Topic
Kubernetes
Kubernetes stopped being interesting as a technology around the time it became boring as infrastructure, and that is precisely when the operational problems got hard. The API is stable. The scheduler works. What breaks now is everything sitting on top: requests set by copy-paste, autoscalers fighting each other, node pools that bin-pack beautifully in a spreadsheet and terribly in production, and eviction behaviour nobody modelled until the first 3 a.m. page.
This section covers the parts of cluster operation where the documentation stops and judgement starts. How to actually pick CPU requests when your workload has a bimodal latency profile. Why the vertical and horizontal autoscalers cannot both own the same signal. What Karpenter changes about node lifecycle, and what it quietly makes worse. Where spot capacity is defensible for stateful services and where it is negligence. Expect specific numbers, real manifests, and a stated position rather than a balanced list of considerations.
Articles in Kubernetes
-
Bin-Packing Is Where Your Cluster Utilisation Actually Comes From
Improving Kubernetes node utilisation through bin-packing: MostAllocated scoring, instance shape selection, topology constraints that fragment capacity, and descheduler-driven consolidation.
-
Spot Instances for Stateful Workloads Are Defensible, With Conditions
When spot capacity is safe for databases, queues, and caches — interruption modelling, capacity-pool diversification, PDB configuration, and the recovery-time math that decides it.
-
Karpenter vs Cluster Autoscaler: What You Actually Trade Away
A practical comparison of Karpenter and cluster-autoscaler: provisioning latency, bin-packing quality, disruption behaviour, consolidation risk, and which one fits which cluster.
-
VPA and HPA Will Fight Each Other Unless You Separate Their Signals
Why vertical and horizontal pod autoscaling oscillate when they share a metric, the specific configurations that are safe, and how to use VPA in recommendation-only mode as a right-sizing engine.
-
Requests Are a Scheduling Contract, Limits Are a Blast Radius
CPU and memory requests and limits solve different problems. A concrete methodology for setting each, why CPU limits usually hurt, and how throttling shows up in your latency percentiles.
Other topics
FinOps Platform Engineering AI Infrastructure Cloud Security