Changeset 2324

Timestamp:

04/15/22 13:05:33 (2 years ago)

Author:

ymipsl

Message:

Solve deadlock or crash occuring when activate second levels of servers.

YM

Location:

XIOS/dev/dev_ym/XIOS_COUPLING/src

Files:

• buffer_client.cpp (modified) (2 diffs)
• context_client.cpp (modified) (3 diffs)
• context_client.hpp (modified) (2 diffs)
• node/context.cpp (modified) (5 diffs)

XIOS/dev/dev_ym/XIOS_COUPLING/src/buffer_client.cpp

@@ -144 +144 @@
     if (hasWindows)
     {
+      if (winState[current]==true) ERROR("CClientBuffer::lockBuffer(void)",<<"Try lo lock client buffer but winState said it is already locked") ;
       MPI_Win_lock(MPI_LOCK_EXCLUSIVE,clientRank_, 0, windows_[current]) ;
       winState[current]=true ;
@@ -155 +156 @@
     if (hasWindows)
     {
+      if (winState[current]==false) ERROR("CClientBuffer::lockBuffer(void)",<<"Try lo unlock client buffer but winState said it is already unlocked") ;
       MPI_Win_unlock(clientRank_, windows_[current]) ;
       winState[current]=false ;

XIOS/dev/dev_ym/XIOS_COUPLING/src/context_client.cpp

@@ -175 +175 @@
       if (isAttachedModeEnabled()) // couldBuffer is always true in attached mode
       {
-        while (checkBuffers(ranks)) context_->globalEventLoop() ;
+        while (checkBuffers(ranks)) callGlobalEventLoop() ;
       
         CXios::getDaemonsManager()->scheduleContext(hashId_) ;
-        while (CXios::getDaemonsManager()->isScheduledContext(hashId_)) context_->globalEventLoop() ;
+        while (CXios::getDaemonsManager()->isScheduledContext(hashId_)) callGlobalEventLoop() ;
       }
       
@@ -287 +287 @@
           checkBuffers();
 
-          context_->globalEventLoop() ;
+          callGlobalEventLoop() ;
         }
 
@@ -301 +301 @@
    }
 
+   void CContextClient::eventLoop(void)
+   {
+      if (!locked_) checkBuffers() ;
+   }
+
+   void CContextClient::callGlobalEventLoop(void)
+   {
+     locked_=true ;
+     context_->globalEventLoop() ;
+     locked_=false ;
+   }
    /*!
    Make a new buffer for a certain connection to server with specific rank

XIOS/dev/dev_ym/XIOS_COUPLING/src/context_client.hpp

@@ -39 +39 @@
       bool checkBuffers(list<int>& ranks);
       bool checkBuffers(void);
+      void eventLoop(void) ;
+      void callGlobalEventLoop() ;
       void releaseBuffers(void);
       bool havePendingRequests(void);
@@ -129 +131 @@
 
       double latency_=0e-2 ;
+
+      bool locked_ = false ; //!< The context client is locked to avoid recursive checkBuffer
   };
 }

XIOS/dev/dev_ym/XIOS_COUPLING/src/node/context.cpp

@@ -684 +684 @@
     setCurrent(getId()) ;
 
-    if (client!=nullptr && !finalized) client->checkBuffers();
+    if (client!=nullptr && !finalized) client->eventLoop();
     
     for (int i = 0; i < clientPrimServer.size(); ++i)
     {
-      if (!finalized) clientPrimServer[i]->checkBuffers();
+      if (!finalized) clientPrimServer[i]->eventLoop();
       if (!finalized) finished &= serverPrimServer[i]->eventLoop(enableEventsProcessing);
     }
 
     for (auto couplerOut : couplerOutClient_)
-      if (!finalized) couplerOut.second->checkBuffers();
+      if (!finalized) couplerOut.second->eventLoop();
     
     for (auto couplerIn : couplerInClient_)
-      if (!finalized) couplerIn.second->checkBuffers();
+      if (!finalized) couplerIn.second->eventLoop();
     
     for (auto couplerOut : couplerOutServer_)
@@ -788 +788 @@
         client->finalize();
         info(100)<<"DEBUG: context "<<getId()<<" Client finalize sent"<<endl ;
-        while (client->havePendingRequests()) client->checkBuffers();
+        while (client->havePendingRequests()) client->eventLoop();
         info(100)<<"DEBUG: context "<<getId()<<" no pending request ok"<<endl ;
         bool notifiedFinalized=false ;
@@ -808 +808 @@
            do
            {
-             clientPrimServer[i]->checkBuffers();
+             clientPrimServer[i]->eventLoop();
              bufferReleased = !clientPrimServer[i]->havePendingRequests();
            } while (!bufferReleased);
@@ -1099 +1099 @@
     {
       for(auto field : fileOutField) slaveServers_.insert(field->getContextClient()) ; 
-      for(auto field : fileInField)  slaveServers_.insert(field->getContextClient()) ;  
     }
 
@@ -1709 +1708 @@
    TRY
    {
-     CEventClient event(getType(),EVENT_ID_UPDATE_CALENDAR);
      for(auto client : slaveServers_) 
      {
+       CEventClient event(getType(),EVENT_ID_UPDATE_CALENDAR);
        if (client->isServerLeader())
        {