Andres Freund <[email protected]> writes:
> On 2025-12-14 23:13:49 +0300, Александр Асафов wrote:
>> I have a feature suggestion to optimize the creation of a partial
>> index. In some cases, it is possible to skip a full table scan. For
>> example:
>> ...
>> How complex would a patch with this optimization be? Will it be
>> necessary to add any attributes to the table metadata, or can we check
>> that the previous command was CREATE TABLE when creating an index?
> I doubt this is feasible, or at least that it's going to be worth the
> complexity. You'd have to make sure that the transaction didn't actually
> insert any rows, and update some in-memory or catalog state if it did. With
> subtransactions etc that could get somewhat complicated.
Even if it could be made to work, I think it likely would be a net
negative for performance. You're suggesting that we add mechanism to
track whether any insertion/update has happened since an ALTER ADD
COLUMN, which means new overhead for every INSERT and UPDATE. Maybe
not a huge amount of overhead, but probably not trivial. That
overhead would only be repaid in the rather unlikely case where a
suitable index gets added soon after the ALTER. Unless you can find
some substantially-more-common scenarios where the overhead would
produce a benefit, it's hard to believe we'd accept such a patch.
regards, tom lane
