Re: optimize file transfer in pg_upgrade

On Thu, Mar 20, 2025 at 03:23:13PM -0500, Nathan Bossart wrote:
> I'm still aiming to commit this sometime early next week.

Committed. Thanks to everyone who chimed in on this thread.

While writing the attributions, I noticed that nobody seems to have
commented specifically on 0001. The closest thing to a review I see is
Greg's note upthread [0]. This patch is a little bigger than what I'd
ordinarily feel comfortable with committing unilaterally, but it's been
posted in its current form since February 28th, this thread has gotten a
decent amount of traffic since then, and it's not a huge change ("9 files
changed, 96 insertions(+), 37 deletions(-)"). I'm happy to address any
post-commit feedback that folks have. As noted earlier [1], I'm not wild
about how it's implemented, but this is the nicest approach I've thought of
thus far.

I also wanted to draw attention to this note in 0003:

/*
* XXX: The below line is a hack to deal with the fact that we
* presently don't have an easy way to find the corresponding new
* tablespace's path. This will need to be fixed if/when we add
* pg_upgrade support for in-place tablespaces.
*/
new_tablespace = old_tablespace;

I intend to address this in v19, primarily to enable same-version
pg_upgrade testing with non-default tablespaces. My current thinking is
that we should have pg_upgrade also gather the new cluster tablespace
information and map them to the corresponding tablespaces on the old
cluster. This might require some refactoring in pg_upgrade. In any case,
I didn't feel this should block the feature for v18.

[0] https://postgr.es/m/CAKAnmm%2Bi3Q1pZ05N_b8%3DS3B%3DrztQDn--HoW8BRKVtCg53r8NiQ%40mail.gmail.com
[1] https://postgr.es/m/Z9h5Spp76EBygyEL%40nathan

--
nathan