Install on Google Cloud Platform¶

This guide runs through how to set up and install Seldon Core in a Kubernetes cluster running on Google Cloud. By the end, you’ll have Seldon Core up and running and be ready to start deploying machine learning models.

Prerequisites¶

Google Cloud CLI¶

You will need the Google Cloud SDK in order to retrieve your cluster authentication credentials. It can also be used to create clusters and other resources for you:

Google Kubernetes Engine (GKE) Cluster¶

If you haven’t already created a Kubernetes cluster on GKE, you can follow this quickstart guide to get set up with your first cluster. Ensure you create 3 nodes (--num-nodes passed to gcloud container clusters create) to be able to run a minimal example:

Warning

Your GKE cluster needs to be a Standard type cluster. GKE Autopilot clusters currently don’t support installation of ingress controllers.

Kubectl¶

kubectl is the Kubernetes command-line tool. It allows you to run commands against Kubernetes clusters, which we’ll need to do as part of setting up Seldon Core.

Helm¶

Helm is a package manager that makes it easy to find, share and use software built for Kubernetes. If you don’t already have Helm installed locally, you can install it here:

Connect to Your Cluster¶

You can connect to your cluster by running the following gcloud command:

gcloud container clusters get-credentials CLUSTER_NAME

This will configure kubectl to use your google cloud kubernetes cluster. Don’t forget to replace CLUSTER_NAME with whatever you called your cluster when you created it. If you’ve forgotten, you can run gcloud container clusters list.

Note

If you get authentication errors while running the command above, try running gcloud auth login to check you are correctly logged in.

Install Cluster Ingress¶

Ingress is a Kubernetes object that provides routing rules for your cluster. It manages the incomming traffic and routes it to the services running inside the cluster.

Seldon Core supports using either Istio or Ambassador to manage incomming traffic. Seldon Core automatically creates the objects and rules required to route traffic to your deployed machine learning models.

Istio is an open source service mesh. If the term service mesh is unfamiliar to you, it’s worth reading a little more about Istio.

Download Istio

For Linux and macOS, the easiest way to download Istio is using the following command:

curl -L https://istio.io/downloadIstio | sh -

Move to the Istio package directory. For example, if the package is istio-1.11.4:

cd istio-1.11.4

Add the istioctl client to your path (Linux or macOS):

export PATH=$PWD/bin:$PATH

Install Istio

Istio provides a command line tool istioctl to make the installation process easy. The demo configuration profile has a good set of defaults that will work on your local cluster.

istioctl install --set profile=demo -y

The namespace label istio-injection=enabled instructs Istio to automatically inject proxies alongside anything we deploy in that namespace. We’ll set it up for our default namespace:

kubectl label namespace default istio-injection=enabled

Create Istio Gateway

In order for Seldon Core to use Istio’s features to manage cluster traffic, we need to create an Istio Gateway by running the following command:

Warning

You will need to copy the entire command from the code block below

kubectl apply -f - << END
apiVersion: networking.istio.io/v1alpha3
kind: Gateway
metadata:
name: seldon-gateway
namespace: istio-system
spec:
selector:
    istio: ingressgateway # use istio default controller
servers:
- port:
    number: 80
    name: http
    protocol: HTTP
    hosts:
    - "*"
END

For custom configuration and more details on installing seldon core with Istio please see the Istio Ingress page.

Ambassador is a Kubernetes ingress controller and API gateway. It routes incomming traffic to the underlying kubernetes workloads through configuration. Install Ambassador following their docs.

Install Seldon Core¶

Before we install Seldon Core, we’ll create a new namespace seldon-system for the operator to run in:

kubectl create namespace seldon-system

We’re now ready to install Seldon Core in our cluster. Run the following command for your choice of Ingress:

helm install seldon-core seldon-core-operator \
    --repo https://storage.googleapis.com/seldon-charts \
    --set usageMetrics.enabled=true \
    --set istio.enabled=true \
    --namespace seldon-system
helm install seldon-core seldon-core-operator \
    --repo https://storage.googleapis.com/seldon-charts \
    --set usageMetrics.enabled=true \
    --set ambassador.enabled=true \
    --namespace seldon-system

You can check that your Seldon Controller is running by doing:

kubectl get pods -n seldon-system

You should see a seldon-controller-manager pod with STATUS=Running.

Accessing your models¶

Congratulations! Seldon Core is now fully installed and operational. Before you move on to deploying models, make a note of your cluster IP and port:

export INGRESS_HOST=$(kubectl -n istio-system get service istio-ingressgateway -o jsonpath='{.status.loadBalancer.ingress[0].ip}')
export INGRESS_PORT=$(kubectl -n istio-system get service istio-ingressgateway -o jsonpath='{.spec.ports[?(@.name=="http2")].port}')
export INGRESS_URL=$INGRESS_HOST:$INGRESS_PORT
echo $INGRESS_URL

This is the public address you will use to access models running in your cluster.

export INGRESS_HOST=$(kubectl -n ambassador get service ambassador -o jsonpath='{.status.loadBalancer.ingress[0].ip}')
export INGRESS_PORT=$(kubectl -n ambassador get service ambassador -o jsonpath='{.spec.ports[?(@.name=="http")].port}')
export INGRESS_URL=$INGRESS_HOST:$INGRESS_PORT
echo $INGRESS_URL

This is the public address you will use to access models running in your cluster.

You are now ready to deploy models to your cluster.