Uninstalling

This page covers a Quasar-owned install. A legacy Compose install is stopped and removed with Compose.

Removing the seed does not remove Quasar

Deleting the seed (its Dockge or Arcane stack, or docker rm) removes only the seed. Quasar keeps running, and the console shows the seed as not found until you start it again. Managing the stack never removes data.

Stop, keeping everything

Agent first, then the control plane and Postgres where they run:

docker stop quasar-node-agent
docker stop quasar-control-plane quasar-postgres

docker start brings them back (Postgres first). Nothing is removed.

Remove a GPU host from the console

1. In Admin ▸ Fleet ▸ Hosts, open the host and press Remove host. It drains the host, waits for sessions to end, then removes the node agent and recovery actor. Homes, volumes and the seed stay.
2. Forget the host once the console says it was removed.
3. Remove its seed from the machine when ready. It stays idle until then.

• If the agent isn’t connected (no heartbeat for about a minute), the console can’t remove it. Run uninstall on the machine.
• If removal stops part-way, press Retry removal, or run uninstall on the machine.
• To bring it back, run a new Add host command with the same node name. It keeps its history.
• Its users’ homes stay claimed by it. If it isn’t coming back, forget it and release each claim; those users start with empty homes. If it is coming back with the same storage root, leave the claims alone.

Uninstall a machine

Works on any machine, including the first, with the console down. Use the seed’s image:

docker run --rm -it --security-opt label=disable \
  -v /var/run/docker.sock:/var/run/docker.sock \
  -v quasar-machine:/var/lib/quasar-machine \
  "$(docker inspect -f '{{.Config.Image}}' quasar-seed)" uninstall

It removes the containers this installation created and keeps your data: the database volume, quasar-machine, the agent’s identity and the homes. It refuses during an update. If interrupted, run it again; it prints the next command each time.

`--purge` deletes the data

--purge also removes this installation’s volumes, network and machine state. It refuses while a seed exists, so remove the seed first. On Podman the seed is a systemd unit that restarts a removed container: run systemctl stop quasar-seed (--user when rootless), and delete its quasar-seed.container file (/etc/containers/systemd/, or ~/.config/containers/systemd/ when rootless) if you are not installing again, since the unit starts at every boot:

img=$(docker inspect -f '{{.Config.Image}}' quasar-seed)
docker rm -f quasar-seed                  # or remove the stack in Dockge or Arcane
docker run --rm -it --security-opt label=disable \
  -v /var/run/docker.sock:/var/run/docker.sock \
  -v quasar-machine:/var/lib/quasar-machine "$img" uninstall --purge
docker volume rm quasar-machine

• It asks for the node name, or takes --confirm <node name>.
• With Quasar’s own database, it first dumps to the quasar-final-dump volume (never purged) or --dump-to <absolute host directory>. If the dump fails, nothing is deleted.
• Your own database is never touched.

Full removal

1. Take a final backup if you might want the data. See Backup and restore.

2. Remove each GPU host from the console, then uninstall the first machine (with --purge if you want the data gone). Remove any remaining seed.

3. Remove the final dump when you no longer want it: docker volume rm quasar-final-dump.

4. Remove the home directories. These are everyone’s saves. Check each host’s root under Admin ▸ Fleet ▸ Hosts and verify the exact path:

   sudo rm -rf -- <verified-managed-home-root>

5. Remove the images: quasar-recovery, quasar-control-plane, quasar-node-agent, and Postgres on the first machine.

   docker image ls --filter reference='ghcr.io/accreleus/quasar/*'
   docker image rm <verified-image-id-or-digest>

   App images Quasar pulled or built are also there. Be careful with a blanket prune.

6. Remove the .env holding your own database’s settings, if you used one.

Host changes to undo

None of the above removes these. Check each file first: something else may rely on it.

Change	Undo
Socket buffer sysctl, if you set it	sudo rm /etc/sysctl.d/99-quasar.conf
User-namespace sysctls (written by the enroll command’s fixes)	Read /etc/sysctl.d/99-quasar-userns.conf, remove it, then reboot or set each value back to your distribution’s default (on Ubuntu 24.04+: sudo sysctl -w kernel.apparmor_restrict_unprivileged_userns=1).
uinput module load (older installs)	sudo rm /etc/modules-load.d/uinput.conf
Unraid	Remove the sysctl -w net.core.wmem_default=2097152 and modprobe uinput lines from /boot/config/go.
App-container AppArmor profile	sudo rm /etc/apparmor.d/quasar-app if present, then reboot to unload it.
TLS certificate	Remove it from any trust stores you added it to.