What problem does it solve? When RDMA networking breaks in a Kubernetes cluster running the NVIDIA Network Operator, finding the root cause is slow: pods lose network attachments, SR-IOV VFs disappear, IPAM pools exhaust, and operator pods crash without obvious signals. This Skill provides a systematic diagnostic flow that maps symptoms to the exact commands and log files needed to isolate the fault. ## Core Features & Use Cases - Guided Diagnostic Flow: Step-by-step checks starting from must-gather collection, through operator health verification, to targeted checks for network attachment, SR-IOV, IPAM, and connectivity issues. - Symptom-to-Cause Decision Tree: A lookup table mapping common symptoms (CrashLoopBackOff, stuck Pending pods, missing VFs, IP allocation failures) to the specific check command and likely root cause. - Stuck Resource Cleanup: Commands to find orphaned finalizers, force-delete terminating namespaces, and remove evicted pods. - Use Case: A cluster admin notices application pods are stuck Pending with no RDMA network. Following the Skill, they run the SR-IOV sync checks, discover the PCI BDF in NETOP_NETLIST is wrong, fix the config, and verify VFs appear. ## Quick Start Ask the assistant to troubleshoot why pods in your Kubernetes cluster are not getting RDMA network from the NVIDIA Network Operator.
