Re: Add a greedy join search algorithm to handle large join problems

Hi Tomas,

Thank you for the question.
The 15-table and 20-table results I shared were obtained using a synthetic
join workload designed to stress join-order planning and measure
planning-time scaling, rather than a JOB or TPC-H query.
Each query is essentially a left-deep chain of equality joins over simple
tables. For reference, the structure is equivalent to:

15-table join

SELECT count(*)

FROM t1

JOIN t2 ON t1.id = t2.id

JOIN t3 ON t2.id = t3.id

...

JOIN t15 ON t14.id = t15.id;

20-table join

SELECT count(*)

FROM t1

JOIN t2 ON t1.id = t2.id

JOIN t3 ON t2.id = t3.id

...

JOIN t20 ON t19.id = t20.id;

Regarding planner settings:

-geqo_threshold was set to:

a high value (e.g., 100) to force DP

a low value (e.g., 2) to allow GEQO/GOO

-enable_goo_join_search was toggled on/off depending on the comparison
being measured.

-Other planner parameters, including join_collapse_limit, were left at
their default values.

So these experiments mainly evaluate planning-time scaling and basic plan
sanity on a controlled join graph, rather than realistic workload plan
quality.

I’m currently preparing additional tests using selected JOB queries to
provide more meaningful plan-quality comparisons and will share those
results once available.

Regards
Lakshmi