metrics: fix deadlock in injection_queue_depth_multi_thread test (#6916)

Author: jofas

tokio/tests/rt_metrics.rs

@@ -2,7 +2,8 @@
 #![warn(rust_2018_idioms)]
 #![cfg(all(feature = "full", not(target_os = "wasi"), target_has_atomic = "64"))]
 
-use std::sync::{Arc, Barrier};
+use std::sync::mpsc;
+use std::time::Duration;
 use tokio::runtime::Runtime;
 
 #[test]
@@ -68,36 +69,51 @@ fn global_queue_depth_multi_thread() {
     let rt = threaded();
     let metrics = rt.metrics();
 
-    let barrier1 = Arc::new(Barrier::new(3));
-    let barrier2 = Arc::new(Barrier::new(3));
-
-    // Spawn a task per runtime worker to block it.
-    for _ in 0..2 {
-        let barrier1 = barrier1.clone();
-        let barrier2 = barrier2.clone();
-        rt.spawn(async move {
-            barrier1.wait();
-            barrier2.wait();
-        });
-    }
-
-    barrier1.wait();
+    for _ in 0..10 {
+        if let Ok(_blocking_tasks) = try_block_threaded(&rt) {
+            for i in 0..10 {
+                assert_eq!(i, metrics.global_queue_depth());
+                rt.spawn(async {});
+            }
 
-    let mut fail: Option<String> = None;
-    for i in 0..10 {
-        let depth = metrics.global_queue_depth();
-        if i != depth {
-            fail = Some(format!("{i} is not equal to {depth}"));
-            break;
+            return;
         }
-        rt.spawn(async {});
     }
 
-    barrier2.wait();
+    panic!("exhausted every try to block the runtime");
+}
 
-    if let Some(fail) = fail {
-        panic!("{fail}");
+fn try_block_threaded(rt: &Runtime) -> Result<Vec<mpsc::Sender<()>>, mpsc::RecvTimeoutError> {
+    let (tx, rx) = mpsc::channel();
+
+    let blocking_tasks = (0..rt.metrics().num_workers())
+        .map(|_| {
+            let tx = tx.clone();
+            let (task, barrier) = mpsc::channel();
+
+            // Spawn a task per runtime worker to block it.
+            rt.spawn(async move {
+                tx.send(()).unwrap();
+                barrier.recv().ok();
+            });
+
+            task
+        })
+        .collect();
+
+    // Make sure the previously spawned tasks are blocking the runtime by
+    // receiving a message from each blocking task.
+    //
+    // If this times out we were unsuccessful in blocking the runtime and hit
+    // a deadlock instead (which might happen and is expected behaviour).
+    for _ in 0..rt.metrics().num_workers() {
+        rx.recv_timeout(Duration::from_secs(1))?;
     }
+
+    // Return senders of the mpsc channels used for blocking the runtime as a
+    // surrogate handle for the tasks. Sending a message or dropping the senders
+    // will unblock the runtime.
+    Ok(blocking_tasks)
 }
 
 fn current_thread() -> Runtime {