]>
Commit | Line | Data |
---|---|---|
73abe331 | 1 | /************************************************************************** |
2 | * Copyright(c) 1998-1999, ALICE Experiment at CERN, All rights reserved. * | |
3 | * * | |
4 | * Author: The ALICE Off-line Project. * | |
5 | * Contributors are mentioned in the code where appropriate. * | |
6 | * * | |
7 | * Permission to use, copy, modify and distribute this software and its * | |
8 | * documentation strictly for non-commercial purposes is hereby granted * | |
9 | * without fee, provided that the above copyright notice appears in all * | |
10 | * copies and that both the copyright notice and this permission notice * | |
11 | * appear in the supporting documentation. The authors make no claims * | |
12 | * about the suitability of this software for any purpose. It is * | |
13 | * provided "as is" without express or implied warranty. * | |
14 | **************************************************************************/ | |
15 | ||
ff3781ad | 16 | /* $Id$ */ |
73abe331 | 17 | |
18 | // | |
19 | // This class is the main manager for AliShuttle. | |
20 | // It organizes the data retrieval from DCS and call the | |
b948db8d | 21 | // interface methods of AliPreprocessor. |
73abe331 | 22 | // For every detector in AliShuttleConfgi (see AliShuttleConfig), |
23 | // data for its set of aliases is retrieved. If there is registered | |
b948db8d | 24 | // AliPreprocessor for this detector then it will be used |
25 | // accroding to the schema (see AliPreprocessor). | |
26 | // If there isn't registered AliPreprocessor than the retrieved | |
73abe331 | 27 | // data is stored automatically to the undelying AliCDBStorage. |
28 | // For detSpec is used the alias name. | |
29 | // | |
30 | ||
31 | #include "AliShuttle.h" | |
32 | ||
33 | #include "AliCDBManager.h" | |
34 | #include "AliCDBStorage.h" | |
35 | #include "AliCDBId.h" | |
84090f85 | 36 | #include "AliCDBRunRange.h" |
37 | #include "AliCDBPath.h" | |
5164a766 | 38 | #include "AliCDBEntry.h" |
73abe331 | 39 | #include "AliShuttleConfig.h" |
eba76848 | 40 | #include "DCSClient/AliDCSClient.h" |
73abe331 | 41 | #include "AliLog.h" |
b948db8d | 42 | #include "AliPreprocessor.h" |
5164a766 | 43 | #include "AliShuttleStatus.h" |
2bb7b766 | 44 | #include "AliShuttleLogbookEntry.h" |
73abe331 | 45 | |
57f50b3c | 46 | #include <TSystem.h> |
58bc3020 | 47 | #include <TObject.h> |
b948db8d | 48 | #include <TString.h> |
57f50b3c | 49 | #include <TTimeStamp.h> |
73abe331 | 50 | #include <TObjString.h> |
57f50b3c | 51 | #include <TSQLServer.h> |
52 | #include <TSQLResult.h> | |
53 | #include <TSQLRow.h> | |
cb343cfd | 54 | #include <TMutex.h> |
9827400b | 55 | #include <TSystemDirectory.h> |
56 | #include <TSystemFile.h> | |
a986b218 | 57 | #include <TFile.h> |
9827400b | 58 | #include <TGrid.h> |
59 | #include <TGridResult.h> | |
73abe331 | 60 | |
e7f62f16 | 61 | #include <TMonaLisaWriter.h> |
62 | ||
5164a766 | 63 | #include <fstream> |
64 | ||
cb343cfd | 65 | #include <sys/types.h> |
66 | #include <sys/wait.h> | |
67 | ||
1bd183ba | 68 | #include <signal.h> |
69 | ||
73abe331 | 70 | ClassImp(AliShuttle) |
71 | ||
b948db8d | 72 | //______________________________________________________________________________________________ |
73 | AliShuttle::AliShuttle(const AliShuttleConfig* config, | |
74 | UInt_t timeout, Int_t retries): | |
4f0ab988 | 75 | fConfig(config), |
76 | fTimeout(timeout), fRetries(retries), | |
77 | fPreprocessorMap(), | |
2bb7b766 | 78 | fLogbookEntry(0), |
eba76848 | 79 | fCurrentDetector(), |
85a80aa9 | 80 | fStatusEntry(0), |
cb343cfd | 81 | fMonitoringMutex(0), |
eba76848 | 82 | fLastActionTime(0), |
e7f62f16 | 83 | fLastAction(), |
9827400b | 84 | fMonaLisa(0), |
85 | fTestMode(kNone), | |
ffa29e93 | 86 | fReadTestMode(kFALSE), |
87 | fOutputRedirected(kFALSE) | |
73abe331 | 88 | { |
89 | // | |
90 | // config: AliShuttleConfig used | |
73abe331 | 91 | // timeout: timeout used for AliDCSClient connection |
92 | // retries: the number of retries in case of connection error. | |
93 | // | |
94 | ||
57f50b3c | 95 | if (!fConfig->IsValid()) AliFatal("********** !!!!! Invalid configuration !!!!! **********"); |
be48e3ea | 96 | for(int iSys=0;iSys<4;iSys++) { |
57f50b3c | 97 | fServer[iSys]=0; |
be48e3ea | 98 | if (iSys < 3) |
2c15234c | 99 | fFXSlist[iSys].SetOwner(kTRUE); |
57f50b3c | 100 | } |
2bb7b766 | 101 | fPreprocessorMap.SetOwner(kTRUE); |
be48e3ea | 102 | |
103 | for (UInt_t iDet=0; iDet<NDetectors(); iDet++) | |
104 | fFirstUnprocessed[iDet] = kFALSE; | |
105 | ||
cb343cfd | 106 | fMonitoringMutex = new TMutex(); |
58bc3020 | 107 | } |
108 | ||
b948db8d | 109 | //______________________________________________________________________________________________ |
57f50b3c | 110 | AliShuttle::~AliShuttle() |
58bc3020 | 111 | { |
9827400b | 112 | // |
113 | // destructor | |
114 | // | |
58bc3020 | 115 | |
b948db8d | 116 | fPreprocessorMap.DeleteAll(); |
be48e3ea | 117 | for(int iSys=0;iSys<4;iSys++) |
57f50b3c | 118 | if(fServer[iSys]) { |
119 | fServer[iSys]->Close(); | |
120 | delete fServer[iSys]; | |
eba76848 | 121 | fServer[iSys] = 0; |
57f50b3c | 122 | } |
2bb7b766 | 123 | |
124 | if (fStatusEntry){ | |
125 | delete fStatusEntry; | |
126 | fStatusEntry = 0; | |
127 | } | |
cb343cfd | 128 | |
129 | if (fMonitoringMutex) | |
130 | { | |
131 | delete fMonitoringMutex; | |
132 | fMonitoringMutex = 0; | |
133 | } | |
73abe331 | 134 | } |
135 | ||
b948db8d | 136 | //______________________________________________________________________________________________ |
57f50b3c | 137 | void AliShuttle::RegisterPreprocessor(AliPreprocessor* preprocessor) |
58bc3020 | 138 | { |
73abe331 | 139 | // |
b948db8d | 140 | // Registers new AliPreprocessor. |
73abe331 | 141 | // It uses GetName() for indentificator of the pre processor. |
142 | // The pre processor is registered it there isn't any other | |
143 | // with the same identificator (GetName()). | |
144 | // | |
145 | ||
eba76848 | 146 | const char* detName = preprocessor->GetName(); |
147 | if(GetDetPos(detName) < 0) | |
148 | AliFatal(Form("********** !!!!! Invalid detector name: %s !!!!! **********", detName)); | |
149 | ||
150 | if (fPreprocessorMap.GetValue(detName)) { | |
151 | AliWarning(Form("AliPreprocessor %s is already registered!", detName)); | |
73abe331 | 152 | return; |
153 | } | |
154 | ||
eba76848 | 155 | fPreprocessorMap.Add(new TObjString(detName), preprocessor); |
73abe331 | 156 | } |
b948db8d | 157 | //______________________________________________________________________________________________ |
3301427a | 158 | Bool_t AliShuttle::Store(const AliCDBPath& path, TObject* object, |
84090f85 | 159 | AliCDBMetaData* metaData, Int_t validityStart, Bool_t validityInfinite) |
73abe331 | 160 | { |
9827400b | 161 | // Stores a CDB object in the storage for offline reconstruction. Objects that are not needed for |
162 | // offline reconstruction, but should be stored anyway (e.g. for debugging) should NOT be stored | |
163 | // using this function. Use StoreReferenceData instead! | |
164 | // It calls StoreLocally function which temporarily stores the data locally; when the preprocessor | |
165 | // finishes the data are transferred to the main storage (Grid). | |
b948db8d | 166 | |
3301427a | 167 | return StoreLocally(fgkLocalCDB, path, object, metaData, validityStart, validityInfinite); |
84090f85 | 168 | } |
169 | ||
170 | //______________________________________________________________________________________________ | |
3301427a | 171 | Bool_t AliShuttle::StoreReferenceData(const AliCDBPath& path, TObject* object, AliCDBMetaData* metaData) |
84090f85 | 172 | { |
9827400b | 173 | // Stores a CDB object in the storage for reference data. This objects will not be available during |
174 | // offline reconstrunction. Use this function for reference data only! | |
175 | // It calls StoreLocally function which temporarily stores the data locally; when the preprocessor | |
176 | // finishes the data are transferred to the main storage (Grid). | |
85a80aa9 | 177 | |
3301427a | 178 | return StoreLocally(fgkLocalRefStorage, path, object, metaData); |
85a80aa9 | 179 | } |
180 | ||
181 | //______________________________________________________________________________________________ | |
3301427a | 182 | Bool_t AliShuttle::StoreLocally(const TString& localUri, |
85a80aa9 | 183 | const AliCDBPath& path, TObject* object, AliCDBMetaData* metaData, |
184 | Int_t validityStart, Bool_t validityInfinite) | |
185 | { | |
9827400b | 186 | // Store object temporarily in local storage. Parameters are passed by Store and StoreReferenceData functions. |
187 | // when the preprocessor finishes the data are transferred to the main storage (Grid). | |
188 | // The parameters are: | |
189 | // 1) Uri of the backup storage (Local) | |
190 | // 2) the object's path. | |
191 | // 3) the object to be stored | |
192 | // 4) the metaData to be associated with the object | |
193 | // 5) the validity start run number w.r.t. the current run, | |
194 | // if the data is valid only for this run leave the default 0 | |
195 | // 6) specifies if the calibration data is valid for infinity (this means until updated), | |
196 | // typical for calibration runs, the default is kFALSE | |
197 | // | |
198 | // returns 0 if fail, 1 otherwise | |
84090f85 | 199 | |
9827400b | 200 | if (fTestMode & kErrorStorage) |
201 | { | |
202 | Log(fCurrentDetector, "StoreLocally - In TESTMODE - Simulating error while storing locally"); | |
203 | return kFALSE; | |
204 | } | |
205 | ||
3301427a | 206 | const char* cdbType = (localUri == fgkLocalCDB) ? "CDB" : "Reference"; |
2bb7b766 | 207 | |
85a80aa9 | 208 | Int_t firstRun = GetCurrentRun() - validityStart; |
84090f85 | 209 | if(firstRun < 0) { |
9827400b | 210 | AliWarning("First valid run happens to be less than 0! Setting it to 0."); |
84090f85 | 211 | firstRun=0; |
212 | } | |
213 | ||
214 | Int_t lastRun = -1; | |
215 | if(validityInfinite) { | |
216 | lastRun = AliCDBRunRange::Infinity(); | |
217 | } else { | |
218 | lastRun = GetCurrentRun(); | |
219 | } | |
220 | ||
3301427a | 221 | // Version is set to current run, it will be used later to transfer data to Grid |
222 | AliCDBId id(path, firstRun, lastRun, GetCurrentRun(), -1); | |
2bb7b766 | 223 | |
224 | if(! dynamic_cast<TObjString*> (metaData->GetProperty("RunUsed(TObjString)"))){ | |
225 | TObjString runUsed = Form("%d", GetCurrentRun()); | |
9e080f92 | 226 | metaData->SetProperty("RunUsed(TObjString)", runUsed.Clone()); |
2bb7b766 | 227 | } |
84090f85 | 228 | |
3301427a | 229 | Bool_t result = kFALSE; |
84090f85 | 230 | |
3301427a | 231 | if (!(AliCDBManager::Instance()->GetStorage(localUri))) { |
232 | Log("SHUTTLE", Form("StoreLocally - Cannot activate local %s storage", cdbType)); | |
84090f85 | 233 | } else { |
3301427a | 234 | result = AliCDBManager::Instance()->GetStorage(localUri) |
84090f85 | 235 | ->Put(object, id, metaData); |
236 | } | |
237 | ||
238 | if(!result) { | |
239 | ||
9827400b | 240 | Log(fCurrentDetector, Form("StoreLocally - Can't store object <%s>!", id.ToString().Data())); |
3301427a | 241 | } |
2bb7b766 | 242 | |
3301427a | 243 | return result; |
244 | } | |
84090f85 | 245 | |
3301427a | 246 | //______________________________________________________________________________________________ |
247 | Bool_t AliShuttle::StoreOCDB() | |
248 | { | |
9827400b | 249 | // |
250 | // Called when preprocessor ends successfully or when previous storage attempt failed (kStoreError status) | |
251 | // Calls underlying StoreOCDB(const char*) function twice, for OCDB and Reference storage. | |
252 | // Then calls StoreRefFilesToGrid to store reference files. | |
253 | // | |
254 | ||
255 | if (fTestMode & kErrorGrid) | |
256 | { | |
257 | Log("SHUTTLE", "StoreOCDB - In TESTMODE - Simulating error while storing in the Grid"); | |
258 | Log(fCurrentDetector, "StoreOCDB - In TESTMODE - Simulating error while storing in the Grid"); | |
259 | return kFALSE; | |
260 | } | |
261 | ||
c88ad5db | 262 | Log("SHUTTLE","StoreOCDB - Storing OCDB data ..."); |
86aa42c3 | 263 | Bool_t resultCDB = StoreOCDB(fgkMainCDB); |
264 | ||
c88ad5db | 265 | Log("SHUTTLE","StoreOCDB - Storing reference data ..."); |
3301427a | 266 | Bool_t resultRef = StoreOCDB(fgkMainRefStorage); |
9827400b | 267 | |
c88ad5db | 268 | Log("SHUTTLE","StoreOCDB - Storing reference files ..."); |
269 | Bool_t resultRefFiles = CopyFilesToGrid("reference"); | |
270 | ||
271 | Bool_t resultMetadata = kTRUE; | |
272 | if(fCurrentDetector == "GRP") | |
273 | { | |
274 | Log("StoreOCDB - SHUTTLE","Storing Run Metadata file ..."); | |
275 | resultMetadata = CopyFilesToGrid("metadata"); | |
276 | } | |
9827400b | 277 | |
c88ad5db | 278 | return resultCDB && resultRef && resultRefFiles && resultMetadata; |
3301427a | 279 | } |
280 | ||
281 | //______________________________________________________________________________________________ | |
282 | Bool_t AliShuttle::StoreOCDB(const TString& gridURI) | |
283 | { | |
284 | // | |
285 | // Called by StoreOCDB(), performs actual storage to the main OCDB and reference storages (Grid) | |
286 | // | |
287 | ||
288 | TObjArray* gridIds=0; | |
289 | ||
290 | Bool_t result = kTRUE; | |
291 | ||
292 | const char* type = 0; | |
293 | TString localURI; | |
294 | if(gridURI == fgkMainCDB) { | |
295 | type = "OCDB"; | |
296 | localURI = fgkLocalCDB; | |
297 | } else if(gridURI == fgkMainRefStorage) { | |
298 | type = "reference"; | |
299 | localURI = fgkLocalRefStorage; | |
300 | } else { | |
301 | AliError(Form("Invalid storage URI: %s", gridURI.Data())); | |
302 | return kFALSE; | |
303 | } | |
304 | ||
305 | AliCDBManager* man = AliCDBManager::Instance(); | |
306 | ||
307 | AliCDBStorage *gridSto = man->GetStorage(gridURI); | |
308 | if(!gridSto) { | |
309 | Log("SHUTTLE", | |
310 | Form("StoreOCDB - cannot activate main %s storage", type)); | |
311 | return kFALSE; | |
312 | } | |
313 | ||
314 | gridIds = gridSto->GetQueryCDBList(); | |
315 | ||
316 | // get objects previously stored in local CDB | |
317 | AliCDBStorage *localSto = man->GetStorage(localURI); | |
318 | if(!localSto) { | |
319 | Log("SHUTTLE", | |
320 | Form("StoreOCDB - cannot activate local %s storage", type)); | |
321 | return kFALSE; | |
322 | } | |
323 | AliCDBPath aPath(GetOfflineDetName(fCurrentDetector.Data()),"*","*"); | |
324 | // Local objects were stored with current run as Grid version! | |
325 | TList* localEntries = localSto->GetAll(aPath.GetPath(), GetCurrentRun(), GetCurrentRun()); | |
326 | localEntries->SetOwner(1); | |
327 | ||
328 | // loop on local stored objects | |
329 | TIter localIter(localEntries); | |
330 | AliCDBEntry *aLocEntry = 0; | |
331 | while((aLocEntry = dynamic_cast<AliCDBEntry*> (localIter.Next()))){ | |
332 | aLocEntry->SetOwner(1); | |
333 | AliCDBId aLocId = aLocEntry->GetId(); | |
334 | aLocEntry->SetVersion(-1); | |
335 | aLocEntry->SetSubVersion(-1); | |
336 | ||
337 | // If local object is valid up to infinity we store it only if it is | |
338 | // the first unprocessed run! | |
339 | if (aLocId.GetLastRun() == AliCDBRunRange::Infinity() && | |
340 | !fFirstUnprocessed[GetDetPos(fCurrentDetector)]) | |
341 | { | |
342 | Log("SHUTTLE", Form("StoreOCDB - %s: object %s has validity infinite but " | |
343 | "there are previous unprocessed runs!", | |
344 | fCurrentDetector.Data(), aLocId.GetPath().Data())); | |
4508ab8b | 345 | result = kFALSE; |
3301427a | 346 | continue; |
347 | } | |
348 | ||
349 | // loop on Grid valid Id's | |
350 | Bool_t store = kTRUE; | |
351 | TIter gridIter(gridIds); | |
352 | AliCDBId* aGridId = 0; | |
353 | while((aGridId = dynamic_cast<AliCDBId*> (gridIter.Next()))){ | |
354 | if(aGridId->GetPath() != aLocId.GetPath()) continue; | |
355 | // skip all objects valid up to infinity | |
356 | if(aGridId->GetLastRun() == AliCDBRunRange::Infinity()) continue; | |
357 | // if we get here, it means there's already some more recent object stored on Grid! | |
358 | store = kFALSE; | |
359 | break; | |
360 | } | |
361 | ||
362 | // If we get here, the file can be stored! | |
363 | Bool_t storeOk = gridSto->Put(aLocEntry); | |
364 | if(!store || storeOk){ | |
365 | ||
366 | if (!store) | |
367 | { | |
368 | Log(fCurrentDetector.Data(), | |
369 | Form("StoreOCDB - A more recent object already exists in %s storage: <%s>", | |
370 | type, aGridId->ToString().Data())); | |
371 | } else { | |
372 | Log("SHUTTLE", | |
373 | Form("StoreOCDB - Object <%s> successfully put into %s storage", | |
374 | aLocId.ToString().Data(), type)); | |
2d9019b4 | 375 | Log(fCurrentDetector.Data(), |
376 | Form("StoreOCDB - Object <%s> successfully put into %s storage", | |
377 | aLocId.ToString().Data(), type)); | |
3301427a | 378 | } |
84090f85 | 379 | |
3301427a | 380 | // removing local filename... |
381 | TString filename; | |
382 | localSto->IdToFilename(aLocId, filename); | |
c88ad5db | 383 | Log("SHUTTLE", Form("StoreOCDB - Removing local file %s", filename.Data())); |
3301427a | 384 | RemoveFile(filename.Data()); |
385 | continue; | |
386 | } else { | |
387 | Log("SHUTTLE", | |
388 | Form("StoreOCDB - Grid %s storage of object <%s> failed", | |
389 | type, aLocId.ToString().Data())); | |
2d9019b4 | 390 | Log(fCurrentDetector.Data(), |
391 | Form("StoreOCDB - Grid %s storage of object <%s> failed", | |
392 | type, aLocId.ToString().Data())); | |
3301427a | 393 | result = kFALSE; |
b948db8d | 394 | } |
395 | } | |
3301427a | 396 | localEntries->Clear(); |
2bb7b766 | 397 | |
b948db8d | 398 | return result; |
3301427a | 399 | } |
400 | ||
546242fb | 401 | //______________________________________________________________________________________________ |
402 | Bool_t AliShuttle::CleanReferenceStorage(const char* detector) | |
403 | { | |
2d9019b4 | 404 | // clears the directory used to store reference files of a given subdetector |
546242fb | 405 | |
406 | AliCDBManager* man = AliCDBManager::Instance(); | |
407 | AliCDBStorage* sto = man->GetStorage(fgkLocalRefStorage); | |
2d9019b4 | 408 | TString localBaseFolder = sto->GetBaseFolder(); |
409 | ||
410 | TString targetDir = GetRefFilePrefix(localBaseFolder.Data(), detector); | |
411 | ||
d524ade6 | 412 | Log("SHUTTLE", Form("CleanReferenceStorage - Cleaning %s", targetDir.Data())); |
2d9019b4 | 413 | |
414 | TString begin; | |
415 | begin.Form("%d_", GetCurrentRun()); | |
416 | ||
417 | TSystemDirectory* baseDir = new TSystemDirectory("/", targetDir); | |
418 | if (!baseDir) | |
419 | return kTRUE; | |
420 | ||
421 | TList* dirList = baseDir->GetListOfFiles(); | |
422 | delete baseDir; | |
423 | ||
424 | if (!dirList) return kTRUE; | |
425 | ||
426 | if (dirList->GetEntries() < 3) | |
427 | { | |
428 | delete dirList; | |
429 | return kTRUE; | |
430 | } | |
431 | ||
432 | Int_t nDirs = 0, nDel = 0; | |
433 | TIter dirIter(dirList); | |
434 | TSystemFile* entry = 0; | |
546242fb | 435 | |
2d9019b4 | 436 | Bool_t success = kTRUE; |
546242fb | 437 | |
2d9019b4 | 438 | while ((entry = dynamic_cast<TSystemFile*> (dirIter.Next()))) |
439 | { | |
440 | if (entry->IsDirectory()) | |
441 | continue; | |
442 | ||
443 | TString fileName(entry->GetName()); | |
444 | if (!fileName.BeginsWith(begin)) | |
445 | continue; | |
446 | ||
447 | nDirs++; | |
448 | ||
449 | // delete file | |
450 | Int_t result = gSystem->Unlink(fileName.Data()); | |
451 | ||
452 | if (result) | |
453 | { | |
d524ade6 | 454 | Log("SHUTTLE", Form("CleanReferenceStorage - Could not delete file %s!", fileName.Data())); |
2d9019b4 | 455 | success = kFALSE; |
456 | } else { | |
457 | nDel++; | |
458 | } | |
459 | } | |
460 | ||
461 | if(nDirs > 0) | |
462 | Log("SHUTTLE", Form("CleanReferenceStorage - %d (over %d) reference files in folder %s were deleted.", | |
463 | nDel, nDirs, targetDir.Data())); | |
464 | ||
465 | ||
466 | delete dirList; | |
467 | return success; | |
468 | ||
469 | ||
470 | ||
471 | ||
472 | ||
546242fb | 473 | |
474 | Int_t result = gSystem->GetPathInfo(targetDir, 0, (Long64_t*) 0, 0, 0); | |
475 | if (result == 0) | |
476 | { | |
477 | // delete directory | |
d524ade6 | 478 | result = gSystem->Exec(Form("rm -rf %s", targetDir.Data())); |
546242fb | 479 | if (result != 0) |
480 | { | |
d524ade6 | 481 | Log("SHUTTLE", Form("CleanReferenceStorage - Could not clean directory %s", targetDir.Data())); |
546242fb | 482 | return kFALSE; |
483 | } | |
484 | } | |
485 | ||
486 | result = gSystem->mkdir(targetDir, kTRUE); | |
487 | if (result != 0) | |
488 | { | |
c88ad5db | 489 | Log("SHUTTLE", Form("CleanReferenceStorage - Error creating base directory %s", targetDir.Data())); |
546242fb | 490 | return kFALSE; |
491 | } | |
492 | ||
493 | return kTRUE; | |
494 | } | |
495 | ||
9827400b | 496 | //______________________________________________________________________________________________ |
497 | Bool_t AliShuttle::StoreReferenceFile(const char* detector, const char* localFile, const char* gridFileName) | |
498 | { | |
499 | // | |
3c2a21c8 | 500 | // Stores reference file directly (without opening it). This function stores the file locally. |
9827400b | 501 | // |
3c2a21c8 | 502 | // The file is stored under the following location: |
503 | // <base folder of local reference storage>/<DET>/<RUN#>_<gridFileName> | |
504 | // where <gridFileName> is the second parameter given to the function | |
505 | // | |
9827400b | 506 | |
507 | if (fTestMode & kErrorStorage) | |
508 | { | |
509 | Log(fCurrentDetector, "StoreReferenceFile - In TESTMODE - Simulating error while storing locally"); | |
510 | return kFALSE; | |
511 | } | |
512 | ||
513 | AliCDBManager* man = AliCDBManager::Instance(); | |
514 | AliCDBStorage* sto = man->GetStorage(fgkLocalRefStorage); | |
515 | ||
516 | TString localBaseFolder = sto->GetBaseFolder(); | |
517 | ||
d524ade6 | 518 | TString target = GetRefFilePrefix(localBaseFolder.Data(), detector); |
519 | target.Append(Form("/%d_%s", GetCurrentRun(), gridFileName)); | |
9827400b | 520 | |
d524ade6 | 521 | return CopyFileLocally(localFile, target); |
c88ad5db | 522 | } |
523 | ||
524 | //______________________________________________________________________________________________ | |
525 | Bool_t AliShuttle::StoreRunMetadataFile(const char* localFile, const char* gridFileName) | |
526 | { | |
527 | // | |
528 | // Stores Run metadata file to the Grid, in the run folder | |
529 | // | |
530 | // Only GRP can call this function. | |
531 | ||
532 | if (fTestMode & kErrorStorage) | |
533 | { | |
534 | Log(fCurrentDetector, "StoreRunMetaDataFile - In TESTMODE - Simulating error while storing locally"); | |
535 | return kFALSE; | |
536 | } | |
537 | ||
538 | AliCDBManager* man = AliCDBManager::Instance(); | |
539 | AliCDBStorage* sto = man->GetStorage(fgkLocalRefStorage); | |
540 | ||
541 | TString localBaseFolder = sto->GetBaseFolder(); | |
542 | ||
543 | // Build Run level folder | |
955d2abc | 544 | // folder = /alice/data/year/lhcPeriod/runNb/raw |
c88ad5db | 545 | |
c88ad5db | 546 | |
675f64cd | 547 | TString lhcPeriod = GetLHCPeriod(); |
c88ad5db | 548 | if (lhcPeriod.Length() == 0) |
549 | { | |
550 | Log("SHUTTLE","StoreRunMetaDataFile - LHCPeriod not found in logbook!"); | |
551 | return 0; | |
552 | } | |
a1c5723b | 553 | |
554 | // TODO partitions with one detector only write data into LHCperiod_DET | |
2694f253 | 555 | TString partition = GetRunParameter("detector"); |
a1c5723b | 556 | |
557 | if (partition.Length() > 0 && partition != "ALICE") | |
558 | { | |
559 | lhcPeriod.Append(Form("_%s", partition.Data())); | |
560 | Log(fCurrentDetector, Form("Run data tags merged file will be written in %s", | |
561 | lhcPeriod.Data())); | |
562 | } | |
8884d71e | 563 | |
955d2abc | 564 | TString target = Form("%s/GRP/RunMetadata/alice/data/%d/%s/%09d/raw/%s", |
675f64cd | 565 | localBaseFolder.Data(), GetCurrentYear(), |
d524ade6 | 566 | lhcPeriod.Data(), GetCurrentRun(), gridFileName); |
c88ad5db | 567 | |
d524ade6 | 568 | return CopyFileLocally(localFile, target); |
c88ad5db | 569 | } |
570 | ||
571 | //______________________________________________________________________________________________ | |
d524ade6 | 572 | Bool_t AliShuttle::CopyFileLocally(const char* localFile, const TString& target) |
c88ad5db | 573 | { |
574 | // | |
575 | // Stores file locally. Called by StoreReferenceFile and StoreRunMetadataFile | |
d524ade6 | 576 | // Files are temporarily stored in the local reference storage. When the preprocessor |
577 | // finishes, the Shuttle calls CopyFilesToGrid to transfer the files to AliEn | |
578 | // (in reference or run level folders) | |
c88ad5db | 579 | // |
580 | ||
d524ade6 | 581 | TString targetDir(target(0, target.Last('/'))); |
582 | ||
583 | //try to open base dir folder, if it does not exist | |
2d9019b4 | 584 | void* dir = gSystem->OpenDirectory(targetDir.Data()); |
585 | if (dir == NULL) { | |
586 | if (gSystem->mkdir(targetDir.Data(), kTRUE)) { | |
a1c5723b | 587 | Log("SHUTTLE", Form("CopyFileLocally - Can't open directory <%s>", targetDir.Data())); |
2d9019b4 | 588 | return kFALSE; |
589 | } | |
590 | ||
591 | } else { | |
592 | gSystem->FreeDirectory(dir); | |
593 | } | |
9827400b | 594 | |
7d43a416 | 595 | Int_t result = 0; |
596 | ||
597 | result = gSystem->GetPathInfo(localFile, 0, (Long64_t*) 0, 0, 0); | |
9827400b | 598 | if (result) |
599 | { | |
a1c5723b | 600 | Log("SHUTTLE", Form("CopyFileLocally - %s does not exist", localFile)); |
546242fb | 601 | return kFALSE; |
9827400b | 602 | } |
546242fb | 603 | |
7d43a416 | 604 | result = gSystem->GetPathInfo(target, 0, (Long64_t*) 0, 0, 0); |
605 | if (!result) | |
606 | { | |
a1c5723b | 607 | Log("SHUTTLE", Form("CopyFileLocally - target file %s already exist, removing...", target.Data())); |
7d43a416 | 608 | if (gSystem->Unlink(target.Data())) |
609 | { | |
a1c5723b | 610 | Log("SHUTTLE", Form("CopyFileLocally - Could not remove existing target file %s!", target.Data())); |
7d43a416 | 611 | return kFALSE; |
612 | } | |
613 | } | |
614 | ||
9827400b | 615 | result = gSystem->CopyFile(localFile, target); |
616 | ||
617 | if (result == 0) | |
618 | { | |
a1c5723b | 619 | Log("SHUTTLE", Form("CopyFileLocally - File %s stored locally to %s", localFile, target.Data())); |
9827400b | 620 | return kTRUE; |
621 | } | |
622 | else | |
623 | { | |
a1c5723b | 624 | Log("SHUTTLE", Form("CopyFileLocally - Could not store file %s to %s! Error code = %d", |
546242fb | 625 | localFile, target.Data(), result)); |
9827400b | 626 | return kFALSE; |
627 | } | |
c88ad5db | 628 | |
629 | ||
630 | ||
9827400b | 631 | } |
632 | ||
633 | //______________________________________________________________________________________________ | |
c88ad5db | 634 | Bool_t AliShuttle::CopyFilesToGrid(const char* type) |
9827400b | 635 | { |
636 | // | |
c88ad5db | 637 | // Transfers local files to the Grid. Local files can be reference files |
638 | // or run metadata file (from GRP only). | |
9827400b | 639 | // |
c88ad5db | 640 | // According to the type (ref, metadata) the files are stored under the following location: |
641 | // ref --> <base folder of reference storage>/<DET>/<RUN#>_<gridFileName> | |
642 | // metadata --> <run data folder>/<MetadataFileName> | |
86aa42c3 | 643 | // |
c88ad5db | 644 | |
9827400b | 645 | AliCDBManager* man = AliCDBManager::Instance(); |
646 | AliCDBStorage* sto = man->GetStorage(fgkLocalRefStorage); | |
647 | if (!sto) | |
648 | return kFALSE; | |
649 | TString localBaseFolder = sto->GetBaseFolder(); | |
9827400b | 650 | |
c88ad5db | 651 | TString dir; |
652 | TString alienDir; | |
9827400b | 653 | TString begin; |
9827400b | 654 | |
c88ad5db | 655 | if (strcmp(type, "reference") == 0) |
656 | { | |
657 | dir = GetRefFilePrefix(localBaseFolder.Data(), fCurrentDetector.Data()); | |
658 | AliCDBStorage* gridSto = man->GetStorage(fgkMainRefStorage); | |
659 | if (!gridSto) | |
660 | return kFALSE; | |
661 | TString gridBaseFolder = gridSto->GetBaseFolder(); | |
662 | alienDir = GetRefFilePrefix(gridBaseFolder.Data(), fCurrentDetector.Data()); | |
663 | begin = Form("%d_", GetCurrentRun()); | |
664 | } | |
665 | else if (strcmp(type, "metadata") == 0) | |
666 | { | |
c88ad5db | 667 | |
675f64cd | 668 | TString lhcPeriod = GetLHCPeriod(); |
c88ad5db | 669 | |
670 | if (lhcPeriod.Length() == 0) | |
671 | { | |
672 | Log("SHUTTLE","CopyFilesToGrid - LHCPeriod not found in logbook!"); | |
673 | return 0; | |
674 | } | |
675 | ||
a1c5723b | 676 | // TODO partitions with one detector only write data into LHCperiod_DET |
2694f253 | 677 | TString partition = GetRunParameter("detector"); |
a1c5723b | 678 | |
679 | if (partition.Length() > 0 && partition != "ALICE") | |
680 | { | |
681 | lhcPeriod.Append(Form("_%s", partition.Data())); | |
682 | } | |
683 | ||
955d2abc | 684 | dir = Form("%s/GRP/RunMetadata/alice/data/%d/%s/%09d/raw", |
675f64cd | 685 | localBaseFolder.Data(), GetCurrentYear(), |
c88ad5db | 686 | lhcPeriod.Data(), GetCurrentRun()); |
675f64cd | 687 | alienDir = dir(dir.Index("/alice/data/"), dir.Length()); |
688 | ||
c88ad5db | 689 | begin = ""; |
690 | } | |
691 | else | |
692 | { | |
693 | Log("SHUTTLE", "CopyFilesToGrid - Unexpected: type label must be reference or metadata!"); | |
694 | return kFALSE; | |
695 | } | |
696 | ||
9827400b | 697 | TSystemDirectory* baseDir = new TSystemDirectory("/", dir); |
3d8bc902 | 698 | if (!baseDir) |
699 | return kTRUE; | |
700 | ||
2d9019b4 | 701 | TList* dirList = baseDir->GetListOfFiles(); |
702 | delete baseDir; | |
703 | ||
704 | if (!dirList) return kTRUE; | |
705 | ||
706 | if (dirList->GetEntries() < 3) | |
3d8bc902 | 707 | { |
2d9019b4 | 708 | delete dirList; |
9827400b | 709 | return kTRUE; |
3d8bc902 | 710 | } |
2d9019b4 | 711 | |
546242fb | 712 | if (!gGrid) |
713 | { | |
c88ad5db | 714 | Log("SHUTTLE", "CopyFilesToGrid - Connection to Grid failed: Cannot continue!"); |
2d9019b4 | 715 | delete dirList; |
546242fb | 716 | return kFALSE; |
717 | } | |
718 | ||
2d9019b4 | 719 | Int_t nDirs = 0, nTransfer = 0; |
720 | TIter dirIter(dirList); | |
721 | TSystemFile* entry = 0; | |
722 | ||
9827400b | 723 | Bool_t success = kTRUE; |
3d8bc902 | 724 | Bool_t first = kTRUE; |
9827400b | 725 | |
2d9019b4 | 726 | while ((entry = dynamic_cast<TSystemFile*> (dirIter.Next()))) |
727 | { | |
9827400b | 728 | if (entry->IsDirectory()) |
729 | continue; | |
730 | ||
731 | TString fileName(entry->GetName()); | |
732 | if (!fileName.BeginsWith(begin)) | |
733 | continue; | |
734 | ||
2d9019b4 | 735 | nDirs++; |
736 | ||
3d8bc902 | 737 | if (first) |
738 | { | |
739 | first = kFALSE; | |
c88ad5db | 740 | // check that folder exists, otherwise create it |
3d8bc902 | 741 | TGridResult* result = gGrid->Ls(alienDir.Data(), "a"); |
742 | ||
743 | if (!result) | |
2d9019b4 | 744 | { |
745 | delete dirList; | |
3d8bc902 | 746 | return kFALSE; |
2d9019b4 | 747 | } |
3d8bc902 | 748 | |
546242fb | 749 | if (!result->GetFileName(1)) // TODO: It looks like element 0 is always 0!! |
3d8bc902 | 750 | { |
675f64cd | 751 | // TODO It does not work currently! Bug in TAliEn::Mkdir |
752 | // TODO Manually fixed in local root v5-16-00 | |
c88ad5db | 753 | if (!gGrid->Mkdir(alienDir.Data(),"-p",0)) |
3d8bc902 | 754 | { |
c88ad5db | 755 | Log("SHUTTLE", Form("CopyFilesToGrid - Cannot create directory %s", |
3d8bc902 | 756 | alienDir.Data())); |
2d9019b4 | 757 | delete dirList; |
3d8bc902 | 758 | return kFALSE; |
546242fb | 759 | } else { |
c88ad5db | 760 | Log("SHUTTLE",Form("CopyFilesToGrid - Folder %s created", alienDir.Data())); |
3d8bc902 | 761 | } |
762 | ||
546242fb | 763 | } else { |
c88ad5db | 764 | Log("SHUTTLE",Form("CopyFilesToGrid - Folder %s found", alienDir.Data())); |
3d8bc902 | 765 | } |
766 | } | |
767 | ||
9827400b | 768 | TString fullLocalPath; |
769 | fullLocalPath.Form("%s/%s", dir.Data(), fileName.Data()); | |
770 | ||
771 | TString fullGridPath; | |
772 | fullGridPath.Form("alien://%s/%s", alienDir.Data(), fileName.Data()); | |
773 | ||
a986b218 | 774 | Bool_t result = TFile::Cp(fullLocalPath, fullGridPath); |
9827400b | 775 | |
776 | if (result) | |
777 | { | |
c88ad5db | 778 | Log("SHUTTLE", Form("CopyFilesToGrid - Copying local file %s to %s succeeded!", |
779 | fullLocalPath.Data(), fullGridPath.Data())); | |
9827400b | 780 | RemoveFile(fullLocalPath); |
2d9019b4 | 781 | nTransfer++; |
9827400b | 782 | } |
783 | else | |
784 | { | |
c88ad5db | 785 | Log("SHUTTLE", Form("CopyFilesToGrid - Copying local file %s to %s FAILED!", |
786 | fullLocalPath.Data(), fullGridPath.Data())); | |
9827400b | 787 | success = kFALSE; |
788 | } | |
789 | } | |
2d9019b4 | 790 | |
c88ad5db | 791 | Log("SHUTTLE", Form("CopyFilesToGrid - %d (over %d) files in folder %s copied to Grid.", |
792 | nTransfer, nDirs, dir.Data())); | |
2d9019b4 | 793 | |
794 | ||
795 | delete dirList; | |
9827400b | 796 | return success; |
797 | } | |
798 | ||
2d9019b4 | 799 | //______________________________________________________________________________________________ |
800 | const char* AliShuttle::GetRefFilePrefix(const char* base, const char* detector) | |
801 | { | |
802 | // | |
803 | // Get folder name of reference files | |
804 | // | |
805 | ||
806 | TString offDetStr(GetOfflineDetName(detector)); | |
807 | TString dir; | |
808 | if (offDetStr == "ITS" || offDetStr == "MUON" || offDetStr == "PHOS") | |
809 | { | |
810 | dir.Form("%s/%s/%s", base, offDetStr.Data(), detector); | |
811 | } else { | |
812 | dir.Form("%s/%s", base, offDetStr.Data()); | |
813 | } | |
814 | ||
815 | return dir.Data(); | |
816 | ||
817 | ||
818 | } | |
c88ad5db | 819 | |
3301427a | 820 | //______________________________________________________________________________________________ |
821 | void AliShuttle::CleanLocalStorage(const TString& uri) | |
822 | { | |
9827400b | 823 | // |
824 | // Called in case the preprocessor is declared failed. Remove remaining objects from the local storages. | |
825 | // | |
3301427a | 826 | |
827 | const char* type = 0; | |
828 | if(uri == fgkLocalCDB) { | |
829 | type = "OCDB"; | |
830 | } else if(uri == fgkLocalRefStorage) { | |
546242fb | 831 | type = "Reference"; |
3301427a | 832 | } else { |
833 | AliError(Form("Invalid storage URI: %s", uri.Data())); | |
834 | return; | |
835 | } | |
836 | ||
837 | AliCDBManager* man = AliCDBManager::Instance(); | |
b948db8d | 838 | |
3301427a | 839 | // open local storage |
840 | AliCDBStorage *localSto = man->GetStorage(uri); | |
841 | if(!localSto) { | |
842 | Log("SHUTTLE", | |
843 | Form("CleanLocalStorage - cannot activate local %s storage", type)); | |
844 | return; | |
845 | } | |
846 | ||
847 | TString filename(Form("%s/%s/*/Run*_v%d_s*.root", | |
546242fb | 848 | localSto->GetBaseFolder().Data(), GetOfflineDetName(fCurrentDetector.Data()), GetCurrentRun())); |
3301427a | 849 | |
c88ad5db | 850 | AliDebug(2, Form("filename = %s", filename.Data())); |
3301427a | 851 | |
c88ad5db | 852 | Log("SHUTTLE", Form("Removing remaining local files for run %d and detector %s ...", |
3301427a | 853 | GetCurrentRun(), fCurrentDetector.Data())); |
854 | ||
855 | RemoveFile(filename.Data()); | |
856 | ||
857 | } | |
858 | ||
859 | //______________________________________________________________________________________________ | |
860 | void AliShuttle::RemoveFile(const char* filename) | |
861 | { | |
9827400b | 862 | // |
863 | // removes local file | |
864 | // | |
3301427a | 865 | |
866 | TString command(Form("rm -f %s", filename)); | |
867 | ||
868 | Int_t result = gSystem->Exec(command.Data()); | |
869 | if(result != 0) | |
870 | { | |
871 | Log("SHUTTLE", Form("RemoveFile - %s: Cannot remove file %s!", | |
872 | fCurrentDetector.Data(), filename)); | |
873 | } | |
73abe331 | 874 | } |
875 | ||
b948db8d | 876 | //______________________________________________________________________________________________ |
5164a766 | 877 | AliShuttleStatus* AliShuttle::ReadShuttleStatus() |
878 | { | |
9827400b | 879 | // |
880 | // Reads the AliShuttleStatus from the CDB | |
881 | // | |
5164a766 | 882 | |
2bb7b766 | 883 | if (fStatusEntry){ |
884 | delete fStatusEntry; | |
885 | fStatusEntry = 0; | |
886 | } | |
5164a766 | 887 | |
10a5a932 | 888 | fStatusEntry = AliCDBManager::Instance()->GetStorage(GetLocalCDB()) |
2bb7b766 | 889 | ->Get(Form("/SHUTTLE/STATUS/%s", fCurrentDetector.Data()), GetCurrentRun()); |
5164a766 | 890 | |
2bb7b766 | 891 | if (!fStatusEntry) return 0; |
892 | fStatusEntry->SetOwner(1); | |
5164a766 | 893 | |
2bb7b766 | 894 | AliShuttleStatus* status = dynamic_cast<AliShuttleStatus*> (fStatusEntry->GetObject()); |
895 | if (!status) { | |
896 | AliError("Invalid object stored to CDB!"); | |
897 | return 0; | |
898 | } | |
5164a766 | 899 | |
2bb7b766 | 900 | return status; |
5164a766 | 901 | } |
902 | ||
903 | //______________________________________________________________________________________________ | |
7bfb2090 | 904 | Bool_t AliShuttle::WriteShuttleStatus(AliShuttleStatus* status) |
5164a766 | 905 | { |
9827400b | 906 | // |
907 | // writes the status for one subdetector | |
908 | // | |
2bb7b766 | 909 | |
910 | if (fStatusEntry){ | |
911 | delete fStatusEntry; | |
912 | fStatusEntry = 0; | |
913 | } | |
5164a766 | 914 | |
2bb7b766 | 915 | Int_t run = GetCurrentRun(); |
5164a766 | 916 | |
2bb7b766 | 917 | AliCDBId id(AliCDBPath("SHUTTLE", "STATUS", fCurrentDetector), run, run); |
5164a766 | 918 | |
2bb7b766 | 919 | fStatusEntry = new AliCDBEntry(status, id, new AliCDBMetaData); |
920 | fStatusEntry->SetOwner(1); | |
5164a766 | 921 | |
2bb7b766 | 922 | UInt_t result = AliCDBManager::Instance()->GetStorage(fgkLocalCDB)->Put(fStatusEntry); |
7bfb2090 | 923 | |
2bb7b766 | 924 | if (!result) { |
3301427a | 925 | Log("SHUTTLE", Form("WriteShuttleStatus - Failed for %s, run %d", |
926 | fCurrentDetector.Data(), run)); | |
2bb7b766 | 927 | return kFALSE; |
928 | } | |
e7f62f16 | 929 | |
930 | SendMLInfo(); | |
7bfb2090 | 931 | |
2bb7b766 | 932 | return kTRUE; |
5164a766 | 933 | } |
934 | ||
935 | //______________________________________________________________________________________________ | |
936 | void AliShuttle::UpdateShuttleStatus(AliShuttleStatus::Status newStatus, Bool_t increaseCount) | |
937 | { | |
9827400b | 938 | // |
939 | // changes the AliShuttleStatus for the given detector and run to the given status | |
940 | // | |
5164a766 | 941 | |
2bb7b766 | 942 | if (!fStatusEntry){ |
943 | AliError("UNEXPECTED: fStatusEntry empty"); | |
944 | return; | |
945 | } | |
5164a766 | 946 | |
2bb7b766 | 947 | AliShuttleStatus* status = dynamic_cast<AliShuttleStatus*> (fStatusEntry->GetObject()); |
5164a766 | 948 | |
2bb7b766 | 949 | if (!status){ |
c88ad5db | 950 | Log("SHUTTLE", "UpdateShuttleStatus - UNEXPECTED: status could not be read from current CDB entry"); |
2bb7b766 | 951 | return; |
952 | } | |
5164a766 | 953 | |
2c15234c | 954 | TString actionStr = Form("UpdateShuttleStatus - %s: Changing state from %s to %s", |
eba76848 | 955 | fCurrentDetector.Data(), |
36c99a6a | 956 | status->GetStatusName(), |
eba76848 | 957 | status->GetStatusName(newStatus)); |
cb343cfd | 958 | Log("SHUTTLE", actionStr); |
959 | SetLastAction(actionStr); | |
5164a766 | 960 | |
2bb7b766 | 961 | status->SetStatus(newStatus); |
962 | if (increaseCount) status->IncreaseCount(); | |
5164a766 | 963 | |
2bb7b766 | 964 | AliCDBManager::Instance()->GetStorage(fgkLocalCDB)->Put(fStatusEntry); |
e7f62f16 | 965 | |
966 | SendMLInfo(); | |
5164a766 | 967 | } |
e7f62f16 | 968 | |
969 | //______________________________________________________________________________________________ | |
970 | void AliShuttle::SendMLInfo() | |
971 | { | |
972 | // | |
973 | // sends ML information about the current status of the current detector being processed | |
974 | // | |
975 | ||
976 | AliShuttleStatus* status = dynamic_cast<AliShuttleStatus*> (fStatusEntry->GetObject()); | |
977 | ||
978 | if (!status){ | |
3301427a | 979 | Log("SHUTTLE", "SendMLInfo - UNEXPECTED: status could not be read from current CDB entry"); |
e7f62f16 | 980 | return; |
981 | } | |
982 | ||
983 | TMonaLisaText mlStatus(Form("%s_status", fCurrentDetector.Data()), status->GetStatusName()); | |
984 | TMonaLisaValue mlRetryCount(Form("%s_count", fCurrentDetector.Data()), status->GetCount()); | |
985 | ||
986 | TList mlList; | |
987 | mlList.Add(&mlStatus); | |
988 | mlList.Add(&mlRetryCount); | |
989 | ||
ee6f7523 | 990 | TString mlID; |
991 | mlID.Form("%d", GetCurrentRun()); | |
992 | fMonaLisa->SendParameters(&mlList, mlID); | |
e7f62f16 | 993 | } |
994 | ||
5164a766 | 995 | //______________________________________________________________________________________________ |
996 | Bool_t AliShuttle::ContinueProcessing() | |
997 | { | |
9827400b | 998 | // this function reads the AliShuttleStatus information from CDB and |
999 | // checks if the processing should be continued | |
1000 | // if yes it returns kTRUE and updates the AliShuttleStatus with nextStatus | |
2bb7b766 | 1001 | |
57c1a579 | 1002 | if (!fConfig->HostProcessDetector(fCurrentDetector)) return kFALSE; |
1003 | ||
1004 | AliPreprocessor* aPreprocessor = | |
1005 | dynamic_cast<AliPreprocessor*> (fPreprocessorMap.GetValue(fCurrentDetector)); | |
1006 | if (!aPreprocessor) | |
1007 | { | |
c88ad5db | 1008 | Log("SHUTTLE", Form("ContinueProcessing - %s: no preprocessor registered", fCurrentDetector.Data())); |
57c1a579 | 1009 | return kFALSE; |
1010 | } | |
1011 | ||
2bb7b766 | 1012 | AliShuttleLogbookEntry::Status entryStatus = |
eba76848 | 1013 | fLogbookEntry->GetDetectorStatus(fCurrentDetector); |
2bb7b766 | 1014 | |
1015 | if(entryStatus != AliShuttleLogbookEntry::kUnprocessed) { | |
c88ad5db | 1016 | Log("SHUTTLE", Form("ContinueProcessing - %s is %s", |
2bb7b766 | 1017 | fCurrentDetector.Data(), |
1018 | fLogbookEntry->GetDetectorStatusName(entryStatus))); | |
1019 | return kFALSE; | |
1020 | } | |
1021 | ||
1022 | // if we get here, according to Shuttle logbook subdetector is in UNPROCESSED state | |
be48e3ea | 1023 | |
1024 | // check if current run is first unprocessed run for current detector | |
1025 | if (fConfig->StrictRunOrder(fCurrentDetector) && | |
1026 | !fFirstUnprocessed[GetDetPos(fCurrentDetector)]) | |
1027 | { | |
86aa42c3 | 1028 | if (fTestMode == kNone) |
1029 | { | |
c88ad5db | 1030 | Log("SHUTTLE", Form("ContinueProcessing - %s requires strict run ordering" |
1031 | " but this is not the first unprocessed run!")); | |
86aa42c3 | 1032 | return kFALSE; |
1033 | } | |
1034 | else | |
1035 | { | |
c88ad5db | 1036 | Log("SHUTTLE", Form("ContinueProcessing - In TESTMODE - " |
1037 | "Although %s requires strict run ordering " | |
1038 | "and this is not the first unprocessed run, " | |
1039 | "the SHUTTLE continues")); | |
86aa42c3 | 1040 | } |
be48e3ea | 1041 | } |
1042 | ||
2bb7b766 | 1043 | AliShuttleStatus* status = ReadShuttleStatus(); |
1044 | if (!status) { | |
1045 | // first time | |
1046 | Log("SHUTTLE", Form("ContinueProcessing - %s: Processing first time", | |
1047 | fCurrentDetector.Data())); | |
1048 | status = new AliShuttleStatus(AliShuttleStatus::kStarted); | |
1049 | return WriteShuttleStatus(status); | |
1050 | } | |
1051 | ||
1052 | // The following two cases shouldn't happen if Shuttle Logbook was correctly updated. | |
1053 | // If it happens it may mean Logbook updating failed... let's do it now! | |
1054 | if (status->GetStatus() == AliShuttleStatus::kDone || | |
1055 | status->GetStatus() == AliShuttleStatus::kFailed){ | |
1056 | Log("SHUTTLE", Form("ContinueProcessing - %s is already %s. Updating Shuttle Logbook", | |
1057 | fCurrentDetector.Data(), | |
1058 | status->GetStatusName(status->GetStatus()))); | |
1059 | UpdateShuttleLogbook(fCurrentDetector.Data(), | |
1060 | status->GetStatusName(status->GetStatus())); | |
1061 | return kFALSE; | |
1062 | } | |
1063 | ||
c42350c2 | 1064 | if (status->GetStatus() == AliShuttleStatus::kStoreStarted || status->GetStatus() == AliShuttleStatus::kStoreError) { |
2bb7b766 | 1065 | Log("SHUTTLE", |
c88ad5db | 1066 | Form("ContinueProcessing - %s: Grid storage of one or more " |
1067 | "objects failed. Trying again now", | |
2bb7b766 | 1068 | fCurrentDetector.Data())); |
9827400b | 1069 | UpdateShuttleStatus(AliShuttleStatus::kStoreStarted); |
1070 | if (StoreOCDB()){ | |
c88ad5db | 1071 | Log("SHUTTLE", Form("ContinueProcessing - %s: all objects " |
1072 | "successfully stored into main storage", | |
3301427a | 1073 | fCurrentDetector.Data())); |
a0aaf247 | 1074 | UpdateShuttleStatus(AliShuttleStatus::kDone); |
1075 | UpdateShuttleLogbook(fCurrentDetector, "DONE"); | |
2bb7b766 | 1076 | } else { |
1077 | Log("SHUTTLE", | |
1078 | Form("ContinueProcessing - %s: Grid storage failed again", | |
1079 | fCurrentDetector.Data())); | |
9827400b | 1080 | UpdateShuttleStatus(AliShuttleStatus::kStoreError); |
2bb7b766 | 1081 | } |
1082 | return kFALSE; | |
1083 | } | |
1084 | ||
1085 | // if we get here, there is a restart | |
57c1a579 | 1086 | Bool_t cont = kFALSE; |
2bb7b766 | 1087 | |
1088 | // abort conditions | |
cb343cfd | 1089 | if (status->GetCount() >= fConfig->GetMaxRetries()) { |
57c1a579 | 1090 | Log("SHUTTLE", Form("ContinueProcessing - %s failed %d times in status %s - " |
1091 | "Updating Shuttle Logbook", fCurrentDetector.Data(), | |
2bb7b766 | 1092 | status->GetCount(), status->GetStatusName())); |
1093 | UpdateShuttleLogbook(fCurrentDetector.Data(), "FAILED"); | |
e7f62f16 | 1094 | UpdateShuttleStatus(AliShuttleStatus::kFailed); |
3301427a | 1095 | |
1096 | // there may still be objects in local OCDB and reference storage | |
1097 | // and FXS databases may be not updated: do it now! | |
9827400b | 1098 | |
1099 | // TODO Currently disabled, we want to keep files in case of failure! | |
1100 | // CleanLocalStorage(fgkLocalCDB); | |
1101 | // CleanLocalStorage(fgkLocalRefStorage); | |
1102 | // UpdateTableFailCase(); | |
1103 | ||
1104 | // Send mail to detector expert! | |
c88ad5db | 1105 | Log("SHUTTLE", Form("ContinueProcessing - Sending mail to %s expert...", |
1106 | fCurrentDetector.Data())); | |
9827400b | 1107 | if (!SendMail()) |
1108 | Log("SHUTTLE", Form("ContinueProcessing - Could not send mail to %s expert", | |
1109 | fCurrentDetector.Data())); | |
3301427a | 1110 | |
57c1a579 | 1111 | } else { |
1112 | Log("SHUTTLE", Form("ContinueProcessing - %s: restarting. " | |
1113 | "Aborted before with %s. Retry number %d.", fCurrentDetector.Data(), | |
1114 | status->GetStatusName(), status->GetCount())); | |
9827400b | 1115 | Bool_t increaseCount = kTRUE; |
c88ad5db | 1116 | if (status->GetStatus() == AliShuttleStatus::kDCSError || |
1117 | status->GetStatus() == AliShuttleStatus::kDCSStarted) | |
1118 | increaseCount = kFALSE; | |
675f64cd | 1119 | |
9827400b | 1120 | UpdateShuttleStatus(AliShuttleStatus::kStarted, increaseCount); |
57c1a579 | 1121 | cont = kTRUE; |
2bb7b766 | 1122 | } |
1123 | ||
57c1a579 | 1124 | return cont; |
5164a766 | 1125 | } |
1126 | ||
1127 | //______________________________________________________________________________________________ | |
2bb7b766 | 1128 | Bool_t AliShuttle::Process(AliShuttleLogbookEntry* entry) |
58bc3020 | 1129 | { |
73abe331 | 1130 | // |
b948db8d | 1131 | // Makes data retrieval for all detectors in the configuration. |
2bb7b766 | 1132 | // entry: Shuttle logbook entry, contains run paramenters and status of detectors |
1133 | // (Unprocessed, Inactive, Failed or Done). | |
d477ad88 | 1134 | // Returns kFALSE in case of error occured and kTRUE otherwise |
73abe331 | 1135 | // |
1136 | ||
9827400b | 1137 | if (!entry) return kFALSE; |
2bb7b766 | 1138 | |
1139 | fLogbookEntry = entry; | |
1140 | ||
c88ad5db | 1141 | Log("SHUTTLE", Form("\t\t\t^*^*^*^*^*^*^*^*^*^*^*^* run %d: START ^*^*^*^*^*^*^*^*^*^*^*^*", |
9827400b | 1142 | GetCurrentRun())); |
2bb7b766 | 1143 | |
e7f62f16 | 1144 | // Send the information to ML |
1145 | TMonaLisaText mlStatus("SHUTTLE_status", "Processing"); | |
9827400b | 1146 | TMonaLisaText mlRunType("SHUTTLE_runtype", Form("%s (%s)", entry->GetRunType(), entry->GetRunParameter("log"))); |
e7f62f16 | 1147 | |
1148 | TList mlList; | |
1149 | mlList.Add(&mlStatus); | |
9827400b | 1150 | mlList.Add(&mlRunType); |
e7f62f16 | 1151 | |
ee6f7523 | 1152 | TString mlID; |
1153 | mlID.Form("%d", GetCurrentRun()); | |
1154 | fMonaLisa->SendParameters(&mlList, mlID); | |
3301427a | 1155 | |
9827400b | 1156 | if (fLogbookEntry->IsDone()) |
1157 | { | |
1158 | Log("SHUTTLE","Process - Shuttle is already DONE. Updating logbook"); | |
1159 | UpdateShuttleLogbook("shuttle_done"); | |
1160 | fLogbookEntry = 0; | |
1161 | return kTRUE; | |
1162 | } | |
1163 | ||
1164 | // read test mode if flag is set | |
1165 | if (fReadTestMode) | |
1166 | { | |
3d8bc902 | 1167 | fTestMode = kNone; |
9827400b | 1168 | TString logEntry(entry->GetRunParameter("log")); |
1169 | //printf("log entry = %s\n", logEntry.Data()); | |
1170 | TString searchStr("Testmode: "); | |
1171 | Int_t pos = logEntry.Index(searchStr.Data()); | |
1172 | //printf("%d\n", pos); | |
1173 | if (pos >= 0) | |
1174 | { | |
1175 | TSubString subStr = logEntry(pos + searchStr.Length(), logEntry.Length()); | |
1176 | //printf("%s\n", subStr.String().Data()); | |
1177 | TString newStr(subStr.Data()); | |
1178 | TObjArray* token = newStr.Tokenize(' '); | |
1179 | if (token) | |
1180 | { | |
1181 | //token->Print(); | |
1182 | TObjString* tmpStr = dynamic_cast<TObjString*> (token->First()); | |
1183 | if (tmpStr) | |
1184 | { | |
1185 | Int_t testMode = tmpStr->String().Atoi(); | |
1186 | if (testMode > 0) | |
1187 | { | |
c88ad5db | 1188 | Log("SHUTTLE", Form("Process - Enabling test mode %d", testMode)); |
9827400b | 1189 | SetTestMode((TestMode) testMode); |
1190 | } | |
1191 | } | |
1192 | delete token; | |
1193 | } | |
1194 | } | |
1195 | } | |
c88ad5db | 1196 | |
eba76848 | 1197 | fLogbookEntry->Print("all"); |
57f50b3c | 1198 | |
1199 | // Initialization | |
d477ad88 | 1200 | Bool_t hasError = kFALSE; |
5164a766 | 1201 | |
675f64cd | 1202 | // Set the CDB and Reference folders according to the year and LHC period |
1203 | TString lhcPeriod(GetLHCPeriod()); | |
1204 | if (lhcPeriod.Length() == 0) | |
1205 | { | |
7d4cf768 | 1206 | Log("SHUTTLE","Process - LHCPeriod not found in logbook!"); |
1207 | return 0; | |
675f64cd | 1208 | } |
1209 | ||
1210 | if (fgkMainCDB.Length() == 0) | |
1211 | fgkMainCDB = Form("alien://folder=/alice/data/%d/%s/OCDB?user=alidaq?cacheFold=/tmp/OCDBCache", | |
1212 | GetCurrentYear(), lhcPeriod.Data()); | |
1213 | ||
1214 | if (fgkMainRefStorage.Length() == 0) | |
1215 | fgkMainRefStorage = Form("alien://folder=/alice/data/%d/%s/Reference?user=alidaq?cacheFold=/tmp/OCDBCache", | |
1216 | GetCurrentYear(), lhcPeriod.Data()); | |
1217 | ||
57f50b3c | 1218 | // Loop on detectors in the configuration |
b948db8d | 1219 | TIter iter(fConfig->GetDetectors()); |
2bb7b766 | 1220 | TObjString* aDetector = 0; |
b948db8d | 1221 | |
4508ab8b | 1222 | Bool_t first = kTRUE; |
1223 | ||
be48e3ea | 1224 | while ((aDetector = (TObjString*) iter.Next())) |
1225 | { | |
7bfb2090 | 1226 | fCurrentDetector = aDetector->String(); |
5164a766 | 1227 | |
9e080f92 | 1228 | if (ContinueProcessing() == kFALSE) continue; |
4508ab8b | 1229 | |
1230 | if (first) | |
1231 | { | |
1232 | // only read QueryCDB when needed and only once | |
1233 | AliCDBStorage *mainCDBSto = AliCDBManager::Instance()->GetStorage(fgkMainCDB); | |
1234 | if(mainCDBSto) mainCDBSto->QueryCDB(GetCurrentRun()); | |
1235 | AliCDBStorage *mainRefSto = AliCDBManager::Instance()->GetStorage(fgkMainRefStorage); | |
1236 | if(mainRefSto) mainRefSto->QueryCDB(GetCurrentRun()); | |
1237 | first = kFALSE; | |
1238 | } | |
9e080f92 | 1239 | |
c88ad5db | 1240 | Log("SHUTTLE", Form("\t\t\t****** run %d - %s: START ******", |
2bb7b766 | 1241 | GetCurrentRun(), aDetector->GetName())); |
1242 | ||
9d733021 | 1243 | for(Int_t iSys=0;iSys<3;iSys++) fFXSCalled[iSys]=kFALSE; |
1244 | ||
c88ad5db | 1245 | Log(fCurrentDetector.Data(), "Process - Starting processing"); |
85a80aa9 | 1246 | |
be48e3ea | 1247 | Int_t pid = fork(); |
1248 | ||
1249 | if (pid < 0) | |
1250 | { | |
c88ad5db | 1251 | Log("SHUTTLE", "Process - ERROR: Forking failed"); |
be48e3ea | 1252 | } |
1253 | else if (pid > 0) | |
1254 | { | |
1255 | // parent | |
c88ad5db | 1256 | Log("SHUTTLE", Form("Process - In parent process of %d - %s: Starting monitoring", |
be48e3ea | 1257 | GetCurrentRun(), aDetector->GetName())); |
1258 | ||
1259 | Long_t begin = time(0); | |
1260 | ||
1261 | int status; // to be used with waitpid, on purpose an int (not Int_t)! | |
1262 | while (waitpid(pid, &status, WNOHANG) == 0) | |
1263 | { | |
1264 | Long_t expiredTime = time(0) - begin; | |
1265 | ||
1266 | if (expiredTime > fConfig->GetPPTimeOut()) | |
1267 | { | |
9827400b | 1268 | TString tmp; |
c88ad5db | 1269 | tmp.Form("Process - Process of %s time out. " |
1270 | "Run time: %d seconds. Killing...", | |
1271 | fCurrentDetector.Data(), expiredTime); | |
9827400b | 1272 | Log("SHUTTLE", tmp); |
1273 | Log(fCurrentDetector, tmp); | |
be48e3ea | 1274 | |
1275 | kill(pid, 9); | |
1276 | ||
3301427a | 1277 | UpdateShuttleStatus(AliShuttleStatus::kPPTimeOut); |
be48e3ea | 1278 | hasError = kTRUE; |
1279 | ||
1280 | gSystem->Sleep(1000); | |
1281 | } | |
1282 | else | |
1283 | { | |
be48e3ea | 1284 | gSystem->Sleep(1000); |
9827400b | 1285 | |
1286 | TString checkStr; | |
1287 | checkStr.Form("ps -o vsize --pid %d | tail -n 1", pid); | |
1288 | FILE* pipe = gSystem->OpenPipe(checkStr, "r"); | |
1289 | if (!pipe) | |
1290 | { | |
c88ad5db | 1291 | Log("SHUTTLE", Form("Process - Error: " |
1292 | "Could not open pipe to %s", checkStr.Data())); | |
9827400b | 1293 | continue; |
1294 | } | |
1295 | ||
1296 | char buffer[100]; | |
1297 | if (!fgets(buffer, 100, pipe)) | |
1298 | { | |
c88ad5db | 1299 | Log("SHUTTLE", "Process - Error: ps did not return anything"); |
9827400b | 1300 | gSystem->ClosePipe(pipe); |
1301 | continue; | |
1302 | } | |
1303 | gSystem->ClosePipe(pipe); | |
1304 | ||
1305 | //Log("SHUTTLE", Form("ps returned %s", buffer)); | |
1306 | ||
1307 | Int_t mem = 0; | |
1308 | if ((sscanf(buffer, "%d\n", &mem) != 1) || !mem) | |
1309 | { | |
c88ad5db | 1310 | Log("SHUTTLE", "Process - Error: Could not parse output of ps"); |
9827400b | 1311 | continue; |
1312 | } | |
1313 | ||
1314 | if (expiredTime % 60 == 0) | |
ee6f7523 | 1315 | { |
c88ad5db | 1316 | Log("SHUTTLE", Form("Process - %s: Checking process. " |
1317 | "Run time: %d seconds - Memory consumption: %d KB", | |
1318 | fCurrentDetector.Data(), expiredTime, mem)); | |
ee6f7523 | 1319 | SendAlive(); |
1320 | } | |
9827400b | 1321 | |
1322 | if (mem > fConfig->GetPPMaxMem()) | |
1323 | { | |
1324 | TString tmp; | |
c88ad5db | 1325 | tmp.Form("Process - Process exceeds maximum allowed memory " |
1326 | "(%d KB > %d KB). Killing...", | |
9827400b | 1327 | mem, fConfig->GetPPMaxMem()); |
1328 | Log("SHUTTLE", tmp); | |
1329 | Log(fCurrentDetector, tmp); | |
1330 | ||
1331 | kill(pid, 9); | |
1332 | ||
1333 | UpdateShuttleStatus(AliShuttleStatus::kPPOutOfMemory); | |
1334 | hasError = kTRUE; | |
1335 | ||
1336 | gSystem->Sleep(1000); | |
1337 | } | |
be48e3ea | 1338 | } |
1339 | } | |
1340 | ||
c88ad5db | 1341 | Log("SHUTTLE", Form("Process - In parent process of %d - %s: Client has terminated.", |
be48e3ea | 1342 | GetCurrentRun(), aDetector->GetName())); |
1343 | ||
1344 | if (WIFEXITED(status)) | |
1345 | { | |
1346 | Int_t returnCode = WEXITSTATUS(status); | |
1347 | ||
c88ad5db | 1348 | Log("SHUTTLE", Form("Process - %s: the return code is %d", fCurrentDetector.Data(), |
3301427a | 1349 | returnCode)); |
be48e3ea | 1350 | |
9827400b | 1351 | if (returnCode == 0) hasError = kTRUE; |
be48e3ea | 1352 | } |
1353 | } | |
1354 | else if (pid == 0) | |
1355 | { | |
1356 | // client | |
c88ad5db | 1357 | Log("SHUTTLE", Form("Process - In client process of %d - %s", GetCurrentRun(), |
1358 | aDetector->GetName())); | |
be48e3ea | 1359 | |
c88ad5db | 1360 | Log("SHUTTLE", Form("Process - Redirecting output to %s log",fCurrentDetector.Data())); |
ffa29e93 | 1361 | |
546242fb | 1362 | if ((freopen(GetLogFileName(fCurrentDetector), "a", stdout)) == 0) |
ffa29e93 | 1363 | { |
c88ad5db | 1364 | Log("SHUTTLE", "Process - Could not freopen stdout"); |
ffa29e93 | 1365 | } |
1366 | else | |
1367 | { | |
1368 | fOutputRedirected = kTRUE; | |
1369 | if ((dup2(fileno(stdout), fileno(stderr))) < 0) | |
c88ad5db | 1370 | Log("SHUTTLE", "Process - Could not redirect stderr"); |
ffa29e93 | 1371 | |
1372 | } | |
1373 | ||
5bac2bde | 1374 | TString wd = gSystem->WorkingDirectory(); |
675f64cd | 1375 | TString tmpDir = Form("%s/%s_%d_process", GetShuttleTempDir(), |
1376 | fCurrentDetector.Data(), GetCurrentRun()); | |
5bac2bde | 1377 | |
d524ade6 | 1378 | Int_t result = gSystem->GetPathInfo(tmpDir.Data(), 0, (Long64_t*) 0, 0, 0); |
1379 | if (!result) // temp dir already exists! | |
1380 | { | |
1381 | Log(fCurrentDetector.Data(), | |
1382 | Form("Process - %s dir already exists! Removing...", tmpDir.Data())); | |
1383 | gSystem->Exec(Form("rm -rf %s",tmpDir.Data())); | |
675f64cd | 1384 | } |
1385 | ||
1386 | if (gSystem->mkdir(tmpDir.Data(), 1)) | |
1387 | { | |
1388 | Log(fCurrentDetector.Data(), "Process - could not make temp directory!!"); | |
1389 | gSystem->Exit(1); | |
d524ade6 | 1390 | } |
1391 | ||
1392 | if (!gSystem->ChangeDirectory(tmpDir.Data())) | |
1393 | { | |
1394 | Log(fCurrentDetector.Data(), "Process - could not change directory!!"); | |
1395 | gSystem->Exit(1); | |
1396 | } | |
5bac2bde | 1397 | |
9827400b | 1398 | Bool_t success = ProcessCurrentDetector(); |
5bac2bde | 1399 | |
1400 | gSystem->ChangeDirectory(wd.Data()); | |
b1d18693 | 1401 | |
9827400b | 1402 | if (success) // Preprocessor finished successfully! |
1403 | { | |
fb40ead4 | 1404 | // remove temporary folder or DCS map |
1405 | if (!fConfig->KeepTempFolder()) | |
1406 | { | |
1407 | gSystem->Exec(Form("rm -rf %s",tmpDir.Data())); | |
1408 | } else if (!fConfig->KeepDCSMap()) | |
1409 | { | |
1410 | gSystem->Exec(Form("rm -f %s/DCSMap.root",tmpDir.Data())); | |
1411 | } | |
b1d18693 | 1412 | |
3301427a | 1413 | // Update time_processed field in FXS DB |
1414 | if (UpdateTable() == kFALSE) | |
5bac2bde | 1415 | Log("SHUTTLE", Form("Process - %s: Could not update FXS databases!", |
1416 | fCurrentDetector.Data())); | |
3301427a | 1417 | |
1418 | // Transfer the data from local storage to main storage (Grid) | |
1419 | UpdateShuttleStatus(AliShuttleStatus::kStoreStarted); | |
1420 | if (StoreOCDB() == kFALSE) | |
1421 | { | |
c88ad5db | 1422 | Log("SHUTTLE", |
1423 | Form("\t\t\t****** run %d - %s: STORAGE ERROR ******", | |
3301427a | 1424 | GetCurrentRun(), aDetector->GetName())); |
1425 | UpdateShuttleStatus(AliShuttleStatus::kStoreError); | |
9827400b | 1426 | success = kFALSE; |
3301427a | 1427 | } else { |
c88ad5db | 1428 | Log("SHUTTLE", |
1429 | Form("\t\t\t****** run %d - %s: DONE ******", | |
3301427a | 1430 | GetCurrentRun(), aDetector->GetName())); |
1431 | UpdateShuttleStatus(AliShuttleStatus::kDone); | |
9827400b | 1432 | UpdateShuttleLogbook(fCurrentDetector, "DONE"); |
3301427a | 1433 | } |
c88ad5db | 1434 | } else |
1435 | { | |
1436 | Log("SHUTTLE", | |
1437 | Form("\t\t\t****** run %d - %s: PP ERROR ******", | |
1438 | GetCurrentRun(), aDetector->GetName())); | |
be48e3ea | 1439 | } |
1440 | ||
4b95672b | 1441 | for (UInt_t iSys=0; iSys<3; iSys++) |
1442 | { | |
1443 | if (fFXSCalled[iSys]) fFXSlist[iSys].Clear(); | |
1444 | } | |
1445 | ||
c88ad5db | 1446 | Log("SHUTTLE", Form("Process - Client process of %d - %s is exiting now with %d.", |
9827400b | 1447 | GetCurrentRun(), aDetector->GetName(), success)); |
be48e3ea | 1448 | |
1449 | // the client exits here | |
9827400b | 1450 | gSystem->Exit(success); |
be48e3ea | 1451 | |
1452 | AliError("We should never get here!!!"); | |
1453 | } | |
7bfb2090 | 1454 | } |
5164a766 | 1455 | |
c88ad5db | 1456 | Log("SHUTTLE", Form("\t\t\t^*^*^*^*^*^*^*^*^*^*^*^* run %d: FINISH ^*^*^*^*^*^*^*^*^*^*^*^*", |
2bb7b766 | 1457 | GetCurrentRun())); |
1458 | ||
1459 | //check if shuttle is done for this run, if so update logbook | |
1460 | TObjArray checkEntryArray; | |
1461 | checkEntryArray.SetOwner(1); | |
9e080f92 | 1462 | TString whereClause = Form("where run=%d", GetCurrentRun()); |
b0e53b15 | 1463 | if (!QueryShuttleLogbook(whereClause.Data(), checkEntryArray) || |
1464 | checkEntryArray.GetEntries() == 0) { | |
9e080f92 | 1465 | Log("SHUTTLE", Form("Process - Warning: Cannot check status of run %d on Shuttle logbook!", |
1466 | GetCurrentRun())); | |
1467 | return hasError == kFALSE; | |
1468 | } | |
b948db8d | 1469 | |
9e080f92 | 1470 | AliShuttleLogbookEntry* checkEntry = dynamic_cast<AliShuttleLogbookEntry*> |
1471 | (checkEntryArray.At(0)); | |
2bb7b766 | 1472 | |
9e080f92 | 1473 | if (checkEntry) |
1474 | { | |
1475 | if (checkEntry->IsDone()) | |
be48e3ea | 1476 | { |
9e080f92 | 1477 | Log("SHUTTLE","Process - Shuttle is DONE. Updating logbook"); |
1478 | UpdateShuttleLogbook("shuttle_done"); | |
1479 | } | |
1480 | else | |
1481 | { | |
1482 | for (UInt_t iDet=0; iDet<NDetectors(); iDet++) | |
be48e3ea | 1483 | { |
9e080f92 | 1484 | if (checkEntry->GetDetectorStatus(iDet) == AliShuttleLogbookEntry::kUnprocessed) |
be48e3ea | 1485 | { |
9e080f92 | 1486 | AliDebug(2, Form("Run %d: setting %s as \"not first time unprocessed\"", |
1487 | checkEntry->GetRun(), GetDetName(iDet))); | |
1488 | fFirstUnprocessed[iDet] = kFALSE; | |
be48e3ea | 1489 | } |
1490 | } | |
2bb7b766 | 1491 | } |
1492 | } | |
1493 | ||
1494 | fLogbookEntry = 0; | |
85a80aa9 | 1495 | |
a7160fe9 | 1496 | return hasError == kFALSE; |
73abe331 | 1497 | } |
1498 | ||
b948db8d | 1499 | //______________________________________________________________________________________________ |
9827400b | 1500 | Bool_t AliShuttle::ProcessCurrentDetector() |
73abe331 | 1501 | { |
1502 | // | |
2bb7b766 | 1503 | // Makes data retrieval just for a specific detector (fCurrentDetector). |
73abe331 | 1504 | // Threre should be a configuration for this detector. |
73abe331 | 1505 | |
1d172743 | 1506 | Log("SHUTTLE", Form("ProcessCurrentDetector - Retrieving values for %s, run %d", |
1507 | fCurrentDetector.Data(), GetCurrentRun())); | |
73abe331 | 1508 | |
d524ade6 | 1509 | TString wd = gSystem->WorkingDirectory(); |
1510 | ||
2d9019b4 | 1511 | if (!CleanReferenceStorage(fCurrentDetector.Data())) |
546242fb | 1512 | return kFALSE; |
d524ade6 | 1513 | |
1514 | gSystem->ChangeDirectory(wd.Data()); | |
1515 | ||
1d172743 | 1516 | TMap* dcsMap = new TMap(); |
3301427a | 1517 | |
1518 | // call preprocessor | |
1519 | AliPreprocessor* aPreprocessor = | |
1520 | dynamic_cast<AliPreprocessor*> (fPreprocessorMap.GetValue(fCurrentDetector)); | |
1521 | ||
1522 | aPreprocessor->Initialize(GetCurrentRun(), GetCurrentStartTime(), GetCurrentEndTime()); | |
1523 | ||
1524 | Bool_t processDCS = aPreprocessor->ProcessDCS(); | |
d477ad88 | 1525 | |
651fdaab | 1526 | if (!processDCS) |
1527 | { | |
1d172743 | 1528 | Log(fCurrentDetector, "ProcessCurrentDetector -" |
1529 | " The preprocessor requested to skip the retrieval of DCS values"); | |
651fdaab | 1530 | } |
8b739301 | 1531 | else if (fTestMode & kSkipDCS) |
2c15234c | 1532 | { |
1d172743 | 1533 | Log(fCurrentDetector, "ProcessCurrentDetector - In TESTMODE: Skipping DCS processing"); |
9827400b | 1534 | } |
1535 | else if (fTestMode & kErrorDCS) | |
1536 | { | |
1d172743 | 1537 | Log(fCurrentDetector, "ProcessCurrentDetector - In TESTMODE: Simulating DCS error"); |
3d8bc902 | 1538 | UpdateShuttleStatus(AliShuttleStatus::kDCSStarted); |
9827400b | 1539 | UpdateShuttleStatus(AliShuttleStatus::kDCSError); |
1d172743 | 1540 | delete dcsMap; |
9827400b | 1541 | return kFALSE; |
2c15234c | 1542 | } else { |
3301427a | 1543 | |
1544 | UpdateShuttleStatus(AliShuttleStatus::kDCSStarted); | |
1545 | ||
1d172743 | 1546 | // Query DCS archive |
1547 | Int_t nServers = fConfig->GetNServers(fCurrentDetector); | |
a038aa70 | 1548 | |
1d172743 | 1549 | for (int iServ=0; iServ<nServers; iServ++) |
2c15234c | 1550 | { |
1d172743 | 1551 | |
1552 | TString host(fConfig->GetDCSHost(fCurrentDetector, iServ)); | |
1553 | Int_t port = fConfig->GetDCSPort(fCurrentDetector, iServ); | |
542b6cc8 | 1554 | Int_t multiSplit = fConfig->GetMultiSplit(fCurrentDetector, iServ); |
1555 | ||
1790d4b7 | 1556 | Log(fCurrentDetector, Form("ProcessCurrentDetector -" |
1557 | " Querying DCS Amanda server %s:%d (%d of %d)", | |
1558 | host.Data(), port, iServ+1, nServers)); | |
1d172743 | 1559 | |
1560 | TMap* aliasMap = 0; | |
1561 | TMap* dpMap = 0; | |
1562 | ||
1563 | if (fConfig->GetDCSAliases(fCurrentDetector, iServ)->GetEntries() > 0) | |
2c15234c | 1564 | { |
1d172743 | 1565 | aliasMap = GetValueSet(host, port, |
542b6cc8 | 1566 | fConfig->GetDCSAliases(fCurrentDetector, iServ), |
1567 | kAlias, multiSplit); | |
1d172743 | 1568 | if (!aliasMap) |
1569 | { | |
1570 | Log(fCurrentDetector, | |
1571 | Form("ProcessCurrentDetector -" | |
675f64cd | 1572 | " Error retrieving DCS aliases from server %s." |
1573 | " Sending mail to DCS experts!", host.Data())); | |
1d172743 | 1574 | UpdateShuttleStatus(AliShuttleStatus::kDCSError); |
675f64cd | 1575 | |
fb40ead4 | 1576 | if (!SendMailToDCS()) |
1577 | Log("SHUTTLE", Form("ProcessCurrentDetector - " | |
1578 | "Could not send mail to DCS experts!")); | |
675f64cd | 1579 | |
a038aa70 | 1580 | delete dcsMap; |
1d172743 | 1581 | return kFALSE; |
1582 | } | |
2c15234c | 1583 | } |
a038aa70 | 1584 | |
1d172743 | 1585 | if (fConfig->GetDCSDataPoints(fCurrentDetector, iServ)->GetEntries() > 0) |
a038aa70 | 1586 | { |
1d172743 | 1587 | dpMap = GetValueSet(host, port, |
542b6cc8 | 1588 | fConfig->GetDCSDataPoints(fCurrentDetector, iServ), |
1589 | kDP, multiSplit); | |
1d172743 | 1590 | if (!dpMap) |
1591 | { | |
1592 | Log(fCurrentDetector, | |
1593 | Form("ProcessCurrentDetector -" | |
675f64cd | 1594 | " Error retrieving DCS data points from server %s." |
1595 | " Sending mail to DCS experts!", host.Data())); | |
1d172743 | 1596 | UpdateShuttleStatus(AliShuttleStatus::kDCSError); |
675f64cd | 1597 | |
fb40ead4 | 1598 | if (!SendMailToDCS()) |
1599 | Log("SHUTTLE", Form("ProcessCurrentDetector - " | |
1600 | "Could not send mail to DCS experts!")); | |
675f64cd | 1601 | |
1d172743 | 1602 | if (aliasMap) delete aliasMap; |
1603 | delete dcsMap; | |
1604 | return kFALSE; | |
1605 | } | |
a038aa70 | 1606 | } |
1d172743 | 1607 | |
1608 | // merge aliasMap and dpMap into dcsMap | |
1609 | if(aliasMap) { | |
1610 | TIter iter(aliasMap); | |
a038aa70 | 1611 | TObjString* key = 0; |
1612 | while ((key = (TObjString*) iter.Next())) | |
1d172743 | 1613 | dcsMap->Add(key, aliasMap->GetValue(key->String())); |
1614 | ||
1615 | aliasMap->SetOwner(kFALSE); | |
1616 | delete aliasMap; | |
1617 | } | |
1618 | ||
1619 | if(dpMap) { | |
1620 | TIter iter(dpMap); | |
1621 | TObjString* key = 0; | |
1622 | while ((key = (TObjString*) iter.Next())) | |
1623 | dcsMap->Add(key, dpMap->GetValue(key->String())); | |
1624 | ||
1625 | dpMap->SetOwner(kFALSE); | |
1626 | delete dpMap; | |
a038aa70 | 1627 | } |
73abe331 | 1628 | } |
1629 | } | |
dc25836b | 1630 | |
b1d18693 | 1631 | // save map into file, to help debugging in case of preprocessor error |
fbc112e3 | 1632 | TFile* f = TFile::Open("DCSMap.root","recreate"); |
b1d18693 | 1633 | f->cd(); |
1634 | dcsMap->Write("DCSMap", TObject::kSingleKey); | |
1635 | f->Close(); | |
fbc112e3 | 1636 | delete f; |
b1d18693 | 1637 | |
2bb7b766 | 1638 | // DCS Archive DB processing successful. Call Preprocessor! |
85a80aa9 | 1639 | UpdateShuttleStatus(AliShuttleStatus::kPPStarted); |
a7160fe9 | 1640 | |
a038aa70 | 1641 | UInt_t returnValue = aPreprocessor->Process(dcsMap); |
b948db8d | 1642 | |
3301427a | 1643 | if (returnValue > 0) // Preprocessor error! |
1644 | { | |
c88ad5db | 1645 | Log(fCurrentDetector, Form("ProcessCurrentDetector - " |
1646 | "Preprocessor failed. Process returned %d.", returnValue)); | |
cb343cfd | 1647 | UpdateShuttleStatus(AliShuttleStatus::kPPError); |
a038aa70 | 1648 | dcsMap->DeleteAll(); |
1649 | delete dcsMap; | |
9827400b | 1650 | return kFALSE; |
1651 | } | |
1652 | ||
1653 | // preprocessor ok! | |
1654 | UpdateShuttleStatus(AliShuttleStatus::kPPDone); | |
1655 | Log(fCurrentDetector, Form("ProcessCurrentDetector - %s preprocessor returned success", | |
1656 | fCurrentDetector.Data())); | |
b948db8d | 1657 | |
a038aa70 | 1658 | dcsMap->DeleteAll(); |
1659 | delete dcsMap; | |
b948db8d | 1660 | |
9827400b | 1661 | return kTRUE; |
2bb7b766 | 1662 | } |
1663 | ||
fbc112e3 | 1664 | //______________________________________________________________________________________________ |
1665 | void AliShuttle::CountOpenRuns() | |
1666 | { | |
1667 | // Query DAQ's Shuttle logbook and sends the number of open runs to ML | |
1668 | ||
1669 | // check connection, in case connect | |
1670 | if (!Connect(3)) | |
1671 | return; | |
1672 | ||
1673 | TString sqlQuery; | |
1674 | sqlQuery = Form("select count(*) from %s where shuttle_done=0", fConfig->GetShuttlelbTable()); | |
1675 | ||
1676 | TSQLResult* aResult = fServer[3]->Query(sqlQuery); | |
1677 | if (!aResult) { | |
1678 | AliError(Form("Can't execute query <%s>!", sqlQuery.Data())); | |
1679 | return; | |
1680 | } | |
1681 | ||
1682 | AliDebug(2,Form("Query = %s", sqlQuery.Data())); | |
1683 | ||
1684 | if (aResult->GetRowCount() == 0) { | |
1685 | AliError(Form("No result for query %s received", sqlQuery.Data())); | |
1686 | return; | |
1687 | } | |
1688 | ||
1689 | if (aResult->GetFieldCount() != 1) { | |
1690 | AliError(Form("Invalid field count for query %s received", sqlQuery.Data())); | |
1691 | return; | |
1692 | } | |
1693 | ||
1694 | TSQLRow* aRow = aResult->Next(); | |
1695 | if (!aRow) { | |
1696 | AliError(Form("Could not receive result of query %s", sqlQuery.Data())); | |
1697 | return; | |
1698 | } | |
1699 | ||
1700 | TString result(aRow->GetField(0), aRow->GetFieldLength(0)); | |
1701 | Int_t count = result.Atoi(); | |
1702 | ||
1703 | Log("SHUTTLE", Form("%d unprocessed runs", count)); | |
1704 | ||
1705 | delete aRow; | |
1706 | delete aResult; | |
1707 | ||
1708 | TMonaLisaValue mlStatus("SHUTTLE_openruns", count); | |
1709 | ||
1710 | TList mlList; | |
1711 | mlList.Add(&mlStatus); | |
1712 | ||
1713 | fMonaLisa->SendParameters(&mlList, "__PROCESSINGINFO__"); | |
1714 | } | |
1715 | ||
2bb7b766 | 1716 | //______________________________________________________________________________________________ |
1717 | Bool_t AliShuttle::QueryShuttleLogbook(const char* whereClause, | |
1718 | TObjArray& entries) | |
1719 | { | |
9827400b | 1720 | // Query DAQ's Shuttle logbook and fills detector status object. |
1721 | // Call QueryRunParameters to query DAQ logbook for run parameters. | |
1722 | // | |
2bb7b766 | 1723 | |
fc5a4708 | 1724 | entries.SetOwner(1); |
1725 | ||
2bb7b766 | 1726 | // check connection, in case connect |
fbc112e3 | 1727 | if (!Connect(3)) return kFALSE; |
2bb7b766 | 1728 | |
1729 | TString sqlQuery; | |
441b0e9c | 1730 | sqlQuery = Form("select * from %s %s order by run", fConfig->GetShuttlelbTable(), whereClause); |
2bb7b766 | 1731 | |
be48e3ea | 1732 | TSQLResult* aResult = fServer[3]->Query(sqlQuery); |
2bb7b766 | 1733 | if (!aResult) { |
1734 | AliError(Form("Can't execute query <%s>!", sqlQuery.Data())); | |
1735 | return kFALSE; | |
1736 | } | |
1737 | ||
fc5a4708 | 1738 | AliDebug(2,Form("Query = %s", sqlQuery.Data())); |
1739 | ||
2bb7b766 | 1740 | if(aResult->GetRowCount() == 0) { |
c88ad5db | 1741 | Log("SHUTTLE", "No entries in Shuttle Logbook match request"); |
9827400b | 1742 | delete aResult; |
1743 | return kTRUE; | |
2bb7b766 | 1744 | } |
1745 | ||
1746 | // TODO Check field count! | |
db99d43e | 1747 | const UInt_t nCols = 23; |
2bb7b766 | 1748 | if (aResult->GetFieldCount() != (Int_t) nCols) { |
c88ad5db | 1749 | Log("SHUTTLE", "Invalid SQL result field number!"); |
2bb7b766 | 1750 | delete aResult; |
1751 | return kFALSE; | |
1752 | } | |
1753 | ||
2bb7b766 | 1754 | TSQLRow* aRow; |
1755 | while ((aRow = aResult->Next())) { | |
1756 | TString runString(aRow->GetField(0), aRow->GetFieldLength(0)); | |
1757 | Int_t run = runString.Atoi(); | |
1758 | ||
eba76848 | 1759 | AliShuttleLogbookEntry *entry = QueryRunParameters(run); |
1760 | if (!entry) | |
1761 | continue; | |
2bb7b766 | 1762 | |
1763 | // loop on detectors | |
eba76848 | 1764 | for(UInt_t ii = 0; ii < nCols; ii++) |
1765 | entry->SetDetectorStatus(aResult->GetFieldName(ii), aRow->GetField(ii)); | |
2bb7b766 | 1766 | |
eba76848 | 1767 | entries.AddLast(entry); |
2bb7b766 | 1768 | delete aRow; |
1769 | } | |
1770 | ||
2bb7b766 | 1771 | delete aResult; |
1772 | return kTRUE; | |
1773 | } | |
1774 | ||
1775 | //______________________________________________________________________________________________ | |
eba76848 | 1776 | AliShuttleLogbookEntry* AliShuttle::QueryRunParameters(Int_t run) |
2bb7b766 | 1777 | { |
eba76848 | 1778 | // |
1779 | // Retrieve run parameters written in the DAQ logbook and sets them into AliShuttleLogbookEntry object | |
1780 | // | |
2bb7b766 | 1781 | |
1782 | // check connection, in case connect | |
be48e3ea | 1783 | if (!Connect(3)) |
eba76848 | 1784 | return 0; |
2bb7b766 | 1785 | |
1786 | TString sqlQuery; | |
2c15234c | 1787 | sqlQuery.Form("select * from %s where run=%d", fConfig->GetDAQlbTable(), run); |
2bb7b766 | 1788 | |
be48e3ea | 1789 | TSQLResult* aResult = fServer[3]->Query(sqlQuery); |
2bb7b766 | 1790 | if (!aResult) { |
c88ad5db | 1791 | Log("SHUTTLE", Form("Can't execute query <%s>!", sqlQuery.Data())); |
eba76848 | 1792 | return 0; |
2bb7b766 | 1793 | } |
1794 | ||
eba76848 | 1795 | if (aResult->GetRowCount() == 0) { |
2bb7b766 | 1796 | Log("SHUTTLE", Form("QueryRunParameters - No entry in DAQ Logbook for run %d. Skipping", run)); |
1797 | delete aResult; | |
eba76848 | 1798 | return 0; |
2bb7b766 | 1799 | } |
1800 | ||
eba76848 | 1801 | if (aResult->GetRowCount() > 1) { |
c88ad5db | 1802 | Log("SHUTTLE", Form("QueryRunParameters - UNEXPECTED: " |
1803 | "more than one entry in DAQ Logbook for run %d!", run)); | |
2bb7b766 | 1804 | delete aResult; |
eba76848 | 1805 | return 0; |
2bb7b766 | 1806 | } |
1807 | ||
eba76848 | 1808 | TSQLRow* aRow = aResult->Next(); |
1809 | if (!aRow) | |
1810 | { | |
c88ad5db | 1811 | Log("SHUTTLE", Form("QueryRunParameters - Could not retrieve row for run %d. Skipping", run)); |
eba76848 | 1812 | delete aResult; |
1813 | return 0; | |
1814 | } | |
2bb7b766 | 1815 | |
eba76848 | 1816 | AliShuttleLogbookEntry* entry = new AliShuttleLogbookEntry(run); |
2bb7b766 | 1817 | |
eba76848 | 1818 | for (Int_t ii = 0; ii < aResult->GetFieldCount(); ii++) |
1819 | entry->SetRunParameter(aResult->GetFieldName(ii), aRow->GetField(ii)); | |
2bb7b766 | 1820 | |
eba76848 | 1821 | UInt_t startTime = entry->GetStartTime(); |
1822 | UInt_t endTime = entry->GetEndTime(); | |
1823 | ||
1abfbb60 | 1824 | // if (!startTime || !endTime || startTime > endTime) |
1825 | // { | |
1826 | // Log("SHUTTLE", | |
1827 | // Form("QueryRunParameters - Invalid parameters for Run %d: startTime = %d, endTime = %d. Skipping!", | |
1828 | // run, startTime, endTime)); | |
1829 | // | |
1830 | // Log("SHUTTLE", Form("Marking SHUTTLE done for run %d", run)); | |
1831 | // fLogbookEntry = entry; | |
1832 | // if (!UpdateShuttleLogbook("shuttle_done")) | |
1833 | // { | |
1834 | // AliError(Form("Could not update logbook for run %d !", run)); | |
1835 | // } | |
1836 | // fLogbookEntry = 0; | |
1837 | // | |
1838 | // delete entry; | |
1839 | // delete aRow; | |
1840 | // delete aResult; | |
1841 | // return 0; | |
1842 | // } | |
1843 | ||
1844 | if (!startTime) | |
b0e53b15 | 1845 | { |
1846 | Log("SHUTTLE", | |
1abfbb60 | 1847 | Form("QueryRunParameters - Invalid parameters for Run %d: " |
1848 | "startTime = %d, endTime = %d. Skipping!", | |
1849 | run, startTime, endTime)); | |
b0e53b15 | 1850 | |
1851 | Log("SHUTTLE", Form("Marking SHUTTLE done for run %d", run)); | |
1852 | fLogbookEntry = entry; | |
ee6f7523 | 1853 | if (!UpdateShuttleLogbook("shuttle_ignored")) |
b0e53b15 | 1854 | { |
1855 | AliError(Form("Could not update logbook for run %d !", run)); | |
1856 | } | |
1857 | fLogbookEntry = 0; | |
1858 | ||
1859 | delete entry; | |
1860 | delete aRow; | |
1861 | delete aResult; | |
1862 | return 0; | |
1863 | } | |
1864 | ||
1abfbb60 | 1865 | if (startTime && !endTime) |
1866 | { | |
1867 | // TODO Here we don't mark SHUTTLE done, because this may mean | |
1868 | //the run is still ongoing!! | |
1869 | Log("SHUTTLE", | |
1870 | Form("QueryRunParameters - Invalid parameters for Run %d: " | |
1871 | "startTime = %d, endTime = %d. Skipping (Shuttle won't be marked as DONE)!", | |
1872 | run, startTime, endTime)); | |
1873 | ||
1874 | //Log("SHUTTLE", Form("Marking SHUTTLE done for run %d", run)); | |
1875 | //fLogbookEntry = entry; | |
1876 | //if (!UpdateShuttleLogbook("shuttle_done")) | |
1877 | //{ | |
1878 | // AliError(Form("Could not update logbook for run %d !", run)); | |
1879 | //} | |
1880 | //fLogbookEntry = 0; | |
1881 | ||
1882 | delete entry; | |
1883 | delete aRow; | |
1884 | delete aResult; | |
1885 | return 0; | |
1886 | } | |
1887 | ||
1888 | if (startTime && endTime && (startTime > endTime)) | |
1889 | { | |
1890 | Log("SHUTTLE", | |
1891 | Form("QueryRunParameters - Invalid parameters for Run %d: " | |
1892 | "startTime = %d, endTime = %d. Skipping!", | |
1893 | run, startTime, endTime)); | |
1894 | ||
1895 | Log("SHUTTLE", Form("Marking SHUTTLE done for run %d", run)); | |
1896 | fLogbookEntry = entry; | |
ee6f7523 | 1897 | if (!UpdateShuttleLogbook("shuttle_ignored")) |
1abfbb60 | 1898 | { |
1899 | AliError(Form("Could not update logbook for run %d !", run)); | |
1900 | } | |
1901 | fLogbookEntry = 0; | |
1902 | ||
1903 | delete entry; | |
1904 | delete aRow; | |
1905 | delete aResult; | |
1906 | return 0; | |
1907 | } | |
1908 | ||
b0e53b15 | 1909 | TString totEventsStr = entry->GetRunParameter("totalEvents"); |
1910 | Int_t totEvents = totEventsStr.Atoi(); | |
1911 | if (totEvents < 1) | |
1912 | { | |
eba76848 | 1913 | Log("SHUTTLE", |
b0e53b15 | 1914 | Form("QueryRunParameters - Run %d has 0 events - Skipping!", run)); |
1915 | ||
1916 | Log("SHUTTLE", Form("Marking SHUTTLE done for run %d", run)); | |
1917 | fLogbookEntry = entry; | |
97e3c167 | 1918 | if (!UpdateShuttleLogbook("shuttle_ignored")) |
b0e53b15 | 1919 | { |
1920 | AliError(Form("Could not update logbook for run %d !", run)); | |
1921 | } | |
1922 | fLogbookEntry = 0; | |
1923 | ||
eba76848 | 1924 | delete entry; |
2bb7b766 | 1925 | delete aRow; |
eba76848 | 1926 | delete aResult; |
1927 | return 0; | |
2bb7b766 | 1928 | } |
1929 | ||
eba76848 | 1930 | delete aRow; |
2bb7b766 | 1931 | delete aResult; |
eba76848 | 1932 | |
1933 | return entry; | |
2bb7b766 | 1934 | } |
1935 | ||
a038aa70 | 1936 | //______________________________________________________________________________________________ |
1937 | TMap* AliShuttle::GetValueSet(const char* host, Int_t port, const TSeqCollection* entries, | |
542b6cc8 | 1938 | DCSType type, Int_t multiSplit) |
a038aa70 | 1939 | { |
1940 | // Retrieve all "entry" data points from the DCS server | |
1941 | // host, port: TSocket connection parameters | |
1942 | // entries: list of name of the alias or data point | |
1943 | // type: kAlias or kDP | |
1944 | // returns TMap of values, 0 when failure | |
542b6cc8 | 1945 | |
1946 | AliDCSClient client(host, port, fTimeout, fRetries, multiSplit); | |
b41b252a | 1947 | |
a038aa70 | 1948 | TMap* result = 0; |
b41b252a | 1949 | if (type == kAlias) |
a038aa70 | 1950 | { |
b41b252a | 1951 | result = client.GetAliasValues(entries, GetCurrentStartTime(), |
1952 | GetCurrentEndTime()); | |
1953 | } | |
1954 | else if (type == kDP) | |
1955 | { | |
1956 | result = client.GetDPValues(entries, GetCurrentStartTime(), | |
1957 | GetCurrentEndTime()); | |
1958 | } | |
a038aa70 | 1959 | |
b41b252a | 1960 | if (result == 0) |
1961 | { | |
1962 | Log(fCurrentDetector.Data(), Form("GetValueSet - Can't get entries! Reason: %s", | |
1790d4b7 | 1963 | client.GetErrorString(client.GetResultErrorCode()))); |
1964 | if (client.GetResultErrorCode() == AliDCSClient::fgkServerError) | |
1965 | Log(fCurrentDetector.Data(), Form("GetValueSet - Server error code: %s", | |
1966 | client.GetServerError().Data())); | |
a038aa70 | 1967 | |
b41b252a | 1968 | return 0; |
a038aa70 | 1969 | } |
b41b252a | 1970 | |
a038aa70 | 1971 | return result; |
1972 | } | |
b41b252a | 1973 | |
b948db8d | 1974 | //______________________________________________________________________________________________ |
57f50b3c | 1975 | const char* AliShuttle::GetFile(Int_t system, const char* detector, |
1976 | const char* id, const char* source) | |
b948db8d | 1977 | { |
9827400b | 1978 | // Get calibration file from file exchange servers |
1979 | // First queris the FXS database for the file name, using the run, detector, id and source info | |
1980 | // then calls RetrieveFile(filename) for actual copy to local disk | |
1981 | // run: current run being processed (given by Logbook entry fLogbookEntry) | |
1982 | // detector: the Preprocessor name | |
1983 | // id: provided as a parameter by the Preprocessor | |
1984 | // source: provided by the Preprocessor through GetFileSources function | |
1985 | ||
1986 | // check if test mode should simulate a FXS error | |
1987 | if (fTestMode & kErrorFXSFiles) | |
1988 | { | |
1989 | Log(detector, Form("GetFile - In TESTMODE - Simulating error while connecting to %s FXS", GetSystemName(system))); | |
1990 | return 0; | |
1991 | } | |
1992 | ||
57f50b3c | 1993 | // check connection, in case connect |
9d733021 | 1994 | if (!Connect(system)) |
eba76848 | 1995 | { |
9d733021 | 1996 | Log(detector, Form("GetFile - Couldn't connect to %s FXS database", GetSystemName(system))); |
57f50b3c | 1997 | return 0; |
1998 | } | |
1999 | ||
2000 | // Query preparation | |
9d733021 | 2001 | TString sourceName(source); |
d386d623 | 2002 | Int_t nFields = 3; |
2003 | TString sqlQueryStart = Form("select filePath,size,fileChecksum from %s where", | |
2004 | fConfig->GetFXSdbTable(system)); | |
2005 | TString whereClause = Form("run=%d and detector=\"%s\" and fileId=\"%s\"", | |
2006 | GetCurrentRun(), detector, id); | |
2007 | ||
9d733021 | 2008 | if (system == kDAQ) |
2009 | { | |
d386d623 | 2010 | whereClause += Form(" and DAQsource=\"%s\"", source); |
57f50b3c | 2011 | } |
9d733021 | 2012 | else if (system == kDCS) |
eba76848 | 2013 | { |
9d733021 | 2014 | sourceName="none"; |
57f50b3c | 2015 | } |
9d733021 | 2016 | else if (system == kHLT) |
9e080f92 | 2017 | { |
d386d623 | 2018 | whereClause += Form(" and DDLnumbers=\"%s\"", source); |
9d733021 | 2019 | nFields = 3; |
9e080f92 | 2020 | } |
2021 | ||
9e080f92 | 2022 | TString sqlQuery = Form("%s %s", sqlQueryStart.Data(), whereClause.Data()); |
2023 | ||
2024 | AliDebug(2, Form("SQL query: \n%s",sqlQuery.Data())); | |
2025 | ||
2026 | // Query execution | |
2027 | TSQLResult* aResult = 0; | |
9d733021 | 2028 | aResult = dynamic_cast<TSQLResult*> (fServer[system]->Query(sqlQuery)); |
9e080f92 | 2029 | if (!aResult) { |
9d733021 | 2030 | Log(detector, Form("GetFileName - Can't execute SQL query to %s database for: id = %s, source = %s", |
2031 | GetSystemName(system), id, sourceName.Data())); | |
9e080f92 | 2032 | return 0; |
2033 | } | |
2034 | ||
2035 | if(aResult->GetRowCount() == 0) | |
2036 | { | |
2037 | Log(detector, | |
9d733021 | 2038 | Form("GetFileName - No entry in %s FXS db for: id = %s, source = %s", |
2039 | GetSystemName(system), id, sourceName.Data())); | |
9e080f92 | 2040 | delete aResult; |
2041 | return 0; | |
2042 | } | |
2bb7b766 | 2043 | |
9e080f92 | 2044 | if (aResult->GetRowCount() > 1) { |
2045 | Log(detector, | |
9d733021 | 2046 | Form("GetFileName - More than one entry in %s FXS db for: id = %s, source = %s", |
2047 | GetSystemName(system), id, sourceName.Data())); | |
9e080f92 | 2048 | delete aResult; |
2049 | return 0; | |
2050 | } | |
2051 | ||
9d733021 | 2052 | if (aResult->GetFieldCount() != nFields) { |
9e080f92 | 2053 | Log(detector, |
9d733021 | 2054 | Form("GetFileName - Wrong field count in %s FXS db for: id = %s, source = %s", |
2055 | GetSystemName(system), id, sourceName.Data())); | |
9e080f92 | 2056 | delete aResult; |
2057 | return 0; | |
2058 | } | |
2059 | ||
2060 | TSQLRow* aRow = dynamic_cast<TSQLRow*> (aResult->Next()); | |
2061 | ||
2062 | if (!aRow){ | |
9d733021 | 2063 | Log(detector, Form("GetFileName - Empty set result in %s FXS db from query: id = %s, source = %s", |
2064 | GetSystemName(system), id, sourceName.Data())); | |
9e080f92 | 2065 | delete aResult; |
2066 | return 0; | |
2067 | } | |
2068 | ||
2069 | TString filePath(aRow->GetField(0), aRow->GetFieldLength(0)); | |
2070 | TString fileSize(aRow->GetField(1), aRow->GetFieldLength(1)); | |
d386d623 | 2071 | TString fileChecksum(aRow->GetField(2), aRow->GetFieldLength(2)); |
9e080f92 | 2072 | |
2073 | delete aResult; | |
2074 | delete aRow; | |
2075 | ||
d386d623 | 2076 | AliDebug(2, Form("filePath = %s; size = %s, fileChecksum = %s", |
2077 | filePath.Data(), fileSize.Data(), fileChecksum.Data())); | |
9e080f92 | 2078 | |
9e080f92 | 2079 | // retrieved file is renamed to make it unique |
675f64cd | 2080 | TString localFileName = Form("%s/%s_%d_process/%s_%s_%d_%s_%s.shuttle", |
2081 | GetShuttleTempDir(), detector, GetCurrentRun(), | |
d524ade6 | 2082 | GetSystemName(system), detector, GetCurrentRun(), |
2083 | id, sourceName.Data()); | |
9d733021 | 2084 | |
9e080f92 | 2085 | |
9d733021 | 2086 | // file retrieval from FXS |
4b95672b | 2087 | UInt_t nRetries = 0; |
2088 | UInt_t maxRetries = 3; | |
2089 | Bool_t result = kFALSE; | |
2090 | ||
2091 | // copy!! if successful TSystem::Exec returns 0 | |
2092 | while(nRetries++ < maxRetries) { | |
2093 | AliDebug(2, Form("Trying to copy file. Retry # %d", nRetries)); | |
2094 | result = RetrieveFile(system, filePath.Data(), localFileName.Data()); | |
2095 | if(!result) | |
2096 | { | |
2097 | Log(detector, Form("GetFileName - Copy of file %s from %s FXS failed", | |
9d733021 | 2098 | filePath.Data(), GetSystemName(system))); |
4b95672b | 2099 | continue; |
4f0749a8 | 2100 | } |
9e080f92 | 2101 | |
d386d623 | 2102 | if (fileChecksum.Length()>0) |
4b95672b | 2103 | { |
2104 | // compare md5sum of local file with the one stored in the FXS DB | |
d524ade6 | 2105 | Int_t md5Comp = gSystem->Exec(Form("md5sum %s |grep %s 2>&1 > /dev/null", |
2106 | localFileName.Data(), fileChecksum.Data())); | |
9e080f92 | 2107 | |
4b95672b | 2108 | if (md5Comp != 0) |
2109 | { | |
2110 | Log(detector, Form("GetFileName - md5sum of file %s does not match with local copy!", | |
2111 | filePath.Data())); | |
2112 | result = kFALSE; | |
2113 | continue; | |
2114 | } | |
d386d623 | 2115 | } else { |
2116 | Log(fCurrentDetector, Form("GetFile - md5sum of file %s not set in %s database, skipping comparison", | |
2117 | filePath.Data(), GetSystemName(system))); | |
9d733021 | 2118 | } |
4b95672b | 2119 | if (result) break; |
9e080f92 | 2120 | } |
2121 | ||
4b95672b | 2122 | if(!result) return 0; |
2123 | ||
9d733021 | 2124 | fFXSCalled[system]=kTRUE; |
2125 | TObjString *fileParams = new TObjString(Form("%s#!?!#%s", id, sourceName.Data())); | |
2126 | fFXSlist[system].Add(fileParams); | |
9e080f92 | 2127 | |
675f64cd | 2128 | static TString staticLocalFileName; |
2129 | staticLocalFileName.Form("%s", localFileName.Data()); | |
2130 | ||
c88ad5db | 2131 | Log(fCurrentDetector, Form("GetFile - Retrieved file with id %s and " |
2132 | "source %s from %s to %s", id, source, | |
d524ade6 | 2133 | GetSystemName(system), localFileName.Data())); |
675f64cd | 2134 | |
d524ade6 | 2135 | return staticLocalFileName.Data(); |
2bb7b766 | 2136 | } |
2137 | ||
2138 | //______________________________________________________________________________________________ | |
9d733021 | 2139 | Bool_t AliShuttle::RetrieveFile(UInt_t system, const char* fxsFileName, const char* localFileName) |
9e080f92 | 2140 | { |
9827400b | 2141 | // |
2142 | // Copies file from FXS to local Shuttle machine | |
2143 | // | |
2bb7b766 | 2144 | |
9e080f92 | 2145 | // check temp directory: trying to cd to temp; if it does not exist, create it |
d524ade6 | 2146 | AliDebug(2, Form("Copy file %s from %s FXS into %s", |
2147 | GetSystemName(system), fxsFileName, localFileName)); | |
2148 | ||
2149 | TString tmpDir(localFileName); | |
2150 | ||
2151 | tmpDir = tmpDir(0,tmpDir.Last('/')); | |
9e080f92 | 2152 | |
d524ade6 | 2153 | Int_t noDir = gSystem->GetPathInfo(tmpDir.Data(), 0, (Long64_t*) 0, 0, 0); |
2154 | if (noDir) // temp dir does not exists! | |
2155 | { | |
2156 | if (gSystem->mkdir(tmpDir.Data(), 1)) | |
2157 | { | |
2158 | Log(fCurrentDetector.Data(), "RetrieveFile - could not make temp directory!!"); | |
9e080f92 | 2159 | return kFALSE; |
2160 | } | |
9e080f92 | 2161 | } |
2162 | ||
9d733021 | 2163 | TString baseFXSFolder; |
2164 | if (system == kDAQ) | |
2165 | { | |
2166 | baseFXSFolder = "FES/"; | |
2167 | } | |
2168 | else if (system == kDCS) | |
2169 | { | |
2170 | baseFXSFolder = ""; | |
2171 | } | |
2172 | else if (system == kHLT) | |
2173 | { | |
42fde080 | 2174 | baseFXSFolder = "/opt/FXS/"; |
9d733021 | 2175 | } |
2176 | ||
2177 | ||
d524ade6 | 2178 | TString command = Form("scp -oPort=%d -2 %s@%s:%s%s %s", |
9d733021 | 2179 | fConfig->GetFXSPort(system), |
2180 | fConfig->GetFXSUser(system), | |
2181 | fConfig->GetFXSHost(system), | |
2182 | baseFXSFolder.Data(), | |
2183 | fxsFileName, | |
9e080f92 | 2184 | localFileName); |
2185 | ||
2186 | AliDebug(2, Form("%s",command.Data())); | |
2187 | ||
4b95672b | 2188 | Bool_t result = (gSystem->Exec(command.Data()) == 0); |
9e080f92 | 2189 | |
4b95672b | 2190 | return result; |
9e080f92 | 2191 | } |
2192 | ||
2193 | //______________________________________________________________________________________________ | |
9d733021 | 2194 | TList* AliShuttle::GetFileSources(Int_t system, const char* detector, const char* id) |
2195 | { | |
9827400b | 2196 | // |
2197 | // Get sources producing the condition file Id from file exchange servers | |
4a33bdd9 | 2198 | // if id is NULL all sources are returned (distinct) |
9827400b | 2199 | // |
1bcd28db | 2200 | |
8884d71e | 2201 | if (id) |
2202 | { | |
2203 | Log(detector, Form("GetFileSources - Querying %s FXS for files with id %s produced by %s", GetSystemName(system), id, detector)); | |
2204 | } else { | |
2205 | Log(detector, Form("GetFileSources - Querying %s FXS for files produced by %s", GetSystemName(system), detector)); | |
2206 | } | |
9827400b | 2207 | |
2208 | // check if test mode should simulate a FXS error | |
2209 | if (fTestMode & kErrorFXSSources) | |
2210 | { | |
2211 | Log(detector, Form("GetFileSources - In TESTMODE - Simulating error while connecting to %s FXS", GetSystemName(system))); | |
2212 | return 0; | |
2213 | } | |
2214 | ||
9d733021 | 2215 | if (system == kDCS) |
2216 | { | |
c88ad5db | 2217 | Log(detector, "GetFileSources - WARNING: DCS system has only one source of data!"); |
6297b37d | 2218 | TList *list = new TList(); |
2219 | list->SetOwner(1); | |
2220 | list->Add(new TObjString(" ")); | |
2221 | return list; | |
9d733021 | 2222 | } |
9e080f92 | 2223 | |
2224 | // check connection, in case connect | |
9d733021 | 2225 | if (!Connect(system)) |
2226 | { | |
4a33bdd9 | 2227 | Log(detector, Form("GetFileSources - Couldn't connect to %s FXS database", GetSystemName(system))); |
9d733021 | 2228 | return NULL; |
9e080f92 | 2229 | } |
2230 | ||
9d733021 | 2231 | TString sourceName = 0; |
2232 | if (system == kDAQ) | |
2233 | { | |
2234 | sourceName = "DAQsource"; | |
2235 | } else if (system == kHLT) | |
2236 | { | |
2237 | sourceName = "DDLnumbers"; | |
2238 | } | |
2239 | ||
4a33bdd9 | 2240 | TString sqlQueryStart = Form("select distinct %s from %s where", sourceName.Data(), fConfig->GetFXSdbTable(system)); |
2241 | TString whereClause = Form("run=%d and detector=\"%s\"", | |
2242 | GetCurrentRun(), detector); | |
2243 | if (id) | |
2244 | whereClause += Form(" and fileId=\"%s\"", id); | |
9e080f92 | 2245 | TString sqlQuery = Form("%s %s", sqlQueryStart.Data(), whereClause.Data()); |
2246 | ||
2247 | AliDebug(2, Form("SQL query: \n%s",sqlQuery.Data())); | |
2248 | ||
2249 | // Query execution | |
2250 | TSQLResult* aResult; | |
9d733021 | 2251 | aResult = fServer[system]->Query(sqlQuery); |
9e080f92 | 2252 | if (!aResult) { |
9d733021 | 2253 | Log(detector, Form("GetFileSources - Can't execute SQL query to %s database for id: %s", |
2254 | GetSystemName(system), id)); | |
9e080f92 | 2255 | return 0; |
2256 | } | |
2257 | ||
86aa42c3 | 2258 | TList *list = new TList(); |
2259 | list->SetOwner(1); | |
2260 | ||
9d733021 | 2261 | if (aResult->GetRowCount() == 0) |
2262 | { | |
9e080f92 | 2263 | Log(detector, |
9d733021 | 2264 | Form("GetFileSources - No entry in %s FXS table for id: %s", GetSystemName(system), id)); |
9e080f92 | 2265 | delete aResult; |
86aa42c3 | 2266 | return list; |
9e080f92 | 2267 | } |
2268 | ||
1bcd28db | 2269 | Log(detector, Form("GetFileSources - Found %d sources", aResult->GetRowCount())); |
9e080f92 | 2270 | |
1bcd28db | 2271 | TSQLRow* aRow; |
9d733021 | 2272 | while ((aRow = aResult->Next())) |
2273 | { | |
9e080f92 | 2274 | |
9d733021 | 2275 | TString source(aRow->GetField(0), aRow->GetFieldLength(0)); |
2276 | AliDebug(2, Form("%s = %s", sourceName.Data(), source.Data())); | |
2277 | list->Add(new TObjString(source)); | |
9e080f92 | 2278 | delete aRow; |
2279 | } | |
9d733021 | 2280 | |
9e080f92 | 2281 | delete aResult; |
2282 | ||
2283 | return list; | |
2bb7b766 | 2284 | } |
2285 | ||
4a33bdd9 | 2286 | //______________________________________________________________________________________________ |
2287 | TList* AliShuttle::GetFileIDs(Int_t system, const char* detector, const char* source) | |
2288 | { | |
2289 | // | |
2290 | // Get all ids of condition files produced by a given source from file exchange servers | |
2291 | // | |
2292 | ||
1bcd28db | 2293 | Log(detector, Form("GetFileIDs - Retrieving ids with source %s with %s", source, GetSystemName(system))); |
2294 | ||
4a33bdd9 | 2295 | // check if test mode should simulate a FXS error |
2296 | if (fTestMode & kErrorFXSSources) | |
2297 | { | |
2298 | Log(detector, Form("GetFileIDs - In TESTMODE - Simulating error while connecting to %s FXS", GetSystemName(system))); | |
2299 | return 0; | |
2300 | } | |
2301 | ||
2302 | // check connection, in case connect | |
2303 | if (!Connect(system)) | |
2304 | { | |
2305 | Log(detector, Form("GetFileIDs - Couldn't connect to %s FXS database", GetSystemName(system))); | |
2306 | return NULL; | |
2307 | } | |
2308 | ||
2309 | TString sourceName = 0; | |
2310 | if (system == kDAQ) | |
2311 | { | |
2312 | sourceName = "DAQsource"; | |
2313 | } else if (system == kHLT) | |
2314 | { | |
2315 | sourceName = "DDLnumbers"; | |
2316 | } | |
2317 | ||
2318 | TString sqlQueryStart = Form("select fileId from %s where", fConfig->GetFXSdbTable(system)); | |
2319 | TString whereClause = Form("run=%d and detector=\"%s\"", | |
2320 | GetCurrentRun(), detector); | |
2321 | if (sourceName.Length() > 0 && source) | |
2322 | whereClause += Form(" and %s=\"%s\"", sourceName.Data(), source); | |
2323 | TString sqlQuery = Form("%s %s", sqlQueryStart.Data(), whereClause.Data()); | |
2324 | ||
2325 | AliDebug(2, Form("SQL query: \n%s",sqlQuery.Data())); | |
2326 | ||
2327 | // Query execution | |
2328 | TSQLResult* aResult; | |
2329 | aResult = fServer[system]->Query(sqlQuery); | |
2330 | if (!aResult) { | |
2331 | Log(detector, Form("GetFileIDs - Can't execute SQL query to %s database for source: %s", | |
2332 | GetSystemName(system), source)); | |
2333 | return 0; | |
2334 | } | |
2335 | ||
2336 | TList *list = new TList(); | |
2337 | list->SetOwner(1); | |
2338 | ||
2339 | if (aResult->GetRowCount() == 0) | |
2340 | { | |
2341 | Log(detector, | |
2342 | Form("GetFileIDs - No entry in %s FXS table for source: %s", GetSystemName(system), source)); | |
2343 | delete aResult; | |
2344 | return list; | |
2345 | } | |
2346 | ||
1bcd28db | 2347 | Log(detector, Form("GetFileIDs - Found %d ids", aResult->GetRowCount())); |
2348 | ||
4a33bdd9 | 2349 | TSQLRow* aRow; |
2350 | ||
2351 | while ((aRow = aResult->Next())) | |
2352 | { | |
2353 | ||
2354 | TString id(aRow->GetField(0), aRow->GetFieldLength(0)); | |
2355 | AliDebug(2, Form("fileId = %s", id.Data())); | |
2356 | list->Add(new TObjString(id)); | |
2357 | delete aRow; | |
2358 | } | |
2359 | ||
2360 | delete aResult; | |
2361 | ||
2362 | return list; | |
2363 | } | |
2364 | ||
2bb7b766 | 2365 | //______________________________________________________________________________________________ |
9d733021 | 2366 | Bool_t AliShuttle::Connect(Int_t system) |
2bb7b766 | 2367 | { |
9827400b | 2368 | // Connect to MySQL Server of the system's FXS MySQL databases |
2369 | // DAQ Logbook, Shuttle Logbook and DAQ FXS db are on the same host | |
2370 | // | |
57f50b3c | 2371 | |
9d733021 | 2372 | // check connection: if already connected return |
2373 | if(fServer[system] && fServer[system]->IsConnected()) return kTRUE; | |
57f50b3c | 2374 | |
9d733021 | 2375 | TString dbHost, dbUser, dbPass, dbName; |
57f50b3c | 2376 | |
9d733021 | 2377 | if (system < 3) // FXS db servers |
2378 | { | |
2379 | dbHost = Form("mysql://%s:%d", fConfig->GetFXSdbHost(system), fConfig->GetFXSdbPort(system)); | |
2380 | dbUser = fConfig->GetFXSdbUser(system); | |
2381 | dbPass = fConfig->GetFXSdbPass(system); | |
2382 | dbName = fConfig->GetFXSdbName(system); | |
2383 | } else { // Run & Shuttle logbook servers | |
2384 | // TODO Will the Shuttle logbook server be the same as the Run logbook server ??? | |
2385 | dbHost = Form("mysql://%s:%d", fConfig->GetDAQlbHost(), fConfig->GetDAQlbPort()); | |
2386 | dbUser = fConfig->GetDAQlbUser(); | |
2387 | dbPass = fConfig->GetDAQlbPass(); | |
2388 | dbName = fConfig->GetDAQlbDB(); | |
2389 | } | |
57f50b3c | 2390 | |
9d733021 | 2391 | fServer[system] = TSQLServer::Connect(dbHost.Data(), dbUser.Data(), dbPass.Data()); |
2392 | if (!fServer[system] || !fServer[system]->IsConnected()) { | |
2393 | if(system < 3) | |
2394 | { | |
2395 | AliError(Form("Can't establish connection to FXS database for %s", | |
2396 | AliShuttleInterface::GetSystemName(system))); | |
2397 | } else { | |
2398 | AliError("Can't establish connection to Run logbook."); | |
57f50b3c | 2399 | } |
9d733021 | 2400 | if(fServer[system]) delete fServer[system]; |
2401 | return kFALSE; | |
2bb7b766 | 2402 | } |
57f50b3c | 2403 | |
9d733021 | 2404 | // Get tables |
2405 | TSQLResult* aResult=0; | |
2406 | switch(system){ | |
2407 | case kDAQ: | |
2408 | aResult = fServer[kDAQ]->GetTables(dbName.Data()); | |
2409 | break; | |
2410 | case kDCS: | |
2411 | aResult = fServer[kDCS]->GetTables(dbName.Data()); | |
2412 | break; | |
2413 | case kHLT: | |
2414 | aResult = fServer[kHLT]->GetTables(dbName.Data()); | |
2415 | break; | |
2416 | default: | |
2417 | aResult = fServer[3]->GetTables(dbName.Data()); | |
2418 | break; | |
2419 | } | |
2420 | ||
2421 | delete aResult; | |
2bb7b766 | 2422 | return kTRUE; |
2423 | } | |
57f50b3c | 2424 | |
9e080f92 | 2425 | //______________________________________________________________________________________________ |
9d733021 | 2426 | Bool_t AliShuttle::UpdateTable() |
9e080f92 | 2427 | { |
9827400b | 2428 | // |
2429 | // Update FXS table filling time_processed field in all rows corresponding to current run and detector | |
2430 | // | |
9e080f92 | 2431 | |
9d733021 | 2432 | Bool_t result = kTRUE; |
9e080f92 | 2433 | |
9d733021 | 2434 | for (UInt_t system=0; system<3; system++) |
2435 | { | |
2436 | if(!fFXSCalled[system]) continue; | |
9e080f92 | 2437 | |
9d733021 | 2438 | // check connection, in case connect |
2439 | if (!Connect(system)) | |
2440 | { | |
2441 | Log(fCurrentDetector, Form("UpdateTable - Couldn't connect to %s FXS database", GetSystemName(system))); | |
2442 | result = kFALSE; | |
2443 | continue; | |
9e080f92 | 2444 | } |
9e080f92 | 2445 | |
9d733021 | 2446 | TTimeStamp now; // now |
2447 | ||
2448 | // Loop on FXS list entries | |
2449 | TIter iter(&fFXSlist[system]); | |
2450 | TObjString *aFXSentry=0; | |
2451 | while ((aFXSentry = dynamic_cast<TObjString*> (iter.Next()))) | |
2452 | { | |
2453 | TString aFXSentrystr = aFXSentry->String(); | |
2454 | TObjArray *aFXSarray = aFXSentrystr.Tokenize("#!?!#"); | |
2455 | if (!aFXSarray || aFXSarray->GetEntries() != 2 ) | |
2456 | { | |
2457 | Log(fCurrentDetector, Form("UpdateTable - error updating %s FXS entry. Check string: <%s>", | |
2458 | GetSystemName(system), aFXSentrystr.Data())); | |
2459 | if(aFXSarray) delete aFXSarray; | |
2460 | result = kFALSE; | |
2461 | continue; | |
2462 | } | |
2463 | const char* fileId = ((TObjString*) aFXSarray->At(0))->GetName(); | |
2464 | const char* source = ((TObjString*) aFXSarray->At(1))->GetName(); | |
2465 | ||
2466 | TString whereClause; | |
2467 | if (system == kDAQ) | |
2468 | { | |
2469 | whereClause = Form("where run=%d and detector=\"%s\" and fileId=\"%s\" and DAQsource=\"%s\";", | |
2470 | GetCurrentRun(), fCurrentDetector.Data(), fileId, source); | |
2471 | } | |
2472 | else if (system == kDCS) | |
2473 | { | |
2474 | whereClause = Form("where run=%d and detector=\"%s\" and fileId=\"%s\";", | |
2475 | GetCurrentRun(), fCurrentDetector.Data(), fileId); | |
2476 | } | |
2477 | else if (system == kHLT) | |
2478 | { | |
2479 | whereClause = Form("where run=%d and detector=\"%s\" and fileId=\"%s\" and DDLnumbers=\"%s\";", | |
2480 | GetCurrentRun(), fCurrentDetector.Data(), fileId, source); | |
2481 | } | |
2482 | ||
2483 | delete aFXSarray; | |
9e080f92 | 2484 | |
9d733021 | 2485 | TString sqlQuery = Form("update %s set time_processed=%d %s", fConfig->GetFXSdbTable(system), |
2486 | now.GetSec(), whereClause.Data()); | |
9e080f92 | 2487 | |
9d733021 | 2488 | AliDebug(2, Form("SQL query: \n%s",sqlQuery.Data())); |
9e080f92 | 2489 | |
9d733021 | 2490 | // Query execution |
2491 | TSQLResult* aResult; | |
2492 | aResult = dynamic_cast<TSQLResult*> (fServer[system]->Query(sqlQuery)); | |
2493 | if (!aResult) | |
2494 | { | |
2495 | Log(fCurrentDetector, Form("UpdateTable - %s db: can't execute SQL query <%s>", | |
2496 | GetSystemName(system), sqlQuery.Data())); | |
2497 | result = kFALSE; | |
2498 | continue; | |
2499 | } | |
2500 | delete aResult; | |
9e080f92 | 2501 | } |
9e080f92 | 2502 | } |
2503 | ||
9d733021 | 2504 | return result; |
9e080f92 | 2505 | } |
57f50b3c | 2506 | |
3301427a | 2507 | //______________________________________________________________________________________________ |
2508 | Bool_t AliShuttle::UpdateTableFailCase() | |
2509 | { | |
9827400b | 2510 | // Update FXS table filling time_processed field in all rows corresponding to current run and detector |
2511 | // this is called in case the preprocessor is declared failed for the current run, because | |
2512 | // the fields are updated only in case of success | |
3301427a | 2513 | |
2514 | Bool_t result = kTRUE; | |
2515 | ||
2516 | for (UInt_t system=0; system<3; system++) | |
2517 | { | |
2518 | // check connection, in case connect | |
2519 | if (!Connect(system)) | |
2520 | { | |
2521 | Log(fCurrentDetector, Form("UpdateTableFailCase - Couldn't connect to %s FXS database", | |
2522 | GetSystemName(system))); | |
2523 | result = kFALSE; | |
2524 | continue; | |
2525 | } | |
2526 | ||
2527 | TTimeStamp now; // now | |
2528 | ||
2529 | // Loop on FXS list entries | |
2530 | ||
2531 | TString whereClause = Form("where run=%d and detector=\"%s\";", | |
2532 | GetCurrentRun(), fCurrentDetector.Data()); | |
2533 | ||
2534 | ||
2535 | TString sqlQuery = Form("update %s set time_processed=%d %s", fConfig->GetFXSdbTable(system), | |
2536 | now.GetSec(), whereClause.Data()); | |
2537 | ||
2538 | AliDebug(2, Form("SQL query: \n%s",sqlQuery.Data())); | |
2539 | ||
2540 | // Query execution | |
2541 | TSQLResult* aResult; | |
2542 | aResult = dynamic_cast<TSQLResult*> (fServer[system]->Query(sqlQuery)); | |
2543 | if (!aResult) | |
2544 | { | |
2545 | Log(fCurrentDetector, Form("UpdateTableFailCase - %s db: can't execute SQL query <%s>", | |
2546 | GetSystemName(system), sqlQuery.Data())); | |
2547 | result = kFALSE; | |
2548 | continue; | |
2549 | } | |
2550 | delete aResult; | |
2551 | } | |
2552 | ||
2553 | return result; | |
2554 | } | |
2555 | ||
2bb7b766 | 2556 | //______________________________________________________________________________________________ |
2557 | Bool_t AliShuttle::UpdateShuttleLogbook(const char* detector, const char* status) | |
2558 | { | |
e7f62f16 | 2559 | // |
2560 | // Update Shuttle logbook filling detector or shuttle_done column | |
2561 | // ex. of usage: UpdateShuttleLogbook("PHOS", "DONE") or UpdateShuttleLogbook("shuttle_done") | |
2562 | // | |
57f50b3c | 2563 | |
2bb7b766 | 2564 | // check connection, in case connect |
be48e3ea | 2565 | if(!Connect(3)){ |
2bb7b766 | 2566 | Log("SHUTTLE", "UpdateShuttleLogbook - Couldn't connect to DAQ Logbook."); |
2567 | return kFALSE; | |
57f50b3c | 2568 | } |
2569 | ||
2bb7b766 | 2570 | TString detName(detector); |
2571 | TString setClause; | |
ee6f7523 | 2572 | if (detName == "shuttle_done" || detName == "shuttle_ignored") |
e7f62f16 | 2573 | { |
2bb7b766 | 2574 | setClause = "set shuttle_done=1"; |
e7f62f16 | 2575 | |
ee6f7523 | 2576 | if (detName == "shuttle_done") |
b0e53b15 | 2577 | { |
2578 | // Send the information to ML | |
2579 | TMonaLisaText mlStatus("SHUTTLE_status", "Done"); | |
e7f62f16 | 2580 | |
b0e53b15 | 2581 | TList mlList; |
2582 | mlList.Add(&mlStatus); | |
2583 | ||
ee6f7523 | 2584 | TString mlID; |
2585 | mlID.Form("%d", GetCurrentRun()); | |
2586 | fMonaLisa->SendParameters(&mlList, mlID); | |
b0e53b15 | 2587 | } |
2bb7b766 | 2588 | } else { |
2bb7b766 | 2589 | TString statusStr(status); |
2590 | if(statusStr.Contains("done", TString::kIgnoreCase) || | |
2591 | statusStr.Contains("failed", TString::kIgnoreCase)){ | |
eba76848 | 2592 | setClause = Form("set %s=\"%s\"", detector, status); |
2bb7b766 | 2593 | } else { |
2594 | Log("SHUTTLE", | |
2595 | Form("UpdateShuttleLogbook - Invalid status <%s> for detector %s", | |
2596 | status, detector)); | |
2597 | return kFALSE; | |
2598 | } | |
2599 | } | |
57f50b3c | 2600 | |
2bb7b766 | 2601 | TString whereClause = Form("where run=%d", GetCurrentRun()); |
2602 | ||
441b0e9c | 2603 | TString sqlQuery = Form("update %s %s %s", |
2604 | fConfig->GetShuttlelbTable(), setClause.Data(), whereClause.Data()); | |
57f50b3c | 2605 | |
2bb7b766 | 2606 | AliDebug(2, Form("SQL query: \n%s",sqlQuery.Data())); |
2607 | ||
2608 | // Query execution | |
2609 | TSQLResult* aResult; | |
be48e3ea | 2610 | aResult = dynamic_cast<TSQLResult*> (fServer[3]->Query(sqlQuery)); |
2bb7b766 | 2611 | if (!aResult) { |
2612 | Log("SHUTTLE", Form("UpdateShuttleLogbook - Can't execute query <%s>", sqlQuery.Data())); | |
2613 | return kFALSE; | |
57f50b3c | 2614 | } |
2bb7b766 | 2615 | delete aResult; |
57f50b3c | 2616 | |
2617 | return kTRUE; | |
2618 | } | |
2619 | ||
2620 | //______________________________________________________________________________________________ | |
2bb7b766 | 2621 | Int_t AliShuttle::GetCurrentRun() const |
2622 | { | |
9827400b | 2623 | // |
2624 | // Get current run from logbook entry | |
2625 | // | |
57f50b3c | 2626 | |
2bb7b766 | 2627 | return fLogbookEntry ? fLogbookEntry->GetRun() : -1; |
57f50b3c | 2628 | } |
2629 | ||
2630 | //______________________________________________________________________________________________ | |
2bb7b766 | 2631 | UInt_t AliShuttle::GetCurrentStartTime() const |
2632 | { | |
9827400b | 2633 | // |
2634 | // get current start time | |
2635 | // | |
57f50b3c | 2636 | |
2bb7b766 | 2637 | return fLogbookEntry ? fLogbookEntry->GetStartTime() : 0; |
57f50b3c | 2638 | } |
2639 | ||
2640 | //______________________________________________________________________________________________ | |
2bb7b766 | 2641 | UInt_t AliShuttle::GetCurrentEndTime() const |
2642 | { | |
9827400b | 2643 | // |
2644 | // get current end time from logbook entry | |
2645 | // | |
57f50b3c | 2646 | |
2bb7b766 | 2647 | return fLogbookEntry ? fLogbookEntry->GetEndTime() : 0; |
57f50b3c | 2648 | } |
2649 | ||
675f64cd | 2650 | //______________________________________________________________________________________________ |
2651 | UInt_t AliShuttle::GetCurrentYear() const | |
2652 | { | |
2653 | // | |
2654 | // Get current year from logbook entry | |
2655 | // | |
2656 | ||
2657 | if (!fLogbookEntry) return 0; | |
2658 | ||
2659 | TTimeStamp startTime(GetCurrentStartTime()); | |
2660 | TString year = Form("%d",startTime.GetDate()); | |
2661 | year = year(0,4); | |
2662 | ||
2663 | return year.Atoi(); | |
2664 | } | |
2665 | ||
2666 | //______________________________________________________________________________________________ | |
2667 | const char* AliShuttle::GetLHCPeriod() const | |
2668 | { | |
2669 | // | |
2670 | // Get current LHC period from logbook entry | |
2671 | // | |
2672 | ||
2673 | if (!fLogbookEntry) return 0; | |
2674 | ||
2675 | return fLogbookEntry->GetRunParameter("LHCperiod"); | |
2676 | } | |
2677 | ||
b948db8d | 2678 | //______________________________________________________________________________________________ |
2679 | void AliShuttle::Log(const char* detector, const char* message) | |
2680 | { | |
9827400b | 2681 | // |
2682 | // Fill log string with a message | |
2683 | // | |
b948db8d | 2684 | |
7d4cf768 | 2685 | TString logRunDir = GetShuttleLogDir(); |
2686 | if (GetCurrentRun() >=0) | |
2687 | logRunDir += Form("/%d", GetCurrentRun()); | |
2688 | ||
2689 | void* dir = gSystem->OpenDirectory(logRunDir.Data()); | |
84090f85 | 2690 | if (dir == NULL) { |
7d4cf768 | 2691 | if (gSystem->mkdir(logRunDir.Data(), kTRUE)) { |
36c99a6a | 2692 | AliError(Form("Can't open directory <%s>", GetShuttleLogDir())); |
84090f85 | 2693 | return; |
2694 | } | |
b948db8d | 2695 | |
84090f85 | 2696 | } else { |
2697 | gSystem->FreeDirectory(dir); | |
2698 | } | |
b948db8d | 2699 | |
cb343cfd | 2700 | TString toLog = Form("%s (%d): %s - ", TTimeStamp(time(0)).AsString("s"), getpid(), detector); |
e7f62f16 | 2701 | if (GetCurrentRun() >= 0) |
2702 | toLog += Form("run %d - ", GetCurrentRun()); | |
2bb7b766 | 2703 | toLog += Form("%s", message); |
2704 | ||
84090f85 | 2705 | AliInfo(toLog.Data()); |
ffa29e93 | 2706 | |
2707 | // if we redirect the log output already to the file, leave here | |
2708 | if (fOutputRedirected && strcmp(detector, "SHUTTLE") != 0) | |
2709 | return; | |
b948db8d | 2710 | |
ffa29e93 | 2711 | TString fileName = GetLogFileName(detector); |
e7f62f16 | 2712 | |
84090f85 | 2713 | gSystem->ExpandPathName(fileName); |
2714 | ||
2715 | ofstream logFile; | |
2716 | logFile.open(fileName, ofstream::out | ofstream::app); | |
2717 | ||
2718 | if (!logFile.is_open()) { | |
2719 | AliError(Form("Could not open file %s", fileName.Data())); | |
2720 | return; | |
2721 | } | |
7bfb2090 | 2722 | |
84090f85 | 2723 | logFile << toLog.Data() << "\n"; |
b948db8d | 2724 | |
84090f85 | 2725 | logFile.close(); |
b948db8d | 2726 | } |
2bb7b766 | 2727 | |
ffa29e93 | 2728 | //______________________________________________________________________________________________ |
2729 | TString AliShuttle::GetLogFileName(const char* detector) const | |
2730 | { | |
2731 | // | |
2732 | // returns the name of the log file for a given sub detector | |
2733 | // | |
2734 | ||
2735 | TString fileName; | |
2736 | ||
2737 | if (GetCurrentRun() >= 0) | |
7d4cf768 | 2738 | { |
2739 | fileName.Form("%s/%d/%s_%d.log", GetShuttleLogDir(), GetCurrentRun(), | |
2740 | detector, GetCurrentRun()); | |
2741 | } else { | |
ffa29e93 | 2742 | fileName.Form("%s/%s.log", GetShuttleLogDir(), detector); |
7d4cf768 | 2743 | } |
ffa29e93 | 2744 | |
2745 | return fileName; | |
2746 | } | |
2747 | ||
ee6f7523 | 2748 | //______________________________________________________________________________________________ |
2749 | void AliShuttle::SendAlive() | |
2750 | { | |
2751 | // sends alive message to ML | |
2752 | ||
2753 | TMonaLisaText mlStatus("SHUTTLE_status", "Alive"); | |
2754 | ||
2755 | TList mlList; | |
2756 | mlList.Add(&mlStatus); | |
2757 | ||
2758 | fMonaLisa->SendParameters(&mlList, "__PROCESSINGINFO__"); | |
2759 | } | |
2760 | ||
2bb7b766 | 2761 | //______________________________________________________________________________________________ |
2762 | Bool_t AliShuttle::Collect(Int_t run) | |
2763 | { | |
9827400b | 2764 | // |
2765 | // Collects conditions data for all UNPROCESSED run written to DAQ LogBook in case of run = -1 (default) | |
2766 | // If a dedicated run is given this run is processed | |
2767 | // | |
2768 | // In operational mode, this is the Shuttle function triggered by the EOR signal. | |
2769 | // | |
2bb7b766 | 2770 | |
eba76848 | 2771 | if (run == -1) |
2772 | Log("SHUTTLE","Collect - Shuttle called. Collecting conditions data for unprocessed runs"); | |
2773 | else | |
2774 | Log("SHUTTLE", Form("Collect - Shuttle called. Collecting conditions data for run %d", run)); | |
cb343cfd | 2775 | |
2776 | SetLastAction("Starting"); | |
2bb7b766 | 2777 | |
ee6f7523 | 2778 | // create ML instance |
2779 | if (!fMonaLisa) | |
2780 | fMonaLisa = new TMonaLisaWriter(fConfig->GetMonitorHost(), fConfig->GetMonitorTable()); | |
2781 | ||
ee6f7523 | 2782 | SendAlive(); |
fbc112e3 | 2783 | CountOpenRuns(); |
ee6f7523 | 2784 | |
2bb7b766 | 2785 | TString whereClause("where shuttle_done=0"); |
eba76848 | 2786 | if (run != -1) |
2787 | whereClause += Form(" and run=%d", run); | |
2bb7b766 | 2788 | |
2789 | TObjArray shuttleLogbookEntries; | |
be48e3ea | 2790 | if (!QueryShuttleLogbook(whereClause, shuttleLogbookEntries)) |
2791 | { | |
cb343cfd | 2792 | Log("SHUTTLE", "Collect - Can't retrieve entries from Shuttle logbook"); |
2bb7b766 | 2793 | return kFALSE; |
2794 | } | |
2795 | ||
9e080f92 | 2796 | if (shuttleLogbookEntries.GetEntries() == 0) |
2797 | { | |
2798 | if (run == -1) | |
2799 | Log("SHUTTLE","Collect - Found no UNPROCESSED runs in Shuttle logbook"); | |
2800 | else | |
2801 | Log("SHUTTLE", Form("Collect - Run %d is already DONE " | |
2802 | "or it does not exist in Shuttle logbook", run)); | |
2803 | return kTRUE; | |
2804 | } | |
2805 | ||
be48e3ea | 2806 | for (UInt_t iDet=0; iDet<NDetectors(); iDet++) |
2807 | fFirstUnprocessed[iDet] = kTRUE; | |
2808 | ||
fc5a4708 | 2809 | if (run != -1) |
be48e3ea | 2810 | { |
2811 | // query Shuttle logbook for earlier runs, check if some detectors are unprocessed, | |
2812 | // flag them into fFirstUnprocessed array | |
2813 | TString whereClause(Form("where shuttle_done=0 and run < %d", run)); | |
2814 | TObjArray tmpLogbookEntries; | |
2815 | if (!QueryShuttleLogbook(whereClause, tmpLogbookEntries)) | |
2816 | { | |
2817 | Log("SHUTTLE", "Collect - Can't retrieve entries from Shuttle logbook"); | |
2818 | return kFALSE; | |
2819 | } | |
2820 | ||
2821 | TIter iter(&tmpLogbookEntries); | |
2822 | AliShuttleLogbookEntry* anEntry = 0; | |
2823 | while ((anEntry = dynamic_cast<AliShuttleLogbookEntry*> (iter.Next()))) | |
2824 | { | |
2825 | for (UInt_t iDet=0; iDet<NDetectors(); iDet++) | |
2826 | { | |
2827 | if (anEntry->GetDetectorStatus(iDet) == AliShuttleLogbookEntry::kUnprocessed) | |
2828 | { | |
2829 | AliDebug(2, Form("Run %d: setting %s as \"not first time unprocessed\"", | |
2830 | anEntry->GetRun(), GetDetName(iDet))); | |
2831 | fFirstUnprocessed[iDet] = kFALSE; | |
2832 | } | |
2833 | } | |
2834 | ||
2835 | } | |
2836 | ||
2837 | } | |
2838 | ||
2839 | if (!RetrieveConditionsData(shuttleLogbookEntries)) | |
2840 | { | |
cb343cfd | 2841 | Log("SHUTTLE", "Collect - Process of at least one run failed"); |
a3ca69fe | 2842 | CountOpenRuns(); |
2bb7b766 | 2843 | return kFALSE; |
2844 | } | |
2845 | ||
36c99a6a | 2846 | Log("SHUTTLE", "Collect - Requested run(s) successfully processed"); |
a3ca69fe | 2847 | CountOpenRuns(); |
eba76848 | 2848 | return kTRUE; |
2bb7b766 | 2849 | } |
2850 | ||
2bb7b766 | 2851 | //______________________________________________________________________________________________ |
2852 | Bool_t AliShuttle::RetrieveConditionsData(const TObjArray& dateEntries) | |
2853 | { | |
9827400b | 2854 | // |
2855 | // Retrieve conditions data for all runs that aren't processed yet | |
2856 | // | |
2bb7b766 | 2857 | |
2858 | Bool_t hasError = kFALSE; | |
2859 | ||
2860 | TIter iter(&dateEntries); | |
2861 | AliShuttleLogbookEntry* anEntry; | |
2862 | ||
2863 | while ((anEntry = (AliShuttleLogbookEntry*) iter.Next())){ | |
2864 | if (!Process(anEntry)){ | |
2865 | hasError = kTRUE; | |
2866 | } | |
4b95672b | 2867 | |
2868 | // clean SHUTTLE temp directory | |
d524ade6 | 2869 | //TString filename = Form("%s/*.shuttle", GetShuttleTempDir()); |
2870 | //RemoveFile(filename.Data()); | |
2bb7b766 | 2871 | } |
2872 | ||
2873 | return hasError == kFALSE; | |
2874 | } | |
cb343cfd | 2875 | |
2876 | //______________________________________________________________________________________________ | |
2877 | ULong_t AliShuttle::GetTimeOfLastAction() const | |
2878 | { | |
9827400b | 2879 | // |
2880 | // Gets time of last action | |
2881 | // | |
2882 | ||
cb343cfd | 2883 | ULong_t tmp; |
36c99a6a | 2884 | |
cb343cfd | 2885 | fMonitoringMutex->Lock(); |
be48e3ea | 2886 | |
cb343cfd | 2887 | tmp = fLastActionTime; |
36c99a6a | 2888 | |
cb343cfd | 2889 | fMonitoringMutex->UnLock(); |
36c99a6a | 2890 | |
cb343cfd | 2891 | return tmp; |
2892 | } | |
2893 | ||
2894 | //______________________________________________________________________________________________ | |
2895 | const TString AliShuttle::GetLastAction() const | |
2896 | { | |
9827400b | 2897 | // |
cb343cfd | 2898 | // returns a string description of the last action |
9827400b | 2899 | // |
cb343cfd | 2900 | |
2901 | TString tmp; | |
36c99a6a | 2902 | |
cb343cfd | 2903 | fMonitoringMutex->Lock(); |
2904 | ||
2905 | tmp = fLastAction; | |
2906 | ||
2907 | fMonitoringMutex->UnLock(); | |
2908 | ||
36c99a6a | 2909 | return tmp; |
cb343cfd | 2910 | } |
2911 | ||
2912 | //______________________________________________________________________________________________ | |
2913 | void AliShuttle::SetLastAction(const char* action) | |
2914 | { | |
9827400b | 2915 | // |
cb343cfd | 2916 | // updates the monitoring variables |
9827400b | 2917 | // |
36c99a6a | 2918 | |
cb343cfd | 2919 | fMonitoringMutex->Lock(); |
36c99a6a | 2920 | |
cb343cfd | 2921 | fLastAction = action; |
2922 | fLastActionTime = time(0); | |
2923 | ||
2924 | fMonitoringMutex->UnLock(); | |
2925 | } | |
eba76848 | 2926 | |
2927 | //______________________________________________________________________________________________ | |
2928 | const char* AliShuttle::GetRunParameter(const char* param) | |
2929 | { | |
9827400b | 2930 | // |
2931 | // returns run parameter read from DAQ logbook | |
2932 | // | |
eba76848 | 2933 | |
2934 | if(!fLogbookEntry) { | |
2935 | AliError("No logbook entry!"); | |
2936 | return 0; | |
2937 | } | |
2938 | ||
2939 | return fLogbookEntry->GetRunParameter(param); | |
2940 | } | |
57c1a579 | 2941 | |
d386d623 | 2942 | //______________________________________________________________________________________________ |
9827400b | 2943 | AliCDBEntry* AliShuttle::GetFromOCDB(const char* detector, const AliCDBPath& path) |
d386d623 | 2944 | { |
9827400b | 2945 | // |
2946 | // returns object from OCDB valid for current run | |
2947 | // | |
d386d623 | 2948 | |
9827400b | 2949 | if (fTestMode & kErrorOCDB) |
2950 | { | |
2951 | Log(detector, "GetFromOCDB - In TESTMODE - Simulating error with OCDB"); | |
2952 | return 0; | |
2953 | } | |
2954 | ||
d386d623 | 2955 | AliCDBStorage *sto = AliCDBManager::Instance()->GetStorage(fgkMainCDB); |
2956 | if (!sto) | |
2957 | { | |
9827400b | 2958 | Log(detector, "GetFromOCDB - Cannot activate main OCDB for query!"); |
d386d623 | 2959 | return 0; |
2960 | } | |
2961 | ||
2962 | return dynamic_cast<AliCDBEntry*> (sto->Get(path, GetCurrentRun())); | |
2963 | } | |
2964 | ||
57c1a579 | 2965 | //______________________________________________________________________________________________ |
2966 | Bool_t AliShuttle::SendMail() | |
2967 | { | |
9827400b | 2968 | // |
2969 | // sends a mail to the subdetector expert in case of preprocessor error | |
2970 | // | |
2971 | ||
2972 | if (fTestMode != kNone) | |
2973 | return kTRUE; | |
fb40ead4 | 2974 | |
2975 | if (!fConfig->SendMail()) return kTRUE; | |
57c1a579 | 2976 | |
57c1a579 | 2977 | TString to=""; |
2978 | TIter iterExperts(fConfig->GetResponsibles(fCurrentDetector)); | |
2979 | TObjString *anExpert=0; | |
2980 | while ((anExpert = (TObjString*) iterExperts.Next())) | |
2981 | { | |
2982 | to += Form("%s,", anExpert->GetName()); | |
2983 | } | |
4508ab8b | 2984 | if (to.Length() > 0) |
2985 | to.Remove(to.Length()-1); | |
909732f7 | 2986 | AliDebug(2, Form("to: %s",to.Data())); |
57c1a579 | 2987 | |
86aa42c3 | 2988 | if (to.IsNull()) { |
fb40ead4 | 2989 | Log("SHUTTLE", "List of detector responsibles not set!"); |
36c99a6a | 2990 | return kFALSE; |
2991 | } | |
2992 | ||
fbc112e3 | 2993 | void* dir = gSystem->OpenDirectory(GetShuttleLogDir()); |
2994 | if (dir == NULL) | |
2995 | { | |
2996 | if (gSystem->mkdir(GetShuttleLogDir(), kTRUE)) | |
2997 | { | |
2998 | Log("SHUTTLE", Form("SendMail - Can't open directory <%s>", GetShuttleLogDir())); | |
2999 | return kFALSE; | |
3000 | } | |
3001 | ||
3002 | } else { | |
3003 | gSystem->FreeDirectory(dir); | |
3004 | } | |
3005 | ||
4508ab8b | 3006 | TString bodyFileName; |
3007 | bodyFileName.Form("%s/mail.body", GetShuttleLogDir()); | |
3008 | gSystem->ExpandPathName(bodyFileName); | |
3009 | ||
3010 | ofstream mailBody; | |
3011 | mailBody.open(bodyFileName, ofstream::out); | |
3012 | ||
3013 | if (!mailBody.is_open()) | |
3014 | { | |
3015 | Log("SHUTTLE", Form("Could not open mail body file %s", bodyFileName.Data())); | |
3016 | return kFALSE; | |
3017 | } | |
3018 | ||
fb40ead4 | 3019 | TString cc=""; |
3020 | TIter iterAdmins(fConfig->GetAdmins(AliShuttleConfig::kGlobal)); | |
3021 | TObjString *anAdmin=0; | |
3022 | while ((anAdmin = (TObjString*) iterAdmins.Next())) | |
3023 | { | |
3024 | cc += Form("%s,", anAdmin->GetName()); | |
3025 | } | |
3026 | if (cc.Length() > 0) | |
3027 | cc.Remove(to.Length()-1); | |
3028 | AliDebug(2, Form("cc: %s",to.Data())); | |
57c1a579 | 3029 | |
6a1146c4 | 3030 | TString subject = Form("%s Shuttle preprocessor FAILED in run %d (run type = %s)!", |
3031 | fCurrentDetector.Data(), GetCurrentRun(), GetRunType()); | |
909732f7 | 3032 | AliDebug(2, Form("subject: %s", subject.Data())); |
57c1a579 | 3033 | |
3034 | TString body = Form("Dear %s expert(s), \n\n", fCurrentDetector.Data()); | |
3035 | body += Form("SHUTTLE just detected that your preprocessor " | |
6a1146c4 | 3036 | "failed processing run %d (run type = %s)!!\n\n", |
3037 | GetCurrentRun(), GetRunType()); | |
7d4cf768 | 3038 | body += Form("Please check %s status on the SHUTTLE monitoring page: \n\n", |
3039 | fCurrentDetector.Data()); | |
b0e53b15 | 3040 | if (fConfig->GetRunMode() == AliShuttleConfig::kTest) |
3041 | { | |
3042 | body += Form("\thttp://pcalimonitor.cern.ch:8889/shuttle.jsp?time=168 \n\n"); | |
3043 | } else { | |
4508ab8b | 3044 | body += Form("\thttp://pcalimonitor.cern.ch/shuttle.jsp?instance=PROD&time=168 \n\n"); |
b0e53b15 | 3045 | } |
3046 | ||
7d4cf768 | 3047 | |
3048 | TString logFolder = "logs"; | |
3049 | if (fConfig->GetRunMode() == AliShuttleConfig::kProd) | |
3050 | logFolder += "_PROD"; | |
3051 | ||
3052 | ||
546242fb | 3053 | body += Form("Find the %s log for the current run on \n\n" |
7d4cf768 | 3054 | "\thttp://pcalishuttle01.cern.ch:8880/%s/%d/%s_%d.log \n\n", |
3055 | fCurrentDetector.Data(), logFolder.Data(), GetCurrentRun(), | |
3056 | fCurrentDetector.Data(), GetCurrentRun()); | |
3057 | body += Form("The last 10 lines of %s log file are following:\n\n", fCurrentDetector.Data()); | |
57c1a579 | 3058 | |
909732f7 | 3059 | AliDebug(2, Form("Body begin: %s", body.Data())); |
57c1a579 | 3060 | |
3061 | mailBody << body.Data(); | |
3062 | mailBody.close(); | |
3063 | mailBody.open(bodyFileName, ofstream::out | ofstream::app); | |
3064 | ||
7d4cf768 | 3065 | TString logFileName = Form("%s/%d/%s_%d.log", GetShuttleLogDir(), |
3066 | GetCurrentRun(), fCurrentDetector.Data(), GetCurrentRun()); | |
57c1a579 | 3067 | TString tailCommand = Form("tail -n 10 %s >> %s", logFileName.Data(), bodyFileName.Data()); |
3068 | if (gSystem->Exec(tailCommand.Data())) | |
3069 | { | |
3070 | mailBody << Form("%s log file not found ...\n\n", fCurrentDetector.Data()); | |
3071 | } | |
3072 | ||
3073 | TString endBody = Form("------------------------------------------------------\n\n"); | |
675f64cd | 3074 | endBody += Form("In case of problems please contact the SHUTTLE core team.\n\n"); |
3075 | endBody += "Please do not answer this message directly, it is automatically generated.\n\n"; | |
3076 | endBody += "Greetings,\n\n \t\t\tthe SHUTTLE\n"; | |
3077 | ||
3078 | AliDebug(2, Form("Body end: %s", endBody.Data())); | |
3079 | ||
3080 | mailBody << endBody.Data(); | |
3081 | ||
3082 | mailBody.close(); | |
3083 | ||
3084 | // send mail! | |
3085 | TString mailCommand = Form("mail -s \"%s\" -c %s %s < %s", | |
3086 | subject.Data(), | |
3087 | cc.Data(), | |
3088 | to.Data(), | |
3089 | bodyFileName.Data()); | |
3090 | AliDebug(2, Form("mail command: %s", mailCommand.Data())); | |
3091 | ||
3092 | Bool_t result = gSystem->Exec(mailCommand.Data()); | |
3093 | ||
3094 | return result == 0; | |
3095 | } | |
3096 | ||
3097 | //______________________________________________________________________________________________ | |
3098 | Bool_t AliShuttle::SendMailToDCS() | |
3099 | { | |
3100 | // | |
fb40ead4 | 3101 | // sends a mail to the DCS Amanda experts in case of DCS data point retrieval error |
675f64cd | 3102 | // |
3103 | ||
3104 | if (fTestMode != kNone) | |
3105 | return kTRUE; | |
3106 | ||
fb40ead4 | 3107 | if (!fConfig->SendMail()) return kTRUE; |
3108 | ||
675f64cd | 3109 | void* dir = gSystem->OpenDirectory(GetShuttleLogDir()); |
3110 | if (dir == NULL) | |
3111 | { | |
3112 | if (gSystem->mkdir(GetShuttleLogDir(), kTRUE)) | |
3113 | { | |
3114 | Log("SHUTTLE", Form("SendMailToDCS - Can't open directory <%s>", GetShuttleLogDir())); | |
3115 | return kFALSE; | |
3116 | } | |
3117 | ||
3118 | } else { | |
3119 | gSystem->FreeDirectory(dir); | |
3120 | } | |
3121 | ||
3122 | TString bodyFileName; | |
3123 | bodyFileName.Form("%s/mail.body", GetShuttleLogDir()); | |
3124 | gSystem->ExpandPathName(bodyFileName); | |
3125 | ||
3126 | ofstream mailBody; | |
3127 | mailBody.open(bodyFileName, ofstream::out); | |
3128 | ||
3129 | if (!mailBody.is_open()) | |
3130 | { | |
3131 | Log("SHUTTLE", Form("SendMailToDCS - Could not open mail body file %s", bodyFileName.Data())); | |
3132 | return kFALSE; | |
3133 | } | |
3134 | ||
fb40ead4 | 3135 | TString to=""; |
3136 | TIter iterExperts(fConfig->GetAdmins(AliShuttleConfig::kAmanda)); | |
3137 | TObjString *anExpert=0; | |
3138 | while ((anExpert = (TObjString*) iterExperts.Next())) | |
3139 | { | |
3140 | to += Form("%s,", anExpert->GetName()); | |
3141 | } | |
3142 | if (to.Length() > 0) | |
3143 | to.Remove(to.Length()-1); | |
675f64cd | 3144 | AliDebug(2, Form("to: %s",to.Data())); |
3145 | ||
3146 | if (to.IsNull()) { | |
fb40ead4 | 3147 | Log("SHUTTLE", "List of Amanda server administrators not set!"); |
675f64cd | 3148 | return kFALSE; |
3149 | } | |
3150 | ||
fb40ead4 | 3151 | TString cc=""; |
3152 | TIter iterAdmins(fConfig->GetAdmins(AliShuttleConfig::kGlobal)); | |
3153 | TObjString *anAdmin=0; | |
3154 | while ((anAdmin = (TObjString*) iterAdmins.Next())) | |
3155 | { | |
3156 | cc += Form("%s,", anAdmin->GetName()); | |
3157 | } | |
3158 | if (cc.Length() > 0) | |
3159 | cc.Remove(to.Length()-1); | |
3160 | AliDebug(2, Form("cc: %s",to.Data())); | |
675f64cd | 3161 | |
3162 | TString subject = Form("Retrieval of data points for %s FAILED in run %d !", | |
3163 | fCurrentDetector.Data(), GetCurrentRun()); | |
3164 | AliDebug(2, Form("subject: %s", subject.Data())); | |
3165 | ||
3166 | TString body = Form("Dear DCS experts, \n\n"); | |
3167 | body += Form("SHUTTLE couldn\'t retrieve the data points for detector %s " | |
3168 | "in run %d!!\n\n", fCurrentDetector.Data(), GetCurrentRun()); | |
7d4cf768 | 3169 | body += Form("Please check %s status on the SHUTTLE monitoring page: \n\n", |
3170 | fCurrentDetector.Data()); | |
b0e53b15 | 3171 | if (fConfig->GetRunMode() == AliShuttleConfig::kTest) |
3172 | { | |
3173 | body += Form("\thttp://pcalimonitor.cern.ch:8889/shuttle.jsp?time=168 \n\n"); | |
3174 | } else { | |
3175 | body += Form("\thttp://pcalimonitor.cern.ch/shuttle.jsp?instance=PROD?time=168 \n\n"); | |
3176 | } | |
7d4cf768 | 3177 | |
3178 | TString logFolder = "logs"; | |
3179 | if (fConfig->GetRunMode() == AliShuttleConfig::kProd) | |
3180 | logFolder += "_PROD"; | |
3181 | ||
3182 | ||
675f64cd | 3183 | body += Form("Find the %s log for the current run on \n\n" |
7d4cf768 | 3184 | "\thttp://pcalishuttle01.cern.ch:8880/%s/%d/%s_%d.log \n\n", |
3185 | fCurrentDetector.Data(), logFolder.Data(), GetCurrentRun(), | |
3186 | fCurrentDetector.Data(), GetCurrentRun()); | |
3187 | body += Form("The last 10 lines of %s log file are following:\n\n", fCurrentDetector.Data()); | |
675f64cd | 3188 | |
3189 | AliDebug(2, Form("Body begin: %s", body.Data())); | |
3190 | ||
3191 | mailBody << body.Data(); | |
3192 | mailBody.close(); | |
3193 | mailBody.open(bodyFileName, ofstream::out | ofstream::app); | |
3194 | ||
7d4cf768 | 3195 | TString logFileName = Form("%s/%d/%s_%d.log", GetShuttleLogDir(), GetCurrentRun(), |
3196 | fCurrentDetector.Data(), GetCurrentRun()); | |
675f64cd | 3197 | TString tailCommand = Form("tail -n 10 %s >> %s", logFileName.Data(), bodyFileName.Data()); |
3198 | if (gSystem->Exec(tailCommand.Data())) | |
3199 | { | |
3200 | mailBody << Form("%s log file not found ...\n\n", fCurrentDetector.Data()); | |
3201 | } | |
3202 | ||
3203 | TString endBody = Form("------------------------------------------------------\n\n"); | |
36c99a6a | 3204 | endBody += Form("In case of problems please contact the SHUTTLE core team.\n\n"); |
3205 | endBody += "Please do not answer this message directly, it is automatically generated.\n\n"; | |
546242fb | 3206 | endBody += "Greetings,\n\n \t\t\tthe SHUTTLE\n"; |
57c1a579 | 3207 | |
909732f7 | 3208 | AliDebug(2, Form("Body end: %s", endBody.Data())); |
57c1a579 | 3209 | |
3210 | mailBody << endBody.Data(); | |
3211 | ||
3212 | mailBody.close(); | |
3213 | ||
3214 | // send mail! | |
3215 | TString mailCommand = Form("mail -s \"%s\" -c %s %s < %s", | |
3216 | subject.Data(), | |
3217 | cc.Data(), | |
3218 | to.Data(), | |
3219 | bodyFileName.Data()); | |
909732f7 | 3220 | AliDebug(2, Form("mail command: %s", mailCommand.Data())); |
57c1a579 | 3221 | |
3222 | Bool_t result = gSystem->Exec(mailCommand.Data()); | |
3223 | ||
3224 | return result == 0; | |
3225 | } | |
d386d623 | 3226 | |
441b0e9c | 3227 | //______________________________________________________________________________________________ |
9827400b | 3228 | const char* AliShuttle::GetRunType() |
441b0e9c | 3229 | { |
9827400b | 3230 | // |
3231 | // returns run type read from "run type" logbook | |
3232 | // | |
441b0e9c | 3233 | |
3234 | if(!fLogbookEntry) { | |
3235 | AliError("No logbook entry!"); | |
3236 | return 0; | |
3237 | } | |
3238 | ||
9827400b | 3239 | return fLogbookEntry->GetRunType(); |
441b0e9c | 3240 | } |
3241 | ||
4859271b | 3242 | //______________________________________________________________________________________________ |
3243 | Bool_t AliShuttle::GetHLTStatus() | |
3244 | { | |
3245 | // Return HLT status (ON=1 OFF=0) | |
3246 | // Converts the HLT status from the status string read in the run logbook (not just a bool) | |
3247 | ||
3248 | if(!fLogbookEntry) { | |
3249 | AliError("No logbook entry!"); | |
3250 | return 0; | |
3251 | } | |
3252 | ||
3253 | // TODO implement when HLTStatus is inserted in run logbook | |
3254 | //TString hltStatus = fLogbookEntry->GetRunParameter("HLTStatus"); | |
3255 | //if(hltStatus == "OFF") {return kFALSE}; | |
3256 | ||
3257 | return kTRUE; | |
3258 | } | |
3259 | ||
d386d623 | 3260 | //______________________________________________________________________________________________ |
3261 | void AliShuttle::SetShuttleTempDir(const char* tmpDir) | |
3262 | { | |
9827400b | 3263 | // |
3264 | // sets Shuttle temp directory | |
3265 | // | |
d386d623 | 3266 | |
3267 | fgkShuttleTempDir = gSystem->ExpandPathName(tmpDir); | |
3268 | } | |
3269 | ||
3270 | //______________________________________________________________________________________________ | |
3271 | void AliShuttle::SetShuttleLogDir(const char* logDir) | |
3272 | { | |
9827400b | 3273 | // |
3274 | // sets Shuttle log directory | |
3275 | // | |
d386d623 | 3276 | |
3277 | fgkShuttleLogDir = gSystem->ExpandPathName(logDir); | |
3278 | } |