On 21/03/2016 14:55, Thor Lancelot Simon wrote: > On Mon, Mar 21, 2016 at 02:30:43PM +0000, Roy Marples wrote: >>> >>> In fact, what you really want is just the guts of shlock(1), specifically >>> the shlock -p behavior. I can't see any reason why pidfile shouldn't just >>> do exactly what shlock -p does, except with a C rather than a shell >>> interface. >> >> No, that's exactly what I don't want because if the program crashes the >> pidfile will persist. It will also exist with my proposal, but the > > But the process won't, so the pidfile can be safely removed after the new > copy of the daemon confirms that, and you cannot race to create it because > of the semantics of link(2).
foo starts as PID 10, links /var/foo and writes 10 to it. foo gets SIGSEGV bar then starts as PID 10, links /var/bar and writes 10 to it. foo now wants to restart, but thinks it's already running because it read /var/foo and PID 10 exists - but it's not foo, it's bar. > Please do not introduce a daemon just to service pidfile. And please bear > in mind that locking over NFS is both unreliably implemented and imperfect > by design. How else would you solve your need to work on NFS and my need to lock and handling crashing programs reliably? Roy
