I'll be honest, seeing this posted took the wind out of my sails a bit because I was really looking forward to experimenting with this, but there's so much other work that needs to be done anyway that I'm not going to get too hung up on it. I appreciate the initiative.

However, I don't have the time or energy to review this properly while trying to get 0.9.0 out the door so I think ideally we could land this in 0.9.1 or 0.9.2 since it should be backwards-compatible anyway.

Right, I learned a lot while working on this. I mean there are many ways to solve this problem, if you'd like you can still experiment and see what solution you like best.

I can imagine that this takes some effort to review and don't mind when this is released, I'll rebase when needed.

I know I'm getting way ahead of myself here but I just want to write this down somewhere: I'd be nice if the pool could benefit from query pipelining.

Instead of acquiring a conn -> send query -> wait for results -> release the conn. We could choose a conn and not acquire it so other tasks can still use it. A naive implementation would be to use an AtomicUsize and index into a Box<[DB::Connection]>. If the pool has a small connection count or a lot of tasks want to execute a query, the AtomicUsize would wrap around and send the query to a connection that is potentially still in use (and thus be query pipelining).

This would improve performance and also help with the infamous PoolTimedOut error.

I think the only breaking change this feature would need is to have the Executor trait take a &self.

My naive solution isn't a valid solution since we should have a way to acquire a connection to support Transaction. Maybe we could use an async RwLock or do something with ArcSwap.

I can't think of any other db or pooling crate that supports this behavior so it might be worth experimenting with this.

I have plans to completely re-architect Pool that would also explicitly allow pipelining, as well as significantly improve performance at high contention.

It's my next big project that I hope to land, if not in 0.9.0, then a subsequent release.

Ah looks like you are one step ahead. Exciting stuff!