actix-net/actix-server/src/accept.rs

use std::{io, thread, time::Duration};

use actix_rt::time::Instant;
use mio::{Interest, Poll, Token as MioToken};
use tracing::{debug, error, info};

use crate::{
    availability::Availability,
    socket::MioListener,
    waker_queue::{WakerInterest, WakerQueue, WAKER_TOKEN},
    worker::{Conn, ServerWorker, WorkerHandleAccept, WorkerHandleServer},
    ServerBuilder, ServerHandle,
};

const TIMEOUT_DURATION_ON_ERROR: Duration = Duration::from_millis(510);

struct ServerSocketInfo {
    token: usize,

    lst: MioListener,

    /// Timeout is used to mark the deadline when this socket's listener should be registered again
    /// after an error.
    timeout: Option<actix_rt::time::Instant>,
}

/// Poll instance of the server.
pub(crate) struct Accept {
    poll: Poll,
    waker_queue: WakerQueue,
    handles: Vec<WorkerHandleAccept>,
    srv: ServerHandle,
    next: usize,
    avail: Availability,
    /// use the smallest duration from sockets timeout.
    timeout: Option<Duration>,
    paused: bool,
}

impl Accept {
    pub(crate) fn start(
        sockets: Vec<(usize, MioListener)>,
        builder: &ServerBuilder,
    ) -> io::Result<(WakerQueue, Vec<WorkerHandleServer>, thread::JoinHandle<()>)> {
        let handle_server = ServerHandle::new(builder.cmd_tx.clone());

        // construct poll instance and its waker
        let poll = Poll::new()?;
        let waker_queue = WakerQueue::new(poll.registry())?;

        // start workers and collect handles
        let (handles_accept, handles_server) = (0..builder.threads)
            .map(|idx| {
                // clone service factories
                let factories = builder
                    .factories
                    .iter()
                    .map(|f| f.clone_factory())
                    .collect::<Vec<_>>();

                // start worker using service factories
                ServerWorker::start(idx, factories, waker_queue.clone(), builder.worker_config)
            })
            .collect::<io::Result<Vec<_>>>()?
            .into_iter()
            .unzip();

        let (mut accept, mut sockets) = Accept::new_with_sockets(
            poll,
            waker_queue.clone(),
            sockets,
            handles_accept,
            handle_server,
        )?;

        let accept_handle = thread::Builder::new()
            .name("actix-server acceptor".to_owned())
            .spawn(move || accept.poll_with(&mut sockets))
            .map_err(|err| io::Error::new(io::ErrorKind::Other, err))?;

        Ok((waker_queue, handles_server, accept_handle))
    }

    fn new_with_sockets(
        poll: Poll,
        waker_queue: WakerQueue,
        sockets: Vec<(usize, MioListener)>,
        accept_handles: Vec<WorkerHandleAccept>,
        server_handle: ServerHandle,
    ) -> io::Result<(Accept, Box<[ServerSocketInfo]>)> {
        let sockets = sockets
            .into_iter()
            .map(|(token, mut lst)| {
                // Start listening for incoming connections
                poll.registry()
                    .register(&mut lst, MioToken(token), Interest::READABLE)?;

                Ok(ServerSocketInfo {
                    token,
                    lst,
                    timeout: None,
                })
            })
            .collect::<io::Result<_>>()?;

        let mut avail = Availability::default();

        // Assume all handles are avail at construct time.
        avail.set_available_all(&accept_handles);

        let accept = Accept {
            poll,
            waker_queue,
            handles: accept_handles,
            srv: server_handle,
            next: 0,
            avail,
            timeout: None,
            paused: false,
        };

        Ok((accept, sockets))
    }

    /// blocking wait for readiness events triggered by mio
    fn poll_with(&mut self, sockets: &mut [ServerSocketInfo]) {
        let mut events = mio::Events::with_capacity(256);

        loop {
            if let Err(err) = self.poll.poll(&mut events, self.timeout) {
                match err.kind() {
                    io::ErrorKind::Interrupted => {}
                    _ => panic!("Poll error: {}", err),
                }
            }

            for event in events.iter() {
                let token = event.token();
                match token {
                    WAKER_TOKEN => {
                        let exit = self.handle_waker(sockets);
                        if exit {
                            info!("accept thread stopped");
                            return;
                        }
                    }
                    _ => {
                        let token = usize::from(token);
                        self.accept(sockets, token);
                    }
                }
            }

            // check for timeout and re-register sockets
            self.process_timeout(sockets);
        }
    }

    fn handle_waker(&mut self, sockets: &mut [ServerSocketInfo]) -> bool {
        // This is a loop because interests for command from previous version was
        // a loop that would try to drain the command channel. It's yet unknown
        // if it's necessary/good practice to actively drain the waker queue.
        loop {
            // Take guard with every iteration so no new interests can be added until the current
            // task is done. Take care not to take the guard again inside this loop.
            let mut guard = self.waker_queue.guard();

            #[allow(clippy::significant_drop_in_scrutinee)]
            match guard.pop_front() {
                // Worker notified it became available.
                Some(WakerInterest::WorkerAvailable(idx)) => {
                    drop(guard);

                    self.avail.set_available(idx, true);

                    if !self.paused {
                        self.accept_all(sockets);
                    }
                }

                // A new worker thread has been created so store its handle.
                Some(WakerInterest::Worker(handle)) => {
                    drop(guard);

                    self.avail.set_available(handle.idx(), true);
                    self.handles.push(handle);

                    if !self.paused {
                        self.accept_all(sockets);
                    }
                }

                Some(WakerInterest::Pause) => {
                    drop(guard);

                    if !self.paused {
                        self.paused = true;

                        self.deregister_all(sockets);
                    }
                }

                Some(WakerInterest::Resume) => {
                    drop(guard);

                    if self.paused {
                        self.paused = false;

                        sockets.iter_mut().for_each(|info| {
                            self.register_logged(info);
                        });

                        self.accept_all(sockets);
                    }
                }

                Some(WakerInterest::Stop) => {
                    if !self.paused {
                        self.deregister_all(sockets);
                    }

                    return true;
                }

                // waker queue is drained
                None => {
                    // Reset the WakerQueue before break so it does not grow infinitely
                    WakerQueue::reset(&mut guard);

                    return false;
                }
            }
        }
    }

    fn process_timeout(&mut self, sockets: &mut [ServerSocketInfo]) {
        // always remove old timeouts
        if self.timeout.take().is_some() {
            let now = Instant::now();

            sockets
                .iter_mut()
                // Only sockets that had an associated timeout were deregistered.
                .filter(|info| info.timeout.is_some())
                .for_each(|info| {
                    let inst = info.timeout.take().unwrap();

                    if now < inst {
                        // still timed out; try to set new timeout
                        info.timeout = Some(inst);
                        self.set_timeout(inst - now);
                    } else if !self.paused {
                        // timeout expired; register socket again
                        self.register_logged(info);
                    }

                    // Drop the timeout if server is paused and socket timeout is expired.
                    // When server recovers from pause it will register all sockets without
                    // a timeout value so this socket register will be delayed till then.
                });
        }
    }

    /// Update accept timeout with `duration` if it is shorter than current timeout.
    fn set_timeout(&mut self, duration: Duration) {
        match self.timeout {
            Some(ref mut timeout) => {
                if *timeout > duration {
                    *timeout = duration;
                }
            }
            None => self.timeout = Some(duration),
        }
    }

    #[cfg(not(target_os = "windows"))]
    fn register(&self, info: &mut ServerSocketInfo) -> io::Result<()> {
        let token = MioToken(info.token);
        self.poll
            .registry()
            .register(&mut info.lst, token, Interest::READABLE)
    }

    #[cfg(target_os = "windows")]
    fn register(&self, info: &mut ServerSocketInfo) -> io::Result<()> {
        // On windows, calling register without deregister cause an error.
        // See https://github.com/actix/actix-web/issues/905
        // Calling reregister seems to fix the issue.
        let token = MioToken(info.token);
        self.poll
            .registry()
            .register(&mut info.lst, token, Interest::READABLE)
            .or_else(|_| {
                self.poll
                    .registry()
                    .reregister(&mut info.lst, token, Interest::READABLE)
            })
    }

    fn register_logged(&self, info: &mut ServerSocketInfo) {
        match self.register(info) {
            Ok(_) => debug!("resume accepting connections on {}", info.lst.local_addr()),
            Err(err) => error!("can not register server socket {}", err),
        }
    }

    fn deregister_logged(&self, info: &mut ServerSocketInfo) {
        match self.poll.registry().deregister(&mut info.lst) {
            Ok(_) => debug!("paused accepting connections on {}", info.lst.local_addr()),
            Err(err) => {
                error!("can not deregister server socket {}", err)
            }
        }
    }

    fn deregister_all(&self, sockets: &mut [ServerSocketInfo]) {
        // This is a best effort implementation with following limitation:
        //
        // Every ServerSocketInfo with associated timeout will be skipped and it's timeout is
        // removed in the process.
        //
        // Therefore WakerInterest::Pause followed by WakerInterest::Resume in a very short gap
        // (less than 500ms) would cause all timing out ServerSocketInfos be re-registered before
        // expected timing.
        sockets
            .iter_mut()
            // Take all timeout.
            // This is to prevent Accept::process_timer method re-register a socket afterwards.
            .map(|info| (info.timeout.take(), info))
            // Socket info with a timeout is already deregistered so skip them.
            .filter(|(timeout, _)| timeout.is_none())
            .for_each(|(_, info)| self.deregister_logged(info));
    }

    // Send connection to worker and handle error.
    fn send_connection(&mut self, conn: Conn) -> Result<(), Conn> {
        let next = self.next();
        match next.send(conn) {
            Ok(_) => {
                // Increment counter of WorkerHandle.
                // Set worker to unavailable with it hit max (Return false).
                if !next.inc_counter() {
                    let idx = next.idx();
                    self.avail.set_available(idx, false);
                }
                self.set_next();
                Ok(())
            }
            Err(conn) => {
                // Worker thread is error and could be gone.
                // Remove worker handle and notify `ServerBuilder`.
                self.remove_next();

                if self.handles.is_empty() {
                    error!("no workers");
                    // All workers are gone and Conn is nowhere to be sent.
                    // Treat this situation as Ok and drop Conn.
                    return Ok(());
                } else if self.handles.len() <= self.next {
                    self.next = 0;
                }

                Err(conn)
            }
        }
    }

    fn accept_one(&mut self, mut conn: Conn) {
        loop {
            let next = self.next();
            let idx = next.idx();

            if self.avail.get_available(idx) {
                match self.send_connection(conn) {
                    Ok(_) => return,
                    Err(c) => conn = c,
                }
            } else {
                self.avail.set_available(idx, false);
                self.set_next();

                if !self.avail.available() {
                    while let Err(c) = self.send_connection(conn) {
                        conn = c;
                    }
                    return;
                }
            }
        }
    }

    fn accept(&mut self, sockets: &mut [ServerSocketInfo], token: usize) {
        while self.avail.available() {
            let info = &mut sockets[token];

            match info.lst.accept() {
                Ok(io) => {
                    let conn = Conn { io, token };
                    self.accept_one(conn);
                }
                Err(ref err) if err.kind() == io::ErrorKind::WouldBlock => return,
                Err(ref err) if connection_error(err) => continue,
                Err(err) => {
                    error!("error accepting connection: {}", err);

                    // deregister listener temporary
                    self.deregister_logged(info);

                    // sleep after error. write the timeout to socket info as later
                    // the poll would need it mark which socket and when it's
                    // listener should be registered
                    info.timeout = Some(Instant::now() + Duration::from_millis(500));
                    self.set_timeout(TIMEOUT_DURATION_ON_ERROR);

                    return;
                }
            };
        }
    }

    fn accept_all(&mut self, sockets: &mut [ServerSocketInfo]) {
        sockets
            .iter_mut()
            .map(|info| info.token)
            .collect::<Vec<_>>()
            .into_iter()
            .for_each(|idx| self.accept(sockets, idx))
    }

    #[inline(always)]
    fn next(&self) -> &WorkerHandleAccept {
        &self.handles[self.next]
    }

    /// Set next worker handle that would accept connection.
    #[inline(always)]
    fn set_next(&mut self) {
        self.next = (self.next + 1) % self.handles.len();
    }

    /// Remove next worker handle that fail to accept connection.
    fn remove_next(&mut self) {
        let handle = self.handles.swap_remove(self.next);
        let idx = handle.idx();
        // A message is sent to `ServerBuilder` future to notify it a new worker
        // should be made.
        self.srv.worker_faulted(idx);
        self.avail.set_available(idx, false);
    }
}

/// This function defines errors that are per-connection; if we get this error from the `accept()`
/// system call it means the next connection might be ready to be accepted.
///
/// All other errors will incur a timeout before next `accept()` call is attempted. The timeout is
/// useful to handle resource exhaustion errors like `ENFILE` and `EMFILE`. Otherwise, it could
/// enter into a temporary spin loop.
fn connection_error(e: &io::Error) -> bool {
    e.kind() == io::ErrorKind::ConnectionRefused
        || e.kind() == io::ErrorKind::ConnectionAborted
        || e.kind() == io::ErrorKind::ConnectionReset
}