Re: problems with making relfilenodes 56-bits

Hi,

On 2022-10-03 10:01:25 -0700, Andres Freund wrote:
> On 2022-10-03 08:12:39 -0400, Robert Haas wrote:
> > On Fri, Sep 30, 2022 at 8:20 PM Andres Freund <andres(at)anarazel(dot)de> wrote:
> > I thought about trying to buy back some space elsewhere, and I think
> > that would be a reasonable approach to getting this committed if we
> > could find a way to do it. However, I don't see a terribly obvious way
> > of making it happen.
>
> I think there's plenty potential...

I light dusted off my old varint implementation from [1] and converted the
RelFileLocator and BlockNumber from fixed width integers to varint ones. This
isn't meant as a serious patch, but an experiment to see if this is a path
worth pursuing.

A run of installcheck in a cluster with autovacuum=off, full_page_writes=off
(for increased reproducability) shows a decent saving:

master: 241106544 - 230 MB
varint: 227858640 - 217 MB

The average record size goes from 102.7 to 95.7 bytes excluding the remaining
FPIs, 118.1 to 111.0 including FPIs.

There's plenty other spots that could be converted (e.g. the record length
which rarely needs four bytes), this is just meant as a demonstration.

I used pg_waldump --stats for that range of WAL to measure the CPU overhead. A
profile does show pg_varint_decode_uint64(), but partially that seems to be
offset by the reduced amount of bytes to CRC. Maybe a ~2% overhead remains.

That would be tolerable, I think, because waldump --stats pretty much doesn't
do anything with the WAL.

But I suspect there's plenty of optimization potential in the varint
code. Right now it e.g. stores data as big endian, and the bswap instructions
do show up. And a lot of my bit-maskery could be optimized too.

Greetings,

Andres Freund