From 2eecc8f5d7efb9ce802cb76aea4fa1173aaf9d62 Mon Sep 17 00:00:00 2001
From: Leon <82407168+sed-i@users.noreply.github.com>
Date: Wed, 20 Nov 2024 05:51:45 -0500
Subject: [PATCH] Fix template variable usage (#199)

* Update host_health.rules

Signed-off-by: Leon <82407168+sed-i@users.noreply.github.com>

* Move individual alerts under group

* Rename ext to match contents

---------

Signed-off-by: Leon <82407168+sed-i@users.noreply.github.com>
---
 src/prometheus_alert_rules/high_cpu_iowait.rule  |  8 --------
 src/prometheus_alert_rules/high_cpu_iowait.rules | 11 +++++++++++
 src/prometheus_alert_rules/host_health.rules     |  4 ++--
 src/prometheus_alert_rules/oomkill.rule          |  8 --------
 src/prometheus_alert_rules/oomkill.rules         | 11 +++++++++++
 5 files changed, 24 insertions(+), 18 deletions(-)
 delete mode 100644 src/prometheus_alert_rules/high_cpu_iowait.rule
 create mode 100644 src/prometheus_alert_rules/high_cpu_iowait.rules
 delete mode 100644 src/prometheus_alert_rules/oomkill.rule
 create mode 100644 src/prometheus_alert_rules/oomkill.rules

diff --git a/src/prometheus_alert_rules/high_cpu_iowait.rule b/src/prometheus_alert_rules/high_cpu_iowait.rule
deleted file mode 100644
index eed0cd7..0000000
--- a/src/prometheus_alert_rules/high_cpu_iowait.rule
+++ /dev/null
@@ -1,8 +0,0 @@
-alert: HostCpuHighIowait
-expr: avg by (instance) (rate(node_cpu_seconds_total{mode="iowait"}[5m])) * 100 > 10
-for: 0m
-labels:
-  severity: warning
-annotations:
-  summary: Host CPU high iowait (instance {{ $labels.instance }})
-  description: "CPU iowait > 10%. A high iowait means that you are disk or network bound.\n  VALUE = {{ $value }}\n  LABELS = {{ $labels }}"
diff --git a/src/prometheus_alert_rules/high_cpu_iowait.rules b/src/prometheus_alert_rules/high_cpu_iowait.rules
new file mode 100644
index 0000000..5b2dceb
--- /dev/null
+++ b/src/prometheus_alert_rules/high_cpu_iowait.rules
@@ -0,0 +1,11 @@
+groups:
+- name: HostCpu
+  rules:
+  - alert: HostCpuHighIowait
+    expr: avg by (instance) (rate(node_cpu_seconds_total{mode="iowait"}[5m])) * 100 > 10
+    for: 0m
+    labels:
+      severity: warning
+    annotations:
+      summary: Host CPU high iowait (instance {{ $labels.instance }})
+      description: "CPU iowait > 10%. A high iowait means that you are disk or network bound.\n  VALUE = {{ $value }}\n  LABELS = {{ $labels }}"
diff --git a/src/prometheus_alert_rules/host_health.rules b/src/prometheus_alert_rules/host_health.rules
index 803e2e5..f4a3814 100644
--- a/src/prometheus_alert_rules/host_health.rules
+++ b/src/prometheus_alert_rules/host_health.rules
@@ -7,7 +7,7 @@ groups:
     labels:
       severity: critical
     annotations:
-      summary: Host '{{labels.instance}}' is down.
+      summary: Host '{{ $labels.instance }}' is down.
       description: >-
         Host '{{ $labels.instance }}' is down.
           VALUE = {{ $value }}
@@ -18,7 +18,7 @@ groups:
     labels:
       severity: critical
     annotations:
-      summary: Metrics not received from host '{{labels.instance}}'.
+      summary: Metrics not received from host '{{ $labels.instance }}'.
       description: >-
         The metrics endpoint for host '{{ $labels.instance }}' is unreachable.
           VALUE = {{ $value }}
diff --git a/src/prometheus_alert_rules/oomkill.rule b/src/prometheus_alert_rules/oomkill.rule
deleted file mode 100644
index e7fce3b..0000000
--- a/src/prometheus_alert_rules/oomkill.rule
+++ /dev/null
@@ -1,8 +0,0 @@
-alert: HostOomKillDetected
-expr: increase(node_vmstat_oom_kill[1h]) > 0
-for: 0m
-labels:
-  severity: warning
-annotations:
-  summary: Host OOM kill detected (instance {{ $labels.instance }})
-  description: "OOM kill detected\n  VALUE = {{ $value }}\n  LABELS = {{ $labels }}"
diff --git a/src/prometheus_alert_rules/oomkill.rules b/src/prometheus_alert_rules/oomkill.rules
new file mode 100644
index 0000000..21c1200
--- /dev/null
+++ b/src/prometheus_alert_rules/oomkill.rules
@@ -0,0 +1,11 @@
+groups:
+- name: HostMemory
+  rules:
+  - alert: HostOomKillDetected
+    expr: increase(node_vmstat_oom_kill[1h]) > 0
+    for: 0m
+    labels:
+      severity: warning
+    annotations:
+      summary: Host OOM kill detected (instance {{ $labels.instance }})
+      description: "OOM kill detected\n  VALUE = {{ $value }}\n  LABELS = {{ $labels }}"