Re: Issues with ON CONFLICT UPDATE and REINDEX CONCURRENTLY

Hello,

On 2026-Jan-25, Mihail Nikalayeu wrote:

> Hello, Álvaro!
>
> Fixes are in attachment. I think the comment message and comments are good
> enough to explain the changes.

Great, many thanks for this. The commit message looks quite good, but I
decided to rewrite the comment in the code. What do you think of this?

(There's also a typo fix for one of the previous commits)

> Also, the second commit adds syscache for pg_inherites. I am not very
> confident with it, but it feels correct to me.

Hmm, I think a syscache on (inhrelid, inhseqno) is a bit weird. This
might be okay, but I'm not sure, and I don't think we absolutely need
this right now. That is to say, I'm not rejecting this, but I'm not
going to pursue getting it pushed for now.

> Another approach - put information about parents into [relcache] - I
> can rebuild the patch that way.

I think that change would be a larger revamp that I definitely don't
want to touch at this time.

> Also, for the first commit it is possible to create a batched version of
> get_partition_ancestors (with the same snapshot for multiple indexes).

Yeah, I've had such thoughts too, but I'd rather fix the bug in a
reasonably non invasive way (which perhaps we can consider backpatching
in some not-too-distant future); major rearchitecting like that can
happen later without pressure.

--
Álvaro Herrera 48°01'N 7°57'E — https://www.EnterpriseDB.com/