01-deployment_notes.md

Kubernetes Cluster Setup

Started from https://zero-to-jupyterhub.readthedocs.io/en/latest/amazon/step-zero-aws-eks.html

Notes from each step:

Step 1: EKS Service Role

• did not see a way to add AmazonEC2ContainerRegistryReadOnly, assuming it is unnecessary or a part of the other 2 policies
• Note the https://docs.aws.amazon.com/eks/latest/userguide/getting-started.html instructions are a little bit differnt, but it seems like the outcome is the same.

Step 2: EKS Cluster

• Originally, did this manually through the console, but assume something went wrong because I couldn't get note status (or was I impatient?)
• Then used https://docs.aws.amazon.com/eks/latest/userguide/getting-started.html instructions, which use an Amazon S3 template URL for cloudformation.

Steps 3-5 nothing to add

Step 6 This was a bit confusing because the zero-to-jupyterhub guide makes me think I need to update ~/.kube/config but it looks like that file is generated completely by the aws cli correctly when you run:

aws eks --region {region} update-kubeconfig --name {cluster_name}

per AWS instructions

Steps 7-10 Nothing to add

Deploying Helm

Now using https://docs.dask.org/en/latest/setup/kubernetes.html

helm repo update 
helm install stable/dask
helm upgrade --force --recreate-pods --devel singed-sabertooth stable/dask -f dask_config.yml

See dask_config.yml in this gist.

AWS EKS Cluster

• 3 m4.2xlarge ec2 instances

Other references

• MUR SST to Zarr Gist
• MUR SST OpenDAP
• MUR SST HTTPS
• On using opendap for MUR SST: https://climate-cms.org/2019/01/18/using-opendap.html
• dask_kubernetes
• Spin up AWS Kubernetes cluster for workshop
• Setting up ACCESS Pangeo on AWS #26
• AWS Deployment

Other notes (can ignore)

• When runninng ./3_deploy_helm.sh got the error:

  Error: priorityclasses.scheduling.k8s.io "pangeohub-default-priority" is forbidden: User "system:serviceaccount:kube-system:default" cannot delete priorityclasses.scheduling.k8s.io at the cluster scope

  This might be because of the instruction https://zero-to-jupyterhub.readthedocs.io/en/latest/amazon/step-zero-aws-eks.html to use

  kubectl create clusterrolebinding cluster-system-anonymous --clusterrole=cluster-admin --user=system:anonymous

  I googled the error and ran these commands and it seems to have resolved itself. I can now run ./3_deploy_helm.sh without error.

• When running kubectl commands, always need to add --namespace pangeo

• Using kubectl logs is critical, worker logs will not show up in jupyter

02-dask_config.yml

jupyter:
  name: jupyter
  env:
    - name: EXTRA_CONDA_PACKAGES
      value: numpy zarr netCDF4 xarray s3fs -c conda-forge --yes
    - name: AWS_ACCESS_KEY_ID
      value: xxx
    - name: AWS_SECRET_ACCESS_KEY
      value: xxx

worker:
  name: worker
  env:
    - name: EXTRA_CONDA_PACKAGES
      value: numpy zarr netCDF4 xarray s3fs -c conda-forge --yes
    - name: AWS_ACCESS_KEY_ID
      value: xxx
    - name: AWS_SECRET_ACCESS_KEY
      value: xxx

scheduler:
  name: scheduler
  env:
    - name: EXTRA_CONDA_PACKAGES
      value: numpy zarr netCDF4 xarray s3fs -c conda-forge --yes
    - name: AWS_ACCESS_KEY_ID
      value: xxx
    - name: AWS_SECRET_ACCESS_KEY
      value: xxx

03-mur-sst.ipynb