This final method sets __active to False only after calling disconnect on the underlying connection

[rdavydov]

I recently encountered this error by chance when looking at my logs. I checked the last week's worth of logs and it looks like it had only occurred once. The relevant lines are:

20/04/24 02:17:02 - DEBUG - TwitchChannelPointsMiner.classes.WebSocketsPool - [on_message]: #0 - Received: {"type":"MESSAGE","data":{"topic":"video-playback-by-id.###","message":"{\"server_time\":1713575822,\"type\":\"stream-down\"}"}}
20/04/24 02:17:02 - INFO - TwitchChannelPointsMiner.classes.Chat - [stop]: Leave IRC Chat: ###
20/04/24 02:17:02 - ERROR - TwitchChannelPointsMiner.classes.Chat - [start]: Exception raised: file descriptor cannot be a negative integer (-1). Thread is active: True
// above line repeated 195 times
20/04/24 02:17:02 - INFO - TwitchChannelPointsMiner.classes.entities.Streamer - [set_offline]: Streamer(username=###, channel_id=###, channel_points=###) is Offline!

This is the code that processes that message:

Twitch-Channel-Points-Miner-v2/TwitchChannelPointsMiner/classes/WebSocketsPool.py

Lines 233 to 235 in 62c01c4

	elif message.type == "stream-down":
	if ws.streamers[streamer_index].is_online is True:
	ws.streamers[streamer_index].set_offline()

Looking through the code, in response to a stream-down message the miner will:

1. Call Streamer.set_offline() for the appropriate Streamer
2. This sets Streamer.is_online = False
3. Then it calls Streamer.toggle_chat()
4. Assuming Streamer.settings.chat is ChatPresence.ONLINE it will then call Streamer.leave_chat() (Incidentally this is likely why the issue doesn't show up for ChatPresence.ALWAYS)
5. This calls self.irc_chat.stop() which is a method on ThreadChat in Chat.py
6. This ends up calling self.chat_irc.die() which is a method on ClientIRC:
   

   Twitch-Channel-Points-Miner-v2/TwitchChannelPointsMiner/classes/Chat.py

   Lines 49 to 51 in 62c01c4

   	def die(self, msg="Bye, cruel world!"):
   	self.connection.disconnect(msg)
   	self.__active = False

I believe this is where a race condition happens. This final method sets __active to False only after calling disconnect on the underlying connection. Disconnect ends up invalidating the underlying Socket and I think the File Descriptor will be -1 for a closed Socket, however I'm no expert on this. This is likely the -1 in the error message.

At the same time as this is happening we also have ClientIRC.start. In there is the while loop that processes the IRC connection:

Twitch-Channel-Points-Miner-v2/TwitchChannelPointsMiner/classes/Chat.py

Lines 40 to 47 in 62c01c4

	while self.__active:
	try:
	self.reactor.process_once(timeout=0.2)
	time.sleep(0.01)
	except Exception as e:
	logger.error(
	f"Exception raised: {e}. Thread is active: {self.__active}"
	)

As previously noted by @rdavydov, that call to logger.error is the source of the error we see in the log. So the Exception that caused it must have come from that try block.

Since these 2 things (processing the connection and responding to stream-down) are happening in 2 different Threads it's entirely possible, albeit unlikely, for execution to change Threads mid-disconnection after the socket becomes invalid but before it sets the __active state. And since this while loop uses that state to determine whether to continue trying to process the connection, it could end up trying to read from a closed socket.

I believe a simple fix to this is to set __active = False prior to starting the disconnection, just swap those 2 lines in ClientIRC.die:

def die(self, msg="Bye, cruel world!"):
    self.__active = False
    self.connection.disconnect(msg)

That should allow the while loop to exit early. It doesn't look like anything else relies on __active so there should be no other side-effects.

Originally posted by @mpforce1 in #250 (comment)