Glossary

Workspace - A persistent Podman container assigned to one user, with CPU, memory, and GPU quotas plus an advisory disk alert. Stays running until the user or admin stops it.

Launcher - A service started inside a running workspace via podman exec (JupyterLab, code-server, LabPod Terminal, TensorBoard, MLflow). Starting or stopping a launcher does not restart the workspace.

Template - A container image plus launcher port configuration and default resource values. Can be global (admin-owned, visible to all) or user-owned (visible only to its creator).

Shared mount - An admin-configured host path mounted into all workspaces (usually read-only). Used to share large datasets without duplication.

User-declared port - A named port slot a user reserves inside a workspace for custom processes (Gradio, Streamlit, etc.). Exposed at /ws/<workspace-id>/_user/<name>/ through LabPod’s reverse proxy. This namespace is separate from launcher routes.

Discovered container - A container the user started directly on the host (outside LabPod) that LabPod may include in read-only safety and capacity views. It is not a workspace and has no LabPod proxy, launcher, terminal, logs, declared-port, or lifecycle support.

Rootless Podman - A daemonless container runtime where each container runs as the owning Linux user, not as root. LabPod uses rootless Podman for all workspace containers.

keep-id - The --userns=keep-id Podman flag. Makes the container UID match the host UID, so files created inside a workspace have natural ownership on the host.

CDI (Container Device Interface) - A standard interface for passing hardware devices into containers. LabPod uses CDI to allocate whole NVIDIA GPUs to workspaces: --device nvidia.com/gpu=<index>.

Fractional GPU - A software partitioning mode that lets multiple workspaces share one physical GPU simultaneously. Each workspace declares its memory and compute fraction. Requires a GPU sharing runtime library configured via LABPOD_GPU_SHARING_LIB_PATH.

MIG (Multi-Instance GPU) - NVIDIA hardware partitioning available on A100 and H100 GPUs. Splits one physical GPU into independent slices with separate memory and compute. Pre-configured by the operator using nvidia-smi.

Idle nudge - A display-only badge shown on the home page when a workspace has been idle for more than idle_nudge_days consecutive days. LabPod does not auto-stop idle workspaces.

/ws/ reverse proxy - LabPod’s built-in HTTP proxy that routes browser requests to workspace containers. Launchers use /ws/<workspace-id>/<launcher>/; user-declared ports use /ws/<workspace-id>/_user/<name>/. All workspace traffic passes through your LabPod session cookie. Raw Podman containers have no /ws/c:<container>/ or /ws/_discovered/ route.