Diagnose GKE TPU slowdowns and interruptions.
What it does
Diagnoses infrastructure causes of GKE TPU workload interruptions and performance issues using system metrics.
When to use it
Use for metric-based monitoring of TPU utilization, memory, node health, node-pool availability, interruptions, and recovery; not for non-TPU workloads or non-metric debugging.
How to use it
Give it a GKE TPU monitoring or troubleshooting request; it gathers cluster context and diagnoses infrastructure issues using metrics.
Uses
What you need · 1
Requires a TPU cluster with GKE System Metrics enabled and container port 8431 exposed. Runtime metrics require GKE 1.27.4-gke.900 or later; node-status queries require 1.32.1-gke.1357001 or later.
Access · 2
This skill
GKE
Inspects GKE cluster configuration
Cloud Monitoring
Reads TPU health metrics