Merge pull request #181 from greatest-ape/work-2023-01-29-b

udp: replace PanicSentinel with loop over JoinHandles

CHANGELOG.md

@@ -35,6 +35,10 @@
 * Speed up parsing and serialization of requests and responses by using
   [zerocopy](https://crates.io/crates/zerocopy)
 
+#### Fixed
+
+* Quit whole application if any worker thread quits
+
 ### aquatic_http
 
 #### Added

TODO.md

@@ -2,7 +2,7 @@
 
 ## High priority
 
-* general
+* http and ws
   * add task to generate prometheus exports on regular interval to clean up
     data. this is important if peer_clients is activated
 
@@ -12,7 +12,9 @@
     In that case, a shorter duration (e.g., 30 seconds) would be a good idea.
 
 * general
-  * panic sentinel not working? at least seemingly not in http?
+  * Replace panic sentinel with checking threads like in udp implementation.
+    It seems to be broken
+
 
 ## Medium priority
 
@@ -25,14 +27,6 @@
     * toml v0.7
     * syn v2.0
 
-* quit whole program if any thread panics
-  * But it would be nice not to panic in workers, but to return errors instead.
-    Once JoinHandle::is_finished is available in stable Rust (#90470), an
-    option would be to
-     * Save JoinHandles
-     * When preparing to quit because of PanicSentinel sending SIGTERM, loop
-       through them, extract error and log it
-
 * Run cargo-deny in CI
 
 * aquatic_ws

crates/udp/Cargo.toml

@@ -20,7 +20,7 @@ name = "aquatic_udp"
 [features]
 default = ["prometheus"]
 # Export prometheus metrics
-prometheus = ["metrics", "metrics-util", "metrics-exporter-prometheus"]
+prometheus = ["metrics", "metrics-util", "metrics-exporter-prometheus", "tokio"]
 # Experimental io_uring support (Linux 6.0 or later required)
 io-uring = ["dep:io-uring"]
 # Experimental CPU pinning support
@@ -59,6 +59,7 @@ tinytemplate = "1"
 metrics = { version = "0.22", optional = true }
 metrics-util = { version = "0.16", optional = true }
 metrics-exporter-prometheus = { version = "0.13", optional = true, default-features = false, features = ["http-listener"] }
+tokio = { version = "1", optional = true, features = ["rt", "net", "time"] }
 
 # io-uring feature
 io-uring = { version = "0.6", optional = true }

crates/udp/src/lib.rs

@@ -3,19 +3,20 @@ pub mod config;
 pub mod workers;
 
 use std::collections::BTreeMap;
-use std::thread::Builder;
+use std::fmt::Display;
+use std::thread::{sleep, Builder, JoinHandle};
 use std::time::Duration;
 
 use anyhow::Context;
 use crossbeam_channel::{bounded, unbounded};
-use signal_hook::consts::{SIGTERM, SIGUSR1};
+use signal_hook::consts::SIGUSR1;
 use signal_hook::iterator::Signals;
 
 use aquatic_common::access_list::update_access_list;
 #[cfg(feature = "cpu-pinning")]
 use aquatic_common::cpu_pinning::{pin_current_if_configured_to, WorkerIndex};
 use aquatic_common::privileges::PrivilegeDropper;
-use aquatic_common::{PanicSentinelWatcher, ServerStartInstant};
+use aquatic_common::ServerStartInstant;
 
 use common::{
     ConnectedRequestSender, ConnectedResponseSender, SocketWorkerIndex, State, SwarmWorkerIndex,
@@ -28,12 +29,12 @@ pub const APP_NAME: &str = "aquatic_udp: UDP BitTorrent tracker";
 pub const APP_VERSION: &str = env!("CARGO_PKG_VERSION");
 
 pub fn run(config: Config) -> ::anyhow::Result<()> {
-    let mut signals = Signals::new([SIGUSR1, SIGTERM])?;
+    let mut signals = Signals::new([SIGUSR1])?;
 
     let state = State::new(config.swarm_workers);
     let connection_validator = ConnectionValidator::new(&config)?;
-    let (sentinel_watcher, sentinel) = PanicSentinelWatcher::create_with_sentinel();
     let priv_dropper = PrivilegeDropper::new(config.privileges.clone(), config.socket_workers);
+    let mut join_handles = Vec::new();
 
     update_access_list(&config.access_list, &state.access_list)?;
 
@@ -62,14 +63,13 @@ pub fn run(config: Config) -> ::anyhow::Result<()> {
     }
 
     for i in 0..config.swarm_workers {
-        let sentinel = sentinel.clone();
         let config = config.clone();
         let state = state.clone();
         let request_receiver = request_receivers.remove(&i).unwrap().clone();
         let response_sender = ConnectedResponseSender::new(response_senders.clone());
         let statistics_sender = statistics_sender.clone();
 
-        Builder::new()
+        let handle = Builder::new()
             .name(format!("swarm-{:02}", i + 1))
             .spawn(move || {
                 #[cfg(feature = "cpu-pinning")]
@@ -81,7 +81,6 @@ pub fn run(config: Config) -> ::anyhow::Result<()> {
                 );
 
                 let mut worker = SwarmWorker {
-                    _sentinel: sentinel,
                     config,
                     state,
                     server_start_instant,
@@ -91,13 +90,14 @@ pub fn run(config: Config) -> ::anyhow::Result<()> {
                     worker_index: SwarmWorkerIndex(i),
                 };
 
-                worker.run();
+                worker.run()
             })
             .with_context(|| "spawn swarm worker")?;
+
+        join_handles.push((WorkerType::Swarm(i), handle));
     }
 
     for i in 0..config.socket_workers {
-        let sentinel = sentinel.clone();
         let state = state.clone();
         let config = config.clone();
         let connection_validator = connection_validator.clone();
@@ -106,7 +106,7 @@ pub fn run(config: Config) -> ::anyhow::Result<()> {
         let response_receiver = response_receivers.remove(&i).unwrap();
         let priv_dropper = priv_dropper.clone();
 
-        Builder::new()
+        let handle = Builder::new()
             .name(format!("socket-{:02}", i + 1))
             .spawn(move || {
                 #[cfg(feature = "cpu-pinning")]
@@ -118,46 +118,48 @@ pub fn run(config: Config) -> ::anyhow::Result<()> {
                 );
 
                 workers::socket::run_socket_worker(
-                    sentinel,
                     state,
                     config,
                     connection_validator,
                     server_start_instant,
                     request_sender,
                     response_receiver,
                     priv_dropper,
-                );
+                )
             })
             .with_context(|| "spawn socket worker")?;
+
+        join_handles.push((WorkerType::Socket(i), handle));
     }
 
     if config.statistics.active() {
-        let sentinel = sentinel.clone();
         let state = state.clone();
         let config = config.clone();
 
-        #[cfg(feature = "prometheus")]
-        if config.statistics.run_prometheus_endpoint {
-            use metrics_exporter_prometheus::PrometheusBuilder;
-            use metrics_util::MetricKindMask;
+        let handle = Builder::new()
+            .name("statistics".into())
+            .spawn(move || {
+                #[cfg(feature = "cpu-pinning")]
+                pin_current_if_configured_to(
+                    &config.cpu_pinning,
+                    config.socket_workers,
+                    config.swarm_workers,
+                    WorkerIndex::Util,
+                );
 
-            PrometheusBuilder::new()
-                .idle_timeout(
-                    MetricKindMask::ALL,
-                    Some(Duration::from_secs(config.statistics.interval * 2)),
-                )
-                .with_http_listener(config.statistics.prometheus_endpoint_address)
-                .install()
-                .with_context(|| {
-                    format!(
-                        "Install prometheus endpoint on {}",
-                        config.statistics.prometheus_endpoint_address
-                    )
-                })?;
-        }
+                workers::statistics::run_statistics_worker(config, state, statistics_receiver)
+            })
+            .with_context(|| "spawn statistics worker")?;
 
-        Builder::new()
-            .name("statistics".into())
+        join_handles.push((WorkerType::Statistics, handle));
+    }
+
+    #[cfg(feature = "prometheus")]
+    if config.statistics.active() && config.statistics.run_prometheus_endpoint {
+        let config = config.clone();
+
+        let handle = Builder::new()
+            .name("prometheus".into())
             .spawn(move || {
                 #[cfg(feature = "cpu-pinning")]
                 pin_current_if_configured_to(
@@ -167,14 +169,73 @@ pub fn run(config: Config) -> ::anyhow::Result<()> {
                     WorkerIndex::Util,
                 );
 
-                workers::statistics::run_statistics_worker(
-                    sentinel,
-                    config,
-                    state,
-                    statistics_receiver,
+                use metrics_exporter_prometheus::PrometheusBuilder;
+                use metrics_util::MetricKindMask;
+
+                let rt = ::tokio::runtime::Builder::new_current_thread()
+                    .enable_all()
+                    .build()
+                    .context("build prometheus tokio runtime")?;
+
+                rt.block_on(async {
+                    let (recorder, exporter) = PrometheusBuilder::new()
+                        .idle_timeout(
+                            MetricKindMask::ALL,
+                            Some(Duration::from_secs(config.statistics.interval * 2)),
+                        )
+                        .with_http_listener(config.statistics.prometheus_endpoint_address)
+                        .build()
+                        .context("build prometheus recorder and exporter")?;
+
+                    ::tokio::spawn(async move {
+                        let mut interval = ::tokio::time::interval(Duration::from_secs(5));
+
+                        loop {
+                            interval.tick().await;
+
+                            // Periodically render metrics to make sure
+                            // idles are cleaned up
+                            recorder.handle().render();
+                        }
+                    });
+
+                    exporter.await.context("run prometheus exporter")
+                })
+            })
+            .with_context(|| "spawn prometheus exporter worker")?;
+
+        join_handles.push((WorkerType::Prometheus, handle));
+    }
+
+    // Spawn signal handler thread
+    {
+        let config = config.clone();
+
+        let handle: JoinHandle<anyhow::Result<()>> = Builder::new()
+            .name("signals".into())
+            .spawn(move || {
+                #[cfg(feature = "cpu-pinning")]
+                pin_current_if_configured_to(
+                    &config.cpu_pinning,
+                    config.socket_workers,
+                    config.swarm_workers,
+                    WorkerIndex::Util,
                 );
+
+                for signal in &mut signals {
+                    match signal {
+                        SIGUSR1 => {
+                            let _ = update_access_list(&config.access_list, &state.access_list);
+                        }
+                        _ => unreachable!(),
+                    }
+                }
+
+                Ok(())
             })
-            .with_context(|| "spawn statistics worker")?;
+            .context("spawn signal worker")?;
+
+        join_handles.push((WorkerType::Signals, handle));
     }
 
     #[cfg(feature = "cpu-pinning")]
@@ -185,21 +246,47 @@ pub fn run(config: Config) -> ::anyhow::Result<()> {
         WorkerIndex::Util,
     );
 
-    for signal in &mut signals {
-        match signal {
-            SIGUSR1 => {
-                let _ = update_access_list(&config.access_list, &state.access_list);
-            }
-            SIGTERM => {
-                if sentinel_watcher.panic_was_triggered() {
-                    return Err(anyhow::anyhow!("worker thread panicked"));
+    loop {
+        for (i, (_, handle)) in join_handles.iter().enumerate() {
+            if handle.is_finished() {
+                let (worker_type, handle) = join_handles.remove(i);
+
+                match handle.join() {
+                    Ok(Ok(())) => {
+                        return Err(anyhow::anyhow!("{} stopped", worker_type));
+                    }
+                    Ok(Err(err)) => {
+                        return Err(err.context(format!("{} stopped", worker_type)));
+                    }
+                    Err(_) => {
+                        return Err(anyhow::anyhow!("{} panicked", worker_type));
+                    }
                 }
-
-                break;
             }
-            _ => unreachable!(),
         }
+
+        sleep(Duration::from_secs(5));
     }
+}
+
+enum WorkerType {
+    Swarm(usize),
+    Socket(usize),
+    Statistics,
+    Signals,
+    #[cfg(feature = "prometheus")]
+    Prometheus,
+}
 
-    Ok(())
+impl Display for WorkerType {
+    fn fmt(&self, f: &mut std::fmt::Formatter<'_>) -> std::fmt::Result {
+        match self {
+            Self::Swarm(index) => f.write_fmt(format_args!("Swarm worker {}", index + 1)),
+            Self::Socket(index) => f.write_fmt(format_args!("Socket worker {}", index + 1)),
+            Self::Statistics => f.write_str("Statistics worker"),
+            Self::Signals => f.write_str("Signals worker"),
+            #[cfg(feature = "prometheus")]
+            Self::Prometheus => f.write_str("Prometheus worker"),
+        }
+    }
 }