Daniel Kahn Gillmor wrote: > On Tue 2017-07-11 21:45:22 -0400, Simon Deziel wrote: > > Having a PID file seems to be the easiest way to make both systemd and > > SysV happy. > > fwiw, i think that systemd would be happier without a pidfile. using a > pidfile introduces a failure mode (as we see here) that is entirely > unnecessary in a system with proper process supervision. > > The fact that systemd *can* do something about a pidfile doesn't mean > that anyone should prefer it. I really hope we can drop pidfiles where > we don't need them.
Unfortunately, we have to continue supporting the sysvinit script (the only plausible "compelling reason" to remove the sysvinit script per #746715 I can think of would be the removal of sysvinit itself from the archive), and the only way to configure Unbound's 'pidfile' parameter is via the config. So we either need to configure systemd to delete Unbound's pidfile, or we could develop and contribute a patch upstream that allows overriding the pidfile via the command-line. Then we could disable the pidfile by default (for modern init systems), and enable the pidfile via the command-line (for sysvinit) in the init script. -- Robert Edmonds edmo...@debian.org