Christopher Smith <x@xman.org> writes:
> begin
> lock table foo in access exclusive mode
> select count(*) from foo where key1 = bar, key2 = baz
> if count > 1
> rollback
> else
> insert into foo (key1, key2, other) values (bar, baz, stuff)
> commit
> Now, table foo has a primary key made up of key1 & key2. Without
> concurrent access, the code works great. However, provided I use enough
> threads, I inevitably get back errors from the database indicating that
> I've tried to insert a record which violates the uniquness of the
> primary key.
Um ... surely that should be "if count > 0" ? Or was that just a
transcription error?
This approach certainly ought to work as desired given the exclusive
lock, so a silly typo seems like a plausible explanation...
regards, tom lane