> > Any chance the PID filename could be hard linked to the format:
> > postmaster-%d.pid where %d is the start time of the postmaster?
>
> And then how does any program except the postmaster know which .pid
> file is the active one? There's also the little problem that this
> defeats the use of the .pid file as a data-directory lock file.
PostgreSQL writes out postmaster.pid in the data directory then
link()'s it to postmaster-%d.pid. I don't see where the problem is in
that. pg_ctl could be rewritten in C and read the startup time out of
a semaphore, but I don't know how easy it would be to get working on
boxen with multiple copies of postgres running. -sc
--
Sean Chittenden