hytop 0.1.8

Lightweight monitoring toolkit for Hygon DCU clusters

hytop - monitoring tools

Quick start

Install from PyPI

Using pipx (recommended):

pipx install hytop
hytop --help

Using uv:

uv tool install hytop
hytop --help

Install from source

uv:

uv run hytop --help

pip:

pip install .
hytop --help

pipx:

pipx install .
hytop --help

Prerequisites

• Python >= 3.10
• Python packages: rich, typer
• Passwordless SSH for remote

hytop

# Show the version number
hytop --version

# Specify a timeout for the subcommand
hytop --timeout 300 [COMMAND]

# 0.5-second interval and 5-second rolling window for the subcommand
hytop -n 0.5 --window 5 [COMMAND]

# Specify a list of nodes for the subcommand
hytop -H node01,node02 [COMMAND]

# Specify a list of nodes with non-standard ssh ports for the subcommand
hytop -H node01:3333,node02:3333 [COMMAND]

SSH transport

hytop uses a lightweight SSH pull model and enables SSH connection reuse by default in the core layer (applies to all subcommands using SSH collection):

• ControlMaster=auto
• ControlPersist=30s
• ControlPath=~/.ssh/hytop-%C
• ServerAliveInterval=5
• ServerAliveCountMax=1

hytop gpu

A lightweight script for live hy-smi polling with rolling averages across multiple hosts. It features a modern terminal UI and can be used as a blocking scheduler for GPU jobs.

Usage

Simple examples:

# Local node, all GPUs
hytop gpu

# Two nodes, 0.5-second interval
hytop -H node01,node02 -n 0.5 gpu

# Exit with code 0 when all monitored GPUs are available
hytop gpu --devices 0,1 --wait-idle

# Wait for GPUs to be idle for 30 seconds before exiting
hytop gpu --devices 0,1 --wait-idle --wait-idle-seconds 30

# Wait at most 300s for availability (exit 124 on timeout)
hytop gpu --devices 0,1 --wait-idle --timeout 300

# Fine-grained columns (output order follows show-flag order)
hytop gpu --showtemp --showpower
hytop gpu --showpower --showtemp

Queue jobs in shared environments:

if hytop -H node01,node02 gpu --timeout 300 --wait-idle; then
  echo "GPUs available, starting workload..."
  # YOUR COMMAND HERE (e.g., python train.py)
else
  echo "Error: GPUs not available in time, aborting pipeline."
fi

Exit codes

Designed to be script-friendly:

• 0: Availability condition met (GPUs are idle).
• 124: Timeout reached before the availability condition was met.
• 130: Interrupted by the user (Ctrl+C).
• 2: Argument or input error.

Fine-grained metric flags

hytop gpu uses formatted hy-smi --json output and supports a subset of hy-smi --show* flags:

• --showtemp: GPU core temperature (Temp)
• --showpower: average package power (AvgPwr, plus AvgPwr@window)
• --showsclk: sclk frequency (sclk)
• --showmemuse: VRAM usage (VRAM%)
• --showuse: GPU utilization (GPU%, plus GPU%@window)

If no --show* flags are specified, hytop defaults to: --showtemp --showpower --showsclk --showmemuse --showuse.

hytop net

Lightweight pull-based network monitor for Ethernet and InfiniBand across one or more hosts.

Usage

# Local host, auto-discover eth+ib interfaces
hytop net

# Two hosts, 0.5-second interval
hytop -H node01,node02 -n 0.5 net

# IB-only monitoring
hytop net --kind ib

# Include only selected interfaces
hytop net --ifaces eth0,mlx5_0/p1

# Stop after 60 seconds (returns 124 on timeout)
hytop --timeout 60 net

Development

Clone the repo and run make setup to create the virtual environment, install all dependencies (including dev), and configure pre-commit hooks:

make setup

Common development commands:

make format     # Auto-fix and format code (ruff)
make lint       # Check code style and errors without modifying files
make test       # Run all unit tests (pytest)
make bump part=patch  # Bump version (patch/minor/major or X.Y.Z)
make clean      # Remove build caches and the virtual environment

Version bump

Version is managed automatically via bump-my-version. Running the bump command will:

1. Update __version__ in src/hytop/__init__.py
2. Update current_version in pyproject.toml
3. Create a commit (e.g., [hytop] Bump version: 0.1.1 → 0.1.2)
4. Create a tag (e.g., hytop-0.1.2)

make bump part=patch          # 0.1.1 -> 0.1.2
make bump part=minor          # 0.1.2 -> 0.2.0
make bump part=major          # 0.2.0 -> 1.0.0
make bump part=1.2.3          # set an explicit version

Publish

Releases are automatically published to PyPI via GitHub Actions when pushing a version tag.

# 1. Bump version (auto-commits and auto-tags)
make bump part=patch

# 2. Push commits and tags to trigger GitHub Actions release
git push --follow-tags

To test building distributions locally:

make build