Last tested: Apr 15, 2020
You may notice that Looker servers aren't closing the connection to the internal DB, which causes the open connection count on the DB to rise continuously and hit the DB connection limit.
This is intended behavior. We keep connections around and re-use them to avoid the overhead associated with creating a new connection so this is actually a performance enhancement.
You'll need to make sure their DB can accept number of nodes in a cluster * ~100 number of connections, which is the most connections Looker will try to open.
Not sure if 100 is the actual number, but 100 is a safe number to use.