> Ok. I'm going to outline my rationale, and if you're probably
> going to disagree, but I feel compelled to try anyway :-)
Always.
> > In any production environment, you are going to be running
> > postgresql as a service. In that case it's not up to the
> console ctrl
> > handler to handle these events, but the service manager.
> Since it's not
> > a console program, I assume (no, haven't tested) that it
> will/may fail
> > in this cse.
>
> Try it. It won't fail (the multi-threaded port I've got does
> the very same thing).
Hmm. Ok. That changes things a bit.
<snip>
> > Any integration with the Service Control Manager will come later.
> > Depending on how that it done, this might change. I suggest
> we revisit
> > it then.
>
> I'll turn that argument back on you. Let it be fatal, and,
> depending on how the SCM is done, maybe revisit it (which you
> probably won't need to).
>
> Bottom line, this call is "never" going to fail. But, IMNSHO,
> if it did fail, then something is seriously wrong, and I'd
> just as soon as not have that postmaster/backend going
> anywhere near my data, thank you very much.
Ok. I'm not married to the idea of it not being fatal :-) Let's go with
fatal. I'll either update my patchset when I get a chance (probably this
weekend - no more patches from me until then), or I'll submit a new
patch after the current one is in (since it "never" happens, it should
be no problem temporarily).
//Magnus