On Thu, Oct 09, 2025 at 04:04:46PM +0100, Mircea Cadariu wrote:
> Did you mean just the tests, without the fix? If so, please find the updated
> tests only patch attached.
The CI has reported that this test returns incorrect results when
using debug_parallel_query = regress, which is the case of the freebsd
task. Forcing debug_parallel_query = off in the test is able to cool
things down.
I did not really get the use of wait_for_log(). The query sequences
are well-timed, hence a set of log_contains() should be enough with
a correct offset to point to a position in the logs.
Instead of SELECT 1 and always using 4999 in the OFFSET clause, let's
use more patterns, like queries with text values related to the
sequence run, or dedicated numbers. Such changes make debugging
easier.
It's a bit late here, and I am not completely done yet. Here is for
now what I have for the tests. I still need to check more its
stability with more runs, and also I need to do a second pass. The CI
looks stable here at least.
--
Michael