Kubecost uses a shared storage bucket to store metrics from clusters, known as durable storage, in order to provide a single-pane-of-glass for viewing cost across many clusters. Multi-cluster is an enterprise feature of Kubecost.
There are multiple methods to provide Kubecost access to an S3 bucket. This guide has two examples:
Using a Kubernetes secret
Attaching an AWS Identity and Access Management (IAM) role to the service account used by Prometheus
Both methods require an S3 bucket. Our example bucket is named kc-thanos-store.
This is a simple S3 bucket with all public access blocked. No other bucket configuration changes should be required.
Once created, add an IAM policy to access this bucket. See our AWS Thanos IAM Policy doc for instructions.
To use the Kubernetes secret method for allowing access, create a YAML file named object-store.yaml with contents similar to the following example. See region to endpoint mappings here.
Instead of using a secret key in a file, many will want to use this method.
Attach the policy to the Thanos pods service accounts. Your object-store.yaml should follow the format below when using this option, which does not contain the secret_key and access_key fields.
Then, follow this AWS guide to enable attaching IAM roles to pods.
You can define the IAM role to associate with a service account in your cluster by creating a service account in the same namespace as Kubecost and adding an annotation to it of the form eks.amazonaws.com/role-arn: arn:aws:iam::<AWS_ACCOUNT_ID>:role/<IAM_ROLE_NAME> as described here.
Once that annotation has been created, configure the following:
You can encrypt the S3 bucket where Kubecost data is stored in AWS via S3 and KMS. However, because Thanos can store potentially millions of objects, it is suggested that you use bucket-level encryption instead of object-level encryption. More details available in these external docs:
Visit the Configuring Thanos doc for troubleshooting help.