"Matthew T. O'Connor" <matthew@zeut.net> writes:
> Tom Lane wrote:
>> the only full solution will involve backends doing some extra work at
>> subtransaction commit/abort so that they can report properly classified
>> update counts.
> Any guess as to the performance implications?
Pushing some counts from one place to another doesn't seem that
expensive, but it'd be nice to avoid scanning a lot of unrelated
table-stats entries to find the ones that have to be adjusted.
Not sure what it'll take exactly.
Or we could blow it off for the time being. Certainly, getting
things right at the top-transaction level would already be a big
leg up in accuracy from where we are, and I don't think that would
be hard at all.
regards, tom lane