From the rust std net library:
let listener = TcpListener::bind(("127.0.0.1", port)).unwrap();
info!("Opened socket on localhost port {}", port);
// accept connections and process them serially
for stream in listener.incoming() {
break;
}
info!("closed socket");
How does one make the listener stop listening? It says in the API that when the listener is dropped, it stops. But how do we drop it if incoming() is a blocking call? Preferably without external crates like tokio/mio.
You'll want to put the TcpListener into non-blocking mode using the set_nonblocking() method, like so:
use std::io;
use std::net::TcpListener;
let listener = TcpListener::bind("127.0.0.1:7878").unwrap();
listener.set_nonblocking(true).expect("Cannot set non-blocking");
for stream in listener.incoming() {
match stream {
Ok(s) => {
// do something with the TcpStream
handle_connection(s);
}
Err(ref e) if e.kind() == io::ErrorKind::WouldBlock => {
// Decide if we should exit
break;
// Decide if we should try to accept a connection again
continue;
}
Err(e) => panic!("encountered IO error: {}", e),
}
}
Instead of waiting for a connection, the incoming() call will immediately return a Result<> type. If Result is Ok(), then a connection was made and you can process it. If the Result is Err(WouldBlock), this isn't actually an error, there just wasn't a connection pending at the exact moment incoming() checked the socket.
Note that in the WouldBlock case, you may want to put a sleep() or something before continuing, otherwise your program will rapidly poll the incoming() function checking for a connection, resulting in high CPU usage.
Code example adapted from here
The standard library doesn't provide an API for this, but there are a few strategies you can use to work around it:
Shut down reads on the socket
You can use platform-specific APIs to shutdown reads on the socket which will cause the incoming iterator to return an error. You can then break out of handling connections when the error is received. For example, on a Unix system:
use std::net::TcpListener;
use std::os::unix::io::AsRawFd;
use std::thread;
let listener = TcpListener::bind("localhost:0")?;
let fd = listener.as_raw_fd();
let handle = thread::spawn(move || {
for connection in listener.incoming() {
match connection {
Ok(connection) => { /* handle connection */ }
Err(_) => break,
}
});
libc::shutdown(fd, libc::SHUT_RD);
handle.join();
Force the listener to wake up
Another (cross-platform) trick is to set a variable indicating that you want to stop listening, and then connect to the socket yourself to force the listening thread to wake up. When the listening thread wakes up, it checks the "stop listening" variable, and then exits cleanly if it's set.
use std::net::{TcpListener, TcpStream};
use std::sync::atomic::{AtomicBool, Ordering};
use std::sync::Arc;
use std::thread;
let listener = TcpListener::bind("localhost:0")?;
let local_addr = listener.local_addr()?;
let shutdown = Arc::new(AtomicBool::new(false));
let server_shutdown = shutdown.clone();
let handle = thread::spawn(move || {
for connection in listener.incoming() {
if server_shutdown.load(Ordering::Relaxed) {
return;
}
match connection {
Ok(connection) => { /* handle connection */ }
Err(_) => break,
}
}
});
shutdown.store(true, Ordering::Relaxed);
let _ = TcpStream::connect(local_addr);
handle.join().unwrap();
You can poll your socket with an eventfd, which used for signaling.
I wrote a helper for this.
let shutdown = EventFd::new();
let listener = TcpListener::bind("0.0.0.0:12345")?;
let incoming = CancellableIncoming::new(&listener, &shutdown);
for stream in incoming {
// Your logic
}
// While in other thread
shutdown.add(1); // Light the shutdown signal, now your incoming loop exits gracefully.
use nix;
use nix::poll::{poll, PollFd, PollFlags};
use nix::sys::eventfd::{eventfd, EfdFlags};
use nix::unistd::{close, write};
use std;
use std::net::{TcpListener, TcpStream};
use std::os::unix::io::{AsRawFd, RawFd};
pub struct EventFd {
fd: RawFd,
}
impl EventFd {
pub fn new() -> Self {
EventFd {
fd: eventfd(0, EfdFlags::empty()).unwrap(),
}
}
pub fn add(&self, v: i64) -> nix::Result<usize> {
let b = v.to_le_bytes();
write(self.fd, &b)
}
}
impl AsRawFd for EventFd {
fn as_raw_fd(&self) -> RawFd {
self.fd
}
}
impl Drop for EventFd {
fn drop(&mut self) {
let _ = close(self.fd);
}
}
// -----
//
pub struct CancellableIncoming<'a> {
listener: &'a TcpListener,
eventfd: &'a EventFd,
}
impl<'a> CancellableIncoming<'a> {
pub fn new(listener: &'a TcpListener, eventfd: &'a EventFd) -> Self {
Self { listener, eventfd }
}
}
impl<'a> Iterator for CancellableIncoming<'a> {
type Item = std::io::Result<TcpStream>;
fn next(&mut self) -> Option<std::io::Result<TcpStream>> {
use nix::errno::Errno;
let fd = self.listener.as_raw_fd();
let evfd = self.eventfd.as_raw_fd();
let mut poll_fds = vec![
PollFd::new(fd, PollFlags::POLLIN),
PollFd::new(evfd, PollFlags::POLLIN),
];
loop {
match poll(&mut poll_fds, -1) {
Ok(_) => break,
Err(nix::Error::Sys(Errno::EINTR)) => continue,
_ => panic!("Error polling"),
}
}
if poll_fds[0].revents().unwrap() == PollFlags::POLLIN {
Some(self.listener.accept().map(|p| p.0))
} else if poll_fds[1].revents().unwrap() == PollFlags::POLLIN {
None
} else {
panic!("Can't be!");
}
}
}
Related
I have a futures::stream::Stream which produces elements in the form <(State, impl std::fmt::Binary)> (Binary is an arbitrary placeholder for a trait I want to use):
let peers = (0..10).map(move |peer| async move {
let delay = core::time::Duration::from_secs(2); // should be 'random'
tokio::time::sleep(delay).await;
if peer % 2 == 0 {
stream::iter(Ok::<i32, std::io::Error>(peer).into_iter())
} else {
let custom_error = std::io::Error::new(std::io::ErrorKind::Other, "oh no!");
stream::iter(Err::<i32, std::io::Error>(custom_error).into_iter())
}
})
.collect::<FuturesUnordered<_>>()
.flatten()
.map(|peer| (State { foo: "foo".into(), bar: "bar".into() }, peer));
The peers above should correspond to a stream of peers which are connected successfully. Since I do not know until runtime how many peers are connected I can't store this in a Vec<(State, impl ...)> or similar.
Is it possible to somehow do a series of tasks concurrently which modifies the State internally for each peer where the task completion is determined by the first peer that completes the task? Similar to a race for each task.
I thought the following might work:
use futures::{
stream::{self, FuturesUnordered},
StreamExt, FutureExt,
};
#[derive(Debug, Clone)]
struct State {
foo: String,
bar: String
}
#[tokio::main]
async fn main() {
let futures = (0..10).map(move |peer| async move {
let mut delay = core::time::Duration::from_secs(2);
if peer == 0 {
delay = core::time::Duration::from_secs(100); // slow peer
}
tokio::time::sleep(delay).await;
if peer % 2 == 0 {
stream::iter(Ok::<i32, std::io::Error>(peer).into_iter())
} else {
let custom_error = std::io::Error::new(std::io::ErrorKind::Other, "oh no!");
stream::iter(Err::<i32, std::io::Error>(custom_error).into_iter())
}
})
.collect::<FuturesUnordered<_>>()
.flatten()
.map(|peer| (State { foo: "foo".into(), bar: "bar".into() }, peer));
// first task
let notify = std::rc::Rc::new(tokio::sync::Notify::new());
let futures = futures.map(|(mut state, x)| {
let notify = notify.clone();
async move {
tokio::select! {
biased;
_ = async {
println!("processing task #1 for peer {:b} with state {:?}", x, state);
let delay = core::time::Duration::from_secs(2);
tokio::time::sleep(delay).await;
state.foo = "test1".to_owned();
notify.notify_waiters();
} => {
(state, x)
}
_ = notify.notified() => { (state, x) }
}
}
}).buffer_unordered(10).collect::<Vec<(State, _)>>().await;
// second task
let notify = std::rc::Rc::new(tokio::sync::Notify::new());
let futures = stream::iter(futures);
let futures = futures.map(|(mut state, x)| {
let notify = notify.clone();
async move {
tokio::select! {
biased;
_ = async {
println!("processing task #2 for peer {:b} with state {:?}", x, state);
let delay = core::time::Duration::from_secs(2);
tokio::time::sleep(delay).await;
notify.notify_waiters();
} => {
(state, x)
}
_ = notify.notified() => { (state, x) }
}
}
}).buffer_unordered(10).collect::<Vec<(State, _)>>().await;
}
playground link
But it will be stuck on the first task because it is waiting for the slow peer with 100 seconds delay. Ideally, I want to prematurely finish the collect once the task is done. I have tried using take_until with notify.notified():
let futures = futures.map(|(mut state, x)| {
let notify = notify.clone();
async move {
tokio::select! {
...
}
}
}).buffer_unordered(10).take_until(notify.notified()).collect::<Vec<(State, _)>>().await;
but this will discard the other peers and leave only 1 peer in futures. I think this is because the outer notify.notified() takes precedence over the inner notify.notified() used in the tokio::select! statement.
Is there a way to reuse a futures::stream::Stream and simultaneously modify the elements which I have tried doing above?
Or is there a more idiomatic solution to what I am trying to achieve here?
Problem is in a microservice with Tokio, where connect to db and other stuff async, but when some connection failed, microservice dont end work. Its a great when you need this case, but I need to end work of this microservice when connection lost... so could you help me how to safety shutdown process...?
src/main.rs
use tokio; // 1.0.0+
fn main() {
let rt = tokio::runtime::Builder::new_multi_thread()
.worker_threads(workers_number)
.enable_all()
.build()
.unwrap();
rt.block_on(async {
// health cheacker connection
let health_checker = match HealthChecker::new(some_configuration).await?;
// some connection to db
// ...
// transport client connection
// ...
// so when connection failed or lost I need to
// end process like `std::process::abort()`
// but I cant use it, because its unsafe
let mut task_handler = vec![];
// create some task
join_all(task_handler).await;
});
}
anyone have some ideas?
You can call any of the Runtime shutdown methods shutdown_timeout or shutdown_background.
If it is needed some king of waiting, you could spawn a task with a tokio::sync::oneshot that will trigger the shutdown when signaled.
use core::time::Duration;
use crate::tokio::time::sleep;
use tokio;
fn main() {
let rt = tokio::runtime::Builder::new_multi_thread()
.enable_all()
.build()
.unwrap();
let handle = rt.handle().clone();
let (s, r) = tokio::sync::oneshot::channel();
rt.spawn(async move {
sleep(Duration::from_secs(1)).await;
s.send(0);
});
handle.block_on(async move {
r.await;
rt.shutdown_background();
});
}
Playground
I have a small program that executes the aws s3 cli commands but with different arguments. I'm using the Command crate and the the command makes a network call and returns some response. At first I have this synchronous & single-threaded implementation:
fn make_call<'a>(_name: &'a str, _bucket_poll: &mut BucketPoll<'a>) -> Option<BucketDetails<'a>> {
let invoke_result = invoke_network_call(_name);
let mut bucket = BucketDetails::new(_name);
match invoke_result {
Ok(invoke_str) => {
bucket.output = invoke_str;
_bucket_poll.insert_bucket(bucket.clone());
_bucket_poll.successful_count += 1;
Some(bucket)
}
Err(_) => {
_bucket_poll.insert_bucket(bucket);
None
}
}
}
// I invoke this function in sequential order, something like
make_call('name_1');
make_call('name_2');
make_call('name_3');
Because I don't really care at which order this function is executed, I decided to learn Tokio to help with performance. I changed the make_call function to be async:
async fn make_call_race() -> ExecutionResult {
let bucket_poll = BucketPoll::new();
let bucket_poll_guard = Arc::new(Mutex::new(bucket_poll));
loop {
let bucket_details = tokio::select! {
Some(bucket_details) = make_call_async("name_1", &bucket_poll_guard) => bucket_details,
Some(bucket_details) = make_call_async("name_2", &bucket_poll_guard) => bucket_details,
Some(bucket_details) = make_call_async("name_3", &bucket_poll_guard) => bucket_details,
Some(bucket_details) = make_call_async("name_4", &bucket_poll_guard) => bucket_details,
else => { break }
};
success_printer(bucket_details);
}
// more printing, no more network calls
ExecutionResult::Success
}
make_call_async is essentially the same as make_call:
async fn make_call_async<'a>(
_name: &'a str,
_bucket_poll_guard: &'a Arc<Mutex<BucketPoll<'a>>>,
) -> Option<BucketDetails<'a>> {
{
if let Ok(bucket_poll_guard) = _bucket_poll_guard.lock() {
if bucket_poll_guard.has_polled(_name) {
return None;
}
}
}
let invoke_result = invoke_network_call(_name);
let mut bucket = BucketDetails::new(_name);
match invoke_result {
Ok(invoke_str) => {
bucket.output = invoke_str;
{
if let Ok(mut bucket_poll_guard) = _bucket_poll_guard.lock() {
bucket_poll_guard.insert_bucket(bucket.clone());
bucket_poll_guard.successful_count += 1;
}
}
Some(bucket)
}
Err(_) => {
{
if let Ok(mut bucket_poll_guard) = _bucket_poll_guard.lock() {
bucket_poll_guard.insert_bucket(bucket);
}
}
None
}
}
}
When I run the async version, I do see that my network calls are made a random order but I do not notice any speedups. I increased the number of network calls to ~50ish invocations but the runtime is nearly the same if not slightly worse. As I am new to async programming and Rust in general, I would like to understand why my async implementation does not seem to offer any improvement.
Extra:
Here is the invoke_network_call method:
fn invoke_network_call(_name: &str) -> core::result::Result<String, AwsCliError> {
let output = Command::new("aws")
.arg("s3")
.arg("ls")
.arg(_name)
.output()
.expect("Could not list s3 objects");
if !output.status.success() {
err_printer(format!("Failed to list s3 objects for bucket {}.", _name));
return Err(AwsCliError);
}
let output_str = get_stdout_string_from_output(&output);
Ok(output_str)
}
EDIT: yorodm's comment makes sense. What I did was use Tokio's Command instead of std::process's Command and made the invoke_network_call async. This reduced my runtime by half. Thank you!
You could rewrite invoke_network_call using an async version of Command.
async fn invoke_network_call(_name: &str) -> core::result::Result<String, AwsCliError> {
let output = tokio::process::Command::new("aws")
.arg("s3")
.arg("ls")
.arg(_name)
.output()
.await
.expect("Could not list s3 objects");
if !output.status.success() {
err_printer(format!("Failed to list s3 objects for bucket {}.", _name));
return Err(AwsCliError);
}
let output_str = get_stdout_string_from_output(&output);
Ok(output_str)
}
Thus removing the blocking std::process::Command call. However I would say that if you're going to access AWS services you should go with rusoto
I'm creating a task which will spawn other tasks. Some of them will take some time, so they cannot be awaited, but they can run in parallel:
src/main.rs
use crossbeam::crossbeam_channel::{bounded, select};
#[tokio::main]
async fn main() {
let (s, r) = bounded::<usize>(1);
tokio::spawn(async move {
let mut counter = 0;
loop {
let loop_id = counter.clone();
tokio::spawn(async move { // why this one was not fired?
println!("inner task {}", loop_id);
}); // .await.unwrap(); - solves issue, but this is long task which cannot be awaited
println!("loop {}", loop_id);
select! {
recv(r) -> rr => {
// match rr {
// Ok(ee) => {
// println!("received from channel {}", loop_id);
// tokio::spawn(async move {
// println!("received from channel task {}", loop_id);
// });
// },
// Err(e) => println!("{}", e),
// };
},
// more recv(some_channel) ->
}
counter = counter + 1;
}
});
// let s_clone = s.clone();
// tokio::spawn(async move {
// s_clone.send(2).unwrap();
// });
loop {
// rest of the program
}
}
I've noticed strange behavior. This outputs:
loop 0
I was expecting it to also output inner task 0.
If I send a value to channel, the output will be:
loop 0
inner task 0
loop 1
This is missing inner task 1.
Why is inner task spawned with one loop of delay?
The first time I noticed such behavior with 'received from channel task' delayed one loop, but when I reduced code to prepare sample this started to happen with 'inner task'. It might be worth mentioning that if I write second tokio::spawn right to another, only the last one will have this issue. Is there something I should be aware when calling tokio::spawn and select!? What causes this one loop of delay?
Cargo.toml dependencies
[dependencies]
tokio = { version = "0.2", features = ["full"] }
crossbeam = "0.7"
Rust 1.46, Windows 10
select! is blocking, and the docs for tokio::spawn say:
The spawned task may execute on the current thread, or it may be sent to a different thread to be executed.
In this case, the select! "future" is actually a blocking function, and spawn doesn't use a new thread (either in the first invocation or the one inside the loop).
Because you don't tell tokio that you are going to block, tokio doesn't think another thread is needed (from tokio's perspective, you only have 3 futures which should never block, so why would you need another thread anyway?).
The solution is to use the tokio::task::spawn_blocking for the select!-ing closure (which will no longer be a future, so async move {} is now move || {}).
Now tokio will know that this function actually blocks, and will move it to another thread (while keeping all the actual futures in other execution threads).
use crossbeam::crossbeam_channel::{bounded, select};
#[tokio::main]
async fn main() {
let (s, r) = bounded::<usize>(1);
tokio::task::spawn_blocking(move || {
// ...
});
loop {
// rest of the program
}
}
Link to playground
Another possible solution is to use a non-blocking channel like tokio::sync::mpsc, on which you can use await and get the expected behavior, like this playground example with direct recv().await or with tokio::select!, like this:
use tokio::sync::mpsc;
#[tokio::main]
async fn main() {
let (mut s, mut r) = mpsc::channel::<usize>(1);
tokio::spawn(async move {
loop {
// ...
tokio::select! {
Some(i) = r.recv() => {
println!("got = {}", i);
}
}
}
});
loop {
// rest of the program
}
}
Link to playground
I have asynchronous code which calls synchronous code that takes a while to run, so I followed the suggestions outlined in What is the best approach to encapsulate blocking I/O in future-rs?. However, my asynchronous code has a timeout, after which I am no longer interested in the result of the synchronous calculation:
use std::{thread, time::Duration};
use tokio::{task, time}; // 0.2.10
// This takes 1 second
fn long_running_complicated_calculation() -> i32 {
let mut sum = 0;
for i in 0..10 {
thread::sleep(Duration::from_millis(100));
eprintln!("{}", i);
sum += i;
// Interruption point
}
sum
}
#[tokio::main]
async fn main() {
let handle = task::spawn_blocking(long_running_complicated_calculation);
let guarded = time::timeout(Duration::from_millis(250), handle);
match guarded.await {
Ok(s) => panic!("Sum was calculated: {:?}", s),
Err(_) => eprintln!("Sum timed out (expected)"),
}
}
Running this code shows that the timeout fires, but the synchronous code also continues to run:
0
1
Sum timed out (expected)
2
3
4
5
6
7
8
9
How can I stop running the synchronous code when the future wrapping it is dropped?
I don't expect that the compiler will magically be able to stop my synchronous code. I've annotated a line with "interruption point" where I'd be required to manually put some kind of check to exit early from my function, but I don't know how to easily get a notification that the result of spawn_blocking (or ThreadPool::spawn_with_handle, for pure futures-based code) has been dropped.
You can pass an atomic boolean which you then use to flag the task as needing cancellation. (I'm not sure I'm using an appropriate Ordering for the load/store calls, that probably needs some more consideration)
Here's a modified version of your code that outputs
0
1
Sum timed out (expected)
2
Interrupted...
use std::sync::atomic::{AtomicBool, Ordering};
use std::sync::Arc;
use std::{thread, time::Duration};
use tokio::{task, time}; // 0.2.10
// This takes 1 second
fn long_running_complicated_calculation(flag: &AtomicBool) -> i32 {
let mut sum = 0;
for i in 0..10 {
thread::sleep(Duration::from_millis(100));
eprintln!("{}", i);
sum += i;
// Interruption point
if !flag.load(Ordering::Relaxed) {
eprintln!("Interrupted...");
break;
}
}
sum
}
#[tokio::main]
async fn main() {
let some_bool = Arc::new(AtomicBool::new(true));
let some_bool_clone = some_bool.clone();
let handle =
task::spawn_blocking(move || long_running_complicated_calculation(&some_bool_clone));
let guarded = time::timeout(Duration::from_millis(250), handle);
match guarded.await {
Ok(s) => panic!("Sum was calculated: {:?}", s),
Err(_) => {
eprintln!("Sum timed out (expected)");
some_bool.store(false, Ordering::Relaxed);
}
}
}
playground
It's not really possible to get this to happen automatically on the dropping of the futures / handles with current Tokio. Some work towards this is being done in http://github.com/tokio-rs/tokio/issues/1830 and http://github.com/tokio-rs/tokio/issues/1879.
However, you can get something similar by wrapping the futures in a custom type.
Here's an example which looks almost the same as the original code, but with the addition of a simple wrapper type in a module. It would be even more ergonomic if I implemented Future<T> on the wrapper type that just forwards to the wrapped handle, but that was proving tiresome.
mod blocking_cancelable_task {
use std::sync::atomic::{AtomicBool, Ordering};
use std::sync::Arc;
use tokio::task;
pub struct BlockingCancelableTask<T> {
pub h: Option<tokio::task::JoinHandle<T>>,
flag: Arc<AtomicBool>,
}
impl<T> Drop for BlockingCancelableTask<T> {
fn drop(&mut self) {
eprintln!("Dropping...");
self.flag.store(false, Ordering::Relaxed);
}
}
impl<T> BlockingCancelableTask<T>
where
T: Send + 'static,
{
pub fn new<F>(f: F) -> BlockingCancelableTask<T>
where
F: FnOnce(&AtomicBool) -> T + Send + 'static,
{
let flag = Arc::new(AtomicBool::new(true));
let flag_clone = flag.clone();
let h = task::spawn_blocking(move || f(&flag_clone));
BlockingCancelableTask { h: Some(h), flag }
}
}
pub fn spawn<F, T>(f: F) -> BlockingCancelableTask<T>
where
T: Send + 'static,
F: FnOnce(&AtomicBool) -> T + Send + 'static,
{
BlockingCancelableTask::new(f)
}
}
use std::sync::atomic::{AtomicBool, Ordering};
use std::{thread, time::Duration};
use tokio::time; // 0.2.10
// This takes 1 second
fn long_running_complicated_calculation(flag: &AtomicBool) -> i32 {
let mut sum = 0;
for i in 0..10 {
thread::sleep(Duration::from_millis(100));
eprintln!("{}", i);
sum += i;
// Interruption point
if !flag.load(Ordering::Relaxed) {
eprintln!("Interrupted...");
break;
}
}
sum
}
#[tokio::main]
async fn main() {
let mut h = blocking_cancelable_task::spawn(long_running_complicated_calculation);
let guarded = time::timeout(Duration::from_millis(250), h.h.take().unwrap());
match guarded.await {
Ok(s) => panic!("Sum was calculated: {:?}", s),
Err(_) => {
eprintln!("Sum timed out (expected)");
}
}
}
playground