1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
322
323
324
325
326
327
328
329
330
331
332
333
334
335
336
337
338
339
340
341
342
343
344
345
346
347
348
349
350
351
352
353
354
355
356
357
358
359
360
361
362
363
364
365
366
367
368
369
370
371
372
373
374
375
376
377
378
379
380
381
382
383
384
385
386
387
388
389
390
391
392
393
394
395
396
397
398
399
400
401
402
403
404
405
406
407
408
409
410
411
412
413
414
415
416
417
418
419
420
421
422
423
424
425
426
427
428
429
430
431
432
433
434
435
436
437
438
439
440
441
442
443
444
445
446
447
448
449
450
451
452
453
454
455
456
457
458
459
460
461
462
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
}