Skip to content

Install problems

First, re-run the same command. Every step checks before it acts, so a re-run continues where the last one stopped. Most interrupted installs (network drops, a closed SSH session) need nothing more.

The installer says Cause and fix
run as a user with sudo, not as root Sign in as your normal user (with sudo) and run it again
git is missing / curl is missing sudo apt install git curl
the docker compose plugin is missing DGX OS ships it; on other images sudo apt install docker-compose-plugin
/var/lib has N GiB free; models and images need at least 150 GiB Free disk space, or move Docker’s data to a larger disk
--garrison needs read access to … The access token from your install command is not exported in this shell, or it has expired. Ask Garrison sales or support for a new one
no terminal to accept the licences on The installer cannot prompt (automation). Add --accept-licences: you accept the NVIDIA and model licences it lists
no default route to take the management address from Pass --address <the appliance's LAN IP>
containers not managed here are running: … A warning, not an error. Garrison leaves them alone; make sure they leave memory and ports free

The installer ends with 13 checks. The usual failures:

Check Cause and fix
backend reached the gateway The AI platform is still starting the model (the first start takes up to 15 minutes). Wait, then re-run
embeddings through the gateway The embedding model is still loading, or there was not enough memory for it. sudo oaf status shows it
endpoint logs ingest / prometheus scrapes the console The monitoring stack is restarting; re-run. If it persists, see Collecting diagnostics
Symptom Cause and fix
Browser warns about the certificate Expected until you trust the appliance CA: First sign-in
“network error” at sign-in The console was built with an absolute API address. Re-run the installer (it rebuilds the console)
Dashboards in Observe are blank The browser does not trust the appliance certificate for embedded content: trust the CA, or use Open in Grafana
Sign-in fails with the admin password Read it again from .env (sudo grep ADMIN_PASSWORD /opt/garrison/.env). It is only set on the first install; re-runs keep it