Add test watchdog

Signed-off-by: Ludvig Liljenberg <4257730+ludfjig@users.noreply.github.com>

Author: ludfjig

src/hyperlight_host/tests/integration_test.rs

@@ -1050,6 +1050,7 @@ fn interrupt_random_kill_stress_test() {
                 let sandbox_wrapper = guard.sandbox_with_snapshot.as_mut().unwrap();
                 let sandbox = &mut sandbox_wrapper.sandbox;
                 let interrupt_handle = sandbox.interrupt_handle();
+                let watchdog_handle = interrupt_handle.clone();
 
                 // If we decided to kill, spawn a thread that will kill at a random time
                 // Use a barrier to ensure the killer thread waits until we're about to call the guest
@@ -1110,7 +1111,32 @@ fn interrupt_random_kill_stress_test() {
                     );
                 }
 
+                // Watchdog: if the guest call doesn't return within 300s,
+                // force-kill it and print diagnostics.
+                let watchdog_thread_id = thread_id;
+                let watchdog_iteration = iteration;
+                let watchdog_should_kill = should_kill;
+                let watchdog_cancel = Arc::new(AtomicBool::new(false));
+                let watchdog_cancel_clone = watchdog_cancel.clone();
+                let watchdog = thread::spawn(move || {
+                    for _ in 0..300 {
+                        thread::sleep(Duration::from_secs(1));
+                        if watchdog_cancel_clone.load(Ordering::Relaxed) {
+                            return;
+                        }
+                    }
+                    println!(
+                        "WATCHDOG: Thread {} iteration {} HUNG for 300s (should_kill={}), force-killing",
+                        watchdog_thread_id, watchdog_iteration, watchdog_should_kill
+                    );
+                    watchdog_handle.kill();
+                });
+
                 let result = sandbox.call::<u64>("SpinForMs", GUEST_CALL_DURATION_MS);
+
+                // Signal watchdog to exit and wait for it
+                watchdog_cancel.store(true, Ordering::Relaxed);
+                let _ = watchdog.join();
                 trace!(
                     "[THREAD-{}] Iteration {}: Guest call returned: {:?}",
                     thread_id,