feat: Add more metrics around phoenix and game traffic

src/cli.rs

@@ -223,6 +223,7 @@ impl Cli {
         #[cfg(target_os = "linux")]
         let mut sig_term_fut = signal::unix::signal(signal::unix::SignalKind::terminate())?;
 
+        crate::metrics::shutdown_initiated().set(false as _);
         tokio::spawn(async move {
             #[cfg(target_os = "linux")]
             let sig_term = sig_term_fut.recv();
@@ -234,6 +235,7 @@ impl Cli {
                 _ = sig_term => "SIGTERM",
             };
 
+            crate::metrics::shutdown_initiated().set(true as _);
             tracing::info!(%signal, "shutting down from signal");
             // Don't unwrap in order to ensure that we execute
             // any subsequent shutdown tasks.

src/components/proxy/io_uring_shared.rs

@@ -445,6 +445,7 @@ impl IoUringLoop {
         std::thread::Builder::new()
             .name(thread_name)
             .spawn(move || {
+                crate::metrics::game_traffic_tasks().inc();
                 let _guard = tracing::dispatcher::set_default(&dispatcher);
 
                 let tokens = slab::Slab::with_capacity(concurrent_sends + 1 + 1);
@@ -490,14 +491,14 @@ impl IoUringLoop {
                         }
                         Err(error) => {
                             tracing::error!(%error, "io-uring submit_and_wait failed");
-                            return;
+                            break 'io;
                         }
                     }
                     cq.sync();
 
                     if let Err(error) = loop_ctx.process_backlog(&submitter) {
                         tracing::error!(%error, "failed to process io-uring backlog");
-                        return;
+                        break 'io;
                     }
 
                     // Now actually process all of the completed io requests
@@ -570,6 +571,8 @@ impl IoUringLoop {
 
                     loop_ctx.sync();
                 }
+
+                crate::metrics::game_traffic_task_closed().inc();
             })?;
 
         Ok(())

src/components/proxy/packet_router/reference.rs

@@ -37,6 +37,7 @@ impl super::DownstreamReceiveWorkerConfig {
         let (tx, mut rx) = tokio::sync::oneshot::channel();
 
         let worker = uring_spawn!(thread_span, async move {
+            crate::metrics::game_traffic_tasks().inc();
             let mut last_received_at = None;
             let socket = crate::net::DualStackLocalSocket::new(port)
                 .unwrap()
@@ -143,6 +144,7 @@ impl super::DownstreamReceiveWorkerConfig {
                         }
                     }
                     _ = &mut rx => {
+                        crate::metrics::game_traffic_task_closed().inc();
                         tracing::debug!("Closing downstream socket loop, shutdown requested");
                         return;
                     }

src/metrics.rs

@@ -135,6 +135,97 @@ impl<'a> From<Option<&'a MetricsIpNetEntry>> for AsnInfo<'a> {
     }
 }
 
+pub(crate) fn shutdown_initiated() -> &'static IntGauge {
+    static SHUTDOWN_INITATED: Lazy<IntGauge> = Lazy::new(|| {
+        prometheus::register_int_gauge_with_registry! {
+            prometheus::opts! {
+                "shutdown_initiated",
+                "Shutdown process has been started",
+            },
+            registry(),
+        }
+        .unwrap()
+    });
+
+    &SHUTDOWN_INITATED
+}
+
+pub(crate) fn game_traffic_tasks() -> &'static IntCounter {
+    static GAME_TRAFFIC_TASKS: Lazy<IntCounter> = Lazy::new(|| {
+        prometheus::register_int_counter_with_registry! {
+            prometheus::opts! {
+                "game_traffic_tasks",
+                "The amount of game traffic tasks that have spawned",
+            },
+            registry(),
+        }
+        .unwrap()
+    });
+
+    &GAME_TRAFFIC_TASKS
+}
+
+pub(crate) fn game_traffic_task_closed() -> &'static IntCounter {
+    static GAME_TRAFFIC_TASK_CLOSED: Lazy<IntCounter> = Lazy::new(|| {
+        prometheus::register_int_counter_with_registry! {
+            prometheus::opts! {
+                "game_traffic_task_closed",
+                "The amount of game traffic tasks that have shutdown",
+            },
+            registry(),
+        }
+        .unwrap()
+    });
+
+    &GAME_TRAFFIC_TASK_CLOSED
+}
+
+pub(crate) fn phoenix_requests() -> &'static IntCounter {
+    static PHOENIX_REQUESTS: Lazy<IntCounter> = Lazy::new(|| {
+        prometheus::register_int_counter_with_registry! {
+            prometheus::opts! {
+                "phoenix_requests",
+                "The amount of phoenix requests",
+            },
+            registry(),
+        }
+        .unwrap()
+    });
+
+    &PHOENIX_REQUESTS
+}
+
+pub(crate) fn phoenix_task_closed() -> &'static IntGauge {
+    static PHOENIX_TASK_CLOSED: Lazy<IntGauge> = Lazy::new(|| {
+        prometheus::register_int_gauge_with_registry! {
+            prometheus::opts! {
+                "phoenix_task_closed",
+                "Whether the phoenix task has shutdown",
+            },
+            registry(),
+        }
+        .unwrap()
+    });
+
+    &PHOENIX_TASK_CLOSED
+}
+
+pub(crate) fn phoenix_server_errors(error: &str) -> IntCounter {
+    static PHOENIX_SERVER_ERRORS: Lazy<IntCounterVec> = Lazy::new(|| {
+        prometheus::register_int_counter_vec_with_registry! {
+            prometheus::opts! {
+                "phoenix_server_errors",
+                "The amount of errors attempting to spawn the phoenix HTTP server",
+            },
+            &["error"],
+            registry(),
+        }
+        .unwrap()
+    });
+
+    PHOENIX_SERVER_ERRORS.with_label_values(&[error])
+}
+
 pub(crate) fn processing_time(direction: Direction) -> Histogram {
     static PROCESSING_TIME: Lazy<HistogramVec> = Lazy::new(|| {
         prometheus::register_histogram_vec_with_registry! {

src/net/phoenix.rs

@@ -107,6 +107,7 @@ pub fn spawn<M: Clone + Measurement + Sync + Send + 'static>(
                                                             "application/json",
                                                         );
 
+                                                    crate::metrics::phoenix_requests().inc();
                                                     tracing::trace!("serving phoenix request");
                                                     Ok::<_, std::convert::Infallible>(
                                                         Response::builder()
@@ -127,13 +128,16 @@ pub fn spawn<M: Clone + Measurement + Sync + Send + 'static>(
                                                 .serve_connection(conn, svc)
                                                 .await
                                             {
+                                                let error_display = err.to_string();
+                                                crate::metrics::phoenix_server_errors(&error_display).inc();
                                                 tracing::error!(
-                                                    "failed to reponse to phoenix request: {err}"
+                                                    "failed to respond to phoenix request: {error_display}"
                                                 );
                                             }
                                         });
                                     }
                                     _ = &mut srx => {
+                                        crate::metrics::phoenix_task_closed().set(true as _);
                                         tracing::info!("shutting down phoenix HTTP service");
                                         break;
                                     }