Kubernetes Podcast from Google

Episode: Kubernetes 1.34 Of Wind & Will, With Vyom Yadav

Hosts: Abdel Sghiouar, Kaslin Fields

Guest: Vyom Yadav (Software Engineer, Canonical; Kubernetes 1.34 Release Lead)
Release Date: August 27, 2025

Episode Overview

This episode of the Kubernetes Podcast dives deep into Kubernetes 1.34, themed "Of Wind & Will," with release lead Vyom Yadav. The discussion explores the meaning behind the nautical theme, offers a detailed breakdown of major features, deprecations, and updates in the release, and highlights how this cycle reflects the community’s ongoing efforts to address real-world friction points. The conversation also touches on major enhancements such as KML (a YAML alternative), improvements for resource allocation and tracing, and new approaches to multi-tenancy and resource management.

Key Highlights & Discussion Points

[03:39] The Story & Theme of Kubernetes 1.34 — "Of Wind & Will"

• Vyom introduces the theme’s origin:

  "Every release cycle we inherit a lot of winds from previous release cycles... Sometimes this wind fills our sails, sometimes it destabilizes the ship... The ship wouldn’t be able to steer without the sailors on board." (Vyom Yadav, 03:39)

• The release is dedicated to the contributors ("sailors") who guide the project, not just the features ("wind").
• The theme and release title—Of Wind & Will—honors both legacy and the determination of the Kubernetes community.

"This is a release that is powered by the wind around us and the will within us."
(Vyom Yadav, 05:08)

[06:39] How Release Work Begins

• Work for each release typically starts immediately or just days after the prior cycle ends, especially for the release lead.
• The assembly of the release team is based on priority of sub-teams, with enhancements prioritized early.

[09:09] What’s New in 1.34: Feature Breakdown

Enhancements & Stats

• 58 enhancements in total (23 stable, 22 beta, 13 alpha).
• "Slightly lower than last release... but very close." (Vyom Yadav, 09:09)

[10:30] Key Deprecations

• CGroup Driver Auto-Discovery ([KEP-4033]):
  • Helps automate cgroup driver selection to reduce administrator pain points.
  • Deprecates manual kubelet config; requires updated containerd (v1.7+).
  • "Getting the cgroup driver correct has been a pain point... this KEP solves this problem." (Vyom Yadav, 09:36)
• Service Routing "PreferLocal" Renamed ([KEP-3015]):
  • "PreferLocal" is now "PreferSameZone" (more descriptive).
  • Adds "PreferSameNode" for finer control; old alias is deprecated but not removed from API.
  • "Prefer close is quite ambiguous... PreferSameZone makes a lot of sense especially in cloud providers." (Abdel Sghiouar, 12:14)

[12:45] KML (Kubernetes Manifest Language) — A New YAML Alternative

• KML is a strict, less error-prone YAML subset for manifests.
• Reduces whitespace sensitivity and solves classic YAML pitfalls (like the "Norway bug").
• Makes large manifests easier to read and maintain; improved compatibility with tooling.
• Not replacing YAML: KML is acceptable anywhere YAML is.
• "YAML isn’t going anywhere... KML is actually YAML... It’s just a particular subset in the YAML grammar." (Vyom Yadav, 15:25)
• "It’s slightly more clean than just like simply YAML... I have to get my hands on it on very large manifests." (Abdel Sghiouar, 15:43)

[16:42] Pod-level Service Account Tokens for Registry Authentication

• Now possible to set image pull authentication at the pod level using projected service account tokens and OIDC.
• Enhances support for multi-tenancy and security; no more node-bound secrets.
• Allows more granular role-based access control for registry pulls.
• "A really good aspect of this is you don’t have any secrets that are bound to the node, which means... a pod does not have access to the images of some other pod." (Vyom Yadav, 18:22)

[19:58] Dynamic Resource Allocation (DRA) Goes Stable

• DRA core becomes generally available (GA), supporting more advanced and dynamic handling of GPUs, FPGAs, and similar resources.
• Shifts from opaque device parameters to structured claims using CEL (Common Expression Language), enabling better scheduling and autoscaling.
• "This is a major milestone for DRA because it has gone through its ups and downs and now is finally going stable." (Vyom Yadav, 23:57)
• Previous DRA implementation was reverted to enable autoscaler compatibility; the new approach bridges that gap.

[24:20] Control Plane Tracing for Observability

• API Server and Kubelet tracing both hit GA in this release ([KEPs 647 & 2831]).
• Enables span and trace context export to standard collectors (e.g., OpenTelemetry).
• Supports defining sampling rates and custom trace endpoints.
• "Both of these KEPs going GA at the same time improve the observability aspect of Kubernetes quite a lot." (Vyom Yadav, 25:55)
• Focus: Tracing is for control plane actions (API calls, reconcilers), not data path/logging.

[27:35] Pod-level Resource Requests and Limits

• Enables specification of resource limits at the pod level, not just per-container.
• Useful for future features (e.g., pod quotas).
• "Having it at the top level of pod and not caring about how the container split it." (Vyom Yadav, 28:55)

[29:11-29:50] Reflections and Closing Thoughts

• Vyom appreciates the balanced focus of 1.34: not just chasing AI features, but addressing wider user friction.
• Expresses relief and plans for a well-deserved vacation post-release.
• "I like the narrative behind [Wind & Will]. So kudos on that." (Abdel Sghiouar, 29:11)
• "It has been a very long grind and I want to take a break." (Vyom Yadav, 29:50)

Notable Quotes

• "This is a release that is powered by the wind around us and the will within us."
  — Vyom Yadav, [05:08]
• "KML is actually YAML. It is YAML and all YAML parsers can work with KML."
  — Vyom Yadav, [15:25]
• "A really good aspect of this is you don’t have any secrets that are bound to the node, which means... a pod does not have access to the images of some other pod."
  — Vyom Yadav, [18:22]
• "Both of these KEPs going GA at the same time improve the observability aspect of Kubernetes quite a lot."
  — Vyom Yadav, [25:55]

Timestamps & Segment Guide

• [03:39] — Theme and narrative of 1.34
• [06:39] — Timing and process for a release cycle
• [09:09] — Overview of new features, enhancements, and stats
• [10:30] — Deprecation highlights: cgroup driver, service routing terminology
• [12:45] — KML (better YAML) introduction and impacts
• [16:42] — Pod-level service account tokens for image pull
• [19:58] — Dynamic Resource Allocation (DRA) becomes stable
• [24:20] — Control plane (API server/Kubelet) tracing GA
• [27:35] — Pod-level resource requests/limits
• [29:11] — Host/guest reflections and close

Takeaways

• Kubernetes 1.34 is characterized by mature, highly targeted enhancements that directly address community needs, usability, security, and observability, as well as thoughtful deprecations and terminology clarifications.
• The theme underscores Kubernetes’ collaborative spirit, honoring not just code but the community’s resilience and effort.
• Features like KML, pod-level service account tokens, DRA’s stability, and enhanced tracing reflect this round’s philosophy: empower users, ease operations, honor contributors.