Is there some little known trick when you wan't to use PostgreSQL's advisory locks from Perl?

I first created a function that does:
1. get a lock (exclusive)
2. get some stuff from a table that's not marked as taken, update those rows as taken, and with info what pid took them, with a couple of other things for which I can't lock the rows or their id's and need basically a semaphore
3. unlock

And when I realized that more than one process is ending up with same table row as taken by them (writing to their output table was failing with duplicate key error) - I though, doup, the lock is released (and next process in queue gets the lock) before the transaction is commit-ed and there you go a racing condition.

So I changed it to something like (in reality function call is wrapped in eval, and there is a if($@) ...rollback/commit after that):

my $prepared_lock = $db->prepare("SELECT pg_advisory_lock(?)");
my $prepared_unlock = $db->prepare("SELECT pg_advisory_unlock(?)");

....
$prepared_lock->execute(123);
$db->{AutoCommit} = 1;
$db->begin_work();
$prepared_function_call->execute(....);
$db->commit();
$prepared_unlock->execute(123);
[download]

Well if that was working - I wouldn't be writing here :) Is there something I missed in some document or somewhere? I checked my sanity by trying out the things manually with several psql's open, and from there it's working - second lock is blocked until first is released. I'm in process of trying to $db->do("SELECT pg_ad...."); instead of prepare/execute. And if that doesn't change the way it's working - I'm planning on writing a dead simple test code that will try to get two locks on two separate DB connections. Any suggestions?