In-Place K8s Workload Right-sizing

Unlike traditional pod resizing, in-place pod resizing allows adjusting containers' CPU and memory allocations without restarting the pod itself. With this update, automation seamlessly triggers a pod resize by updating the desired CPU and memory if the recommended requests or limits do not match the currently allocated resources.

"Traditional pod resizing requires restarting workloads, which may disrupt services and impact stability," said Eli Birger, CTO of PerfectScale by DoiT. "What we observe is that teams often prefer to stay over-provisioned, allocating more resources to avoid restarts and ensure their services run smoothly. This approach frequently leads to underutilized resources and inefficiencies, without strong guarantees of stability. In-place resizing helps engineers overcome this challenge by resizing pods without restarts, enabling teams to continuously optimize resources for any type of pods while maintaining stable, reliable clusters."

In your day-to-day operations In-Place delivers:

✅ Improved reliability
Reduce the operational risks of pod restarts when applying critical configuration changes, maintaining a stable and reliable K8s.

✅ Efficient resource utilization
Remove the friction between cost-efficiency and resilience and right-size pods safely, enabling quick optimization without service disruption.

✅ Operational productivity
Implement changes instantly and automatically by scaling services up or down based on the actual resource demand, without the overhead of restarts.

We’re excited to announce that PerfectScale now supports In-Place Kubernetes Pod Right-Sizing, making your optimization process even more flexible, predictable, and aligned with your business goals.

Let’s explore the full story on why it matters for your K8s workloads and how to take advantage of this update!

How to start leveraging the update

In order to start leveraging the new capabilities with PerfectScale, the only prerequisite is a cluster that supports in-place changes, which requires Kubernetes version 1.33 or later with the beta feature gate enabled.

💡No extra setup is required! PerfectScale automatically applies changes in place, delivering results right away.

For a deeper dive into how in-place changes are implemented, check out the official Kubernetes documentation.

In-place rightsizing with PerfectScale Automation

In-place pod resizing combined with PerfectScale Automation provides a complete solution for graceful and effective workload right-sizing.

Instead of struggling with manual effort to identify and eliminate inefficiencies and relying on risky, guesstimated decisions, PerfectScale automatically analyzes workloads and applies data-driven recommendations in real-time, ensuring resources are continuously optimized. This not only removes operational overhead for teams but also drives safe, ongoing cost reduction without compromising performance and stability.

‍

Maintaining pod resources consistency

In fact, the in-place pod resizing allows you to avoid pod recreation when updating resources, which significantly improves cluster stability and prevents disruptions due to restarts. As a result of in-place resizing, multiple pod replicas could end up with different sizes, creating inconsistency. This can occur due to predefined resource update policies, Kubelet delays, etc. As a result of the inconsistency, replicas of the same pod may compete for shared resources in an imbalanced way, leading to performance degradation, unpredictable service behavior, imbalanced load distribution, and, ultimately, service disruptions.

To address this challenge, PerfectScale takes a unique approach: when autonomously right-sizing workloads in-place, it ensures pod resources remain consistent. This guarantees consistency, preserves uniformity, and prevents performance degradation, enabling you to maximize the benefits of in-place resizing.

Leveraging In-Place with real-time control

In-place optimization reduces the risks associated with restarts, but it doesn’t remove the need for control.

By leveraging advanced PerfectScale safety mechanisms, such as a maintenance window, automation frequency, etc., you can ensure that changes will happen outside peak production hours while a policy-driven algorithm aligns the optimization with your business goals.

The combination of In-Place and PerfectScale allows you to mitigate the overhead associated with the risks of restarts, while giving you flexibility, governance, and real-time control over the entire workload sizing process to keep operations safe, minimize downtimes, and free your team to focus on strategic tasks.

Optimizing continuously with hybrid workload sizing

When a pod’s resource request exceeds the available node capacity, Kubernetes won’t apply in-place resizing. Instead, it keeps the pod running without the requested resources, which can impact application stability and lead to performance degradation. This also prevents your autoscaling solution, like Karpenter, from working effectively, leading to inefficiencies, wasted capacity, and risk of performance issues.

To prevent this and ensure workloads always have the resources they need, properly scheduled and running smoothly, PerfectScale’s advanced algorithm automatically switches to a traditional, mutating-hook approach whenever in-place resizing isn’t applicable. With strong alignment to your autoscaler configuration, such as Karpenter, PerfectScale’s hybrid scaling mechanism helps you continuously maintain cluster stability and maximize optimization outcomes.

Applying In-Place when horizontal scaling fails

There might be a case where you can’t scale your clusters horizontally. This can happen when the number of replicas is fixed due to application architecture, stateful workloads, operational constraints, etc.

In such situations, vertical scaling becomes the go-to approach, allowing pods to get the resources they need while respecting these limitations. While traditionally, vertical scaling comes with a restart concern, In-Place resizing eliminates it, allowing workloads to scale up and down without the risk of downtime. By automating workload right-sizing, PerfectScale enables you effortlessly provision your services with the resources they need, ensuring safe and efficient operations, all without any manual effort!

Bottom Line

In-place resizing is a powerful feature, but it also comes with risks if not handled properly. PerfectScale eliminates these risks, ensuring you stay continuously optimized without compromising stability and performance.

Ready to see it in action? Learn more in our documentation or book a technical session with our experts.

New to PerfectScale by DoiT? Get started today for free!