High Availability Kubecost

Running Kubecost in high availability (HA) mode is a feature that relies on multiple Kubecost replica pods implementing the ETL Bucket Backup feature combined with a Leader/Follower implementation which ensures that there always exists exactly one leader across all replicas.

Leader + Follower

The Leader/Follower implementation leverages a coordination.k8s.io/v1 Lease resource to manage the election of a leader when necessary. To control access of the backup from the ETL pipelines, a RWStorageController is implemented to ensure the following:

• Followers block on all backup reads, and poll bucket storage for any backup reads every 30 seconds.

• Followers no-op on any backup writes.

• Followers who receive Queries in a backup store will not stack on pending reads, preventing external queries from blocking.

• Followers promoted to Leader will drop all locks and receive write privileges.

• Leaders behave identically to a single Kubecost install.

Configuring high availability

In order to enable the leader/follower and HA features, the following must also be configured:

• Replicas are set to a value greater than 1

• ETL FileStore is Enabled (enabled by default)

• ETL Bucket Backup is configured

For example, using our Helm chart, the following is an acceptable configuration:

helm install kubecost kubecost/cost-analyzer --namespace kubecost \
	--set kubecostDeployment.leaderFollower.enabled=true \ 
	--set kubecostDeployment.replicas=5 \
	--set kubecostModel.etlBucketConfigSecret=kubecost-bucket-secret

This can also be done in the values.yaml file within the chart:

kubecostModel:
  image: "gcr.io/kubecost1/cost-model"
  imagePullPolicy: Always
  # ... 
  # ETL should be enabled with etlFileStoreEnabled: true 
  etl: true
  etlFileStoreEnabled: true 
  # ...
  # ETL Bucket Backup should be configured by passing the configuration secret name
  etlBucketConfigSecret: kubecost-bucket-secret

# Used for HA mode in Enterprise tier
kubecostDeployment:
  # Select a number of replicas of Kubecost pods to run 
  replicas: 5
  # Enable Leader/Follower Election 
  leaderFollower:
    enabled: true