Re: Add new option 'all' to pg_stat_reset_shared()

Hi,

On 2023-11-03 00:55:00 +0530, Bharath Rupireddy wrote:
> On Wed, Nov 1, 2023 at 4:24 AM Michael Paquier <michael@paquier.xyz> wrote:
> >
> > On Tue, Oct 31, 2023 at 04:26:18PM +0900, torikoshia wrote:
> > > Yes, calling pg_stat_reset_shared() for all stats types can do what I wanted
> > > to do.
> > > But calling it with 6 different parameters seems tiresome and I thought it
> > > would be convenient to have a parameter to delete all cluster-wide
> > > statistics at once.
> > > I may be wrong, but I imagine that it's more common to want to delete all of
> > > the statistics for an entire cluster rather than just a portion of it.

Yes, agreed. However, I'd suggest adding pg_stat_reset_shared(), instead of
pg_stat_reset_shared('all') for this purpose.


> > If more flexibility is wanted in this function, could it be an option
> > to consider a flavor like pg_stat_reset_shared(text[]), where it is
> > possible to specify a list of shared stats types to reset?  Perhaps
> > there are no real use cases for it, just wanted to mention it anyway
> > regarding the fact that it could have benefits to refactor this code
> > to use a bitwise operator for its internals with bit flags for each
> > type.

I don't think there is much point in such an API - if the caller actually
wants to delete individual stats, it's not too hard to loop.

But most of the time resetting individual stats doesn't make sense. IMO it was
a mistake to ever add the ability. But that ship has sailed.


> I don't see a strong reason to introduce yet-another API when someone
> can just call things in a loop.

I don't agree at all. That requires callers to know the set of possible values
that stats need to be reset for - which has grown over time. But nearly all
the time the right thing to do is to reset *all* shared stats, not just some.

> I could recollect a recent analogy - a
> proposal to have a way to define multiple custom wait events with a
> single function call instead of callers defining in a loop didn't draw
> much interest.

That's not analoguous - in your example the caller by definition knows the set
of wait events it wants to create. Introducing a batch API wouldn't change
that.  But in case of resetting all stats the caller does *not* inherently
know the set of stats types.

Greetings,

Andres Freund