1 | #include "globalScopeData.hpp" |
---|
2 | #include "xios_spl.hpp" |
---|
3 | #include "cxios.hpp" |
---|
4 | #include "client.hpp" |
---|
5 | #include <boost/functional/hash.hpp> |
---|
6 | #include "type.hpp" |
---|
7 | #include "context.hpp" |
---|
8 | #include "context_client.hpp" |
---|
9 | #include "oasis_cinterface.hpp" |
---|
10 | #include "mpi.hpp" |
---|
11 | #include "timer.hpp" |
---|
12 | #include "buffer_client.hpp" |
---|
13 | #include "string_tools.hpp" |
---|
14 | |
---|
15 | namespace xios |
---|
16 | { |
---|
17 | |
---|
18 | MPI_Comm CClient::intraComm ; |
---|
19 | MPI_Comm CClient::interComm ; |
---|
20 | std::list<MPI_Comm> CClient::contextInterComms; |
---|
21 | int CClient::serverLeader ; |
---|
22 | bool CClient::is_MPI_Initialized ; |
---|
23 | int CClient::rank_ = INVALID_RANK; |
---|
24 | StdOFStream CClient::m_infoStream; |
---|
25 | StdOFStream CClient::m_errorStream; |
---|
26 | MPI_Comm& CClient::getInterComm(void) { return (interComm); } |
---|
27 | |
---|
28 | ///--------------------------------------------------------------- |
---|
29 | /*! |
---|
30 | * \fn void CClient::initialize(const string& codeId, MPI_Comm& localComm, MPI_Comm& returnComm) |
---|
31 | * Function creates intraComm (CClient::intraComm) for client group with id=codeId and interComm (CClient::interComm) between client and server groups. |
---|
32 | * \param [in] codeId identity of context. |
---|
33 | * \param [in/out] localComm local communicator. |
---|
34 | * \param [in/out] returnComm (intra)communicator of client group. |
---|
35 | */ |
---|
36 | |
---|
37 | void CClient::initialize(const string& codeId, MPI_Comm& localComm, MPI_Comm& returnComm) |
---|
38 | { |
---|
39 | int initialized ; |
---|
40 | MPI_Initialized(&initialized) ; |
---|
41 | if (initialized) is_MPI_Initialized=true ; |
---|
42 | else is_MPI_Initialized=false ; |
---|
43 | int rank ; |
---|
44 | bool ensembleManagement ; |
---|
45 | int ensembleSize ; |
---|
46 | |
---|
47 | // don't use OASIS |
---|
48 | if (!CXios::usingOasis) |
---|
49 | { |
---|
50 | // localComm isn't given |
---|
51 | if (localComm == MPI_COMM_NULL) |
---|
52 | { |
---|
53 | if (!is_MPI_Initialized) |
---|
54 | { |
---|
55 | MPI_Init(NULL, NULL); |
---|
56 | } |
---|
57 | CTimer::get("XIOS").resume() ; |
---|
58 | CTimer::get("XIOS init/finalize").resume() ; |
---|
59 | boost::hash<string> hashString ; |
---|
60 | |
---|
61 | unsigned long hashClient=hashString(codeId) ; |
---|
62 | unsigned long hashServer=hashString(CXios::xiosCodeId) ; |
---|
63 | unsigned long* hashAll ; |
---|
64 | int size ; |
---|
65 | int myColor ; |
---|
66 | int i,c ; |
---|
67 | MPI_Comm newComm ; |
---|
68 | |
---|
69 | MPI_Comm_size(CXios::globalComm,&size) ; |
---|
70 | MPI_Comm_rank(CXios::globalComm,&rank_); |
---|
71 | |
---|
72 | hashAll=new unsigned long[size] ; |
---|
73 | |
---|
74 | MPI_Allgather(&hashClient,1,MPI_LONG,hashAll,1,MPI_LONG,CXios::globalComm) ; |
---|
75 | |
---|
76 | map<unsigned long, int> colors ; |
---|
77 | map<unsigned long, int> leaders ; |
---|
78 | |
---|
79 | for(i=0,c=0;i<size;i++) |
---|
80 | { |
---|
81 | if (colors.find(hashAll[i])==colors.end()) |
---|
82 | { |
---|
83 | colors[hashAll[i]] =c ; |
---|
84 | leaders[hashAll[i]]=i ; |
---|
85 | c++ ; |
---|
86 | } |
---|
87 | } |
---|
88 | |
---|
89 | // Verify whether we are on server mode or not |
---|
90 | CXios::setNotUsingServer(); |
---|
91 | for (i=0; i < size; ++i) |
---|
92 | { |
---|
93 | if (hashServer == hashAll[i]) |
---|
94 | { |
---|
95 | CXios::setUsingServer(); |
---|
96 | break; |
---|
97 | } |
---|
98 | } |
---|
99 | |
---|
100 | myColor=colors[hashClient]; |
---|
101 | MPI_Comm_split(CXios::globalComm,myColor,rank_,&intraComm) ; |
---|
102 | |
---|
103 | if (CXios::usingServer) |
---|
104 | { |
---|
105 | int clientLeader=leaders[hashClient] ; |
---|
106 | serverLeader=leaders[hashServer] ; |
---|
107 | int intraCommSize, intraCommRank ; |
---|
108 | MPI_Comm_size(intraComm,&intraCommSize) ; |
---|
109 | MPI_Comm_rank(intraComm,&intraCommRank) ; |
---|
110 | info(50)<<"intercommCreate::client "<<rank_<<" intraCommSize : "<<intraCommSize |
---|
111 | <<" intraCommRank :"<<intraCommRank<<" clientLeader "<< serverLeader<<endl ; |
---|
112 | MPI_Intercomm_create(intraComm, 0, CXios::globalComm, serverLeader, 0, &interComm) ; |
---|
113 | //rank_ = intraCommRank; |
---|
114 | } |
---|
115 | else |
---|
116 | { |
---|
117 | MPI_Comm_dup(intraComm,&interComm) ; |
---|
118 | } |
---|
119 | delete [] hashAll ; |
---|
120 | } |
---|
121 | // localComm argument is given |
---|
122 | else |
---|
123 | { |
---|
124 | if (CXios::usingServer) |
---|
125 | { |
---|
126 | //ERROR("void CClient::initialize(const string& codeId,MPI_Comm& localComm,MPI_Comm& returnComm)", << " giving a local communictor is not compatible with using server mode") ; |
---|
127 | } |
---|
128 | else |
---|
129 | { |
---|
130 | MPI_Comm_dup(localComm,&intraComm) ; |
---|
131 | MPI_Comm_dup(intraComm,&interComm) ; |
---|
132 | } |
---|
133 | } |
---|
134 | } |
---|
135 | // using OASIS |
---|
136 | else |
---|
137 | { |
---|
138 | |
---|
139 | // manage ensemble with oasis |
---|
140 | // code id is splitted follow the base code Id and the num number using the ":" separator |
---|
141 | // codeId == baseCodeId:member |
---|
142 | // member must integer convertible |
---|
143 | vector<string> splittedCodeId=splitRegex(codeId,"\\s*:\\s*") ; |
---|
144 | if (splittedCodeId.size()>2) ERROR("void CClient::initialize(const string& codeId, MPI_Comm& localComm, MPI_Comm& returnComm)", |
---|
145 | <<"CodeId is badly formed "<<codeId) ; |
---|
146 | string baseCodeId=splittedCodeId[0] ; |
---|
147 | int member ; |
---|
148 | if (splittedCodeId.size()==2) |
---|
149 | { |
---|
150 | try |
---|
151 | { |
---|
152 | member = stoi(splittedCodeId[1]) ; |
---|
153 | } |
---|
154 | catch(...) |
---|
155 | { |
---|
156 | ERROR("void CClient::initialize(const string& codeId, MPI_Comm& localComm, MPI_Comm& returnComm)", |
---|
157 | <<"CodeId is badly formed "<<codeId) ; |
---|
158 | } |
---|
159 | ensembleManagement=true ; |
---|
160 | } |
---|
161 | else ensembleManagement=false ; |
---|
162 | |
---|
163 | string codesId=CXios::getin<string>("oasis_codes_id") ; |
---|
164 | vector<string> oasisCodesId=splitRegex(codesId,"\\s*,\\s*") ; |
---|
165 | for(auto& oasisCodeId : oasisCodesId) |
---|
166 | { |
---|
167 | info(10)<<"oasisCodeId "<<oasisCodeId<<endl ; |
---|
168 | vector<string> splittedOasisCodeId = splitRegex(oasisCodeId,"\\s*:\\s*") ; |
---|
169 | string baseOasisCodeId=splittedOasisCodeId[0] ; |
---|
170 | info(10)<<"baseOasisCodeId "<<baseOasisCodeId<<" baseCodeId "<<baseCodeId<<endl ; |
---|
171 | if (baseOasisCodeId==baseCodeId) |
---|
172 | { |
---|
173 | if (splittedOasisCodeId.size()==2) |
---|
174 | { |
---|
175 | try |
---|
176 | { |
---|
177 | ensembleSize = stoi(splittedOasisCodeId[1]) ; |
---|
178 | info(10)<<"ensembleSize "<<ensembleSize<<endl ; |
---|
179 | } |
---|
180 | catch(...) |
---|
181 | { |
---|
182 | ERROR("void CClient::initialize(const string& codeId, MPI_Comm& localComm, MPI_Comm& returnComm)", |
---|
183 | <<"CodeId is badly formed "<<codeId) ; |
---|
184 | } |
---|
185 | if (ensembleManagement==false) ERROR("void CClient::initialize(const string& codeId, MPI_Comm& localComm, MPI_Comm& returnComm)", |
---|
186 | <<"inconsistancy for ensemble management between local codeId and OASIS codes Id : " |
---|
187 | <<codeId<<" "<< codesId); |
---|
188 | } |
---|
189 | else if (ensembleManagement==true) ERROR("void CClient::initialize(const string& codeId, MPI_Comm& localComm, MPI_Comm& returnComm)", |
---|
190 | <<"inconsistancy for ensemble management between local codeId and OASIS codes Id " |
---|
191 | <<codeId<<" "<< codesId); |
---|
192 | } |
---|
193 | } |
---|
194 | |
---|
195 | |
---|
196 | // localComm isn't given |
---|
197 | if (localComm == MPI_COMM_NULL) |
---|
198 | { |
---|
199 | if (!is_MPI_Initialized) oasis_init(codeId) ; |
---|
200 | if(ensembleManagement) |
---|
201 | { |
---|
202 | vector<string> intraCommStr(ensembleSize) ; |
---|
203 | vector<int> rootRanks ; |
---|
204 | for(int i=0;i<ensembleSize;i++) intraCommStr[i]=baseCodeId+":"+std::to_string(i) ; |
---|
205 | oasis_get_multi_intracomm(localComm, intraCommStr, rootRanks) ; |
---|
206 | } |
---|
207 | else oasis_get_localcomm(localComm) ; |
---|
208 | } |
---|
209 | MPI_Comm_dup(localComm,&intraComm) ; |
---|
210 | |
---|
211 | CTimer::get("XIOS").resume() ; |
---|
212 | CTimer::get("XIOS init/finalize").resume() ; |
---|
213 | |
---|
214 | if (CXios::usingServer) |
---|
215 | { |
---|
216 | MPI_Status status ; |
---|
217 | MPI_Comm_rank(intraComm,&rank_) ; |
---|
218 | |
---|
219 | if(ensembleManagement) |
---|
220 | { |
---|
221 | vector<string> intraCommStr(ensembleSize+1) ; |
---|
222 | vector<int> rootRanks ; |
---|
223 | for(int i=0;i<ensembleSize;i++) intraCommStr[i]=baseCodeId+":"+std::to_string(i) ; |
---|
224 | intraCommStr[ensembleSize] = CXios::xiosCodeId ; |
---|
225 | |
---|
226 | oasis_get_multi_intracomm(localComm, intraCommStr, rootRanks) ; |
---|
227 | |
---|
228 | int myColor= 0 ; // 0:client 1:server |
---|
229 | int myrank ; |
---|
230 | MPI_Comm splittedComm ; |
---|
231 | MPI_Comm_rank(localComm,&myrank) ; |
---|
232 | MPI_Comm_split(localComm,myColor,myrank, &splittedComm) ; |
---|
233 | MPI_Intercomm_create(splittedComm,0,localComm,rootRanks[ensembleSize],10,&interComm) ; |
---|
234 | |
---|
235 | } |
---|
236 | else oasis_get_intercomm(interComm,CXios::xiosCodeId) ; |
---|
237 | if (rank_==0) MPI_Recv(&serverLeader,1, MPI_INT, 0, 0, interComm, &status) ; |
---|
238 | MPI_Bcast(&serverLeader,1,MPI_INT,0,intraComm) ; |
---|
239 | } |
---|
240 | else MPI_Comm_dup(intraComm,&interComm) ; |
---|
241 | } |
---|
242 | |
---|
243 | MPI_Comm_dup(intraComm,&returnComm) ; |
---|
244 | } |
---|
245 | |
---|
246 | ///--------------------------------------------------------------- |
---|
247 | /*! |
---|
248 | * \fn void CClient::registerContext(const string& id, MPI_Comm contextComm) |
---|
249 | * \brief Sends a request to create a context to server. Creates client/server contexts. |
---|
250 | * \param [in] id id of context. |
---|
251 | * \param [in] contextComm. |
---|
252 | * Function is only called by client. |
---|
253 | */ |
---|
254 | void CClient::registerContext(const string& id, MPI_Comm contextComm) |
---|
255 | { |
---|
256 | CContext::setCurrent(id) ; |
---|
257 | CContext* context=CContext::create(id); |
---|
258 | StdString idServer(id); |
---|
259 | idServer += "_server"; |
---|
260 | |
---|
261 | if (CXios::isServer && !context->hasServer) |
---|
262 | // Attached mode |
---|
263 | { |
---|
264 | MPI_Comm contextInterComm ; |
---|
265 | MPI_Comm_dup(contextComm,&contextInterComm) ; |
---|
266 | CContext* contextServer = CContext::create(idServer); |
---|
267 | |
---|
268 | // Firstly, initialize context on client side |
---|
269 | context->initClient(contextComm,contextInterComm, contextServer); |
---|
270 | |
---|
271 | // Secondly, initialize context on server side |
---|
272 | contextServer->initServer(contextComm,contextInterComm, context); |
---|
273 | |
---|
274 | // Finally, we should return current context to context client |
---|
275 | CContext::setCurrent(id); |
---|
276 | |
---|
277 | contextInterComms.push_back(contextInterComm); |
---|
278 | } |
---|
279 | else |
---|
280 | { |
---|
281 | int size,rank,globalRank ; |
---|
282 | size_t message_size ; |
---|
283 | int leaderRank ; |
---|
284 | MPI_Comm contextInterComm ; |
---|
285 | |
---|
286 | MPI_Comm_size(contextComm,&size) ; |
---|
287 | MPI_Comm_rank(contextComm,&rank) ; |
---|
288 | MPI_Comm_rank(CXios::globalComm,&globalRank) ; |
---|
289 | if (rank!=0) globalRank=0 ; |
---|
290 | |
---|
291 | CMessage msg ; |
---|
292 | msg<<idServer<<size<<globalRank ; |
---|
293 | // msg<<id<<size<<globalRank ; |
---|
294 | |
---|
295 | int messageSize=msg.size() ; |
---|
296 | char * buff = new char[messageSize] ; |
---|
297 | CBufferOut buffer((void*)buff,messageSize) ; |
---|
298 | buffer<<msg ; |
---|
299 | |
---|
300 | MPI_Send((void*)buff,buffer.count(),MPI_CHAR,serverLeader,1,CXios::globalComm) ; |
---|
301 | |
---|
302 | MPI_Intercomm_create(contextComm,0,CXios::globalComm,serverLeader,10+globalRank,&contextInterComm) ; |
---|
303 | info(10)<<"Register new Context : "<<id<<endl ; |
---|
304 | MPI_Comm inter ; |
---|
305 | MPI_Intercomm_merge(contextInterComm,0,&inter) ; |
---|
306 | MPI_Barrier(inter) ; |
---|
307 | MPI_Comm_free(&inter); |
---|
308 | |
---|
309 | context->initClient(contextComm,contextInterComm) ; |
---|
310 | |
---|
311 | contextInterComms.push_back(contextInterComm); |
---|
312 | delete [] buff ; |
---|
313 | |
---|
314 | } |
---|
315 | } |
---|
316 | |
---|
317 | /*! |
---|
318 | * \fn void CClient::callOasisEnddef(void) |
---|
319 | * \brief Send the order to the servers to call "oasis_enddef". It must be done by each compound of models before calling oasis_enddef on client side |
---|
320 | * Function is only called by client. |
---|
321 | */ |
---|
322 | void CClient::callOasisEnddef(void) |
---|
323 | { |
---|
324 | bool oasisEnddef=CXios::getin<bool>("call_oasis_enddef",true) ; |
---|
325 | if (!oasisEnddef) ERROR("void CClient::callOasisEnddef(void)", <<"Function xios_oasis_enddef called but variable <call_oasis_enddef> is set to false."<<endl |
---|
326 | <<"Variable <call_oasis_enddef> must be set to true"<<endl) ; |
---|
327 | if (CXios::isServer) |
---|
328 | // Attached mode |
---|
329 | { |
---|
330 | // nothing to do |
---|
331 | } |
---|
332 | else |
---|
333 | { |
---|
334 | int rank ; |
---|
335 | int msg=0 ; |
---|
336 | |
---|
337 | MPI_Comm_rank(intraComm,&rank) ; |
---|
338 | if (rank==0) |
---|
339 | { |
---|
340 | MPI_Send(&msg,1,MPI_INT,0,5,interComm) ; // tags oasis_endded = 5 |
---|
341 | } |
---|
342 | |
---|
343 | } |
---|
344 | } |
---|
345 | |
---|
346 | |
---|
347 | void CClient::finalize(void) |
---|
348 | { |
---|
349 | int rank ; |
---|
350 | int msg=0 ; |
---|
351 | |
---|
352 | MPI_Comm_rank(intraComm,&rank) ; |
---|
353 | |
---|
354 | if (!CXios::isServer) |
---|
355 | { |
---|
356 | MPI_Comm_rank(intraComm,&rank) ; |
---|
357 | if (rank==0) |
---|
358 | { |
---|
359 | MPI_Send(&msg,1,MPI_INT,0,0,interComm) ; |
---|
360 | } |
---|
361 | } |
---|
362 | |
---|
363 | |
---|
364 | MPI_Comm_free(&interComm); |
---|
365 | for (std::list<MPI_Comm>::iterator it = contextInterComms.begin(); it != contextInterComms.end(); it++) |
---|
366 | MPI_Comm_free(&(*it)); |
---|
367 | MPI_Comm_free(&intraComm); |
---|
368 | |
---|
369 | CTimer::get("XIOS init/finalize").suspend() ; |
---|
370 | CTimer::get("XIOS").suspend() ; |
---|
371 | |
---|
372 | if (!is_MPI_Initialized) |
---|
373 | { |
---|
374 | if (CXios::usingOasis) oasis_finalize(); |
---|
375 | else MPI_Finalize() ; |
---|
376 | } |
---|
377 | |
---|
378 | info(20) << "Client side context is finalized"<<endl ; |
---|
379 | report(0) <<" Performance report : Whole time from XIOS init and finalize: "<< CTimer::get("XIOS init/finalize").getCumulatedTime()<<" s"<<endl ; |
---|
380 | report(0) <<" Performance report : total time spent for XIOS : "<< CTimer::get("XIOS").getCumulatedTime()<<" s"<<endl ; |
---|
381 | report(0)<< " Performance report : time spent for waiting free buffer : "<< CTimer::get("Blocking time").getCumulatedTime()<<" s"<<endl ; |
---|
382 | report(0)<< " Performance report : Ratio : "<< CTimer::get("Blocking time").getCumulatedTime()/CTimer::get("XIOS init/finalize").getCumulatedTime()*100.<<" %"<<endl ; |
---|
383 | report(0)<< " Performance report : This ratio must be close to zero. Otherwise it may be usefull to increase buffer size or numbers of server"<<endl ; |
---|
384 | // report(0)<< " Memory report : Current buffer_size : "<<CXios::bufferSize<<endl ; |
---|
385 | report(0)<< " Memory report : Minimum buffer size required : " << CClientBuffer::maxRequestSize << " bytes" << endl ; |
---|
386 | report(0)<< " Memory report : increasing it by a factor will increase performance, depending of the volume of data wrote in file at each time step of the file"<<endl ; |
---|
387 | report(100)<<CTimer::getAllCumulatedTime()<<endl ; |
---|
388 | } |
---|
389 | |
---|
390 | /*! |
---|
391 | * Return global rank without oasis and current rank in model intraComm in case of oasis |
---|
392 | */ |
---|
393 | int CClient::getRank() |
---|
394 | { |
---|
395 | return rank_; |
---|
396 | } |
---|
397 | |
---|
398 | /*! |
---|
399 | * Open a file specified by a suffix and an extension and use it for the given file buffer. |
---|
400 | * The file name will be suffix+rank+extension. |
---|
401 | * |
---|
402 | * \param fileName[in] protype file name |
---|
403 | * \param ext [in] extension of the file |
---|
404 | * \param fb [in/out] the file buffer |
---|
405 | */ |
---|
406 | void CClient::openStream(const StdString& fileName, const StdString& ext, std::filebuf* fb) |
---|
407 | { |
---|
408 | StdStringStream fileNameClient; |
---|
409 | int numDigit = 0; |
---|
410 | int size = 0; |
---|
411 | int rank; |
---|
412 | MPI_Comm_size(CXios::globalComm, &size); |
---|
413 | while (size) |
---|
414 | { |
---|
415 | size /= 10; |
---|
416 | ++numDigit; |
---|
417 | } |
---|
418 | |
---|
419 | if (CXios::usingOasis) |
---|
420 | { |
---|
421 | MPI_Comm_rank(CXios::globalComm,&rank); |
---|
422 | fileNameClient << fileName << "_" << std::setfill('0') << std::setw(numDigit) << rank << ext; |
---|
423 | } |
---|
424 | else |
---|
425 | fileNameClient << fileName << "_" << std::setfill('0') << std::setw(numDigit) << getRank() << ext; |
---|
426 | |
---|
427 | |
---|
428 | fb->open(fileNameClient.str().c_str(), std::ios::out); |
---|
429 | if (!fb->is_open()) |
---|
430 | ERROR("void CClient::openStream(const StdString& fileName, const StdString& ext, std::filebuf* fb)", |
---|
431 | << std::endl << "Can not open <" << fileNameClient.str() << "> file to write the client log(s)."); |
---|
432 | } |
---|
433 | |
---|
434 | /*! |
---|
435 | * \brief Open a file stream to write the info logs |
---|
436 | * Open a file stream with a specific file name suffix+rank |
---|
437 | * to write the info logs. |
---|
438 | * \param fileName [in] protype file name |
---|
439 | */ |
---|
440 | void CClient::openInfoStream(const StdString& fileName) |
---|
441 | { |
---|
442 | std::filebuf* fb = m_infoStream.rdbuf(); |
---|
443 | openStream(fileName, ".out", fb); |
---|
444 | |
---|
445 | info.write2File(fb); |
---|
446 | report.write2File(fb); |
---|
447 | } |
---|
448 | |
---|
449 | //! Write the info logs to standard output |
---|
450 | void CClient::openInfoStream() |
---|
451 | { |
---|
452 | info.write2StdOut(); |
---|
453 | report.write2StdOut(); |
---|
454 | } |
---|
455 | |
---|
456 | //! Close the info logs file if it opens |
---|
457 | void CClient::closeInfoStream() |
---|
458 | { |
---|
459 | if (m_infoStream.is_open()) m_infoStream.close(); |
---|
460 | } |
---|
461 | |
---|
462 | /*! |
---|
463 | * \brief Open a file stream to write the error log |
---|
464 | * Open a file stream with a specific file name suffix+rank |
---|
465 | * to write the error log. |
---|
466 | * \param fileName [in] protype file name |
---|
467 | */ |
---|
468 | void CClient::openErrorStream(const StdString& fileName) |
---|
469 | { |
---|
470 | std::filebuf* fb = m_errorStream.rdbuf(); |
---|
471 | openStream(fileName, ".err", fb); |
---|
472 | |
---|
473 | error.write2File(fb); |
---|
474 | } |
---|
475 | |
---|
476 | //! Write the error log to standard error output |
---|
477 | void CClient::openErrorStream() |
---|
478 | { |
---|
479 | error.write2StdErr(); |
---|
480 | } |
---|
481 | |
---|
482 | //! Close the error log file if it opens |
---|
483 | void CClient::closeErrorStream() |
---|
484 | { |
---|
485 | if (m_errorStream.is_open()) m_errorStream.close(); |
---|
486 | } |
---|
487 | } |
---|