Skip to content

All posts tagged “kubernetes”

Tag: kubernetes

Three Kubernetes Projects Shipped a Brake Pedal This Month. Why?

In one week of late-August 2026, Agent Sandbox, Kueue, and LeaderWorkerSet each shipped a limiter on their own elasticity: a warm pool that refills more slowly, a field that used to be editable and is now frozen, and an autoscaler that ships with a guard against doing what it was asked. The reason they converged now is pod churn against a control plane built for slower turnover and accelerator scarcity that rules out idle capacity—two pressures that turn a refill loop into a thundering herd of its own making. Read at the level of what each controller actually does—the exact flags, the quota invariant, the no-shrink floor—it's the same lesson three times. A control loop you add to smooth a system will, past some load, become the fastest-moving thing in it, and the grown-up move is to bound it. Plus the honest counter-case, because the same week also shipped features that add elasticity.

Compute Scarcity Is Permanent. Build a Ladder.

Every resilience pattern you know assumes the machine shows up. Accelerator scarcity breaks that assumption, and the pattern catalogs have not caught up. Here is the Compute Fallback Ladder—an ordered set of rungs your workload can run on, a selector that picks the highest obtainable one, and a promotion path back up when capacity returns.

Your Agents Aren't Too Insecure. They're Too Awake.

Agent density is a lifecycle problem, not an isolation problem. Google's own GKE Agent Sandbox benchmark puts the sandbox at 44% and suspend and resume at up to 3.5x—the sandbox is the smaller half. And upstream Kubernetes 1.37 is quietly turning checkpoint and restore into a kubelet primitive. Here's why the economics live in the lifecycle, not the box.

Schedule the Accelerators: A GPU/TPU Puzzle

I'm hiring a Senior Developer Relations Engineer for GKE and AI Infrastructure. Instead of describing the work, I built it: a small, playable cluster scheduler that teaches the real problem behind running AI at scale—gang scheduling, memory ceilings, spot preemption, and a TPU slice with actual topology.