SelectConnection catches all errors and hides details.

[tango-taylor]

Originally from an SO question but I'll report it here via @lukebakken's request.

SelectConnection seems to catch all errors and doesn't provide any stack trace information. This makes debugging difficult.

For instance, here's an example where it divides by 0 in the _on_connection_open callback.

import pika

class AsyncServer():
    def __init__(self):
        self._channel = None

    def run(self):
        self._connection = self._connect()
        self._connection.ioloop.start()

    def _connect(self) -> pika.SelectConnection:
        return pika.SelectConnection(
            parameters=pika.ConnectionParameters(host='localhost'),
            on_open_callback=self._on_connection_open,
            on_close_callback=self._on_connection_closed
        )

    def _on_connection_open(self, connection):
        # Cause an error
        x = 1/0

    def _on_connection_closed(self, connection, reason):
        print("Reason for closing: ",reason)
        print("Exception class: ",reason.__class__)
        self._connection.ioloop.stop()


if __name__ == '__main__':
    recv = AsyncServer()
    recv.run()

And the output:

Reason for closing:  Stream connection lost: ZeroDivisionError('division by zero')
Exception class:  <class 'pika.exceptions.StreamLostError'>

Notice the exception is a StreamLostError instead of a ZeroDivisionError which squashes all details of the inner exception. traceback.print_exc() won't work here because the exception has already been caught.

Some possible fixes:

• Add a flag to disable exception handling and raise them to the top like BlockingConnection does
• Send traceback information to the callback provided by on_close_callback

[akselerando]

Is there a way to work around this issue? I have the same problem, which makes it difficult to find where the exception originally was raised.

[lukebakken]

This issue is on my radar, but I am very busy working on the next .NET client release.