microdao-daarion/ops/grafana_voice_dashboard.json

{
  "__inputs": [
    {
      "name": "DS_PROMETHEUS",
      "label": "Prometheus",
      "description": "Prometheus datasource — point to your Prometheus instance",
      "type": "datasource",
      "pluginId": "prometheus",
      "pluginName": "Prometheus"
    }
  ],
  "title": "DAARION Voice SLO Dashboard",
  "uid": "voice-slo",
  "description": "Voice pipeline SLO: TTFA, LLM latency, TTS health, queue underflows. Aligns with ops/voice_alerts.yml and config/slo_policy.yml.",
  "tags": ["voice", "slo", "daarion"],
  "timezone": "browser",
  "refresh": "30s",
  "time": { "from": "now-1h", "to": "now" },
  "panels": [

    {
      "id": 1,
      "title": "⏱ Time-to-First-Audio p50 / p95",
      "description": "SLO: voice_fast_uk p95 ≤ 5000ms | voice_quality_uk p95 ≤ 7000ms",
      "type": "timeseries",
      "gridPos": { "x": 0, "y": 0, "w": 12, "h": 8 },
      "datasource": { "type": "prometheus", "uid": "${DS_PROMETHEUS}" },
      "fieldConfig": {
        "defaults": {
          "unit": "ms",
          "custom": { "lineWidth": 2 },
          "thresholds": {
            "mode": "absolute",
            "steps": [
              { "value": null,  "color": "green"  },
              { "value": 5000,  "color": "yellow" },
              { "value": 7000,  "color": "red"    }
            ]
          }
        }
      },
      "targets": [
        {
          "expr": "histogram_quantile(0.50, rate(voice_ttfa_ms_bucket{voice_profile='voice_fast_uk'}[$__rate_interval]))",
          "legendFormat": "fast p50"
        },
        {
          "expr": "histogram_quantile(0.95, rate(voice_ttfa_ms_bucket{voice_profile='voice_fast_uk'}[$__rate_interval]))",
          "legendFormat": "fast p95"
        },
        {
          "expr": "histogram_quantile(0.95, rate(voice_ttfa_ms_bucket{voice_profile='voice_quality_uk'}[$__rate_interval]))",
          "legendFormat": "quality p95"
        }
      ],
      "options": {
        "legend": { "displayMode": "table", "placement": "bottom", "calcs": ["lastNotNull", "max"] }
      }
    },

    {
      "id": 2,
      "title": "🤖 LLM Latency by Model",
      "description": "LLM inference time per model. Use to identify slow models and trigger auto-promote.",
      "type": "timeseries",
      "gridPos": { "x": 12, "y": 0, "w": 12, "h": 8 },
      "datasource": { "type": "prometheus", "uid": "${DS_PROMETHEUS}" },
      "fieldConfig": {
        "defaults": { "unit": "ms" }
      },
      "targets": [
        {
          "expr": "histogram_quantile(0.50, rate(voice_llm_ms_bucket[$__rate_interval])) by (model)",
          "legendFormat": "{{ model }} p50"
        },
        {
          "expr": "histogram_quantile(0.95, rate(voice_llm_ms_bucket[$__rate_interval])) by (model)",
          "legendFormat": "{{ model }} p95"
        }
      ],
      "options": {
        "legend": { "displayMode": "table", "placement": "bottom", "calcs": ["lastNotNull", "max"] }
      }
    },

    {
      "id": 3,
      "title": "🔊 TTS Health: Synthesis Time + Error Rate",
      "description": "SLO: tts_first_ms p95 ≤ 2000ms. Error rate > 0.05/s → alert.",
      "type": "timeseries",
      "gridPos": { "x": 0, "y": 8, "w": 12, "h": 8 },
      "datasource": { "type": "prometheus", "uid": "${DS_PROMETHEUS}" },
      "fieldConfig": {
        "defaults": { "unit": "ms" },
        "overrides": [
          {
            "matcher": { "id": "byName", "options": "errors/s" },
            "properties": [
              { "id": "custom.axisPlacement", "value": "right" },
              { "id": "unit", "value": "short" },
              { "id": "color", "value": { "fixedColor": "red", "mode": "fixed" } }
            ]
          }
        ]
      },
      "targets": [
        {
          "expr": "histogram_quantile(0.50, rate(voice_tts_first_ms_bucket[$__rate_interval]))",
          "legendFormat": "tts_first p50"
        },
        {
          "expr": "histogram_quantile(0.95, rate(voice_tts_first_ms_bucket[$__rate_interval]))",
          "legendFormat": "tts_first p95"
        },
        {
          "expr": "histogram_quantile(0.95, rate(voice_tts_compute_ms_bucket[$__rate_interval])) by (engine)",
          "legendFormat": "{{ engine }} compute p95"
        },
        {
          "expr": "rate(voice_tts_errors_total[$__rate_interval])",
          "legendFormat": "errors/s"
        }
      ],
      "options": {
        "legend": { "displayMode": "table", "placement": "bottom", "calcs": ["lastNotNull", "max"] }
      }
    },

    {
      "id": 4,
      "title": "📊 Queue Underflows + E2E Latency",
      "description": "Underflow = playback outran TTS synthesis (silence gap). E2E SLO: p95 ≤ 9000ms.",
      "type": "timeseries",
      "gridPos": { "x": 12, "y": 8, "w": 12, "h": 8 },
      "datasource": { "type": "prometheus", "uid": "${DS_PROMETHEUS}" },
      "fieldConfig": {
        "defaults": { "unit": "ms" },
        "overrides": [
          {
            "matcher": { "id": "byName", "options": "underflows/min" },
            "properties": [
              { "id": "custom.axisPlacement", "value": "right" },
              { "id": "unit", "value": "short" },
              { "id": "color", "value": { "fixedColor": "orange", "mode": "fixed" } }
            ]
          }
        ]
      },
      "targets": [
        {
          "expr": "histogram_quantile(0.95, rate(voice_e2e_ms_bucket{voice_profile='voice_fast_uk'}[$__rate_interval]))",
          "legendFormat": "e2e fast p95"
        },
        {
          "expr": "histogram_quantile(0.95, rate(voice_e2e_ms_bucket{voice_profile='voice_quality_uk'}[$__rate_interval]))",
          "legendFormat": "e2e quality p95"
        },
        {
          "expr": "rate(voice_queue_underflows_total[$__rate_interval]) * 60",
          "legendFormat": "underflows/min"
        }
      ],
      "options": {
        "legend": { "displayMode": "table", "placement": "bottom", "calcs": ["lastNotNull", "max"] }
      }
    },

    {
      "id": 5,
      "title": "🚦 SLO Status (Stat)",
      "type": "stat",
      "gridPos": { "x": 0, "y": 16, "w": 24, "h": 4 },
      "datasource": { "type": "prometheus", "uid": "${DS_PROMETHEUS}" },
      "fieldConfig": {
        "defaults": {
          "thresholds": {
            "mode": "absolute",
            "steps": [
              { "value": null,  "color": "green"  },
              { "value": 5000,  "color": "yellow" },
              { "value": 7000,  "color": "red"    }
            ]
          },
          "mappings": []
        }
      },
      "options": {
        "reduceOptions": { "calcs": ["lastNotNull"] },
        "orientation": "horizontal",
        "colorMode": "background"
      },
      "targets": [
        {
          "expr": "histogram_quantile(0.95, rate(voice_ttfa_ms_bucket{voice_profile='voice_fast_uk'}[10m]))",
          "legendFormat": "TTFA fast p95"
        },
        {
          "expr": "histogram_quantile(0.95, rate(voice_tts_first_ms_bucket[10m]))",
          "legendFormat": "TTS first p95"
        },
        {
          "expr": "histogram_quantile(0.95, rate(voice_e2e_ms_bucket{voice_profile='voice_fast_uk'}[10m]))",
          "legendFormat": "E2E fast p95"
        },
        {
          "expr": "rate(voice_tts_errors_total[10m])",
          "legendFormat": "TTS errors/s"
        }
      ]
    }
  ]
}