ecip-observability-stack 1.0.0

package/dashboards/cache-performance.json

@@ -0,0 +1,129 @@
+{
+  "description": "ECIP Knowledge Store (M03) — Cache hit rate by type and repo, Redis/pgvector performance",
+  "editable": true,
+  "panels": [
+    {
+      "title": "Cache Performance",
+      "type": "row",
+      "gridPos": { "h": 1, "w": 24, "x": 0, "y": 0 },
+      "collapsed": false
+    },
+    {
+      "title": "Cache Hit Rate",
+      "type": "gauge",
+      "datasource": "Prometheus",
+      "gridPos": { "h": 8, "w": 12, "x": 0, "y": 1 },
+      "fieldConfig": {
+        "defaults": {
+          "unit": "percentunit",
+          "min": 0,
+          "max": 1,
+          "thresholds": {
+            "mode": "absolute",
+            "steps": [
+              { "color": "red", "value": null },
+              { "color": "yellow", "value": 0.6 },
+              { "color": "green", "value": 0.8 }
+            ]
+          }
+        }
+      },
+      "targets": [
+        {
+          "expr": "cache_hit_rate{job=\"ecip-knowledge-store\"}",
+          "legendFormat": "{{cache_type}} — {{repo}}"
+        }
+      ]
+    },
+    {
+      "title": "Cache Hit Rate Over Time",
+      "type": "timeseries",
+      "datasource": "Prometheus",
+      "gridPos": { "h": 8, "w": 12, "x": 12, "y": 1 },
+      "fieldConfig": {
+        "defaults": { "unit": "percentunit" }
+      },
+      "targets": [
+        {
+          "expr": "cache_hit_rate{job=~\"ecip-knowledge-store|ecip-query-service\"}",
+          "legendFormat": "{{cache_type}} — {{repo}}"
+        }
+      ]
+    },
+    {
+      "title": "Knowledge Store Write Duration (p95)",
+      "type": "timeseries",
+      "datasource": "Prometheus",
+      "gridPos": { "h": 8, "w": 12, "x": 0, "y": 9 },
+      "fieldConfig": {
+        "defaults": {
+          "unit": "ms",
+          "thresholds": {
+            "mode": "absolute",
+            "steps": [
+              { "color": "green", "value": null },
+              { "color": "yellow", "value": 150 },
+              { "color": "red", "value": 200 }
+            ]
+          }
+        }
+      },
+      "targets": [
+        {
+          "expr": "histogram_quantile(0.95, sum(rate(knowledge_store_write_duration_ms_bucket[5m])) by (le, store_type))",
+          "legendFormat": "p95 — {{store_type}}"
+        }
+      ]
+    },
+    {
+      "title": "pgvector Query Duration (p95)",
+      "type": "timeseries",
+      "datasource": "Prometheus",
+      "gridPos": { "h": 8, "w": 12, "x": 12, "y": 9 },
+      "fieldConfig": {
+        "defaults": { "unit": "ms" }
+      },
+      "targets": [
+        {
+          "expr": "histogram_quantile(0.95, sum(rate(knowledge_store_write_duration_ms_bucket{store_type=\"pgvector\"}[5m])) by (le, namespace))",
+          "legendFormat": "p95 — {{namespace}}"
+        }
+      ]
+    },
+    {
+      "title": "HNSW Rebuild Duration",
+      "type": "timeseries",
+      "datasource": "Prometheus",
+      "gridPos": { "h": 8, "w": 24, "x": 0, "y": 17 },
+      "fieldConfig": {
+        "defaults": { "unit": "ms" }
+      },
+      "targets": [
+        {
+          "expr": "histogram_quantile(0.95, sum(rate(hnsw_rebuild_duration_ms_bucket[5m])) by (le, repo))",
+          "legendFormat": "{{repo}}"
+        }
+      ]
+    }
+  ],
+  "refresh": "30s",
+  "schemaVersion": 39,
+  "tags": ["ecip", "m03", "knowledge-store", "cache"],
+  "templating": {
+    "list": [
+      {
+        "name": "repo",
+        "type": "query",
+        "datasource": "Prometheus",
+        "query": "label_values(cache_hit_rate, repo)",
+        "refresh": 2,
+        "includeAll": true,
+        "multi": true
+      }
+    ]
+  },
+  "time": { "from": "now-1h", "to": "now" },
+  "title": "ECIP — Cache Performance",
+  "uid": "ecip-cache-performance",
+  "version": 1
+}

package/dashboards/cross-repo-fanout.json

@@ -0,0 +1,93 @@
+{
+  "description": "ECIP Query Service (M04) — Cross-repo fan-out depth distribution, cycle warnings",
+  "editable": true,
+  "panels": [
+    {
+      "title": "Cross-Repo Fan-out",
+      "type": "row",
+      "gridPos": { "h": 1, "w": 24, "x": 0, "y": 0 },
+      "collapsed": false
+    },
+    {
+      "title": "Fan-out Depth Distribution",
+      "type": "histogram",
+      "datasource": "Prometheus",
+      "gridPos": { "h": 8, "w": 12, "x": 0, "y": 1 },
+      "fieldConfig": {
+        "defaults": { "unit": "short" }
+      },
+      "targets": [
+        {
+          "expr": "sum(rate(cross_repo_fanout_count_bucket[5m])) by (le, depth)",
+          "legendFormat": "depth={{depth}}"
+        }
+      ]
+    },
+    {
+      "title": "Fan-out Depth Over Time",
+      "type": "timeseries",
+      "datasource": "Prometheus",
+      "gridPos": { "h": 8, "w": 12, "x": 12, "y": 1 },
+      "fieldConfig": {
+        "defaults": { "unit": "short" }
+      },
+      "targets": [
+        {
+          "expr": "histogram_quantile(0.95, sum(rate(cross_repo_fanout_count_bucket[5m])) by (le))",
+          "legendFormat": "p95 depth"
+        },
+        {
+          "expr": "histogram_quantile(0.50, sum(rate(cross_repo_fanout_count_bucket[5m])) by (le))",
+          "legendFormat": "p50 depth"
+        }
+      ]
+    },
+    {
+      "title": "Fan-out by Repo (top 10)",
+      "type": "bargauge",
+      "datasource": "Prometheus",
+      "gridPos": { "h": 8, "w": 12, "x": 0, "y": 9 },
+      "fieldConfig": {
+        "defaults": { "unit": "short" }
+      },
+      "targets": [
+        {
+          "expr": "topk(10, sum(rate(cross_repo_fanout_count_count[5m])) by (repo))",
+          "legendFormat": "{{repo}}",
+          "instant": true
+        }
+      ]
+    },
+    {
+      "title": "Depth > 2 Warnings (Cycle Risk)",
+      "type": "timeseries",
+      "datasource": "Prometheus",
+      "gridPos": { "h": 8, "w": 12, "x": 12, "y": 9 },
+      "fieldConfig": {
+        "defaults": {
+          "unit": "short",
+          "thresholds": {
+            "mode": "absolute",
+            "steps": [
+              { "color": "green", "value": null },
+              { "color": "red", "value": 1 }
+            ]
+          }
+        }
+      },
+      "targets": [
+        {
+          "expr": "sum(rate(cross_repo_fanout_count_bucket{le=\"+Inf\"}[5m])) - sum(rate(cross_repo_fanout_count_bucket{le=\"2\"}[5m]))",
+          "legendFormat": "Depth > 2 queries"
+        }
+      ]
+    }
+  ],
+  "refresh": "30s",
+  "schemaVersion": 39,
+  "tags": ["ecip", "m04", "cross-repo", "fanout"],
+  "time": { "from": "now-6h", "to": "now" },
+  "title": "ECIP — Cross-Repo Fan-out",
+  "uid": "ecip-cross-repo-fanout",
+  "version": 1
+}

package/dashboards/event-bus-dlq.json

@@ -0,0 +1,129 @@
+{
+  "description": "ECIP Event Bus (M07) — DLQ depth, DLQ age, retry counts",
+  "editable": true,
+  "panels": [
+    {
+      "title": "Event Bus DLQ",
+      "type": "row",
+      "gridPos": { "h": 1, "w": 24, "x": 0, "y": 0 },
+      "collapsed": false
+    },
+    {
+      "title": "DLQ Depth by Topic",
+      "type": "stat",
+      "datasource": "Prometheus",
+      "gridPos": { "h": 8, "w": 12, "x": 0, "y": 1 },
+      "fieldConfig": {
+        "defaults": {
+          "unit": "short",
+          "thresholds": {
+            "mode": "absolute",
+            "steps": [
+              { "color": "green", "value": null },
+              { "color": "yellow", "value": 50 },
+              { "color": "red", "value": 100 }
+            ]
+          }
+        }
+      },
+      "targets": [
+        {
+          "expr": "event_bus_dlq_depth",
+          "legendFormat": "{{topic}}"
+        }
+      ]
+    },
+    {
+      "title": "DLQ Depth Over Time",
+      "type": "timeseries",
+      "datasource": "Prometheus",
+      "gridPos": { "h": 8, "w": 12, "x": 12, "y": 1 },
+      "fieldConfig": {
+        "defaults": { "unit": "short" }
+      },
+      "targets": [
+        {
+          "expr": "event_bus_dlq_depth",
+          "legendFormat": "{{topic}}"
+        }
+      ]
+    },
+    {
+      "title": "Webhook Processing Latency (p95)",
+      "type": "timeseries",
+      "datasource": "Prometheus",
+      "gridPos": { "h": 8, "w": 12, "x": 0, "y": 9 },
+      "fieldConfig": {
+        "defaults": { "unit": "ms" }
+      },
+      "targets": [
+        {
+          "expr": "histogram_quantile(0.95, sum(rate(webhook_processing_duration_ms_bucket[5m])) by (le, topic))",
+          "legendFormat": "p95 — {{topic}}"
+        }
+      ]
+    },
+    {
+      "title": "Kafka Consumer Lag",
+      "type": "timeseries",
+      "datasource": "Prometheus",
+      "gridPos": { "h": 8, "w": 12, "x": 12, "y": 9 },
+      "fieldConfig": {
+        "defaults": { "unit": "short" }
+      },
+      "targets": [
+        {
+          "expr": "sum(kafka_consumergroup_lag{group=~\"ecip-event-bus.*\"}) by (topic)",
+          "legendFormat": "{{topic}}"
+        }
+      ]
+    },
+    {
+      "title": "DLQ Retry Counts",
+      "type": "timeseries",
+      "datasource": "Prometheus",
+      "gridPos": { "h": 8, "w": 12, "x": 0, "y": 17 },
+      "fieldConfig": {
+        "defaults": { "unit": "short" }
+      },
+      "targets": [
+        {
+          "expr": "sum(rate(event_bus_dlq_retries_total[5m])) by (topic)",
+          "legendFormat": "{{topic}}"
+        }
+      ]
+    },
+    {
+      "title": "DLQ Message Age (oldest)",
+      "type": "stat",
+      "datasource": "Prometheus",
+      "gridPos": { "h": 8, "w": 12, "x": 12, "y": 17 },
+      "fieldConfig": {
+        "defaults": {
+          "unit": "s",
+          "thresholds": {
+            "mode": "absolute",
+            "steps": [
+              { "color": "green", "value": null },
+              { "color": "yellow", "value": 3600 },
+              { "color": "red", "value": 86400 }
+            ]
+          }
+        }
+      },
+      "targets": [
+        {
+          "expr": "max(event_bus_dlq_oldest_message_age_seconds) by (topic)",
+          "legendFormat": "{{topic}}"
+        }
+      ]
+    }
+  ],
+  "refresh": "30s",
+  "schemaVersion": 39,
+  "tags": ["ecip", "m07", "event-bus", "dlq"],
+  "time": { "from": "now-3h", "to": "now" },
+  "title": "ECIP — Event Bus DLQ",
+  "uid": "ecip-event-bus-dlq",
+  "version": 1
+}

package/dashboards/lsp-daemon-health.json

@@ -0,0 +1,104 @@
+{
+  "description": "ECIP Analysis Engine (M02) — LSP daemon status, restart rate, OOM events",
+  "editable": true,
+  "panels": [
+    {
+      "title": "LSP Daemon Health",
+      "type": "row",
+      "gridPos": { "h": 1, "w": 24, "x": 0, "y": 0 },
+      "collapsed": false
+    },
+    {
+      "title": "LSP Daemon Restart Rate (/hour)",
+      "type": "stat",
+      "datasource": "Prometheus",
+      "gridPos": { "h": 8, "w": 8, "x": 0, "y": 1 },
+      "fieldConfig": {
+        "defaults": {
+          "unit": "short",
+          "thresholds": {
+            "mode": "absolute",
+            "steps": [
+              { "color": "green", "value": null },
+              { "color": "yellow", "value": 1 },
+              { "color": "red", "value": 2 }
+            ]
+          }
+        }
+      },
+      "targets": [
+        {
+          "expr": "sum(rate(lsp_daemon_restarts_total[1h])) by (repo, language)",
+          "legendFormat": "{{repo}} — {{language}}"
+        }
+      ]
+    },
+    {
+      "title": "LSP Daemon Restarts Over Time",
+      "type": "timeseries",
+      "datasource": "Prometheus",
+      "gridPos": { "h": 8, "w": 16, "x": 8, "y": 1 },
+      "fieldConfig": {
+        "defaults": { "unit": "short" }
+      },
+      "targets": [
+        {
+          "expr": "sum(increase(lsp_daemon_restarts_total[1h])) by (repo, language)",
+          "legendFormat": "{{repo}} — {{language}}"
+        }
+      ]
+    },
+    {
+      "title": "LSP Daemon Memory Usage",
+      "type": "timeseries",
+      "datasource": "Prometheus",
+      "gridPos": { "h": 8, "w": 12, "x": 0, "y": 9 },
+      "fieldConfig": {
+        "defaults": { "unit": "bytes" }
+      },
+      "targets": [
+        {
+          "expr": "container_memory_working_set_bytes{pod=~\"lsp-daemon.*\"}",
+          "legendFormat": "{{pod}}"
+        }
+      ]
+    },
+    {
+      "title": "LSP Daemon CPU Usage",
+      "type": "timeseries",
+      "datasource": "Prometheus",
+      "gridPos": { "h": 8, "w": 12, "x": 12, "y": 9 },
+      "fieldConfig": {
+        "defaults": { "unit": "short" }
+      },
+      "targets": [
+        {
+          "expr": "rate(container_cpu_usage_seconds_total{pod=~\"lsp-daemon.*\"}[5m])",
+          "legendFormat": "{{pod}}"
+        }
+      ]
+    },
+    {
+      "title": "OOM Kill Events",
+      "type": "timeseries",
+      "datasource": "Prometheus",
+      "gridPos": { "h": 8, "w": 24, "x": 0, "y": 17 },
+      "fieldConfig": {
+        "defaults": { "unit": "short" }
+      },
+      "targets": [
+        {
+          "expr": "sum(increase(kube_pod_container_status_last_terminated_reason{container=~\"lsp-daemon.*\", reason=\"OOMKilled\"}[1h])) by (pod)",
+          "legendFormat": "OOM — {{pod}}"
+        }
+      ]
+    }
+  ],
+  "refresh": "30s",
+  "schemaVersion": 39,
+  "tags": ["ecip", "m02", "lsp-daemon", "health"],
+  "time": { "from": "now-3h", "to": "now" },
+  "title": "ECIP — LSP Daemon Health",
+  "uid": "ecip-lsp-daemon-health",
+  "version": 1
+}

package/dashboards/mcp-call-graph.json

@@ -0,0 +1,114 @@
+{
+  "description": "ECIP Query Service (M04) + MCP Server (M05) — Fan-out topology, latency per target_repo",
+  "editable": true,
+  "panels": [
+    {
+      "title": "MCP Call Graph",
+      "type": "row",
+      "gridPos": { "h": 1, "w": 24, "x": 0, "y": 0 },
+      "collapsed": false
+    },
+    {
+      "title": "MCP Call Duration p95 by Tool",
+      "type": "timeseries",
+      "datasource": "Prometheus",
+      "gridPos": { "h": 8, "w": 12, "x": 0, "y": 1 },
+      "fieldConfig": {
+        "defaults": {
+          "unit": "ms",
+          "thresholds": {
+            "mode": "absolute",
+            "steps": [
+              { "color": "green", "value": null },
+              { "color": "yellow", "value": 500 },
+              { "color": "red", "value": 800 }
+            ]
+          }
+        }
+      },
+      "targets": [
+        {
+          "expr": "histogram_quantile(0.95, sum(rate(mcp_call_duration_ms_bucket[5m])) by (le, tool_name))",
+          "legendFormat": "{{tool_name}}"
+        }
+      ]
+    },
+    {
+      "title": "MCP Call Duration p95 by Target Repo",
+      "type": "timeseries",
+      "datasource": "Prometheus",
+      "gridPos": { "h": 8, "w": 12, "x": 12, "y": 1 },
+      "fieldConfig": {
+        "defaults": { "unit": "ms" }
+      },
+      "targets": [
+        {
+          "expr": "histogram_quantile(0.95, sum(rate(mcp_call_duration_ms_bucket[5m])) by (le, target_repo))",
+          "legendFormat": "{{target_repo}}"
+        }
+      ]
+    },
+    {
+      "title": "MCP Call Rate",
+      "type": "timeseries",
+      "datasource": "Prometheus",
+      "gridPos": { "h": 8, "w": 12, "x": 0, "y": 9 },
+      "fieldConfig": {
+        "defaults": { "unit": "reqps" }
+      },
+      "targets": [
+        {
+          "expr": "sum(rate(mcp_call_duration_ms_count[5m])) by (tool_name)",
+          "legendFormat": "{{tool_name}}"
+        }
+      ]
+    },
+    {
+      "title": "MCP Auth Failure Rate",
+      "type": "timeseries",
+      "datasource": "Prometheus",
+      "gridPos": { "h": 8, "w": 12, "x": 12, "y": 9 },
+      "fieldConfig": {
+        "defaults": { "unit": "short" }
+      },
+      "targets": [
+        {
+          "expr": "sum(rate(auth_failure_total{module=\"M05\"}[5m]))",
+          "legendFormat": "Auth Failures/s"
+        }
+      ]
+    },
+    {
+      "title": "MCP Fan-out Trace View",
+      "type": "nodeGraph",
+      "datasource": "Tempo",
+      "gridPos": { "h": 10, "w": 24, "x": 0, "y": 17 },
+      "targets": [
+        {
+          "queryType": "traceql",
+          "query": "{ span.ecip.module = \"M05\" }"
+        }
+      ]
+    }
+  ],
+  "refresh": "30s",
+  "schemaVersion": 39,
+  "tags": ["ecip", "m04", "m05", "mcp", "call-graph"],
+  "templating": {
+    "list": [
+      {
+        "name": "target_repo",
+        "type": "query",
+        "datasource": "Prometheus",
+        "query": "label_values(mcp_call_duration_ms_bucket, target_repo)",
+        "refresh": 2,
+        "includeAll": true,
+        "multi": true
+      }
+    ]
+  },
+  "time": { "from": "now-1h", "to": "now" },
+  "title": "ECIP — MCP Call Graph",
+  "uid": "ecip-mcp-call-graph",
+  "version": 1
+}