diff --git a/paddle/fluid/memory/allocation/allocator_facade.cc b/paddle/fluid/memory/allocation/allocator_facade.cc
index e2730a1b825e9a277b72b9e5a28f503cc645953a..e2649a7fd334d6dab6dde44792024f251da84a4f 100644
--- a/paddle/fluid/memory/allocation/allocator_facade.cc
+++ b/paddle/fluid/memory/allocation/allocator_facade.cc
@@ -415,6 +415,23 @@ class AllocatorFacadePrivate {
   void SetDefaultStream(const platform::CUDAPlace& place, gpuStream_t stream) {
     const std::shared_ptr<StreamSafeCUDAAllocator>& allocator =
         GetDefaultStreamSafeCUDAAllocator(place);
+
+    // NOTE(Ruibiao): The default stream will be set when the CUDADeviceContext
+    // created. Normally, the DeviceContextPool is a global singleton and one
+    // Place only correspond to one DeviceContext. However, to support
+    // multi-stream scheduling, standalone executor creates two extra
+    // DeviceContextPools for H2D and D2H stream in StreamAnalyzer, which make
+    // one Place correspond to multiple DeviceContext and unexpectedly reset the
+    // default stream in runtime. To avoid this behavior, we do not allow
+    // changing default stream after initially setting.
+    if (allocator->GetDefaultStream() != nullptr) {
+      VLOG(5) << "The default stream for StreamSafeCUDAAllocator("
+              << allocator.get() << ") in " << place << " has been set to "
+              << allocator->GetDefaultStream()
+              << " before, not allow to change now.";
+      return;
+    }
+
     allocator->SetDefaultStream(stream);
     VLOG(8) << "Set default stream to " << stream
             << " for StreamSafeCUDAAllocator(" << allocator.get() << ") in "