Re: Postgres undeterministically uses a bad plan, how to convince it otherwise?

No, the planner may not cause execution.  While I could imagine extending EXPLAIN to somehow retrieve and maybe even try alternative plans that have been fully constructed today I'm not holding my breath.

There is little reason for the project to give any real weight to "assuming the query to be immutable".  We do want to fix the planner to behave better if it is mis-behaving, otherwise you do have access to cost parameters, and potentially other planner toggles if you've truly run into an intractable problem.

David J.