Uncaught TypeError when using keep-alive HTTP agent

[cdhowie]

By simply adding this code to our application:

AWS.config.update({
    httpOptions: {
        agent: new http.Agent({ keepAlive: true })
    }
});

The following exception is thrown some time later, after numerous successful operations:

     Uncaught TypeError: cb is not a function
      at afterWrite (_stream_writable.js:354:3)
      at onwrite (_stream_writable.js:345:7)
      at WritableState.onwrite (_stream_writable.js:89:5)
      at Socket._writeGeneric (net.js:706:5)
      at Socket._writev (net.js:711:8)
      at doWrite (_stream_writable.js:298:12)
      at clearBuffer (_stream_writable.js:387:5)
      at Socket.Writable.uncork (_stream_writable.js:237:7)
      at ClientRequest._flushOutput (_http_outgoing.js:673:10)
      at ClientRequest.OutgoingMessage._flush (_http_outgoing.js:647:16)
      at _http_client.js:156:10
      at callSocketMethod (_http_client.js:552:7)
      at ClientRequest.onSocket (_http_client.js:557:7)
      at tickOnSocket (_http_client.js:523:7)
      at onSocketNT (_http_client.js:535:5)

Changing keepAlive to false eliminates the errors.

Node 4.5.0 and aws-sdk 2.6.2.

[chrisradek]

@cdhowie
Are you making calls across multiple services? Do you know if it's a specific operation that always throws the exception? How long is 'some time later'?

[cdhowie]

These service tests exclusively run against DynamoDB. At the point where it fails, there are multiple concurrent requests and they are all DeleteItem requests. It will consistently fail on that test, and I haven't been able to locate any bug in my code (yet).

It probably makes it through 50 or so requests before failing, and I know that some of the other tests involve concurrent deletions. Everything works fine until I enable keepalive.

[cdhowie]

My spidey sense says this is actually a bug in Node's http.Agent class, but I'm not super-familiar with Node's internals so I'm not well-equipped to go about debugging them to make sure. So I'm not sure if I should leave this issue open or go report it to Node. :/

The first stack frame shown is this Node-internal code:

function afterWrite(stream, state, finished, cb) {
  if (!finished)
    onwriteDrain(stream, state);
  state.pendingcb--;
  cb();
  finishMaybe(stream, state);
}

The cb() invocation is what's failing. This parameter gets set earlier from state.writecb, which, according to the comments in that file, are supplied when invoking write() on the stream. Apparently the callback is required, at least in Node v4? That doesn't match the docs, but it wouldn't be the first time the docs are wrong.

[cdhowie]

I am seeing the same behavior on Node 6.6.0 using the stock http.Agent with keepAlive: true.

[chrisradek]

Closing old issues. Please open a new issue if you are still encountering this behavior.

[lock]

This thread has been automatically locked since there has not been any recent activity after it was closed. Please open a new issue for related bugs and link to relevant comments in this thread.