Added parallel safe keyword to function definitions

[jsanz]

Since Postgres 9.6 the PARALLEL SAFE modifiers for function definition enables the execution of queries using one or more workers in parallel (more details here).

This straight forward PR adds this modifier to this repo functions to avoid them potentially blocking parallel execution when present.

I've done tests on all functions to ensure they don't block parallel executions with the following procedure:

• Created a sample database on a postgres:12 Docker container with Postgis 3 installed
• Created a sample table points with 1M random points and a random kpi integer column
• Imported the layer ne_10m_admin_1_states_provinces from Natural Earth to have also a multipolygon table
• Relaxed the conditions for parallel execution:

[local] postgres@bench=#  set parallel_setup_cost = 10;
SET
Time: 0.322 ms
[local] postgres@bench=# set parallel_tuple_cost = 0.001;
SET

• Run sample queries with the keyword enabled until the parallel execution was activated
• Removed the keyword and run the same query for comparison.

The tests below shared are just meant to showcase that PARALLEL SAFE can help the faster execution of queries and is not an indicator of the improvement introduced since there are many factors that can affect that.

It's worth noting that this makes these functions only available for Postgres 9.6 onwards so maybe we could consider having both versions in separated folders. Let me know if that would make sense to modify the PR that way.

Query examples

For each query, the execution time in milliseconds with and without the PARALLEL SAFE keyword is shared.

MakeArc

explain analyze 
select makearc(geom_lag, geom, geom_lead)
from (
        select lag(geom) over (partition by kpi % 3 order by kpi) geom_lag,
                geom,
                lead(geom) over (partition by kpi % 3 order by kpi) geom_lead
        from points
) a 
union all
select makearc(geom_lag, geom, geom_lead)
from (
        select lag(geom) over (partition by kpi % 4 order by kpi) geom_lag,
                geom,
                lead(geom) over (partition by kpi % 4 order by kpi) geom_lead
        from points
) a;

• With parallel: 1148
• Without parallel: 19395

MercBuffer

explain analyze verbose 
select count(*) 
from points 
where st_area(mercbuffer(geom,kpi)) > kpi^10;

• With parallel: 8942
• Without parallel: 25351

MercDWithin

explain analyze verbose 
select count(*) 
from points 
where
MercDWithin(
        st_transform(geom,3857),
        st_transform(
                ST_SetSRID(
                        ST_MakePoint( -71.1043443253471,42.3150676015829)
                        ,4326)
                ,3857)
        ,100000);

• With parallel: 1222
• Without parallel: 2885

MercLength

explain analyze verbose 
select count(*) 
from points 
where
MercLength(
        ST_MakeLine(
                st_transform(geom,3857),
                st_transform(
                        ST_SetSRID(ST_MakePoint(-71,42),4326)
                        ,3857)
        )
) > kpi^10;

• With parallel: 2043
• Without parallel: 4978

OrientedEnvelope

explain analyze  
select sum(r) 
from
(
        select sum(st_area(orientedenvelope(geom))) as r
        from points 
        group by kpi % 50000
        union all
        select sum(st_area(orientedenvelope(geom))) as r
        from points 
        group by kpi % 70000
) a;

• With parallel: 8727
• Without parallel: 22693

Sieve

explain analyze
select sum(st_area(sieve(geom,0.1)))
from countries
where iso_a2 like 'ES'
union all
select sum(st_area(sieve(geom,0.1)))
from countries
where iso_a2 like 'IT';

• With parallel: 8368
• Without parallel: 12244

SmartShrink

explain analyze
select sum(st_area(smartshrink(geom,0.1)))
from countries where ogc_fid % 7 = 0
UNION ALL
select sum(st_area(smartshrink(geom,0.1)))
from countries where ogc_fid % 5 = 0;

• With parallel: 15129
• Without parallel: 21061

TileBBox

explain analyze
select count(*)
from countries
where st_centroid(geom) && tilebbox(6,28,26)
union
select count(*)
from points
where st_centroid(geom) && tilebbox(6,28,27);

• With parallel: 470
• Without parallel: 879

ToPoint

explain analyze
select count(*)
from countries
where ToPoint(geom) && tilebbox(6,28,26)
union
select count(*)
from points
where ToPoint(geom) && tilebbox(6,28,27);

• With parallel: 3857
• Without: 4948

[nyurik]

awesome research job, thank you @jsanz !!!