From 9f9d15e2854bcfd2cb56631f9255f7b5dc2c1f03 Mon Sep 17 00:00:00 2001
From: lilong12 <lilong12@baidu.com>
Date: Thu, 17 Sep 2020 15:03:35 +0800
Subject: [PATCH] fix the bug of non-exit, test=develop (#27350)

---
 paddle/fluid/framework/device_worker.h     | 1 +
 paddle/fluid/framework/pipeline_trainer.cc | 1 +
 paddle/fluid/framework/section_worker.cc   | 2 --
 3 files changed, 2 insertions(+), 2 deletions(-)

diff --git a/paddle/fluid/framework/device_worker.h b/paddle/fluid/framework/device_worker.h
index 04befbe1ca0..efe6fa1b2da 100644
--- a/paddle/fluid/framework/device_worker.h
+++ b/paddle/fluid/framework/device_worker.h
@@ -441,6 +441,7 @@ class SectionWorker : public DeviceWorker {
     skip_vars_ = skip_vars;
   }
   static void ResetBatchId() { batch_id_ = 0; }
+  static void ResetThreadCompletedFlag() { threads_completed = false; }
 
   static std::atomic<int> cpu_id_;
 
diff --git a/paddle/fluid/framework/pipeline_trainer.cc b/paddle/fluid/framework/pipeline_trainer.cc
index 758b728fd9c..d7506edbf4c 100644
--- a/paddle/fluid/framework/pipeline_trainer.cc
+++ b/paddle/fluid/framework/pipeline_trainer.cc
@@ -251,6 +251,7 @@ void PipelineTrainer::Finalize() {
   }
   root_scope_->DropKids();
   SectionWorker::ResetBatchId();
+  SectionWorker::ResetThreadCompletedFlag();
 }
 
 Scope* PipelineTrainer::GetWorkerScope(int thread_id) {
diff --git a/paddle/fluid/framework/section_worker.cc b/paddle/fluid/framework/section_worker.cc
index 03b7afbb877..b9a3cac0ec4 100644
--- a/paddle/fluid/framework/section_worker.cc
+++ b/paddle/fluid/framework/section_worker.cc
@@ -196,7 +196,6 @@ void SectionWorker::TrainFiles() {
         if (threads_completed) {
           VLOG(3) << "thread " << thread_id_ << " completed.";
           lk.unlock();
-          threads_completed = false;
           return;
         }
         lk.unlock();
@@ -459,7 +458,6 @@ void SectionWorker::TrainFilesWithProfiler() {
                     << ", mean_time: " << op_total_time[i] / op_count[i];
           }
           VLOG(0) << "================================";
-          threads_completed = false;
           return;
         }
         lk.unlock();
-- 
GitLab