Atomix Helm Charts

This project provides Helm charts for Atomix.

• Atomix
• Atomix Operator

Atomix

The Atomix chart is implemented primarily as a StatefulSet with a headless service for persistent pod identifiers. The chart supports:

• Any number of replicas
• Configurable pod disruption budget
• Optional host-based anti-affinity policy
• Configurable storage class
• Arbitrary Atomix configuration in HOCON or JSON format
• Upgrades through the RollingUpdate strategy

Installation

The Atomix chart is not currently hosted anywhere. To install the chart, clone the repository and cd into the atomix-k8s directory:

git clone --branch master https://github.com/atomix/atomix-helm.git && cd atomix-helm

To install the chart, run helm install:

helm install atomix

To install the chart on minikube, disable anti-affinity:

helm install --set podAntiAffinity.enabled=false atomix

Configuration

Atomix cluster

The Atomix configuration is a set of root values used to override the configuration for the Atomix pods themselves:

• replicas (default 3)
• logLevel (default INFO)
• heap (default 2G)
• config

The config field sets the base node configuration for all Atomix nodes. The configuration format is HOCON and should exclude the cluster configuration, which is managed by the chart itself. By default, the following configuration is used:

managementGroup {
  type: raft
  name: system
  partitionSize: 3
  partitions: 1
  members: ${atomix.members}
}
partitionGroups.raft {
  type: raft
  name: raft
  partitionSize: 3
  partitions: ${atomix.replicas}
  members: ${atomix.members}
}

To override the configuration, create a new yaml file and provide it to Helm when installing the chart, e.g.:

atomix.yaml

config: |
  # Raft management group for strongly consistent primary elections
  managementGroup {
    type: raft
    partitionSize: 3
    partitions: 1
    members: ${atomix.members}
  }

  # Raft partition group for storing consistent primitives
  partititonGroups.consensus {
    type: raft
    partitionSize: 3
    partitions: ${atomix.replicas}
    members: ${atomix.members}
    storage.segmentSize: 32MB
    storage.maxEntrySize: 1KB
  }

  # Primary-backup data group for in-memory replication backed by Raft elections
  partitionGroups.data {
    type: primary-backup
    partitions: 31
  }

  # Define a default protocol
  defaultProtocol {
    type: multi-raft
    group: consensus
    readConsistency: SEQUENTIAL
    recoveryStrategy: RECOVER
  }

  # Sets the default protocol for DistributedSet, DistributedList, DistributedQueue
  primitiveDefaults.set.protocol: ${defaultProtocol}
  primitiveDefaults.list.protocol: ${defaultProtocol}
  primitiveDefaults.queue.protocol: ${defaultProtocol}

  # Simple primitive configuration to use a multi-primary protocol and
  # enable caching for all `AtomicMap`s named `foo`
  primitives.foo {
    type: atomic-map
    cache.enabled: true
    protocol {
      type: multi-primary
      group: data
      backups: 2
      maxRetries: 2
    }
  }

Some useful variables are provided by the chart for use in HOCON files:

• ${atomix.replicas} - the configured number of replicas in the StatefulSet
• ${atomix.members} - the list of member IDs in the StatefulSet for use in Raft partition group configurations

When used in the atomix.config configuration, Atomix will replace the provided variables with the correct values for the running chart.

To install the chart with a custom configuration, use the --values option:

helm install --values=atomix.yaml atomix`

image

The image used by the chart is configurable via the image object:

• image.repository (default atomix/atomix)
• image.tag (default latest)
• image.pullPolicy (default IfNotPresent)
• image.pullSecrets (default [])

podAntiAffinity

The podAntiAffinity object sets the options for the StatefulSet's anti-affinity policy:

• podAntiAffinity.enabled (default true)

podDisruptionBudget

The podDisruptionBudget configures the number of nodes that can be taken offline by k8s:

• podDisruptionBudget.enabled (default true)
• podDisruptionBudget.maxUnavailable (default 1)
• podDisruptionBudget.minAvailable

persistence

The persistence object is used to override the persistence configuration for the chart. The available persistence options are:

• persistence.accessModes
• persistence.size (default 10Gi)
• persistence.annotations (default {})
• persistence.storageClass

It's recommended that local-storage be used for clusters where Raft partition groups are used. It's essential that Raft logs for a single pod be stored on a single host.

Upgrades

The Atomix chart supports upgrades through a simple RollingUpdate strategy. To upgrade Atomix, use the helm upgrade command, overriding the image.tag value:

helm install --set image.tag=3.0.6 atomix
helm upgrade --set image.tag=latest atomix

Atomix Operator

The atomix-operator chart provides templates for deploying the Atomix Operator to enable the deployment of custom AtomixCluster resources. The operator is able to manage the Atomix cluster to provide for much more involved cluster architectures, allowing partition groups to be added to and removed from the cluster, facilitating benchmarking, and providing chaos engineering features.

Installation

The Atomix Operator chart is not currently hosted anywhere. To install the chart, clone the repository and cd into the atomix-helm directory:

git clone --branch master https://github.com/atomix/atomix-helm.git && cd atomix-helm

To install the chart, run helm install:

helm install atomix-operator

Once the chart has been installed, you should be able to list AtomixCluster resources:

kubectl get atomixclusters

Configuration

The operator chart supports few configuration options since the majority of the configurable options are provided through custom resources provided by the operator itself.

image

The image used by the chart is configurable via the image object:

• image.repository (default atomix/atomix-operator)
• image.tag (default latest)
• image.pullPolicy (default IfNotPresent)
• image.pullSecrets (default [])

Acknowledgements

Atomix is developed as part of the ONOS project at the Open Networking Foundation.