Modern Kubernetes teams do a lot of “right” things, including auto‑scaling, rightsizing, golden images, and sensible defaults. And yet the cloud bill doesn’t budge. Why? Because reclaiming CPU and memory within pods doesn’t always collapse nodes. Savings are “soft” unless the cluster actually sheds capacity.

Sedai’s newest Kubernetes capabilities close that gap. We surface granular details about which cluster components are the biggest offenders on your bill, and we compact clusters to free up and eliminate wasted nodes. With Kubernetes v1.33’s in‑place pod resizing, we apply resource changes instantly — no restarts, no reschedules. And, with more granular cost attribution, you can tie more of your bill back to the workloads that drive it.

This blog will take you through some of the latest Kubernetes optimization features in Sedai.

What’s new

Cluster Compaction: Rightsizing that actually reduces nodes

After right-sizing pods, clusters often look like Swiss cheese: pockets of free capacity scattered across nodes. Traditional autoscalers focus on unused nodes; they rarely repack partially used ones. Sedai’s Cluster Compaction actively re-packs workloads to free whole nodes, then (with optional cloud credentials) removes those idle nodes.

• How it works: Works broadly via Kubernetes APIs; final node deletion uses cloud-specific APIs (optional permission).
• Plays well with autoscalers: Think of autoscaling as capacity sizing and cluster compaction as capacity defragmentation. They’re better together.

The result? Turns “soft savings” into hard savings by eliminating nodes once workloads are consolidated.

In‑Place Pod Resizing (K8s v1.33+): Instant, disruption‑free tuning

Kubernetes 1.33 introduces the ability to change CPU/memory without restarting a pod. Sedai now uses this to apply rightsizing in place.

Requirements: Cluster running Kubernetes v1.33+; Sedai detects and uses the capability automatically.

The result? Faster optimization cycles, no downtime, and higher confidence to apply frequent, incremental tuning in production.

Cost Attribution beyond compute: GPU, network, and storage

Kubernetes cost isn’t just CPU and RAM. With Sedai, teams can now see more of the bill tied back to the workloads that drive it, including:

• GPU: Attribute GPU usage, including fractional/time‑sliced scenarios, to consuming workloads. 
• Storage: Tie EBS/EFS/etc. costs to pods/services that mount volumes.
• Network: Distinguish internal vs. egress traffic and attribute costs correctly.

The result? Better, more granular visibility into your Kubernetes spend.

FAQs

1. Does Sedai replace my autoscaler?
No. Autoscalers add/remove capacity based on load. Sedai tunes autoscaler settings to make them work better, and complements autoscalers by defragmenting partially used nodes after rightsizing so entire nodes can be removed.

2. Do I need to provide extra permissions to use any of these features?
Only for automated node deletion during Cluster Compaction. Everything else runs with standard Kubernetes access.

3. Does Sedai optimize GPUs?
We attribute GPU costs now (including fractional/time‑sliced scenarios). Full GPU optimization is on our roadmap.

4. What’s required for in‑place resizing?
A cluster on Kubernetes v1.33+. Sedai automatically uses the capability; no special setup beyond upgrading.

‍

Get Started

Ready to turn rightsizing into real reductions, and prove it on your bill? Let us show you how.

Book a demo