Deploy ECK on GKE Autopilot

ECK

This page shows how to run ECK on GKE Autopilot.

1. It is recommended that each Kubernetes host’s virtual memory kernel settings be modified. Refer to Virtual memory.
2. It is recommended that Elasticsearch Pods have an initContainer that waits for virtual memory settings to be in place.
3. For Elastic Agent/Beats there are storage limitations to be considered.
4. Ensure you are using a node class that is applicable for your workload by adding a cloud.google.com/compute-class label in a nodeSelector. Refer to GKE Autopilot documentation.

If you are intending to run production workloads on GKE Autopilot then vm.max_map_count should be set. The recommended way to set this kernel setting on the Autopilot hosts depends on your ECK version:

• ECK Use a DaemonSet. You must be running at least version 1.25 when on the regular channel or using the rapid channel, which currently runs version 1.27.

  Warning

  Use the provided Daemonset exactly as specified, with a max_map_count value of 262144, or it could be rejected by the Autopilot control plane.

• ECK Use a custom ComputeClass. Using a custom ComputeClass allows you to set a higher value for max_map_count due to limitations on the DaemonSet.

Refer to Install ECK for more information on installation options.

Create an Elasticsearch cluster. The information that you need to provide in your spec depends on whether you've increased your virtual memory kernel setting, and the method that you used.

cat <<EOF | kubectl apply -f -
apiVersion: elasticsearch.k8s.elastic.co/v1
kind: Elasticsearch
metadata:
  name: elasticsearch-sample
spec:
  version: 9.2.3
  nodeSets:
  - name: default
    count: 1
    podTemplate:
      spec:
        nodeSelector:
          cloud.google.com/compute-class: "elasticsearch"
EOF
		

cat <<EOF | kubectl apply -f -
apiVersion: elasticsearch.k8s.elastic.co/v1
kind: Elasticsearch
metadata:
  name: elasticsearch-sample
spec:
  version: 9.2.3
  nodeSets:
  - name: default
    count: 1
    podTemplate:
      spec:
        # This init container ensures that the `max_map_count` setting has been applied before starting Elasticsearch.
        # This is not required, but is encouraged when using the Daemonset to set max_map_count.
        # Do not use this if setting config.node.store.allow_mmap: false
        initContainers:
        - name: max-map-count-check
          command: ['sh', '-c', "while true; do mmc=$(cat /proc/sys/vm/max_map_count); if [ ${mmc} -eq 262144 ]; then exit 0; fi; sleep 1; done"]
EOF
		

If you didn't increase your virtual memory, then you need to set node.store.allow_mmap to false.

cat <<EOF | kubectl apply -f -
apiVersion: elasticsearch.k8s.elastic.co/v1
kind: Elasticsearch
metadata:
  name: elasticsearch-sample
spec:
  version: 9.2.3
  nodeSets:
  - name: default
    count: 1
    config:
      node.store.allow_mmap: false
EOF
		

When running Elastic Agent and Beats within GKE Autopilot there are storage constraints to be considered. No HostPath volumes are allowed, which the ECK operator defaults to when unset for both Deployments and DaemonSets. Instead use Kubernetes ephemeral volumes.

Refer to Recipes to deploy Elasticsearch, Kibana, Elastic Fleet Server and Elastic Agent and/or Beats within GKE Autopilot.