A Kubernetes pod repeatedly restarted with exit code 137 despite not hitting its memory limit, revealing that node‑level memory pressure and QoS‑based eviction caused the pod to be killed, and outlining how to diagnose and prevent such OOMKill events.
A pod in a Kubernetes cluster repeatedly restarted with exit code 137 despite staying well below its 6 Gi memory limit, prompting an investigation that uncovered the role of QoS classes, oom_score calculations, and node‑level memory pressure in the eviction process.
This article walks through a real‑world Kubernetes incident where a node’s disk usage exceeded the eviction threshold, causing pods to enter the Evicted state, and details the investigation steps, root‑cause analysis, and practical remediation actions.
This article explains why Kubernetes pods get evicted, covering preemptive eviction, node‑pressure eviction, pod scheduling, priority classes, QoS tiers, alternative eviction methods, and how to monitor evictions with Prometheus, providing concrete commands and examples.
This article explains why Kubernetes does not automatically rebalance pods, demonstrates how to use the Descheduler, Node Problem Detector, and Cluster Autoscaler together to detect node pressure, evict overloaded pods, and scale down underutilized nodes for improved cluster efficiency.
This article explains why Kubernetes pods are evicted, covering preemption, node‑pressure eviction, pod scheduling, priority classes, QoS tiers, other eviction methods, and how to monitor evictions with Prometheus, providing practical examples and command‑line snippets.
This article explains how to investigate large numbers of evicted Kubernetes pods caused by disk pressure, examines the role of SkyWalking log files stored in EmptyDir volumes, and provides both temporary and permanent solutions to free space and prevent future evictions.
