Keyboard shortcuts

Press or to navigate between chapters

Press S or / to search in the book

Press ? to show this help

Press Esc to hide this help

Monitoring

Operational monitoring combines CLI health checks, GPU status tracking, Prometheus metrics, and Metabase dashboards. The Health Observer daemon provides continuous background monitoring with FMEA-scored incident tracking.

Quick Status

# Overall health (all service checks)
uv run gaius-cli --cmd "/health" --format json

# GPU and endpoint status (6x RTX 4090)
uv run gaius-cli --cmd "/gpu status" --format json

# Active incidents with RPN scores
uv run gaius-cli --cmd "/health incidents" --format json

# Prometheus metrics (inference throughput, latency)
uv run gaius-cli --cmd "/observe" --format json

Monitoring Stack

LayerToolWhat it monitorsAccess
Health/healthService availability, gRPC connectivityCLI/MCP
Endpoints/gpu statusvLLM endpoints, GPU memory/tempCLI/MCP
IncidentsHealth ObserverContinuous FMEA-scored failure detectionEngine daemon
MetricsPrometheusInference latency p95, throughput, error ratesPromQL (port 9090)
DashboardsMetabaseLineage, operations, KB geometryWeb UI (port 3000)
TelemetryOpenTelemetryDistributed traces, metric instrumentationOTel Collector

Key Metrics

MetricQuery PatternAlert Threshold
Inference latency p95histogram_quantile(0.95, ...)> 5000ms
GPU utilizationFLOPS-weighted Welford mean< 10% (underutilized)
Active incidentsgaius_gaius_incidents_active> 0
FMEA RPN scoresgaius_gaius_fmea_rpn_score> 200 (escalation)
Pipeline backloggaius_gaius_pipeline_pending_cards> 50

All rate calculations use 10-minute windows to survive bursty workloads.

TUI ObservePanel

The TUI’s ObservePanel displays real-time metrics with 15-second refresh intervals. Sparklines show 5 minutes of history. Thresholds trigger color changes (green → yellow → red).

See Also