonnet 4/3.7. Having clarity (and ideally resolution) around this timeout behavior would be instrumental for maintaining trust in the platform