[systemd-devel] Restarting a socket starts the associated service

[Tim Cuthbertson]

On Tue, Mar 25, 2014 at 3:05 PM, Tim Cuthbertson <tim at gfxmonk.net> wrote:
> Hi all,
>
> I've got a socket-activated service, called `myapp-main.service` and
> `myapp-main.socket`. These are part of myapp.target:
>
> $ cat myapp.target
> [Install]
> WantedBy=multi-user.target
>
> $ cat myapp-main.service
> [Unit]
> After=local-fs.target
> After=network.target
> Requires=myapp-main.socket
>
> [Service]
> ExecStart=< .. uninteresting .. >
>
> $ cat user/myapp-main.socket
> [Socket]
> ListenStream=9776
>
> [Unit]
> PartOf=myapp.target
>
> [Install]
> WantedBy=myapp.target
>
> --------------------------
>
> When I install the units, I run:
>
> $ systemctl reenable myapp-main.service myapp-main.socket myapp.target
> $ systemctl daemon-reload
> $ systemctl reload-or-try-restart myapp-main.service myapp-main.socket
> myapp.target
> $ systemctl start myapp.target
>
> This ensures that required units are started, while also ensuring that
> if the service did happen to be running already, it is restarted.
>
> On the first install, myapp-main.socket is made active, but
> myapp-main.service is not (since it's socket activated, and nothing
> uses it). However on the second install it is now running.
>
> It seems to be that if you run `systemctl restart` on a socket that is
> currently listening, it starts the associated service even if it was
> not previously running. Is this intentional? Is there some different
> approach I should use to ensure that:
>
> a) all running units in my list of units that I've installed are
> reloaded / restarted
> b) no services are unnecessarily started
>
> Cheers,
>  - Tim.

I forgot to mention: fedora 20, with systemd 208