Steven Fackler

The [Javadoc](https://docs.oracle.com/javase/7/docs/api/java/sql/Connection.html#setNetworkTimeout(java.util.concurrent.Executor,%20int)) for `Connection::setNetworkTimeout` is kind of interesting in that it discusses the timeout only applying to reads and not writes. I guess in practice the outbound TCP buffer doesn't...

Thinking about this a bit more, it actually seems to me like TCP keepalive may be the better approach here compared to some kind of read/query timeout. It doesn't make...

We could definitely add a variant of is_valid that takes a timeout, but I'd want to be pretty careful about what timeouts are actually passed to it. For example, if...

I think adding `ManageTimeout::is_valid_timeout` with a default impl seems best.

At a philosophical level, I have a couple of concerns: More broadly than this single issue, I think `UnwindSafe`/`RefUnwindSafe` are huge wastes of time and mental energy. In 5+ years...

This crate's readme mentions r2d2: https://github.com/AlexPikalov/cdrs

Sure, if you make a type that implements this trait: https://docs.rs/r2d2/0.8.4/r2d2/trait.ManageConnection.html

What's the rationale for this? Error types are generally not cloneable.

But how do you concretely want to use the clonability of those things?

By default jemalloc is built with a prefix, which I think the environment variable includes.