Stateful Elastic Search Workload on Kubernetes

This repository will deploy a stateful elastic search instance on kubernetes leveraging managed application capabilities.

Azure Kubernetes

Kubernetes Design

Enabled uptime SLA
System Node Pool (CriticalAddonsOnly)
User Node Pool (no state)
Tainted Elastic Node Pool (state)
Auto-scaling (Nodes not elastic)
CNI Networking
Ephemeral Disks for Node OS
Cluster Monitoring

Installation

Service Definition

Use the deploy to Azure button which creates a service definition, then deploy instances of the Elastic Stamp from the service definition.

Manual Deployment

The main.bicep deployment creates an AKS cluster with monitoring, policy and flux enabled.

Guide

1. clone the repo

git clone https://github.com/danielscholl/aks-elastic.git
cd aks-elastic

2. Deploy the infrastructure to your azure subscription

Optional - Look for and ensure VM Support for the desired VM Size in the Region. az vm list-skus --location $AZURE_LOCATION --size Standard_D --all --output table |grep none

RESOURCE_GROUP='elastic-playground'
AZURE_LOCATION='eastus2'

az group create --name $RESOURCE_GROUP --location $AZURE_LOCATION
DEPLOYMENT=$(az deployment group create -g $RESOURCE_GROUP -f main.bicep -o json)
CLUSTER_NAME=$(echo $DEPLOYMENT | jq -r '.properties.outputs.aksClusterName.value')
az aks get-credentials -n $CLUSTER_NAME -g $RESOURCE_GROUP --overwrite-existing


# Validate Nodes availability over Zones
kubectl get nodes
kubectl describe nodes -l agentpool=npsystem | grep -i topology.kubernetes.io/zone
kubectl get nodes -l purpose=elastic
kubectl describe nodes -l purpose=elastic | grep -i topology.kubernetes.io/zone
kubectl describe nodes -l purpose=elastic | grep -i agentpool

kubectl -n nginx port-forward svc/nginx-ingress-controller 8080:80 &
curl -H "Host: podinfo.staging" http://localhost:8080
{
  "hostname": "podinfo-59489db7b5-lmwpn",
  "version": "6.2.3"
}

Elastice Search

Elastic Search Design

ElasticSearch has 3 main components that make up the cluster Client/Coordinating, Masters, and Data Nodes. you can read more about what each one does in elastic public docs.

Client/Coordinating Nodes Act as a reverse proxy for the clusters, this is what the external world interacts with. its deployed as a k8s deployment with horizontal pod autoscaling enabled, we will try to have a client in each node to minimize data movement across nodes, we can minimize but we can't prevent it from happening.
Master Nodes stores the metadata about the data nodes, it will be deployment as a k8s deployment, ideally we need 3.
Data Nodes this is where the magic is, this is where the indices are stored and replicated. this would be our Statefulset with persistent volume to persist the data.

Here is how the cluster will look like

danielscholl / aks-elastic