https://kubefleet.dev/blog/2026/Recent content in 2026 Posts on KubeFleetHugoenhttps://kubefleet.dev/blog/2026/04/07/kubefleet-performance-and-scalability-report-q1-2026/Tue, 07 Apr 2026 00:00:00 +0000https://kubefleet.dev/blog/2026/04/07/kubefleet-performance-and-scalability-report-q1-2026/<h2 id="tldr">TL;DR</h2> <ul> <li>With proper confirmation, you can use KubeFleet to manage a large-scale multi-cluster enviornment environment with up to <strong>1,000 member clusters, 1,000 placements, and 100 concurrent progressive rollouts</strong>.</li> <li>To support deployments at such a scale, <ul> <li>on your KubeFleet hub cluster: <ul> <li>make sure that the API server and its etcd storage backend are configured to handle higher volume of requests. In this evaluation, we see that: <ul> <li>the API server may take 10+ cores of CPU and 30+ GB of memory when KubeFleet is busy processing a large number of placements and progressive rollouts concurrently;</li> <li>the KubeFleet API objects in total consume approximately 2 GB of space on the etcd storage backend.</li> </ul> </li> <li>make sure that the KubeFleet hub agent is allocated with ample CPU and memory resources: <ul> <li>the agent needs 8-12 cores and 16-24 GB of memory to run smoothly in a large-scale multi-cluster environment.</li> </ul> </li> </ul> </li> <li>the KubeFleet member agent can run reasonably well with a much smaller resource allocation (e.g., 1 core and 2 GB of memory).</li> </ul> </li> <li>KubeFleet can reliably process placements and progressive rollouts fast in a larger-scale environment: <ul> <li>Typically a placement that picks 100 member clusters in a fleet of 1,000 clusters using a label selector can be processed within 30 seconds.</li> <li>A 3-staged, non-gated progressive rollout with 50% in-stage concurrency can be completed within 2 minutes. Running 100 of such rollouts concurrently usually takes less than 4 minutes to complete.</li> </ul> </li> <li>As with any Kubernetes controllers, the KubeFleet agents, especially the hub agent, need to re-sync resources periodically or when they restart. In a larger-scale environment with many (1,000) placements, the re-processing should take less than 15 minutes to complete. During this period, you might experience some level of degraded responsiveness in the system.</li> <li>We are committed to continuously optimizing KubeFleet&rsquo;s performance and scalability; this report is the result of one of the many rounds of evaluations we plan to conduct as KubeFleet continues to evolve. The team aims to support larger-scale deployments better, with faster processing of placements and rollouts, and less resource consumption on both the API server end and the KubeFleet agent end. The team is currently working to revamp how KubeFleet handles heartbeat signals and cluster property collection for a smoother experience. Please reach out to us if you have any concerns or suggestions in the domain of KubeFleet performance and scalability.</li> </ul> <div class="alert alert-primary" role="alert"> <h4 class="alert-heading">A side note</h4> <p>Your experience with KubeFleet may vary due to a variety of factors, such as:</p>