graphistry-helm

Run Graphistry in Kubernetes using this live helm repository and supporting automation scripts & documentation

See CHANGELOG.md for version history. For contributing to this repository as a developer, see DEVELOP.md

Private docker image repositories

We recommend using a private repository to avoid rate-limiting and improve security:

• Setup a free DockerHub account and generate a token for use as a service account
• Notify Graphistry of the DockerHub account ID and get confirmation of read-access to the Graphistry Docker images
• Pick one of the instructions below

Azure Container Registry (ACR)

Option 1 (Recommended): Automatic - Azure Pipelines

• Fork this repository
• In Azure Pipelines, connect to your forked repository and load pipeline azure-pipelines.acr-mirror.yml
• In the Azure Pipelines UI, add pipeline variables as defined in the script
• Run the pipeline

Updates:

• Update the pipeline by pulling the latest changes of this repository into your fork
• Get new Graphistry versions by updating variable GRAPHISTRY_VERSION and rerunning the pipeline

Option 2: Manual

• Set up the Azure Container Registry
• Login to Azure: az login
• Run:

APP_BUILD_TAG=latest ACR_NAME=myacr DOCKERHUB_USERNAME=mydockerhubuser DOCKERHUB_TOKEN=mydockerhubtoken ./acr-bootstrap/import-image-into-acr-from-dockerhub.sh 

Kubernetes secrets

Azure

Create a Azure Container Registry Container principal ID by running the following command with your ACR information to create a kube secret with the ACR principal ID(This script assumes a default namespace, to change, edit the script):

ACR_NAME=myacr AZSUBSCRIPTION="my subscription name" SERVICE_PRINCIPAL_NAME=acrk8sprincipal CONTAINER_REGISTRY_NAME=myacrk8sregistry ./acr-bootstrap/make_acr_principal_and_create_secret.sh

Any other Kubernetes cluster (assumes a default namespace)

kubectl create secret docker-registry acr-secret \
--namespace default \
--docker-server=<CONTAINER_REGISTRY_NAME>.azurecr.io \
--docker-username=<Docker username> \
--docker-password=<Docker password> 

Add gpu daemonset to cluster

Note: Be sure to add the nvidia device plugin daemonset to the cluster before deployment.
kubectl create -f https://raw.githubusercontent.com/NVIDIA/k8s-device-plugin/master/nvidia-device-plugin.yml

once daemonset has been installed and started if successful will see nvidia.com/gpu in nodes capacity here
kubectl get nodes -ojson | jq .items[].status.capacity

install Nginx ingress controller

Note: Be sure to add the nginx ingress controller to the cluster before deployment.
helm upgrade --install ingress-nginx ingress-nginx --repo https://kubernetes.github.io/ingress-nginx --namespace ingress-nginx --create-namespace

install Longhorn NFS

Note: Be sure to add Longhorn to the cluster before deployment.
helm repo add longhorn https://charts.longhorn.io
helm repo update
kubectl create namespace longhorn-system \ kubectl apply -f https://raw.githubusercontent.com/longhorn/longhorn/master/deploy/prerequisite/longhorn-iscsi-installation.yaml -n longhorn-system
helm upgrade -i longhorn longhorn/longhorn --namespace longhorn-system

Setting the node selector and the acr container registry for deployment

Note: Be sure to change the azurecontainerregistry value in values.yaml to the name of your acr as well as setting the nodeSelector value to your preferred node to deploy the cluster onto.

kubectl get nodes

once you have a node selected, run the following command and find the hostname of the node to use with the nodeSelector value:

kubectl describe node <node name>

and then set the nodeSelector value to the hostname of the selected node along with your acr container registry name.:

helm upgrade -i my-graphistry-chart graphistry-helm/Graphistry-Helm-Chart \
 --set azurecontainerregistry.name=<container-registry-name>.azurecr.io \
 --set nodeSelector."kubernetes\\.io/hostname"=<node hostname> \ 
 --set domain = <FQDN or node external IP ex: example.com> \
 --set imagePullSecrets=<secrets_name>  (has to go last) 

Note: different labels can be used for the nodeSelector value, but some labels between the nodes may not be unique.

ReadTheDocs