Re: pg_postmaster_reload_time() patch

On Wed, Apr 30, 2008 at 8:16 AM, Alvaro Herrera
<alvherre(at)commandprompt(dot)com> wrote:
> George Gensure escribió:
>
>
> > I've done a quick write up for reload time reporting from the
> > administration TODO. I was a little paranoid with the locking, but
> > didn't want problems to occur with signals on the postmaster and the
> > read side.
>
> I'd say too much -- postmaster runs with signals blocked all the time
> (except during select()) so this is not necessary there.
>
> Regarding the locking on backends, I admit I am not sure if this is
> really a problem enough that you need a spinlock for it. Anyway we tend
> not to use spinlocks too much -- probably an LWLock would be more
> apropos, if a lock is really needed. (A bigger question is whether the
> reload time should be local for each backend, or exposed globally
> through MyProc. I don't think it's interesting enough to warrant that,
> but perhaps others think differently.)
>
> Lastly, I didn't read the patch close enough to tell if it would work on
> both the EXEC_BACKEND case and the regular one.
>
> --
> Alvaro Herrera http://www.CommandPrompt.com/
> The PostgreSQL Company - Command Prompt, Inc.
>

I've reworked the patch in response to comments.

The new function name is pg_conf_load_time()
I'm now using LWLocks only on the backend in order to protect the
PgReloadTime from mid copy reads. This may prove to be unnecessary,
since the code to handle HUPs seems to be executed synchronously on
the backend, but I'll let someone else tell me its safe before
removing it.

-George