Cloud Native Now

Kubernetes can make workloads remarkably resilient, but when a router, firewall, DNS service or WAN link fails, the platform may keep running while the people responsible for restoring full service have lost the ability to reach the systems they need to fix.

Redundancy does not automatically create recoverability, and a monitoring platform can tell an engineer that a switch is unreachable without guaranteeing a reliable path back to it, which is why the recovery path should be as independent as practical from the production path it is meant to repair.

Why observability does not equal control, and how to test for loss of control: zpr.io/vnyNNzSahvRi

#Kubernetes #CloudNative #SRE #Resilience #PlatformEngineering

Reliable AI depends on the infrastructure behind it.

Cloud Native Now 2026 is available on demand, with 16 sessions exploring the cloud-native technologies supporting AI applications and autonomous agents.

Hear from practitioners on improving reliability, securing AI development and understanding how applications behave in production. Explore how Kubernetes and data platforms fit into that work.

Get practical perspectives on building and operating the AI stack, with free access to the full event.

Explore the sessions 👉 cloudnativenowevent.com

#CloudNative #AIInfrastructure #PlatformEngineering #AgenticAI

When you say you run production without a centralized log archive, every SRE assumes you mean short-term retention or scrubbed sensitive fields, but this platform has no mechanism for collecting, forwarding or archiving application logs at all.

The constraint was set during the design phase, on the belief that observability cannot become a means of collecting information about users, so nodes send only metrics with strict label allowlists and verbose logging is switched on just long enough to diagnose a specific incident.

See how the architecture holds up in practice, and where the trade-offs actually land: zpr.io/ZLWdh4YK5raD

#SRE #Kubernetes #Observability #PrivacyEngineering #DevOps

Buoyant has made available an open-source control plane for managing AI workloads running on top of a vLLM inference engine inside a Kubernetes cluster, which CEO William Morgan says makes it simpler to manage those workloads in a distributed environment and scale them more cost-effectively.

It targets one of the great paradoxes of AI, that GPUs remain in short supply while utilization typically sits in the single digits, using cache data, queueing status and per-tenant attribution to get more out of a fixed pool before optionally sending requests to the cloud under conditions an IT team defines.

Read the full article on how Kinference fits self-hosted AI workloads: zpr.io/YdnR4DPBaF2w

#Kubernetes #AI #vLLM #OpenSource #GPU #PlatformEngineering

Can agentic migration move at enterprise speed and still stay governed?

@Cognizant believes it can, provided AI reasoning is used only where it adds real value and deterministic automation handles the rest. Add quality gates, human review and clear token-cost controls, and speed and governance stop pulling in opposite directions. See the full architecture in this on-demand session.

Watch now: bit.ly/4ivzPi6

#AgenticAI #CloudMigration