From 4b04f70b6f56af9f451543a08856a44aa6be9ab8 Mon Sep 17 00:00:00 2001
From: Mike Lang <mikelang3000@gmail.com>
Date: Wed, 6 Nov 2019 18:58:12 -0800
Subject: [PATCH] overview dashboard: Add system-level metrics

---
 monitoring/dashboards/overview.jsonnet | 29 ++++++++++++++++++++++++++
 1 file changed, 29 insertions(+)

diff --git a/monitoring/dashboards/overview.jsonnet b/monitoring/dashboards/overview.jsonnet
index e83137e..5e64acd 100644
--- a/monitoring/dashboards/overview.jsonnet
+++ b/monitoring/dashboards/overview.jsonnet
@@ -160,6 +160,35 @@ grafana.dashboard({
             },
           },
         ],
+        // Third row - process-level health
+        [
+          {
+            name: "CPU usage",
+            axis: {min: 0, label: "cores", format: grafana.formats.percent},
+            expressions: {
+              "{{instance}} {{job}}": |||
+                sum by (instance, job) (
+                  rate(process_cpu_seconds_total[2m])
+                )
+              |||
+            },
+          },
+          {
+            name: "Memory usage (RSS)",
+            axis: {min: 0, format: grafana.formats.bytes},
+            expressions: {
+              "{{instance}} {{job}}": "process_resident_memory_bytes",
+            },
+          },
+          {
+            name: "Process restarts",
+            axis: {min: 0, label: "restarts within last minute"},
+            tooltip: "Multiple restarts within 15sec will be missed, and only counted as one.",
+            expressions: {
+              "{{instance}} {{job}}": "changes(process_start_time_seconds[1m])",
+            },
+          },
+        ],
       ],
     },