diff --git a/static_core/libpandabase/taskmanager/utils/sp_sc_lock_free_queue.h b/static_core/libpandabase/taskmanager/utils/sp_sc_lock_free_queue.h
index 2431d03ef76e2adeba817aba2e51f531d6c350f7..d0fc86fe11cf2960ac86f77c9915f3783ea5a684 100644
--- a/static_core/libpandabase/taskmanager/utils/sp_sc_lock_free_queue.h
+++ b/static_core/libpandabase/taskmanager/utils/sp_sc_lock_free_queue.h
@@ -199,13 +199,15 @@ public:
     {
         static_assert(ALLOCATION_TYPE == QueueElemAllocationType::INPLACE);
         auto inode = static_cast<QueueBigNode *>(node);
-        TSAN_ANNOTATE_IGNORE_WRITES_BEGIN();
-        // Atomic with relaxed order reason: gets correct value
-        auto count = inode->toDeleteCount.fetch_sub(1, std::memory_order_relaxed);
+
+        // There is atomic because the queue is used by multiple consumers using mutex, but this function is executed concurrently
+        // Atomic with release order reason: we need to sychronize an unprotected access to the node content with deletion of the node
+        auto count = inode->toDeleteCount.fetch_sub(1, std::memory_order_release);
         if (count == 1) {
+            // Atomic with acquire order reason: need to synchronize the last fetch_sub with other fetch_sub's (see above)
+            inode->toDeleteCount.load(std::memory_order_acquire);
             DeleteQueueBigNode(inode);
         }
-        TSAN_ANNOTATE_IGNORE_WRITES_END();
     }
 
 private: