Installation
How a platform operator installs Geo-MLOps on a single server, secures it and opens the first tenant
This chapter is for the platform operator. It walks you, in order, through putting Geo-MLOps on a single server, making the security settings you must change, and signing in as the global administrator to open the first tenant.
Geo-MLOps consists of one app started with Docker Compose and a k3s cluster where the training, build and serving pods run. The API, MLflow and the container registry are all inside the app, so there are few separate servers to install.
Order of this chapter
| Step | Page | What you do |
|---|---|---|
| 1 | Requirements | Check hardware, software and ports |
| 2 | Components | See in a diagram what runs where |
| 3 | Install with Docker Compose | Create the configuration file and start the stack |
| 4 | Prepare k3s and GPUs | Prepare the cluster and the NVIDIA device plugin |
| 5 | Required security settings | Replace the default signing keys and passwords |
| 6 | First sign-in and initial setup | Prepare tenants, members, GPUs and shared images |
| 7 | Upgrade and migration | Move to a new version |
| 8 | Backup and restore | Schedule a nightly backup and learn how to restore |
| 9 | Reverse proxy caveats | When the server sits behind a CDN or reverse proxy |
| 10 | Air-gapped installation | When you install on a site with no internet |
After installation, the Administration chapter covers tenants, members and tokens in detail, and the Reference lists every setting.