It’s commonplace that if you’ve run services on Linux and wanted something to auto-restart if it crashed, then you’ve looked at monit. You may not know, however, that systemd can provide the ability to restart a failed process without adding another daemon. In this example, we’ll keep Tomcat running if it stops for any reason, unless we
systemctl stop tomcat using a systemd override:
1 2 3 4 5 6 7 8 9 10 11
You don’t need to restart Tomcat; systemd will begin to honor the
Restart setting immediately. You can view all of these values before and after to verify the changes were honored:
1 2 3 4 5
I recommend reading the documentation about each of the options, but a quick explanation:
Restart=always: Unless running
systemctl stop, systemd will always attempt to start the process if it dies for any reason. That means if you were to
pkill -f tomcator any other kill command, regardless of the SIG code, it will be restarted. This would mean you should always be using
systemctlto stop. If you’d like the ability to stop the process another way and only want systemd to restart if the application crashed or exited with a non-zero exit code, you can use
Restart=on-failureor another value. Check the docs for your use-case.
RestartSec=30: First, the documentation above lists
RestartUSec; I try and stick with the docs where I can so you can follow along with the official guidance rather than something I have stumbled upon that may not work in the future. This value is how long to sleep before restarting the service. The default is only
100ms, so I like to set it higher to allow for a busy operation to continue before systemd gets too aggressive with restarting.
TimeoutStopSecinstruct systemd how long to wait for the stop and start commands before considering the operation failed and trying again. Change as needed, but I set it to 4 minutes to allow for our Tomcat applications to completely initialize.
It’s worth noting that you can run the interactive command
systemctl edit tomcat and it will open
/etc/systemd/system/tomcat.service.d/override.conf in your
$EDITOR for adding in the contents above, eliminating the need to remember the directory structure and where to put override directives. I prefer adding files by hand as shown above for two reasons: 1) adding a file directly is obviously easier in a non-interactive server initialization script, and 2) you can have multiple files with the pattern
/etc/systemd/system/name.service.d/*.conf. For organizing purposes, you could have a
restart.conf where you specify your custom Restarting parameters, and another
environment.conf file where you might add more environmental variables.