On Thu, Aug 22, 2024 at 10:01 PM Tom Lane <tgl@sss.pgh.pa.us> wrote:
Maxim Boguk <maxim.boguk@gmail.com> writes: > Ok there are reproducer test case on the stock postgresql 16.4 config:
Hmm, you can see both behaviors on the small version of t1, just by varying the comparison constant in the WHERE clause. For me, it'll use only one worker with "where a<1", and not parallelize at all with "where a<0". It looks like it's deciding that it's not worth starting workers when too few rows are expected to be returned. That would be unsurprising with a normal setting of parallel_setup_cost, but it does seem odd with parallel_setup_cost=0.
In any case, I think this isn't about the big table being big but about changing the range of values of "a", which changes the selectivity of "where a<10" 100-fold.
(I tested on HEAD not v16)
regards, tom lane
I especially designed both tests in a way that the expected amount of rows - is the same in both cases.
With sufficiently large table - I didn't manage find a combinations of settings to force parallel execution (im my real case - the database huge and mostly cold on ssd raid - so parallel execution provides almost linear speedup with amount of parallel workers even on 1000 rows... not talking about 10k-10m rows, but I cannot convince the PostgreSQL use parallel index scan for 2m returned rows case).
