Kubernetes
Learn more about Pelanor’s eBPF Kubernetes agent and how to install it on your K8s clusters.
Pelanor’s Kubernetes sensor delivers granular cost allocation for every cluster, namespace, workload, and even individual network endpoint.
Key Capabilities
- Cost by namespace & workload – CPU, memory, storage, and network
- Network cost attribution – cluster → namespace → workload → endpoint
- eBPF-powered traffic mapping
- DNS – map cloud resources (e.g., ALB, RDS) to workloads
- HTTP – link workloads to S3 buckets / object stores
- SQL – attribute database queries to the calling pods
Architecture Overview
A lightweight DaemonSet runs on each node. Using eBPF plus the Kubernetes API it gathers:
Data collected | Purpose |
---|---|
K8s object state (namespaces, deployments, pods, etc.) | Identity & ownership |
Pod/container CPU & memory requests & usage | Cost & right-sizing |
Per-pod network statistics | Internal vs external traffic |
DNS, HTTP, SQL traces | Resource → workload mapping |
The DaemonSet consumes ~50 MB memory and < 2% CPU per node. It captures only metadata, never payloads.
Installation
After signing in to Pelanor, go to Integrations → Kubernetes and copy the pre-filled Helm command.
Required values
Parameter | Notes |
---|---|
customerId | Permanent, unique to your Pelanor tenant. |
deployKey | Unique per Kubernetes integration. Create one per cluster for better status tracking, or reuse if desired. |
clusterName | Human-readable & unique. For EKS, the cluster ARN works well. |
clusterName
must be unique across all clusters in your tenant; duplicates will merge data.
Troubleshooting
Helm chart installed but some pods are not scheduled
Check for taints
Ensure pods aren’t unschedulable due to node taints (configure tolerations if needed).
Verify node resources
Confirm that all nodes have enough available CPU and memory to schedule the Pelanor pods.