Multi-Cluster / Long Term Storage
Kubecost leverages Thanos to enable durable storage for three different purposes:
To enable Thanos, follow these steps:
This step creates the object-store.yaml file that contains your durable storage target (e.g. GCS, S3, etc.) configuration and access credentials. The details of this file are documented thoroughly in Thanos documentation.
We have guides for using cloud-native storage for the largest cloud providers. Other providers can be similarly configured.
Use the appropriate guide for your cloud provider:
Create a secret with the .yaml file generated in the previous step:
kubectl create secret generic kubecost-thanos -n kubecost --from-file=./object-store.yaml
Each cluster needs to be labelled with a unique Cluster ID, which is done in two places.
The Thanos subchart includes
thanos-compact, and service discovery for
thanos-sidecar. These components are recommended when deploying Thanos on the primary cluster.
helm upgrade kubecost kubecost/cost-analyzer \
--namespace kubecost \
-f https://raw.githubusercontent.com/kubecost/cost-analyzer-helm-chart/master/cost-analyzer/values-thanos.yaml \
thanos-storecontainer is configured to request 2.5GB memory, this may be reduced for smaller deployments.
thanos-storeis only used on the primary Kubecost cluster.
To verify installation, check to see all Pods are in a READY state. View Pod logs for more detail and see common troubleshooting steps below.
Thanos sends data to the bucket every 2 hours. Once 2 hours have passed, logs should indicate if data has been sent successfully or not.
You can monitor the logs with:
kubectl logs --namespace kubecost -l app=prometheus -l component=server --prefix=true --container thanos-sidecar --tail=-1 | grep uploaded
Monitoring logs this way should return results like this:
[pod/kubecost-prometheus-server-xxx/thanos-sidecar] level=debug ts=2022-06-09T13:00:10.084904136Z caller=objstore.go:206 msg="uploaded file" from=/data/thanos/upload/BUCKETID/chunks/000001 dst=BUCKETID/chunks/000001 bucket="tracing: kc-thanos-store"
As an aside, you can validate the Prometheus metrics are all configured with correct cluster names with:
kubectl logs --namespace kubecost -l app=prometheus -l component=server --prefix=true --container thanos-sidecar --tail=-1 | grep external_labels
To troubleshoot the IAM Role Attached to the serviceaccount, you can create a Pod using the same service account used by the thanos-sidecar (default is
- image: amazon/aws-cli
command: ['sleep', '500']
kubectl apply -f s3-pod.yaml
kubectl exec -i -t s3-pod -- aws s3 ls s3://kc-thanos-store
This should return a list of objects (or at least not give a permission error).
If a cluster is not successfully writing data to the bucket, review
thanos-sidecarlogs with the following command:
kubectl logs kubecost-prometheus-server-<your-pod-id> -n kubecost -c thanos-sidecar
Logs in the following format are evidence of a successful bucket write:
level=debug ts=2019-12-20T20:38:32.288251067Z caller=objstore.go:91 msg="uploaded file" from=/data/thanos/upload/BUCKET-ID/meta.json dst=debug/metas/BUCKET-ID.json bucket=kc-thanos
If thanos-query can't connect to both the sidecar and the store, you may want to directly specify the store gRPC service address instead of using DNS discovery (the default). You can quickly test if this is the issue by running:
kubectl edit deployment kubecost-thanos-query -n kubecost
to the container args. This will cause a query restart and you can visit
/storesagain to see if the store has been added.
If it has, you'll want to use these addresses instead of DNS more permanently by setting .Values.thanos.query.stores in values-thanos.yaml.
- name: GOGC
# Maximum number of queries processed concurrently by query node.
# Maximum number of select requests made concurrently per a query.
- name: GOGC
A common error is as follows, which means you do not have the correct access to the supplied bucket:
Assuming Pods are running, use port forwarding to connect to the
kubectl port-forward svc/kubecost-thanos-query-http 8080:10902 --namespace kubecost
If you navigate to Stores using the top navigation bar, you should be able to see the status of both the
thanos-sidecarwhich accompanied the Prometheus server:
Also note that the sidecar should identify with the unique
cluster_idprovided in your values.yaml in the previous step. Default value is
The default retention period for when data is moved into the object storage is currently 2h - This configuration is based on Thanos suggested values. By default, it will be 2 hours before data is written to the provided bucket.
Instead of waiting 2h to ensure that thanos was configured correctly, the default log level for the thanos workloads is
debug(it's very light logging even on debug). You can get logs for the
thanos-sidecar, which is part of the
thanos-store. The logs should give you a clear indication of whether or not there was a problem consuming the secret and what the issue is. For more on Thanos architecture, view this resource.