gctl Reference

gctl changes scheduler behavior at runtime.

Usage

gctl <command> [args]
gctl completion <shell>

Commands

gctl show-gpus

Show per-GPU status, including whether a GPU is restricted.

gctl show-gpus

gctl gpu-process ignore --gpu <index> --pid <pid>

Ignore a running GPU process when gflow evaluates whether a GPU is blocked by an unmanaged workload.

This is a runtime-only override. gflowd restart or reload clears it automatically.

gctl gpu-process ignore --gpu 0 --pid 1234

gctl gpu-process unignore --gpu <index> --pid <pid>

Remove a runtime GPU-process ignore override.

gctl gpu-process unignore --gpu 0 --pid 1234

gctl gpu-process list

List active runtime GPU-process ignore overrides.

gctl gpu-process list

gctl set-gpus <gpu_spec>

Restrict which GPUs the scheduler can allocate for new jobs.

<gpu_spec> examples:

• all
• 0,2,4
• 0-3
• 0-1,3,5-6

gctl set-gpus 0,2
gctl set-gpus all

gctl set-limit <job_or_group_id> <limit>

Set max concurrency for a job group.

gctl set-limit <job_id> 2
gctl set-limit <group_id> 2

gctl reserve create

Create a GPU reservation for a specific user.

By GPU count (scheduler allocates dynamically):

gctl reserve create --user alice --gpus 2 --start '2026-01-28 14:00' --duration 2h

By specific GPU indices (reserve exact GPUs):

gctl reserve create --user alice --gpu-spec 0,2 --start '2026-01-28 14:00' --duration 2h
gctl reserve create --user bob --gpu-spec 0-3 --start '2026-01-28 16:00' --duration 1h

--start supports ISO8601 (e.g. 2026-01-28T14:00:00Z) or YYYY-MM-DD HH:MM (local time). Times must be on :00 or :30; durations are multiples of 30 minutes.

gctl reserve list

List reservations.

gctl reserve list
gctl reserve list --active
gctl reserve list --user alice --status active
gctl reserve list --timeline --range 48h

gctl reserve get <reservation_id>

Show details for a reservation.

gctl reserve get <reservation_id>

gctl reserve cancel <reservation_id>

Cancel a reservation.

gctl reserve cancel <reservation_id>