I've noticed what appears to be index corruption on composite primary key indexes during my testing. Data deletes, updates, and inserts are applied from delta data that is loaded into temporary tables. The duplicate key error occurs at different points in time and isn't isolated to any single table, although all affected tables have a composite primary key index. For example, if machine1 and machine2 both start off from the same starting state, then machine1 might fail on delta 100 while machine2 won't fail until delta 125. The error in the log is "duplicate key value violates unique constraint" with a reference to the composite primary key index. The issue occurs on different machines, so that would seemingly rule out hardware failure. The fact that it only happens on tables with composite primary keys is highly suspicious. It occurs reproducibly enough that it seems to be a pgsql bug, or maybe a pgsql running on a certain set of hardware + software.
Some info on the platform I'm using: PostgreSQL 8.4.4 Red Hat 4.1.2-44 Linux 2.6.18-128.el5 #1 SMP Wed Dec 17 11:41:38 EST 2008 x86_64 x86_64 x86_64 GNU/Linux Intel(R) Xeon(R) CPU E5420 @ 2.50GHz Does anyone know if this a known issue? Any help would be much appreciated.
