"Ross J. Reedstrom" <reedstrm(at)rice(dot)edu> writes:
> I just tested this by artificially creating a user with duplicate usesysid
> (I had to UPDATE pg_shadow to do it : the CREATE USER command is guarded
> against this) and got exactly the symptoms you report:
FWIW, this sort of problem should be impossible in 7.2 and beyond: we
finally got around to attaching a unique index to pg_shadow.usesysid.
regards, tom lane