This patch is a ground up rewrite of the existing work-stealing thread
pool. The goal is to reduce overhead while simplifying code when
possible.
At a high level, the following architectural changes were made:
- The local run queues were switched for bounded circle buffer queues.
- Reduce cross-thread synchronization.
- Refactor task constructs to use a single allocation and always include
a join handle (#887).
- Simplify logic around putting workers to sleep and waking them up.
**Local run queues**
Move away from crossbeam's implementation of the Chase-Lev deque. This
implementation included unnecessary overhead as it supported
capabilities that are not needed for the work-stealing thread pool.
Instead, a fixed size circle buffer is used for the local queue. When
the local queue is full, half of the tasks contained in it are moved to
the global run queue.
**Reduce cross-thread synchronization**
This is done via many small improvements. Primarily, an upper bound is
placed on the number of concurrent stealers. Limiting the number of
stealers results in lower contention. Secondly, the rate at which
workers are notified and woken up is throttled. This also reduces
contention by preventing many threads from racing to steal work.
**Refactor task structure**
Now that Tokio is able to target a rust version that supports
`std::alloc` as well as `std::task`, the pool is able to optimize how
the task structure is laid out. Now, a single allocation per task is
required and a join handle is always provided enabling the spawner to
retrieve the result of the task (#887).
**Simplifying logic**
When possible, complexity is reduced in the implementation. This is done
by using locks and other simpler constructs in cold paths. The set of
sleeping workers is now represented as a `Mutex<VecDeque<usize>>`.
Instead of optimizing access to this structure, we reduce the amount the
pool must access this structure.
Secondly, we have (temporarily) removed `threadpool::blocking`. This
capability will come back later, but the original implementation was way
more complicated than necessary.
**Results**
The thread pool benchmarks have improved significantly:
Old thread pool:
```
test chained_spawn ... bench: 2,019,796 ns/iter (+/- 302,168)
test ping_pong ... bench: 1,279,948 ns/iter (+/- 154,365)
test spawn_many ... bench: 10,283,608 ns/iter (+/- 1,284,275)
test yield_many ... bench: 21,450,748 ns/iter (+/- 1,201,337)
```
New thread pool:
```
test chained_spawn ... bench: 147,943 ns/iter (+/- 6,673)
test ping_pong ... bench: 537,744 ns/iter (+/- 20,928)
test spawn_many ... bench: 7,454,898 ns/iter (+/- 283,449)
test yield_many ... bench: 16,771,113 ns/iter (+/- 733,424)
```
Real-world benchmarks improve significantly as well. This is testing the hyper hello
world server using: `wrk -t1 -c50 -d10`:
Old scheduler:
```
Running 10s test @ http://127.0.0.1:3000
1 threads and 50 connections
Thread Stats Avg Stdev Max +/- Stdev
Latency 371.53us 99.05us 1.97ms 60.53%
Req/Sec 114.61k 8.45k 133.85k 67.00%
1139307 requests in 10.00s, 95.61MB read
Requests/sec: 113923.19
Transfer/sec: 9.56MB
```
New scheduler:
```
Running 10s test @ http://127.0.0.1:3000
1 threads and 50 connections
Thread Stats Avg Stdev Max +/- Stdev
Latency 275.05us 69.81us 1.09ms 73.57%
Req/Sec 153.17k 10.68k 171.51k 71.00%
1522671 requests in 10.00s, 127.79MB read
Requests/sec: 152258.70
Transfer/sec: 12.78MB
```
As discussed in #1620, the attribute names for `#[tokio::main]` and
`#[tokio::test]` aren't great. Specifically, they both use
`single_thread` and `multi_thread`, as opposed to names that match the
runtime names: `current_thread` and `threadpool`. This PR changes the
former to the latter.
Fixes#1627.
`is_terminated` must return `true` until the future has been polled at least once to make sure that the associated block in select is called even after the delay has elapsed.
You use `Delay` in a `select!` by [fusing it](https://docs.rs/futures-preview/0.3.0-alpha.19/futures/future/trait.FutureExt.html#method.fuse):
```rust
let delay = tokio::timer::delay(/* ... */);
let delay = delay.fuse();
select! {
_ = delay => {
/* work here */
}
}
```
When polling the task, the current waker is saved to the oneshot state.
When the handle is migrated to a new task and polled again, the waker
must be swaped from the old waker to the new waker. In some cases, there
is a potential for the old waker to leak.
This bug was caught by loom with the recently added memory leak
detection.
Use a counter to count notifications. This protects against spurious
wakeups by pthreads and other libraries. The state transitions now
track num_idle precisely.
The standard library's `io` module has small utilities such as `repeat`,
`empty`, and `sink`, which return `Read` and `Write` implementations.
These can come in handy in some circiumstances. `tokio::io` has no
equivalents that implement `AsyncRead`/`AsyncWrite`.
This commit adds `repeat`, `empty`, and `sink` helpers to `tokio::io`.
In the past, it was not possible to choose to use the multi-threaded
tokio `Runtime` in tests, which meant that any test that transitively
used `executor::threadpool::blocking` would fail with
```
'blocking' annotation used from outside the context of a thread pool
```
This patch adds a runtime annotation attribute to `#[tokio::test]` just
like `#[tokio::main]` has, which lets users opt in to the threadpool
runtime over `current_thread` (the default).
The algorithm backing `AtomicWaker` effectively uses a spin lock backed
by notifying & yielding the current task. This adds a `spin_lock_hint`
annotation to cover this case.
While, in practice, the omission of `spin_lock_hint` would not cause
problems, there are platforms that do not handle spin locks very well
and could enter a deadlock in pathological cases.
- Adds a minimum `rt-current-thread` optional feature that exports
`tokio::runtime::current_thread`.
- Adds a `macros` optional feature to enable the `#[tokio::main]` and
`#[tokio::test]` attributes.
- Adjusts `#[tokio::main]` macro to select a runtime "automatically" if
a specific strategy isn't specified. Allows using the macro with only
the rt-current-thread feature.
* Removes most pin-projection related unsafe code.
* Removes manual Unpin implementations.
As references always implement Unpin, there is no need to implement
Unpin manually.
* Adds tests to check that Unpin requirement does not change accidentally
because changing Unpin requirements will be breaking changes.
`BufWriter` and `BufReader` did not previously forward the "opposite" trait (`AsyncRead` for `BufWriter` and `AsyncWrite` for `BufReader`). This meant that there was no way to have both directions buffered at once. This patch fixes that, and introduces a convenience type + constructor for this double-wrapped construct.
This adds `Barrier` to `tokio-sync`, which is an asynchronous alternative to [`std::sync::Barrier`](https://doc.rust-lang.org/std/sync/struct.Barrier.html). It is a synchronization primitive that allows multiple futures to "rendezvous" at certain points in their execution.
Currently, when threads in the blocking pool shutdown due to being idle
the counter tracking threads is not decremented. This prevents new threads
from being spawned to replace the shutdown threads.
This renames `Lock` to `Mutex`, and brings the API more in line with `std::sync::Mutex`.
In partcular, locking now only takes `&self`, with the expectation that you place the `Mutex` in an `Arc` (or something similar) to share it between threads.
Fixes#1544.
Part of #1210.
Carl Lerche