USAGE:(1)

πŸ“¦ version 10.0, in ubuntu stonking

🌸 NAME 🌸

usage: β€” the ROCm Deployment Health Check tool, bestie. This is the rdhc manual, section 1, and it's here to tell you literally everything about running the deployment health check. 🩺✨


πŸ’… DESCRIPTION πŸ’…

okay so first things first, here's how you actually run this thing:

usage: ./rdhc.py [options]

It's the ROCm Deployment Health Check Tool. It pokes at your ROCm install and tells you what's up. 🫢

✨ options: ✨

  • -h, --help β€” shows this help message and dips. that's it, that's the whole flag.
  • --quick β€” runs the quick tests only. this is the default vibe, so you're already doing it. πŸ’…
  • --all β€” the full default test suite PLUS it compiles and executes a simple program for each component. the glow-up version of the test run. πŸ”₯
  • -v, --verbose β€” turns on verbose output. you want the tea? here's all the tea. 🍡
  • -s, --silent β€” silent mode. errors only. if nothing's wrong, you hear nothing at all. iconic, honestly. 🀫
  • -j, --json FILE β€” exports results to a JSON file. take your data home with you.
  • -d, --dir DIR β€” the directory path for temporary files. and listen: it must not already exist. it gets created with mode 0700. default is an autogenerated secure temp directory in /tmp (and no, it does not get autocleaned). want to reuse a directory? remove it first: rm -rf <dir>.
  • --cleanup β€” removes the auto-generated temp directory after the tests finish. ⚠️ this only applies when -d is not specified.
  • --rocm-install-prefix DIR β€” your ROCm installation prefix. if you set it, it overrides ROCM_PATH; otherwise ROCM_PATH or /opt/rocm gets used.
  • --skip-multinode-readiness β€” skips the multinode/RDMA readiness check. use this on nodes with no InfiniBand or RDMA stack. πŸ¦‹
  • --skip-atomic-operations β€” skips the PCIe atomic operations check. use it in containers or whenever lspci can't read capabilities.
  • --skip-optional-cluster-checks β€” skips multinode readiness AND atomic operations. it's literally the same as passing both --skip-* flags above at once. two birds, one flag. πŸ‘‘
  • --kernel-params-warnings-only β€” for the kernel parameter checks (numa, iommu, pci realloc, etc.), this treats failures as warnings (is_error=false) instead of counting them as errors. soft launch instead of a whole scandal. πŸ™ˆ

Peep the README at @<ROCM_INSTALL_PATH>/share/rdhc/README.md for more.

🌈 Usage examples 🌈

# Run quick test (default tests only)
./rdhc.py

# Run all tests including compile and execute the rocm-example program for each component
./rdhc.py --all

# Run all tests with verbose output
./rdhc.py --all -v

# Enable verbose output
./rdhc.py -v

# Run in silent mode (only errors shown)
./rdhc.py -s

# Export results to a specific JSON file
./rdhc.py --all --json rdhc-results.json

# Specify a directory for temp files (directory must not exist)
./rdhc.py --all -d /home/user/rdhc-dir/

# Auto-generate temp directory and clean it up after tests
./rdhc.py --all --cleanup

# Use a custom ROCm install prefix
./rdhc.py --rocm-install-prefix /usr/local/rocm

# Skip optional checks (e.g. single-node, containers without full PCI/RDMA)
./rdhc.py --skip-optional-cluster-checks
# (or: --skip-multinode-readiness --skip-atomic-operations)

# Kernel cmdline/sysctl checks: failures as warnings only (no FAIL on mismatch)
./rdhc.py --kernel-params-warnings-only

NOTE: some checks need root privileges β€” like lspci -vvv for the atomic operations check. run as root when you need to, or use --skip-atomic-operations if you're running unprivileged. no cap, some tests just want the crown. πŸ‘‘
NOTE for Ubuntu 24.04 (Python 3.12) users: because of the enhanced security policies, you must use a virtual environment:
# Create and activate virtual environment (one-time setup)
python3 -m venv ~/rdhc-venv
source ~/rdhc-venv/bin/activate
pip3 install -r requirements.txt

# Run the tool
./rdhc.py
./rdhc.py --all

and if you try something spicy that doesn't exist, the tool will simply say:

rdhc: error: unrecognized arguments: --version

not the tool reading you to filth, just the tool being honest. πŸ’€


okay that's the whole page, mwah πŸ’– this was rdhc's manual, USAGE:(1), dated September 2026 β€” love her, she ate, periodt. πŸŽ€πŸ¦‹πŸ’œ

πŸ“¦ where this page comes from