Thanks for the ideas, I tested them:
The essential question here is when the function is executed, so you should use
EXPLAIN (VERBOSE) to see that.
Append (cost=0.00..5.12 rows=4 width=32)
-> Gather (cost=0.00..1.26 rows=1 width=32)
Output: plpgsql_function(...)
Workers Planned: 8
-> Parallel Seq Scan on public.table t (cost=0.00..1.01 rows=1 width=174)
Output: ...
-> Gather (cost=0.00..1.26 rows=1 width=32)
Output: plpgsql_function(...)
Workers Planned: 8
-> Parallel Seq Scan on public.table t_1 (cost=0.00..1.01 rows=1 width=174)
Output: ...
Possible explanations:
- The function is executed after the "Gather" node.
The question is - could we do something to fix it?
Perhaps you didn't define it as PARALLEL SAFE.
The function is marked as "PARALLEL RESTRICTED” because it’s uses temp tables (and I tested it as PARALLEL SAFE with the same result… parallelisation doesn’t work anyway).
- Perhaps the tables are small.
Yes, but these settings applied when the table is created:
analyze table;
set parallel_setup_cost = 0;
set parallel_tuple_cost = 0;
set force_parallel_mode = on;
alter table table set (parallel_workers = 8);
P.S. Actually, I just need to run in parallel mode one function with a set of different arguments to utilise all available CPUs. That’s strange but I couldn’t google a way to do it.