Re: [COMMITTERS] pgsql: Implement table partitioning.

This check is odd (tablecmds.c ATExecAttachPartition line 13861):

/* Temp parent cannot have a partition that is itself not a temp */
if (rel->rd_rel->relpersistence == RELPERSISTENCE_TEMP &&
attachrel->rd_rel->relpersistence != RELPERSISTENCE_TEMP)
ereport(ERROR,
(errcode(ERRCODE_WRONG_OBJECT_TYPE),
errmsg("cannot attach a permanent relation as partition of temporary relation \"%s\"",
RelationGetRelationName(rel))));

This seems to work (i.e. it's allowed to create a temp partition on a
permanent parent and not vice-versa, which you'd think makes sense) but
it's illusory, because if two sessions try to create temp partitions for
overlapping values, the second session fails with a silly error message.
To be more precise, do this in one session:

CREATE TABLE p (a int, b int) PARTITION BY RANGE (a);
CREATE TEMP TABLE p1 PARTITION OF p FOR VALUES FROM (0) TO (10);

then without closing that one, in another session repeat the second
command:

alvherre=# CREATE TEMP TABLE p1 PARTITION OF p FOR VALUES FROM (0) TO (10);
ERROR: partition "p1" would overlap partition "p1"

which is not what I would have expected.

Maybe there are combinations of different persistence values that can be
allowed to differ (an unlogged partition is probably OK with a permanent
parent), but I don't think the current check is good enough.

--
Álvaro Herrera https://www.2ndQuadrant.com/
PostgreSQL Development, 24x7 Support, Remote DBA, Training & Services