Working with Conda

Databricks can support Conda based environments. FSDH provides two options for users to work with Conda.

1. Project specific Docker image with Conda support and a predefined Conda environment. The Docker image needs to be co-developed with the FSDH support team and pushed to GitHub Container Registry (GHCR).
2. Generic Docker image with Conda support. Users will need to install packages in their notebooks.

For illustration purposes, the following steps are based on option 1.

Step 1: Create environment YAML

Sample code for env.yml. Skip to Step 3 if using an existing Docker image.

name: fsdh-sample
channels:
  - bioconda
  - default
dependencies:
  - python=3.8.16
  - pip=23.0.1
  - six=1.16.0
  - ipython=8.12.0
  - nomkl=3.0
  - numpy=1.24.3
  - pandas=1.1.5
  - traitlets=5.7.1
  - wheel=0.38.4
  - hifiasm=0.16.1
  - pip:
    - pyarrow==1.0.1

Step 2. Build and Push the Image

The FSDH team builds and pushes the image to GitHub. Skip to Step 3 if using an existing Docker image.

docker build -t fsdh-sample .
export GHCR_PAT="XXX"
echo $GHCR_PAT|docker login ghcr.io -u <username> --password-stdin
docker tag fsdh-sample ghcr.io/ssc-sp/fsdh-sample:latest
docker push ghcr.io/ssc-sp/fsdh-sample:latest

Step 3. Create a cluster

1. Ask your admin to enable Container Service for your Databricks workspace.
2. Create a cluster with access mode "No Isolation Shared"
3. Choose a runtime of 10.4-LTS, 9.1-LTS or 7.3-LTS
4. Under Advance -> Docker, use image ghcr.io/ssc-sp/fsdh-sample:latest

Step 4. Validate the Cluster

Run the following code:

%sh
conda list