K8S Cluster Running Environment Configuration

— Dockerfile boilerplates for configuring kubernetes (K8S) cluster running environment
— Following instructions are based on the user name pchen6 with the boilerplate PIP-DL-Env, please replace all pchen6 to your own user name for experimentation

Configuration Procedures

Step 1: Prepare Dockerfile

• Access Dockerfile within the PIP-DL-Env folder (further adapt it for your purpose)

$ git clone https://github.com/idso-fa1-pathology/k8s-cluster-env.git
$ cd k8s-cluster-env/PIP-DL-Env

Step 2. Build Docker Image

• Build docker image from Dockerfile, the first-time building process can take long (5-30 mins)

$ docker build --platform linux/x86_64 -t pipdlenv:pchen6 . 

Step 3. Push Docker Image to HPC-Harbor

• On HPC-Harbor, create a public project named as your user name (e.g., pchen6), for storing docker images
• On your local machine, first login to hpcharbor, next tag the built docker image as hpcharbor format, and then push the tagged docker image to hpcharbor

$ docker login hpcharbor.mdanderson.edu                                       # login HPC-Harbor
$ docker tag pipdlenv:pchen6 hpcharbor.mdanderson.edu/pchen6/pipdlenv:pchen6  # tag docker image as hpcharbor format
$ docker push hpcharbor.mdanderson.edu/pchen6/pipdlenv:pchen6                 # push image to hpcharbor

Step 4. Launch K8S Instance for DL Model Development

• Visit http://hpcexhaproxy.mdanderson.edu/jupyter
• Scroll to the bottom, and toggle on "Custom Environment", and set configurations
  • CPUs: default 32, range from 1 to 256
  • DGX Type: default A100, options are: A100 and H100
  • GPUs: default 1, range from 1 to 4
  • Image: no default, the format is hpcharbor.mdanderson.edu/<project_name>/<image_name>:<image_tag> (e.g., hpcharbor.mdanderson.edu/pchen6/pipdlenv:pchen6)
• Click the yellow "Start", and wait 1-2 minutes for environment initiation, retry if encountering Timeout