Re: Revisiting {CREATE INDEX, REINDEX} CONCURRENTLY improvements

On Sun, 9 Nov 2025 at 19:02, Mihail Nikalayeu <mihailnikalayeu(at)gmail(dot)com> wrote:
>
> Hello!
>
> This is a rebased version.
>
> Also I decided to keep only part 3 for now, because we need some
> common solution to keep the horizon advance for both INDEX and REPACK
> operations [0].

I'm not sure a complete and common approach is that easy between CIC
and REPACK CONCURRENTLY.

Specifically, indexes don't need to deal with the exact visibility
info of a tuple, and can let VACUUM take care of any false positives
(now-dead tuples), while REPACK does need to deal with all of that
that (xmin/xmax/xcid). Considering that REPACK is still going to rely
on primitives provided by logical replication, it would be not much
different from reducing the lifetime of the snapshots used by Logical
Replication's initial sync, and I'd rather not have to wait for that
to get implemented.

The only thing I can think of that might be shareable between the two
is the tooling in heapscan to every so often call into a function that
registers a new snapshot, but I think that's a comparatively minor
change on top of what was implemented for CIC, one that REPACK can
deal with on its own.

Kind regards,

Matthias van de Meent
Databricks (https://www.databricks.com)