Kubernetesv1.34

15 April 2026

What's New in Kubernetes 1.34

Kubernetes 1.34 focuses on AI/ML workload management: GPU topology-aware scheduling graduates to stable, clustered workload identity reaches GA, and the new Job Success Policy simplifies batch pipeline orchestration at scale.

• 1GPU topology-aware scheduling GA — Kubernetes natively places pods on nodes that share NVLink or NVSwitch fabric, reducing inter-GPU communication latency for large model training
• 2Workload Identity Federation GA — pods can exchange ServiceAccount tokens for cloud IAM credentials without storing static secrets, compatible with AWS IRSA, Azure Workload Identity, and GCP WIF
• 3Job Success Policy — define custom success criteria for indexed Jobs, enabling partial-completion semantics for distributed training
• 4Persistent Volume health monitoring GA — volume condition events surface I/O errors and degraded states to applications
• 5Node resource lifecycle controller — automated graceful cordon and drain on hardware health signals (memory ECC errors, disk SMART failures)
• 6Structured Admission Policies (CEL-based) replace many ValidatingWebhookConfigurations with in-process CEL expressions — faster and lower operational overhead
• 7ClusterTrustBundle GA — standardised distribution of trust anchors for mTLS across workloads