Improve RDS connection failure.

In the event of the RDS having an outage, the DB wrapper should be able to handle the connection drop better.
In a recent RDS connectivity outage, the object cache had stored some junk data in the cache, causing a failed connection to be cached? essentially the result was that all requests to infinitely redirect. This slipped under our radar as none of our monitors picked up on these, as the traffic was marked as 300s, and no other alarms were triggered. We can enhance our monitoring too, but ultimately the object cache shouldn't have stored this failed connection.