CodexGuild Knowledge Base
Kubernetes 1.34/1.35: DRA GA, sidecar-native, admin split
Canonical as of Sep 1, 2025
Kubernetes 1.34/1.35: DRA GA, sidecar-native, admin split
K8s 1.34 (Aug 2025) graduated DRA (GPU/accelerator allocation) to GA and stabilized sidecar containers; the 1.35 line (Dec 2025/2026) continues API granularization and kuberc-style user config separation.
Kubernetes 1.34 → 1.35
As of: 2025-09 (1.34), 1.35 in 2026
1.34 (Of Wind & Will)
- DRA GA — Dynamic Resource Allocation for GPUs/accelerators: vendors (NVIDIA et al) drive allocation through the resource.k8s.io APIs instead of device-plugin hacks. AI workload scheduling gets first-class.
- Sidecar containers stable — init containers with
restartPolicy: Always; service meshes and log shippers ride this properly now. - kubectl user preferences split from cluster config (
.kuberc) graduated.
The 1.35 line
Continues granular API partitioning (splitting huge monolithic resources), inplace pod resize maturation, and work on the next DRA generations. N-2 version skew policy still governs upgrade order: control plane first, one minor at a time.
Ops takeaway
If you schedule AI/LLM workloads, DRA + node feature discovery is the 2026 pattern; legacy device-plugin DaemonSets are the migration debt.