[fleet_executor] fix message bus bug (#37507)

PaddlePaddle · Nov 24, 2021 · 10d8d6b · 10d8d6b
1 parent 7de99d8
commit 10d8d6b
Show file tree

Hide file tree

Showing 3 changed files with 3 additions and 3 deletions.
diff --git a/paddle/fluid/distributed/fleet_executor/message_bus.cc b/paddle/fluid/distributed/fleet_executor/message_bus.cc
@@ -111,8 +111,7 @@ void MessageBus::ListenPort() {
 #if defined(PADDLE_WITH_DISTRIBUTE) && defined(PADDLE_WITH_PSCORE) && \
     !defined(PADDLE_WITH_ASCEND_CL)
   // function keep listen the port and handle the message
-  InterceptorMessageServiceImpl interceptor_message_service;
-  PADDLE_ENFORCE_EQ(server_.AddService(&interceptor_message_service,
+  PADDLE_ENFORCE_EQ(server_.AddService(&interceptor_message_service_,
                                        brpc::SERVER_DOESNT_OWN_SERVICE),
                     0, platform::errors::Unavailable(
                            "Message bus: init brpc service error."));

diff --git a/paddle/fluid/distributed/fleet_executor/message_bus.h b/paddle/fluid/distributed/fleet_executor/message_bus.h
@@ -89,6 +89,7 @@ class MessageBus final {
 
 #if defined(PADDLE_WITH_DISTRIBUTE) && defined(PADDLE_WITH_PSCORE) && \
     !defined(PADDLE_WITH_ASCEND_CL)
+  InterceptorMessageServiceImpl interceptor_message_service_;
   // brpc server
   brpc::Server server_;
 #endif

diff --git a/paddle/fluid/distributed/fleet_executor/runtime_graph.cc b/paddle/fluid/distributed/fleet_executor/runtime_graph.cc
@@ -142,7 +142,7 @@ void RuntimeGraph::SplitProgramBasedFunctionality(const ProgramDesc& program) {
   int pipeline_stage = coord.pp_idx;
   int64_t num_pipeline_stages = exe_desc_.pp_degree();
   // TODO(fleet_executor dev): start up steps should be a config `num_slots`
-  int64_t start_up_steps = num_pipeline_stages - pipeline_stage - 1;
+  int64_t start_up_steps = num_pipeline_stages - pipeline_stage;
   int64_t num_micro_batches = exe_desc_.num_micro_batches();
   int64_t task_id = cur_rank * functionality_order.size();
   for (std::size_t i = 0; i < functionality_order.size(); ++i) {