Kubernetes Scheduling and Capacity
31 incident problems about Kubernetes Scheduling and Capacity. Start with the reviewed ones.
먼저 읽을 가이드
추천 문제
All problems (31)
K8S-088StatefulSet ordinal reuses a stale PVC and replays old node identity into the quorum clusterThe pod comes up, but the distributed system behaves incorrectly because the restored ordinal is carrying a PVC from a previous logical member.KubernetesAdvanced23 minProK8S-059PriorityClass protects system pods but starves customer workloads during node pressureThe cluster keeps critical control components alive as intended, but the chosen priority and preemption policy leave business workloads permanently unschedulable.KubernetesAdvanced25 minProK8S-054Volume attach succeeds but the pod never mountsThe control plane reports the volume attachment as healthy, yet the pod stays stuck on a subset of nodes because the node-side plugin is not present everywhere.KubernetesAdvanced27 minProK8S-060HPA and PodDisruptionBudget combine to deadlock a low-replica rolloutAutoscaling and disruption control each look reasonable alone, but together they prevent the deployment from making forward progress during an update.KubernetesAdvanced28 minProK8S-029CSI driver recovers volume attach only after controller restartPersistent volume attachment gets stuck until the controller is restarted, pointing to an unhealthy reconciliation loop.KubernetesAdvanced29 minProK8S-032PodDisruptionBudget blocks maintenance drain on a low-replica deploymentThe node drain command hangs even though workloads seem healthy, because a tight disruption budget and replica count leave no legal eviction path.KubernetesIntermediate20 minProK8S-083HPA scales on CPU but the true bottleneck is a single-threaded queue consumer partition assignmentAutoscaling appears to help, yet throughput never rises because the workload's shard or partition ownership keeps all useful work pinned to one replica.KubernetesAdvanced22 minProK8S-061Topology spread and zone taints keep replicas pending after a node drainCapacity still exists in the cluster, but the workload cannot recover because spread rules and taints together eliminate every legal placement choice.KubernetesAdvanced24 minProK8S-065One node rejects new podsThe node still reports in, but admission and management paths fail because kubelet trust renewal broke when time drift invalidated the rotation flow.KubernetesAdvanced25 minProK8S-037TopologySpreadConstraints leave replicas pending although one node has roomThe cluster has enough raw CPU and memory, but zone spread requirements and maxSkew prevent any legal placement for the remaining replicas.KubernetesAdvanced26 minProK8S-040Node eviction startsApplication storage looks healthy on the persistent volume, but a temporary working directory on the node root filesystem silently fills and triggers eviction pressure.KubernetesIntermediate24 minPro