]> granicus.if.org Git - postgresql/commit
Simplify partitioned table creation vs. relcache
authorAlvaro Herrera <alvherre@alvh.no-ip.org>
Wed, 5 Sep 2018 17:36:13 +0000 (14:36 -0300)
committerAlvaro Herrera <alvherre@alvh.no-ip.org>
Wed, 5 Sep 2018 17:36:13 +0000 (14:36 -0300)
commit2fbdf1b38bc54b297e0f885ca97e0c8f5c922e72
tree2b92c726c42be7dbbac6a88cc19d6341d33234de
parentf5a6509bb1ec5222a707205941a40f280f3e6e15
Simplify partitioned table creation vs. relcache

In the original code, we were storing the pg_inherits row for a
partitioned table too early: enough that we had a hack for relcache to
avoid falling flat on its face while reading such a partial entry.  If
we finish the pg_class creation first and *then* store the pg_inherits
entry, we don't need that hack.

Also recognize that pg_class.relpartbound is not marked NOT NULL and
therefore it's entirely possible to read null values, so having only
Assert() protection isn't enough.  Change those to if/elog tests
instead.  This qualifies as a robustness fix, so backpatch to pg11.

In passing, remove one access that wasn't actually needed, and reword
one message to be like all the others that check for the same thing.

Reviewed-by: Amit Langote
Discussion: https://postgr.es/m/20180903213916.hh6wasnrdg6xv2ud@alvherre.pgsql
src/backend/commands/tablecmds.c
src/backend/partitioning/partbounds.c
src/backend/utils/cache/partcache.c