Скачать или смотреть Beyond VPA: Rightsizing Done Right | Why VPA Fails in Production (and What to Do Instead)

Kubernetes VPA was a great idea in theory, but in production environments, it breaks down fast.

In the second episode of our video series, Nic breaks down why Kubernetes Vertical Pod Autoscaler fails when it matters most, and what production-safe rightsizing actually looks like.

We’ll walk through real-world failure scenarios, compare VPA vs. real-time automation side by side, and show how context-aware rightsizing avoids downtime, restarts, and paged SREs.

If you’ve ever disabled VPA auto mode “just to be safe” - this one’s for you.

⏱️ Chapters
00:00 – Intro: Why VPA Looked Promising
VPA’s original promise and why teams wanted it to work.

00:09 – The 5 Core Architectural Flaws of VPA
Global decisions, historical bias, disruption, lack of context, and HPA conflicts.

00:33 – The Real Production Killer: Disruption at the Worst Time
Why VPA causes incidents exactly when your system is under pressure.

00:55 – Real Failure Scenario: Batch Jobs & Evictions
How VPA misreads load spikes and kills workloads mid-execution.

01:28 – Why Teams Disable VPA Auto Mode
VPA doesn’t understand business context.

01:50 – What “Rightsizing Done Right” Actually Requires
Introducing real-time, context-aware optimization.

02:24 – VPA vs ScaleOps: Side-by-Side Comparison
Historian vs first responder: what happens during a live CPU spike.

02:54 – Real-Time Reaction vs Delayed Evictions
Why immediate adjustment beats reactive eviction every time.

03:24 – Context Matters: Databases vs Stateless Apps
Why treating every workload the same is a recipe for downtime.

04:06 – Continuous, Hands-Free Optimization
How aggressive optimization can still be safe.

04:40 – Maximum Efficiency Without Sacrificing Reliability
Tight where you can, padded where you must.

05:09 – Automation vs Chaos
Why production-safe automation is the only option at scale.

🔗 Learn more
If you want to see how production-safe rightsizing works in practice, check out ScaleOps:
👉 https://scaleops.com