I have a systemd service unit with OnCalendar timer assigned to it. That service invokes JVM, which runs code, that has to be started only on a specific schedule.
This systemd unit (and several others) are managed by the ansible playbook, which uses openstack/ansible-role-systemd_service role. It restarts systemd services every time I make some changes in their unit files.
Also, there's a handler, which restarts all the services every time Gitlab CI pipeline runs this ansible playbook and uploads a new .jar artifact to a server.
Is there any systemd built-in mechanism to restrict restarting service with timer assigned to it? Still, I have to be able to stop/start those services manually when it's needed.
Seems like I can fork that ansible role and add some check, but I wonder for some native and more straight-forward way to achieve this.
Seems I've found the way:
Create and don't enable .service unit, that would contain next:
[Unit]
After = network-online.target
...
[Service]
Type = simple
...
SuccessExitStatus = 143
RestartForceExitStatus = 143
Restart = on-failure
RemainAfterExit = False
StandardOutput = null
StandardError = syslog
[̶I̶n̶s̶t̶a̶l̶l̶]
W̶a̶n̶t̶e̶d̶B̶y̶ ̶=̶ ̶m̶u̶l̶t̶i̶-̶u̶s̶e̶r̶.̶t̶a̶r̶g̶e̶t̶
Create and enable .timer unit, that would contain next:
[Timer]
O̶n̶B̶o̶o̶t̶S̶e̶c̶ ̶=̶ ̶0̶m̶i̶n̶
OnCalendar = *-*-* 18:00:00
Persistent = yes
[Install]
WantedBy=multi-user.target
A little explanation:
After = network-online.target
Needed to make sure that service would start only after getting online.
Restart = on-failure
SuccessExitStatus = 143
RestartForceExitStatus = 143
Needed to be able to kill the main process and make systemd restart the service, or stop the service w/o systemd marking it failed. It's because JVM returning 143 exit code when the main process receives SIGTERM and stops gracefully.
RemainAfterExit = False
Needed for .timer to understand that .service stoped. Either way, .timer would hang in the running state, and won't run service next time.
StandardOutput = null
StandardError = syslog
Needed to stop the main process from writing its INFO log to syslog, and write there only ERRORs.
[̶I̶n̶s̶t̶a̶l̶l̶]
W̶a̶n̶t̶e̶d̶B̶y̶ ̶=̶ ̶m̶u̶l̶t̶i̶-̶u̶s̶e̶r̶.̶t̶a̶r̶g̶e̶t̶
Omitted cause the service has to be disabled. Either way, it would be stared on system boot.
Persistent = yes
Needed to be able to start service later that day after system boot, if the target time was missed.
O̶n̶B̶o̶o̶t̶S̶e̶c̶ ̶=̶ ̶0̶m̶i̶n̶
Omitted cause OnCalendar was used.
All those changes seem to have done the trick.