The major breaking change in Mio v0.8 is TcpSocket type being removed.
Replacing Mio's TcpSocket we switch to the socket2 library which
provides a similar type Socket, as well as SockRef, which provide all
options TcpSocket provided (and more!).
Tokio's TcpSocket type is now backed by Socket2 instead of Mio's
TcpSocket. The main pitfall here is that socket2 isn't non-blocking by
default, which Mio obviously is. As a result we have to do potentially
blocking calls more carefully, specifically we need to handle
would-block-like errors when connecting the TcpSocket ourselves.
One benefit for this change is that adding more socket options to
TcpSocket is now merely a single function call away (in most cases
anyway).
PR #3881 factored out the spawning of local tasks on a `LocalSet` into a
function `spawn_local_inner`, so that the implementation could be shared
with the `task::Builder` API. But, that PR neglected to add a
`#[track_caller]` attribute to `spawn_local_inner`, so the `tracing`
spans for local tasks are all generated with `spawn_local_inner` as
their spawn location, rather than forwarding the actual spawn location
from the calling function.
This causes pretty useless results when using `tokio-console` with code
that spawns a number of local tasks, such as Actix
(https://reddit.com/r/rust/comments/snt5fq/can_tokioconsole_profile_actixrt/)
This commit fixes the issue by adding the missing `#[track_caller]`
attribute.
`Instant::duration_since`, `Instant::elapsed`, and `Instant::sub` may
panic. This is especially dangerous when `Instant::now` travels back in
time. While this isn't supposed to happen, this behavior is highly
platform-dependent (e.g., rust-lang/rust#86470).
This change modifies the behavior of `tokio::time::Instant` to prevent
this class of panic, as proposed for `std::time::Instant` in
rust-lang/rust#89926.
When backporting patches to LTS branches, we often run into CI failures due to
changes in rust. Newer rust versions add more lints, which break CI. We really
don't want to also have to backport patches that fix CI, so instead, LTS branches
should pin the stable rust version in CI (e.g. #4434).
This PR restructures the CI config files to make it a bit easier to set a specific rust
version in CI.
* Several of tokio's features (e.g. the channel implementation) do not
need a runtime to work, and can be compiled and used for
wasm32-unknown-unknown targets
* This change enables running tests for the `sync` and `macros` features
so that we can note any regressions there
This patch reduces the number of times worker threads wake up without having
work to do in the multi-threaded scheduler. Unnecessary wake-ups are expensive
and slow down the scheduler. I have observed this change reduce no-op wakes
by up to 50%.
The multi-threaded scheduler is work-stealing. When a worker has tasks to process,
and other workers are idle (parked), these idle workers must be unparked so that
they can steal work from the busy worker. However, unparking threads is expensive,
so there is an optimization that avoids unparking a worker if there already exists
workers in a "searching" state (the worker is unparked and looking for work). This
works pretty well, but transitioning from 1 "searching" worker to 0 searching workers
introduces a race condition where a thread unpark can be lost:
* thread 1: last searching worker about to exit searching state
* thread 2: needs to unpark a thread, but skip because there is a searching worker.
* thread 1: exits searching state w/o seeing thread 2's work.
Because this should be a rare condition, Tokio solves this by always unparking a
new worker when the current worker:
* is the last searching worker
* is transitioning out of searching
* has work to process.
When the newly unparked worker wakes, if the race condition described above
happened, "thread 2"'s work will be found. Otherwise, it will just go back to sleep.
Now we come to the issue at hand. A bug incorrectly set a worker to "searching"
when the I/O driver unparked the thread. In a situation where the scheduler was
only partially under load and is able to operate with 1 active worker, the I/O driver
would unpark the thread when new I/O events are received, incorrectly transition
it to "searching", find new work generated by inbound I/O events, incorrectly
transition itself from the last searcher -> no searchers, and unpark a new thread.
This new thread would wake, find no work and go back to sleep.
Note that, when the scheduler is fully saturated, this change will make no impact
as most workers are always unparked and the optimization to avoid unparking
threads described at the top apply.
Re-applies #4377 and fixes the bug resulting in Hyper's double panic.
Revert: #4394
Original PR:
This PR does some refactoring to the current-thread scheduler bringing it closer to the structure of the
multi-threaded scheduler. More specifically, the core scheduler data is stored in a Core struct and that
struct is passed around as a "token" indicating permission to do work. The Core structure is also stored
in the thread-local context.
This refactor is intended to support #4373, making it easier to track counters in more locations in the
current-thread scheduler.
I tried to keep commits small, but the "set Core in thread-local context" is both the biggest commit and
the key one.
b-naber