If you write a program that has thousands of persistent websockets open, you do not want all those thousands of connections to each have their own, persistent, long lived database transaction. You will not like your process listing nor will you be happy about the amount of table and row locking going on. Transactions and connection scope should be short lived, and if you're using a connection pool, you definitely are going for that. As my blog post stated, you'd use a thread pool for the database access part, given that the http-facing side is async.