Re: Proposal : Parallel Merge Join

On Tue, Mar 7, 2017 at 5:16 AM, Dilip Kumar <dilipbalaut(at)gmail(dot)com> wrote:
> I am confused about whether to call
> "get_cheapest_parallel_safe_total_inner" with
> innerrel->cheapest_parameterized_paths like we do in case of
> hash_inner_and_outer or with
> innerrel->pathlist. The reason behind I am calling this with complete
> pathlist (innerrel->pathlist) is that inside generate_mergejoin_paths
> it calls "get_cheapest_path_for_pathkeys" for complete pathlist and if
> we decide not to call consider_parallel_mergejoin if the cheapest
> parallel safe path in innerrel->cheapest_parameterized_paths is NULL
> then it will not be fair (we might have some parallel safe paths in
> innerrel->pathlist).

You're right to be confused, because that seems to be a bug in the
existing code. There seems to be no guarantee that the cheapest
parallel-safe path will be in the cheapest_parameterized_paths list.
I'll go fix that.

As a matter of style, when testing a value of type "bool", write if
(x) or if (!x). When testing a variable of some other type, say int,
write if (x == 0) or if (x != 0) or whatever.

--
Robert Haas
EnterpriseDB: http://www.enterprisedb.com
The Enterprise PostgreSQL Company