Is there a way to detect why the planing is taking this long?
The database is a 9.6.1 with 32GB of shared_buffers, and 1GB of maintanance_work_mem, and machine CPU is below 80% all the time.
What is default_statistics_target set to for your cluster or the involved tables/columns? When raised above the default 100, planning time can increase exponentially it seems, as the planner examines the stats and weighs the many options.