From 580c497261dd03026a83266d2a383d7546fbbfd8 Mon Sep 17 00:00:00 2001
From: Vitaly Zhuravlev <v-zhuravlev@users.noreply.github.com>
Date: Tue, 28 Mar 2023 06:58:17 +0800
Subject: [PATCH] Add NodeSystemSaturation and NodeMemoryMajorPagesFaults

Signed-off-by: Vitaly Zhuravlev <v-zhuravlev@users.noreply.github.com>
---
 docs/node-mixin/alerts/alerts.libsonnet | 37 ++++++++++++++++++++++++-
 1 file changed, 36 insertions(+), 1 deletion(-)

diff --git a/docs/node-mixin/alerts/alerts.libsonnet b/docs/node-mixin/alerts/alerts.libsonnet
index 2ea61ba1..071cfd8d 100644
--- a/docs/node-mixin/alerts/alerts.libsonnet
+++ b/docs/node-mixin/alerts/alerts.libsonnet
@@ -323,6 +323,41 @@
               description: 'CPU usage at {{ $labels.instance }} has been above 80% for the last 15 minutes, is currently at {{ printf "%.2f" $value }}%.',
             },
           },
+          {
+            alert: 'NodeSystemSaturation',
+            expr: |||
+              node_load1{%(nodeExporterSelector)s}
+              / count without (cpu, mode) (node_cpu_seconds_total{%(nodeExporterSelector)s, mode="idle"}) > 2
+            ||| % $._config,
+            'for': '15m',
+            labels: {
+              severity: 'critical',
+            },
+            annotations: {
+              summary: 'System saturated, load per core is very high.',
+              description: |||
+                System load per core at {{ $labels.instance }} has been above 2 for the last 15 minutes, is currently at {{ printf "%.2f" $value }}.
+                This might indicate this instance resources saturation and can cause it becoming unresponsive.
+              |||,
+            },
+          },
+          {
+            alert: 'NodeMemoryMajorPagesFaults',
+            expr: |||
+              rate(node_vmstat_pgmajfault{%(nodeExporterSelector)s}[5m]) > 500
+            ||| % $._config,
+            'for': '15m',
+            labels: {
+              severity: 'warning',
+            },
+            annotations: {
+              summary: 'Memory major page faults are occurring at very high rate.',
+              description: |||
+                Memory major pages are occurring at very high rate at {{ $labels.instance }}, 500 major page faults per second for the last 15 minutes, is currently at {{ printf "%.2f" $value }}.
+                Please check that there is enough memory available at this instance.
+              |||,
+            },
+          },
           {
             alert: 'NodeMemoryHighUtilization',
             expr: |||
@@ -352,7 +387,7 @@
               summary: 'Disk IO queue is high.',
               description: |||
                 Disk IO queue (aqu-sq) is high on {{ $labels.device }} at {{ $labels.instance }}, has been above 10 for the last 15 minutes, is currently at {{ printf "%.2f" $value }}.
-                This symptom might indicate disk saturation.,
+                This symptom might indicate disk saturation.
               |||,
             },
           },