+ @MoritzMuehlenhoff from ops who have experience with systemd!

Also note that there is base::service_unit define in ops/puppet which lets you pick the init script based on the current distro, which could ease transition.

one takeaway from T124197: Port Ganglia aggregator setup to systemd is how to handle multiple instances of the same service, like ganglia aggregator. Using systemd instances seems to work well via base::service_unit, one other thing it would be nice to have for operator comfort is being able to manipulate all instances at once when e.g. puppet is disabled.

one way I think to achieve this is to provide a dummy service e.g. SERVICE-instances only for dependency management purposes

[Unit]
Description=SERVICE-instances

[Service]
Type=oneshot
RemainAfterExit=true
ExecStart=/bin/true

and in turn the templated instance unit file would be PartOf=SERVICE-instances and WantedBy=SERVICE-instances. Note that the SERVICE-instances won't be started at boot, it is there just as a convenience as each instance is already ensure=>running and enabled to start at boot by puppet