LWN: Comments on "Haas: DO or UNDO - there is no VACUUM"

Haas: DO or UNDO - there is no VACUUM

intgr — Tue, 06 Feb 2018 07:51:04 +0000

> VACUUM (which involves a full table scan from time to time, and that's what hurts with huge tables)

That's not strictly true, there have been many improvements to VACUUM; with recent PostgreSQL versions, VACUUUM only scans those parts of the table heap that has changes. For the details see http://rhaas.blogspot.com.ee/2018/01/the-state-of-vacuum....

Haas: DO or UNDO - there is no VACUUM

oldtomas — Thu, 01 Feb 2018 08:37:42 +0000

Well... yes and no. Traditional (PostgreSQL) VACUUM (which involves a full table scan from time to time, and that's what hurts with huge tables) would be gone under this new scheme.

Records wouldn't move around in the default case (OK, when the new version is too big to fit into the old slot...), which would be lighter on indices.

Still some index cleanup will be needed. As for reclaiming space, it all sits neatly in the log, which isn't scattered across the table and ist thus much easier to throw away.

So it might pretty well work out. Given Robert Haas' (deserved!) reputation, I'm pretty sure it will.

Mind you... write ahead MVCC is so deeply ingrained with PostgreSQL history (remember time travel?) that I'm feeling some nostalgia now :-)

Haas: DO or UNDO - there is no VACUUM

xtifr — Tue, 30 Jan 2018 18:58:25 +0000

"[W]henever possible, we [...]" sounds like "there is no need for VACUUM" may be a *bit* of an exaggeration.

Not that I'm complaining, mind. Sounds like a good approach. At least as an option.