That presumes that Postgres will always make good decisions based on up-to-date statistics, which is unfortunately not true. The statistics are just a sample ac

That presumes that Postgres will always make good decisions based on up-to-date statistics, which is unfortunately not true. The statistics are just a sample across a subset of your data, collecting things like value frequencies, for a single table. It does not track inter-tabular correlations, which means that it can never truly plan joins optimally because it does not know at the outset how many rows will match any given row in a join. I have hit many cases where a query will use a nested loop and take 5-10 seconds to run where an index scan would have taken just a few milliseconds. Some discussion here: http://archives.postgresql.org/message-id/603c8f070902180908j3ae46774g535d96ece2c90e74@mail.gmail.com http://archives.postgresql.org/message-id/603c8f070902180908...