> To my taste this is worse, because it breaks the direct filiation,
>which means "type=simple" isn't really true anymore. Good on systemd to
>accept a different MAINPID even with type=simple, but having the daemon
>run as a grandchild of the supervisor when it doesn't have to feels
>more hackish than accepting a notification from another process.
>
> Honestly, I'd rather document in the source that the NotifyAccess option
>should be changed. The default seems very unnecessarily restrictive.
Nyeh. The pid doesn't matter here because the supervisor is pid 1 so
it will get a SIGCHLD anyway if the intermediary process is dead. Still,
I feel like it's cleaner to have the long-running process have the same
pid as the ExecStart=, the same way you'd keep the same pid as ./run
under s6. NotifyAccess= sounds like it's configurable precisely to allow
shenanigans like sdnotify-wrapper, so I don't see a reason not to use
it.
--
Laurent
Received on Wed Jun 05 2024 - 20:11:12 CEST