From 3905dff581df91fdf0c0be70af9afad5d41a264a Mon Sep 17 00:00:00 2001
From: =?UTF-8?q?G=C3=B6rz=2C=20Friedrich?= <friedrich.goerz@tolina.local>
Date: Tue, 28 Mar 2023 12:47:47 +0000
Subject: [PATCH] DEV-471: added push metrics part to restre playbook

---
 restore-remote-database-backup.yml            | 31 +++++++++++++++++++
 .../config/prometheus/alert.rules.j2          | 24 ++++++++++++++
 2 files changed, 55 insertions(+)

diff --git a/restore-remote-database-backup.yml b/restore-remote-database-backup.yml
index 326d345..8364c80 100644
--- a/restore-remote-database-backup.yml
+++ b/restore-remote-database-backup.yml
@@ -247,6 +247,37 @@
       vars:
         record_to_remove: '{{ inventory_hostname }}'
 
+#############################################################
+#   Sending metric to prometheus push gateway
+#############################################################
+
+- hosts: "{{ stage }}-virtual-host-to-read-groups-vars"
+  serial: "{{ serial_number | default(1) }}"
+  gather_facts: false
+  connection: local
+  run_once: true
+  vars:
+    prometheus_pushgw_host: prometheus-pushgateway.monitoring:9091 # pushgw-endpoint within k8s cluster running AWX
+  tasks:
+    # attention: do not remove manually set newline char at the end of metric line
+    # in case of removing: request will die with HTTP 400 bad request
+    - name: "Set Fact"
+      set_fact:
+        metric: "nightly_restore_successful_generic {{ lookup('pipe', 'date +%s') }}\n"
+
+  # body => trim needed to remove new line char
+  # due to created yaml string
+    - name: "Send metric to prometheus pushgw"
+      uri:
+        url: "http://{{ prometheus_pushgw_host }}/metrics/job/restore_test/creator/awx-run/database_engine/{{ database_engine }}"
+        method: POST
+        body: "{{ metric }}"
+        status_code: [200]
+      register: send_prome_metric
+      retries: 5
+      delay: 5
+      until: send_prome_metric.status in [200]
+
 #############################################################
 #   Sending smardigo management message to process
 #############################################################
diff --git a/templates/prometheus/config/prometheus/alert.rules.j2 b/templates/prometheus/config/prometheus/alert.rules.j2
index 397c5ca..0386a87 100644
--- a/templates/prometheus/config/prometheus/alert.rules.j2
+++ b/templates/prometheus/config/prometheus/alert.rules.j2
@@ -174,6 +174,30 @@ groups:
       summary: "Instance <{{ '{{' }} $labels.instance {{ '}}' }}> nightly backup failed."
       description: "PostgreSQL backup failed."
 
+  - alert: nightly_restoretest_failed_postgres
+    expr: |
+      (time() - nightly_restore_successful_generic{job="restore_test",database_engine="postgres"}) > 30 * 3600 or
+      absent(nightly_restore_successful_generic{job="restore_test",database_engine="postgres"})
+    for: 2m
+    labels:
+      severity: critical
+    annotations:
+      identifier: '{{ '{{' }} $labels.instance {{ '}}' }}'
+      summary: "Instance <{{ '{{' }} $labels.instance {{ '}}' }}> nightly restore test failed."
+      description: "PostgreSQL restore test failed."
+
+  - alert: nightly_restoretest_failed_maria
+    expr: |
+      (time() - nightly_restore_successful_generic{job="restore_test",database_engine="maria"}) > 30 * 3600 or
+      absent(nightly_restore_successful_generic{job="restore_test",database_engine="maria"})
+    for: 2m
+    labels:
+      severity: critical
+    annotations:
+      identifier: '{{ '{{' }} $labels.instance {{ '}}' }}'
+      summary: "Instance <{{ '{{' }} $labels.instance {{ '}}' }}> nightly restore test failed."
+      description: "MariaDB restore test failed."
+
   - alert: megaraid_smart_errors
     expr: megaraid_smart_errors > 1
     for: 2m