Paddle

Python Atmospheric Dynamics: Discovering and Learning about Exoplanets. An open-source, user-friendly Python version of canoe.

Test the package

Testing the package can be done very easily by following the steps below. All we need is to create a python virtual environment, install the package, and run the test script.

1. Create a python virtual environment

   python -m venv pyenv
   

2. Install paddle package

   pip install paddle
   

3. Run test

   cd tests
   python test_saturn_adiabat.py
   

Develop with Docker

You may need to install Docker to compose up and install the package inside a container if your device or operating system does not support certain dependencies. Follow the instructions below to install docker and docker-compose plugin.

1. Install docker with compose

   curl -fsSL https://get.docker.com | sudo sh
   

2. Start docker using the command below or open docker desktop if applicable.

   sudo systemctl start docker
   

After installing Docker, you can use the Makefile commands below to manage your docker containers from the terminal. By default, the container will mount the current directory to /paddle inside the container.

Mounting a local directory allows you to edit files on your local machine while running and testing the code inside the container; or use the container as a development environment and sync files to your local machine.

If you want to change the mounted directory, you can create a docker-compose.overrides.yml file based on the provided docker-compose.overrides.yml.tmp template file.

• Create a docker container

  make up
  

• Start a docker container (only if previously created)

  make start
  

• Terminate a docker container

  make down
  

• Build a new docker image (rarely used)

  make build
  

If you use VSCode, it is recommended to install extensions including Remote Development, Dev Containers and Container Tools for a better development experience within a container.

1. Install the extensions mentioned above.

2. Start the container using make up or make start.

3. Open VSCode and click on the "Containers" icon on the left sidebar. Note that if you have too many extensions installed, the icon may be hidden under the "..." menu.

   

4. Right click on the running container named paddle with a green triangle icon (indicating it is running), and select "Attach Visual Studio Code" (see above images). This will open a new VSCode window connected to the container.

5. Open either the default folder paddle mounted from your local machine, or any custom workspace folder you have set up inside the docker-compose.overrides.yml file. Now you can start developing inside the container as if you were working on your local machine!

   

For Developers

Follow the steps below to set up your development environment.

1. Clone the repository

   https://github.com/elijah-mullens/paddle
   

2. Cache your github credential. This will prevent you from being prompted for your username and password every time you push changes to github.

   git config credential.helper 'cache --timeout=86400'
   

3. Create a python virtual environment.

   python -m venv .pyenv
   

4. Install paddle package

   # Install the package normally
   pip install paddle
   
   # [Alternatively] if you want to install in editable mode
   pip install -e .
   

5. Install pre-commit hook. This will automatically format your code before each commit to ensure consistent code style.

   pip install pre-commit
   pre-commit install
   

Cubed-Sphere Remapping

paddle can remap Snapy stitched cubed-sphere NetCDF outputs to a rectilinear lat-lon grid with TempestRemap-generated offline weights.

See docs/README_REMAP.md for a step-by-step remapping guide, including the real workspace/data example files, 0.5 degree output, variable selection, and TempestRemap cache usage.

Example:

paddle-cs-remap \
  /home/chengcli/scix/workspace/data/sub_neptune_tidallock.out1.00290.nc \
  /home/chengcli/scix/workspace/data/sub_neptune_tidallock.out2.00290.nc \
  remapped.nc \
  --nlat 181 \
  --nlon 360

The remapper treats Snapy outputs as the 2D stitched 3x2 cubed-sphere mosaic written by Snapy's NetCDF output code. Scalar fields are remapped directly. vel1, vel2, vel3 are rotated to geographic vel_east, vel_north, vel_up before remapping. Bilinear interpolation is the default for Snapy's cell-centered samples; pass --method conservative when integral preservation is required.

Install TempestRemap

Preferred option:

conda install -c conda-forge tempest-remap

Helper script:

./scripts/install_tempestremap.sh conda

Source-build fallback:

./scripts/install_tempestremap.sh source "$HOME/.local/tempestremap"
export PATH="$HOME/.local/tempestremap/bin:$PATH"

Monitor Remote Machines

paddle monitor displays CPU, GPU, process, and local disk usage for a list of machines accessible through password-less SSH.

Create ~/.config/paddle/nodelist with one SSH host or alias per line:

# Dart workstations
dart1
dart2
dart3

Run a snapshot, continuously refresh the display, or emit JSON:

paddle monitor
paddle monitor --watch 5
paddle monitor --json
paddle monitor --nodelist path/to/nodelist --top 10 --timeout 15

The remote machines need standard Linux tools and /proc. GPU metrics are collected when nvidia-smi is available. Nodes are probed concurrently using multiple threads. Individual unreachable machines are reported without preventing collection from the remaining nodes.

Submit Jobs to Remote Machines

paddle submit starts a detached command on an explicitly selected machine from the same nodelist used by paddle monitor:

paddle submit dart1 -- python train.py --epochs 10
paddle submit dart1 --cwd ~/work -- python train.py
paddle submit dart1 --log ./train.log --json -- python train.py
paddle submit dart1 --file train.py --file input.yaml -- python train.py
paddle submit dart1 --nodelist path/to/nodelist --timeout 15 -- ./run.sh

The command validates that the host is in the nodelist and connects using password-less SSH. Without --log, the command runs in the foreground and its output is displayed in the submitting terminal. Pressing Ctrl-C terminates the SSH connection and terminates the remote foreground command's process group and descendant processes. With --log REMOTE_PATH, the command runs in the remote background, writes output to that remote path, and reports its remote PID.

Jobs run from the remote home directory by default; use --cwd to select another remote directory. Before executing the command, the remote launcher sources ${HOME}/.bash_profile, including aliases when their startup scripts permit non-interactive use. Arguments after -- are executed literally; use an explicit shell command such as sh -lc 'python train.py | tee summary.txt' when pipes, redirects, or other shell expressions are needed.

Repeat --file PATH to copy local regular files into the remote working directory before execution. Each file keeps its local permission bits and is placed under its basename, atomically replacing an existing remote file with the same name. Directories and duplicate basenames are rejected.

Troubleshooting

1. If you have Docker installed but do not have Docker Compose, remove your current Docker installation, which could be docker or docker.io, and re-install it following the guide provided in the Develop with Docker section above.

2. If you run out of disk space while building the image. You can relocate the default location for container images: Here is a simple recipe:

   1. Generate config (if you don’t already have one):

   sudo mkdir -p /etc/containerd
   containerd config default | sudo tee /etc/containerd/config.toml >/dev/null
   

   2. Edit /etc/containerd/config.toml and set:

   root = "/home/containerd"   # location where you have space
   state = "/run/containerd"
   

   3. Restart

   sudo systemctl restart containerd