Re: recovering from "found xmin ... from before relfrozenxid ..."

Ashutosh Sharma <ashu(dot)coek88(at)gmail(dot)com> writes:
> On Wed, Sep 16, 2020 at 1:25 AM Tom Lane <tgl(at)sss(dot)pgh(dot)pa(dot)us> wrote:
>> * Should any of the other tables in the test be converted to temp?

> Are you trying to say that we can achieve the things being done in
> test-case 1 and 2 by having a single temp table and we should aim for
> it because it will make the test-case more efficient and easy to
> maintain?

Well, I'm just looking at the comment that says the reason for the
begin/rollback structure is to keep autovacuum's hands off the table.
In most if not all of the other places where we need that, the preferred
method is to make the table temp or mark it with (autovacuum = off).
While this way isn't wrong exactly, nor inefficient, it does seem
a little restrictive. For instance, you can't easily test cases that
involve intentional errors.

Another point is that we have a few optimizations that apply to tables
created in the current transaction. I'm not sure whether any of them
would fire in this test case, but if they do (now or in the future)
that might mean you aren't testing the usual scenario for use of
pg_surgery, which is surely not going to be a new-in-transaction
table. (That might be an argument for preferring autovacuum = off
over a temp table, too.)

Like I said, I don't have a big problem with leaving the rest of the
test as-is. It just seems to be doing things in an unusual way for
no very good reason.

regards, tom lane