source: XIOS/branchs/xios-2.5/src/context_client.cpp @ 2207

Last change on this file since 2207 was 2079, checked in by ymipsl, 3 years ago

bug fix when checking event synchronisation
YM

  • Property copyright set to
    Software name : XIOS (Xml I/O Server)
    http://forge.ipsl.jussieu.fr/ioserver
    Creation date : January 2009
    Licence : CeCCIL version2
    see license file in root directory : Licence_CeCILL_V2-en.txt
    or http://www.cecill.info/licences/Licence_CeCILL_V2-en.html
    Holder : CEA/LSCE (Laboratoire des Sciences du CLimat et de l'Environnement)
    CNRS/IPSL (Institut Pierre Simon Laplace)
    Project Manager : Yann Meurdesoif
    yann.meurdesoif@cea.fr
  • Property svn:eol-style set to native
File size: 16.0 KB
RevLine 
[591]1#include "xios_spl.hpp"
[300]2#include "context_client.hpp"
3#include "context_server.hpp"
4#include "event_client.hpp"
5#include "buffer_out.hpp"
6#include "buffer_client.hpp"
7#include "type.hpp"
8#include "event_client.hpp"
9#include "context.hpp"
[382]10#include "mpi.hpp"
[347]11#include "timer.hpp"
[401]12#include "cxios.hpp"
[1130]13#include "server.hpp"
[300]14
[335]15namespace xios
[300]16{
[512]17    /*!
18    \param [in] parent Pointer to context on client side
19    \param [in] intraComm_ communicator of group client
20    \param [in] interComm_ communicator of group server
[983]21    \cxtSer [in] cxtSer Pointer to context of server side. (It is only used in case of attached mode).
[512]22    */
[595]23    CContextClient::CContextClient(CContext* parent, MPI_Comm intraComm_, MPI_Comm interComm_, CContext* cxtSer)
[917]24     : mapBufferSize_(), parentServer(cxtSer), maxBufferedEvents(4)
[300]25    {
[595]26      context = parent;
27      intraComm = intraComm_;
28      interComm = interComm_;
29      MPI_Comm_rank(intraComm, &clientRank);
30      MPI_Comm_size(intraComm, &clientSize);
[509]31
[595]32      int flag;
33      MPI_Comm_test_inter(interComm, &flag);
34      if (flag) MPI_Comm_remote_size(interComm, &serverSize);
35      else  MPI_Comm_size(interComm, &serverSize);
[509]36
[1232]37      computeLeader(clientRank, clientSize, serverSize, ranksServerLeader, ranksServerNotLeader);
38
39      timeLine = 0;
40    }
41
42    void CContextClient::computeLeader(int clientRank, int clientSize, int serverSize,
43                                       std::list<int>& rankRecvLeader,
44                                       std::list<int>& rankRecvNotLeader)
45    {
46      if ((0 == clientSize) || (0 == serverSize)) return;
47
[595]48      if (clientSize < serverSize)
49      {
50        int serverByClient = serverSize / clientSize;
51        int remain = serverSize % clientSize;
52        int rankStart = serverByClient * clientRank;
[300]53
[595]54        if (clientRank < remain)
55        {
56          serverByClient++;
57          rankStart += clientRank;
58        }
59        else
60          rankStart += remain;
61
62        for (int i = 0; i < serverByClient; i++)
[1232]63          rankRecvLeader.push_back(rankStart + i);
[1021]64
[1232]65        rankRecvNotLeader.resize(0);
[1158]66      }
[595]67      else
68      {
69        int clientByServer = clientSize / serverSize;
70        int remain = clientSize % serverSize;
71
72        if (clientRank < (clientByServer + 1) * remain)
73        {
74          if (clientRank % (clientByServer + 1) == 0)
[1232]75            rankRecvLeader.push_back(clientRank / (clientByServer + 1));
[1021]76          else
[1232]77            rankRecvNotLeader.push_back(clientRank / (clientByServer + 1));
[595]78        }
79        else
80        {
81          int rank = clientRank - (clientByServer + 1) * remain;
82          if (rank % clientByServer == 0)
[1232]83            rankRecvLeader.push_back(remain + rank / clientByServer);
[1021]84          else
[1232]85            rankRecvNotLeader.push_back(remain + rank / clientByServer);
[595]86        }
87      }
[300]88    }
89
[512]90    /*!
91    In case of attached mode, the current context must be reset to context for client
92    \param [in] event Event sent to server
93    */
[300]94    void CContextClient::sendEvent(CEventClient& event)
95    {
[731]96      list<int> ranks = event.getRanks();
[1054]97
[1377]98      if (CXios::checkEventSync)
99      {
[2079]100        int typeId, classId, typeId_in, classId_in ;
101        size_t timeLine_out;
[1377]102        typeId_in=event.getTypeId() ;
103        classId_in=event.getClassId() ;
[1475]104//        MPI_Allreduce(&timeLine,&timeLine_out, 1, MPI_UINT64_T, MPI_SUM, intraComm) ; // MPI_UINT64_T standardized by MPI 3
[2079]105        MPI_Allreduce(&timeLine,&timeLine_out, 1, MPI_UNSIGNED_LONG_LONG, MPI_SUM, intraComm) ; 
[1377]106        MPI_Allreduce(&typeId_in,&typeId, 1, MPI_INT, MPI_SUM, intraComm) ;
107        MPI_Allreduce(&classId_in,&classId, 1, MPI_INT, MPI_SUM, intraComm) ;
108        if (typeId/clientSize!=event.getTypeId() || classId/clientSize!=event.getClassId() || timeLine_out/clientSize!=timeLine)
109        {
110           ERROR("void CContextClient::sendEvent(CEventClient& event)",
111               << "Event are not coherent between client.");
112        }
113      }
114
[595]115      if (!event.isEmpty())
[300]116      {
[731]117        list<int> sizes = event.getSizes();
[300]118
[1130]119        // We force the getBuffers call to be non-blocking on classical servers
[1054]120        list<CBufferOut*> buffList;
[1130]121        bool couldBuffer = getBuffers(ranks, sizes, buffList, (!CXios::isClient && (CServer::serverLevel == 0) ));
[1184]122//        bool couldBuffer = getBuffers(ranks, sizes, buffList, CXios::isServer );
[509]123
[1054]124        if (couldBuffer)
125        {
126          event.send(timeLine, sizes, buffList);
[731]127
[1054]128          checkBuffers(ranks);
129
130          if (isAttachedModeEnabled()) // couldBuffer is always true in attached mode
131          {
132            waitEvent(ranks);
133            CContext::setCurrent(context->getId());
134          }
135        }
136        else
137        {
138          tmpBufferedEvent.ranks = ranks;
139          tmpBufferedEvent.sizes = sizes;
140
141          for (list<int>::const_iterator it = sizes.begin(); it != sizes.end(); it++)
142            tmpBufferedEvent.buffers.push_back(new CBufferOut(*it));
[1377]143          info(100)<<"DEBUG : temporaly event created : timeline "<<timeLine<<endl ;
[1054]144          event.send(timeLine, tmpBufferedEvent.sizes, tmpBufferedEvent.buffers);
145        }
[300]146      }
147
[1054]148      timeLine++;
149    }
150
151    /*!
152     * Send the temporarily buffered event (if any).
153     *
[1158]154     * \return true if a temporarily buffered event could be sent, false otherwise
[1054]155     */
156    bool CContextClient::sendTemporarilyBufferedEvent()
157    {
158      bool couldSendTmpBufferedEvent = false;
159
160      if (hasTemporarilyBufferedEvent())
[511]161      {
[1054]162        list<CBufferOut*> buffList;
163        if (getBuffers(tmpBufferedEvent.ranks, tmpBufferedEvent.sizes, buffList, true)) // Non-blocking call
164        {
165          list<CBufferOut*>::iterator it, itBuffer;
166
167          for (it = tmpBufferedEvent.buffers.begin(), itBuffer = buffList.begin(); it != tmpBufferedEvent.buffers.end(); it++, itBuffer++)
168            (*itBuffer)->put((char*)(*it)->start(), (*it)->count());
169
[1377]170          info(100)<<"DEBUG : temporaly event sent "<<endl ;
[1054]171          checkBuffers(tmpBufferedEvent.ranks);
172
173          tmpBufferedEvent.clear();
174
175          couldSendTmpBufferedEvent = true;
176        }
[511]177      }
178
[1054]179      return couldSendTmpBufferedEvent;
[300]180    }
[509]181
[512]182    /*!
183    If client is also server (attached mode), after sending event, it should process right away
184    the incoming event.
185    \param [in] ranks list rank of server connected this client
186    */
[300]187    void CContextClient::waitEvent(list<int>& ranks)
188    {
[595]189      parentServer->server->setPendingEvent();
190      while (checkBuffers(ranks))
[300]191      {
[595]192        parentServer->server->listen();
193        parentServer->server->checkPendingRequest();
[300]194      }
[386]195
[595]196      while (parentServer->server->hasPendingEvent())
[386]197      {
[595]198       parentServer->server->eventLoop();
[386]199      }
[300]200    }
201
[512]202    /*!
[1054]203     * Get buffers for each connection to the servers. This function blocks until there is enough room in the buffers unless
204     * it is explicitly requested to be non-blocking.
205     *
206     * \param [in] serverList list of rank of connected server
207     * \param [in] sizeList size of message corresponding to each connection
208     * \param [out] retBuffers list of buffers that can be used to store an event
209     * \param [in] nonBlocking whether this function should be non-blocking
210     * \return whether the already allocated buffers could be used
[512]211    */
[1071]212    bool CContextClient::getBuffers(const list<int>& serverList, const list<int>& sizeList, list<CBufferOut*>& retBuffers,
213                                    bool nonBlocking /*= false*/)
[300]214    {
[1054]215      list<int>::const_iterator itServer, itSize;
[595]216      list<CClientBuffer*> bufferList;
[1054]217      map<int,CClientBuffer*>::const_iterator it;
[595]218      list<CClientBuffer*>::iterator itBuffer;
[884]219      bool areBuffersFree;
[300]220
[595]221      for (itServer = serverList.begin(); itServer != serverList.end(); itServer++)
[300]222      {
[595]223        it = buffers.find(*itServer);
224        if (it == buffers.end())
[300]225        {
[595]226          newBuffer(*itServer);
227          it = buffers.find(*itServer);
[509]228        }
[595]229        bufferList.push_back(it->second);
[300]230      }
[347]231
232      CTimer::get("Blocking time").resume();
[884]233      do
[300]234      {
[884]235        areBuffersFree = true;
[595]236        for (itBuffer = bufferList.begin(), itSize = sizeList.begin(); itBuffer != bufferList.end(); itBuffer++, itSize++)
[884]237          areBuffersFree &= (*itBuffer)->isBufferFree(*itSize);
238
239        if (!areBuffersFree)
[300]240        {
[884]241          checkBuffers();
[1130]242          if (CServer::serverLevel == 0)
243            context->server->listen();
[1071]244
[1130]245          else if (CServer::serverLevel == 1)
246          {
247            context->server->listen();
248            for (int i = 0; i < context->serverPrimServer.size(); ++i)
249              context->serverPrimServer[i]->listen();
[1378]250            CServer::contextEventLoop(false) ; // avoid dead-lock at finalize...
[1130]251          }
252
253          else if (CServer::serverLevel == 2)
254            context->server->listen();
255
[300]256        }
[1054]257      } while (!areBuffersFree && !nonBlocking);
[1130]258
[347]259      CTimer::get("Blocking time").suspend();
260
[1054]261      if (areBuffersFree)
[300]262      {
[1054]263        for (itBuffer = bufferList.begin(), itSize = sizeList.begin(); itBuffer != bufferList.end(); itBuffer++, itSize++)
264          retBuffers.push_back((*itBuffer)->getBuffer(*itSize));
[300]265      }
[1054]266
267      return areBuffersFree;
[300]268   }
[509]269
[512]270   /*!
271   Make a new buffer for a certain connection to server with specific rank
272   \param [in] rank rank of connected server
273   */
[300]274   void CContextClient::newBuffer(int rank)
275   {
[1201]276      if (!mapBufferSize_.count(rank))
277      {
278        error(0) << "WARNING: Unexpected request for buffer to communicate with server " << rank << std::endl;
279        mapBufferSize_[rank] = CXios::minBufferSize;
280        maxEventSizes[rank] = CXios::minBufferSize;
281      }
282      CClientBuffer* buffer = buffers[rank] = new CClientBuffer(interComm, rank, mapBufferSize_[rank], maxEventSizes[rank], maxBufferedEvents);
283      // Notify the server
284      CBufferOut* bufOut = buffer->getBuffer(sizeof(StdSize));
285      bufOut->put(mapBufferSize_[rank]); // Stupid C++
286      buffer->checkBuffer();
[509]287   }
[300]288
[512]289   /*!
290   Verify state of buffers. Buffer is under pending state if there is no message on it
291   \return state of buffers, pending(true), ready(false)
292   */
[300]293   bool CContextClient::checkBuffers(void)
294   {
[595]295      map<int,CClientBuffer*>::iterator itBuff;
296      bool pending = false;
[1130]297      for (itBuff = buffers.begin(); itBuff != buffers.end(); itBuff++)
298        pending |= itBuff->second->checkBuffer();
[595]299      return pending;
[509]300   }
[300]301
[512]302   //! Release all buffers
[1071]303   void CContextClient::releaseBuffers()
[300]304   {
[595]305      map<int,CClientBuffer*>::iterator itBuff;
[1077]306      for (itBuff = buffers.begin(); itBuff != buffers.end(); itBuff++)
[1139]307      {
[1077]308          delete itBuff->second;
[1139]309      }
[1077]310      buffers.clear();
[509]311   }
[300]312
[512]313   /*!
314   Verify state of buffers corresponding to a connection
315   \param [in] ranks list rank of server to which client connects to
316   \return state of buffers, pending(true), ready(false)
317   */
[300]318   bool CContextClient::checkBuffers(list<int>& ranks)
319   {
[595]320      list<int>::iterator it;
321      bool pending = false;
322      for (it = ranks.begin(); it != ranks.end(); it++) pending |= buffers[*it]->checkBuffer();
323      return pending;
[509]324   }
[300]325
[512]326   /*!
[917]327    * Set the buffer size for each connection. Warning: This function is collective.
328    *
329    * \param [in] mapSize maps the rank of the connected servers to the size of the correspoinding buffer
330    * \param [in] maxEventSize maps the rank of the connected servers to the size of the biggest event
[512]331   */
[917]332   void CContextClient::setBufferSize(const std::map<int,StdSize>& mapSize, const std::map<int,StdSize>& maxEventSize)
[509]333   {
334     mapBufferSize_ = mapSize;
[1201]335     maxEventSizes = maxEventSize;
[917]336
337     // Compute the maximum number of events that can be safely buffered.
338     double minBufferSizeEventSizeRatio = std::numeric_limits<double>::max();
339     for (std::map<int,StdSize>::const_iterator it = mapSize.begin(), ite = mapSize.end(); it != ite; ++it)
340     {
[1474]341       double ratio = double(it->second) / maxEventSizes[it->first];
[917]342       if (ratio < minBufferSizeEventSizeRatio) minBufferSizeEventSizeRatio = ratio;
343     }
344     MPI_Allreduce(MPI_IN_PLACE, &minBufferSizeEventSizeRatio, 1, MPI_DOUBLE, MPI_MIN, intraComm);
345
346     if (minBufferSizeEventSizeRatio < 1.0)
[1201]347     {
[917]348       ERROR("void CContextClient::setBufferSize(const std::map<int,StdSize>& mapSize, const std::map<int,StdSize>& maxEventSize)",
349             << "The buffer sizes and the maximum events sizes are incoherent.");
[1201]350     }
351     else if (minBufferSizeEventSizeRatio == std::numeric_limits<double>::max())
352       minBufferSizeEventSizeRatio = 1.0; // In this case, maxBufferedEvents will never be used but we want to avoid any floating point exception
[917]353
354     maxBufferedEvents = size_t(2 * minBufferSizeEventSizeRatio) // there is room for two local buffers on the server
355                          + size_t(minBufferSizeEventSizeRatio)  // one local buffer can always be fully used
356                          + 1;                                   // the other local buffer might contain only one event
[509]357   }
358
[1158]359  /*!
360  Get leading server in the group of connected server
361  \return ranks of leading servers
362  */
363  const std::list<int>& CContextClient::getRanksServerNotLeader(void) const
364  {
365    return ranksServerNotLeader;
366  }
[1021]367
[1158]368  /*!
369  Check if client connects to leading server
370  \return connected(true), not connected (false)
371  */
372  bool CContextClient::isServerNotLeader(void) const
373  {
374    return !ranksServerNotLeader.empty();
375  }
[1021]376
[595]377  /*!
378  Get leading server in the group of connected server
379  \return ranks of leading servers
380  */
381  const std::list<int>& CContextClient::getRanksServerLeader(void) const
382  {
383    return ranksServerLeader;
384  }
[509]385
[595]386  /*!
387  Check if client connects to leading server
388  \return connected(true), not connected (false)
389  */
390  bool CContextClient::isServerLeader(void) const
391  {
392    return !ranksServerLeader.empty();
393  }
[300]394
[704]395  /*!
396   * Check if the attached mode is used.
397   *
398   * \return true if and only if attached mode is used
399   */
400  bool CContextClient::isAttachedModeEnabled() const
401  {
402    return (parentServer != 0);
403  }
[697]404
[512]405   /*!
[1130]406   * Finalize context client and do some reports. Function is non-blocking.
[512]407   */
[1130]408  void CContextClient::finalize(void)
[1054]409  {
410    map<int,CClientBuffer*>::iterator itBuff;
411    bool stop = false;
[731]412
[1054]413    CTimer::get("Blocking time").resume();
414    while (hasTemporarilyBufferedEvent())
415    {
416      checkBuffers();
417      sendTemporarilyBufferedEvent();
418    }
419    CTimer::get("Blocking time").suspend();
[509]420
[1054]421    CEventClient event(CContext::GetType(), CContext::EVENT_ID_CONTEXT_FINALIZE);
422    if (isServerLeader())
423    {
424      CMessage msg;
425      const std::list<int>& ranks = getRanksServerLeader();
426      for (std::list<int>::const_iterator itRank = ranks.begin(), itRankEnd = ranks.end(); itRank != itRankEnd; ++itRank)
[1377]427      {
428        info(100)<<"DEBUG : Sent context Finalize event to rank "<<*itRank<<endl ;
[1054]429        event.push(*itRank, 1, msg);
[1377]430      }
[1054]431      sendEvent(event);
432    }
433    else sendEvent(event);
[509]434
[1054]435    CTimer::get("Blocking time").resume();
[1130]436//    while (!stop)
[1054]437    {
438      checkBuffers();
439      if (hasTemporarilyBufferedEvent())
440        sendTemporarilyBufferedEvent();
[511]441
[1054]442      stop = true;
[1130]443//      for (itBuff = buffers.begin(); itBuff != buffers.end(); itBuff++) stop &= !itBuff->second->hasPendingRequest();
[1054]444    }
445    CTimer::get("Blocking time").suspend();
446
447    std::map<int,StdSize>::const_iterator itbMap = mapBufferSize_.begin(),
448                                          iteMap = mapBufferSize_.end(), itMap;
[1071]449
[1054]450    StdSize totalBuf = 0;
451    for (itMap = itbMap; itMap != iteMap; ++itMap)
452    {
453      report(10) << " Memory report : Context <" << context->getId() << "> : client side : memory used for buffer of each connection to server" << endl
454                 << "  +) To server with rank " << itMap->first << " : " << itMap->second << " bytes " << endl;
455      totalBuf += itMap->second;
456    }
457    report(0) << " Memory report : Context <" << context->getId() << "> : client side : total memory used for buffer " << totalBuf << " bytes" << endl;
458
[1130]459    //releaseBuffers(); // moved to CContext::finalize()
[1054]460  }
[1130]461
[1139]462
463  /*!
464  */
[1130]465  bool CContextClient::havePendingRequests(void)
466  {
467    bool pending = false;
468    map<int,CClientBuffer*>::iterator itBuff;
469    for (itBuff = buffers.begin(); itBuff != buffers.end(); itBuff++)
470      pending |= itBuff->second->hasPendingRequest();
471    return pending;
472  }
473
474
[509]475}
Note: See TracBrowser for help on using the repository browser.