Kubernetes 1.35 in Practice: What Is Worth Adopting First

Kubernetes 1.35 — Timbernetes — landed in December, and managed providers picked it up quickly. Three items justify attention ahead of the rest.

In-place pod resource updates

CPU and memory adjustments without restarting the pod. For stateful workloads and anything with a slow warm-up, this removes a category of scheduled disruption that teams have engineered around for years.

KYAML

A stricter, less ambiguous subset of YAML, graduated to beta and enabled by default in kubectl output. It eliminates the class of failures where whitespace or an unquoted string is silently reinterpreted — the so-called Norway bug, where an unquoted country code parses as a boolean. Every KYAML file is still valid YAML, so adoption carries no compatibility cost.

Precise traffic distribution

The vague PreferClose setting is replaced by two explicit options: PreferSameNode keeps traffic on the originating node when an endpoint exists there, and PreferSameZone keeps it within the zone. If you have been paying cross-zone data transfer charges on chatty internal services, this is a direct cost lever.

The thing to plan for

1.35 is the last release supporting containerd 1.x. Moving to containerd 2.0 or later is a prerequisite for the next upgrade, not an optional cleanup — and node-level changes take longer to roll than control plane ones.