Running the DAX & XMLA Server

Run server

The server scans --models-root for every *.json TMSL file and loads each as a separate model. With no flags at all, it defaults to ./demodata for both models and datasets — generate the demo dataset first with cargo run --bin generate-demodata, then:

cargo run --bin dax-rs

Or point it at your own models directory:

cargo run --bin dax-rs -- --models-root /path/to/models --datasets-root /path/to/data

Key flags:

• --models-root <dir> — directory scanned for *.json TMSL model files, each loaded as a separate model (default: ./demodata)
• --datasets-root <dir> — directory containing parquet data files (default: ./demodata)
• --port <n> — listen port (default: 3000)
• --config <file> — YAML config file path (default: server.yaml if present)

Docker

Build the image (stable Rust, no extra polars perf flags):

docker build -t dax-rs .

Build with polars’ performant feature (more fast paths, slower compile — recommended for production):

docker build -t dax-rs --build-arg POLARS_FEATURES=performant .

Build a nightly image (SIMD + specialization; requires the nightly toolchain, pulled in automatically via RUST_CHANNEL). performant and nightly are independent — combine them explicitly:

docker build -t dax-rs:rust-nightly \
  --build-arg RUST_CHANNEL=nightly \
  --build-arg POLARS_FEATURES=performant,nightly \
  .

Run with a directory of models (every *.json file found under --models-root/DAX_MODELS_ROOT is loaded as a separate model):

docker run -p 3000:3000 \
  -v /path/to/models:/models \
  -e DAX_MODELS_ROOT=/models \
  -e DAX_DATASETS_ROOT=/models \
  dax-rs

Run with S3 datasets:

docker run -p 3000:3000 \
  -v /path/to/models:/models \
  -e DAX_MODELS_ROOT=/models \
  -e DAX_DATASETS_TYPE=s3 \
  -e DAX_DATASETS_BUCKET=my-bucket \
  -e DAX_DATASETS_REGION=eu-west-1 \
  -e DAX_DATASETS_ACCESS_KEY_ID=AKIA... \
  -e DAX_DATASETS_SECRET_ACCESS_KEY=... \
  dax-rs

The server exposes GET /health which returns {"status":"ok"} and is used as the Docker healthcheck. The dashboard is available at http://localhost:3000/.