Interrupting task causes 'bad' connection to go back to pool

[yazd]

In an attempt to implement timeout for requestHTTP, I wrote the function below:

T timeout(T)(T delegate() expr, Duration period)
{
  auto event = createManualEvent();
  bool finishedNormally = false;

  Exception ex;
  T result;

  auto task = runTask({
    try { result = expr(); finishedNormally = true; }
    catch (Exception e) { ex = e; }
    finally { event.emit(); }
  });

  event.wait(period, 0);

  if (!finishedNormally)
    task.interrupt();

  if (ex !is null) throw ex;
  if (finishedNormally) return result;
  else throw new Exception("task interrupted");
}

I expected this to work correctly, however, as evidenced by the piece of code below, this is buggy:

import vibe.d;

void main()
{
  runTask({
    auto settings = new HTTPServerSettings();
    settings.port = 9889;

    listenHTTP(
      settings,
      (HTTPServerRequest req, HTTPServerResponse res)
      {
        sleep(5.seconds); // comment this for things to work correctly
        res.writeJsonBody(req.query["q"]);
      }
    );
  });

  runTask({
    timeout({ makeRequest("first_request"); return true; }, 2.seconds);
  });

  setTimer(6.seconds, {
    makeRequest("second_request");
  });

  runEventLoop();
}

void makeRequest(string query)
{
  auto res = requestHTTP("http://localhost:9889/?q=%s".format(query));
  scope(exit) res.dropBody();

  auto response = res.readJson().to!string();
  if (response != query)
    logWarn("unexpected response! Got '%s' while expecting '%s'.", response, query);
  else
    logInfo("everything is working correctly");
}

The code above creates a server that responds with the request's query param q after 5 seconds. However, the client timeouts after 2 seconds. And sends another request after 6 seconds from the start. The problem shows when the response from the first request is received by the second request. I believe that this is due to the connection being returned to the pool and reused by the second request.

The output from the code above is:

Listening for requests on http://[::]:9889/
Failed to listen on 0.0.0.0:9889
Task terminated with uncaught exception: Task interrupted.
unexpected response! Got 'first_request' while expecting 'second_request'.

Is this correct behavior from vibe.d and timeout should be implemented differently? Or in this case, because the connection wasn't finalized correctly, it shouldn't be returned to the pool and reused?

[yazd]

Any advice or possible workaround?