jupyterlab_kernel_terminal_workspace_culler_extension

Automatically cull idle kernels, terminals, and workspaces after configurable timeout periods. Helps manage system resources by cleaning up unused resources that accumulate during long JupyterLab usage.

Features

• Idle kernel culling - Shut down kernels idle beyond timeout (checks execution_state and last_activity)
• Idle terminal culling - Close terminals with no WebSocket activity beyond timeout, and any terminal idle beyond a maximum (7 days by default)
• Workspace culling - Remove stale JupyterLab workspaces, named and auto-generated (auto-0, auto-k, etc.), based on last modified time; the default workspace is kept
• Configurable timeouts - All timeouts adjustable via JupyterLab Settings
• Notifications - Optional toast notifications when resources are culled (requires jupyterlab-notifications)
• Server-side detection - Uses tornado PeriodicCallback for accurate activity tracking

Default Settings

How Idle Detection Works

Kernels: Checked for execution_state (busy kernels are never culled) and last_activity timestamp. A kernel is idle when it's not executing and hasn't had activity beyond the timeout.

Terminals: A terminal referenced by any existing workspace is not culled - the workspace must be culled first, which releases its terminals unless another surviving workspace still references them (the cascade). Beyond that, only terminals with no active browser tab are culled by default (controlled by "Only Cull Disconnected Terminals" setting). When a terminal tab is open, it maintains a WebSocket connection and the terminal is kept. Once the tab is closed or disconnected, the terminal becomes eligible for culling one full idle timeout later. None of this protection lasts past the terminal maximum idle ("Terminal Maximum Idle", 7 days by default): a terminal with no input or output for that long is culled even with an open tab or a workspace reference. A terminal counts as culled only once the server has actually removed it: a terminal whose shell has exited but whose pty is still held open by a surviving process is closed the way an end-of-file would have closed it, and a terminal that survives a cull anyway is attempted once and then left alone.

Workspaces: Based on the workspace file's last_modified timestamp. Every workspace except the default one is culled once idle past the workspace timeout: the auto-named workspaces (auto-0, auto-k, etc.) JupyterLab creates when you open multiple windows, and named workspaces you saved yourself. Culling a workspace releases the terminals it referenced, so they can be culled in the same pass (the cascade).

Note: Terminal culling signals the terminal's own shell, starting with SIGHUP and escalating through SIGINT and SIGTERM to SIGKILL if it does not exit. Processes started with nohup, screen or tmux are detached from that shell and survive culling.

Installation

Requires JupyterLab 4.0.0 or higher.

pip install jupyterlab-kernel-terminal-workspace-culler-extension

Configuration

Open JupyterLab Settings (Settings -> Settings Editor) and search for "Resource Culler" to adjust timeouts and enable/disable culling for each resource type.

Logs

Culling actions are logged at INFO level with [Culler] prefix:

[Culler] CULLING KERNEL abc123 - idle 65.2 minutes (threshold: 60)
[Culler] Kernel abc123 culled successfully
[Culler] CULLING TERMINAL 1 - idle 62.1 minutes (threshold: 60)
[Culler] Terminal 1 culled successfully

Run JupyterLab with --log-level=INFO to see culling activity.

FAQ

Q: My long-running calculation was killed. How do I prevent this?

Two options:

1. Increase timeout: Go to Settings -> Settings Editor -> Resource Culler and increase the kernel/terminal timeout; for a terminal that stays open in a tab, increase "Terminal Maximum Idle"
2. Use a terminal multiplexer: Run calculations inside screen or tmux - these survive terminal culling

# Using screen
screen -S mysession
python long_calculation.py
# Detach with Ctrl+A, D

# Using tmux
tmux new -s mysession
python long_calculation.py
# Detach with Ctrl+B, D

Q: Will closing my browser tab kill my running process?

For terminals: By default a terminal with an open browser tab is kept until it has had no input or output for the terminal maximum idle (7 days by default), even while a command runs; after the tab closes, it is culled once the idle timeout expires unless a workspace still references it; closing the browser tab leaves the terminal in the saved workspace, which keeps it until that workspace is culled or the terminal maximum idle passes. Foreground processes receive SIGHUP. Use nohup, screen, or tmux for processes that must survive.

For kernels: The kernel continues running. Activity is tracked server-side, so a busy kernel won't be culled even if the browser is closed.

Q: What happens to processes started with nohup?

They survive terminal culling. nohup makes processes ignore SIGHUP, which is the signal sent when a terminal closes.

Q: How do I disable culling entirely?

Go to Settings -> Settings Editor -> Resource Culler and uncheck "Enable Kernel Culling" and "Enable Terminal Culling".

Q: Can I see when resources were culled?

Yes. Run JupyterLab with --log-level=INFO to see [Culler] log messages. If you have jupyterlab-notifications installed, you'll also see toast notifications.

CLI

The extension includes a command-line tool for listing and culling resources from the terminal.

# Show help
jupyterlab_kernel_terminal_workspace_culler

# List all resources and their idle times
jupyterlab_kernel_terminal_workspace_culler list

# List as JSON
jupyterlab_kernel_terminal_workspace_culler list --json

# Show what would be culled (dry run)
jupyterlab_kernel_terminal_workspace_culler cull --dry-run

# Cull idle resources
jupyterlab_kernel_terminal_workspace_culler cull

# Cull and output as JSON
jupyterlab_kernel_terminal_workspace_culler cull --json

# Custom timeouts (minutes)
jupyterlab_kernel_terminal_workspace_culler cull --kernel-timeout 30 --terminal-timeout 120

# Cull workspaces idle > 1 minute (default is 7 days)
jupyterlab_kernel_terminal_workspace_culler cull --workspace-timeout 1

# Also cull terminals with an open browser tab or a workspace reference (default: skip them)
jupyterlab_kernel_terminal_workspace_culler cull --include-connected

By default cull skips terminals with an open browser tab or a workspace reference; unlike the extension, it does not apply the terminal maximum idle. Pass --include-connected to cull them anyway. A terminal is printed as culled only when the server removed it; one that is still registered afterwards is printed as failed. list marks the workspaces the server never culls as (protected).

The CLI auto-discovers running Jupyter servers. You can also set environment variables:

• JUPYTER_SERVER_URL - server URL (e.g., http://localhost:8888/)
• JUPYTER_TOKEN - authentication token

Uninstall

pip uninstall jupyterlab-kernel-terminal-workspace-culler-extension