From eae91f0d2befe47e4a0cf439e8f0306f7feb0cb3 Mon Sep 17 00:00:00 2001
From: seddikalaouiismaili <seddikalaouiismaili@gmail.com>
Date: Fri, 12 Mar 2021 01:38:33 +0100
Subject: [PATCH] install nrpe check for systemd units

---
 .../files/scripts/check_systemd_units         | 48 +++++++++++++++++++
 roles/nagios_client/tasks/main.yml            |  8 ++++
 2 files changed, 56 insertions(+)
 create mode 100644 roles/nagios_client/files/scripts/check_systemd_units

diff --git a/roles/nagios_client/files/scripts/check_systemd_units b/roles/nagios_client/files/scripts/check_systemd_units
new file mode 100644
index 0000000000..accdc9ab20
--- /dev/null
+++ b/roles/nagios_client/files/scripts/check_systemd_units
@@ -0,0 +1,48 @@
+#!/usr/bin/env bash
+
+# Description : script to check the status of systemd units
+# if they failed, try to restart the service once !!
+
+# Author : Seddik Alaoui Ismaili
+# Version : 1.0
+
+
+# Exits code
+warning_exit="1"
+ok_exit="0"
+
+# Unit list
+unit_list=(pagure_ci
+pagure_ev
+pagure_fast_worker
+pagure_loadjson
+pagure_logcom
+pagure_medium_worker
+pagure_milter
+pagure_mirror
+pagure_slow_worker
+pagure_webhook
+pagure_worker
+pagure_mirror_project_in.timer)
+
+#Element's arrays
+failed_array=()
+active_array=()
+
+# Check units's status
+ echo -e "here the lenght of array : ${#active_array[@]}"
+for element in ${unit_list[@]}; do
+  status=$(systemctl status ${element} |grep -E "Active:"  | awk '{ print $2 }')
+  if [ $status == failed ]; then
+    systemctl restart ${element} && active_array+=($element) || failed_array+=($element)
+  fi
+done
+
+# check the lenght of array and print result/exit code for nagios
+if [ ${#failed_array[@]} -ne "0" ]; then
+  echo -e "WARNING - Failed systemd units after restart : ${failed_array[@]}"
+  exit ${warning_exit}
+elif [ ${#failed_array[@]} -eq "0" ]; then
+  echo -e "OK - Systemd units are active"
+  exit ${ok_exit}
+fi
diff --git a/roles/nagios_client/tasks/main.yml b/roles/nagios_client/tasks/main.yml
index 970bfdda85..6a5e0f6436 100644
--- a/roles/nagios_client/tasks/main.yml
+++ b/roles/nagios_client/tasks/main.yml
@@ -54,6 +54,14 @@
   tags:
   - nagios_client
 
+- name: install nrpe check for systemd unit
+  copy: src="script/{{ item }}" dest="{{ libdir }}/nagios/plugins/{{ item }}" mode=0775 owner=nagios group=nagios
+  with_items:
+    - check_systemd_units
+  when: inventory_hostname.startswith('pagure')
+  tags:
+    - nagios_client
+
 # create dirs
 # puppet used to make /var/spool/nagios (owned by nagios.nagios) mode 750
 # and /usr/lib/nagios/plugins (owned by root) mode 755 - but we don't know WHY