scripts/xymon/xymon.libs.alert.sh

#!/bin/sh
# Purpose {{{
## If Xymon server says that a service is in error on a remote host, try to restart this 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=$(mktemp -d -t xymon-libs-alert-XXXXXX.tmp)
debug_stdout="${temp_dir}/debug.stdout"
debug_stderr="${temp_dir}/debug.stderr"
service_list="${temp_dir}/services.error.list"
# }}}

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

# Manage only libs probe {{{
if [ "${BBSVCNAME}" = "libs" ]; then
	[ "${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 host need to reboot {{{
if echo "${BBALPHAMSG}" | grep -qE "\\&yellow Machine should be reboot.*" ; then
	[ "${DEBUG}" -eq "0" ] && printf '\e[1;35m%-6s\e[m\n' "DEBUG : Test kernel — The host need to be rebooted." >> "${debug_stdout}"
fi
# }}}

# Check if a service need to restart
if echo "${BBALPHAMSG}" | grep -qE "\\&yellow The following processes.*" ; then
	[ "${DEBUG}" -eq "0" ] && printf '\e[1;35m%-6s\e[m\n' "DEBUG : Test service — Some services need to be restarted." >> "${debug_stdout}"
	# Get the list of binaries path
	echo "${BBALPHAMSG}" | sed -n 's/^  \(\/.*\) (.*)/\1/p' > "${service_list}"

	while IFS= read -r bin_path; do
		[ "${DEBUG}" -eq "0" ] && printf '\e[1;35m%-6s\e[m\n' "DEBUG : while bin_path loop — ${bin_path} use old libs." >> "${debug_stdout}"
		# Match binaries path and services name {{{
		case "${bin_path}" in
			# Apache2
			'/usr/sbin/apache2' )
				service_name="apache2"
				;;
			# Arpwatch
			'/usr/sbin/arpwatch' )
				service_name="arpwatch"
				;;
			# Dbus
			'/usr/bin/dbus-daemon' )
				service_name="dbus"
				;;
			# Postfix
			'/usr/lib/postfix/sbin/pickup' | '/usr/lib/postfix/qmgr' | '/usr/lib/postfix/sbin/tlsmgr' | '/usr/lib/postfix/sbin/qmgr' )
				service_name="postfix"
				;;
			# Nginx
			'/usr/sbin/nginx' )
				service_name="nginx"
				;;
			# Netdata
			'/usr/sbin/netdata' )
				service_name="netdata"
				;;
			# Ntp
			'/usr/sbin/ntpd' )
				service_name="ntp"
				;;
			# Systemd-journald
			'/lib/systemd/systemd-journald' )
				service_name="systemd-journald"
				;;
			# Systemd-logind
			'/lib/systemd/systemd-logind' )
				service_name="systemd-logind"
				;;
			# 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 bin_path loop — ${service_name} need to be restarted." >> "${debug_stdout}"
			[ "${DEBUG}" -eq "0" ] && printf '\e[1;35m%-6s\e[m\n' "DEBUG : while bin_path loop — ssh -n -o StrictHostKeyChecking=no ${REMOTE_SSH_USER}@${BBHOSTNAME} sudo systemctl restart ${service_name}.service" >> "${debug_stdout}"
			ssh -n -o StrictHostKeyChecking=no ${REMOTE_SSH_USER}@${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 bin_path loop — service for ${bin_path} is not managed." >> "${debug_stdout}"
			else
				[ "${DEBUG}" -eq "0" ] && printf '\e[1;35m%-6s\e[m\n' "DEBUG : while bin_path loop — ${service_name} was already restarted." >> "${debug_stdout}"
			fi
		fi
		# }}}
	done < "${service_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}"
	# }}}
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