createSqlPool's idle timeout seems much too short

https://hackage.haskell.org/package/persistent-2.7.3.1/docs/src/Database-Persist-Sql-Run.html#createSqlPool calls createPool with 1 stripe and a 20s idle timeout on resources. I would say that

a) This is much too small

and, to contradict myself somewhat

b) There is no one-size-fits-all.

I would at least bump this up to 5 minutes. Anything but a heavily loaded site essentially negates any benefit of using a pool, constantly forming new connections. I've instrumented the use of a 20s timeout in a pool at work, and while we are constantly serving requests, in the last hour we've opened 160 new database connections.

Methodology

I simply added a Prometheus.incCounter call to the function passed to create in resource-pool.

Here is our traffic (average requests/s with a 1h look back)

Here is a a counter of new database connections (average new connections per second).

And here is a graph showing every time the new connection counter increases:

Start with createSqlPool in Database-Persist-Sql-Run.html and inspect the createPool call, focusing on its one-stripe configuration and 20-second idle timeout. Review the issue's connection-creation measurements and determine a timeout behavior that addresses the reported churn while accounting for the stated lack of a one-size-fits-all setting; done means the agreed behavior is implemented and verified.