Dynamic Resource Allocation Goes GA: How to Run AI Workloads on Kubernetes the Right Way
Kubernetes 1.34 brings Dynamic Resource Allocation to GA, enabling proper GPU sharing, topology-aware scheduling, and gang scheduling for AI/ML workloads.
Tag: scheduling
Kubernetes v1.35 as an AI Workload Platform: What Actually Changes for Operators
Kubernetes v1.35 continues a trend: clusters are increasingly asked to run mixed AI workloads (training, batch, and latency-sensitive inference) alongside traditional services. Here’s what’s new that matters for platform teams—especially around scheduling, resizing, and safer config workflows.
Node Readiness Controller: a practical fix for ‘Ready’ not meaning ready in Kubernetes
Kubernetes’ Node Ready condition is a blunt instrument. The new Node Readiness Controller adds declarative, taint-based readiness gates so nodes only enter the scheduling pool when platform-specific dependencies (CNI, storage, GPU drivers, local agents) are truly healthy.
Kubernetes Node Readiness Controller: breaking the “Ready” bit into actionable signals
A new Node Readiness Controller proposal reframes node health as a set of dependency-aware readiness signals—making scheduling and remediation more precise than the classic Ready/NotReady binary.
Node Readiness Controller: a new, declarative gate for safer Kubernetes node bootstrapping
Kubernetes’ binary Node Ready signal is often too coarse for modern clusters. The new Node Readiness Controller proposes a declarative, taint-driven way to keep workloads off nodes until the platform-specific dependencies you care about are truly healthy.
Kubernetes Node Readiness Controller: making “Ready” less binary
Kubernetes’ new Node Readiness Controller tackles a long-standing problem: “Ready” is binary, but modern nodes fail in nuanced ways. What’s changing, why it matters, and how to roll it out.