Skip to content

Handle responses that span multiple packets - #104

Merged
itzg merged 2 commits into
itzg:mainfrom
eliasehmann:fix/multipacket-responses
Nov 15, 2026
Merged

itzg merged 2 commits into
itzg:mainfrom
eliasehmann:fix/multipacket-responses

Conversation

@eliasehmann

Copy link
Copy Markdown
Contributor

Fixes #9, fixes #42. Probably also what is behind the Minecraft reports in #28.

Problem

A response longer than 4096 bytes is split by the server over several packets, but only the first one was read. Long output such as help on a modded server got cut off, and in an interactive session the leftover packets were then shown as the responses to the following commands, together with "Weird. This response is for another request."

Change

As discussed in #9, the packet handling now lives in this repo (new rcon package) and the dependency on james4k/rcon is gone.

To find the end of a response, the client follows each command with an empty SERVERDATA_RESPONSE_VALUE packet and reads until the reply to that one arrives, as described at https://minecraft.wiki/w/RCON#Fragmentation. Minecraft answers it with "Unknown request 0" and Source servers echo it back, so nothing gets executed.

Some details that are worth a look:

  • The extra packet is only sent after the first packet of the response has arrived, never together with the command. The Minecraft server expects exactly one packet per socket read and closes the connection otherwise.
  • I could not test against servers other than Minecraft, so the wait for the reply to the extra packet is limited to one second without receiving anything. When that runs out, what was received so far is returned and the client reads a single packet per command for the rest of the connection, which is the previous behavior.
  • Packets whose request ID does not belong to the current command are skipped, so nothing can leak into the next response. With that, the "Weird. This response is for another request." message is gone.
  • If the connection is lost after the response has started (e.g. stop), the response received so far is still printed.
  • "Failed to send command" / "Failed to read command" became a single "Failed to execute command".

Testing

Unit tests run against an in-process fake server that mimics the Minecraft one (splits at 4096 bytes, requires one packet per read) and can also ignore the extra packet or close the connection after responding.

I also tried it against itzg/minecraft-server (vanilla) with 400 scoreboard objectives, where scoreboard objectives list is about 9.5 kB:

  • before: 4097 bytes in one-shot mode; in interactive mode the rest showed up as the response to the next commands
  • after: complete response in both modes, and the following commands show their own response

Short commands take the same time as before, and stop still prints "Stopping the server".

I did not test against non-Minecraft servers (Ark, Palworld, Source).

A response longer than 4096 bytes is split by the server over several
packets, but only the first one was read. The output of commands such as
"help" on a modded server got cut off, and in an interactive session the
remaining packets were then taken for the responses to the commands that
followed, along with "Weird. This response is for another request."

The RCON packet handling now lives in this repo, replacing the
james4k/rcon library. Since the protocol does not mark the last packet
of a response, the client follows up each command with an empty packet
of the response type and reads until the reply to that one arrives, as
described at https://minecraft.wiki/w/RCON#Fragmentation. That packet is
only sent once the response to the command has started to arrive, since
the Minecraft server closes the connection when it gets two packets in
a single read.

For servers that do not answer such a packet, the wait for its reply is
limited to a second without receiving anything. After that the client
goes back to reading a single packet per command for the rest of the
connection. Packets that do not belong to the current command are
skipped either way.

Fixes itzg#9, fixes itzg#42
@itzg itzg added the bug label Nov 14, 2026
@itzg

itzg commented Nov 14, 2026

Copy link
Copy Markdown
Owner

Removing dependency on james4k/rcon at first seemed like a big, scary thing...but wow, that repo is now 5 years without changes.

@itzg itzg left a comment

Copy link
Copy Markdown
Owner

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Awesome and long overdue modernization.

Comment thread rcon/rcon.go Resolved
Comment thread rcon/rcon.go Resolved
Comment thread rcon/rcon_test.go Resolved
Comment thread rcon/rcon_test.go Resolved
@itzg itzg mentioned this pull request Nov 14, 2026
@itzg itzg added release/minor On merge, creates a new release with a minor version bump. status/ready to merge labels Nov 15, 2026
@itzg
itzg merged commit 1dcfffa into itzg:main Nov 15, 2026
2 checks passed
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

bug enhancement release/minor On merge, creates a new release with a minor version bump. status/ready to merge

Projects

None yet

Development

Successfully merging this pull request may close these issues.

Response buffer of 4096 is too small for some commands handling multipacket responses

2 participants