Changeset 1377 for XIOS

Timestamp:

12/13/17 16:06:32 (6 years ago)

Author:

ymipsl

Message:

• Add more debugging information at info_level 100
• Enforce synchronisation check for event on client side, using parameter "check_event_sync" at true

YM

Location:

XIOS/dev/XIOS_DEV_CMIP6/src

Files:

• context_client.cpp (modified) (5 diffs)
• cxios.cpp (modified) (2 diffs)
• cxios.hpp (modified) (1 diff)
• event_client.cpp (modified) (2 diffs)
• event_client.hpp (modified) (1 diff)
• server.cpp (modified) (2 diffs)

XIOS/dev/XIOS_DEV_CMIP6/src/context_client.cpp

@@ -96 +96 @@
       list<int> ranks = event.getRanks();
 
+      if (CXios::checkEventSync)
+      {
+        int typeId, classId, typeId_in, classId_in, timeLine_out;
+        typeId_in=event.getTypeId() ;
+        classId_in=event.getClassId() ;
+        MPI_Allreduce(&timeLine,&timeLine_out, 1, MPI_UINT64_T, MPI_SUM, intraComm) ;
+        MPI_Allreduce(&typeId_in,&typeId, 1, MPI_INT, MPI_SUM, intraComm) ;
+        MPI_Allreduce(&classId_in,&classId, 1, MPI_INT, MPI_SUM, intraComm) ;
+        if (typeId/clientSize!=event.getTypeId() || classId/clientSize!=event.getClassId() || timeLine_out/clientSize!=timeLine)
+        {
+           ERROR("void CContextClient::sendEvent(CEventClient& event)",
+               << "Event are not coherent between client.");
+        }
+      }
+
       if (!event.isEmpty())
       {
@@ -124 +139 @@
           for (list<int>::const_iterator it = sizes.begin(); it != sizes.end(); it++)
             tmpBufferedEvent.buffers.push_back(new CBufferOut(*it));
-
+          info(100)<<"DEBUG : temporaly event created : timeline "<<timeLine<<endl ;
           event.send(timeLine, tmpBufferedEvent.sizes, tmpBufferedEvent.buffers);
         }
@@ -151 +166 @@
             (*itBuffer)->put((char*)(*it)->start(), (*it)->count());
 
+          info(100)<<"DEBUG : temporaly event sent "<<endl ;
           checkBuffers(tmpBufferedEvent.ranks);
 
@@ -230 +246 @@
             for (int i = 0; i < context->serverPrimServer.size(); ++i)
               context->serverPrimServer[i]->listen();
+//            CServer::contextEventLoop(false) ; // avoid dead-lock at finalize...
           }
 
@@ -406 +423 @@
       const std::list<int>& ranks = getRanksServerLeader();
       for (std::list<int>::const_iterator itRank = ranks.begin(), itRankEnd = ranks.end(); itRank != itRankEnd; ++itRank)
+      {
+        info(100)<<"DEBUG : Sent context Finalize event to rank "<<*itRank<<endl ;
         event.push(*itRank, 1, msg);
+      }
       sendEvent(event);
     }

XIOS/dev/XIOS_DEV_CMIP6/src/cxios.cpp

@@ -37 +37 @@
   CRegistry* CXios::globalRegistry = 0;
   double CXios::recvFieldTimeout = 300.0;
+  bool CXios::checkEventSync=false ;
  
   //! Parse configuration file and create some objects from it
@@ -77 +78 @@
     if (recvFieldTimeout < 0.0)
       ERROR("CXios::parseXiosConfig()", "recv_field_timeout cannot be negative.");
+
+    checkEventSync = getin<bool>("check_event_sync", checkEventSync);
 
     globalComm=MPI_COMM_WORLD ;

XIOS/dev/XIOS_DEV_CMIP6/src/cxios.hpp

@@ -52 +52 @@
      static CRegistry* globalRegistry ; //!< global registry which is wrote by the root process of the servers
      static double recvFieldTimeout; //!< Time to wait for data before issuing an error when receiving a field
+     static bool checkEventSync; //!< For debuuging, check if event are coherent and synchrone on client side
 
     public:

XIOS/dev/XIOS_DEV_CMIP6/src/event_client.cpp

@@ -5 +5 @@
 #include "type.hpp"
 #include "mpi.hpp"
+#include "cxios.hpp"
 
 namespace xios
@@ -49 +50 @@
      std::list<CMessage*>::iterator itMsg = messages.begin();
 
+     if (CXios::checkEventSync) info(100)<<"Send event "<<timeLine<<" classId : "<<classId<<"  typeId : "<<typeId<<endl ;
      for (; itBuff != buffers.end(); ++itBuff, ++itSizes, ++itSenders, ++itMsg)
      {

XIOS/dev/XIOS_DEV_CMIP6/src/event_client.hpp

@@ -21 +21 @@
       std::list<int> getRanks(void);
       std::list<int> getSizes(void);
-
+      int getClassId(void) { return classId; }
+      int getTypeId(void) { return typeId; }
+      
     private:
       int classId;

XIOS/dev/XIOS_DEV_CMIP6/src/server.cpp

@@ -734 +734 @@
      }
 
-     void CServer::contextEventLoop(void)
+     void CServer::contextEventLoop(bool enableEventsProcessing /*= true*/)
      {
        bool isFinalized ;
@@ -748 +748 @@
          }
          else
-           it->second->checkBuffersAndListen();
+           it->second->checkBuffersAndListen(enableEventsProcessing);
        }
      }