Building Drava from source¶

The C++ runtime depends on several native libraries. On the ALCF JLSE cluster, use docs/jlse.md instead — the dependencies are preinstalled there and that guide is copy-paste ready. This page is the generic build for other environments.

The example apps and the drava-pipeline CLI are pure Python and run anywhere; only the C++ runtime (the drava module) needs the build below.

Dependencies¶

A C/C++ compiler with C++20 support (tested with LLVM/Clang).
xkrt — the task runtime Drava is built on.
yaml-cpp — pipeline config parsing.
SWIG — generates the Python bindings.
A no-GIL Python build (3.13+ compiled with --disable-gil).
Optional: a NATS server + nats.c client for the JetStream transport.
Optional: NVML/CUDA for GPU energy reporting.

Build yaml-cpp¶

git clone https://github.com/jbeder/yaml-cpp.git
cd yaml-cpp && mkdir build && cd build
CC=clang CXX=clang++ cmake .. -DYAML_BUILD_SHARED_LIBS=ON \
    -DCMAKE_INSTALL_PREFIX=$HOME/opt/yaml-cpp-install
make -j && make install

(Optional) NATS for the JetStream transport¶

# NATS server
curl -fsSL https://binaries.nats.dev/nats-io/nats-server/v2@v2.11.6 | sh

# NATS C client
git clone https://github.com/nats-io/nats.c.git
cd nats.c && mkdir build && cd build
cmake .. -DNATS_BUILD_STREAMING=OFF -DCMAKE_INSTALL_PREFIX=$HOME/opt/nats
make -j && make install

Build Drava¶

export NATS_ROOT=$HOME/opt/nats     # only if using the JetStream transport
export NVML_ROOT=$CUDA_HOME         # only for GPU energy
mkdir build && cd build
CC=clang CXX=clang++ cmake -DCMAKE_BUILD_TYPE=Debug ..
make -j
export PYTHONPATH="$(pwd):$PYTHONPATH"   # so `import drava` finds the built module

CMake prints whether the NATS and NVML backends were enabled. Confirm the module imports:

python -c "import drava; print('drava OK')"

Next: running the examples.