work

ops/devops/orchestrator/grafana/orchestrator-overview.json

@@ -0,0 +1,42 @@
+{
+  "schemaVersion": 39,
+  "title": "Orchestrator Overview",
+  "panels": [
+    {
+      "type": "stat",
+      "title": "Queue Depth",
+      "datasource": "Prometheus",
+      "fieldConfig": {"defaults": {"unit": "none"}},
+      "targets": [{"expr": "sum(job_queue_depth)"}]
+    },
+    {
+      "type": "timeseries",
+      "title": "Queue Depth by Job Type",
+      "datasource": "Prometheus",
+      "targets": [{"expr": "job_queue_depth"}],
+      "fieldConfig": {"defaults": {"unit": "none"}}
+    },
+    {
+      "type": "timeseries",
+      "title": "Failures per minute",
+      "datasource": "Prometheus",
+      "targets": [{"expr": "rate(job_failures_total[5m])"}],
+      "fieldConfig": {"defaults": {"unit": "short"}}
+    },
+    {
+      "type": "timeseries",
+      "title": "Leases per second",
+      "datasource": "Prometheus",
+      "targets": [{"expr": "rate(lease_extensions_total[5m])"}],
+      "fieldConfig": {"defaults": {"unit": "ops"}}
+    },
+    {
+      "type": "timeseries",
+      "title": "Job latency p95",
+      "datasource": "Prometheus",
+      "targets": [{"expr": "histogram_quantile(0.95, sum(rate(job_latency_seconds_bucket[5m])) by (le))"}],
+      "fieldConfig": {"defaults": {"unit": "s"}}
+    }
+  ],
+  "time": {"from": "now-6h", "to": "now"}
+}