Kubernetes 1.33 Preview Now Available on Leafcloud
We've added preview support for Kubernetes 1.33 to our Gardener-managed platform. Teams can now test the latest upstream release while maintaining European data sovereignty and running on climate-positive infrastructure.
By
Published on
Preview Support for Kubernetes 1.33
Preview support for Kubernetes 1.33 is now available across our Gardener-managed clusters. The upstream release landed in April 2025, and we’ve been validating it against our infrastructure over the past months.
Technical Improvements in 1.33
The 1.33 release brings several practical improvements for production workloads. The enhanced scheduler includes better resource bin-packing for GPU workloads, which matters when you’re running multi-GPU training jobs or inference deployments on our A100 or H100 nodes. Pod scheduling decisions now account for topology awareness improvements, meaning the scheduler makes smarter choices about which nodes get which workloads based on actual hardware locality.
Job management has been refined with better handling of job completion and failure conditions. This is useful for batch processing workloads—think large-scale data processing or periodic ML training runs where you need reliable completion signals and proper cleanup. The StatefulSet updates include improved rolling update behavior, which reduces disruption when updating stateful applications like databases or message queues.
There’s also expanded support for dynamic resource allocation, which allows containers to request resources more flexibly during runtime. For GPU-heavy workloads, this means better utilization when containers need varying amounts of GPU memory or compute capacity across different phases of execution.
What We Validated
Our engineering team tested the 1.33 release against our existing infrastructure stack: Gardener control planes, OpenStack compute and networking, and the GPU Operator configurations we use for NVIDIA hardware provisioning. We verified that cluster upgrades work smoothly, that GPU scheduling behaves as expected, and that the new scheduler features don’t conflict with our existing node configurations.
Preview support means the release is ready for testing on non-production clusters. We’re collecting feedback from early adopters before we promote 1.33 to general availability for production workloads.
If you want to test Kubernetes 1.33 on our infrastructure, sign up at leaf.cloud or schedule a call to discuss your cluster setup.
Find out more about version 1.33 here
Related: