Fabric server does not shutdown on stop command

[DearFox]

Describe the bug

When you stop the Fabric server, the server does not shut down completely, resulting in the need to use C^ to shut down the server. This breaks automatic server restarts.

To Reproduce

1. Start the fabric-server-mc.1.19.3-loader.0.14.13-launcher.0.11.1 with fabric-api-0.73.0+1.19.3 and the newly built Geyser-Fabric from the master branch
2. Log in to the server via Bedrock (I'm using the Android version)
3. Everything works? Log out of the server.
4. Write to the server console /stop
5. The server stops but the console is still open and the process is not completely stopped.

Expected behaviour

The server is completely stopped and the console is closed (as it usually happens when the server is stopped)

Screenshots / Videos

No response

Server Version and Plugins

No response

Geyser Dump

https://dump.geysermc.org/P0ex2z5ML3tyXoWyJRSX5MQ2meMR8qsq

Geyser Version

2.1.0-SNAPSHOT (git-master-af5d03f)

Minecraft: Bedrock Edition Device/Version

No response

Additional Context

I downloaded the master branch and built it.
Then I added the missing mappings and languages folders to Geyser-Fabric.jar
Everything works great except /stop command (only if someone connected to the server via bedrock)

[Ej747]

I get this too. Does yours look similar to this?

[DearFox]

I get this too. Does yours look similar to this?
Yes, something like this

[06:00:00] [Server thread/INFO]: Stopping the server
[06:00:00] [Server thread/INFO]: Выключение Geyser.
[06:00:00] [Server thread/INFO]: Geyser успешно выключен.
[06:00:00] [Server thread/INFO]: Stopping server
[06:00:00] [Server thread/INFO]: Saving players
[06:00:00] [Server thread/INFO]: Saving worlds
[06:00:01] [Server thread/INFO]: Saving chunks for level 'ServerLevel[world]'/minecraft:overworld
[06:00:01] [Server thread/INFO]: Saving chunks for level 'ServerLevel[world]'/minecraft:the_nether
[06:00:01] [Server thread/INFO]: Saving chunks for level 'ServerLevel[world]'/minecraft:the_end
[06:00:01] [Server thread/INFO]: ThreadedAnvilChunkStorage (world): All chunks are saved
[06:00:01] [Server thread/INFO]: ThreadedAnvilChunkStorage (DIM-1): All chunks are saved
[06:00:01] [Server thread/INFO]: ThreadedAnvilChunkStorage (DIM1): All chunks are saved
[06:00:01] [Server thread/INFO]: ThreadedAnvilChunkStorage: All dimensions are saved
[06:00:01] [Server thread/INFO]: Shutting down EasyAuth.
[06:00:01] [Server thread/INFO]: Database connection closed successfully

[DearFox]

I also found out about this in the Fabric discord
Perhaps this will be of some help in solving this problem.

[onebeastchris]

I cannot reproduce this shutdown issue with just Geyser-Fabric. However, this issue does occur with floodgate-fabric installed, caused most likely by bstats there. Not Geyser-Fabric though; so no idea how Geyser-fabric alone does this in your case

[DearFox]

I cannot reproduce this shutdown issue with just Geyser-Fabric. However, this issue does occur with floodgate-fabric installed, caused most likely by bstats there. Not Geyser-Fabric though; so no idea how Geyser-fabric alone does this in your case

In your tests, at least 1 player entered the server with minecraft bedrock (android)?
and I don't use floodgate-fabric on my server.
and i just checked, same thing happens on quilt server 1.19.3

[onebeastchris]

Ah, I didn't actually test with players online, just the shutdown behavior itself. I'll see if I can reproduce it with a player join; I'll post here later

[onebeastchris]

Thanks for that info, I'm now able to replicate this:
https://paste.gg/p/anonymous/84e53bf6b98e429a93e982727079163e here a list of all the running threads after running /stop, looks like a Geyser player thread isn't closed properly.

[DearFox]

2.1.0-SNAPSHOT (git-master-9b3b2fb) the problem is still there

[petersv5]

Still there in build 238.

[Konicai]

Likely caused by this: GeyserMC/MCProtocolLib#699

[petersv5]

jstack of Minecraft with Fabric, Geyser and Floodgate hung after a stop:
jstack.txt

[petersv5]

I tried adding a shutdown of the EventLoopPool for the geyser player thread event pool. That got rid of some of the lingering threads. However, there are still some lingering threads:
jstack2.txt

[petersv5]

Likely caused by this: GeyserMC/MCProtocolLib#699

I suspect it has to do with TcpClientSocket in PacketLib creating an EventLoopGroup (static class variable) that is never shut down. See GeyserMC/PacketLib#49.

[petersv5]

I was looking at the wrong PacketLib implementation. I see that in the version used from MCProtocolLib there is a Shutdown hook that is supposed to close the EventLoopGroup.

I will check if this is called properly.

[petersv5]

Shutting down the EventLoopGroup in TcpClientSession removed one of the blocking threads.
The only blocking thread seems to have been created during a call to requestSkinAndCape() which in turn calls CompletableFuture.supplyAsync(). This thread seems not to be stopped

[petersv5]

By shutting down the EXECUTOR_SERVICE in SkinProvider as well I got a clean shutdown even after a Bedrock player had been connected.

Should I file this collection of hack-ish changes as a PR? It is probably needs a lot of rework by someone that understands how the components are supposed to interact inside Geyser.

[Camotoy]

Absolutely feel free to make a PR and we'll work from there if changes are needed.

[petersv5]

I created #4029 (#4029) for Geyser and GeyserMC/MCProtocolLib#744 for MCProtocolLib.

I would again like to stress that these are PoC hacks to verify that it was indeed the thread pools that were keeping the server from stopping after a Bedrock player had been on the server.

[petersv5]

I updated the PRs above with updates as mentioned. Most important is that /geyser reload works now.