Tenants
Create tenants, edit names, KPI targets and feature toggles, deactivate, delete permanently and follow deletion progress
Required permission: Global admin
A tenant is an isolated workspace used by one team (or one customer). Datasets, experiments, models and deployments are all separated per tenant, and each tenant gets one Kubernetes namespace and one MLflow workspace. This is the only screen where you can create, change, suspend and delete tenants.
Each row shows the tenant code and name, a status badge, and links to the next jobs that make the tenant usable (Users (사용자), Shared images (공유 이미지), GPU wiring (GPU 배선)). If any kind or framework has an empty default image slot, the row gets an "N defaults missing (기본 누락 N개)" badge; pressing it takes you to Shared images. Training submissions and builds for that slot are rejected.
Create a tenant
- Press Create tenant (테넌트 생성) at the top right.
- Enter the Code (코드) and Name (이름). The code may contain only letters, digits and hyphens between them, is 2–16 characters long, and is stored in upper case. As you type, the names derived from the code (namespace
tenant-acme, MLflow workspaceacme) appear right below. - Under Features (기능), turn off the modules this tenant will not use (all are on by default).
- ① Under Default shared images (기본 공유 이미지), choose the defaults for the training runtime (one per framework), the dataset stager and the serving builder. Usually you leave the preselected values as they are.
Create tenant — code, name, features, ① default shared images - Press Create (생성). Press Assign first admin (첫 관리자 지정) in the completion notice to assign the tenant's first ADMIN on the Users screen.
Change the name, KPI targets and features
- Press Edit (편집) on the row.
- Change the name, the KPI targets (KPI 목표) and the Features (기능) switches, then press Save (저장).
Edit tenant — name, KPI targets, feature toggles
- KPI targets: the dashboard's KPI cards compare against these values and flag shortfalls. If left empty, the card shows "Target not set (목표 미설정)". Enter
%items as a value from 0 to 100. - Features: turning one off removes that module from the menus of the tenant's members, and the server also blocks the related API. Global administrators pass the feature toggles on the server.
Deactivate — suspend for a while
Deactivation only cuts off access and leaves the tenant's data as it is. Members of an inactive tenant cannot enter its workspace. You can undo it at any time with Activate (활성화).
- Press Deactivate (비활성화) on the row.
- Press Deactivate (비활성화) once more in the confirmation dialog.
Confirm tenant deactivation — you can reactivate it at any time
Delete permanently
Deletion removes the tenant and every resource in it. It cannot be undone, so if you only want to suspend the tenant, use deactivation.
- Press Delete (삭제) on the row. The dialog asks the server and shows the resources to be deleted (삭제될 리소스): serving endpoints, running PODs, datasets and their size, training runs, MLflow experiments/runs, registered models, the namespace and the MLflow workspace. Items at 0 are still shown so you can confirm there is "none".
- The Delete permanently (영구 삭제) button unlocks only after you type the tenant code exactly in the confirmation field.
Delete tenant — lists what will be removed and requires typing the code (not executed here) - Pressing Delete permanently (영구 삭제) takes you to the deletion progress screen. Cleanup can take a few minutes, so you can close the window.
The deletion progress screen runs the steps below in order and shows the result and log of each step.
- Stop serving endpoints
- Stop training processes and delete working directories
- Stop pipeline Jobs
- Delete the Kubernetes namespace
- Delete the MLflow workspace
- Delete object storage
- Delete database records
Deletions in progress or recently finished stay in the In progress · recent deletions (진행 중·최근 삭제) panel above the tenant list, and you can reopen them with View progress (진행 보기) / View result (결과 보기). On the row of a tenant being deleted, the edit, deactivate and delete buttons are locked.