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

On Mon, Feb 16, 2026 at 4:56 PM lakshmi <lakshmigcdac(at)gmail(dot)com> wrote:

> Hi Tomas,
>
> Thank you for the clarification. You’re right that the synthetic test
> doesn’t strongly constrain join order and, with the default
> join_collapse_limit , the DP comparison was limited.
>
> I’m now preparing follow-up tests using selected JOB queries and adjusted
> planner settings to provide a fair and more realistic comparison. I’ll
> share the results soon.
>

Hi Tomas,

Following up on my previous message, I ran some additional tests
using a few selected JOB queries with DP, GEQO and GOO under the same
planner settings.

Configuration used:

- join_collapse_limit = 100
- from_collapse_limit = 100
- max_parallel_workers_per_gather = 0
- PostgreSQL 19devel with the GOO v5 patch

Here is a summary of the results:

Query DP Plan(ms) DP Exec(ms) GEQO Plan(ms) GEQO Exec(ms) GOO
>> Plan(ms) GOO Exec(ms)
>>
>> -----------------------------------------------------------------------------------------------
>> 12b 4.97 0.16 54.57 0.11 4.18
>> 0.12
>> 24a 88.94 0.21 67.10 0.12 6.26
>> 0.12
>> 26b 68.58 0.20 44.44 0.10 3.90
>> 0.07
>> 29a 1903.9 0.24 14.73 0.22 25.47
>> 0.26
>> 33c 163.25 0.28 52.93 0.18 3.98
>> 0.07
>
>
Overall, GOO consistently shows much lower planning time than DP, and
is comparable to GEQO on larger join queries. Execution times are generally
similar across the planners.

Regards,
Lakshmi