Coiled account for Fast-cGAN software environment and doing forecast inference

[nishadhka]

Overview

This issue documents the setup process for running the Fast-CGAN model entirely on a cloud-based environment using Coiled and Docker. The goal is to containerize the software environment, push the container to Google Artifact Registry, and configure a Coiled software environment to run Fast-CGAN seamlessly.

---

Steps to Build and Push the Docker Image

The following commands build the Fast-CGAN Docker image and push it to Google Artifact Registry:

docker build -t us-east1-docker.pkg.dev/sewaa-416306/coiled/fast-cgan:v0 -f coiled-dockerfile .
docker push us-east1-docker.pkg.dev/sewaa-416306/coiled/fast-cgan:v0

Authentication and Configuration

To authenticate with Google Cloud and configure Docker for pushing images:

export GOOGLE_APPLICATION_CREDENTIALS="/path/to/key.json"
gcloud auth activate-service-account --key-file=$GOOGLE_APPLICATION_CREDENTIALS
gcloud auth configure-docker us-east1-docker.pkg.dev

Verify the authentication:

gcloud auth list --filter=status:ACTIVE --format="value(account)" | grep iam.gserviceaccount.com

---

Setting Up Coiled Environment

Once the Docker image is built and pushed, we define a Coiled software environment to use this container:

import coiled

coiled.create_software_environment(
    name="fast-cgan-v0",
    container="us-east1-docker.pkg.dev/sewaa-416306/coiled/fast-cgan:v0",
)

To start a Jupyter notebook in Coiled using this environment:

coiled notebook start --name fast-cgan-notebook --vm-type n2-standard-2 --software fast-cgan-v0

---

Dockerfile for Fast-CGAN Environment

The coiled-dockerfile below details the Fast-CGAN setup, including dependencies, model files, and integration with Coiled.

ARG PYTHON_VERSION=3.10

FROM alpine/git AS builder
ARG API_REPO=https://github.com/icpac-igad/fast-cgan.git
ARG GAN_REPO=https://github.com/jaysnm/ensemble-cgan.git
ARG GAN_BRANCH=Jurre_brishti

RUN git clone --depth 1 ${API_REPO} /tmp/api && \
    git clone --depth 1 -b ${GAN_BRANCH} ${GAN_REPO} /tmp/cgan

FROM python:${PYTHON_VERSION}-slim AS runner

ARG USER_NAME=cgan
ARG USER_ID=1000
ARG GROUP_ID=1000
ARG WORK_HOME=/opt/cgan

# Install system libraries
RUN apt-get update -y && \
    apt-get install -y --no-install-recommends git rsync ssh ca-certificates pkg-config wget \
    libgdal-dev libgeos-dev libproj-dev gdal-bin libcgal-dev libxml2-dev libsqlite3-dev  \
    gcc g++ dvipng libfontconfig-dev libjpeg-dev libspng-dev libx11-dev libgbm-dev \
    libeccodes-dev libeccodes-tools && \
    mkdir -p ${WORK_HOME}/.local/bin ${WORK_HOME}/.ssh

RUN groupadd --gid ${GROUP_ID} ${USER_NAME} && \
    useradd --home-dir ${WORK_HOME} --uid ${USER_ID} --gid ${GROUP_ID} ${USER_NAME} && \
    chown -Rf ${USER_NAME}:${USER_NAME} ${WORK_HOME}

USER ${USER_NAME}
WORKDIR ${WORK_HOME}
ENV PATH=${WORK_HOME}/.local/bin:$PATH

# Download and extract model initialization data
RUN mkdir -p ${WORK_HOME}/data/models-config/jurre-brishti && \
    mkdir -p ${WORK_HOME}/data/models-config/mvua-kubwa && \
    wget -v https://cgan.icpac.net/ftp/models-config/jurre-brishti.tar.gz -P /tmp && \
    wget -v https://cgan.icpac.net/ftp/models-config/mvua-kubwa.tar.gz -P /tmp && \
    tar -xvzf /tmp/jurre-brishti.tar.gz -C ${WORK_HOME}/data/models-config/jurre-brishti && \
    tar -xvzf /tmp/mvua-kubwa.tar.gz -C ${WORK_HOME}/data/models-config/mvua-kubwa && \
    rm /tmp/jurre-brishti.tar.gz /tmp/mvua-kubwa.tar.gz

# Copy cloned repositories
COPY --from=builder --chown=${USER_NAME}:root /tmp/cgan ${WORK_HOME}/ensemble-cgan
RUN cd ${WORK_HOME}/ensemble-cgan && pip install --upgrade pip && pip install --no-cache-dir -e .

COPY --from=builder --chown=${USER_NAME}:root /tmp/api/pyproject.toml /tmp/api/poetry.lock /tmp/api/README.md ${WORK_HOME}/
COPY --from=builder --chown=${USER_NAME}:root /tmp/api/fastcgan ${WORK_HOME}/fastcgan

# Install Fast-CGAN API and dependencies
RUN cd ${WORK_HOME} && pip install --no-cache-dir -e . && \
    pip install eccodes cfgrib && \
    pip install coiled dask distributed dask-kubernetes google-cloud-storage && \
    touch ${WORK_HOME}/.env

# Create directories for data storage
RUN mkdir -p ${WORK_HOME}/jobs ${WORK_HOME}/forecasts ${WORK_HOME}/logs

# Set environment variables
ENV WORK_HOME=${WORK_HOME}
ENV JOBS_DATA_DIR=${WORK_HOME}/jobs
ENV FORECASTS_DATA_DIR=${WORK_HOME}/forecasts
ENV LOGS_DIR=${WORK_HOME}/logs

---

Guide for New Members: Setting Up Coiled and Running Fast-CGAN

1. Coiled Setup and Login

Step 1: Install Coiled Library

Ensure you have Python 3.8+ installed, then install Coiled:

pip install coiled

Step 2: Log into Coiled

Each user needs to log in to Coiled using the following command:

coiled login

This will open a browser window prompting you to authenticate with Coiled. After logging in, you’ll be connected to the shared workspace.

Step 3: Verify Workspace Access

Ensure that you are part of the correct Coiled workspace:

coiled whoami

If needed, switch to the workspace assigned for Fast-CGAN (replace workspace-name with the actual name: gcp-sewaa-nka):

coiled set workspace workspace-name

---

2. Creating a Jupyter Notebook in Coiled

Once logged in, each member can create their own notebook environment in Coiled.

Step 4: Start a Notebook

To start a Jupyter notebook using the pre-built Fast-CGAN software environment:

coiled notebook start --name fast-cgan-notebook --vm-type n2-standard-2 --software fast-cgan-v0

This will:

• Spin up a Jupyter Notebook on Coiled Cloud
• Use the pre-configured fast-cgan-v0 software environment
• Assign a n2-standard-2 VM for execution

Once the notebook is started, Coiled will display a URL. Open this in your browser to access the notebook.

---

6. Shutting Down the Notebook

When finished, members should stop their notebook session to free up resources:

coiled notebook stop --name fast-cgan-notebook

Troubleshooting

• If Coiled authentication fails, re-run:

  coiled login

• If the notebook doesn't start, check workspace permissions:

  coiled whoami

• If packages are missing inside the notebook:

  pip install --user <package_name>

---