Dmitriy Sarafannikov <dsarafannikov@yandex.ru> writes:
>> Maybe we need another type of snapshot that would accept any
>> non-vacuumable tuple. I really don't want SnapshotAny semantics here,
> If I understood correctly, this new type of snapshot would help if
> there are long running transactions which can see this tuples.
> But if there are not long running transactions, it will be the same.
> Am i right?
Right. You haven't shown us much about the use-case you're concerned
with, so it's not clear what's actually needed.
> And what about don’t fetch actual min and max values from indexes
> whose columns doesn’t involved in join?
We don't fetch that info unless we need it.
I'm not entirely certain, but there could be cases where a single
planning cycle ends up fetching that data more than once. (There's
caching at the RestrictInfo level, but that might not be enough.)
So a line of thought that might be worth looking into is adding a
lower layer of caching to make sure it's not done more than once per
plan. Again, whether this saves anything would depend a lot on
specific use-cases.
regards, tom lane