Alex Solovey <a.solovey@xxxxxxxxx> writes: > We have pretty big production database (running PostgreSQL 8.3.1) with > many partitioned tables. In most cases, they work well (since 8.2.1 at > least) -- constraint exclusion is able to select correct partitions. > However, there is an exception: queries on partitioned tables using > PostgreSQL 'UPDATE Foo ... FROM Bar' syntax extension. Hmm, the immediate problem is that cost_mergejoin is coming out with a silly cost (NaN) because of division by zero. The attached patch should get it back to 8.2-equivalent behavior. But really we're missing a bet because the sub-joins ought to get discarded entirely when we know they must be empty. There are various places testing for this but it looks like make_join_rel() needs to do it too. regards, tom lane Index: src/backend/optimizer/path/costsize.c =================================================================== RCS file: /cvsroot/pgsql/src/backend/optimizer/path/costsize.c,v retrieving revision 1.191 diff -c -r1.191 costsize.c *** src/backend/optimizer/path/costsize.c 1 Jan 2008 19:45:50 -0000 1.191 --- src/backend/optimizer/path/costsize.c 24 Mar 2008 20:55:42 -0000 *************** *** 1385,1390 **** --- 1385,1396 ---- Selectivity joininfactor; Path sort_path; /* dummy for result of cost_sort */ + /* Protect some assumptions below that rowcounts aren't zero */ + if (outer_path_rows <= 0) + outer_path_rows = 1; + if (inner_path_rows <= 0) + inner_path_rows = 1; + if (!enable_mergejoin) startup_cost += disable_cost; - Sent via pgsql-general mailing list (pgsql-general@xxxxxxxxxxxxxx) To make changes to your subscription: http://www.postgresql.org/mailpref/pgsql-general
