On Wed, Jan 27, 2021 at 2:13 PM Hou, Zhijie <houzj.fnst@cn.fujitsu.com> wrote:
>
> Hi,
>
> When testing the patch with the following kind of sql.
>
> ---
> Insert into part_table select 1;
> Insert into part_table select generate_series(1,10000,1);
> Insert into part_table select * from testfunc();
> ---
>
> we usually use these sqls to initialize the table or for testing purpose.
>
> Personally I think we do not need to do the parallel safety-check for these cases,
> because there seems no chance for the select part to consider parallel.
>
> I thought we aim to not check the safety unless parallel is possible.
> , So I was thinking is it possible to avoid the check it these cases ?
>
> I did some quick check on the code, An Immature ideal is to check if there is RTE_RELATION in query.
> If no we do not check the safety-check.
>
> I am not sure is it worth to do that, any thoughts ?
>
Yes, I think it's worth it. It's surprising that there's not really
any optimizations for these with just the current Postgres parallel
SELECT functionality (as there's currently no way to divide the work
for these amongst the workers, even if the function/expression is
parallel-safe).
For the additional parallel-safety checks for INSERT, currently we
check that RTE_SUBQUERY is in the range-table. So I think we can
additionally check that RTE_RELATION is in the subquery range-table
(otherwise treat it as unsafe).
Regards,
Greg Nancarrow
Fujitsu Australia
