scripts/xymon/xymon.files.alert.sh

#!/bin/sh
# Purpose {{{
## If Xymon server says that a file is in error on a remote host, try to restart the related service.
## 1. Create a ssh keyring for xymon user {{{
# sudo mkdir -p -- /var/lib/xymon/.ssh/
# sudo ssh-keygen -f /var/lib/xymon/.ssh/id_rsa -N '' -q
# sudo chown -R xymon:xymon /var/lib/xymon/.ssh/
## }}}
## 2. Remote user {{{
# Ensure to have the ${REMOTE_SSH_USER} available on remote hosts and allowed to connect with SSH.
# Restrict the SSH access to a single SSH key from the Xymon server IP (~${REMOTE_SSH_USER}/.ssh/authorized_keys) :
## from="IP.SRV.XYM.ON" ssh-rsa AAAAA…
# Allow sudo commands to restart services (/etc/sudoers.d/xymon-ssh) :
## xymon-ssh ALL=(root:root) NOPASSWD: /bin/systemctl restart *
## }}}

# }}}
# Vars {{{
DEBUG=1

LOCAL_SSH_USER="xymon"
REMOTE_SSH_USER="xymon-ssh"

temp_dir=$(mktemp -d -t xymon-files-alert-XXXXXX.tmp)
debug_stdout="${temp_dir}/debug.stdout"
debug_stderr="${temp_dir}/debug.stderr"
file_list="${temp_dir}/services.error.list"
# }}}

# Create log files
touch "${debug_stdout}" "${debug_stderr}"

# Manage only files probe {{{
if [ "${BBSVCNAME}" = "files" ]; then
	[ "${DEBUG}" -eq "0" ] && printf '\e[1;35m%-6s\e[m\n' "DEBUG : ${BBHOSTNAME} — ${BBSVCNAME} error" >> "${debug_stdout}"
else
	[ "${DEBUG}" -eq "0" ] && printf '\e[1;35m%-6s\e[m\n' "DEBUG : ${BBHOSTNAME} — ${BBSVCNAME} probe is not managed." >> "${debug_stderr}"
	[ "${DEBUG}" -eq "0" ] || rm -rf -- "${temp_dir}"
	exit 0
fi
# }}}

# Check if a file exceeds it's modification time {{{
if echo "${BBALPHAMSG}" | grep -qE "File was modified.*ago - should be.*" ; then
	[ "${DEBUG}" -eq "0" ] && printf '\e[1;35m%-6s\e[m\n' "DEBUG : Test file — Some files exceeds their modification time." >> "${debug_stdout}"
	# First restart Rsyslog service {{{
	[ "${DEBUG}" -eq "0" ] && printf '\e[1;35m%-6s\e[m\n' "DEBUG : Test service — First restart rsyslog service." >> "${debug_stdout}"
	[ "${DEBUG}" -eq "0" ] && printf '\e[1;35m%-6s\e[m\n' "DEBUG : Test service — ssh -n -o StrictHostKeyChecking=no -o UserKnownHostsFile=/dev/null ${REMOTE_SSH_USER}@${BBHOSTNAME} sudo systemctl restart rsyslog.service" >> "${debug_stdout}"
	ssh -n -o StrictHostKeyChecking=no -o UserKnownHostsFile=/dev/null "${REMOTE_SSH_USER}"@"${BBHOSTNAME}" "sudo systemctl restart rsyslog.service" >> "${debug_stdout}" 2>> "${debug_stderr}"
	# }}}
	## Get the list of files path
	## TODO : Set a pattern for both yellow and red colors
	echo "${BBALPHAMSG}" | sed -n 's;^\&yellow <a.*>\(.*\)</a>;\1;p' > "${file_list}"

	while IFS= read -r file_path; do
		[ "${DEBUG}" -eq "0" ] && printf '\e[1;35m%-6s\e[m\n' "DEBUG : while file_path loop — ${file_path} exceeds it's modification time." >> "${debug_stdout}"
		# Match files path and services name {{{
		case "${file_path}" in
			# Cron
			'/var/log/cron.log' )
				service_name="cron"
				;;
			# default
			* )
				service_name="NOT.MANAGED"
				;;
		esac
		# }}}

		# Restart service if needed {{{
		if [ "${service_name}" != "NOT.MANAGED" ] && [ "${service_name}" != "${previous_service_name}" ]; then
			[ "${DEBUG}" -eq "0" ] && printf '\e[1;35m%-6s\e[m\n' "DEBUG : while file_path loop — ${service_name} need to be restarted." >> "${debug_stdout}"
			[ "${DEBUG}" -eq "0" ] && printf '\e[1;35m%-6s\e[m\n' "DEBUG : while file_path loop — ssh -n -o StrictHostKeyChecking=no ${REMOTE_SSH_USER}@${BBHOSTNAME} sudo systemctl restart ${service_name}.service" >> "${debug_stdout}"
			ssh -n -o StrictHostKeyChecking=no -o UserKnownHostsFile=/dev/null "${REMOTE_SSH_USER}"@"${BBHOSTNAME}" "sudo systemctl restart ${service_name}.service" >> "${debug_stdout}" 2>> "${debug_stderr}"
			previous_service_name="${service_name}"
		else
			if [ "${service_name}" = "NOT.MANAGED" ]; then
				[ "${DEBUG}" -eq "0" ] && printf '\e[1;35m%-6s\e[m\n' "DEBUG : while file_path loop — service for ${file_path} is not managed." >> "${debug_stdout}"
			else
				[ "${DEBUG}" -eq "0" ] && printf '\e[1;35m%-6s\e[m\n' "DEBUG : while file_path loop — ${service_name} was already restarted." >> "${debug_stdout}"
			fi
		fi
		# }}}
	done < "${file_list}"

	# Also restart xymon-client service {{{
	[ "${DEBUG}" -eq "0" ] && printf '\e[1;35m%-6s\e[m\n' "DEBUG : Test service — xymon-client also need to be restarted." >> "${debug_stdout}"
	[ "${DEBUG}" -eq "0" ] && printf '\e[1;35m%-6s\e[m\n' "DEBUG : Test service — ssh -n -o StrictHostKeyChecking=no -o UserKnownHostsFile=/dev/null ${REMOTE_SSH_USER}@${BBHOSTNAME} sudo systemctl restart xymon-client.service" >> "${debug_stdout}"
	ssh -n -o StrictHostKeyChecking=no -o UserKnownHostsFile=/dev/null "${REMOTE_SSH_USER}"@"${BBHOSTNAME}" "sudo systemctl restart xymon-client.service" >> "${debug_stdout}" 2>> "${debug_stderr}"
	# }}}
else
	[ "${DEBUG}" -eq "0" ] && printf '\e[1;35m%-6s\e[m\n' "DEBUG : Test file — All files seems up to date." >> "${debug_stdout}"
fi
# }}}

# Remove empty error file
[ -s "${debug_stderr}" ] || rm -f "${debug_stderr}"

# Remove temp_dir if DEBUG is disable
[ "${DEBUG}" -eq "0" ] || rm -rf -- "${temp_dir}"

exit 0