On 2025-Apr-01, Antonin Houska wrote:
> Besides that, it occurred to me that 0005 ("Preserve visibility
> information of the concurrent data changes.") will probably introduce
> significant overhead. The problem is that the table we're repacking is
> treated like a catalog, for reorderbuffer.c to generate snapshots that
> we need to replay UPDATE / DELETE commands on the new table.
>
> contrib/test_decoding can be used to demonstrate the difference
> between ordinary and catalog tables:
>
> [.. ordinary ..]
> Execution Time: 3521.190 ms
> [.. catalog ..]
> Execution Time: 6561.634 ms
Significant indeed. Thinking about the scenarios in which I envision
people using REPACK CONCURRENTLY (mostly, cases where very large tables
have accumulated considerable amounts of bloat) and considering the size
of the patch, I think the case for treating it as concurrent-safe is not
credible, at least not at this stage -- not only because of this
performance impact, but also because of the additional code complexity,
which I'm really doubtful we can address at this stage. I would suggest
to put that patch aside for now, maybe with a doc warning that
"repacking a table would cause visibility information to be lost"; and
then address that aspect later on, after this feature has gone through
some battle-hardening.
--
Álvaro Herrera PostgreSQL Developer — https://www.EnterpriseDB.com/
"Las navajas y los monos deben estar siempre distantes" (Germán Poo)
