Is this the expected way to parallelize?

[ahadnagy]

Hi everyone,

I'm trying to get a few TPC-H-like queries running by building the query plans manually.

For example, TPC-H 13:

auto plan = PlanBuilder()
             .values({tables.get("customer")})
             .hashJoin(
                 {0},
                 {0},
                 PlanBuilder()
                     .values({tables.get("orders")})
                     .planNode(),
                 R"(not REGEXP_LIKE(o_comment, '^special.*?requests'))",
                 {0, 1, 2, 3})
             .singleAggregation({0}, {"count(o_orderkey)"})
             .project({"c_custkey", "a0"}, {"c_custkey", "c_count"})
             .singleAggregation({1}, {"count(1)"})
             .project({"c_count", "a0"}, {"c_count", "custdist"})
             .orderBy({0, 1}, {kDescNullsLast, kDescNullsLast}, false)
             .limit(0, 1, false)
             .planNode();
auto result = readSingleValue(plan);

I tried to parallelize the execution using the CursorParameters:

params.maxDrivers = 16;
params.numResultDrivers = 1;

This appears to be working, although it's on a single core ~95% of the time. params.numResultDrivers = 1; seem fishy to me, but without that the execution hangs.

Is this the correct approach, or am I missing something?
Thanks in advance!