[localserver] Streaming Lambda functions can only be invoked once.

[sebsto]

Expected behavior

When using the LocalServer for testing locally a Streaming Lambda function, we expect the user to invoke the Lambda function multiple times without having to restart the server between invocations.

This is the behavior of the server with non-streaming Lambda function.

Actual behavior

When invoking the local test server (POST /invoke) with a streaming Lambda function, the Local Lambda function is correctly run and returns the expected value.

However, the local server never handles the subsequent GET /next issued by the lambda runtime client.

This makes impossible to invoke the function twice on the local test server.

Steps to reproduce

cd Examples/Streaming
swift run

From another terminal

curl -v --output test1.txt --header "Content-Type: application/json" --data '"1111"' http://127.0.0.1:7000/invoke

Wait three seconds and you will see result in test1.txt

$ cat test1.txt 
{"headers":{"Content-Type":"text/plain","x-my-custom-header":"streaming-example"},"statusCode":418}Number: 1
Number: 2
Number: 3
Streaming complete!

Now invoke the function a second time

curl -v --output test2.txt --header "Content-Type: application/json" --data '"2222"' http://127.0.0.1:7000/invoke

The call blocks and never returns.

On the server trace, we can see

2025-10-16T13:34:34+0200 trace LambdaRuntime : URI=POST /invoke requestId=615386897861416 response requestId=615386897861416 [AWSLambdaRuntime] /invoke returning
2025-10-16T13:34:34+0200 trace LambdaRuntime : [AWSLambdaRuntime] Done handling the connection
2025-10-16T13:34:40+0200 trace LambdaRuntime : [AWSLambdaRuntime] Handling a new connection
2025-10-16T13:34:40+0200 trace LambdaRuntime : Body="1111" URI=POST /invoke [AWSLambdaRuntime] Processing request
2025-10-16T13:34:40+0200 trace LambdaRuntime : URI=POST /invoke requestId=615396149012541 [AWSLambdaRuntime] /invoke received invocation, pushing it to the pool and wait for a lambda response

The runtime never sends the next GET /next request

Here is the trace from a non-streaming function for comparison:

2025-10-16T13:40:57+0200 trace LambdaRuntime: aws-request-id=615770142842041 [AWSLambdaRuntime] Handler finished processing invocation
2025-10-16T13:40:57+0200 trace LambdaRuntime: [AWSLambdaRuntime] Waiting for next invocation
2025-10-16T13:40:57+0200 trace LambdaRuntime: URI=GET /2018-06-01/runtime/invocation/next [AWSLambdaRuntime] Processing request
2025-10-16T13:40:57+0200 trace LambdaRuntime: URI=GET /2018-06-01/runtime/invocation/next [AWSLambdaRuntime] /next waiting for /invoke
2025-10-16T13:40:57+0200 trace LambdaRuntime: [AWSLambdaRuntime] Done handling the connection

If possible, minimal yet complete reproducer code (or URL to code)

No response

What version of this project (swift-aws-lambda-runtime) are you using?

2.2.0

Swift version

n/a

Amazon Linux 2 docker image version

n/a

[sebsto]

When deployed on AWS, the runtime behaves correctly

➜  Streaming git:(main) ✗ curl https://gs2njl3irkfrhmk3cneplyww740ivwqs.lambda-url.us-east-1.on.aws --user "$AWS_ACCESS_KEY_ID":"$AWS_SECRET_ACCESS_KEY"  --aws-sigv4 "aws:amz:us-east-1:lambda" --no-buffer             
Number: 1
Number: 2
Number: 3
Streaming complete!
➜  Streaming git:(main) ✗ curl https://gs2njl3irkfrhmk3cneplyww740ivwqs.lambda-url.us-east-1.on.aws --user "$AWS_ACCESS_KEY_ID":"$AWS_SECRET_ACCESS_KEY"  --aws-sigv4 "aws:amz:us-east-1:lambda" --no-buffer
Number: 1
Number: 2
Number: 3
Streaming complete!

And the two invocations happen on the same microVM.

[sebsto]

Looking at the handleConnection method, when a streaming response is detected:

HEAD: Pushes a response to the pool (lines 244-250)
BODY: Pushes each chunk to the pool (lines 257-261)
END: Pushes final response to the pool (lines 267-271)
But there's no acknowledgment sent back to the Lambda runtime client for streaming requests! The streaming path completely bypasses processRequestAndSendResponse where the acknowledgment would normally be sent.

Compare this to the non-streaming path in processRequestAndSendResponse (lines 390-408):

// tell the Lambda function we accepted the response
return try await sendResponse(
    .init(id: requestId, status: .accepted, final: true),
    outbound: outbound,
    logger: logger
)

[sebsto]

Fixed in #590