On Sun, Feb 05, 2012 at 04:46:53AM -0800, Doug Barton wrote: > On 02/05/2012 01:59, Pawel Jakub Dawidek wrote: > > > I seem to miss positives of the other approach. Leaving stale PIDs in > > pidfile is something we should avoid at all costs, so recommending that > > in the manual page is not the best recommendation. > > Which is worse ... potentially stale pidfiles that get cleaned up at > every boot, or stale directories that never do?
Every boot might be very rare situation on servers. Those directories should be cleaned when application is deinstalled and not when process exits. > I'm also not sure why you think this method will leave behind a stale > pidfile. The idea is that the pidfile is pre-created with the ownership > that daemon is going to su to, for the express purpose of allowing it to > delete the pidfile when the process exits. If you're saying that this > method doesn't work then please point out the problem ASAP because > numerous ports rc.d scripts do this now. Great, but this is not how UNIX permissions work. To remove directory entry you have to have rights to modify the directory. Having write permission to file within the directory won't allow you to remove it. -- Pawel Jakub Dawidek http://www.wheelsystems.com FreeBSD committer http://www.FreeBSD.org Am I Evil? Yes, I Am! http://tupytaj.pl
pgpBzEtwkYHv3.pgp
Description: PGP signature
