CME on Netty Server IO Thread

[ams2990]

I'm listening to PacketType.Play.Server.ENTITY_EQUIPMENT packets and editing the contents to remove information that Mojang never should send to clients. You can the code I'm using here.

I have observed that players occasionally disconnect with the following super helpful log entries:

[20:37:20] [Netty Server IO #2/ERROR]: java.util.ConcurrentModificationException 
[20:37:20] [Server thread/INFO]: DeathWalking lost connection: Disconnected
[20:37:20] [Server thread/INFO]: DeathWalking left the game.

It's pretty hard to reproduce, unfortunately. It seems to reproduce more often during PvP, so I gathered four other people on a test server, spawned a bunch of armor and pots, and whacked each other for a while. We got screenshots of the client disconnect message, two very similar ones: http://imgur.com/a/etwzX.

I'm not just sure where the fault lies. My guess is something on the Netty threads is throwing the CME and the packet that's being sent to the client is malformed, causing the disconnect.

[dmulloy2]

This looks to be the same issue as dmulloy2/ProtocolLib#36. Can you set debug=true in the server.properties and debug: true in the ProtocolLib config? It should print out a (more) helpful message.

[dmulloy2]

That code didn't seem to work. I modified it a bit here and didn't get any CME's.

[ams2990]

Like I said, it's hard to reproduce. It tends to happen in groups, but I can't quite figure out exactly what triggers it. When I did get the above error, I had ProtocolLib debug: true. I didn't have Minecraft debug=true, but I'll have to try that and see if I get anything more interesting.

[dmulloy2]

Any more info on this?

[ams2990]

Sorry, my free time has been consumed by Starcraft and Fallout :-)

I'll set debug=true on the main server and see what happens.

[dmulloy2]

I'd like to get this fixed before releasing 3.6.5, can I get the debug info?

[ams2990]

Sorry for the delayed response. I've had debug=true on and haven't gotten any more info. I got one earlier today and all we get is the name of the exception. No trace body :/

[ams2990]

Given how infrequently this reproduces, my guess is that this is a race condition. Is it possible the packet is being read so it can be sent while I'm still modifying it?

[dmulloy2]

It could be a race condition, but ProtocolLib should hold on to the packet until you're finished modifying it. It's likely that the CME is internal (PL or Minecraft itself), since any exceptions from onPacketSending and onPacketReceiving should be caught by PL. Unfortunately, there's not much I can do without a stack body :/

[dmulloy2]

Any more information on this?

[ams2990]

Sorry, I couldn't get any useful debugging information. I eventually just disabled the feature, since it wasn't worth disconnecting users over.

[Kneesnap]

I'm having this issue too, interestingly also with ENTITY_EQUIPMENT. Sometimes it's NullPointerException, sometimes its ConcurrentModificationException. It's really frustrating.