Changeset 492 for XIOS/trunk

Timestamp:

10/02/14 17:34:02 (10 years ago)

Author:

ymipsl

Message:

Add event scheduler functionnality in order to schedule events from different context, that cause Deadlock or crash when using collective MPI communication in netcdf/hdf5 library.

YM

Location:

XIOS/trunk/src

Files:

• context_server.cpp (modified) (4 diffs)
• context_server.hpp (modified) (1 diff)
• event_scheduler.cpp (added)
• event_scheduler.hpp (added)
• server.cpp (modified) (7 diffs)
• server.hpp (modified) (2 diffs)
• xmlioserver_spl.hpp (modified) (1 diff)

XIOS/trunk/src/context_server.cpp

@@ -14 +14 @@
 #include "timer.hpp"
 #include "cxios.hpp"
+#include "event_scheduler.hpp"
+#include "server.hpp"
+#include <boost/functional/hash.hpp>
 
 
@@ -32 +35 @@
     else  MPI_Comm_size(interComm,&commSize) ;
     currentTimeLine=0 ;
+    scheduled=false ;
     finished=false ;
+    
+    boost::hash<string> hashString ;
+    hashId=hashString(context->getId()) ;
+
   }
   void CContextServer::setPendingEvent(void)
@@ -150 +158 @@
     {
       event=it->second ;
+      
       if (event->isFull())
       {
+        if (!scheduled && !CXios::isServer)
+        {
+          CServer::eventScheduler->registerEvent(currentTimeLine,hashId) ;  
+          scheduled=true ;
+        }
+        else if (CXios::isServer || CServer::eventScheduler->queryEvent(currentTimeLine,hashId) ) 
+        {
          CTimer::get("Process events").resume() ;
          dispatchEvent(*event) ;
@@ -159 +175 @@
          events.erase(it) ;
          currentTimeLine++ ;
-       }
-     }
-   }
+         scheduled = false ;
+        }
+      }
+    }
+  }
 
   CContextServer::~CContextServer()

XIOS/trunk/src/context_server.hpp

@@ -41 +41 @@
     bool finished ;
     bool pendingEvent ;
+    bool scheduled  ;    /*!< event of current timeline is alreading scheduled ? */
+    size_t hashId ;
     ~CContextServer() ;    
   } ;

XIOS/trunk/src/server.cpp

@@ -12 +12 @@
 #include "tracer.hpp"
 #include "timer.hpp"
+#include "event_scheduler.hpp"
 
 namespace xios
@@ -23 +24 @@
     bool CServer::finished=false ;
     bool CServer::is_MPI_Initialized ;
+    CEventScheduler* CServer::eventScheduler ;
 
     void CServer::initialize(void)
@@ -86 +88 @@
            {
              clientLeader=it->second ;
+             int intraCommSize, intraCommRank ;
+             MPI_Comm_size(intraComm,&intraCommSize) ;
+             MPI_Comm_rank(intraComm,&intraCommRank) ;
+             cout<<"intercommCreate::server "<<rank<<" intraCommSize : "<<intraCommSize
+                 <<" intraCommRank :"<<intraCommRank<<"  clientLeader "<< clientLeader<<endl ;
 
              MPI_Intercomm_create(intraComm,0,CXios::globalComm,clientLeader,0,&newComm) ;
@@ -122 +129 @@
           interComm.push_back(newComm) ;
         }
-        oasis_enddef() ;
+              oasis_enddef() ;
       }
 
@@ -129 +136 @@
       if (rank==0) isRoot=true;
       else isRoot=false;
-//      eventLoop() ;
-//      finalize() ;
+      
+      eventScheduler = new CEventScheduler(intraComm) ;
     }
 
@@ -136 +143 @@
     {
       CTimer::get("XIOS").suspend() ;
+      
+      delete eventScheduler ;
+      
       if (!is_MPI_Initialized)
       {
@@ -166 +176 @@
          contextEventLoop() ;
          if (finished && contextList.empty()) stop=true ;
+         if (! CXios::isServer) eventScheduler->checkEvent() ;
        }
        CTimer::get("XIOS server").suspend() ;

XIOS/trunk/src/server.hpp

@@ -5 +5 @@
 #include "context.hpp"
 #include "mpi.hpp"
+#include "event_scheduler.hpp"
 
 namespace xios
@@ -25 +26 @@
        static MPI_Comm intraComm ;
        static list<MPI_Comm> interComm ;
+       static CEventScheduler* eventScheduler ;
 
        struct contextMessage

XIOS/trunk/src/xmlioserver_spl.hpp

@@ -20 +20 @@
 #include <map>
 #include <deque>
+#include <queue>
 #include <valarray>
 // Flux.