Upload production/grafana/dashboards/ml_intern.json

production/grafana/dashboards/ml_intern.json

@@ -0,0 +1,95 @@
+{
+  "dashboard": {
+    "id": null,
+    "title": "ml-intern Production Dashboard",
+    "tags": ["ml-intern", "production"],
+    "timezone": "UTC",
+    "panels": [
+      {
+        "id": 1,
+        "title": "Request Rate (RPM)",
+        "type": "graph",
+        "targets": [
+          {
+            "expr": "sum(rate(ml_intern_requests_total[1m])) by (provider)",
+            "legendFormat": "{{provider}}"
+          }
+        ],
+        "gridPos": {"h": 8, "w": 12, "x": 0, "y": 0}
+      },
+      {
+        "id": 2,
+        "title": "Request Latency (P95)",
+        "type": "graph",
+        "targets": [
+          {
+            "expr": "histogram_quantile(0.95, rate(ml_intern_request_duration_seconds_bucket[5m]))",
+            "legendFormat": "P95"
+          }
+        ],
+        "gridPos": {"h": 8, "w": 12, "x": 12, "y": 0}
+      },
+      {
+        "id": 3,
+        "title": "Active Sessions",
+        "type": "singlestat",
+        "targets": [
+          {
+            "expr": "ml_intern_active_sessions"
+          }
+        ],
+        "gridPos": {"h": 4, "w": 6, "x": 0, "y": 8}
+      },
+      {
+        "id": 4,
+        "title": "Cache Hit Rate",
+        "type": "singlestat",
+        "targets": [
+          {
+            "expr": "sum(rate(ml_intern_cache_hits_total[5m])) / (sum(rate(ml_intern_cache_hits_total[5m])) + sum(rate(ml_intern_cache_misses_total[5m])))",
+            "format": "percentunit"
+          }
+        ],
+        "gridPos": {"h": 4, "w": 6, "x": 6, "y": 8}
+      },
+      {
+        "id": 5,
+        "title": "LLM Cost (USD/hour)",
+        "type": "graph",
+        "targets": [
+          {
+            "expr": "sum(rate(ml_intern_llm_cost_usd_total[1h])) by (provider, model)",
+            "legendFormat": "{{provider}}/{{model}}"
+          }
+        ],
+        "gridPos": {"h": 8, "w": 12, "x": 0, "y": 12}
+      },
+      {
+        "id": 6,
+        "title": "Circuit Breaker States",
+        "type": "graph",
+        "targets": [
+          {
+            "expr": "ml_intern_circuit_breaker_state",
+            "legendFormat": "{{provider}}"
+          }
+        ],
+        "gridPos": {"h": 8, "w": 12, "x": 12, "y": 12}
+      },
+      {
+        "id": 7,
+        "title": "Error Rate",
+        "type": "graph",
+        "targets": [
+          {
+            "expr": "sum(rate(ml_intern_requests_total{status!~\"2..\"}[5m])) / sum(rate(ml_intern_requests_total[5m]))",
+            "legendFormat": "Error Rate"
+          }
+        ],
+        "gridPos": {"h": 8, "w": 12, "x": 0, "y": 20}
+      }
+    ],
+    "time": {"from": "now-1h", "to": "now"},
+    "refresh": "10s"
+  }
+}