[ConnexionDev]

Summary

On ver/26.2 (build 26.2-11 / 068d8c5) a keepalive timeout freezes the whole server.

MinecraftServer calls keepConnectionAlive() on the global server thread for every player. disconnect() then sees that this thread does not own the player, queues the kick onto the player region, and waits with managedBlock() / waitForTasks. The region cannot tick until that server tick returns, so the wait never finishes. The watchdog fires after 10 seconds. While the player is still connected, the same path logs was kicked due to keepalive timeout again on the next tick.

Seen on 2026-10-11, 10-12, 10-16 and 10-18, each time a different player. The stuck stack is only server code:

keepConnectionAlive
-> ServerCommonPacketListenerImpl.disconnect
-> managedBlock
-> MinecraftServer.waitForTasks

This is not covered by #64 or #70. Both fixes are already in 068d8c5, and ver/26.3 still has the same wait.

Change

In ServerCommonPacketListenerImpl:

• From a tick thread that does not own the player, scheduleTask the kick onto that player's region and return. Do not call managedBlock or waitable.get().
• A flag stops the next ticks from logging another timeout and queueing another kick until the region has run the first one.
• Callers that are not a tick thread still wait, same as before.

Test plan

• Build ver/26.2 with this patch applied.
• With players online, let one connection hit the keepalive limit (or kick a player from the server thread). The server tick should continue and the player should leave on the next region tick.
• The console should log the timeout once, not once per tick.
• A kick from an async thread should still complete.

[PureGero]

Please send the thread dumps from these crashes