Update to tokio 0.2 and futures 0.3

[leoyvens]

This brings into the future of futures by upgrading to tokio 0.2, futures 0.3, hyper 0.13 and various other dependency upgrades. This not only allows us to use async/.await but also pushes us towards that by having updated dependencies. Some parts of the code were updated to use async/.await to serve as examples. This is working fine as far as I have tested it.

Using std::future and async

There's now a Future trait that lives in std, though helpers are still in the futures crate. The major difference from the one in futures 0.1 is that it has a single Output associated type, instead of Item and Error (same for streams). It is easy to convert back and forth between a std future that has Output = Result<_, _> and the old futures by calling .compat().

The async syntax can be thought of as a future literal, and works on fns and blocks (but not closures, yet). To use async in trait fns the async-trait crate is required, I haven't tried that yet. In blocks you usually want async move. Inside an async fn or block, you can use .await on futures as it were the old and_then, but more flexible. Using future combinators is still useful, so checkout the docs for FutureExt and TryFutureExt to catch up on them. There is no new syntax for streams, yet. Sometimes the compiler will complain about Unpin, I haven't totally understood the pinning stuff but wrapping the future in a Box::pin usually fixes that.

Spawning tasks and panic handling

Usually we want a tokio task panic to abort the process so that the node has a chance to recover by restarting. Except in some situations such as in the runtime or in queries. So instead of using tokio::spawn directly, we should choose between:

• graph::spawn which runs on the default threadpool and aborts on panic.
• graph::spawn_blocking which is run on the blocking threadpool and aborts on panic.
• graph::spawn_blocking_allow_panic which is run on the blocking threadpool and allows panics. A panic will result in an error in the JoinHandle.

Note that all of these now return a JoinHandle for joining the task, which can be very useful.

Resolves #937, since we're now careful to use the blocking thread pool when called for, and we're also back to the default thread pool size.

Resolves #805, panics in query processing are no longer fatal.

[That3Percent]

Mostly suggestions which I leave to your judgment, but there are a couple blocking issues here that need be addressed, particularly tests not running and futures not being polled.

[leoyvens]

@That3Percent Thank you for the review, I believe I've addressed all comments.

[Jannis]

There's a wip commit in this PR. Can we change that? 😉

[That3Percent]

Nice. LGTM

[leoyvens]

@Jannis You've caught me, I squashed that and rebased. After tests pass, I'll merge.

[Jannis]

I'm doing a review pass myself. Can you wait for that? I'll make it quick. 😄

[leoyvens]

@Jannis np at all, thanks for also reviewing.

[Jannis]

Looking good! I merely have a few questions.

[leoyvens]

@Jannis I've addressed your review.