26.07.2024 15:41, Andrew Dunstan wrote:
One way to workaround this is to disable debug_parallel_query in the test
and another I find possible is to set max_parallel_workers = 0.
But wouldn't either of those just be masking the problem?
Yes, I'm inclined to consider this behavior a problem (what if the table
contained 1M rows?), that's why I called those solutions workarounds.
Of course, there are parallel_setup_cost and parallel_tuple_cost
parameters, which can prevent this from happening in the wild, but still...
Best regards.
Alexander
