Skip to content

The only tool you need to know what is happening and how to fix it.

2.2 KB
Raw Blame

Traceway for Kubernetes

Ready-to-apply OpenTelemetry Collector manifests that report a whole cluster to a Traceway project: host metrics per node, pod resource usage, pod logs, cluster state, and Kubernetes events.

Full walkthrough: docs.tracewayapp.com/learn/kubernetes

Files

File What it deploys
traceway-kubernetes.yaml traceway-node-agent (DaemonSet) and traceway-cluster-agent (Deployment), plus namespace, ServiceAccount and RBAC
traceway-gateway.yaml Optional in-cluster OTLP endpoint your applications export to, so the project token lives in one Secret
helm/node-agent-values.yaml The same node agent as values for the upstream open-telemetry/opentelemetry-collector chart
helm/cluster-agent-values.yaml The same cluster agent as chart values

Install

kubectl create namespace traceway

kubectl create secret generic traceway-token \
  --namespace traceway \
  --from-literal=token='<your-project-token>'

# Set TRACEWAY_CLUSTER_NAME and TRACEWAY_ENDPOINT in the traceway-settings
# ConfigMap at the top of the file first.
kubectl apply -f traceway-kubernetes.yaml

The project token is the one from your Traceway project's Settings page. TRACEWAY_ENDPOINT is your instance's OTLP base URL, ending in /api/otel.

What you get

  • One instance row per node in the organization overview, grouped under TRACEWAY_CLUSTER_NAME, with CPU, memory, disk and network.
  • k8s.pod.* metrics for per-pod dashboards.
  • Container stdout/stderr on the Logs page.
  • Cluster state metrics and Kubernetes events.

Notes

  • TRACEWAY_CLUSTER_NAME becomes k8s.cluster.name. Use a distinct value per cluster; that string is what the overview groups by.
  • The node agent sets service.name to the node name, which becomes the server_name tag. That is what gives one instance row per node.
  • The three system.*.utilization metrics are off by default in the hostmetrics receiver and are enabled explicitly in these configs. Without them the organization overview has no CPU, memory or disk to show.
  • Keep traceway-cluster-agent at one replica.
  • The collector image is pinned to a released version. Bump it deliberately.