K8ssandra Operator architecture

A deeper dive into the K8ssandra Operator architecture.

K8ssandra Operator provides several Custom Resource Definitions (CRDs), notably K8ssandraCluster. It is the primary custom resource (CR) with which you will with interact when using K8ssandra Operator. K8ssandraCluster is inherently complex given the number of software components than can be configured and deployed. This topic provides information about the K8ssandraCluster CR.

The following diagram provides an overview of the various objects that make up a K8ssandraCluster. Key parts are highlighted with arrows pointing to the relevant objects. The following sections of this document will explain the various parts of the diagram.

Architecture diagram

Kubernetes Clusters

The components that make up a K8ssandraCluster can be spread across multiple Kubernetes clusters. One of those clusters must be designated as the control plane cluster. The other clusters are data plane clusters. In the diagram we have three Kubernetes clusters - control-plane, east, and west. The latter two make up the data plane.

The K8ssandraCluster object must be created in the control plane cluster.

K8ssandra Operator is deployed in each cluster. The operator consists of several controllers. Some controllers, notably the K8ssandraCluster controller, only run in the control plane cluster. This is the main reason that the K8ssandraCluster object needs be created in the control plane cluster.

Other controllers run in both the control plane and in the data plane. This includes controllers for the following types:

  • CassandraDataCenter
  • Stargate
  • Reaper
  • MedusaBackupJob
  • MedusaRestoreJob

Cassandra Datacenters

The cassandra.datacenters property is an array. K8ssandra Operator creates a CassandraDatacenter for each element. Here is the relevant portion of the manifest from the diagram:

cassandra:
  datacenters:
    - metadata:
        name: dc1
      k8sContext: east
      size: 3
    - metadata:
        name: dc2
      k8sContext: west
      size: 3  

There are two elements in the array. K8ssnandra Operator will create CassandraDatacenters named dc1 and dc2. Each one will have a size of 3.

(Add link to configuration doc when it’s available.)

The k8sContext property specifies in which Kubernetes cluster the CassandraDatacenter should be created. K8sandra Operator creates dc1 in the east cluster and dc2 in the west cluster. If k8sContext is not specified, the operator creates the CassandraDatacenter in the control plane cluster. To understand what values can be specified for k8sContext we have to talk about ClientConfigs, which are covered in the next section.

Cass Operator provides the CassandraDatacenter CRD and manages CassandraDatacenter objects. After K8ssandra Operator (the K8ssandraCluster controller to be precise) creates the CassandraDatacenter, Cass Operator creates a StatefulSet.

Note: There can be multiple StatefulSets for a CassandraDatacenter. The example in the diagram above only includes one for simplicty.

After Cass Operator creates the StatefulSet, the StatefulSet controller (which is part of Kubernetes itself0 creates the Cassandra pods. The Cassandra pod consists of several containers. The cassandra container is the main one where the Cassandra process runs.

ClientConfig

A ClientConfig is another CRD provided by K8ssandra Operator; however, it is not part of a K8ssandraCluster. It is essentially a reference to secret that contains a kubeconfig. The kubeconfig should contain one or more context elements which provide details for connecting to a Kubernetes cluster. The k8sContext property is a reference to one of these contexts.

The operator uses ClientConfigs to establish connections to remote clusters. It is best to think of a ClientConfig as an operator configuration that is shared across all K8ssandraClusters.

Note: See [Remote Cluster Connection Management][https://github.com/k8ssandra/k8ssandra-operator/tree/main/docs/remote-k8s-access#readme] for more information about ClientConfigs.

Telemetry

Next we look at the telemetry property. Here is the relevant portion of the manifest in the diagram:

cassandra:
  datacenters:
    ...
  telemetry:
    prometheus:
      enabled:
        true

When telemetry is enabled, K8ssandra Operator creates a ServiceMonitor for each CassandraDatacenter. The ServiceMonitor CRD is provided by Prometheus Operator. You must install the CRD in order for K8ssandra Operator to be able to create Service Monitors. This can most easily be done by installing Prometheus Operator.

Note: See Monitoring using Prometheus for more information about using ServiceMonitors.

Stargate

K8ssandra Operator provides a Stargate CRD. Stargate objects are managed by the Stargate controller.

Here is the relevant portion of the manifest for the stargate property:

cassandra:
  ...
stargate:
  size: 1  

The presence of the stargate property tells the operator to enable and configure Stargate for each CassandraDatacenter. The size property configures the number of Stargate nodes per datacenter.

(TODO: Add link to configuration doc when it’s available.)

Note: See stargate.io to learn more abour Stargate.

After the K8ssandraCluster controller creates the Stargate object, the Stargate controller creates a Deployment. After the Deployment is created, the Deployment controller, which is part of Kubernetes itself, creates the Stargate pod. This pod runs the Stargate process.

Reaper

K8ssandra Operator provides a Reaper CRD. Reaper objects are managed by the Reaper controller.

Here is the relevant portion of the manifest for the reaper property:

cassandra:
  ...
stargate:
  ...
reaper: {}    

The presence of the reaper property tells the operator to enable and configure Repaer for each CassandraDatacenter. The empty object {} simply means that Reaper will be configured with default settings.

Note: See cassandra-reaper.io to learn more about Reaper.

(TODO: Add link to configuration doc when it’s available.)

After the K8ssadraCluster controller creates the Reaper object, the Reaper controller creates a Deployment. After the Deployment is created, the Deployment controller, which is part of Kubernetes itself, creates the Reaper pod. This pod runs the Reaper process.

Medusa

K8ssandra Operator provides MedusaBackupJob and MedusaRestoreJob CRDs for performing backup/restore operations with Medusa. MedusaBackupJob and MedusaRestoreJob objects are managed by their own, respective controllers.

Here is the relevant portion of the manifest for the medusa property:

cassandra:
  ...
stargate:
  ...
reaper: {}
medusa:
  storageProperties:
    storageProvider: local
  ...

Note: See https://github.com/thelastpickle/cassandra-medusa to learn more about Medusa.

The medusa property is different from the telemetry, stargate, and reaper properties in that it does not result in the operator creating an additional object. Instead K8ssandra Operator configures the CassandraDatacenter to enable and deploy Medusa.

The Cassandra pod has a medusa-restore init container and a medusa sidecar container. The former performs restores and the latter performs backups.

Note: See Deploying Medusa for more information on configuring and performing backups and restores.

ReplicatedSecret

K8ssandra Operator provides a ReplicatedSecret CRD. ReplicatedSecrets are managed by the SecretSync controller. Unlike other properties, there is no property in the K8ssandraCluster spec to enable and configure a ReplicatedSecret.

The operator automatically creates a ReplicatedSecret for each K8ssandraCluster. This object determines which secrets in the control plane cluster are replicated to the data plane clusters. The best example of this is the default superuser secret for Cassandra.

There is a superuser secret for each K8ssandraCluster. It can be created and provided by the user; otherwise, the operator generates a default one. The SecretSync controller ensures that the secret is replicated to each of the data plane clusters.

(TODO: Add link to secrets management doc when it’s available.)


Last modified September 18, 2024: Release v1.20.0 (9dc420a)