Re: Add 64-bit XIDs into PostgreSQL 15

On Mon, Nov 28, 2022 at 4:53 PM Peter Geoghegan <pg(at)bowt(dot)ie> wrote:
> Imagine if we actually had 64-bit XIDs -- let's assume for a moment
> that it's a done deal. This raises a somewhat awkward question: do you
> just let the system get further and further behind on freezing,
> forever? We can all agree that 2 billion XIDs is very likely the wrong
> time to start refusing new XIDs -- because it just isn't designed with
> debt in mind. But what's the right time, if any? How much debt is too
> much?

I simply don't see a reason to ever stop the server entirely. I don't
even agree with the idea of slowing down XID allocation, let alone
refusing it completely. When the range of allocated XIDs become too
large, several bad things happen. First, we become unable to allocate
new XIDs without corrupting the database. Second, pg_clog and other
SLRUs become uncomfortably large. There may be some other things too
that I'm not thinking about. But these things are not all equally bad.
If these were medical problems, being unable to allocate new XIDs
without data corruption would be a heart attack, and SLRUs getting
bigger on disk would be acne. You don't handle problems of such wildly
differing severity in the same way. When someone is having a heart
attack, an ambulance rushes them to the hospital, running red lights
as necessary. When someone has acne, you don't take them to the same
hospital in the same ambulance and drive it at a slower rate of speed.
You do something else entirely, and it's something that is in every
way much less dramatic. There's no such thing as an attack of acne
that's so bad that it requires an ambulance ride, but even a mild
heart attack should result in a fast trip to the ER. So here. The two
problems are so qualitatively different that the responses should also
be qualitatively different.

> Admittedly this argument works a lot better with the failsafe than it
> does with xidStopLimit. Both are removed by the patch.

I don't think the failsafe stuff should be removed, but it should
probably be modified in some way. Running out of XIDs is the only
valid reason for stopping the world, at least IMO, but it is
definitely NOT the only reason for vacuuming more aggressively.

--
Robert Haas
EDB: http://www.enterprisedb.com