Re: In-placre persistance change of a relation

I think there are some good ideas here. I started to take a look at the
patches, and I've attached a rebased version of the patch set. Apologies
if I am repeating any discussions from upthread.

First, I tested the time difference in ALTER TABLE SET UNLOGGED/LOGGED with
the patch applied, and the results looked pretty impressive.

before:
postgres=# alter table test set unlogged;
ALTER TABLE
Time: 5108.071 ms (00:05.108)
postgres=# alter table test set logged;
ALTER TABLE
Time: 6747.648 ms (00:06.748)

after:
postgres=# alter table test set unlogged;
ALTER TABLE
Time: 25.609 ms
postgres=# alter table test set logged;
ALTER TABLE
Time: 1241.800 ms (00:01.242)

My first question is whether 0001 is a prerequisite to 0002. I'm assuming
it is, but the reason wasn't immediately obvious to me. If it's just
nice-to-have, perhaps we could simplify the patch set a bit. I see that
Heikki had some general concerns with the marker file approach [0], so
perhaps it is at least worth brainstorming some alternatives if we _do_
need it.

[0] https://postgr.es/m/9827ebd3-de2e-fd52-4091-a568387b1fc2%40iki.fi

--
Nathan Bossart
Amazon Web Services: https://aws.amazon.com