Kubernetes: Taming the Watch Cache Memory Problem

Two independent pull requests this week tackle the same root cause—kube-apiserver's watch cache holding too much decoded data in memory—while a cluster of related API machinery fixes address stale reads and leaked event handlers.

Duration: PT2M44S

Episode overview

This episode is a short developer briefing from Kubernetes.

It explains recent repository work in plain language.

  • Show: Kubernetes
  • Published: 2026-09-13T13:01:49Z
  • Audio duration: PT2M44S

Transcript excerpt

This excerpt keeps the crawler page concise. Listen to the episode or use the RSS feed for the full update.

Good day, and welcome to Kubernetes, the daily developer briefing for September 13th, 2026.

Today's clearest signal: kube-apiserver memory pressure is getting serious attention, with two separate contributors independently proposing fixes for the same underlying problem.

PR 142071 from wdoekes and PR 142066 from Pcmhacker-piro both target issue 142062: watch cache retaining fully decoded objects in memory, causing out-of-memory crashes even when object counts are moderate. One adds a max average object size flag, the other a max bytes per resource flag—both implementing byte-budget…

The second theme is correctness in the API machinery's read paths. PR 142069 fixes a race where custom resource handlers could serve stale storage after a spec change, because discovery and the CRD handler weren't ordered against each other—now the read path is level-triggered to close that gap. PR 142073 fixes a…

Elsewhere, node-level reliability got attention too. PR 142067 stops kubelet from reading cgroup data for pods whose cgroup directory is already gone during teardown, fixing spurious errors under in-place resource scaling. And two flake fixes—PR 142064 for kubeadm's config file tests and PR…

What…

Nearby episodes from Kubernetes

  1. Backport Season and the Great Feature Gate Cleanup
  2. Chasing Down Integer Overflow
  3. Quantity Math Gets an Overflow Fix, Scheduler Gains Placement Awareness
  4. Quantity Parsing Gets a Deep Cleanup
  5. Consistency Fixes and the Composite Pod Group Rollout
  6. Quiet Correctness Fixes Across the Stack
  7. Weekly Recap - API Cleanup, Go 1.27, and Scheduler Hardening
  8. Stale Cache Reads Get a Second Look