feat(auth): add exponential backoff with jitter on token refresh failures (fixes #1629)

Author: AustinBenoit

auth/src/desktop/auth_desktop.cc

@@ -681,6 +681,8 @@ void IdTokenRefreshThread::Initialize(AuthData* auth_data) {
   thread_ = firebase::Thread(
       [](IdTokenRefreshThread* refresh_thread) {
         Auth* auth = refresh_thread->auth;
+        int retry_backoff_ms = kMinRetryBackoffMs;
+
         while (!refresh_thread->is_shutting_down()) {
           // Note:  Make sure to always make future_impl.mutex the innermost
           // lock, to prevent deadlocks!
@@ -714,10 +716,20 @@ void IdTokenRefreshThread::Initialize(AuthData* auth_data) {
               // is completed.
               future_sem.Wait();
 
-              // If token refresh failed (e.g. network outage), wait before retrying
-              // to prevent a 0ms tight retry loop and high CPU usage.
-              if (future.error() != 0 && !refresh_thread->is_shutting_down()) {
-                refresh_thread->wakeup_sem_.TimedWait(kMsPerTokenRefreshFailedRetry);
+              if (future.error() != 0) {
+                // Token refresh failed (e.g. network outage). Wait with exponential backoff
+                // to prevent tight retry loops and avoid overloading backend servers upon reconnect.
+                if (!refresh_thread->is_shutting_down()) {
+                  int jitter_ms = rand() % 1000;
+                  int wait_time_ms = retry_backoff_ms + jitter_ms;
+                  // Double backoff for consecutive failures up to max 30 minutes.
+                  retry_backoff_ms = std::min(kMaxRetryBackoffMs, retry_backoff_ms * 2);
+                  refresh_thread->wakeup_sem_.TimedWait(wait_time_ms);
+                }
+                continue;
+              } else {
+                // Refresh succeeded! Reset backoff to minimum interval.
+                retry_backoff_ms = kMinRetryBackoffMs;
               }
             } else {
               auth->auth_data_->future_impl.mutex().Release();

auth/src/desktop/auth_desktop.h

@@ -179,8 +179,9 @@ struct AuthImpl {
 const int kMinutesPerTokenRefresh = 58;
 const int kMsPerTokenRefresh =
     kMinutesPerTokenRefresh * internal::kMillisecondsPerMinute;
-// Retry delay when token refresh fails (e.g., due to network outage).
-const int kMsPerTokenRefreshFailedRetry = 30 * 1000;
+// Exponential backoff parameters when token refresh fails (e.g. network outage).
+const int kMinRetryBackoffMs = 30 * 1000;        // 30 seconds base delay
+const int kMaxRetryBackoffMs = 30 * 60 * 1000;   // 30 minutes max delay
 
 void InitializeUserDataPersist(AuthData* auth_data);
 void DestroyUserDataPersist(AuthData* auth_data);