StackState core integration
The Kubernetes integration is used to create a near real-time synchronization of topology and associated internal services from a Kubernetes cluster to StackState. This StackPack allows monitoring of the following:
- Nodes, pods, containers and services
- Configmaps, secrets and volumes
- In StackState:
The Kubernetes integration collects topology data in a Kubernetes cluster as well as metrics and events. To achieve this, different types of StackState Agent are used:
StackState Cluster Agent is deployed as a Deployment. There is one instance for the entire Kubernetes cluster:
- Topology and events data for all resources in the cluster are retrieved from the Kubernetes API.
- Control plane metrics are retrieved from the Kubernetes API.
StackState Agent V2 is deployed as a DaemonSet with one instance on each node in the Kubernetes cluster:
- Host information is retrieved from the Kubernetes API.
- Container information is collected from the Docker daemon.
- Metrics are retrieved from kubelet running on the node.
By default, metrics are also retrieved from kube-state-metrics if that is deployed on the same node as the StackState Agent pod. This can cause issues on a large Kubernetes cluster. To avoid this, it is advisable to enable cluster checks so that metrics are gathered from kube-state-metrics by a dedicated StackState ClusterCheck Agent.
Deployed only when
clusterChecks.enabledis set to
values.yamlwhen the StackState Cluster Agent is deployed. When deployed, default is one instance per cluster. When enabled, cluster checks configured on the StackState Cluster Agent are run by one of the deployed StackState ClusterCheck Agent pods. This is useful to run checks that do not need to run on a specific node and monitor non-containerized workloads such as:
- Out-of-cluster datastores and endpoints (for example, RDS or CloudSQL).
- Load-balanced cluster services (for example, Kubernetes services).
The following prerequisites are required to install the Kubernetes StackPack and deploy the StackState Agent and Cluster Agent:
- A Kubernetes Cluster must be up and running.
- A recent version of Helm 3.
- A user with permissions to create privileged pods, ClusterRoles and ClusterRoleBindings:
- ClusterRole and ClusterRoleBinding are needed to grant StackState Agents permissions to access the Kubernetes API.
- StackState Agents need to run in a privileged pod to be able to gather information on network connections and host information.
Install the Kubernetes StackPack from the StackState UI StackPacks > Integrations screen. You will need to provide the following parameters:
- Kubernetes Cluster Name - A name to identify the cluster. This does not need to match the cluster name used in
kubeconfig, however, that is usually a good candidate for a unique name.
If the Agent StackPack is not already installed, this will be automatically installed together with the Kubernetes StackPack. This is required to work with the StackState Agent, which will need to be deployed on each node in the Kubernetes cluster.
For the Kubernetes integration to retrieve topology, events and metrics data, you will need to have the following installed on your Kubernetes cluster:
- A StackState Agent on each node in the cluster
- StackState Cluster Agent on one node
The StackState Agent, Cluster Agent and kube-state-metrics can be installed together using the Cluster Agent Helm Chart:
- 1.If you do not already have it, you will need to add the StackState helm repository to the local helm client:helm repo add stackstate https://helm.stackstate.iohelm repo update
- 2.Deploy the StackState Agent, Cluster Agent and kube-state-metrics with the helm command provided in the StackState UI after you have installed the StackPack. For large Kubernetes clusters, consider enabling cluster checks to run the kubernetes_state check in a StackState ClusterCheck Agent pod.
To check the status of the Kubernetes integration, check that the StackState Cluster Agent (
cluster-agent) pod and all of the StackState Agent (
cluster-agent-agent) pods have status ready.
❯ kubectl get deployment,daemonset --namespace stackstate
NAME READY UP-TO-DATE AVAILABLE AGE
deployment.apps/stackstate-cluster-agent 1/1 1 1 5h14m
NAME DESIRED CURRENT READY UP-TO-DATE AVAILABLE NODE SELECTOR AGE
daemonset.apps/stackstate-cluster-agent-agent 10 10 10 10 10 <none> 5h14m
To enable cluster checks and the cluster check Agent pods, create a
values.yamlfile to deploy the
cluster-agentHelm chart and add the following YAML segment:
The kubernetes_state check is responsible for gathering metrics from kube-state-metrics and sending them to StackState. It is configured on the StackState Cluster Agent and runs in the StackState Agent pod that is on the same node as the kube-state-metrics pod.
In a default deployment, the pod running the StackState Cluster Agent and every deployed StackState Agent need to be able to run the check. In a large Kubernetes cluster, this can consume a lot of memory as every pod must be configured with sufficient CPU and memory requests and limits. Since only one of those Agent pods will actually run the check, a lot of CPU and memory resources will be allocated, but will not be used.
To remedy that situation, the kubernetes_state check can be configured to run as a cluster check. The YAML segment below shows how to do that in the
values.yamlfile used to deploy the
# clusterChecks.enabled -- Enables the cluster checks functionality _and_ the clustercheck pods.
# agent.config.override -- Disables kubernetes_state check on regular agent pods.
- name: auto_conf.yaml
# clusterAgent.config.override -- Defines kubernetes_state check for clusterchecks agents. Auto-discovery
# with ad_identifiers does not work here. Use a specific URL instead.
- name: conf.yaml
- kube_state_url: http://YOUR_KUBE_STATE_METRICS_SERVICE_NAME:8080/metrics
The Kubernetes integration retrieves the following data:
The Kubernetes integration retrieves all events from the Kubernetes cluster. The table below shows which event category will be assigned to each event type in StackState:
The Kubernetes integration makes all metrics from the Kubernetes cluster available in StackState. Relevant metrics are automatically mapped to the associated components.
All retrieved metrics can be browsed or added to a component as a telemetry stream. Select the data source StackState Metrics and type
kubernetesin the Select box to get a full list of all available metrics.
Add a Kubernetes metrics stream to a component
The Kubernetes integration retrieves components and relations for the Kubernetes cluster.
The following Kubernetes topology data is available in StackState as components:
The following relations between components are retrieved:
- Container → Volume
- CronJob → Job
- DaemonSet → Pod
- Deployment → ReplicaSet
- Job → Pod
- Ingress → Service
- Namespace → CronJob, DaemonSet, Deployment, Job, ReplicaSet, Service, StatefulSet
- Node → Cluster relation
- Pod → ConfigMap, Container, Deployment, Node, Secret, Volume
- ReplicaSet → Pod
- Service → ExternalService, Pod
- StatefulSet → Pod
- Direct communication between processes
- Process → Process communication via Kubernetes service
- Process → Process communication via headless Kubernetes service
The Kubernetes integration does not retrieve any traces data.
All tags defined in Kubernetes will be retrieved and added to the associated components and relations in StackState.
The StackState Agent talks to the kubelet and kube-state-metrics API.
The StackState Agent and Cluster Agent connect to the Kubernetes API to retrieve cluster wide information and Kubernetes events. The following API endpoints used:
A number of actions are added to StackState when the Kubernetes StackPack is installed. They are available from the Actions section on the right of the screen when a Kubernetes component is selected or from the component context menu, displayed when you hover over a Kubernetes component in the Topology Perspective
Details of installed actions can be found in the StackState UI Settings > Actions > Component Actions screen.
When the Kubernetes integration is enabled, the following Kubernetes views are available in StackState for each cluster:
- Kubernetes - Applications -
- Kubernetes - Infrastructure -
- Kubernetes - Namespaces -
- Kubernetes - Workload Controllers -
The code for the StackState Agent Kubernetes check is open source and available on GitHub at:
To uninstall the Kubernetes StackPack, go to the StackState UI StackPacks > Integrations > Kubernetes screen and click UNINSTALL. All Kubernetes StackPack specific configuration will be removed from StackState.
To uninstall the StackState Cluster Agent and the StackState Agent from your Kubernetes cluster, run a Helm uninstall:
helm uninstall <release_name> --namespace <namespace>
# If you used the standard install command provided when you installed the StackPack
helm uninstall stackstate-cluster-agent --namespace stackstate
Kubernetes StackPack v3.9.1 (2021-04-02)
- Improvement: Enable auto grouping on generated views.
- Improvement: Update documentation.
- Improvement: Common bumped from 2.4.1 to 2.5.0
- Improvement: StackState min version bumped to 4.3.0
Kubernetes StackPack v3.8.0 (2021-03-08)
- Feature: Namespaces are now a component in StackState with a namespaces view for each cluster
- Feature: New component actions for quick navigation on workloads, pods and namespaces
- Feature: Added a "Pod Scheduled" metric stream to pods
- Feature: Secrets are now a component in StackState
- Improvement: The "Desired vs Ready" checks on workloads now use the "Ready replicas" stream instead of the replicas stream.
- Improvement: Use standard (blue) Kubernetes icons
- Bug fix: Fixed Kubernetes synchronization when a component had no labels but only tags
Kubernetes StackPack v3.7.2 (2020-08-18)
- Feature: Introduced the Release notes pop up for customer
Kubernetes StackPack v3.7.1 (2020-08-10)
- Feature: Introduced Kubernetes specific component identifiers
Kubernetes StackPack v3.7.0 (2020-08-04)
- Improvement: Deprecated stackpack specific layers and introduced a new common layer structure.
Kubernetes StackPack v3.6.0 (2020-06-19)
- Improvement: Set the stream priorities on all streams.