> If you're using SSD storage, or if the DB is small compared with shared_buffers or RAM, then random_page_cost should be closer to seq_page_cost. I don't *think* we are using SSDs but I'll need to confirm that though. > How large are the indexes? problem_id_idx1 ? Using the query from here: https://wiki.postgresql.org/wiki/Index_Maintenance#Index_size.2Fusage_statistics Output here: https://gist.github.com/indy-singh/e33eabe5cc937043c93b42a8783b3bfb I've setup a repo here where it is possible to reproduce the weird behaviour I'm getting:- https://github.com/indy-singh/postgres-duplicate-where-conditon That contains the data (amended to remove any private information) as well as the statements need to recreate tables, indices, and constraints, I think after some trial and error this is something to do with the size of the table and statistics. I've been trying to put together a Short, Self Contained, Correct example (http://sscce.org/) and the problem only appears when fill problem_instance.message with junk, but I have to do it in two steps as outlined in the README in repo. Indy
