Metrics integration (CPU/memory via metrics-server)

[syedazeez337]

Summary

Display CPU and memory usage per pod/node by querying the Kubernetes metrics-server API.

Motivation

k9s shows real-time CPU/memory metrics. This is expected by users monitoring workloads. metrics-server is installed on most clusters (EKS, GKE, AKS all ship it by default).

Requirements

Metrics Source

• Query metrics.k8s.io/v1beta1 API via kube-rs
• Pod metrics: PodMetrics (cpu, memory per container)
• Node metrics: NodeMetrics (cpu, memory, capacity)
• kube-rs has kube::api::Api<PodMetrics> support

Display

• Optional columns: CPU and MEM (e.g., 45m, 128Mi)
• Show as percentage of request/limit when available: 45m/100m (45%)
• For nodes: show usage vs allocatable: 2.1/4 cores, 6.2/8 Gi
• Color: green (<70%), yellow (70-90%), red (>90%)

Config

[metrics]
enabled = true                  # default true — disable for clusters without metrics-server
refresh_interval_secs = 15      # how often to refresh metrics
show_percentage = true          # show % of request/limit

Implementation

• Create periodic metrics refresh task (separate from watcher)
• Store metrics in a HashMap<(namespace, name), PodMetrics> shared via Arc<RwLock<>>
• Merge metrics into K8sItem display at render time
• Don't block item display waiting for metrics — show "-" until first fetch

Graceful Degradation

• If metrics-server not installed: API returns 404 → disable metrics silently
• Log: "metrics-server not available, disabling CPU/MEM columns"
• No column shown if no metrics data

Acceptance Criteria

• Pod CPU/memory displayed from metrics-server
• Node CPU/memory with capacity shown
• Color-coded usage levels
• Configurable refresh interval
• Graceful fallback when metrics-server unavailable
• Doesn't block or slow down resource listing
• Tests: mock metrics API response