Deploying ElasticSearch on Kubernetes

I'm a huge fan of Kubernetes. I'm also a huge fan of ElasticSearch. So I was a bit let down when the process of deploying wasn't very seamless. In this post I'll show you an easy and scalable Kubernetes config that works perfectly.

Prerequisites

1. An up and running kubernetes cluster (see my guide here).
2. Kubectl installed (docs)

The easy stuff

The first and easiest step is adding ResourceDefinitions to your cluster. This can be done with an awesome oneliner:

kubectl --kubeconfig KUBECONFIG_LOCATION apply -f https://download.elastic.co/downloads/eck/1.2.1/all-in-one.yaml

The copy paste stuff

This is the config for running the ElasticSearch cluster on your Kube Cluster.

# This sample sets up an Elasticsearch cluster with 2 nodes.
apiVersion: elasticsearch.k8s.elastic.co/v1
kind: Elasticsearch
metadata:
  name: production-elasticsearch
spec:
  version: 7.7.1
  nodeSets:
    - name: default
      config:
        xpack.security.authc:
          anonymous:
            username: anonymous
            roles: superuser
            authz_exception: false
        # most Elasticsearch configuration parameters are possible to set, e.g: node.attr.attr_name: attr_value
        node.master: true
        node.data: true
        node.ingest: true
        node.store.allow_mmap: false
      podTemplate:
        metadata:
          labels:
            # additional labels for pods
            app: APP_NAME
            tier: elasticsearch
        spec:
          containers:
            - name: elasticsearch
              # specify resource limits and requests
              resources:
                limits:
                  memory: 0.7G
                  cpu: 0.2
                requests:
                  memory: 0.7G
                  cpu: 0.2
              env:
                - name: ES_JAVA_OPTS
                  value: "-Xms325m -Xmx325m"
      count: 2
  http:
    tls:
      selfSignedCertificate:
        disabled: true

Two important things to note:

1. env:
       - name: ES_JAVA_OPTS
         value: "-Xms325m -Xmx325m" 

These values reference two things: Xms (Minimum Heap Size) and Xmx (Maximum Heap Size). The heap size refers to the JVM Heap Size, and by default ElasticSearch sets it to 1Gb. They should be the same size and when moving to production you need to set it to no more than 50% or your instance's RAM.

2. http:
      tls:
        selfSignedCertificate:
          disabled: true 

This is set to false, because my ElasticSearch cluster is running on a closed system (VPN) on kubernetes. Meaning than it's not reachable from the internet. This makes https useless, so I prefer the simplicity of http.

Save this config as deploy-elastic.yaml and run it using:

kubectl --kubeconfig KUBECONFIG_LOCATION apply -f deploy-elastic.yaml

Connecting to Elastic

After the ElasticSearch cluster finished deploying you can easily connect to it with the following URL:

http://production-elasticsearch-es-http.default.svc.cluster.local:9200

Do note that this URL is only reachable when you are inside the kubernetes cluster. If you want to publicly expose your ElasticSearch cluster I highly recommend a different setup! One that uses HTTPS and has users setup.

You can reach it using an application that's already running in your cluster, or using a VPN.

Scaling up

This config is not connected to an autoscaling group. If you want to scale your cluster to run more than two instances, you can simple change

count: 2

to a different integer.

Some potentially problematic stuff

Kubernetes is mostly designed to run stateless containers. So running something like a database isn't really recommended. Some user do use ElasticSearch as a database, which is fine most of the time, especially when you run it on a dedicated VPS. You should truly not do this when running ElasticSearch on Kubernetes. Treat your ElasticSearch cluster as an ephemeral index.

You should always be able to reconstruct the search index when necessary.

For the people in the back: DO NOT USE ELASTICSEARCH AS A DATABASE.

Wrapping up

So with that out of the way, I hope you found this useful. I'm aware that there's also a very well documented helm chart available for ElasticSearch. Do you want me to make a post on that? Let me know!