Re: increased max_parallel_workers_per_gather results in fewer workers? - Mailing list pgsql-performance

On 4. Jun 2020, at 20:37, Philip Semanchuk <philip@americanefficient.com> wrote:

[...]

This brings up a couple of questions —
1) I’ve read that this is Postgres’ formula for the max # of workers it will consider for a table —

 max_workers = log3(table size / min_parallel_table_scan_size)

Does that use the raw table size, or does the planner use statistics to estimate the size of the subset of the table that will be read before allocating workers?

"table size" is the number of PSQL pages, i.e. relation-size / 8 kB. This comes from statistics.

OK, so it sounds like the planner does *not* use the values in pg_stats when planning workers, true?

I’m still trying to understand one thing I’ve observed. I can run the query that produced the plan in the gist I linked to above with max_parallel_workers_per_gather=6 and the year param = 2018, and I get 6 workers. When I set the year param=2022 I get only one worker. Same tables, same query, different parameter. That suggests to me that the planner is using pg_stats when allocating workers, but I can imagine there might be other things going on that I don’t understand. (I haven’t ruled out that this might be an AWS-specific quirk, either.)