Consider renaming cascade**d**.service to cascade.service?

[maertsen]

This is feedback from dogfooding Cascade internally, going only by public documentation.

The thing that tripped me up twice, last week (when systemctl enable cascade.service), and just now once again (journalctl -u cascade.service), is that the systemd service is called cascaded.service (matching the daemon process), and not cascade.service (like the name of the project). I understand why it’s named that way, but somehow it goes against what I’m used to.

I think that's primarily to do with how common systemd services I use follow the .service scheme. One example is ssh.service starting sshd. I'm sure there are counterexamples, just reporting my experience.

I am curious about the preferences of others, but I would be in favor of renaming the .service file to cascade.service.

[ximon18]

I think this makes a lot of sense: rename the service to -d but keep the daemon binary name with d.

[maertsen]

counterexample provided by Philip: BIND9 on Debian ships a named.service.

My personal perspective is that's one more thing to remember, which we can avoid. But the counter-perspective is to have the .service match the process name.

[jpmens]

Responded to on the fediverse, but here for posterity:

I too would welcome the change, and I find "cascade" more logical, as that what the software is called.

It would certainly not inconvenience me, as I build from source at the moment, and run it under supervisord so no problem here!

The "named" counterargument is, well, yes, but then on some systems ISTR the service is called 'bind9', so there's that?

[pemensik]

I think d suffix is kind of unwanted with systemd and .service suffix. Bind9 had name named since early sysV scripts. I would recommend using cascade.service as a primary documented way. You can use Alias=cascaded.service in the unit, so both ways will work. Because you don't have historical need to keep old service name, I would recommend to omit d letter. After all in systemd most of services are daemons.

Is it pure coincidence that both nsd and unbound end with d, so there is no such dilemma? In theory, it should be named unboundd. I am glad it is not.

[pemensik]

Btw. On Fedora rawhide named.service is accessible by bind.service too. The plan is to rename it to bind9.service too, like Debian has it. But primary reason is we have multiple packages providing named.service and it is done to not conflict with it.

[maertsen]

Thanks all! We've decided to rename and not add an Alias. To be continued!