Centuries ago, Nostradamus foresaw when tgl@sss.pgh.pa.us (Tom Lane) would write:
> Since there's no performance difference at pg_dump time, I can't see any
> advantage to freezing your decision then.
This parallels the common suggestion of throwing an ANALYZE in at the
bottom of a pg_dump script. On that particular note, I'd think it
preferable to analyze after loading each table, since the data for the
specific table will still be in memory. But that's a _bit_ of a
change of subject.
This looks like something where a "hook" would be valuable such that
there is something in the pg_dump that can be configured AFTER the
fact to control how it's loaded.
It would surely seem valuable to have a way of making loads go As Fast
As Possible, even with the possibility of "breakneck speed" offering
the possibility of actually getting seriously injured (breaking one's
neck?). If the hardware fails during the recovery, consider that you
were _recovering_ from a _backup_; that surely ought to be an
eminently redoable operation, quite unlike accepting a random SQL
request from a user.
I have done some "recoveries" recently (well, more precisely,
"installs") by taking a tarball of a pre-existing database and
dropping it into place. I had no problem with the fact that if my
hand slipped and hit ^C at the wrong moment ("quelle horreur!"), I
would be forced to restart the "cd $TARGETDIR; tar xfvz Flex.tgz"
process.
I would be pretty "game" for a near-single-user-mode approach that
would turn off some of the usual functionality that we knew we didn't
need because the data source was an already-committed-and-FK-checked
set of data.
--
output = reverse("ac.notelrac.teneerf" "@" "454aa")
http://www.ntlug.org/~cbbrowne/spiritual.html
"Another result of the tyranny of Pascal is that beginners don't use
function pointers." --Rob Pike