Certain 1.8 PvP modded clients do not like Velocity

[astei]

I have seen some sporadic reports that some PvP-oriented clients for Minecraft (Badlion and Lunar, at least) may be having some issues when connected to a Velocity proxy. Further investigation is needed.

[roccodev]

What type of issues exactly?

I have a proxy running on 1.1.8 with some Lunar (and occasionally Badlion) traffic, but I have never seen any unexpected issues. (apart from the "connection reset by peer"s and read timeouts, but their volume is pretty normal).

I can inspect logs if further investigation is required.

[CoreyShupe]

connection reset by peer"s and read timeouts

@RoccoDev The frequency of this happening scales as you have more users and more server hopping. We're noticing this in Minehut where specifically these clients are receiving these issues at a much higher rate than other users.

[roccodev]

Well, connection resets and broken pipes are common when the user abruptly closes the connection (e.g. by closing their game or crashing).

Now, there might be a feature in the client that lets you join a server while being connected to a different one (e.g. join a friend, or view the multi-player list in the pause menu), so there is a chance that it's not closing the connection properly on the old server, which would in this case cause a read timeout.

If it's not that, I guess they're making changes to the network handler TCP options or Netty configuration, or they could be delaying the keep alive packets. (ping spoofing?)

Alas, this could have been easily investigated by looking at the source/decompiled version, but apparently people run obfuscated mods which have no reason to be.

[CoreyShupe]

Well, connection resets and broken pipes are common when the user abruptly closes the connection (e.g. by closing their game or crashing).

Now, there might be a feature in the client that lets you join a server while being connected to a different one (e.g. join a friend, or view the multi-player list in the pause menu), so there is a chance that it's not closing the connection properly on the old server, which would in this case cause a read timeout.

If it's not that, I guess they're making changes to the network handler TCP options or Netty configuration, or they could be delaying the keep alive packets. (ping spoofing?)

Alas, this could have been easily investigated by looking at the source/decompiled version, but apparently people run obfuscated mods which have no reason to be.

Definitely possible they're doing something funky with netty. Generally we're letting the backend cleanup the connections and not necessarily doing anything funky with it. Unsure what the issue could be.