Re: Reducing memory consumption for pending inval messages

I wrote:
> It turns out that the existing implementation in inval.c is quite
> inefficient when a lot of individual commands each register just
> a few invalidations --- but a few invalidations per command is
> pretty typical.

Per the cfbot, here's a rebase over 3788c6678 (actually just
undoing its effects on inval.c, since that code is removed here).

			regards, tom lane