Directions for installing Helm for your particular OS flavor can be found on the Helm GitHub page.

Once that is done, it will be necessary to update the main Helm chart repository. This main repository contains subcharts for LogScale.

We depend on the Confluent Helm Charts as dependencies, which are included automatically when running the installation below.

$ helm repo add humio https://humio.github.io/humio-helm-charts
$ helm repo update

Now create a values.yaml file. Adjust the version, resources and JVM memory as appropriate for the nodes on which the pods will be scheduled. The jvm.xmx and jvm.maxDirectMemorySize should each be half of the allocated memory.

humio-core:
  enabled: true

  # The number of LogScale pods
  replicas: 3

  # Use a custom version of LogScale.
  image: humio/humio-core:<version>

  # Custom partitions
  ingest:
    initialPartitionsPerNode: 4
  storage:
    initialPartitionsPerNode: 4

  # Custom CPU/Memory resources
  resources:
    limits:
    cpu: 30
    memory: 220Gi
    requests:
    cpu: 28
    memory: 220Gi

  # Custom JVM memory settings (these will depend on resources defined)
  jvm:
    xss: 2m
    xms: 4g
    xmx: 110g
    maxDirectMemorySize: 110g
    extraArgs: -XX:+UseParallelGC -XX:+UnlockDiagnosticVMOptions -XX:CompileCommand=dontinline,com/humio/util/HotspotUtilsJ.dontInline -Xlog:gc+jni=debug:stdout -Dakka.log-config-on-start=on -Xlog:gc*:stdout:time,tags

  affinity:
    # Affinity policy to prevent multiple LogScale pods per node (recommended)
    podAntiAffinity:
      requiredDuringSchedulingIgnoredDuringExecution:
      - labelSelector:
          matchExpressions:
          - key: app
            operator: In
            values:
            - humio-core
        topologyKey: "kubernetes.io/hostname"
global:
  sharedTokens:
    fluentbit: {kubernetes: in-cluster}

These settings will tell Helm to create a default three-node LogScale cluster with Kafka and ZooKeeper. It will also create a Fluent Bit daemonset that will collect logs from any pods running in the Kubernetes cluster, and autodiscover the LogScale endpoint and token. We recommend installing LogScale into its own namespace; in this example we're using the logging namespace

$ helm install humio humio/humio-helm-charts \
   --namespace logging \
   --values values.yaml

$ kubectl get service humio-humio-core-http -n logging -o go-template --template='http://{{(index .status.loadBalancer.ingress 0 ).ip}}:8080'

$ minikube service humio-humio-core-http -n logging --url

$ kubectl get secret developer-user-password -n logging -o=template --template={{.data.password}} | base64 -D

To update using a non-rolling strategy, you'll have to delete the LogScale pods temporarily and then bring them back with the new version.

To update using a rolling strategy, you can rely on the statefulset to deploy the changes.

If you decide to uninstall Helm Chart, you can do so by executing the following from the command-line:

$ helm delete --purge humio
$ kubectl delete namespace logging --cascade=true