In this case, the init script has restart != stop + start. Had systemd called restart it probably would have worked.
On 6/15/16, Felipe Sateler <fsate...@debian.org> wrote: > Control: reassign -1 winbind 2:4.1.17+dfsg-2+deb8u2 > > On 15 June 2016 at 11:52, Joshua Hudson <joshud...@gmail.com> wrote: >> Package: systemd >> Version: 215-17+deb8u4 >> Severity: important >> >> # systemctl restart winbind (to get it to reread its config file) starts >> a >> broken winbind, apparently because ti doesn't ensure the previous copy >> shut down cleanly first. winbind is a multi-process daemon and it looks >> like systemd only kills the root process; this would cause the new copy >> to have trouble starting up. The old processes are visible in ps for a >> long time. >> >> Repeatedly running systemctl restart winbind eventually works; however >> this is a real pain to figure out when winbind is not functioning for >> some other reason. (In this case the clock had wandered out of range.) > > Systemd cannot assume it can kill all processes, unless the service > declares that. Currently winbind does not have a native systemd > service, but an init script. Thus, the init script should ensure > winbind is really exited before returning. > > This is a bug in the winbind init script. Thus reassigning to winbind. > > -- > > Saludos, > Felipe Sateler >