Обсуждение: FDW LIM IT pushdown
Hello hackers, First timer here with a question: I’ve searched through various e-mail threads and documents and could not find if pushdown of LIMIT clauses to FDW is implemented. Seen some conversation about that a couple of years ago when that was treated as a feature request but nothing since then. Citus does that according to its documentation but for now we are trying to base our solution on “plain” Postgres. If it is not implemented - could you, guys, provide me with some pointers to source code where I could look at and try toimplement that? Thanks, Michal
Sorry, I wasn’t precise - my question is about foreign partitions - LIMIT on ordinary tables is supported. Thanks, Michal > On 6 Oct 2023, at 18:02, Michał Kłeczek <michal@kleczek.org> wrote: > > Hello hackers, > > First timer here with a question: > > I’ve searched through various e-mail threads and documents and could not find if pushdown of LIMIT clauses to FDW is implemented. > Seen some conversation about that a couple of years ago when that was treated as a feature request but nothing since then. > Citus does that according to its documentation but for now we are trying to base our solution on “plain” Postgres. > > If it is not implemented - could you, guys, provide me with some pointers to source code where I could look at and tryto implement that? > > > Thanks, > Michal
Hi Michal, On Fri, Oct 6, 2023 at 9:34 PM Michał Kłeczek <michal@kleczek.org> wrote: > > Hello hackers, > > First timer here with a question: > > I’ve searched through various e-mail threads and documents and could not find if pushdown of LIMIT clauses to FDW is implemented. > Seen some conversation about that a couple of years ago when that was treated as a feature request but nothing since then. > Citus does that according to its documentation but for now we are trying to base our solution on “plain” Postgres. > > If it is not implemented - could you, guys, provide me with some pointers to source code where I could look at and tryto implement that? > I started looking at code from grouping_planner and reached create_ordinary_grouping_paths(). It calls create_partitionwise_grouping_paths() to push aggregate and grouping down into partitions and from there it is pushed down into FDW. Looking at create_limit_path(), I don't similar treatment. So there may be some cases where we are not pushing down final LIMIT. However create_append_path() uses PlannerInfo::limit_tuples or root->limit_tuples when creating append path node. So either it's being used for costing or for pushing it down to the partitions. This isn't a full answer, but I hope these pointers would help you. -- Best Wishes, Ashutosh Bapat