gflowd Reference

gflowd manages the local gflow daemon.

Usage

gflowd [options] [command]
gflowd completion <shell>

Common Examples

# Initialize config interactively
gflowd init

# Initialize config non-interactively with defaults
gflowd init --yes

# Start the daemon
gflowd up

# Start with restricted GPUs and random allocation
gflowd up --gpus 0,2 --gpu-allocation-strategy random

# Start with faster GPU occupancy polling
gflowd up --gpu-poll-interval-secs 3

# Reload without downtime
gflowd reload

# Restart the daemon with a new GPU restriction
gflowd restart --gpus 0-3

# Check status or stop the daemon
gflowd status
gflowd down

Global Options

• -c, --config <path>: use a custom config file
• --cleanup: clean up the configuration file
• -v/-vv/-vvv/-vvvv: increase daemon logging verbosity
• -q: reduce daemon logging verbosity

Commands

gflowd init

Create or update the configuration file via a guided wizard.

gflowd init [--yes] [--force] [--advanced] [--gpus <indices>] [--host <host>] [--port <port>] [--timezone <tz>] [--gpu-allocation-strategy <strategy>] [--gpu-poll-interval-secs <seconds>]

Options:

• --yes: accept all defaults without prompts
• --force: overwrite an existing config file
• --advanced: configure advanced options such as notifications
• --gpus <indices>: restrict scheduler-visible GPUs, for example 0,2 or 0-2
• --host <host>: daemon host (default: localhost)
• --port <port>: daemon port (default: 59000)
• --timezone <tz>: store a timezone like Asia/Shanghai or UTC; use local to leave it unset
• --gpu-allocation-strategy <strategy>: sequential or random
• --gpu-poll-interval-secs <seconds>: poll NVML for GPU occupancy changes every N seconds (default: 10, minimum: 1)

gflowd up

Start the daemon in a tmux session.

gflowd up [--gpus <indices>] [--gpu-allocation-strategy <strategy>] [--gpu-poll-interval-secs <seconds>]

gflowd reload

Reload the daemon with zero downtime.

gflowd reload [--gpus <indices>] [--gpu-allocation-strategy <strategy>] [--gpu-poll-interval-secs <seconds>]

Use this when you want to refresh the running daemon without stopping it first.

gflowd restart

Stop the daemon and start it again.

gflowd restart [--gpus <indices>] [--gpu-allocation-strategy <strategy>] [--gpu-poll-interval-secs <seconds>]

Use this when a full restart is acceptable or needed.

gflowd status

Show whether the daemon is running and responding to health checks.

gflowd status

gflowd down

Stop the daemon.

gflowd down

gflowd completion <shell>

Generate shell completion scripts.

gflowd completion bash
gflowd completion zsh
gflowd completion fish

Notes

• --gpus affects which GPUs the scheduler may allocate for new work.
• --gpu-allocation-strategy accepts sequential or random.
• --gpu-poll-interval-secs controls how quickly unmanaged GPU occupancy changes are detected.
• gflowd up, reload, and restart all accept the same GPU-related overrides.

See Also

• Configuration
• GPU Management
• Quick Reference