Justin Pitts <jpitts(at)bplglobal(dot)net> writes:
> My guess is that I am not provoking a 'SI queue overrun'
The 100 temp table creations probably will do that just fine.
> Am I completely off base about how this should be reproducing?
Two points: the session you hope to have crash *must* be in serializable
mode, and the crash would actually happen in the transaction after the
one that's rolled back.
The error doesn't have to be a serialization error, so in principle
you should be able to make it fail with something as simple as
begin;
select 1/0;
rollback;
select * from foo;
as long as the ROLLBACK is done with a prepared statement and you've
forced a SI overrun since the ROLLBACK was prepared.
regards, tom lane