Server crashes on restart from inside the caves.

Here's the info from claude:
 

  When you file the report, it would help to include:

  - Description: "Rollback from within the caves causes a reconnection loop lasting 25–30 minutes. The client connects directly to the caves shard port (10999) after rollback

  instead of routing through the master shard (10998), causing repeated ID_DST_USER_CONNECTION_FAILED errors."

  - Steps to reproduce: Have all players in the caves → initiate a rollback → observe connection loop

  - Workaround: Going through a portal to the surface before rolling back avoids the issue

  - Logs: Attach your client_log.txt, master_server_log.txt, and caves_server_log.txt from ~/Documents/Klei/DoNotStarveTogether/

  - Platform: macOS (Darwin 25.2.0), Apple M1 Max

  - Game version: 713564

What's Actually Happening

  This isn't a traditional crash — it's a reconnection loop that can last 25–30 minutes. Here's the exact sequence found in your logs:

  The Trigger

  When you do a rollback from within the caves, the rollback command is sent (c_rollback(1)) and both the master (surface) server and the caves server restart cleanly to load the

   previous save. That part works fine.

  The Race Condition

  After the rollback restart, your client tries to reconnect directly to the caves server (port 10999) instead of going through the master server (port 10998). This is because

  your last known active shard was the caves.

  [01:05:13]: [Connect] DirectConnect to 127.0.0.1|10999   ← caves port

  [01:05:13]: PushNetworkDisconnectEvent With Reason: "ID_DST_USER_CONNECTION_FAILED"

  The caves server just restarted and isn't ready to accept direct connections yet. It rejects you.

  Compare this to a rollback initiated from the surface, where the client correctly connects to the master server instead:

  [01:25:12]: [Connect] DirectConnect to 127.0.0.1|10998   ← master port ✓

  The Loop

  After the failed connection, the client sends a DoRestart: false IPC signal, which tells both servers to restart again for another reconnection attempt. This creates a cycle:

  1. Servers restart → caves not ready → ID_DST_USER_CONNECTION_FAILED

  2. Client signals DoRestart: false → servers restart again

  3. Repeat for ~25-30 minutes

  The logs show dozens of Reset() calls in rapid succession after each cave rollback (visible between [01:05:25] and [01:32:04] in your client log for the first rollback

  attempt).

  Secondary Symptoms

  The caves server log also shows Invalid RPC sender list errors at shutdown — these appear to come from the Insight mod (workshop-2189004162) trying to communicate between

  shards during the abrupt disconnect when the master goes down.