AMP 'Proteus' 2.8.0 Release Notes

AMP 2.8.0.10 ‘Proteus’

:warning: It’s very important that you follow the update guide for this release:
https://discourse.cubecoders.com/t/how-to-update-amp-to-the-latest-version/2297

There have been many vulnerabilities the past week listed online. Some impact AMP due to being built in .NET and others impacted the AMP container images. These have all been addressed in this update, so please ensure you follow the update guide above to get all the needed bits.

With the infinitely rising prices of RAM, this release focuses on memory optimizations and bug fixes. The key optimizations were focused around .NET 10, packaging uncompressed binaries (Kernels can share the memory more readily), and a major sleep mode overhaul.

Sleep mode is temporarily disabled for all generic templates to avoid unexpected sleeping or false wakes while we make the needed adjustments. We will be testing templates over the next few weeks to support the improved sleep functionality so servers can be turned off automatically when no players are on.

.NET 10

  • All of AMP has been upgraded to now run on .NET 10 which includes many optimizations from .NET 9/10
  • AMP binaries are no longer compressed - They take more disk space, but use less memory while running since AMP no longer unpacks itself into RAM on startup. This amounts to over 100+ MB of freed RAM per instance and even more that can be freed when the system is low on memory

Graceful Shutdowns

  • Stopping a server now waits for it to actually finish shutting down (up to 10 minutes) instead of killing it after a short per-template timeout - Fixes servers such as ARK: Survival Ascended losing progress when stopped for a backup, because the save was cut off part way
  • If AMP knows the stop command never reached the server (e.g. RCON not connected), it now terminates it straight away and logs why, instead of waiting
  • A second stop request while a server is already stopping is ignored
  • App.ExitTimeout (templates) setting is retired
  • Minecraft, Rust and Generic Module servers all use the same stop logic now
  • Container stop timeouts (Docker and Podman) and the host’s own backstop timeouts were raised above the new limit, so they only catch a stuck AMP process rather than cutting off a save
  • The systemd unit’s stop timeout was raised to match
  • On Windows, the AMP service now waits for instances to finish stopping before reporting itself stopped
  • Only ADS can now be installed as a Windows service - ampinstmgr refuses to set up a service for an individual game instance

Minecraft Changes

  • Fixed Sponge installs failing on modern versions - Sponge downloads now use Sponge’s own downloads API instead of their maven repository at their request
  • Version lists are now only fetched for the server type you have selected, at most once per hour - A refresh button has been added to fetch them again on demand. This extremely reduces the impact version checking has on the public APIs
  • Fixed join/leave tracking creating ghost or duplicate users for players who renamed, or when the server hides IPs or appends a UUID to the name
  • Fixed chat messages being logged twice when System chat was enabled
  • Fixed Forge and NeoForge servers refusing to start if the version list couldn’t be fetched when AMP started - AMP now retries the fetch when starting
  • Fixed [Server] being stripped from server chat messages in the console

Backups

  • The “Backup While Running” toggle now always shows on the Backups tab
  • New Minecraft instances now default to backing up while running, since Minecraft supports quiesce

Generic Module Changes

  • Universal sleep now only wakes the server when a player is actually trying to join - Previously any packet (server list pings, port scanners) could wake it
  • Sleeping servers now answer Steam server queries with the server’s real last response, so server browsers show correct information while asleep
  • Fixed sleep wake triggering more than once for the same connection
  • Steam server queries now work over IPv6
  • Added UpdateSourceItems for SteamCMD so templates can download individual depots with download_depot - Halo 3: Project Reclaimer will be the first template to use this functionality to avoid downloading all of the Master Chief Collection
  • Added a Sleep option to the Startup Mode setting - The server starts directly into sleep mode, answering server browser queries from its last known reply until a player joins. Falls back to a normal start if the server has never run before
  • Fixed sleep mode crashing the instance when a server went back to sleep
  • Fixed sleep mode sometimes failing to start because the game’s ports were still being released - AMP now retries briefly instead of stopping the server
  • Fixed sleeping servers answering with stale information cached from an earlier run (old name, map etc.) - The reply is now refreshed from the live server
  • Fixed killing a sleeping server leaving its ports in use, which stopped it from starting again
  • Fixed sleep mode on servers that use the same port for game traffic and Steam queries (e.g. Space Engineers, Starbound) - Joining players previously couldn’t wake the server

SteamCMD

  • AMP now logs a message when an update runs without a stored Steam login
  • AMP now logs a message when Steam’s API is unreachable during the update check, instead of failing silently
  • Fixed the scheduled update check always reporting “no update” when the server’s folders contain symlinks or folders AMP can’t read (e.g. Dune: Awakening)
  • Fixed servers with a depot that has no public manifest being reported as out of date on every check, which made auto-update restart them in a loop

Containers

  • {{InstanceId}} and {{InstanceName}} are now expanded in mount bind host paths, including when editing existing binds
  • Mount binds can now point to a single file, not just a folder
  • Containers now use the host’s timezone instead of always running in UTC
  • Unused AMP container images left behind after an image update are now cleaned up automatically - Only AMP-labelled images are removed, and this can free several GB on hosts that have been running for a while
  • /dev/ntsync is now passed into containers when the host has it, improving Wine/Proton performance
  • Fixed duplicate mount binds (two binds that expand to the same path) stopping the instance from starting - They are now rejected with a clear message when saving, and existing duplicates are skipped with a warning
  • Container instances report the host’s ampinstmgr version. ampinstmgr removed from the images to save space

Instance Management

  • Fixed deleting an instance from a controller showing a “Task 3” error even though the instance was removed
  • Fixed the last instance coming back after an ADS restart - The instance list is now saved when the final instance is deleted
  • Fixed ADS crashing when the hourly template repository refresh hit an error

Windows Instance Manager

  • Replaced the old Metro look with the native Windows 11 Fluent theme
  • Fixed the Instance Manager hanging when closed
  • Fixed “Manage Services” not reopening the Instance Manager correctly
  • Upgrading with the MSI now keeps a custom install folder rather than resetting it to the default
  • Upgrading with the MSI now skips the setup type screen, so installed features stay as they were
  • Fixed a controller treating a target’s queued task as finished - Bulk upgrades and other remote actions could fail with an unknown error or report success too early
  • Fixed starting a remote instance that needs to update first returning an error
  • Remote actions that time out now say the target didn’t respond in time and that the action may still complete, rather than a generic error
  • Fixed a deleted instance coming back after an ADS restart if removing its users failed
  • Fixed ADS template repository refresh hanging forever when a repository needs a login - Git no longer waits for a credentials prompt
  • Fixed a git error when adding a new template repository
  • Fixed disabling a schedule trigger that was already disabled turning it back on

Misc

  • The SFTP idle timeout now has a 30 second minimum - Lower values were causing clients to be disconnected while still in use
  • Removed the special notice page from ADS
  • LDAP group membership is now re-synced on every login, so group changes apply without recreating the user
  • Settings now report their min/max limits to the UI and API, so the limits can be shown and checked before saving
  • Fixed a crash when an RCON client connected and disconnected immediately
  • Fixed an upgrade or restart losing its exit code if the server received a shutdown signal (SIGTERM) part way through

getamp Changes 4.2.0

  • The package manager is now picked from the OS family in os-release - Fixes installs on Arch-based distros (e.g. CachyOS) that have an unrelated apt package installed
  • Exit code 100 is now 93, so it no longer clashes with package manager exit codes
  • Unattended installs now use a random system password when one isn’t supplied
  • ntsync is now enabled on all new installs, and is also enabled by getamp update
3 Likes