Fix race in event_tracer/repeat unit tests (#139)

* Fix race in event_tracer
* Repeat unit tests 100 times in CI to catch instability

Author: alan-george-lk

.github/workflows/tests.yml

@@ -198,18 +198,22 @@ jobs:
       # ---------- Run unit tests ----------
       - name: Run unit tests (Unix)
         if: runner.os != 'Windows'
-        timeout-minutes: 1
+        timeout-minutes: 10
         shell: bash
         run: |
           build-release/bin/livekit_unit_tests \
+            --gtest_repeat=100 \
+            --gtest_brief=1 \
             --gtest_output=xml:build-release/unit-test-results.xml
 
       - name: Run unit tests (Windows)
         if: runner.os == 'Windows'
-        timeout-minutes: 1
+        timeout-minutes: 10
         shell: pwsh
         run: |
           build-release\bin\livekit_unit_tests.exe `
+            --gtest_repeat=100 `
+            --gtest_brief=1 `
             --gtest_output="xml:build-release\unit-test-results.xml"
 
       # ---------- Install + start livekit-server for integration tests ----------

src/trace/event_tracer.cpp

@@ -408,8 +408,12 @@ void StopTracing() {
   // Disable tracing first to stop new events
   g_tracing_enabled.store(false, std::memory_order_release);
 
-  // Signal writer thread to shut down
-  g_shutdown_requested.store(true);
+  // Signal writer thread to shut down. The shutdown flag must be set while
+  // holding g_mutex so the writer thread cannot miss the notification
+  {
+    const std::scoped_lock<std::mutex> lock(g_mutex);
+    g_shutdown_requested.store(true);
+  }
   g_cv.notify_one();
 
   // Wait for writer thread to finish