Overview

The unified cloud-K8s release introduces a significant improvement to Kubernetes storage cost allocation for AWS environments. The enhancement aligns storage cost attribution with real pod-level ownership of persistent volumes, replacing the previous node-level proportional allocation model. The result is a more accurate and meaningful breakdown of storage costs across workloads.

Background

Previously, storage costs for all EBS volumes attached to a node, including root volumes, were distributed across all hosted pods, weighted by pod uptime. This approach did not reflect real usage patterns, especially when a single pod exclusively owns and utilizes a PersistentVolumeClaim (PVC). We refined the storage allocation model to better reflect true ownership and reduce noise caused by irrelevant volume types.

Key Enhancements

1. Direct Volume-to-Pod Cost Allocation PVC-backed EBS volumes are now fully allocated to the specific pod that claims them, aligning with the expectation of single-ownership PVCs. This eliminates the previous node-wide cost spreading and provides precise visibility into pod storage costs.
2. Root Volume Handling Root EBS volumes (non-PVC volumes) continue to be shared across all pods on the node, as they are typically not meaningful for workload-level cost insights. Due to incomplete visibility of root-volume metrics in Prometheus, this behavior intentionally mirrors the existing implementation.
3. Simplification for Non-EBS Storage EFS and other storage types are considered negligible for this stage and are not included in allocation.
4. Improved Volume Mapping PVCs are mapped to their actual EBS Volume IDs using newly collected Prometheus metrics. Volumes without a PVC mapping are assumed to be node root volumes and handled with shared allocation.
5. Waste Allocation (Deferred) Waste allocation for storage (capacity vs. utilization) is not yet included.

Preconditions

⚙️

This enhancement applies only when:

• Running in AWS.
• Using prometheus-agent version 0.3.22, this is the Prometheus Agent version in use, which supports the newly required PVC-to-volume metrics; otherwise, volumes are shared between all pods evenly. (and not based on storage PVC)
• Using the Unified-K8s implementation.

If any of these conditions are not met, the system reverts to its previous allocation behavior.

Technical Notes

• Volume cost allocation now supports grouping by volume ownership, rather than node hosting.
• No changes were required to “allocatable” resource handling since storage waste is out of scope for this phase.
• No generic allocation refactor was needed, though future enhancements (e.g., data transfer attribution) may revisit this area.