Autopause enters repeated pause/knock loop due to RCON activity (GTNH)

[Knufle]

I’m running a GT New Horizons (GTNH) server using itzg/minecraft-server with ENABLE_AUTOPAUSE=TRUE.

When the last player disconnects, autopause correctly pauses the JVM, but then the server repeatedly wakes up due to RCON connections originating from localhost, immediately re-pauses, and logs tick time warnings. This continues indefinitely while no players are connected.

This appears to be caused by internal or external RCON activity being interpreted as a “knock”, even though no client is actually attempting to join.

---

Observed behavior

The server enters the following loop:

1. Autopause detects no players
2. JVM is paused
3. An RCON connection from ::1 occurs
4. World is saved via RCON
5. JVM resumes
6. Minecraft logs a tick time warning due to time jump
7. Autopause detects a “knock”
8. No player connects → server pauses again

This repeats every few minutes.

Log excerpt

[Autopause loop] All clients disconnected - pausing in 60 seconds
[Autopause loop] No client reconnected - pausing

[RCON Listener #1/INFO]: Rcon connection from: /0:0:0:0:0:0:0:1
[Server thread/INFO]: [Rcon: Saved the world]
[Autopause] Pausing Java process

[Server thread/WARN]: Can't keep up! Did the system time change, or is the server overloaded?
Running 5112ms behind, skipping 102 tick(s)

[Autopause loop] Server was knocked - waiting for clients or timeout
[Autopause loop] No client connected since startup / knocked - pausing

This happens repeatedly while the server is idle.

---

Expected behavior

When no players are connected:

• Autopause should pause the JVM
• The server should remain paused

---

Container details

Image

itzg/minecraft-server:java25
@sha256:5c10b9b6bfe7311e652afbd833626ec46dba66ee517ced9b7cdc39b0092d9789

General

TYPE=GTNH
VERSION=LATEST
EULA=TRUE

Java

JAVA_VERSION=jdk-25.0.1+8
JAVA_HOME=/opt/java/openjdk

Memory

INIT_MEMORY=4G
MAX_MEMORY=6G

Autopause configuration

ENABLE_AUTOPAUSE=TRUE
AUTOPAUSE_TIMEOUT_EST=60
AUTOPAUSE_TIMEOUT_INIT=60
MAX_TICK_TIME=-1

Networking

25565/tcp exposed (IPv4 + IPv6)
25565/udp exposed (IPv4 + IPv6)

User / permissions

UID=1000
GID=1000

---

Notes

• No players are connected when this occurs
• The RCON connections are always from localhost (::1)

---

Questions

1. Is there a supported way to prevent RCON activity from waking a paused server? And what would be the impact of that?
2. Is this behavior expected with heavy modpacks like GTNH, or is this considered a bug?

[itzg]

Your logs aren't indicating that rcon activity is doing the knocking on the port. Rcon is used immediately before pausing to flush the world data to disk.

You'll need to use a network monitor like Wireshark to see what is contacting the Minecraft port.

[itzg]

...with that said, I can't fully convince myself that rcon should be allowed to wake it. The original reason for that might have been to support the backup container; however, that now waits to resume backups until the server is unpaused.

I think about changing that.

[Knufle]

Thanks for the clarification. I went ahead and verified this with packet capture on the Docker host.

I ran tcpdump on the host interface listening for tcp port 25565 while the server was paused. The wakeups are not caused by RCON or backup activity. Instead, they are triggered by external TCP connections performing what appears to be a standard Minecraft handshake/status query:

• Incoming SYN from arbitrary internet hosts
• Small payloads (~20–30 bytes) consistent with a status request
• Server responds with a large payload (status JSON)
• Connection closes without any login phase

These connections come from random cloud providers/servers on the public internet, not from local containers, the backup container, or RCON. Capturing on the Docker bridge showed no internal traffic when the wake occurred.

This confirms that autopause currently wakes the server on any inbound TCP activity, including unauthenticated status/handshake probes, even when no player actually joins.

So far, this lines up with your observation that RCON isn’t the cause. From a behavior perspective, this makes autopause ineffective for publicly exposed servers due to constant background scanning and server list pings.

If helpful, I can share sanitized tcpdump excerpts, but the main takeaway is that the wake trigger is external network probing rather than in-container activity.

Are you aware of any way to circumvent this?

[itzg]

If helpful, I can share sanitized tcpdump excerpts, but the main takeaway is that the wake trigger is external network probing rather than in-container activity.

No need. This is an extremely common occurrence thanks to all the annoying bots and port scanners out there in the wild.

Yes, the way to mitigate it is by putting https://github.com/itzg/mc-router in front of the server since it'll only only connections through that are directed at your specific server hostname. It can also filter by client address, etc.

[Knufle]

Alright! Thanks a lot, I've put itzg/mc-router in front of the server, and now everything is working fine :)

Just one final question, with AUTOPAUSE, I can see that CPU usage is down to 0% when paused, but is there a way to also free up the memory? Or is that not possible?

[itzg]

You can use auto stop to completely free up the process. Since you're now using mc-router, the auto scaler option there is going to work out much better.