Skip to main content

Clusters

What is a Cluster​

A cluster in OpsWorker represents a connected Kubernetes cluster that OpsWorker can investigate. Each cluster has an OpsWorker Kubernetes Agent installed and belongs to a workspace.

OpsWorker supports all major Kubernetes distributions:

  • Amazon EKS
  • Google GKE
  • Azure AKS
  • Self-hosted Kubernetes (kubeadm, k3s, Rancher, OpenShift)

Connection Model​

OpsWorker uses an outbound-only connection model. The Kubernetes Agent inside your cluster initiates all communication; OpsWorker never connects inbound to your cluster.

sequenceDiagram
    participant Agent as K8s Agent (your cluster)
    participant SQS as AWS SQS
    participant OW as OpsWorker Cloud

    Agent->>SQS: Poll for commands (outbound HTTPS)
    OW->>SQS: Send investigation commands
    Agent->>SQS: Return collected data (outbound HTTPS)
    SQS->>OW: Deliver results

This means:

  • No inbound ports need to be opened on your cluster.
  • No VPN or tunneling is required.
  • No kubeconfig or cluster credentials are stored in OpsWorker's cloud.
  • Communication is encrypted with TLS over standard HTTPS (port 443).

Cluster Authentication​

Each cluster is authenticated with a cluster token and an API key, both generated by the OpsWorker portal when you add a cluster. They are provided to the agent during Helm installation. The cluster token is the cluster's permanent identifier; it cannot be rotated. To replace it, remove and re-add the cluster.

Per-Cluster Integrations​

Integrations are configured at the cluster level. Each cluster can have different integrations enabled:

  • Alert sources: Prometheus AlertManager, Grafana Alerting, Datadog.
  • Metrics: Grafana MCP (for dashboard and metric queries during investigations and chat).
  • Source control: GitHub, GitLab (for change correlation and pull/merge request creation).

Cluster Status​

The portal displays the connection status for each cluster:

StatusMeaning
PendingCluster added but the agent has not yet connected
SuccessAgent is running and communicating with OpsWorker (shown as connected)
FailureThe agent reported a problem during onboarding
TimeoutThe agent did not connect within the expected window

A cluster can only be deleted from the portal while it is in the Pending state.

Next Steps​