From 8fb9ab2213b2b7ba6a9710749065b107854ffbb0 Mon Sep 17 00:00:00 2001
From: Deepak Tiwari <deepak@obmondo.com>
Date: Wed, 18 Dec 2024 15:01:36 +0530
Subject: [PATCH] Updated the docker rules and fixed the test case for docker
 container status

---
 .../prometheus-linuxaid/rules/docker.yaml     | 15 ++++++++++
 .../templates/prometheusRule-docker.yaml      |  9 ++++++
 .../prometheus-linuxaid/tests/docker.yaml     | 28 +++++++++++++++++++
 .../prometheus-linuxaid/values.yaml           |  1 +
 4 files changed, 53 insertions(+)
 create mode 100644 argocd-helm-charts/prometheus-linuxaid/rules/docker.yaml
 create mode 100644 argocd-helm-charts/prometheus-linuxaid/templates/prometheusRule-docker.yaml
 create mode 100644 argocd-helm-charts/prometheus-linuxaid/tests/docker.yaml

diff --git a/argocd-helm-charts/prometheus-linuxaid/rules/docker.yaml b/argocd-helm-charts/prometheus-linuxaid/rules/docker.yaml
new file mode 100644
index 00000000..bb3675ff
--- /dev/null
+++ b/argocd-helm-charts/prometheus-linuxaid/rules/docker.yaml
@@ -0,0 +1,15 @@
+groups:
+  - name: monitor::system::docker
+    rules:
+      - alert: monitor::system::docker::status
+        expr: |
+          (
+            sum by (certname, name) (increase(container_last_seen{name!~"^runner.*"}[30m])) < 1200
+          ) and on(certname) obmondo_monitoring{alert_id="monitor::system::docker::status"} > 0
+        for: 5m
+        labels:
+          alert_id: monitor::system::docker::status
+          severity: critical
+        annotations:
+          summary: "Docker container {{$labels.name}} is down on host {{ $labels.certname }}"
+          description: "Docker container **{{ $labels.name }}** is down since from 10min on server **{{ $labels.certname }}**"
diff --git a/argocd-helm-charts/prometheus-linuxaid/templates/prometheusRule-docker.yaml b/argocd-helm-charts/prometheus-linuxaid/templates/prometheusRule-docker.yaml
new file mode 100644
index 00000000..4b2d44fe
--- /dev/null
+++ b/argocd-helm-charts/prometheus-linuxaid/templates/prometheusRule-docker.yaml
@@ -0,0 +1,9 @@
+{{- if .Values.prometheusRule.docker }}
+apiVersion: monitoring.coreos.com/v1
+kind: PrometheusRule
+metadata:
+  name: docker-rules
+  namespace: monitoring-{{ .Values.customerid }}
+spec:
+  {{- $.Files.Get "rules/docker.yaml" | nindent 4 }}
+{{- end }}
diff --git a/argocd-helm-charts/prometheus-linuxaid/tests/docker.yaml b/argocd-helm-charts/prometheus-linuxaid/tests/docker.yaml
new file mode 100644
index 00000000..0542fefa
--- /dev/null
+++ b/argocd-helm-charts/prometheus-linuxaid/tests/docker.yaml
@@ -0,0 +1,28 @@
+---
+evaluation_interval: 1m
+
+rule_files:
+  - ../rules/docker.yaml
+
+tests:
+  - interval: 1m
+    input_series:
+      - series: obmondo_monitoring{certname="bantha.enableit", alert_id="monitor::system::docker::status"}
+        values: 1x1800
+      - series: container_last_seen{certname="bantha.enableit", name="packagesign-script-1"}
+        values: '1734500308+35x1800'
+      - series: container_last_seen{certname="bantha.enableit", name="jenkins"}
+        values: '1734500308+60x1800'
+
+    alert_rule_test:
+      - alertname: monitor::system::docker::status
+        eval_time: 30m
+        exp_alerts:
+          - exp_labels:
+              severity: critical
+              certname: bantha.enableit
+              name: packagesign-script-1
+              alert_id: monitor::system::docker::status
+            exp_annotations:
+              summary: 'Docker container packagesign-script-1 is down on host bantha.enableit'
+              description: 'Docker container **packagesign-script-1** is down since from 10min on server **bantha.enableit**'
diff --git a/argocd-helm-charts/prometheus-linuxaid/values.yaml b/argocd-helm-charts/prometheus-linuxaid/values.yaml
index 194973c9..96af9924 100644
--- a/argocd-helm-charts/prometheus-linuxaid/values.yaml
+++ b/argocd-helm-charts/prometheus-linuxaid/values.yaml
@@ -32,6 +32,7 @@ prometheusRule:
   cpu: true
   dellhw: true
   disk: true
+  docker: true
   drbd: true
   elasticsearch: true
   file_size: true