DAVAI

This project contains:

• the davai command-line interface, to create davai experiments and run them,
• the davai-specific utilities, resources and AlgoComponents for Vortex (plugin)
• the davai tasks templates, jobs sequences and config files
• the documentation

This repository is the result of the merge of:

• the davai sub-package from Vortex (1.x)
• the DAVAI-env repository (command-line interface)
• the DAVAI-tests repository (utilities, tasks & jobs)

Installation

• on belenos@MF

• on Atos/bologna@ECMWF

• or to setup your own install:

  • Create a venv:

    python -m venv ~/venvs/davai`
    source ~/venvs/davai
    

  • Then either install your own working version:
    • Clone this repository, e.g. in ~/repositories/ and install it:

      git clone https://github.com/ACCORD-NWP/DAVAI.git
      pip install [-e] DAVAI
      

    • Or from PyPI:

      pip install davai
      

Quick start

For a quick start:

1. Prepare an experiment based on version <v> of the Davai tests, to validate an IAL Git reference <r>:
   davai-new_xp <r> [-v <v>] [-h]
   Notes:
   • you may need to specify the path to your IAL repository by argument, cf. options with -h.
   • to know what version of the tests to use, cf. below. From CY50+ onwards, a default version can be infered from the IAL repository.
2. Go to the prompted directory of the experiment (ending with .../dv-<nnnn>-<platform>@<user>/davai/nrv/) and source the associated python virtualenv as prompted:
   source venv/bin/activate
3. Run the tests: davai-run_xp and monitor (standard output for the build, then job scheduler).
4. If you need to re-build & re-run tests:

• davai-build [-h]
• davai-run_tests [-h]

More details to be found in the documentation, in particular in case of a multi-repos contribution.

Tests versions and reference experiments

From CY50+, a default version of Davai can be found in IAL repo, file .davai_default_version.

However, new versions can be recommended, unrelated to changes in the IAL. For that matter, it will be referenced here

Documentation

Cf. here.

Dependencies

DAVAI is mainly written in Python. Make sure you have Python3.10 at least. Python dependencies are included in the pyproject.toml file. They include in particular:

• IAL-build: wrappers around git and gmkpack (and eventually other building tools) to build IAL executables from Git
• IAL-expertise: tools to analyse automatically the outputs of NWP configurations -- norms, Jo-tables, fields in FA/GRIB files, ...
• EPyGrAM: a Python library for handling output data from the IAL models; it is used here within IAL-expertise. Note that the Vortex package, currently used in version 1.x, is not yet available in these dependencies; cf. below.

It also uses Git, make sure you have a "recent enough" version of it, or some commands may not work properly. DAVAI works over a number of NWP packages, tools, softwares, that need to be pre-installed on the machine with their own procedures. These include:

• Vortex: scripting system used for the definition of tasks (resources, executables launch, ...) and the running of the jobs. It embeds a number of necesary-as-well sub-packages.
• Gmkpack: build system for IAL. Documentation is outdated.

These packages may already be pre-installed on MF's HPC or other platforms where DAVAI is already ported. Cf. the "Local complements" in "Installation" section, in this case.

To install them on a new machine, cf. the projects' installation instructions.

Lexicon

• DAVAI = "Device Aiming at the VAlidation of IAL"
• IAL = IFS-Arpege-LAM