Install CLIO Core.

pip

Platforms: Linux x86_64 and aarch64 (glibc 2.28+), Windows x64, macOS 14+ on Apple Silicon

The wheel is self-contained. Its dependencies are linked statically on Linux and bundled on macOS and Windows. The only extra software you might need is a FUSE driver to mount the filesystem.

$ pip install iowarp-core
$ python -c "import iowarp_core; print(iowarp_core.get_version())"

The wheel puts these commands on your PATH: clio_run, clio_cte_fuse (Linux and Windows), clio_cte_bench and clio_run_thrpt_bench. It also gives you the Python modules iowarp_core, clio_cee and clio_cte_core_ext.

The first time Python imports iowarp_core, it copies the default configuration to ~/.clio/clio.yaml. It never overwrites an existing file.

conda

Platforms: linux-64, Python 3.12

Packages are published to the iowarp channel on Anaconda.org for linux-64 and Python 3.12. They are built from the release preset and do not include the FUSE adapter.

$ conda create -n iowarp -c iowarp -c conda-forge python=3.12 iowarp-core
$ conda activate iowarp
$ clio_run --help

Build the recipe yourself

To enable other features, build the recipe yourself. IOWARP_PRESET picks a preset from CMakePresets.json:

$ git clone --recurse-submodules https://github.com/iowarp/clio-core.git
$ cd clio-core
$ conda install -n base -y conda-build -c conda-forge
$ IOWARP_PRESET=release conda build installers/conda/ -c conda-forge --output-folder build/conda-output
$ conda install -c conda-forge build/conda-output/*/iowarp-core-*.conda

Docker

Platforms: amd64, arm64

The iowarp/deploy-cpu image is multi-arch (amd64 and arm64). It includes MPI, the HDF5 VOL connector, ADIOS2 and compression. It runs as the iowarp user. Its default command is a shell, so pass clio_run start yourself.

$ docker pull iowarp/deploy-cpu:latest
$ docker run -d -p 9413:9413 -p 8080:8080 -e CLIO_VIZ_BIND=0.0.0.0 \
    --memory=8g --name iowarp iowarp/deploy-cpu:latest clio_run start

Port 9413 is the runtime RPC port and 8080 is the dashboard. :latest is rebuilt from the main branch.

As a long-running service

For a long-running service, mount your own config and a volume for state. That keeps the persistent tier, the metadata log and the search index across docker compose down:

services:
  iowarp:
    image: iowarp/deploy-cpu:latest
    volumes:
      - ./clio.yaml:/etc/iowarp/clio.yaml:ro
      - iowarp-state:/home/iowarp/.clio
    environment:
      - CLIO_SERVER_CONF=/etc/iowarp/clio.yaml
      - CLIO_VIZ_BIND=0.0.0.0
    ports:
      - "9413:9413"
      - "8080:8080"
    mem_limit: 8g
    command: ["clio_run", "start"]
    restart: unless-stopped
volumes:
  iowarp-state:

CLIO Core uses memfd_create() for shared memory on Linux, so /dev/shm needs no tuning. Only mem_limit matters.

Spack

Platforms: Linux x86_64 and aarch64

1. Install a recent Spack

   shell

   $ git clone --depth=2 https://github.com/spack/spack.git
   $ . spack/share/spack/setup-env.sh

2. Add the IOWarp repository

   shell

   $ git clone --recurse-submodules https://github.com/iowarp/clio-core.git
   $ spack repo add clio-core/installers/spack

3. Install

   shell

   $ spack install iowarp +fuse

   The package tracks the main (default) and dev branches. Variants include +fuse, +python, +adios2, +mochi, +compress, +encrypt, +cuda, +rocm, +s3 and +gcs. HDF5, MPI-IO and ELF interception are on by default in Spack.

Packages

Platforms: Ubuntu 24.04, Fedora 40, Windows x64

Each tagged release on GitHub Releases carries .deb (Ubuntu 24.04) and .rpm (Fedora 40) packages and an AppImage for x86_64 and arm64, plus a ZIP and an installer for Windows x64. They include the FUSE adapter but not the Python bindings. They do not create ~/.clio/clio.yaml. Without that file the runtime uses built-in defaults. Set CLIO_SERVER_CONF to use your own file.

Source

Platforms: Linux, macOS, Windows

You need a C++20 compiler (GCC 11+ or Clang 14+) and CMake 3.20 or newer.

$ sudo apt install -y build-essential cmake ninja-build pkg-config git \
    python3 python3-pip python3-venv libfuse3-dev fuse3 \
    libyaml-cpp-dev libcereal-dev libmsgpack-dev libsodium-dev libzmq3-dev
$ git clone --recurse-submodules https://github.com/iowarp/clio-core.git
$ cd clio-core
$ cmake --preset release-fuse
$ cmake --build build -j"$(nproc)"
$ sudo cmake --install build

$ brew install --cask macfuse
$ git clone --recurse-submodules https://github.com/iowarp/clio-core.git
$ cd clio-core
$ ./CI/ci-deps.sh --only-deps release
$ conda activate iowarp
$ cmake --preset release-mac-fuse -DCLIO_CORE_ENABLE_CONDA=ON
$ cmake --build build-mac-fuse -j"$(sysctl -n hw.ncpu)"

PS> git clone --recurse-submodules https://github.com/iowarp/clio-core.git
PS> cd clio-core
PS> cmake --preset windows-release -DCLIO_CTE_ENABLE_FUSE_ADAPTER=ON
PS> cmake --build build --config Release

Presets

Feature options

When an adapter's dependency is missing, the build skips it with a warning rather than failing. Check the configure output for the features you need.