On Sat, Nov 24, 2018 at 10:18:17AM +0900, Michael Paquier wrote: > On Sat, Nov 24, 2018 at 12:32:36PM +1300, David Rowley wrote: >> On Fri, 23 Nov 2018 at 17:03, David Rowley <david.row...@2ndquadrant.com> >> wrote: >> > Here's a patch for that. Parking here until January's commitfest. >> >> And another, now rebased atop of de38ce1b8 (which I probably should >> have waited for).
Thanks for the patch, David. I can see that this patch makes the code more consistent for partitioned tables and partitioned indexes when it comes to tablespace handling, which is a very good thing. -ATExecPartedIdxSetTableSpace(Relation rel, Oid newTableSpace) +ATExecSetTableSpaceNoStorage(Relation rel, Oid newTableSpace) NoStorage looks strange as routine name for this case. Would something like ATExecPartedRelSetTableSpace be more adapted perhaps? + else if (stmt->partbound) + { + RangeVar *parent; + Relation parentrel; + + /* + * For partitions, when no other tablespace is specified, we default + * the tablespace to the parent partitioned table's. + */ Okay, so the direct parent is what counts, and not the top-most parent. Could you add a test with multiple level of partitioned tables, like: - parent is in tablespace 1. - child is created in tablespace 2. - grandchild is created, which should inherit tablespace 2 from the child, but not tablespace 1 from the parent. In the existing example, as one partition is used to test the top-most parent and another for the new default, it looks cleaner to create a third partition which would be itself a partitioned table. -- Michael
signature.asc
Description: PGP signature