fix bug for ctrl server, not response for pushplan

examples/mnist/train/job.prototxt

@@ -7,6 +7,7 @@ default_data_type: kFloat
 use_async_cpu_stream: false
 global_fs_conf {
   hdfs_conf {
+    namenode: "hdfs://192.168.1.11:9000"
   }
 }
 train_conf {

examples/mnist/train/resource.prototxt

 machine {
   addr: "192.168.1.11"
-  port: 9000
+  port: 6666
   name: "192.168.1.11"
 }
 
 machine {
   addr: "192.168.1.13"
-  port: 9000
+  port: 6666
   name: "192.168.1.13"
 }
 

oneflow/core/control/ctrl_server.cpp

@@ -131,16 +131,18 @@ void CtrlServer::WaitUntilDoneHandler(
 void CtrlServer::PushPlanHandler(
     CtrlCall<PushPlanRequest, PushPlanResponse>* call) {
   plan_.reset(new Plan(call->request().plan()));
-  for (auto call : pending_plan_calls_) {
-    *(call->mut_response()->mutable_plan()) = *plan_;
-    call->SendResponse();
+  for (auto pending_call : pending_plan_calls_) {
+    *(pending_call->mut_response()->mutable_plan()) = *plan_;
+    pending_call->SendResponse();
   }
+  call->SendResponse();
   ENQUEUE_REQUEST(PushPlan);
 }
 
 void CtrlServer::ClearPlanHandler(
     CtrlCall<ClearPlanRequest, ClearPlanResponse>* call) {
   plan_.reset();
+  call->SendResponse();
   ENQUEUE_REQUEST(ClearPlan);
 }
 
@@ -158,16 +160,18 @@ void CtrlServer::PullPlanHandler(
 void CtrlServer::PushPortHandler(
     CtrlCall<PushPortRequest, PushPortResponse>* call) {
   port_ = call->request().port();
-  for (auto call : pending_port_calls_) {
-    call->mut_response()->set_port(port_);
-    call->SendResponse();
+  for (auto pending_call : pending_port_calls_) {
+    pending_call->mut_response()->set_port(port_);
+    pending_call->SendResponse();
   }
+  call->SendResponse();
   ENQUEUE_REQUEST(PushPort);
 }
 
 void CtrlServer::ClearPortHandler(
     CtrlCall<ClearPortRequest, ClearPortResponse>* call) {
   port_ = -1;
+  call->SendResponse();
   ENQUEUE_REQUEST(ClearPort);
 }
 

oneflow/core/persistence/file_system.cpp

@@ -2,6 +2,7 @@
 #include <errno.h>
 #include "oneflow/core/common/str_util.h"
 #include "oneflow/core/job/job_desc.h"
+#include "oneflow/core/persistence/hadoop/hadoop_file_system.h"
 #include "oneflow/core/persistence/posix/posix_file_system.h"
 #include "oneflow/core/persistence/windows/windows_file_system.h"
 
@@ -91,8 +92,7 @@ struct GlobalFSConstructor {
       CHECK_EQ(JobDesc::Singleton()->resource().machine().size(), 1);
       gfs = LocalFS();
     } else if (gfs_conf.has_hdfs_conf()) {
-      // static fs::FileSystem* fs = new
-      // fs::HadoopFileSystem(gfs_conf.hdfs_conf()); return fs;
+      gfs = new HadoopFileSystem(gfs_conf.hdfs_conf());
     } else {
       UNEXPECTED_RUN();
     }

oneflow/core/persistence/hadoop/hadoop_file_system.cpp

@@ -128,9 +128,9 @@ void LibHDFS::LoadAndBind() {
 #else
   const char* kLibHdfsDso = "libhdfs.so";
 #endif
-  char* hdfs_home = getenv("HADOOP_HDFS_HOME");
+  char* hdfs_home = getenv("HADOOP_HOME");
   if (hdfs_home == nullptr) {
-    PLOG(WARNING) << "Environment variable HADOOP_HDFS_HOME not set";
+    PLOG(WARNING) << "Environment variable HADOOP_HOME not set";
     status_ = false;
     return;
   }

scripts/run_scheduler.sh

@@ -8,13 +8,14 @@ SCHEDULER_CMD='GLOG_logtostderr=0 GLOG_log_dir=./log GLOG_v=0 GLOG_logbuflevel=-
 set +e
 for host in "${hosts[@]}"
 do
-  ssh $USER@$host "/usr/sbin/fuser -k 9000/tcp"
+  ssh $USER@$host "/usr/sbin/fuser -k 6666/tcp"
+  ssh $USER@$host "mkdir ~/oneflow_temp"
 done
 set -e
 
 for host in "${hosts[@]}"
 do
-  ssh $USER@$host 'rm -rf ~/oneflow_temp && mkdir ~/oneflow_temp'
+  ssh $USER@$host 'rm -rf ~/oneflow_temp/*'
   scp ./compiler ./runtime ./scheduler ./*.prototxt $USER@$host:~/oneflow_temp
   ssh $USER@$host "cd ~/oneflow_temp; $SCHEDULER_CMD -this_machine_name=$host &"
 done