The Kubernetes community has delivered another milestone release with Kubernetes v1.33, codenamed “Octarine.” Octarine is a fictional color described in Terry Pratchett’s Discworld novels, called “the color of magic” — a mystical, invisible hue that only wizards and magical beings can see.
Kubernetes v.133 Octarine image

As AI and multi-cloud adoption accelerate, features like dynamic resource allocation, sidecars, NUMA scheduling, and automatic scaling without disruption mean that Kubernetes is handling things users don’t even see.

Let’s dive into the key changes, what they look like, and why they matter (or you can click on the feature below to jump to the relevant section).

FeatureStageWhy It Matters
Sidecar ContainersStableSimplifies lifecycle management for proxies/loggers; reduces operational complexity
In-place Resource Resize for PodsBetaEnables vertical scaling without restarts or downtime
Dynamic Resource Allocation (DRA)BetaImproves GPU and hardware resource scheduling for AI/ML workloads
NUMA-Aware SchedulingBetaBoosts performance by placing workloads near memory for AI/HPC
Asynchronous PreemptionBetaEnhances scheduler responsiveness under heavy AI/ML cluster loads
ClusterTrustBundlesBetaEases multi-cluster trust management and internal certificate handling
AdminNetworkPolicyBetaAllows platform-wide network enforcement independent of namespace controls
Pod Replacement PolicyAlphaReduces update disruption by controlling pod replacements based on conditions
Volume PopulatorsGASimplifies PersistentVolume setup with pre-loaded data sources
Topology-Aware Routing (PreferClose)GAOptimizes network traffic by routing within the same zone
.kuberc File for kubectlAlphaCustomizes kubectl defaults per user for easier CLI usage

Top 3 Highlights of Kubernetes 1.33

If we had to pick just three things to tell you about this release:

1. Sidecar Containers Are Now Stable

Sidecar containers — used for logging agents, proxies, and metrics collectors — are now officially supported and stable in Kubernetes. Sidecars now have well-defined behavior for restart policies, readiness probes, and OOM handling.

Until now, developers had to use workarounds to manage sidecar lifecycle with main containers. With sidecars stable, Kubernetes ensures predictable startup and shutdown behavior, reducing operational complexity for service meshes, observability, and custom networking.

Here’s an example of a sidecar configuration:

				
					apiVersion: v1
kind: Pod
metadata:
  name: my-app
spec:
  containers:
  - name: main-app
    image: my-app-image
  - name: sidecar-logger
    image: logger-image
    lifecycle:
      preStop:
        exec:
          command: ["/bin/cleanup-logs"]
				
			

2. In-place Resource Resize for Vertical Scaling of Pods (Beta)

Kubernetes 1.33 allows the Vertical Pod Autoscaler (VPA) to update CPU and memory resource allocations without restarting pods.

Traditionally, resizing required deleting and recreating pods, causing disruptions. In-place VPA means you can now dynamically rightsize resource usage for running applications without downtime — ideal for scaling real-world production workloads smoothly.

Example configuration:

				
					apiVersion: autoscaling.k8s.io/v1
kind: VerticalPodAutoscaler
metadata:
  name: my-vpa
spec:
  targetRef:
    apiVersion: "apps/v1"
    kind: Deployment
    name: my-deployment
  updatePolicy:
    updateMode: "Auto"
				
			

3. Dynamic Resource Allocation (DRA) Improvements (Beta)

The DRA API, critical for GPU and hardware accelerators, has been refined in v1.33. The new beta version simplifies how Kubernetes allocates special hardware resources like GPUs and high-speed network cards.

As AI/ML and HPC workloads become mainstream, dynamically allocating GPUs is essential. DRA now makes it easier to schedule workloads that need hardware devices without manual intervention or waste.

Example resource claim:

				
					apiVersion: resource.k8s.io/v1beta1
kind: ResourceClaim
metadata:
  name: gpu-claim
spec:
  resourceClassName: nvidia-a100
				
			

Machine Learning & AI Updates

As AI and HPC workloads grow more complex, Kubernetes 1.33 brings key improvements to scheduling, device management, and resource efficiency, including:

Dynamic Resource Allocation (DRA) Beta Improvements

The DRA improvements in Kubernetes 1.33 bring simpler APIs, broader compatibility, and better support for AI/ML applications that rely on GPUs, FPGAs, and custom devices.

The update makes DRA easier to adopt for third-party schedulers and better integrates it with Kubernetes core features.

NUMA-Aware Scheduling (Beta)

Non-Uniform Memory Access (NUMA) aware scheduling enables Kubernetes to place workloads closer to their memory, improving performance for AI, ML, and high-performance computing (HPC) workloads.

Why It Matters:

NUMA-aware placement ensures lower latency and higher throughput for resource-intensive applications — critical for training large AI models efficiently.

Asynchronous Preemption (Beta)

New asynchronous preemption improves how the Kubernetes scheduler handles resource contention. Instead of blocking the entire scheduling cycle when evicting lower-priority pods, Kubernetes now preempts asynchronously, improving throughput under load.

Why It Matters:

In AI-heavy clusters with thousands of pods, faster preemption keeps the system responsive even during heavy job loads.

Security Updates

ClusterTrustBundles (Beta)

ClusterTrustBundles allow you to manage trusted X.509 certificate bundles within the cluster itself, rather than outside systems.

Why It Matters:

Makes multi-cluster trust management easier and more portable, especially for service meshes or hybrid cloud setups.

Admin Network Policy (Beta)

New AdminNetworkPolicy objects provide cluster-wide network controls independent of namespaces.

Why It Matters:

Strengthens Kubernetes network security by allowing platform administrators to enforce policies even when namespace-level NetworkPolicies are missing.

Example:

				
					apiVersion: networking.k8s.io/v1alpha1
kind: AdminNetworkPolicy
metadata:
  name: deny-egress
spec:
  priority: 100
  ingress:
    - action: Allow
				
			

Platform Simplification and Operations

Pod Replacement Policy (Alpha)

A new PodReplacementPolicy allows users to control whether pods are proactively replaced based on pod conditions, making updates less disruptive.

Volume Populators (GA)

Volume Populators allow users to pre-populate PersistentVolumes with content from custom sources — like datasets for ML models.

Why It Matters:

Simplifies preparing storage without manual snapshotting.

Topology-Aware Routing (GA)

PreferClose topology-aware routing is now stable, optimizing traffic routing to endpoints in the same zone whenever possible.

kubectl .kuberc File (Alpha)

kubectl now supports a .kuberc file that stores default arguments and config, making it easier for users to customize kubectl behavior across clusters.

Takeaways from Kubernetes 1.33

As enterprises increasingly run critical and AI-heavy workloads on Kubernetes, these improvements make clusters easier to scale, safer to operate, and better aligned with modern cloud-native applications.

As teams increasingly adopt Kubernetes, they face challenges in configuring, monitoring and optimizing clusters within complex containerized environments.

Most teams manage these complexities with a combination of manual monitoring, third-party tools, and basic metrics provided by native Kubernetes dashboards — requiring them to switch between different tools and analyze data from multiple sources.

nOps + Kubernetes are better together section image

nOps is a complete Kubernetes solution: Visibility, Management & Optimization

nOps is the only solution on the market offering end-to-end, comprehensive Kubernetes monitoring and optimization capabilities are all in one platform. It rightsizes containers to match usage, efficiently binpacks and places nodes, and once clusters are optimized, implements the optimal pricing strategy. At every level, you use less and pay less for what you use – all with complete EKS visibility.

Key features include:

  • Critical metrics & benchmarkingfor pricing optimization, utilization rates, waste optimization down to the pod, node or container level
  • Container Cost Allocation: nOps processes massive amounts of your data to automatically unify and allocate your Kubernetes costs in the context of all your other cloud spending.
  • Container Insights & Rightsizing. View your cost breakdown, number of clusters, and the utilization of your containers to quickly assess the scale of your clusters, where costs are coming from, and where the waste is.
  • Autoscaling Optimization: nOps continually reconfigures your preferred autoscaler (Cluster Autoscaler or Karpenter) to keep your workloads optimized at all times for minimal engineering effort.
  • Spot Savings: automatically run your workloads on the optimal blend of On-Demand, Savings Plans and Spot instances, with automated instance selection & real-time instance reconsideration

nOps was recently ranked #1 with five stars in G2’s cloud cost management category, and we optimize $2+ billion in cloud spend for our customers.

Join our customers using nOps to understand your cloud costs and leverage automation with complete confidence by booking a demo with one of our AWS experts.