Kubernetes 1.35 Released with In-Place Pod Resize and AI-Optimized Scheduling

Kubernetes 1.35: Mutability and AI/ML Optimization

The Cloud Native Computing Foundation (CNCF) recently released Kubernetes 1.35, “Timbernetes,” focusing on resource mutability and improvements for AI/ML workloads. This release features the general availability of In-Place Pod Resize, allowing CPU and memory adjustments without pod restarts.

Historically, scaling Kubernetes pods required recreating them, leading to service disruptions and increased resource consumption. In-Place Pod Resize addresses this by enabling dynamic resource allocation to running pods, improving application efficiency and reducing downtime. The impact of frequent pod restarts can be significant, especially for stateful applications, incurring costs associated with data synchronization and service discovery.

Key Insights

• In-Place Pod Resize GA: Eliminates pod restarts during resource adjustments, 2025
• Gang Scheduling: Designates co-scheduled pods for AI, similar to Volcano/Kueue promoting efficiency.
• /flagz & /statusz Enhancements: Provides machine-parsable HTTP output for improved observability.

Practical Applications

• Use Case: AI/ML training jobs use Gang Scheduling to ensure all necessary pods are scheduled together for optimal performance.
• Pitfall: Relying on deprecated Ingress NGINX without migrating to Gateway API risks losing support by March 2026.

References:

• https://www.infoq.com/news/2025/12/kubernetes-1-35/