run_forever()'s restart condition never met

in this loop on line 208 of stream.py, self._running is never set to false in the exception. retries starts at 0, then incremented in the exception. This has it compare 1>3 which is false. This means that self._running is never set to false. Then in finally block on line 226, the break statement on line 228 will never be reached. Wouldn’t this stop it from restarting on the next pass of the loop?

        while True: # stream.py - line 208
            try:
                if not self._running:
                    await self._start_ws()
                    self._running = True
                    retries = 0
                    await self._consume()
            except websockets.WebSocketException as wse:
                retries += 1
                if retries > int(os.environ.get('APCA_RETRY_MAX', 3)):
                    await self.close()
                    self._running = False
                    raise ConnectionError("max retries exceeded")
                if retries > 1:
                    await asyncio.sleep(
                        int(os.environ.get('APCA_RETRY_WAIT', 3)))
                log.warn('websocket error, restarting connection: ' +
                         str(wse))
            finally:
                print(retries)
                if not self._running:
                    break
                await asyncio.sleep(0.01)```

[Chrisrdouglas]

@srgustafson8 Thank you for making a pull request with a fix. As for the 1006 error, I’m wondering if that has something to do with the way web sockets are implemented server side. If I hadn’t already switched to Polygon for my needs then I’d use Wireshark to sniff the traffic on my machine and start trouble shooting the issue that way.

Alpaca really needs engineers who are testing and using the software they’re providing 😕

[srgustafson8]

I have been taking a look at this and while the cause of the 1006 errors remain a mystery to me, the reconnect should kick in after one and get you going again.

I’m not a python expert, but @Chrisrdouglas assessment looks along the right lines. There was an update at some point since which brought await self._consume() out of the if not self._running: statement, but that doesn’t seem to reconnect the websocket in the case of a 1006 error.

I’ll raise a pull request with a fix and see what the maintainers think.