A project I'm currently working on requires a queueing solution. I'm of the opinion that you should start small and go from there-- so naturally I decided on RPC'ing into a singleton service. I knew this wouldn't scale, and recently I decided it was time to replace this kludge.
But. I don't want to maintain a dedicated queue in my cluster just yet. So in the vein of my original implementation, I decided to build the smallest implementation of Queueing I could think of. The implementation relies on row-level locking in Postgres, as described here.
Nothing here is new, but the idea is simple enough that I decided it was worth doing myself rather than depending on something new.