Investigate and fix service and timer leaks.

[joeysk2012]

There are still issues with service and timer leaks when udev rules fail or the unit exits on code 1 for some reason.
This could lead to extra policy-routes@.service or refresh-policy-routes@.timer to remain on the system after the interface has been deleted. policy-routes@.service should no longer be running but rather remain because RemainAfterExit=yes. Leaked refresh-policy-routes@.timer are more sinister because they can process every 60 seconds if not disabled or stopped.

[ddermendzhiev]

Can you provide more details on the scenarios in which you observed the leaks?

[ddermendzhiev]

Also, is there a spec for these units that defines their expected lifecycle? Should units be automatically cleaned up when the iface is removed, even if the udev remove rule fails?

[ddermendzhiev]

For context: #149

[joeysk2012]

1. udev rules starts the policy-routes@.service. Wants=refresh-policy-routes@%i.timer starts the timer inpolicy-routes@.service If the service unit fails out or times out, Restart=on-failure attempts to restart the service, thereby activating the timer again even if the udev remove rule had been triggered sometimes before.
2. The solution is to stop the Restart and remove all of the associated restart properties in policy-routes@.service. It seems like Restart=on-failure was there since the beginning and never careful considered if it should be there or not.
3. Another option would be remove the Wants= and move the enabling of timer to the udev rules. This, way the timer will only be activated by the udev rules and not by Wants=. However, this was the way it was before. Using Wants was added to show a more clear relationship between the two services. don't reconfigure networkd on "stop" #107
4. Another option would be to BindsTo= on the timer to have a strong relationship with the service and timer. When the service disabled, stopped, or restarted, the timer will follow the state of the service. This might be a good alternative option but might be too strong, if someone restarts the service then the timer is restarted at the same time and cause a race conition.