On Wed, Jul 11, 2018 at 03:49:59PM +0530, amul sul wrote: > On the master head, getConstraints() function skips FK constraints for > a partitioned table because of tbinfo->hastriggers is false. > > While creating FK constraints on the partitioned table, the FK triggers are > only > created on leaf partitions and skipped for the partitioned tables.
Oops. Good catch. > To fix this, either bypass the aforementioned condition of getConstraints() or > set pg_class.relhastriggers to true for the partitioned table which doesn't > seem > to be the right solution, IMO. Thoughts? Changing pg_class.relhastriggers is out of scope because as far as I know partitioned tables have no triggers, so the current value is correct, and that would be a catalog change at this stage which would cause any existing deployments of v11 to complain about the inconsistency. I think that this should be fixed client-side as the attached does. I have just stolen this SQL set from Alvaro to check the consistency of the dumps created: create table prim (a int primary key); create table partfk (a int references prim) partition by range (a); create table partfk1 partition of partfk for values from (0) to (100); create table partfk2 partition of partfk for values from (100) to (200); Thoughts? -- Michael
diff --git a/src/bin/pg_dump/pg_dump.c b/src/bin/pg_dump/pg_dump.c
index 463639208d..74a1270169 100644
--- a/src/bin/pg_dump/pg_dump.c
+++ b/src/bin/pg_dump/pg_dump.c
@@ -7131,7 +7131,12 @@ getConstraints(Archive *fout, TableInfo tblinfo[], int numTables)
{
TableInfo *tbinfo = &tblinfo[i];
- if (!tbinfo->hastriggers ||
+ /*
+ * For partitioned tables, foreign keys have no triggers so they
+ * must be included anyway in case some foreign keys are defined.
+ */
+ if ((!tbinfo->hastriggers &&
+ tbinfo->relkind != RELKIND_PARTITIONED_TABLE) ||
!(tbinfo->dobj.dump & DUMP_COMPONENT_DEFINITION))
continue;
signature.asc
Description: PGP signature
