pingora-core 0.8.0

Pingora's APIs and traits for the core network protocols.
Documentation
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
463
464
465
466
467
468
469
470
471
472
473
474
475
476
477
478
479
480
481
482
483
484
485
486
487
488
489
490
491
492
493
494
495
496
497
498
499
500
501
502
503
504
505
506
507
508
509
510
511
512
513
514
515
516
517
518
519
520
521
522
523
524
525
526
527
528
529
530
531
532
533
534
535
536
537
538
539
540
541
542
543
544
545
546
547
548
549
550
551
552
553
554
555
556
557
558
559
560
561
562
563
564
565
566
567
568
569
570
571
572
573
574
575
576
577
578
579
580
581
582
583
584
585
586
587
588
589
590
591
592
593
594
595
596
597
598
599
600
601
602
603
604
605
606
607
608
609
610
611
612
613
614
615
616
617
618
619
620
621
622
623
624
625
626
627
628
629
630
631
632
633
634
635
636
637
638
639
640
641
642
643
644
645
646
647
648
649
650
651
652
653
654
655
656
657
658
659
660
661
662
663
664
665
666
667
668
669
670
671
672
673
674
675
676
677
678
679
680
681
682
683
684
685
686
687
688
689
690
691
692
693
694
695
696
697
698
699
700
701
702
703
704
705
706
707
708
709
710
711
712
713
714
715
716
717
718
719
720
721
722
723
724
725
726
727
728
729
730
731
732
733
734
735
736
737
738
739
740
741
742
743
744
745
746
747
748
749
750
751
752
753
754
755
756
757
758
759
760
761
762
763
764
765
766
767
768
769
770
771
772
773
774
775
776
777
778
779
780
781
782
783
784
785
786
787
788
789
790
791
792
793
794
795
796
797
798
799
800
801
802
803
804
805
806
807
808
809
810
811
812
813
814
815
816
817
818
819
820
821
822
823
824
825
826
827
828
// Copyright 2026 Cloudflare, Inc.
//
// Licensed under the Apache License, Version 2.0 (the "License");
// you may not use this file except in compliance with the License.
// You may obtain a copy of the License at
//
// http://www.apache.org/licenses/LICENSE-2.0
//
// Unless required by applicable law or agreed to in writing, software
// distributed under the License is distributed on an "AS IS" BASIS,
// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
// See the License for the specific language governing permissions and
// limitations under the License.

//! Server process and configuration management

mod bootstrap_services;
pub mod configuration;
#[cfg(unix)]
mod daemon;
#[cfg(unix)]
pub(crate) mod transfer_fd;

use async_trait::async_trait;
#[cfg(unix)]
use daemon::daemonize;
use daggy::NodeIndex;
use log::{debug, error, info, warn};
use parking_lot::Mutex;
use pingora_runtime::Runtime;
use pingora_timeout::fast_timeout;
#[cfg(feature = "sentry")]
use sentry::ClientOptions;
use std::sync::Arc;
use std::thread;
use std::time::SystemTime;
#[cfg(unix)]
use tokio::signal::unix;
use tokio::sync::{broadcast, watch, Mutex as TokioMutex};
use tokio::time::{sleep, Duration};

use crate::prelude::background_service;
use crate::server::bootstrap_services::{Bootstrap, BootstrapService, SentryInitService};
use crate::services::{
    DependencyGraph, ServiceHandle, ServiceReadyNotifier, ServiceReadyWatch, ServiceWithDependents,
};
use configuration::{Opt, ServerConf};
use std::collections::HashMap;
#[cfg(unix)]
pub use transfer_fd::Fds;

use pingora_error::{Error, ErrorType, Result};

/* Time to wait before exiting the program.
This is the graceful period for all existing sessions to finish */
const EXIT_TIMEOUT: u64 = 60 * 5;
/* Time to wait before shutting down listening sockets.
This is the graceful period for the new service to get ready */
const CLOSE_TIMEOUT: u64 = 5;

enum ShutdownType {
    Graceful,
    Quick,
}

/// Internal wrapper for services with dependency metadata.
pub(crate) struct ServiceWrapper {
    ready_notifier: Option<ServiceReadyNotifier>,
    service: Box<dyn ServiceWithDependents>,
    service_handle: ServiceHandle,
}

/// The execution phase the server is currently in.
#[derive(Clone, Debug)]
#[non_exhaustive]
pub enum ExecutionPhase {
    /// The server was created, but has not started yet.
    Setup,

    /// Services are being prepared.
    ///
    /// During graceful upgrades this phase acquires the listening FDs from the old process.
    Bootstrap,

    /// Bootstrap has finished, listening FDs have been transferred.
    BootstrapComplete,

    /// The server is running and is listening for shutdown signals.
    Running,

    /// A QUIT signal was received, indicating that a new process wants to take over.
    ///
    /// The server is trying to send the fds to the new process over a Unix socket.
    GracefulUpgradeTransferringFds,

    /// FDs have been sent to the new process.
    /// Waiting a fixed amount of time to allow the new process to take the sockets.
    GracefulUpgradeCloseTimeout,

    /// A TERM signal was received, indicating that the server should shut down gracefully.
    GracefulTerminate,

    /// The server is shutting down.
    ShutdownStarted,

    /// Waiting for the configured grace period to end before shutting down.
    ShutdownGracePeriod,

    /// Wait for runtimes to finish.
    ShutdownRuntimes,

    /// The server has stopped.
    Terminated,
}

/// The receiver for server's shutdown event. The value will turn to true once the server starts
/// to shutdown
pub type ShutdownWatch = watch::Receiver<bool>;
#[cfg(unix)]
pub type ListenFds = Arc<TokioMutex<Fds>>;

/// The type of shutdown process that has been requested.
#[derive(Debug)]
pub enum ShutdownSignal {
    /// Send file descriptors to the new process before starting runtime shutdown with
    /// [ServerConf::graceful_shutdown_timeout_seconds] timeout.
    GracefulUpgrade,
    /// Wait for [ServerConf::grace_period_seconds] before starting runtime shutdown with
    /// [ServerConf::graceful_shutdown_timeout_seconds] timeout.
    GracefulTerminate,
    /// Shutdown with no timeout for runtime shutdown.
    FastShutdown,
}

/// Watcher of a shutdown signal, e.g., [UnixShutdownSignalWatch] for Unix-like
/// platforms.
#[async_trait]
pub trait ShutdownSignalWatch {
    /// Returns the desired shutdown type once one has been requested.
    async fn recv(&self) -> ShutdownSignal;
}

/// A Unix shutdown watcher that awaits for Unix signals.
///
/// - `SIGQUIT`: graceful upgrade
/// - `SIGTERM`: graceful terminate
/// - `SIGINT`: fast shutdown
#[cfg(unix)]
pub struct UnixShutdownSignalWatch;

#[cfg(unix)]
#[async_trait]
impl ShutdownSignalWatch for UnixShutdownSignalWatch {
    async fn recv(&self) -> ShutdownSignal {
        let mut graceful_upgrade_signal = unix::signal(unix::SignalKind::quit()).unwrap();
        let mut graceful_terminate_signal = unix::signal(unix::SignalKind::terminate()).unwrap();
        let mut fast_shutdown_signal = unix::signal(unix::SignalKind::interrupt()).unwrap();

        tokio::select! {
            _ = graceful_upgrade_signal.recv() => {
                ShutdownSignal::GracefulUpgrade
            },
            _ = graceful_terminate_signal.recv() => {
                ShutdownSignal::GracefulTerminate
            },
            _ = fast_shutdown_signal.recv() => {
                ShutdownSignal::FastShutdown
            },
        }
    }
}

/// Arguments to configure running of the pingora server.
pub struct RunArgs {
    /// Signal for initating shutdown
    #[cfg(unix)]
    pub shutdown_signal: Box<dyn ShutdownSignalWatch>,
}

impl Default for RunArgs {
    #[cfg(unix)]
    fn default() -> Self {
        Self {
            shutdown_signal: Box::new(UnixShutdownSignalWatch),
        }
    }

    #[cfg(windows)]
    fn default() -> Self {
        Self {}
    }
}

/// The server object
///
/// This object represents an entire pingora server process which may have multiple independent
/// services (see [crate::services]). The server object handles signals, reading configuration,
/// zero downtime upgrade and error reporting.
pub struct Server {
    // This is a way to add services that have to be run before any others
    // without requiring dependencies to be set directly
    init_services: Vec<Box<dyn ServiceWithDependents + 'static>>,

    services: HashMap<NodeIndex, ServiceWrapper>,
    shutdown_watch: watch::Sender<bool>,
    // TODO: we many want to drop this copy to let sender call closed()
    shutdown_recv: ShutdownWatch,

    /// Tracks the execution phase of the server during upgrades and graceful shutdowns.
    ///
    /// Users can subscribe to the phase with [`Self::watch_execution_phase()`].
    execution_phase_watch: broadcast::Sender<ExecutionPhase>,

    /// Specification of service level dependencies
    dependencies: Arc<Mutex<DependencyGraph>>,

    /// Service initialization
    bootstrap: Arc<Mutex<Bootstrap>>,

    /// The parsed server configuration
    pub configuration: Arc<ServerConf>,
    /// The parser command line options
    pub options: Option<Opt>,
}

// TODO: delete the pid when exit

impl Server {
    /// Acquire a receiver for the server's execution phase.
    ///
    /// The receiver will produce values for each transition.
    pub fn watch_execution_phase(&self) -> broadcast::Receiver<ExecutionPhase> {
        self.execution_phase_watch.subscribe()
    }

    #[cfg(unix)]
    async fn main_loop(&self, run_args: RunArgs) -> ShutdownType {
        // waiting for exit signal

        self.execution_phase_watch
            .send(ExecutionPhase::Running)
            .ok();

        match run_args.shutdown_signal.recv().await {
            ShutdownSignal::FastShutdown => {
                info!("SIGINT received, exiting");
                ShutdownType::Quick
            }
            ShutdownSignal::GracefulTerminate => {
                // we receive a graceful terminate, all instances are instructed to stop
                info!("SIGTERM received, gracefully exiting");
                // graceful shutdown if there are listening sockets
                info!("Broadcasting graceful shutdown");
                match self.shutdown_watch.send(true) {
                    Ok(_) => {
                        info!("Graceful shutdown started!");
                    }
                    Err(e) => {
                        error!("Graceful shutdown broadcast failed: {e}");
                    }
                }
                info!("Broadcast graceful shutdown complete");

                self.execution_phase_watch
                    .send(ExecutionPhase::GracefulTerminate)
                    .ok();

                ShutdownType::Graceful
            }
            ShutdownSignal::GracefulUpgrade => {
                // TODO: still need to select! on signals in case a fast shutdown is needed
                // aka: move below to another task and only kick it off here
                info!("SIGQUIT received, sending socks and gracefully exiting");

                self.execution_phase_watch
                    .send(ExecutionPhase::GracefulUpgradeTransferringFds)
                    .ok();

                if let Some(fds) = self.listen_fds() {
                    let fds = fds.lock().await;
                    info!("Trying to send socks");
                    // XXX: this is blocking IO
                    match fds.send_to_sock(self.configuration.as_ref().upgrade_sock.as_str()) {
                        Ok(_) => {
                            info!("listener sockets sent");
                        }
                        Err(e) => {
                            error!("Unable to send listener sockets to new process: {e}");
                            // sentry log error on fd send failure
                            #[cfg(all(not(debug_assertions), feature = "sentry"))]
                            sentry::capture_error(&e);
                        }
                    }
                    self.execution_phase_watch
                        .send(ExecutionPhase::GracefulUpgradeCloseTimeout)
                        .ok();
                    sleep(Duration::from_secs(CLOSE_TIMEOUT)).await;
                    info!("Broadcasting graceful shutdown");
                    // gracefully exiting
                    match self.shutdown_watch.send(true) {
                        Ok(_) => {
                            info!("Graceful shutdown started!");
                        }
                        Err(e) => {
                            error!("Graceful shutdown broadcast failed: {e}");
                            // switch to fast shutdown
                            return ShutdownType::Graceful;
                        }
                    }
                    info!("Broadcast graceful shutdown complete");
                    ShutdownType::Graceful
                } else {
                    info!("No socks to send, shutting down.");
                    ShutdownType::Graceful
                }
            }
        }
    }

    #[cfg(windows)]
    async fn main_loop(&self, _run_args: RunArgs) -> ShutdownType {
        // waiting for exit signal

        self.execution_phase_watch
            .send(ExecutionPhase::Running)
            .ok();

        match tokio::signal::ctrl_c().await {
            Ok(()) => {
                info!("Ctrl+C received, gracefully exiting");
                // graceful shutdown if there are listening sockets
                info!("Broadcasting graceful shutdown");
                match self.shutdown_watch.send(true) {
                    Ok(_) => {
                        info!("Graceful shutdown started!");
                    }
                    Err(e) => {
                        error!("Graceful shutdown broadcast failed: {e}");
                    }
                }
                info!("Broadcast graceful shutdown complete");

                self.execution_phase_watch
                    .send(ExecutionPhase::GracefulTerminate)
                    .ok();

                ShutdownType::Graceful
            }
            Err(e) => {
                error!("Unable to listen for shutdown signal: {}", e);
                ShutdownType::Quick
            }
        }
    }

    #[cfg(feature = "sentry")]
    #[cfg_attr(docsrs, doc(cfg(feature = "sentry")))]
    /// The Sentry ClientOptions.
    ///
    /// Panics and other events sentry captures will be sent to this DSN **only in release mode**
    pub fn set_sentry_config(&mut self, sentry_config: ClientOptions) {
        self.bootstrap.lock().set_sentry_config(Some(sentry_config));
    }

    /// Get the configured file descriptors for listening
    #[cfg(unix)]
    fn listen_fds(&self) -> Option<ListenFds> {
        self.bootstrap.lock().get_fds()
    }

    #[allow(clippy::too_many_arguments)]
    fn run_service(
        mut service: Box<dyn ServiceWithDependents>,
        #[cfg(unix)] fds: Option<ListenFds>,
        shutdown: ShutdownWatch,
        threads: usize,
        work_stealing: bool,
        listeners_per_fd: usize,
        ready_notifier: ServiceReadyNotifier,
        dependency_watches: Vec<ServiceReadyWatch>,
    ) -> Runtime
// NOTE: we need to keep the runtime outside async since
        // otherwise the runtime will be dropped.
    {
        let service_runtime = Server::create_runtime(service.name(), threads, work_stealing);
        let service_name = service.name().to_string();
        service_runtime.get_handle().spawn(async move {
            // Wait for all dependencies to be ready
            let mut time_waited_opt: Option<Duration> = None;
            for mut watch in dependency_watches {
                let start = SystemTime::now();

                if watch.wait_for(|&ready| ready).await.is_err() {
                    error!(
                        "Service '{}' dependency channel closed before ready",
                        service_name
                    );
                }

                *time_waited_opt.get_or_insert_default() += start.elapsed().unwrap_or_default()
            }

            if let Some(time_waited) = time_waited_opt {
                service.on_startup_delay(time_waited);
            }

            // Start the actual service, passing the ready notifier
            service
                .start_service(
                    #[cfg(unix)]
                    fds,
                    shutdown,
                    listeners_per_fd,
                    ready_notifier,
                )
                .await;
            info!("service '{}' exited.", service_name);
        });
        service_runtime
    }

    /// Create a new [`Server`], using the [`Opt`] and [`ServerConf`] values provided
    ///
    /// This method is intended for pingora frontends that are NOT using the built-in
    /// command line and configuration file parsing, and are instead using their own.
    ///
    /// If a configuration file path is provided as part of `opt`, it will be ignored
    /// and a warning will be logged.
    pub fn new_with_opt_and_conf(raw_opt: impl Into<Option<Opt>>, mut conf: ServerConf) -> Server {
        let opt = raw_opt.into();
        if let Some(opts) = &opt {
            if let Some(c) = opts.conf.as_ref() {
                warn!("Ignoring command line argument using '{c}' as configuration, and using provided configuration instead.");
            }
            conf.merge_with_opt(opts);
        }

        let (tx, rx) = watch::channel(false);

        let execution_phase_watch = broadcast::channel(100).0;
        let bootstrap = Arc::new(Mutex::new(Bootstrap::new(
            &opt,
            &conf,
            &execution_phase_watch,
        )));

        Server {
            services: Default::default(),
            init_services: Default::default(),
            shutdown_watch: tx,
            shutdown_recv: rx,
            execution_phase_watch,
            configuration: Arc::new(conf),
            options: opt,
            dependencies: Arc::new(Mutex::new(DependencyGraph::new())),
            bootstrap,
        }
    }

    /// Create a new [`Server`].
    ///
    /// Only one [`Server`] needs to be created for a process. A [`Server`] can hold multiple
    /// independent services.
    ///
    /// Command line options can either be passed by parsing the command line arguments via
    /// `Opt::parse_args()`, or be generated by other means.
    pub fn new(opt: impl Into<Option<Opt>>) -> Result<Server> {
        let opt = opt.into();
        let (tx, rx) = watch::channel(false);

        let execution_phase_watch = broadcast::channel(100).0;
        let conf = if let Some(opt) = opt.as_ref() {
            opt.conf.as_ref().map_or_else(
                || {
                    // options, no conf, generated
                    ServerConf::new_with_opt_override(opt).ok_or_else(|| {
                        Error::explain(ErrorType::ReadError, "Conf generation failed")
                    })
                },
                |_| {
                    // options and conf loaded
                    ServerConf::load_yaml_with_opt_override(opt)
                },
            )
        } else {
            ServerConf::new()
                .ok_or_else(|| Error::explain(ErrorType::ReadError, "Conf generation failed"))
        }?;

        let bootstrap = Arc::new(Mutex::new(Bootstrap::new(
            &opt,
            &conf,
            &execution_phase_watch,
        )));

        Ok(Server {
            services: Default::default(),
            init_services: Default::default(),
            shutdown_watch: tx,
            shutdown_recv: rx,
            execution_phase_watch,
            configuration: Arc::new(conf),
            options: opt,
            dependencies: Arc::new(Mutex::new(DependencyGraph::new())),
            bootstrap,
        })
    }

    /// Add a service that all other services will wait on before starting.
    fn add_init_service(&mut self, service: impl ServiceWithDependents + 'static) {
        let boxed_service = Box::new(service);
        self.init_services.push(boxed_service);
    }

    /// Add the init services as dependencies for all existing services
    fn apply_init_service_dependencies(&mut self) {
        let services = self
            .services
            .values()
            .map(|service| service.service_handle.clone())
            .collect::<Vec<_>>();
        let global_deps = self
            .init_services
            .drain(..)
            .collect::<Vec<_>>()
            .into_iter()
            .map(|dep| self.add_boxed_service(dep))
            .collect::<Vec<_>>();
        for service in services {
            service.add_dependencies(&global_deps);
        }
    }

    /// Add a service to this server.
    ///
    /// Returns a [`ServiceHandle`] that can be used to declare dependencies.
    ///
    /// # Example
    ///
    /// ```rust,ignore
    /// let db_id = server.add_service(database_service);
    /// let api_id = server.add_service(api_service);
    ///
    /// // Declare that API depends on database
    /// api_id.add_dependency(&db_id);
    /// ```
    pub fn add_service(&mut self, service: impl ServiceWithDependents + 'static) -> ServiceHandle {
        self.add_boxed_service(Box::new(service))
    }

    /// Add a pre-boxed service to this server.
    ///
    /// Returns a [`ServiceHandle`] that can be used to declare dependencies.
    ///
    /// # Example
    ///
    /// ```rust,ignore
    /// let db_id = server.add_service(database_service);
    /// let api_id = server.add_service(api_service);
    ///
    /// // Declare that API depends on database
    /// api_id.add_dependency(&db_id);
    /// ```
    pub fn add_boxed_service(
        &mut self,
        service_box: Box<dyn ServiceWithDependents>,
    ) -> ServiceHandle {
        let name = service_box.name().to_string();

        // Create a readiness notifier for this service
        let (tx, rx) = watch::channel(false);

        let id = self.dependencies.lock().add_node(name.clone(), rx.clone());

        let service_handle = ServiceHandle::new(id, name, rx, &self.dependencies);

        let wrapper = ServiceWrapper {
            ready_notifier: Some(ServiceReadyNotifier::new(tx)),
            service: service_box,
            service_handle: service_handle.clone(),
        };

        self.services.insert(id, wrapper);

        service_handle
    }

    /// Similar to [`Self::add_service()`], but take a list of services.
    ///
    /// Returns a `Vec<ServiceHandle>` for all added services.
    pub fn add_services(
        &mut self,
        services: Vec<Box<dyn ServiceWithDependents>>,
    ) -> Vec<ServiceHandle> {
        services
            .into_iter()
            .map(|service| self.add_boxed_service(service))
            .collect()
    }

    /// Prepare the server to start
    ///
    /// When trying to zero downtime upgrade from an older version of the server which is already
    /// running, this function will try to get all its listening sockets in order to take them over.
    pub fn bootstrap(&mut self) {
        self.bootstrap.lock().bootstrap();
    }

    /// Create a service that will run to prepare the service to start
    ///
    /// The created service will handle the zero-downtime upgrade from an older version of the server
    /// to this one. It will try to get all its listening sockets in order to take them over.
    ///
    /// Other bootstrapping functionality like sentry initialization will also be handled, but as a
    /// service that will complete before any other service starts.
    pub fn bootstrap_as_a_service(&mut self) -> ServiceHandle {
        let bootstrap_service =
            background_service("Bootstrap Service", BootstrapService::new(&self.bootstrap));

        let sentry_service = background_service(
            "Sentry Init Service",
            SentryInitService::new(&self.bootstrap),
        );

        self.add_init_service(sentry_service);

        self.add_service(bootstrap_service)
    }

    /// Start the server using [Self::run] and default [RunArgs].
    ///
    /// This function will block forever until the server needs to quit. So this would be the last
    /// function to call for this object.
    ///
    /// Note: this function may fork the process for daemonization, so any additional threads created
    /// before this function will be lost to any service logic once this function is called.
    pub fn run_forever(self) -> ! {
        self.run(RunArgs::default());

        std::process::exit(0)
    }

    /// Run the server until execution finished.
    ///
    /// This function will run until the server has been instructed to shut down
    /// through a signal, and will then wait for all services to finish and
    /// runtimes to exit.
    ///
    /// Note: if daemonization is enabled in the config, this function will
    /// never return.
    /// Instead it will either start the daemon process and exit, or panic
    /// if daemonization fails.
    pub fn run(mut self, run_args: RunArgs) {
        self.apply_init_service_dependencies();

        info!("Server starting");

        let conf = self.configuration.as_ref();

        #[cfg(unix)]
        if conf.daemon {
            info!("Daemonizing the server");
            fast_timeout::pause_for_fork();
            daemonize(&self.configuration);
            fast_timeout::unpause();
        }

        #[cfg(windows)]
        if conf.daemon {
            panic!("Daemonizing under windows is not supported");
        }

        // Holds tuples of runtimes and their service name.
        let mut runtimes: Vec<(Runtime, String)> = Vec::new();

        // Get services in topological order (dependencies first)
        let startup_order = match self.dependencies.lock().topological_sort() {
            Ok(order) => order,
            Err(e) => {
                error!("Failed to determine service startup order: {}", e);
                std::process::exit(1);
            }
        };

        // Log service names in startup order
        let service_names: Vec<String> = startup_order
            .iter()
            .map(|(_, service)| service.name.clone())
            .collect();
        info!("Starting services in dependency order: {:?}", service_names);

        // Start services in dependency order
        for (service_id, service) in startup_order {
            let mut wrapper = match self.services.remove(&service_id) {
                Some(w) => w,
                None => {
                    warn!(
                        "Service ID {:?}-{} in startup order but not found",
                        service_id, service.name
                    );
                    continue;
                }
            };

            let threads = wrapper.service.threads().unwrap_or(conf.threads);
            let name = wrapper.service.name().to_string();

            // Extract dependency watches from the ServiceHandle
            let dependencies = self
                .dependencies
                .lock()
                .get_dependencies(wrapper.service_handle.id);

            // Get the readiness notifier for this service by taking it from the Option.
            // Since service_id is the index, we can directly access it.
            // We take() the notifier, leaving None in its place.
            let ready_notifier = wrapper
                .ready_notifier
                .take()
                .expect("Service notifier should exist");

            if !dependencies.is_empty() {
                info!(
                    "Service '{name}' will wait for dependencies: {:?}",
                    dependencies.iter().map(|s| &s.name).collect::<Vec<_>>()
                );
            } else {
                info!("Starting service: {}", name);
            }

            let dependency_watches = dependencies
                .iter()
                .map(|s| s.ready_watch.clone())
                .collect::<Vec<_>>();

            let runtime = Server::run_service(
                wrapper.service,
                #[cfg(unix)]
                self.listen_fds(),
                self.shutdown_recv.clone(),
                threads,
                conf.work_stealing,
                self.configuration.listener_tasks_per_fd,
                ready_notifier,
                dependency_watches,
            );
            runtimes.push((runtime, name));
        }

        // blocked on main loop so that it runs forever
        // Only work steal runtime can use block_on()
        let server_runtime = Server::create_runtime("Server", 1, true);
        #[cfg(unix)]
        let shutdown_type = server_runtime
            .get_handle()
            .block_on(self.main_loop(run_args));
        #[cfg(windows)]
        let shutdown_type = server_runtime
            .get_handle()
            .block_on(self.main_loop(run_args));

        self.execution_phase_watch
            .send(ExecutionPhase::ShutdownStarted)
            .ok();

        if matches!(shutdown_type, ShutdownType::Graceful) {
            self.execution_phase_watch
                .send(ExecutionPhase::ShutdownGracePeriod)
                .ok();

            let exit_timeout = self
                .configuration
                .as_ref()
                .grace_period_seconds
                .unwrap_or(EXIT_TIMEOUT);
            info!("Graceful shutdown: grace period {}s starts", exit_timeout);
            thread::sleep(Duration::from_secs(exit_timeout));
            info!("Graceful shutdown: grace period ends");
        }

        // Give tokio runtimes time to exit
        let shutdown_timeout = match shutdown_type {
            ShutdownType::Quick => Duration::from_secs(0),
            ShutdownType::Graceful => Duration::from_secs(
                self.configuration
                    .as_ref()
                    .graceful_shutdown_timeout_seconds
                    .unwrap_or(5),
            ),
        };

        self.execution_phase_watch
            .send(ExecutionPhase::ShutdownRuntimes)
            .ok();

        let shutdowns: Vec<_> = runtimes
            .into_iter()
            .map(|(rt, name)| {
                info!("Waiting for runtimes to exit!");
                let join = thread::spawn(move || {
                    rt.shutdown_timeout(shutdown_timeout);
                    thread::sleep(shutdown_timeout)
                });
                (join, name)
            })
            .collect();
        for (shutdown, name) in shutdowns {
            info!("Waiting for service runtime {} to exit", name);
            if let Err(e) = shutdown.join() {
                error!("Failed to shutdown service runtime {}: {:?}", name, e);
            }
            debug!("Service runtime {} has exited", name);
        }
        info!("All runtimes exited, exiting now");

        self.execution_phase_watch
            .send(ExecutionPhase::Terminated)
            .ok();
    }

    fn create_runtime(name: &str, threads: usize, work_steal: bool) -> Runtime {
        if work_steal {
            Runtime::new_steal(threads, name)
        } else {
            Runtime::new_no_steal(threads, name)
        }
    }
}