rust-server-flux

The Rust dedicated server image for Flux: runonflux/rust-server-flux. One image for Vanilla and Modded (Carbon). Written from scratch on steamcmd/steamcmd:ubuntu-24, in the same family as runonflux/garrysmod-server-flux and runonflux/7dtd-server-flux.

Why the game is not in the image

The images it replaces (pfeiffermax/rust-game-server, and runonflux/rust-carbon-server built on it) carry the game inside. A Facepunch update then means a new image, and FluxOS rolls a new :latest out to every server as a soft redeploy, hours later and whether or not that server needed it. Until it arrives, players are refused with "Wrong Connection Protocol: Server update required!", and restarting the server does nothing.

Here SteamCMD installs the game (app 258550) into the install volume on the first start and updates it on every start, and flux-updatewatch.sh restarts the server, with a warning in game, as soon as Steam lists a new build. This image changes only when its own scripts do.

What it does

• Every start: SteamCMD update, then Carbon for a modded server, then RustDedicated as a child of the supervisor (PID 1). A crash is restarted in place; five within an hour end the container with 42 so the platform can move it.
• Update watch (FLUX_AUTO_UPDATE, on by default): Steam is asked every 10 minutes. A new build means a 5 minute warning in chat, then a save and a restart onto it. A Carbon server restarts for a new Carbon release too, and for Carbon itself if it could not be fetched at the last start. An update that did not take is not retried for an hour.
• Daily restart (FLUX_RESTART_TIME, off by default): the same warning, save and restart.
• Carbon (FLUX_FRAMEWORK=carbon): the production_build release, fetched from GitHub on every start when it moved, cached on the install volume (the last one is used when GitHub is unreachable). The runtime (managed, tools, native) is refreshed; plugins, configs, data, lang and extensions are the owner's and never touched. The game runs as root, so a plugin FluxOS writes as root loads at once (the old image needed an inotify chown for that).
• A save cut off mid-write (a .sav.new left on disk) is repaired before every start: the whole copy becomes the .sav, the other is kept as broken-<name>.<epoch>. A truncated .sav otherwise fails to load and can leave the server hung at boot with the CPU pegged.
• docker stop: server.save over RCON, then SIGTERM. RustDedicated does not save on SIGTERM (it exits in about two seconds and the world is the last autosave), which is what the images this one replaces did on every stop.
• flux-rcon "<command>" and flux-restart "<reason>" [warn minutes] inside the container.

App spec

containerData: g:/srv/rust/server/<identity>|ml:game:/srv/rust|m:carbon:/srv/rust/carbon

• g: the world (server/<identity>), synced to the second instance. Same path as every older Rust spec, so a server keeps its world when it moves onto this image.
• ml:game the install. Local: never synced (FluxOS 8.19+ keeps ml: out of Syncthing) and never in a backup. A node that takes the server over installs the game itself.
• m:carbon the Carbon tree with the owner's plugins. On a Vanilla server it is simply unused.

Docker mounts the parent first, so the two folders inside /srv/rust show through the install.

The arguments are the old images' (+rcon.port 28016 +rcon.password ${RCON_PASSWORD} +server.identity ...), with ${VARS} expanded as they were.

Variable	Default
FLUX_FRAMEWORK	(vanilla)	carbon for a modded server
FLUX_AUTO_UPDATE	on	false to wait for a restart instead
FLUX_UPDATE_CHECK_MINUTES	10	2..1440
FLUX_UPDATE_WARN_MINUTES	5	0..30
FLUX_RESTART_TIME / TZ	(off) / UTC	HH:MM daily restart
FLUX_RESTART_WARN_MINUTES	5	0..30
STEAM_BRANCH	public	a Steam beta branch
SKIP_UPDATE	off	do not run SteamCMD when the game is installed

Measured (2026-10-23, build 26139947)

• First install: 6.1 GB, 69 s on a fast line. An update with nothing to do: 11 s.
• Carbon 2.0.262: a 21 MB download, 2 s.
• Moving a server off runonflux/rust-carbon-server (same world and Carbon volumes, an empty install volume): game installed in 73 s, up in 110 s, the 381 saved entities loaded, the uMod plugin already in the folder loaded. Files the old image left owned by 10001 are no obstacle.
• docker stop: world saved over RCON, exit 0, 4.4 s in all.
• A new build seen by the update watch: warned, saved, restarted, world loaded from the save.
• SteamCMD's depotcache must survive a redeploy. An update from build A needs A's manifests, and Steam refuses an old manifest to an anonymous login ("Failed to get manifest request code, 'Access Denied'", state 0x6), even with validate. The image keeps depotcache on the install volume; when it is gone anyway, the update is redone as a fresh install over the top: the whole 6 GB again and about 10 GB on the disk at the peak, which still fits the smallest plan (15 GB).

Build and test

./tests/test-flux.sh                     # pure logic
docker build -t rust-server-flux:local .
./tests/test-image.sh                    # the supervisor around a stand-in Steam and game
docker run --rm -v "$PWD:/mnt" -w /mnt koalaman/shellcheck:stable -x scripts/*.sh tests/*.sh tests/image/*.sh

Versions

• 1.0.0 (2026-10-23): first release.
• 1.0.1 (2026-10-23): repairs a save that was cut off while being written, before each start.
• 1.0.2 (2026-10-23): Carbon servers hung at boot forever: Carbon's self-updater forks curl, and on Flux nodes that fork never reaches exec. It is now turned off in carbon/config.json before every start (the image keeps Carbon updated itself), and game processes left behind by a server that ended are killed before the next one starts.