2019年1月9日(水) 13:18 Kyotaro HORIGUCHI <horiguchi.kyot...@lab.ntt.co.jp>: > > At Sun, 30 Dec 2018 12:31:22 +0900, Kohei KaiGai <kai...@heterodb.com> wrote > in <CAOP8fzY1Oqf-LGdrZT+TAu+JajwPGn1uYnpWWUPL=2liatt...@mail.gmail.com> > > 2018年12月30日(日) 4:12 Tom Lane <t...@sss.pgh.pa.us>: > > On the other hands, the later hook must be dedicated to add regular paths, > > and also provides a chance for extensions to manipulate pre-built path-list > > including Gather-path. > > As long as I know, pg_hint_plan uses the set_rel_pathlist_hook to enforce > > a particular path-node, including Gather-node, by manipulation of the cost > > value. Horiguchi-san, is it right? > > Mmm. I haven't expected that it is mentioned here. > > Actually in the hook, it changes enable_* planner variables, or > directory manipuraltes path costs or even can clear and > regenerate the path list and gather paths for the parallel > case. It will be happy if we had a chance to manpurate partial > paths before genrating gahter paths. > So, is it sufficient if set_rel_pathlist_hook is just relocated in front of the generate_gather_paths? If we have no use case for the second hook, here is little necessity to have the post_rel_pathlist_hook() here. (At least, PG-Strom will use the first hook only.)
Thanks, -- HeteroDB, Inc / The PG-Strom Project KaiGai Kohei <kai...@heterodb.com>