]> git.uio.no Git - u/mrichter/AliRoot.git/blame - SHUTTLE/AliShuttle.cxx
git://git.uio.no / u / mrichter / AliRoot.git / blame
? search:
summary | shortlog | log | commit | commitdiff | tree
blob | blame (incremental) | history | HEAD
CMake: Define configuration variable as CACHE
[u/mrichter/AliRoot.git] / SHUTTLE / AliShuttle.cxx
CommitLineData
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
3ff44dcb 31#include <stdexcept>
73abe331 32#include "AliShuttle.h"
33
34#include "AliCDBManager.h"
35#include "AliCDBStorage.h"
36#include "AliCDBId.h"
84090f85 37#include "AliCDBRunRange.h"
38#include "AliCDBPath.h"
5164a766 39#include "AliCDBEntry.h"
73abe331 40#include "AliShuttleConfig.h"
eba76848 41#include "DCSClient/AliDCSClient.h"
73abe331 42#include "AliLog.h"
b948db8d 43#include "AliPreprocessor.h"
5164a766 44#include "AliShuttleStatus.h"
2bb7b766 45#include "AliShuttleLogbookEntry.h"
73abe331 46
57f50b3c 47#include <TSystem.h>
58bc3020 48#include <TObject.h>
b948db8d 49#include <TString.h>
57f50b3c 50#include <TTimeStamp.h>
73abe331 51#include <TObjString.h>
57f50b3c 52#include <TSQLServer.h>
53#include <TSQLResult.h>
54#include <TSQLRow.h>
cb343cfd 55#include <TMutex.h>
9827400b 56#include <TSystemDirectory.h>
57#include <TSystemFile.h>
a986b218 58#include <TFile.h>
9827400b 59#include <TGrid.h>
60#include <TGridResult.h>
503ff24f 61#include <TMap.h>
73abe331 62
e7f62f16 63#include <TMonaLisaWriter.h>
64
5164a766 65#include <fstream>
66
cb343cfd 67#include <sys/types.h>
68#include <sys/wait.h>
69
1bd183ba 70#include <signal.h>
3ff44dcb 71using namespace std;
1bd183ba 72
73abe331 73ClassImp(AliShuttle)
74
b948db8d 75//______________________________________________________________________________________________
76AliShuttle::AliShuttle(const AliShuttleConfig* config,
77 UInt_t timeout, Int_t retries):
4f0ab988 78fConfig(config),
79fTimeout(timeout), fRetries(retries),
80fPreprocessorMap(),
2bb7b766 81fLogbookEntry(0),
eba76848 82fCurrentDetector(),
926cbc0e 83fFirstProcessing(0),
339adafa 84fFXSError(-1),
85a80aa9 85fStatusEntry(0),
cb343cfd 86fMonitoringMutex(0),
eba76848 87fLastActionTime(0),
e7f62f16 88fLastAction(),
9827400b 89fMonaLisa(0),
90fTestMode(kNone),
ffa29e93 91fReadTestMode(kFALSE),
92fOutputRedirected(kFALSE)
73abe331 93{
94 //
95 // config: AliShuttleConfig used
73abe331 96 // timeout: timeout used for AliDCSClient connection
97 // retries: the number of retries in case of connection error.
98 //
99
57f50b3c 100 if (!fConfig->IsValid()) AliFatal("********** !!!!! Invalid configuration !!!!! **********");
efec19bd 101 for(int iSys=0;iSys<5;iSys++) {
57f50b3c 102 fServer[iSys]=0;
efec19bd 103 if (iSys < 4)
2c15234c 104 fFXSlist[iSys].SetOwner(kTRUE);
57f50b3c 105 }
2bb7b766 106 fPreprocessorMap.SetOwner(kTRUE);
be48e3ea 107
108 for (UInt_t iDet=0; iDet<NDetectors(); iDet++)
109 fFirstUnprocessed[iDet] = kFALSE;
110
cb343cfd 111 fMonitoringMutex = new TMutex();
58bc3020 112}
113
b948db8d 114//______________________________________________________________________________________________
57f50b3c 115AliShuttle::~AliShuttle()
58bc3020 116{
9827400b 117 //
118 // destructor
119 //
58bc3020 120
b948db8d 121 fPreprocessorMap.DeleteAll();
efec19bd 122 for(int iSys=0;iSys<5;iSys++)
57f50b3c 123 if(fServer[iSys]) {
124 fServer[iSys]->Close();
125 delete fServer[iSys];
eba76848 126 fServer[iSys] = 0;
57f50b3c 127 }
2bb7b766 128
129 if (fStatusEntry){
130 delete fStatusEntry;
131 fStatusEntry = 0;
132 }
cb343cfd 133
134 if (fMonitoringMutex)
135 {
136 delete fMonitoringMutex;
137 fMonitoringMutex = 0;
138 }
73abe331 139}
140
b948db8d 141//______________________________________________________________________________________________
57f50b3c 142void AliShuttle::RegisterPreprocessor(AliPreprocessor* preprocessor)
58bc3020 143{
73abe331 144 //
b948db8d 145 // Registers new AliPreprocessor.
73abe331 146 // It uses GetName() for indentificator of the pre processor.
147 // The pre processor is registered it there isn't any other
148 // with the same identificator (GetName()).
149 //
150
eba76848 151 const char* detName = preprocessor->GetName();
152 if(GetDetPos(detName) < 0)
153 AliFatal(Form("********** !!!!! Invalid detector name: %s !!!!! **********", detName));
154
155 if (fPreprocessorMap.GetValue(detName)) {
156 AliWarning(Form("AliPreprocessor %s is already registered!", detName));
73abe331 157 return;
158 }
159
eba76848 160 fPreprocessorMap.Add(new TObjString(detName), preprocessor);
73abe331 161}
b948db8d 162//______________________________________________________________________________________________
3301427a 163Bool_t AliShuttle::Store(const AliCDBPath& path, TObject* object,
84090f85 164 AliCDBMetaData* metaData, Int_t validityStart, Bool_t validityInfinite)
73abe331 165{
9827400b 166 // Stores a CDB object in the storage for offline reconstruction. Objects that are not needed for
167 // offline reconstruction, but should be stored anyway (e.g. for debugging) should NOT be stored
168 // using this function. Use StoreReferenceData instead!
169 // It calls StoreLocally function which temporarily stores the data locally; when the preprocessor
170 // finishes the data are transferred to the main storage (Grid).
b948db8d 171
3301427a 172 return StoreLocally(fgkLocalCDB, path, object, metaData, validityStart, validityInfinite);
84090f85 173}
174
175//______________________________________________________________________________________________
3301427a 176Bool_t AliShuttle::StoreReferenceData(const AliCDBPath& path, TObject* object, AliCDBMetaData* metaData)
84090f85 177{
9827400b 178 // Stores a CDB object in the storage for reference data. This objects will not be available during
179 // offline reconstrunction. Use this function for reference data only!
180 // It calls StoreLocally function which temporarily stores the data locally; when the preprocessor
181 // finishes the data are transferred to the main storage (Grid).
85a80aa9 182
3301427a 183 return StoreLocally(fgkLocalRefStorage, path, object, metaData);
85a80aa9 184}
185
186//______________________________________________________________________________________________
3301427a 187Bool_t AliShuttle::StoreLocally(const TString& localUri,
85a80aa9 188 const AliCDBPath& path, TObject* object, AliCDBMetaData* metaData,
189 Int_t validityStart, Bool_t validityInfinite)
190{
9827400b 191 // Store object temporarily in local storage. Parameters are passed by Store and StoreReferenceData functions.
192 // when the preprocessor finishes the data are transferred to the main storage (Grid).
193 // The parameters are:
194 // 1) Uri of the backup storage (Local)
195 // 2) the object's path.
196 // 3) the object to be stored
197 // 4) the metaData to be associated with the object
198 // 5) the validity start run number w.r.t. the current run,
199 // if the data is valid only for this run leave the default 0
200 // 6) specifies if the calibration data is valid for infinity (this means until updated),
201 // typical for calibration runs, the default is kFALSE
202 //
203 // returns 0 if fail, 1 otherwise
84090f85 204
924fb6e9 205
9827400b 206 if (fTestMode & kErrorStorage)
207 {
208 Log(fCurrentDetector, "StoreLocally - In TESTMODE - Simulating error while storing locally");
209 return kFALSE;
210 }
211
3301427a 212 const char* cdbType = (localUri == fgkLocalCDB) ? "CDB" : "Reference";
2bb7b766 213
85a80aa9 214 Int_t firstRun = GetCurrentRun() - validityStart;
84090f85 215 if(firstRun < 0) {
9827400b 216 AliWarning("First valid run happens to be less than 0! Setting it to 0.");
84090f85 217 firstRun=0;
218 }
219
220 Int_t lastRun = -1;
221 if(validityInfinite) {
222 lastRun = AliCDBRunRange::Infinity();
223 } else {
224 lastRun = GetCurrentRun();
225 }
226
3301427a 227 // Version is set to current run, it will be used later to transfer data to Grid
228 AliCDBId id(path, firstRun, lastRun, GetCurrentRun(), -1);
2bb7b766 229
230 if(! dynamic_cast<TObjString*> (metaData->GetProperty("RunUsed(TObjString)"))){
231 TObjString runUsed = Form("%d", GetCurrentRun());
9e080f92 232 metaData->SetProperty("RunUsed(TObjString)", runUsed.Clone());
2bb7b766 233 }
84090f85 234
3301427a 235 Bool_t result = kFALSE;
84090f85 236
3301427a 237 if (!(AliCDBManager::Instance()->GetStorage(localUri))) {
238 Log("SHUTTLE", Form("StoreLocally - Cannot activate local %s storage", cdbType));
84090f85 239 } else {
924fb6e9 240 Int_t logLevel = AliLog::GetGlobalLogLevel();
241 AliLog::SetGlobalLogLevel(AliLog::kError);
3301427a 242 result = AliCDBManager::Instance()->GetStorage(localUri)
84090f85 243 ->Put(object, id, metaData);
924fb6e9 244 AliLog::SetGlobalLogLevel((AliLog::EType_t)logLevel);
84090f85 245 }
246
247 if(!result) {
248
9827400b 249 Log(fCurrentDetector, Form("StoreLocally - Can't store object <%s>!", id.ToString().Data()));
3301427a 250 }
2bb7b766 251
924fb6e9 252
3301427a 253 return result;
254}
84090f85 255
3301427a 256//______________________________________________________________________________________________
257Bool_t AliShuttle::StoreOCDB()
258{
9827400b 259 //
260 // Called when preprocessor ends successfully or when previous storage attempt failed (kStoreError status)
261 // Calls underlying StoreOCDB(const char*) function twice, for OCDB and Reference storage.
262 // Then calls StoreRefFilesToGrid to store reference files.
263 //
264
5e993b6f 265 UpdateShuttleStatus(AliShuttleStatus::kStoreStarted);
56d3e9e4 266
9827400b 267 if (fTestMode & kErrorGrid)
268 {
269 Log("SHUTTLE", "StoreOCDB - In TESTMODE - Simulating error while storing in the Grid");
270 Log(fCurrentDetector, "StoreOCDB - In TESTMODE - Simulating error while storing in the Grid");
271 return kFALSE;
272 }
273
c88ad5db 274 Log("SHUTTLE","StoreOCDB - Storing OCDB data ...");
5e993b6f 275 Int_t resultCDB = StoreOCDB(fgkMainCDB);
86aa42c3 276
c88ad5db 277 Log("SHUTTLE","StoreOCDB - Storing reference data ...");
5e993b6f 278 Int_t resultRef = StoreOCDB(fgkMainRefStorage);
9827400b 279
c88ad5db 280 Log("SHUTTLE","StoreOCDB - Storing reference files ...");
281 Bool_t resultRefFiles = CopyFilesToGrid("reference");
282
283 Bool_t resultMetadata = kTRUE;
284 if(fCurrentDetector == "GRP")
285 {
7802e884 286 Log("SHUTTLE","StoreOCDB - Storing Run Metadata file ...");
c88ad5db 287 resultMetadata = CopyFilesToGrid("metadata");
288 }
9827400b 289
5e993b6f 290 Int_t storeResult = 0;
291
292 if (resultCDB < 0 || resultRef < 0 || resultRefFiles == kFALSE || resultMetadata == kFALSE)
293 storeResult = -1;
294 else if (resultCDB > 0 || resultRef > 0)
295 storeResult = 1;
296
297 if (storeResult < 0)
298 {
299 Log("SHUTTLE",
300 Form("\t\t\t****** run %d - %s: STORAGE ERROR ******",
301 GetCurrentRun(), fCurrentDetector.Data()));
302 UpdateShuttleStatus(AliShuttleStatus::kStoreError);
303 }
304 else if (storeResult > 0)
305 {
306 Log("SHUTTLE",
307 Form("\t\t\t****** run %d - %s: STORAGE DELAYED ******",
308 GetCurrentRun(), fCurrentDetector.Data()));
309 UpdateShuttleStatus(AliShuttleStatus::kStoreDelayed);
310 }
311 else if (storeResult == 0)
312 {
313 Log("SHUTTLE",
314 Form("\t\t\t****** run %d - %s: DONE ******",
315 GetCurrentRun(), fCurrentDetector.Data()));
316 UpdateShuttleStatus(AliShuttleStatus::kDone);
317 UpdateShuttleLogbook(fCurrentDetector, "DONE");
318 }
319
320 return (storeResult == 0);
3301427a 321}
322
323//______________________________________________________________________________________________
5e993b6f 324Int_t AliShuttle::StoreOCDB(const TString& gridURI)
3301427a 325{
326 //
327 // Called by StoreOCDB(), performs actual storage to the main OCDB and reference storages (Grid)
328 //
5e993b6f 329 // Return code:
330 // -2 initialization error
331 // -1 storage error
332 // 0 success
333 // 1 storage delayed (e.g. previous unprocessed runs)
334 //
3301427a 335
336 TObjArray* gridIds=0;
337
338 Bool_t result = kTRUE;
5e993b6f 339 Bool_t delayed = kFALSE;
3301427a 340
341 const char* type = 0;
342 TString localURI;
343 if(gridURI == fgkMainCDB) {
344 type = "OCDB";
345 localURI = fgkLocalCDB;
346 } else if(gridURI == fgkMainRefStorage) {
347 type = "reference";
348 localURI = fgkLocalRefStorage;
349 } else {
350 AliError(Form("Invalid storage URI: %s", gridURI.Data()));
5e993b6f 351 return -2;
3301427a 352 }
353
354 AliCDBManager* man = AliCDBManager::Instance();
355
356 AliCDBStorage *gridSto = man->GetStorage(gridURI);
357 if(!gridSto) {
358 Log("SHUTTLE",
359 Form("StoreOCDB - cannot activate main %s storage", type));
5e993b6f 360 return -2;
3301427a 361 }
c4918594 362 gridSto->SetMirrorSEs(fgkMirrorSEs.Data());
3301427a 363
364 gridIds = gridSto->GetQueryCDBList();
365
366 // get objects previously stored in local CDB
367 AliCDBStorage *localSto = man->GetStorage(localURI);
368 if(!localSto) {
369 Log("SHUTTLE",
370 Form("StoreOCDB - cannot activate local %s storage", type));
5e993b6f 371 return -2;
3301427a 372 }
373 AliCDBPath aPath(GetOfflineDetName(fCurrentDetector.Data()),"*","*");
374 // Local objects were stored with current run as Grid version!
375 TList* localEntries = localSto->GetAll(aPath.GetPath(), GetCurrentRun(), GetCurrentRun());
376 localEntries->SetOwner(1);
377
378 // loop on local stored objects
379 TIter localIter(localEntries);
380 AliCDBEntry *aLocEntry = 0;
381 while((aLocEntry = dynamic_cast<AliCDBEntry*> (localIter.Next()))){
382 aLocEntry->SetOwner(1);
383 AliCDBId aLocId = aLocEntry->GetId();
384 aLocEntry->SetVersion(-1);
385 aLocEntry->SetSubVersion(-1);
6c0c9948 386
387 Log(fCurrentDetector.Data(), Form("Attempting to store %s", aLocId.ToString().Data()));
3301427a 388
389 // If local object is valid up to infinity we store it only if it is
390 // the first unprocessed run!
391 if (aLocId.GetLastRun() == AliCDBRunRange::Infinity() &&
392 !fFirstUnprocessed[GetDetPos(fCurrentDetector)])
393 {
394 Log("SHUTTLE", Form("StoreOCDB - %s: object %s has validity infinite but "
395 "there are previous unprocessed runs!",
396 fCurrentDetector.Data(), aLocId.GetPath().Data()));
5e993b6f 397 Log(fCurrentDetector.Data(), Form("StoreOCDB - %s: object %s has validity infinite but "
398 "there are previous unprocessed runs!",
399 fCurrentDetector.Data(), aLocId.GetPath().Data()));
400 delayed = kTRUE;
3301427a 401 continue;
402 }
403
404 // loop on Grid valid Id's
405 Bool_t store = kTRUE;
406 TIter gridIter(gridIds);
407 AliCDBId* aGridId = 0;
6c0c9948 408 while ((aGridId = dynamic_cast<AliCDBId*> (gridIter.Next()))) {
409 if (aGridId->GetPath() != aLocId.GetPath())
410 continue;
3301427a 411 // skip all objects valid up to infinity
6c0c9948 412 if (aGridId->GetLastRun() == AliCDBRunRange::Infinity())
413 continue;
414
3301427a 415 // if we get here, it means there's already some more recent object stored on Grid!
6c0c9948 416 Log(fCurrentDetector.Data(),
417 Form("StoreOCDB - A more recent object already exists in %s storage: <%s>",
418 type, aGridId->ToString().Data()));
419
3301427a 420 store = kFALSE;
421 break;
422 }
423
6c0c9948 424 Bool_t storeOk = kFALSE;
425 if (store)
426 {
427 Log(fCurrentDetector.Data(), Form("Prechecks succeeded. Ready to store %s", aLocId.ToString().Data()));
428 storeOk = gridSto->Put(aLocEntry);
429 if (storeOk) {
3301427a 430 Log("SHUTTLE",
6c0c9948 431 Form("StoreOCDB - Object <%s> successfully put into %s storage",
432 aLocId.ToString().Data(), type));
2d9019b4 433 Log(fCurrentDetector.Data(),
434 Form("StoreOCDB - Object <%s> successfully put into %s storage",
6c0c9948 435 aLocId.ToString().Data(), type));
436 } else {
437 Log("SHUTTLE",
438 Form("StoreOCDB - Grid %s storage of object <%s> failed",
439 type, aLocId.ToString().Data()));
440 Log(fCurrentDetector.Data(),
441 Form("StoreOCDB - Grid %s storage of object <%s> failed",
442 type, aLocId.ToString().Data()));
443 result = kFALSE;
3301427a 444 }
6c0c9948 445 }
446
447 if (!store || storeOk) {
448 // removing local file...
3301427a 449 TString filename;
450 localSto->IdToFilename(aLocId, filename);
c88ad5db 451 Log("SHUTTLE", Form("StoreOCDB - Removing local file %s", filename.Data()));
3301427a 452 RemoveFile(filename.Data());
b948db8d 453 }
454 }
3301427a 455 localEntries->Clear();
2bb7b766 456
5e993b6f 457 Int_t returnCode = 0;
458
459 if (result == kFALSE)
460 returnCode = -1;
461 else if (delayed != kFALSE)
462 returnCode = 1;
463
464 Log("SHUTTLE", Form("StoreOCDB - Returning with %d (result = %d, delayed = %d)", returnCode, result, delayed));
465 Log(fCurrentDetector.Data(), Form("StoreOCDB - Returning with %d (result = %d, delayed = %d)", returnCode, result, delayed));
466
467 return returnCode;
3301427a 468}
469
546242fb 470//______________________________________________________________________________________________
471Bool_t AliShuttle::CleanReferenceStorage(const char* detector)
472{
2d9019b4 473 // clears the directory used to store reference files of a given subdetector
546242fb 474
475 AliCDBManager* man = AliCDBManager::Instance();
476 AliCDBStorage* sto = man->GetStorage(fgkLocalRefStorage);
2d9019b4 477 TString localBaseFolder = sto->GetBaseFolder();
478
479 TString targetDir = GetRefFilePrefix(localBaseFolder.Data(), detector);
480
d524ade6 481 Log("SHUTTLE", Form("CleanReferenceStorage - Cleaning %s", targetDir.Data()));
2d9019b4 482
483 TString begin;
484 begin.Form("%d_", GetCurrentRun());
485
486 TSystemDirectory* baseDir = new TSystemDirectory("/", targetDir);
487 if (!baseDir)
488 return kTRUE;
489
490 TList* dirList = baseDir->GetListOfFiles();
491 delete baseDir;
492
493 if (!dirList) return kTRUE;
494
efec19bd 495 if (dirList->GetEntries() < 3) // to be changed to 4?
2d9019b4 496 {
497 delete dirList;
498 return kTRUE;
499 }
500
501 Int_t nDirs = 0, nDel = 0;
502 TIter dirIter(dirList);
503 TSystemFile* entry = 0;
546242fb 504
2d9019b4 505 Bool_t success = kTRUE;
546242fb 506
2d9019b4 507 while ((entry = dynamic_cast<TSystemFile*> (dirIter.Next())))
508 {
509 if (entry->IsDirectory())
510 continue;
511
512 TString fileName(entry->GetName());
513 if (!fileName.BeginsWith(begin))
514 continue;
515
516 nDirs++;
517
518 // delete file
519 Int_t result = gSystem->Unlink(fileName.Data());
520
521 if (result)
522 {
d524ade6 523 Log("SHUTTLE", Form("CleanReferenceStorage - Could not delete file %s!", fileName.Data()));
2d9019b4 524 success = kFALSE;
525 } else {
526 nDel++;
527 }
528 }
529
530 if(nDirs > 0)
531 Log("SHUTTLE", Form("CleanReferenceStorage - %d (over %d) reference files in folder %s were deleted.",
532 nDel, nDirs, targetDir.Data()));
533
534
535 delete dirList;
536 return success;
537
538
539
540
541
546242fb 542
543 Int_t result = gSystem->GetPathInfo(targetDir, 0, (Long64_t*) 0, 0, 0);
544 if (result == 0)
545 {
546 // delete directory
d524ade6 547 result = gSystem->Exec(Form("rm -rf %s", targetDir.Data()));
546242fb 548 if (result != 0)
549 {
d524ade6 550 Log("SHUTTLE", Form("CleanReferenceStorage - Could not clean directory %s", targetDir.Data()));
546242fb 551 return kFALSE;
552 }
553 }
554
555 result = gSystem->mkdir(targetDir, kTRUE);
556 if (result != 0)
557 {
c88ad5db 558 Log("SHUTTLE", Form("CleanReferenceStorage - Error creating base directory %s", targetDir.Data()));
546242fb 559 return kFALSE;
560 }
561
562 return kTRUE;
563}
564
9827400b 565//______________________________________________________________________________________________
566Bool_t AliShuttle::StoreReferenceFile(const char* detector, const char* localFile, const char* gridFileName)
567{
568 //
3c2a21c8 569 // Stores reference file directly (without opening it). This function stores the file locally.
9827400b 570 //
3c2a21c8 571 // The file is stored under the following location:
572 // <base folder of local reference storage>/<DET>/<RUN#>_<gridFileName>
573 // where <gridFileName> is the second parameter given to the function
574 //
9827400b 575
576 if (fTestMode & kErrorStorage)
577 {
578 Log(fCurrentDetector, "StoreReferenceFile - In TESTMODE - Simulating error while storing locally");
579 return kFALSE;
580 }
581
582 AliCDBManager* man = AliCDBManager::Instance();
583 AliCDBStorage* sto = man->GetStorage(fgkLocalRefStorage);
584
585 TString localBaseFolder = sto->GetBaseFolder();
586
d524ade6 587 TString target = GetRefFilePrefix(localBaseFolder.Data(), detector);
588 target.Append(Form("/%d_%s", GetCurrentRun(), gridFileName));
9827400b 589
d524ade6 590 return CopyFileLocally(localFile, target);
c88ad5db 591}
592
593//______________________________________________________________________________________________
594Bool_t AliShuttle::StoreRunMetadataFile(const char* localFile, const char* gridFileName)
595{
596 //
597 // Stores Run metadata file to the Grid, in the run folder
598 //
599 // Only GRP can call this function.
600
601 if (fTestMode & kErrorStorage)
602 {
603 Log(fCurrentDetector, "StoreRunMetaDataFile - In TESTMODE - Simulating error while storing locally");
604 return kFALSE;
605 }
606
607 AliCDBManager* man = AliCDBManager::Instance();
608 AliCDBStorage* sto = man->GetStorage(fgkLocalRefStorage);
609
610 TString localBaseFolder = sto->GetBaseFolder();
611
612 // Build Run level folder
955d2abc 613 // folder = /alice/data/year/lhcPeriod/runNb/raw
c88ad5db 614
c88ad5db 615
675f64cd 616 TString lhcPeriod = GetLHCPeriod();
c88ad5db 617 if (lhcPeriod.Length() == 0)
618 {
619 Log("SHUTTLE","StoreRunMetaDataFile - LHCPeriod not found in logbook!");
620 return 0;
621 }
a1c5723b 622
623 // TODO partitions with one detector only write data into LHCperiod_DET
2694f253 624 TString partition = GetRunParameter("detector");
a1c5723b 625
626 if (partition.Length() > 0 && partition != "ALICE")
627 {
628 lhcPeriod.Append(Form("_%s", partition.Data()));
629 Log(fCurrentDetector, Form("Run data tags merged file will be written in %s",
630 lhcPeriod.Data()));
631 }
8884d71e 632
b8cf5e0f 633 TString target = Form("%s/GRP/RunMetadata%s%d/%s/%09d/raw/%s",
634 localBaseFolder.Data(), fConfig->GetAlienPath(), GetCurrentYear(),
d524ade6 635 lhcPeriod.Data(), GetCurrentRun(), gridFileName);
c88ad5db 636
d524ade6 637 return CopyFileLocally(localFile, target);
c88ad5db 638}
639
640//______________________________________________________________________________________________
d524ade6 641Bool_t AliShuttle::CopyFileLocally(const char* localFile, const TString& target)
c88ad5db 642{
643 //
644 // Stores file locally. Called by StoreReferenceFile and StoreRunMetadataFile
d524ade6 645 // Files are temporarily stored in the local reference storage. When the preprocessor
646 // finishes, the Shuttle calls CopyFilesToGrid to transfer the files to AliEn
647 // (in reference or run level folders)
c88ad5db 648 //
649
d524ade6 650 TString targetDir(target(0, target.Last('/')));
651
652 //try to open base dir folder, if it does not exist
2d9019b4 653 void* dir = gSystem->OpenDirectory(targetDir.Data());
654 if (dir == NULL) {
655 if (gSystem->mkdir(targetDir.Data(), kTRUE)) {
a1c5723b 656 Log("SHUTTLE", Form("CopyFileLocally - Can't open directory <%s>", targetDir.Data()));
2d9019b4 657 return kFALSE;
658 }
659
660 } else {
661 gSystem->FreeDirectory(dir);
662 }
9827400b 663
7d43a416 664 Int_t result = 0;
665
666 result = gSystem->GetPathInfo(localFile, 0, (Long64_t*) 0, 0, 0);
9827400b 667 if (result)
668 {
a1c5723b 669 Log("SHUTTLE", Form("CopyFileLocally - %s does not exist", localFile));
546242fb 670 return kFALSE;
9827400b 671 }
546242fb 672
7d43a416 673 result = gSystem->GetPathInfo(target, 0, (Long64_t*) 0, 0, 0);
674 if (!result)
675 {
a1c5723b 676 Log("SHUTTLE", Form("CopyFileLocally - target file %s already exist, removing...", target.Data()));
7d43a416 677 if (gSystem->Unlink(target.Data()))
678 {
a1c5723b 679 Log("SHUTTLE", Form("CopyFileLocally - Could not remove existing target file %s!", target.Data()));
7d43a416 680 return kFALSE;
681 }
682 }
683
9827400b 684 result = gSystem->CopyFile(localFile, target);
685
686 if (result == 0)
687 {
a1c5723b 688 Log("SHUTTLE", Form("CopyFileLocally - File %s stored locally to %s", localFile, target.Data()));
9827400b 689 return kTRUE;
690 }
691 else
692 {
a1c5723b 693 Log("SHUTTLE", Form("CopyFileLocally - Could not store file %s to %s! Error code = %d",
546242fb 694 localFile, target.Data(), result));
9827400b 695 return kFALSE;
696 }
c88ad5db 697
698
699
9827400b 700}
701
702//______________________________________________________________________________________________
c88ad5db 703Bool_t AliShuttle::CopyFilesToGrid(const char* type)
9827400b 704{
705 //
c88ad5db 706 // Transfers local files to the Grid. Local files can be reference files
707 // or run metadata file (from GRP only).
9827400b 708 //
c88ad5db 709 // According to the type (ref, metadata) the files are stored under the following location:
710 // ref --> <base folder of reference storage>/<DET>/<RUN#>_<gridFileName>
711 // metadata --> <run data folder>/<MetadataFileName>
86aa42c3 712 //
c88ad5db 713
9827400b 714 AliCDBManager* man = AliCDBManager::Instance();
715 AliCDBStorage* sto = man->GetStorage(fgkLocalRefStorage);
716 if (!sto)
717 return kFALSE;
718 TString localBaseFolder = sto->GetBaseFolder();
9827400b 719
c88ad5db 720 TString dir;
721 TString alienDir;
9827400b 722 TString begin;
9827400b 723
c88ad5db 724 if (strcmp(type, "reference") == 0)
725 {
726 dir = GetRefFilePrefix(localBaseFolder.Data(), fCurrentDetector.Data());
727 AliCDBStorage* gridSto = man->GetStorage(fgkMainRefStorage);
728 if (!gridSto)
729 return kFALSE;
730 TString gridBaseFolder = gridSto->GetBaseFolder();
731 alienDir = GetRefFilePrefix(gridBaseFolder.Data(), fCurrentDetector.Data());
732 begin = Form("%d_", GetCurrentRun());
733 }
734 else if (strcmp(type, "metadata") == 0)
735 {
c88ad5db 736
675f64cd 737 TString lhcPeriod = GetLHCPeriod();
c88ad5db 738
739 if (lhcPeriod.Length() == 0)
740 {
741 Log("SHUTTLE","CopyFilesToGrid - LHCPeriod not found in logbook!");
742 return 0;
743 }
744
a1c5723b 745 // TODO partitions with one detector only write data into LHCperiod_DET
2694f253 746 TString partition = GetRunParameter("detector");
a1c5723b 747
748 if (partition.Length() > 0 && partition != "ALICE")
749 {
750 lhcPeriod.Append(Form("_%s", partition.Data()));
751 }
752
b8cf5e0f 753 dir = Form("%s/GRP/RunMetadata%s%d/%s/%09d/raw",
754 localBaseFolder.Data(), fConfig->GetAlienPath(), GetCurrentYear(),
c88ad5db 755 lhcPeriod.Data(), GetCurrentRun());
b8cf5e0f 756 alienDir = dir(dir.Index(fConfig->GetAlienPath()), dir.Length());
675f64cd 757
c88ad5db 758 begin = "";
759 }
760 else
761 {
762 Log("SHUTTLE", "CopyFilesToGrid - Unexpected: type label must be reference or metadata!");
763 return kFALSE;
764 }
765
9827400b 766 TSystemDirectory* baseDir = new TSystemDirectory("/", dir);
3d8bc902 767 if (!baseDir)
768 return kTRUE;
769
2d9019b4 770 TList* dirList = baseDir->GetListOfFiles();
771 delete baseDir;
772
773 if (!dirList) return kTRUE;
774
775 if (dirList->GetEntries() < 3)
3d8bc902 776 {
2d9019b4 777 delete dirList;
9827400b 778 return kTRUE;
3d8bc902 779 }
2d9019b4 780
546242fb 781 if (!gGrid)
782 {
c88ad5db 783 Log("SHUTTLE", "CopyFilesToGrid - Connection to Grid failed: Cannot continue!");
2d9019b4 784 delete dirList;
546242fb 785 return kFALSE;
786 }
787
2d9019b4 788 Int_t nDirs = 0, nTransfer = 0;
789 TIter dirIter(dirList);
790 TSystemFile* entry = 0;
791
9827400b 792 Bool_t success = kTRUE;
3d8bc902 793 Bool_t first = kTRUE;
9827400b 794
2d9019b4 795 while ((entry = dynamic_cast<TSystemFile*> (dirIter.Next())))
796 {
9827400b 797 if (entry->IsDirectory())
798 continue;
799
800 TString fileName(entry->GetName());
801 if (!fileName.BeginsWith(begin))
802 continue;
803
2d9019b4 804 nDirs++;
805
3d8bc902 806 if (first)
807 {
808 first = kFALSE;
c88ad5db 809 // check that folder exists, otherwise create it
3d8bc902 810 TGridResult* result = gGrid->Ls(alienDir.Data(), "a");
811
812 if (!result)
2d9019b4 813 {
814 delete dirList;
3d8bc902 815 return kFALSE;
2d9019b4 816 }
3d8bc902 817
546242fb 818 if (!result->GetFileName(1)) // TODO: It looks like element 0 is always 0!!
3d8bc902 819 {
675f64cd 820 // TODO It does not work currently! Bug in TAliEn::Mkdir
821 // TODO Manually fixed in local root v5-16-00
c88ad5db 822 if (!gGrid->Mkdir(alienDir.Data(),"-p",0))
3d8bc902 823 {
c88ad5db 824 Log("SHUTTLE", Form("CopyFilesToGrid - Cannot create directory %s",
3d8bc902 825 alienDir.Data()));
2d9019b4 826 delete dirList;
3d8bc902 827 return kFALSE;
546242fb 828 } else {
c88ad5db 829 Log("SHUTTLE",Form("CopyFilesToGrid - Folder %s created", alienDir.Data()));
3d8bc902 830 }
831
546242fb 832 } else {
c88ad5db 833 Log("SHUTTLE",Form("CopyFilesToGrid - Folder %s found", alienDir.Data()));
3d8bc902 834 }
835 }
836
9827400b 837 TString fullLocalPath;
838 fullLocalPath.Form("%s/%s", dir.Data(), fileName.Data());
839
840 TString fullGridPath;
841 fullGridPath.Form("alien://%s/%s", alienDir.Data(), fileName.Data());
842
a986b218 843 Bool_t result = TFile::Cp(fullLocalPath, fullGridPath);
9827400b 844
845 if (result)
846 {
c88ad5db 847 Log("SHUTTLE", Form("CopyFilesToGrid - Copying local file %s to %s succeeded!",
848 fullLocalPath.Data(), fullGridPath.Data()));
9827400b 849 RemoveFile(fullLocalPath);
2d9019b4 850 nTransfer++;
9827400b 851 }
852 else
853 {
c88ad5db 854 Log("SHUTTLE", Form("CopyFilesToGrid - Copying local file %s to %s FAILED!",
855 fullLocalPath.Data(), fullGridPath.Data()));
9827400b 856 success = kFALSE;
857 }
858 }
2d9019b4 859
c88ad5db 860 Log("SHUTTLE", Form("CopyFilesToGrid - %d (over %d) files in folder %s copied to Grid.",
861 nTransfer, nDirs, dir.Data()));
2d9019b4 862
863
864 delete dirList;
9827400b 865 return success;
866}
867
2d9019b4 868//______________________________________________________________________________________________
869const char* AliShuttle::GetRefFilePrefix(const char* base, const char* detector)
870{
871 //
872 // Get folder name of reference files
873 //
874
875 TString offDetStr(GetOfflineDetName(detector));
3dd835d8 876 static TString dir;
2d9019b4 877 if (offDetStr == "ITS" || offDetStr == "MUON" || offDetStr == "PHOS")
878 {
879 dir.Form("%s/%s/%s", base, offDetStr.Data(), detector);
880 } else {
881 dir.Form("%s/%s", base, offDetStr.Data());
882 }
883
884 return dir.Data();
2d9019b4 885}
c88ad5db 886
3301427a 887//______________________________________________________________________________________________
888void AliShuttle::CleanLocalStorage(const TString& uri)
889{
9827400b 890 //
891 // Called in case the preprocessor is declared failed. Remove remaining objects from the local storages.
892 //
3301427a 893
894 const char* type = 0;
895 if(uri == fgkLocalCDB) {
896 type = "OCDB";
897 } else if(uri == fgkLocalRefStorage) {
546242fb 898 type = "Reference";
3301427a 899 } else {
900 AliError(Form("Invalid storage URI: %s", uri.Data()));
901 return;
902 }
903
904 AliCDBManager* man = AliCDBManager::Instance();
b948db8d 905
3301427a 906 // open local storage
907 AliCDBStorage *localSto = man->GetStorage(uri);
908 if(!localSto) {
909 Log("SHUTTLE",
910 Form("CleanLocalStorage - cannot activate local %s storage", type));
911 return;
912 }
913
914 TString filename(Form("%s/%s/*/Run*_v%d_s*.root",
546242fb 915 localSto->GetBaseFolder().Data(), GetOfflineDetName(fCurrentDetector.Data()), GetCurrentRun()));
3301427a 916
c88ad5db 917 AliDebug(2, Form("filename = %s", filename.Data()));
3301427a 918
c88ad5db 919 Log("SHUTTLE", Form("Removing remaining local files for run %d and detector %s ...",
3301427a 920 GetCurrentRun(), fCurrentDetector.Data()));
921
922 RemoveFile(filename.Data());
923
924}
925
926//______________________________________________________________________________________________
927void AliShuttle::RemoveFile(const char* filename)
928{
9827400b 929 //
930 // removes local file
931 //
3301427a 932
933 TString command(Form("rm -f %s", filename));
934
935 Int_t result = gSystem->Exec(command.Data());
936 if(result != 0)
937 {
938 Log("SHUTTLE", Form("RemoveFile - %s: Cannot remove file %s!",
939 fCurrentDetector.Data(), filename));
940 }
73abe331 941}
942
b948db8d 943//______________________________________________________________________________________________
5164a766 944AliShuttleStatus* AliShuttle::ReadShuttleStatus()
945{
9827400b 946 //
947 // Reads the AliShuttleStatus from the CDB
948 //
5164a766 949
2bb7b766 950 if (fStatusEntry){
951 delete fStatusEntry;
952 fStatusEntry = 0;
953 }
5164a766 954
3ebbf3c4 955 Int_t path1 = GetCurrentRun()/10000;
e6608c2e 956 try{
957 fStatusEntry = AliCDBManager::Instance()->GetStorage(GetLocalCDB())
958 ->Get(Form("/SHUTTLE/%s/%d", fCurrentDetector.Data(), path1), GetCurrentRun());
959 } catch(std::exception& x) {
e3b6a6e9 960 AliInfo(TString::Format("%s",x.what()));
e6608c2e 961 }
5164a766 962
2bb7b766 963 if (!fStatusEntry) return 0;
964 fStatusEntry->SetOwner(1);
5164a766 965
2bb7b766 966 AliShuttleStatus* status = dynamic_cast<AliShuttleStatus*> (fStatusEntry->GetObject());
967 if (!status) {
968 AliError("Invalid object stored to CDB!");
969 return 0;
970 }
5164a766 971
2bb7b766 972 return status;
5164a766 973}
974
975//______________________________________________________________________________________________
7bfb2090 976Bool_t AliShuttle::WriteShuttleStatus(AliShuttleStatus* status)
5164a766 977{
9827400b 978 //
979 // writes the status for one subdetector
980 //
2bb7b766 981
982 if (fStatusEntry){
983 delete fStatusEntry;
984 fStatusEntry = 0;
985 }
5164a766 986
2bb7b766 987 Int_t run = GetCurrentRun();
3ebbf3c4 988 Int_t path1 = run/10000;
989 TString path1_string = Form("%d",path1);
5164a766 990
3ebbf3c4 991 AliCDBId id(AliCDBPath("SHUTTLE", fCurrentDetector, path1_string), run, run);
5164a766 992
2bb7b766 993 fStatusEntry = new AliCDBEntry(status, id, new AliCDBMetaData);
994 fStatusEntry->SetOwner(1);
5164a766 995
924fb6e9 996 Int_t logLevel = AliLog::GetGlobalLogLevel();
997 AliLog::SetGlobalLogLevel(AliLog::kError);
998
2bb7b766 999 UInt_t result = AliCDBManager::Instance()->GetStorage(fgkLocalCDB)->Put(fStatusEntry);
7bfb2090 1000
924fb6e9 1001 AliLog::SetGlobalLogLevel((AliLog::EType_t)logLevel);
1002
1003 if (!result) {
3301427a 1004 Log("SHUTTLE", Form("WriteShuttleStatus - Failed for %s, run %d",
1005 fCurrentDetector.Data(), run));
2bb7b766 1006 return kFALSE;
1007 }
e7f62f16 1008
8e828d39 1009 SendMLDetInfo();
7bfb2090 1010
2bb7b766 1011 return kTRUE;
5164a766 1012}
1013
1014//______________________________________________________________________________________________
1015void AliShuttle::UpdateShuttleStatus(AliShuttleStatus::Status newStatus, Bool_t increaseCount)
1016{
9827400b 1017 //
1018 // changes the AliShuttleStatus for the given detector and run to the given status
1019 //
5164a766 1020
2bb7b766 1021 if (!fStatusEntry){
1022 AliError("UNEXPECTED: fStatusEntry empty");
1023 return;
1024 }
5164a766 1025
2bb7b766 1026 AliShuttleStatus* status = dynamic_cast<AliShuttleStatus*> (fStatusEntry->GetObject());
5164a766 1027
2bb7b766 1028 if (!status){
c88ad5db 1029 Log("SHUTTLE", "UpdateShuttleStatus - UNEXPECTED: status could not be read from current CDB entry");
2bb7b766 1030 return;
1031 }
5164a766 1032
2c15234c 1033 TString actionStr = Form("UpdateShuttleStatus - %s: Changing state from %s to %s",
eba76848 1034 fCurrentDetector.Data(),
36c99a6a 1035 status->GetStatusName(),
eba76848 1036 status->GetStatusName(newStatus));
cb343cfd 1037 Log("SHUTTLE", actionStr);
1038 SetLastAction(actionStr);
5164a766 1039
2bb7b766 1040 status->SetStatus(newStatus);
1041 if (increaseCount) status->IncreaseCount();
5164a766 1042
924fb6e9 1043 Int_t logLevel = AliLog::GetGlobalLogLevel();
1044 AliLog::SetGlobalLogLevel(AliLog::kError);
1045
2bb7b766 1046 AliCDBManager::Instance()->GetStorage(fgkLocalCDB)->Put(fStatusEntry);
e7f62f16 1047
924fb6e9 1048 AliLog::SetGlobalLogLevel((AliLog::EType_t)logLevel);
1049
8e828d39 1050 SendMLDetInfo();
5164a766 1051}
e7f62f16 1052
1053//______________________________________________________________________________________________
8e828d39 1054void AliShuttle::SendMLDetInfo()
e7f62f16 1055{
1056 //
1057 // sends ML information about the current status of the current detector being processed
1058 //
1059
1060 AliShuttleStatus* status = dynamic_cast<AliShuttleStatus*> (fStatusEntry->GetObject());
1061
1062 if (!status){
8e828d39 1063 Log("SHUTTLE", "SendMLDetInfo - UNEXPECTED: status could not be read from current CDB entry");
e7f62f16 1064 return;
1065 }
1066
1067 TMonaLisaText mlStatus(Form("%s_status", fCurrentDetector.Data()), status->GetStatusName());
1068 TMonaLisaValue mlRetryCount(Form("%s_count", fCurrentDetector.Data()), status->GetCount());
1069
1070 TList mlList;
1071 mlList.Add(&mlStatus);
1072 mlList.Add(&mlRetryCount);
1073
ee6f7523 1074 TString mlID;
1075 mlID.Form("%d", GetCurrentRun());
1076 fMonaLisa->SendParameters(&mlList, mlID);
e7f62f16 1077}
1078
5164a766 1079//______________________________________________________________________________________________
1080Bool_t AliShuttle::ContinueProcessing()
1081{
9827400b 1082 // this function reads the AliShuttleStatus information from CDB and
1083 // checks if the processing should be continued
1084 // if yes it returns kTRUE and updates the AliShuttleStatus with nextStatus
2bb7b766 1085
926cbc0e 1086 if (!fConfig->HostProcessDetector(fCurrentDetector))
1087 return kFALSE;
57c1a579 1088
1089 AliPreprocessor* aPreprocessor =
1090 dynamic_cast<AliPreprocessor*> (fPreprocessorMap.GetValue(fCurrentDetector));
1091 if (!aPreprocessor)
1092 {
e55e9b46 1093 Log("SHUTTLE", Form("ContinueProcessing - %s: no preprocessor registered", fCurrentDetector.Data()));
1094 return kFALSE;
57c1a579 1095 }
1096
2bb7b766 1097 AliShuttleLogbookEntry::Status entryStatus =
eba76848 1098 fLogbookEntry->GetDetectorStatus(fCurrentDetector);
2bb7b766 1099
5e993b6f 1100 if (entryStatus != AliShuttleLogbookEntry::kUnprocessed) {
c88ad5db 1101 Log("SHUTTLE", Form("ContinueProcessing - %s is %s",
2bb7b766 1102 fCurrentDetector.Data(),
1103 fLogbookEntry->GetDetectorStatusName(entryStatus)));
1104 return kFALSE;
1105 }
1106
1107 // if we get here, according to Shuttle logbook subdetector is in UNPROCESSED state
be48e3ea 1108
1109 // check if current run is first unprocessed run for current detector
1110 if (fConfig->StrictRunOrder(fCurrentDetector) &&
1111 !fFirstUnprocessed[GetDetPos(fCurrentDetector)])
1112 {
86aa42c3 1113 if (fTestMode == kNone)
1114 {
c88ad5db 1115 Log("SHUTTLE", Form("ContinueProcessing - %s requires strict run ordering"
0a0b087a 1116 " but this is not the first unprocessed run!",fCurrentDetector.Data()));
86aa42c3 1117 return kFALSE;
1118 }
1119 else
1120 {
c88ad5db 1121 Log("SHUTTLE", Form("ContinueProcessing - In TESTMODE - "
1122 "Although %s requires strict run ordering "
1123 "and this is not the first unprocessed run, "
0a0b087a 1124 "the SHUTTLE continues",fCurrentDetector.Data()));
86aa42c3 1125 }
be48e3ea 1126 }
1127
926cbc0e 1128 // Is the subdetector processed first time for this run?
1129 fFirstProcessing = kFALSE;
1130
2bb7b766 1131 AliShuttleStatus* status = ReadShuttleStatus();
1132 if (!status) {
1133 // first time
1134 Log("SHUTTLE", Form("ContinueProcessing - %s: Processing first time",
1135 fCurrentDetector.Data()));
1136 status = new AliShuttleStatus(AliShuttleStatus::kStarted);
926cbc0e 1137 fFirstProcessing = kTRUE;
2bb7b766 1138 return WriteShuttleStatus(status);
1139 }
1140
ffa25f20 1141 // The following case shouldn't happen if Shuttle Logbook was correctly updated.
2bb7b766 1142 // If it happens it may mean Logbook updating failed... let's do it now!
1143 if (status->GetStatus() == AliShuttleStatus::kDone ||
ffa25f20 1144 status->GetStatus() == AliShuttleStatus::kFailed ||
1145 status->GetStatus() == AliShuttleStatus::kSkipped) {
2bb7b766 1146 Log("SHUTTLE", Form("ContinueProcessing - %s is already %s. Updating Shuttle Logbook",
1147 fCurrentDetector.Data(),
1148 status->GetStatusName(status->GetStatus())));
ffa25f20 1149
1150 if (status->GetStatus() == AliShuttleStatus::kSkipped)
1151 {
1152 UpdateShuttleLogbook(fCurrentDetector.Data(), "DONE");
1153 }
1154 else
1155 UpdateShuttleLogbook(fCurrentDetector.Data(), status->GetStatusName(status->GetStatus()));
1156
2bb7b766 1157 return kFALSE;
1158 }
1159
5e993b6f 1160 if (status->GetStatus() == AliShuttleStatus::kStoreStarted || status->GetStatus() == AliShuttleStatus::kStoreDelayed ||status->GetStatus() == AliShuttleStatus::kStoreError) {
2bb7b766 1161 Log("SHUTTLE",
c88ad5db 1162 Form("ContinueProcessing - %s: Grid storage of one or more "
1163 "objects failed. Trying again now",
2bb7b766 1164 fCurrentDetector.Data()));
5e993b6f 1165 StoreOCDB();
2bb7b766 1166 return kFALSE;
1167 }
1168
1169 // if we get here, there is a restart
57c1a579 1170 Bool_t cont = kFALSE;
2bb7b766 1171
1172 // abort conditions
cb343cfd 1173 if (status->GetCount() >= fConfig->GetMaxRetries()) {
57c1a579 1174 Log("SHUTTLE", Form("ContinueProcessing - %s failed %d times in status %s - "
1175 "Updating Shuttle Logbook", fCurrentDetector.Data(),
2bb7b766 1176 status->GetCount(), status->GetStatusName()));
1177 UpdateShuttleLogbook(fCurrentDetector.Data(), "FAILED");
e7f62f16 1178 UpdateShuttleStatus(AliShuttleStatus::kFailed);
3301427a 1179
1180 // there may still be objects in local OCDB and reference storage
1181 // and FXS databases may be not updated: do it now!
9827400b 1182
1183 // TODO Currently disabled, we want to keep files in case of failure!
1184 // CleanLocalStorage(fgkLocalCDB);
1185 // CleanLocalStorage(fgkLocalRefStorage);
1186 // UpdateTableFailCase();
1187
1188 // Send mail to detector expert!
c88ad5db 1189 Log("SHUTTLE", Form("ContinueProcessing - Sending mail to %s expert...",
926cbc0e 1190 fCurrentDetector.Data()));
fe6a3257 1191 // det experts in to
1192 TString to="";
1193 TIter *iterExperts = 0;
1194 iterExperts = new TIter(fConfig->GetResponsibles(fCurrentDetector));
1195 TObjString *anExpert=0;
1196 while ((anExpert = (TObjString*) iterExperts->Next()))
1197 {
1198 to += Form("%s, \n", anExpert->GetName());
1199 }
1200 delete iterExperts;
1201
1202 if (to.Length() > 0)
1203 to.Remove(to.Length()-3);
1204 AliDebug(2, Form("to: %s",to.Data()));
1205
1206 if (to.IsNull()) {
1207 Log("SHUTTLE", Form("List of %s responsibles not set!", fCurrentDetector.Data()));
1208 return kFALSE;
1209 }
1210
1211 Log(fCurrentDetector.Data(), Form("ContinueProcessing - Sending mail to %s expert(s):",
1212 fCurrentDetector.Data()));
1213 Log(fCurrentDetector.Data(), Form("\n%s", to.Data()));
339adafa 1214 if (!SendMail(kPPEMail))
9827400b 1215 Log("SHUTTLE", Form("ContinueProcessing - Could not send mail to %s expert",
926cbc0e 1216 fCurrentDetector.Data()));
3301427a 1217
57c1a579 1218 } else {
1219 Log("SHUTTLE", Form("ContinueProcessing - %s: restarting. "
1220 "Aborted before with %s. Retry number %d.", fCurrentDetector.Data(),
1221 status->GetStatusName(), status->GetCount()));
9827400b 1222 Bool_t increaseCount = kTRUE;
c88ad5db 1223 if (status->GetStatus() == AliShuttleStatus::kDCSError ||
8da81210 1224 status->GetStatus() == AliShuttleStatus::kDCSStarted ||
5ca8f352 1225 status->GetStatus() == AliShuttleStatus::kFXSError ||
1226 status->GetStatus() == AliShuttleStatus::kOCDBError)
c88ad5db 1227 increaseCount = kFALSE;
675f64cd 1228
9827400b 1229 UpdateShuttleStatus(AliShuttleStatus::kStarted, increaseCount);
57c1a579 1230 cont = kTRUE;
2bb7b766 1231 }
1232
57c1a579 1233 return cont;
5164a766 1234}
1235
8e828d39 1236//______________________________________________________________________________________________
1237void AliShuttle::SendMLRunInfo(const char* status)
1238{
1239 //
1240 // Send information about this run to ML
1241
1242 TMonaLisaText mlStatus("SHUTTLE_status", status);
1243 TString runType(fLogbookEntry->GetRunType());
1244 if (strlen(fLogbookEntry->GetRunParameter("log")) > 0){
1245
1246 runType += "(";
1247 runType += fLogbookEntry->GetRunParameter("log");
1248 runType += ")";
1249 }
728290a4 1250 if (fLogbookEntry->GetDATestMode()){
1251 runType += " (DATest)";
1252 }
8e828d39 1253 TMonaLisaText mlRunType("SHUTTLE_runtype", runType);
1254
1255 TList mlList;
1256 mlList.Add(&mlStatus);
1257 mlList.Add(&mlRunType);
1258
1259 TString mlID;
1260 mlID.Form("%d", GetCurrentRun());
1261 fMonaLisa->SendParameters(&mlList, mlID);
1262}
1263
ef83773f 1264//______________________________________________________________________________________________
1265Int_t AliShuttle::GetMem(Int_t pid)
1266{
1267 // invokes ps to get the memory consumption of the process <pid>
1268 // returns -1 in case of error
1269
1270 TString checkStr;
1271 checkStr.Form("ps -o vsize --pid %d | tail -n 1", pid);
1272 FILE* pipe = gSystem->OpenPipe(checkStr, "r");
1273 if (!pipe)
1274 {
1275 Log("SHUTTLE", Form("Process - Error: "
1276 "Could not open pipe to %s", checkStr.Data()));
1277 return -1;
1278 }
1279
1280 char buffer[100];
1281 if (!fgets(buffer, 100, pipe))
1282 {
1283 Log("SHUTTLE", "Process - Error: ps did not return anything");
1284 gSystem->ClosePipe(pipe);
1285 return -1;
1286 }
1287 gSystem->ClosePipe(pipe);
1288
1289 //Log("SHUTTLE", Form("ps returned %s", buffer));
1290
1291 Int_t mem = 0;
1292 if ((sscanf(buffer, "%d\n", &mem) != 1) || !mem)
1293 {
1294 Log("SHUTTLE", "Process - Error: Could not parse output of ps");
1295 return -1;
1296 }
1297
1298 return mem;
1299}
1300
5164a766 1301//______________________________________________________________________________________________
2bb7b766 1302Bool_t AliShuttle::Process(AliShuttleLogbookEntry* entry)
58bc3020 1303{
73abe331 1304 //
b948db8d 1305 // Makes data retrieval for all detectors in the configuration.
2bb7b766 1306 // entry: Shuttle logbook entry, contains run paramenters and status of detectors
1307 // (Unprocessed, Inactive, Failed or Done).
d477ad88 1308 // Returns kFALSE in case of error occured and kTRUE otherwise
73abe331 1309 //
1310
9827400b 1311 if (!entry) return kFALSE;
2bb7b766 1312
1313 fLogbookEntry = entry;
1314
c88ad5db 1315 Log("SHUTTLE", Form("\t\t\t^*^*^*^*^*^*^*^*^*^*^*^* run %d: START ^*^*^*^*^*^*^*^*^*^*^*^*",
9827400b 1316 GetCurrentRun()));
2bb7b766 1317
339adafa 1318 CountOpenRuns();
1319
8e828d39 1320 // Send the information to ML
1321 SendMLRunInfo("Processing");
3301427a 1322
9827400b 1323 if (fLogbookEntry->IsDone())
1324 {
1325 Log("SHUTTLE","Process - Shuttle is already DONE. Updating logbook");
1326 UpdateShuttleLogbook("shuttle_done");
1327 fLogbookEntry = 0;
1328 return kTRUE;
1329 }
1330
1331 // read test mode if flag is set
1332 if (fReadTestMode)
1333 {
3d8bc902 1334 fTestMode = kNone;
9827400b 1335 TString logEntry(entry->GetRunParameter("log"));
1336 //printf("log entry = %s\n", logEntry.Data());
1337 TString searchStr("Testmode: ");
1338 Int_t pos = logEntry.Index(searchStr.Data());
1339 //printf("%d\n", pos);
1340 if (pos >= 0)
1341 {
1342 TSubString subStr = logEntry(pos + searchStr.Length(), logEntry.Length());
1343 //printf("%s\n", subStr.String().Data());
1344 TString newStr(subStr.Data());
1345 TObjArray* token = newStr.Tokenize(' ');
1346 if (token)
1347 {
1348 //token->Print();
1349 TObjString* tmpStr = dynamic_cast<TObjString*> (token->First());
1350 if (tmpStr)
1351 {
1352 Int_t testMode = tmpStr->String().Atoi();
1353 if (testMode > 0)
1354 {
c88ad5db 1355 Log("SHUTTLE", Form("Process - Enabling test mode %d", testMode));
9827400b 1356 SetTestMode((TestMode) testMode);
1357 }
1358 }
1359 delete token;
1360 }
1361 }
1362 }
c88ad5db 1363
eba76848 1364 fLogbookEntry->Print("all");
57f50b3c 1365
1366 // Initialization
d477ad88 1367 Bool_t hasError = kFALSE;
5164a766 1368
728290a4 1369 // Set the CDB and Reference folders according to the year
1370
1371 // build cdb paths (repeat each time, run might be a DATest run)
1372 if (!fLogbookEntry->GetDATestMode()){
1373 fgkMainCDB.Form("alien://folder=%s%d/OCDB?user=alidaq?cacheFold=/tmp/OCDBCache",
1374 fConfig->GetAlienPath(), GetCurrentYear());
1375
1376 fgkMainRefStorage.Form("alien://folder=%s%d/Reference?user=alidaq?cacheFold=/tmp/OCDBCache",
1377 fConfig->GetAlienPath(), GetCurrentYear());
1378 }
1379 else {
1380 fgkMainCDB.Form("alien://folder=%s%d/DATest/OCDB?user=alidaq?cacheFold=/tmp/OCDBCache",
1381 fConfig->GetAlienPath(), GetCurrentYear());
1382
1383 fgkMainRefStorage.Form("alien://folder=%s%d/DATest/Reference?user=alidaq?cacheFold=/tmp/OCDBCache",
1384 fConfig->GetAlienPath(), GetCurrentYear());
1385 }
1386
1387 AliDebug(2,Form("Main CDB storage = %s",fgkMainCDB.Data()));
1388 AliDebug(2,Form("Main Reference storage = %s",fgkMainRefStorage.Data()));
1389
57f50b3c 1390 // Loop on detectors in the configuration
b948db8d 1391 TIter iter(fConfig->GetDetectors());
2bb7b766 1392 TObjString* aDetector = 0;
b948db8d 1393
4508ab8b 1394 Bool_t first = kTRUE;
1395
be48e3ea 1396 while ((aDetector = (TObjString*) iter.Next()))
1397 {
7bfb2090 1398 fCurrentDetector = aDetector->String();
5164a766 1399
5e993b6f 1400 if (ContinueProcessing() == kFALSE)
1401 continue;
4508ab8b 1402
1403 if (first)
1404 {
1405 // only read QueryCDB when needed and only once
1406 AliCDBStorage *mainCDBSto = AliCDBManager::Instance()->GetStorage(fgkMainCDB);
1407 if(mainCDBSto) mainCDBSto->QueryCDB(GetCurrentRun());
1408 AliCDBStorage *mainRefSto = AliCDBManager::Instance()->GetStorage(fgkMainRefStorage);
1409 if(mainRefSto) mainRefSto->QueryCDB(GetCurrentRun());
1410 first = kFALSE;
1411 }
9e080f92 1412
c88ad5db 1413 Log("SHUTTLE", Form("\t\t\t****** run %d - %s: START ******",
2bb7b766 1414 GetCurrentRun(), aDetector->GetName()));
1415
6e25f828 1416 for(Int_t iSys=0;iSys<4;iSys++) fFXSCalled[iSys]=kFALSE;
ef83773f 1417
1418 Int_t initialMem = GetMem(getpid());
1419 Log("SHUTTLE", Form("Memory consumption before forking is %d", initialMem));
9d733021 1420
c88ad5db 1421 Log(fCurrentDetector.Data(), "Process - Starting processing");
85a80aa9 1422
be48e3ea 1423 Int_t pid = fork();
1424
1425 if (pid < 0)
1426 {
c88ad5db 1427 Log("SHUTTLE", "Process - ERROR: Forking failed");
be48e3ea 1428 }
1429 else if (pid > 0)
1430 {
1431 // parent
c88ad5db 1432 Log("SHUTTLE", Form("Process - In parent process of %d - %s: Starting monitoring",
be48e3ea 1433 GetCurrentRun(), aDetector->GetName()));
1434
1435 Long_t begin = time(0);
1436
1437 int status; // to be used with waitpid, on purpose an int (not Int_t)!
1438 while (waitpid(pid, &status, WNOHANG) == 0)
1439 {
1440 Long_t expiredTime = time(0) - begin;
1441
01c76120 1442 // the run-dependent timeout is the timeout from the configuration plus a twentieth of
1443 // the run duration, e.g. 3 additional minutes for 1h run or 1/2h for a 10h run
1444 Int_t runDepTimeOut = fConfig->GetPPTimeOut() + (GetCurrentEndTime() - GetCurrentStartTime()) * 0.05;
1445 if (expiredTime > runDepTimeOut)
be48e3ea 1446 {
56d3e9e4 1447 TString logMsg;
1448 AliShuttleStatus *currentStatus = ReadShuttleStatus();
1449 AliShuttleStatus::Status newStatus = AliShuttleStatus::kInvalid;
1450
838d9827 1451 if (currentStatus->GetStatus() == AliShuttleStatus::kDCSStarted)
1452 {
1453 // in case the pp goes in TimeOut while retrieving the DCS DPs
1454 // set status to kDCSError
1455
0a0b087a 1456 logMsg.Form("Process - Process of %s timed out while retrieving the DCS DataPoints. Run time: %ld seconds. Killing... and setting status to DCSError.",
838d9827 1457 fCurrentDetector.Data(), expiredTime);
1458 newStatus = AliShuttleStatus::kDCSError;
1459 }
1460 else if (currentStatus->GetStatus() <= AliShuttleStatus::kPPDone)
56d3e9e4 1461 {
1462 // in case pp not yet done set status to kPPTimeOut
1463
0a0b087a 1464 logMsg.Form("Process - Process of %s timed out. Run time: %ld seconds. Killing...",
56d3e9e4 1465 fCurrentDetector.Data(), expiredTime);
1466 newStatus = AliShuttleStatus::kPPTimeOut;
1467 }
1468 else if (currentStatus->GetStatus() == AliShuttleStatus::kStoreStarted)
1469 {
1470 // in case the pp goes in TimeOut while storing the objects in the OCDB
1471 // set status to kStoreError
1472
0a0b087a 1473 logMsg.Form("Process - Process of %s timed out while storing the OCDB object. Run time: %ld seconds. Killing... and setting status to StoreError.",
56d3e9e4 1474 fCurrentDetector.Data(), expiredTime);
1475 newStatus = AliShuttleStatus::kStoreError;
1476 }
1477 else
1478 {
1479 // in other cases don't change the status
1480
0a0b087a 1481 logMsg.Form("Process - Process of %s timed out in status = %s. Run time: %ld seconds. Killing... without changing the status",
56d3e9e4 1482 fCurrentDetector.Data(), currentStatus->GetStatusName(), expiredTime);
1483 }
1484
1485 Log("SHUTTLE", logMsg);
1486 Log(fCurrentDetector, logMsg);
be48e3ea 1487
1488 kill(pid, 9);
1489
56d3e9e4 1490 if (newStatus != AliShuttleStatus::kInvalid)
1491 UpdateShuttleStatus(newStatus);
be48e3ea 1492 hasError = kTRUE;
1493
1494 gSystem->Sleep(1000);
1495 }
1496 else
1497 {
be48e3ea 1498 gSystem->Sleep(1000);
9827400b 1499
ef83773f 1500 Int_t mem = GetMem(pid);
1501
1502 if (mem < 0)
9827400b 1503 continue;
9827400b 1504
ef83773f 1505 mem -= initialMem;
1506 if (mem < 0)
1507 mem = 0;
9827400b 1508
1509 if (expiredTime % 60 == 0)
ee6f7523 1510 {
c88ad5db 1511 Log("SHUTTLE", Form("Process - %s: Checking process. "
0a0b087a 1512 "Run time: %ld seconds - Memory consumption: %d KB",
c88ad5db 1513 fCurrentDetector.Data(), expiredTime, mem));
ee6f7523 1514 SendAlive();
1515 }
9827400b 1516
1517 if (mem > fConfig->GetPPMaxMem())
1518 {
1519 TString tmp;
c88ad5db 1520 tmp.Form("Process - Process exceeds maximum allowed memory "
1521 "(%d KB > %d KB). Killing...",
9827400b 1522 mem, fConfig->GetPPMaxMem());
1523 Log("SHUTTLE", tmp);
1524 Log(fCurrentDetector, tmp);
1525
1526 kill(pid, 9);
1527
1528 UpdateShuttleStatus(AliShuttleStatus::kPPOutOfMemory);
1529 hasError = kTRUE;
1530
1531 gSystem->Sleep(1000);
1532 }
be48e3ea 1533 }
1534 }
1535
c88ad5db 1536 Log("SHUTTLE", Form("Process - In parent process of %d - %s: Client has terminated.",
be48e3ea 1537 GetCurrentRun(), aDetector->GetName()));
1538
1539 if (WIFEXITED(status))
1540 {
1541 Int_t returnCode = WEXITSTATUS(status);
1542
c88ad5db 1543 Log("SHUTTLE", Form("Process - %s: the return code is %d", fCurrentDetector.Data(),
3301427a 1544 returnCode));
be48e3ea 1545
9827400b 1546 if (returnCode == 0) hasError = kTRUE;
be48e3ea 1547 }
1548 }
1549 else if (pid == 0)
1550 {
26b1866e 1551 // child
1552 Log("SHUTTLE", Form("Process - In child process of %d - %s", GetCurrentRun(),
c88ad5db 1553 aDetector->GetName()));
be48e3ea 1554
c88ad5db 1555 Log("SHUTTLE", Form("Process - Redirecting output to %s log",fCurrentDetector.Data()));
ffa29e93 1556
546242fb 1557 if ((freopen(GetLogFileName(fCurrentDetector), "a", stdout)) == 0)
ffa29e93 1558 {
c88ad5db 1559 Log("SHUTTLE", "Process - Could not freopen stdout");
ffa29e93 1560 }
1561 else
1562 {
1563 fOutputRedirected = kTRUE;
1564 if ((dup2(fileno(stdout), fileno(stderr))) < 0)
c88ad5db 1565 Log("SHUTTLE", "Process - Could not redirect stderr");
ffa29e93 1566
1567 }
51c3744e 1568
1569 Log("SHUTTLE", "Executing TGrid::Connect");
1570 TGrid::Connect("alien://");
ffa29e93 1571
5bac2bde 1572 TString wd = gSystem->WorkingDirectory();
3ebbf3c4 1573 Int_t dir_lev1 = GetCurrentRun()/10000;
1574 TString tmpDir = Form("%s/%d/%d/%s_process", GetShuttleTempDir(),
1575 dir_lev1, GetCurrentRun(), fCurrentDetector.Data());
5bac2bde 1576
d524ade6 1577 Int_t result = gSystem->GetPathInfo(tmpDir.Data(), 0, (Long64_t*) 0, 0, 0);
1578 if (!result) // temp dir already exists!
1579 {
1580 Log(fCurrentDetector.Data(),
1581 Form("Process - %s dir already exists! Removing...", tmpDir.Data()));
1582 gSystem->Exec(Form("rm -rf %s",tmpDir.Data()));
675f64cd 1583 }
1584
1585 if (gSystem->mkdir(tmpDir.Data(), 1))
1586 {
1587 Log(fCurrentDetector.Data(), "Process - could not make temp directory!!");
1588 gSystem->Exit(1);
d524ade6 1589 }
1590
1591 if (!gSystem->ChangeDirectory(tmpDir.Data()))
1592 {
1593 Log(fCurrentDetector.Data(), "Process - could not change directory!!");
1594 gSystem->Exit(1);
1595 }
5bac2bde 1596
ffa25f20 1597 Int_t success = ProcessCurrentDetector();
26b1866e 1598
5bac2bde 1599 gSystem->ChangeDirectory(wd.Data());
b1d18693 1600
ffa25f20 1601 if (success == 1) // Preprocessor finished successfully!
9827400b 1602 {
fb40ead4 1603 // remove temporary folder or DCS map
1604 if (!fConfig->KeepTempFolder())
1605 {
1606 gSystem->Exec(Form("rm -rf %s",tmpDir.Data()));
1607 } else if (!fConfig->KeepDCSMap())
1608 {
1609 gSystem->Exec(Form("rm -f %s/DCSMap.root",tmpDir.Data()));
1610 }
b1d18693 1611
3301427a 1612 // Update time_processed field in FXS DB
1613 if (UpdateTable() == kFALSE)
5bac2bde 1614 Log("SHUTTLE", Form("Process - %s: Could not update FXS databases!",
1615 fCurrentDetector.Data()));
3301427a 1616
1617 // Transfer the data from local storage to main storage (Grid)
3301427a 1618 if (StoreOCDB() == kFALSE)
9827400b 1619 success = kFALSE;
5e993b6f 1620 }
ffa25f20 1621 else if (success == 0)
c88ad5db 1622 {
1623 Log("SHUTTLE",
5ca8f352 1624 Form("\t\t\t****** run %d - %s: ERROR ******",
c88ad5db 1625 GetCurrentRun(), aDetector->GetName()));
be48e3ea 1626 }
1627
6e25f828 1628 for (UInt_t iSys=0; iSys<4; iSys++)
4b95672b 1629 {
1630 if (fFXSCalled[iSys]) fFXSlist[iSys].Clear();
1631 }
1632
c88ad5db 1633 Log("SHUTTLE", Form("Process - Client process of %d - %s is exiting now with %d.",
9827400b 1634 GetCurrentRun(), aDetector->GetName(), success));
be48e3ea 1635
1636 // the client exits here
9827400b 1637 gSystem->Exit(success);
be48e3ea 1638
1639 AliError("We should never get here!!!");
1640 }
7bfb2090 1641 }
5164a766 1642
c88ad5db 1643 Log("SHUTTLE", Form("\t\t\t^*^*^*^*^*^*^*^*^*^*^*^* run %d: FINISH ^*^*^*^*^*^*^*^*^*^*^*^*",
2bb7b766 1644 GetCurrentRun()));
1645
1646 //check if shuttle is done for this run, if so update logbook
1647 TObjArray checkEntryArray;
1648 checkEntryArray.SetOwner(1);
9e080f92 1649 TString whereClause = Form("where run=%d", GetCurrentRun());
b0e53b15 1650 if (!QueryShuttleLogbook(whereClause.Data(), checkEntryArray) ||
1651 checkEntryArray.GetEntries() == 0) {
9e080f92 1652 Log("SHUTTLE", Form("Process - Warning: Cannot check status of run %d on Shuttle logbook!",
1653 GetCurrentRun()));
1654 return hasError == kFALSE;
1655 }
b948db8d 1656
9e080f92 1657 AliShuttleLogbookEntry* checkEntry = dynamic_cast<AliShuttleLogbookEntry*>
1658 (checkEntryArray.At(0));
2bb7b766 1659
9e080f92 1660 if (checkEntry)
1661 {
1662 if (checkEntry->IsDone())
be48e3ea 1663 {
9e080f92 1664 Log("SHUTTLE","Process - Shuttle is DONE. Updating logbook");
1665 UpdateShuttleLogbook("shuttle_done");
1666 }
1667 else
1668 {
1669 for (UInt_t iDet=0; iDet<NDetectors(); iDet++)
be48e3ea 1670 {
9e080f92 1671 if (checkEntry->GetDetectorStatus(iDet) == AliShuttleLogbookEntry::kUnprocessed)
be48e3ea 1672 {
9e080f92 1673 AliDebug(2, Form("Run %d: setting %s as \"not first time unprocessed\"",
1674 checkEntry->GetRun(), GetDetName(iDet)));
1675 fFirstUnprocessed[iDet] = kFALSE;
be48e3ea 1676 }
1677 }
8e828d39 1678 SendMLRunInfo("Pending");
2bb7b766 1679 }
1680 }
1681
1682 fLogbookEntry = 0;
85a80aa9 1683
a7160fe9 1684 return hasError == kFALSE;
73abe331 1685}
1686
b948db8d 1687//______________________________________________________________________________________________
ffa25f20 1688Int_t AliShuttle::ProcessCurrentDetector()
73abe331 1689{
1690 //
2bb7b766 1691 // Makes data retrieval just for a specific detector (fCurrentDetector).
73abe331 1692 // Threre should be a configuration for this detector.
73abe331 1693
1d172743 1694 Log("SHUTTLE", Form("ProcessCurrentDetector - Retrieving values for %s, run %d",
1695 fCurrentDetector.Data(), GetCurrentRun()));
73abe331 1696
d524ade6 1697 TString wd = gSystem->WorkingDirectory();
1698
2d9019b4 1699 if (!CleanReferenceStorage(fCurrentDetector.Data()))
ffa25f20 1700 return 0;
d524ade6 1701
1702 gSystem->ChangeDirectory(wd.Data());
1703
3301427a 1704 // call preprocessor
1705 AliPreprocessor* aPreprocessor =
1706 dynamic_cast<AliPreprocessor*> (fPreprocessorMap.GetValue(fCurrentDetector));
1707
8579c6fb 1708 // check if the preprocessor wants to process this run type
1709 if (aPreprocessor->ProcessRunType() == kFALSE)
ffa25f20 1710 {
1711 UpdateShuttleStatus(AliShuttleStatus::kSkipped);
1712 UpdateShuttleLogbook(fCurrentDetector, "DONE");
bd23d48b 1713 if (!UpdateTableSkippedCase(fCurrentDetector.Data()))
1714 {
1715 AliError(Form("Could not update FXS tables for run %d !", GetCurrentRun()));
1716 }
ffa25f20 1717 Log(fCurrentDetector, Form("ProcessCurrentDetector - %s preprocessor is not interested in this run type", fCurrentDetector.Data()));
1718
1719 return 2;
1720 }
1721
5ca8f352 1722 // checking if OCDB is reachable
1723 AliCDBEntry* testEntry = GetFromOCDB("SHUTTLE","GRP/CTP/DummyConfig");
1724 if (!testEntry){
1725 // OCDB is not accessible, going in OCDBError for current detector
1726 AliError("OCDB Test entry not accessible");
1727 UpdateShuttleStatus(AliShuttleStatus::kOCDBError);
1728 return 0;
1729 }
1730
ffa25f20 1731 TMap* dcsMap = new TMap();
1732
3301427a 1733 aPreprocessor->Initialize(GetCurrentRun(), GetCurrentStartTime(), GetCurrentEndTime());
1734
1735 Bool_t processDCS = aPreprocessor->ProcessDCS();
d477ad88 1736
651fdaab 1737 if (!processDCS)
1738 {
1d172743 1739 Log(fCurrentDetector, "ProcessCurrentDetector -"
1740 " The preprocessor requested to skip the retrieval of DCS values");
651fdaab 1741 }
8b739301 1742 else if (fTestMode & kSkipDCS)
2c15234c 1743 {
1d172743 1744 Log(fCurrentDetector, "ProcessCurrentDetector - In TESTMODE: Skipping DCS processing");
9827400b 1745 }
1746 else if (fTestMode & kErrorDCS)
1747 {
1d172743 1748 Log(fCurrentDetector, "ProcessCurrentDetector - In TESTMODE: Simulating DCS error");
3d8bc902 1749 UpdateShuttleStatus(AliShuttleStatus::kDCSStarted);
9827400b 1750 UpdateShuttleStatus(AliShuttleStatus::kDCSError);
1d172743 1751 delete dcsMap;
ffa25f20 1752 return 0;
2c15234c 1753 } else {
3301427a 1754
1755 UpdateShuttleStatus(AliShuttleStatus::kDCSStarted);
1756
1d172743 1757 // Query DCS archive
1758 Int_t nServers = fConfig->GetNServers(fCurrentDetector);
a038aa70 1759
1d172743 1760 for (int iServ=0; iServ<nServers; iServ++)
2c15234c 1761 {
1d172743 1762
1763 TString host(fConfig->GetDCSHost(fCurrentDetector, iServ));
1764 Int_t port = fConfig->GetDCSPort(fCurrentDetector, iServ);
542b6cc8 1765 Int_t multiSplit = fConfig->GetMultiSplit(fCurrentDetector, iServ);
1766
1790d4b7 1767 Log(fCurrentDetector, Form("ProcessCurrentDetector -"
1768 " Querying DCS Amanda server %s:%d (%d of %d)",
1769 host.Data(), port, iServ+1, nServers));
1d172743 1770
1771 TMap* aliasMap = 0;
1772 TMap* dpMap = 0;
26b1866e 1773
1d172743 1774 if (fConfig->GetDCSAliases(fCurrentDetector, iServ)->GetEntries() > 0)
2c15234c 1775 {
5ed060b1 1776 Log(fCurrentDetector, Form("Querying %d DCS aliases", fConfig->GetDCSAliases(fCurrentDetector, iServ)->GetEntries()));
1d172743 1777 aliasMap = GetValueSet(host, port,
542b6cc8 1778 fConfig->GetDCSAliases(fCurrentDetector, iServ),
1779 kAlias, multiSplit);
1d172743 1780 if (!aliasMap)
1781 {
1782 Log(fCurrentDetector,
1783 Form("ProcessCurrentDetector -"
675f64cd 1784 " Error retrieving DCS aliases from server %s."
1785 " Sending mail to DCS experts!", host.Data()));
1d172743 1786 UpdateShuttleStatus(AliShuttleStatus::kDCSError);
675f64cd 1787
339adafa 1788 if (!SendMail(kDCSEMail))
fb40ead4 1789 Log("SHUTTLE", Form("ProcessCurrentDetector - "
95d4aaef 1790 "Could not send mail to DCS experts!"));
675f64cd 1791
a038aa70 1792 delete dcsMap;
ffa25f20 1793 return 0;
1d172743 1794 }
2c15234c 1795 }
a038aa70 1796
1d172743 1797 if (fConfig->GetDCSDataPoints(fCurrentDetector, iServ)->GetEntries() > 0)
a038aa70 1798 {
5ed060b1 1799 Log(fCurrentDetector, Form("Querying %d DCS data points", fConfig->GetDCSDataPoints(fCurrentDetector, iServ)->GetEntries()));
1d172743 1800 dpMap = GetValueSet(host, port,
542b6cc8 1801 fConfig->GetDCSDataPoints(fCurrentDetector, iServ),
1802 kDP, multiSplit);
1d172743 1803 if (!dpMap)
1804 {
1805 Log(fCurrentDetector,
1806 Form("ProcessCurrentDetector -"
675f64cd 1807 " Error retrieving DCS data points from server %s."
1808 " Sending mail to DCS experts!", host.Data()));
1d172743 1809 UpdateShuttleStatus(AliShuttleStatus::kDCSError);
675f64cd 1810
339adafa 1811 if (!SendMail(kDCSEMail))
fb40ead4 1812 Log("SHUTTLE", Form("ProcessCurrentDetector - "
95d4aaef 1813 "Could not send mail to DCS experts!"));
675f64cd 1814
1d172743 1815 if (aliasMap) delete aliasMap;
1816 delete dcsMap;
ffa25f20 1817 return 0;
1d172743 1818 }
a038aa70 1819 }
1d172743 1820
1821 // merge aliasMap and dpMap into dcsMap
1822 if(aliasMap) {
1823 TIter iter(aliasMap);
a038aa70 1824 TObjString* key = 0;
1825 while ((key = (TObjString*) iter.Next()))
1d172743 1826 dcsMap->Add(key, aliasMap->GetValue(key->String()));
1827
1828 aliasMap->SetOwner(kFALSE);
1829 delete aliasMap;
1830 }
1831
1832 if(dpMap) {
1833 TIter iter(dpMap);
1834 TObjString* key = 0;
1835 while ((key = (TObjString*) iter.Next()))
1836 dcsMap->Add(key, dpMap->GetValue(key->String()));
1837
1838 dpMap->SetOwner(kFALSE);
1839 delete dpMap;
a038aa70 1840 }
73abe331 1841 }
1842 }
dc25836b 1843
b1d18693 1844 // save map into file, to help debugging in case of preprocessor error
fbc112e3 1845 TFile* f = TFile::Open("DCSMap.root","recreate");
b1d18693 1846 f->cd();
1847 dcsMap->Write("DCSMap", TObject::kSingleKey);
1848 f->Close();
fbc112e3 1849 delete f;
b1d18693 1850
2bb7b766 1851 // DCS Archive DB processing successful. Call Preprocessor!
85a80aa9 1852 UpdateShuttleStatus(AliShuttleStatus::kPPStarted);
a7160fe9 1853
339adafa 1854 fFXSError = -1; // this variable is kTRUE after ::Process if an FXS error occured
8da81210 1855
a038aa70 1856 UInt_t returnValue = aPreprocessor->Process(dcsMap);
8da81210 1857
339adafa 1858 if (fFXSError!=-1) {
8da81210 1859 UpdateShuttleStatus(AliShuttleStatus::kFXSError);
339adafa 1860 SendMail(kFXSEMail, fFXSError);
8da81210 1861 dcsMap->DeleteAll();
1862 delete dcsMap;
ffa25f20 1863 return 0;
8da81210 1864 }
b948db8d 1865
3301427a 1866 if (returnValue > 0) // Preprocessor error!
1867 {
c88ad5db 1868 Log(fCurrentDetector, Form("ProcessCurrentDetector - "
1869 "Preprocessor failed. Process returned %d.", returnValue));
cb343cfd 1870 UpdateShuttleStatus(AliShuttleStatus::kPPError);
a038aa70 1871 dcsMap->DeleteAll();
1872 delete dcsMap;
ffa25f20 1873 return 0;
9827400b 1874 }
1875
1876 // preprocessor ok!
1877 UpdateShuttleStatus(AliShuttleStatus::kPPDone);
1878 Log(fCurrentDetector, Form("ProcessCurrentDetector - %s preprocessor returned success",
1879 fCurrentDetector.Data()));
b948db8d 1880
a038aa70 1881 dcsMap->DeleteAll();
1882 delete dcsMap;
b948db8d 1883
ffa25f20 1884 return 1;
2bb7b766 1885}
1886
fbc112e3 1887//______________________________________________________________________________________________
1888void AliShuttle::CountOpenRuns()
1889{
1890 // Query DAQ's Shuttle logbook and sends the number of open runs to ML
1891
982278d6 1892 SendAlive();
1893
fbc112e3 1894 // check connection, in case connect
efec19bd 1895 if (!Connect(4))
fbc112e3 1896 return;
1897
1898 TString sqlQuery;
1899 sqlQuery = Form("select count(*) from %s where shuttle_done=0", fConfig->GetShuttlelbTable());
1900
efec19bd 1901 TSQLResult* aResult = fServer[4]->Query(sqlQuery);
fbc112e3 1902 if (!aResult) {
1903 AliError(Form("Can't execute query <%s>!", sqlQuery.Data()));
1904 return;
1905 }
1906
1907 AliDebug(2,Form("Query = %s", sqlQuery.Data()));
1908
1909 if (aResult->GetRowCount() == 0) {
1910 AliError(Form("No result for query %s received", sqlQuery.Data()));
1911 return;
1912 }
1913
1914 if (aResult->GetFieldCount() != 1) {
1915 AliError(Form("Invalid field count for query %s received", sqlQuery.Data()));
1916 return;
1917 }
1918
1919 TSQLRow* aRow = aResult->Next();
1920 if (!aRow) {
1921 AliError(Form("Could not receive result of query %s", sqlQuery.Data()));
1922 return;
1923 }
1924
1925 TString result(aRow->GetField(0), aRow->GetFieldLength(0));
1926 Int_t count = result.Atoi();
1927
1928 Log("SHUTTLE", Form("%d unprocessed runs", count));
1929
1930 delete aRow;
1931 delete aResult;
1932
1933 TMonaLisaValue mlStatus("SHUTTLE_openruns", count);
1934
1935 TList mlList;
1936 mlList.Add(&mlStatus);
1937
1938 fMonaLisa->SendParameters(&mlList, "__PROCESSINGINFO__");
1939}
1940
2bb7b766 1941//______________________________________________________________________________________________
1942Bool_t AliShuttle::QueryShuttleLogbook(const char* whereClause,
1943 TObjArray& entries)
1944{
9827400b 1945 // Query DAQ's Shuttle logbook and fills detector status object.
1946 // Call QueryRunParameters to query DAQ logbook for run parameters.
1947 //
2bb7b766 1948
fc5a4708 1949 entries.SetOwner(1);
1950
2bb7b766 1951 // check connection, in case connect
efec19bd 1952 if (!Connect(4)) return kFALSE;
2bb7b766 1953
1954 TString sqlQuery;
441b0e9c 1955 sqlQuery = Form("select * from %s %s order by run", fConfig->GetShuttlelbTable(), whereClause);
2bb7b766 1956
efec19bd 1957 TSQLResult* aResult = fServer[4]->Query(sqlQuery);
2bb7b766 1958 if (!aResult) {
1959 AliError(Form("Can't execute query <%s>!", sqlQuery.Data()));
1960 return kFALSE;
1961 }
1962
fc5a4708 1963 AliDebug(2,Form("Query = %s", sqlQuery.Data()));
1964
2bb7b766 1965 if(aResult->GetRowCount() == 0) {
c88ad5db 1966 Log("SHUTTLE", "No entries in Shuttle Logbook match request");
9827400b 1967 delete aResult;
1968 return kTRUE;
2bb7b766 1969 }
1970
1971 // TODO Check field count!
0fdc005f 1972 const UInt_t nCols = 27;
2bb7b766 1973 if (aResult->GetFieldCount() != (Int_t) nCols) {
c88ad5db 1974 Log("SHUTTLE", "Invalid SQL result field number!");
2bb7b766 1975 delete aResult;
1976 return kFALSE;
1977 }
1978
2bb7b766 1979 TSQLRow* aRow;
1980 while ((aRow = aResult->Next())) {
1981 TString runString(aRow->GetField(0), aRow->GetFieldLength(0));
1982 Int_t run = runString.Atoi();
1983
eba76848 1984 AliShuttleLogbookEntry *entry = QueryRunParameters(run);
1985 if (!entry)
1986 continue;
2bb7b766 1987
728290a4 1988 // DA test mode flag
1989 TString daTestModeString(aRow->GetField(2), aRow->GetFieldLength(2)); // field 2 = DA test mode flag
1990 Bool_t daTestMode = (Bool_t)daTestModeString.Atoi();
1991 entry->SetDATestMode(daTestMode);
1992
2bb7b766 1993 // loop on detectors
eba76848 1994 for(UInt_t ii = 0; ii < nCols; ii++)
1995 entry->SetDetectorStatus(aResult->GetFieldName(ii), aRow->GetField(ii));
2bb7b766 1996
eba76848 1997 entries.AddLast(entry);
2bb7b766 1998 delete aRow;
1999 }
2000
2bb7b766 2001 delete aResult;
2002 return kTRUE;
2003}
2004
2005//______________________________________________________________________________________________
eba76848 2006AliShuttleLogbookEntry* AliShuttle::QueryRunParameters(Int_t run)
2bb7b766 2007{
eba76848 2008 //
2009 // Retrieve run parameters written in the DAQ logbook and sets them into AliShuttleLogbookEntry object
2010 //
2bb7b766 2011
2012 // check connection, in case connect
efec19bd 2013 if (!Connect(4))
eba76848 2014 return 0;
2bb7b766 2015
2016 TString sqlQuery;
2c15234c 2017 sqlQuery.Form("select * from %s where run=%d", fConfig->GetDAQlbTable(), run);
2bb7b766 2018
efec19bd 2019 TSQLResult* aResult = fServer[4]->Query(sqlQuery);
2bb7b766 2020 if (!aResult) {
c88ad5db 2021 Log("SHUTTLE", Form("Can't execute query <%s>!", sqlQuery.Data()));
eba76848 2022 return 0;
2bb7b766 2023 }
2024
eba76848 2025 if (aResult->GetRowCount() == 0) {
2bb7b766 2026 Log("SHUTTLE", Form("QueryRunParameters - No entry in DAQ Logbook for run %d. Skipping", run));
2027 delete aResult;
eba76848 2028 return 0;
2bb7b766 2029 }
2030
eba76848 2031 if (aResult->GetRowCount() > 1) {
c88ad5db 2032 Log("SHUTTLE", Form("QueryRunParameters - UNEXPECTED: "
2033 "more than one entry in DAQ Logbook for run %d!", run));
2bb7b766 2034 delete aResult;
eba76848 2035 return 0;
2bb7b766 2036 }
2037
eba76848 2038 TSQLRow* aRow = aResult->Next();
2039 if (!aRow)
2040 {
c88ad5db 2041 Log("SHUTTLE", Form("QueryRunParameters - Could not retrieve row for run %d. Skipping", run));
eba76848 2042 delete aResult;
2043 return 0;
2044 }
2bb7b766 2045
eba76848 2046 AliShuttleLogbookEntry* entry = new AliShuttleLogbookEntry(run);
2bb7b766 2047
eba76848 2048 for (Int_t ii = 0; ii < aResult->GetFieldCount(); ii++)
2049 entry->SetRunParameter(aResult->GetFieldName(ii), aRow->GetField(ii));
2bb7b766 2050
c9db68c7 2051 delete aRow;
2052 delete aResult;
2053
eba76848 2054 UInt_t startTime = entry->GetStartTime();
2055 UInt_t endTime = entry->GetEndTime();
89bcf3b4 2056 Bool_t ecsSuccess = entry->GetECSSuccess();
1415d3f9 2057 TString runType = entry->GetRunType();
2058 TString tmpdaqstartTime = entry->GetRunParameter("DAQ_time_start");
f38b266c 2059 TString recordingFlagString = entry->GetRunParameter("GDCmStreamRecording");
2060 UInt_t recordingFlag = recordingFlagString.Atoi();
1415d3f9 2061 UInt_t daqstartTime = tmpdaqstartTime.Atoi();
c9db68c7 2062
af93e5d7 2063 UInt_t now = time(0);
2eb98462 2064 Int_t dcsDelay = fConfig->GetDCSDelay()+fConfig->GetDCSQueryOffset();
8e828d39 2065
2066 Bool_t skip = kFALSE;
2067
1415d3f9 2068 // runs are processed if
2069 // a) runType is PHYSICS and ecsSuccess is set
2070 // b) runType is not PHYSICS and (ecsSuccess is set or DAQ_time_start is non-0)
2071 // effectively this means that all runs are processed that started properly (ecsSucess behaviour is different for PHYSICS and non-PHYSICS runs (check with ECS!)
1bbcc21b 2072 if (startTime != 0 && endTime != 0) {
2073 if (endTime > startTime) {
2074 if (endTime >= now - dcsDelay) {
2075 Log("SHUTTLE", Form("Skipping run %d for now, because DCS buffer time is not yet expired", run));
2076 } else {
f38b266c 2077 if ((runType == "PHYSICS" || runType == "STANDALONE") && recordingFlag == 0){
2078 Log("SHUTTLE", Form("QueryRunParameters - Run type for run %d is %s but the recording is OFF - Skipping!", run, runType.Data()));
2079 skip = kTRUE;
2080 }
2081 else {
2082 if (runType == "PHYSICS") {
2083 if (ecsSuccess) {
2084 return entry;
2085 } else {
2086 Log("SHUTTLE", Form("QueryRunParameters - Run type for run %d is PHYSICS but ECS success flag not set (Reason = %s) - Skipping!", run, entry->GetRunParameter("eor_reason")));
2087 skip = kTRUE;
2088 }
1bbcc21b 2089 } else {
f38b266c 2090 if (ecsSuccess || daqstartTime > 0) {
2091 if (ecsSuccess == kFALSE)
2092 Log("SHUTTLE", Form("Processing run %d although in status ECS failure (Reason: %s), since run type != PHYSICS and DAQ_time_start != 0", run, entry->GetRunParameter("eor_reason")));
2093 return entry;
2094 } else {
2095 Log("SHUTTLE", Form("QueryRunParameters - Run type for run %d is %s, ECS success flag was not set (Reason = %s) and DAQ_time_start was NULL - Skipping!", run, runType.Data(), entry->GetRunParameter("eor_reason")));
2096 skip = kTRUE;
2097 }
1bbcc21b 2098 }
2099 }
2100 }
2101 } else {
2102 Log("SHUTTLE", Form("QueryRunParameters - Invalid parameters for run %d: startTime equal to endTime: %d %d - Skipping!", run, startTime, endTime));
2103 skip = kTRUE;
2104 }
1415d3f9 2105 } else {
2106 Log("SHUTTLE", Form("QueryRunParameters - Invalid parameters for Run %d: "
2107 "startTime = %d, endTime = %d. Skipping (Shuttle won't be marked as DONE)!",
2108 run, startTime, endTime));
89bcf3b4 2109 }
8e828d39 2110
2111 if (skip)
2112 {
2113 Log("SHUTTLE", Form("Marking SHUTTLE skipped for run %d", run));
2114 fLogbookEntry = entry;
2115 if (!UpdateShuttleLogbook("shuttle_skipped"))
2116 {
2117 AliError(Form("Could not update logbook for run %d !", run));
2118 }
bd23d48b 2119 if (!UpdateTableSkippedCase("ALL"))
2120 {
2121 AliError(Form("Could not update FXS tables for run %d !", run));
2122 }
8e828d39 2123 fLogbookEntry = 0;
2124 }
c9db68c7 2125
2126 delete entry;
2127 return 0;
2bb7b766 2128}
2129
a038aa70 2130//______________________________________________________________________________________________
2131TMap* AliShuttle::GetValueSet(const char* host, Int_t port, const TSeqCollection* entries,
542b6cc8 2132 DCSType type, Int_t multiSplit)
a038aa70 2133{
2134 // Retrieve all "entry" data points from the DCS server
2135 // host, port: TSocket connection parameters
2136 // entries: list of name of the alias or data point
2137 // type: kAlias or kDP
2138 // returns TMap of values, 0 when failure
542b6cc8 2139
2140 AliDCSClient client(host, port, fTimeout, fRetries, multiSplit);
b41b252a 2141
a038aa70 2142 TMap* result = 0;
51657f6d 2143 UInt_t startQuery = GetStartTimeDCSQuery();
2144 UInt_t endQuery = GetEndTimeDCSQuery();
2145 if (fCurrentDetector == "GRP" && (endQuery - startQuery) <= 120) { // enlarging DCS query for GRP when a run is shorter than 2 minutes (i.e. the time of forced archival of GRP DPs)
2146 Log(fCurrentDetector.Data(), Form("GetValueSet: run lasting less than 120 seconds, enlarging DCS window for DPs retrival to 130 s"));
2147 startQuery = endQuery - 130; // we add 130 s to be sure that there is something (even if the archival is forced after 120 s)
2148 }
2149
b41b252a 2150 if (type == kAlias)
a038aa70 2151 {
5f3fe74e 2152 //result = client.GetAliasValues(entries, GetCurrentStartTime()-offset,
2153 // GetCurrentEndTime()+offset);
51657f6d 2154
2155 result = client.GetAliasValues(entries, startQuery, endQuery);
b41b252a 2156 }
2157 else if (type == kDP)
2158 {
5f3fe74e 2159 //result = client.GetDPValues(entries, GetCurrentStartTime()-offset,
2160 // GetCurrentEndTime()+offset);
51657f6d 2161 result = client.GetDPValues(entries, startQuery, endQuery);
b41b252a 2162 }
a038aa70 2163
b41b252a 2164 if (result == 0)
2165 {
2166 Log(fCurrentDetector.Data(), Form("GetValueSet - Can't get entries! Reason: %s",
1790d4b7 2167 client.GetErrorString(client.GetResultErrorCode())));
2168 if (client.GetResultErrorCode() == AliDCSClient::fgkServerError)
2169 Log(fCurrentDetector.Data(), Form("GetValueSet - Server error code: %s",
2170 client.GetServerError().Data()));
a038aa70 2171
b41b252a 2172 return 0;
a038aa70 2173 }
b41b252a 2174
a038aa70 2175 return result;
2176}
b41b252a 2177
b948db8d 2178//______________________________________________________________________________________________
57f50b3c 2179const char* AliShuttle::GetFile(Int_t system, const char* detector,
2180 const char* id, const char* source)
b948db8d 2181{
9827400b 2182 // Get calibration file from file exchange servers
2183 // First queris the FXS database for the file name, using the run, detector, id and source info
2184 // then calls RetrieveFile(filename) for actual copy to local disk
2185 // run: current run being processed (given by Logbook entry fLogbookEntry)
2186 // detector: the Preprocessor name
2187 // id: provided as a parameter by the Preprocessor
2188 // source: provided by the Preprocessor through GetFileSources function
2189
2190 // check if test mode should simulate a FXS error
2191 if (fTestMode & kErrorFXSFiles)
2192 {
2193 Log(detector, Form("GetFile - In TESTMODE - Simulating error while connecting to %s FXS", GetSystemName(system)));
2194 return 0;
2195 }
2196
57f50b3c 2197 // check connection, in case connect
9d733021 2198 if (!Connect(system))
eba76848 2199 {
9d733021 2200 Log(detector, Form("GetFile - Couldn't connect to %s FXS database", GetSystemName(system)));
339adafa 2201 fFXSError = system;
57f50b3c 2202 return 0;
2203 }
2204
2205 // Query preparation
9d733021 2206 TString sourceName(source);
d386d623 2207 Int_t nFields = 3;
2208 TString sqlQueryStart = Form("select filePath,size,fileChecksum from %s where",
2209 fConfig->GetFXSdbTable(system));
2210 TString whereClause = Form("run=%d and detector=\"%s\" and fileId=\"%s\"",
2211 GetCurrentRun(), detector, id);
2212
efec19bd 2213 if (system == kDAQ || system == kDQM)
9d733021 2214 {
d386d623 2215 whereClause += Form(" and DAQsource=\"%s\"", source);
57f50b3c 2216 }
9d733021 2217 else if (system == kDCS)
eba76848 2218 {
9d733021 2219 sourceName="none";
57f50b3c 2220 }
9d733021 2221 else if (system == kHLT)
9e080f92 2222 {
d386d623 2223 whereClause += Form(" and DDLnumbers=\"%s\"", source);
9e080f92 2224 }
2225
9e080f92 2226 TString sqlQuery = Form("%s %s", sqlQueryStart.Data(), whereClause.Data());
2227
2228 AliDebug(2, Form("SQL query: \n%s",sqlQuery.Data()));
2229
2230 // Query execution
2231 TSQLResult* aResult = 0;
9d733021 2232 aResult = dynamic_cast<TSQLResult*> (fServer[system]->Query(sqlQuery));
9e080f92 2233 if (!aResult) {
8da81210 2234 Log(detector, Form("GetFile - Can't execute SQL query to %s database for: id = %s, source = %s",
9d733021 2235 GetSystemName(system), id, sourceName.Data()));
339adafa 2236 fFXSError = system;
9e080f92 2237 return 0;
2238 }
2239
8da81210 2240 if (aResult->GetRowCount() == 0)
9e080f92 2241 {
2242 Log(detector,
8da81210 2243 Form("GetFile - No entry in %s FXS db for: id = %s, source = %s",
9d733021 2244 GetSystemName(system), id, sourceName.Data()));
9e080f92 2245 delete aResult;
2246 return 0;
2247 }
2bb7b766 2248
9e080f92 2249 if (aResult->GetRowCount() > 1) {
2250 Log(detector,
8da81210 2251 Form("GetFile - More than one entry in %s FXS db for: id = %s, source = %s",
9d733021 2252 GetSystemName(system), id, sourceName.Data()));
339adafa 2253 fFXSError = system;
9e080f92 2254 delete aResult;
2255 return 0;
2256 }
2257
9d733021 2258 if (aResult->GetFieldCount() != nFields) {
9e080f92 2259 Log(detector,
9d733021 2260 Form("GetFileName - Wrong field count in %s FXS db for: id = %s, source = %s",
2261 GetSystemName(system), id, sourceName.Data()));
339adafa 2262 fFXSError = system;
9e080f92 2263 delete aResult;
2264 return 0;
2265 }
2266
2267 TSQLRow* aRow = dynamic_cast<TSQLRow*> (aResult->Next());
2268
2269 if (!aRow){
8da81210 2270 Log(detector, Form("GetFile - Empty set result in %s FXS db from query: id = %s, source = %s",
9d733021 2271 GetSystemName(system), id, sourceName.Data()));
339adafa 2272 fFXSError = system;
9e080f92 2273 delete aResult;
2274 return 0;
2275 }
2276
2277 TString filePath(aRow->GetField(0), aRow->GetFieldLength(0));
2278 TString fileSize(aRow->GetField(1), aRow->GetFieldLength(1));
d386d623 2279 TString fileChecksum(aRow->GetField(2), aRow->GetFieldLength(2));
9e080f92 2280
2281 delete aResult;
2282 delete aRow;
2283
d386d623 2284 AliDebug(2, Form("filePath = %s; size = %s, fileChecksum = %s",
2285 filePath.Data(), fileSize.Data(), fileChecksum.Data()));
9e080f92 2286
9e080f92 2287 // retrieved file is renamed to make it unique
3ebbf3c4 2288 Int_t dir_lev1 = GetCurrentRun()/10000;
2289 TString localFileName = Form("%s/%d/%d/%s_process/%s_%s_%d_%s_%s.shuttle",
2290 GetShuttleTempDir(), dir_lev1, GetCurrentRun(), detector,
d524ade6 2291 GetSystemName(system), detector, GetCurrentRun(),
2292 id, sourceName.Data());
3ebbf3c4 2293 Log("SHUTTLE",Form("file from FXS = %s",localFileName.Data()));
9d733021 2294
9e080f92 2295
9d733021 2296 // file retrieval from FXS
4b95672b 2297 UInt_t nRetries = 0;
2298 UInt_t maxRetries = 3;
2299 Bool_t result = kFALSE;
2300
2301 // copy!! if successful TSystem::Exec returns 0
8da81210 2302 while (nRetries++ < maxRetries) {
4b95672b 2303 AliDebug(2, Form("Trying to copy file. Retry # %d", nRetries));
2304 result = RetrieveFile(system, filePath.Data(), localFileName.Data());
8da81210 2305 if (!result)
4b95672b 2306 {
8da81210 2307 Log(detector, Form("GetFile - Copy of file %s from %s FXS failed",
9d733021 2308 filePath.Data(), GetSystemName(system)));
4b95672b 2309 continue;
4f0749a8 2310 }
9e080f92 2311
c98417a1 2312 if (fileSize.Length()>0)
2313 {
2314 // compare filesize of local file with the one stored in the FXS DB
926cbc0e 2315 Long_t size = -1;
2316 Int_t sizeComp = gSystem->GetPathInfo(localFileName.Data(), 0, &size, 0, 0);
c98417a1 2317
926cbc0e 2318 if (sizeComp != 0 || size != fileSize.Atoi())
c98417a1 2319 {
8da81210 2320 Log(detector, Form("GetFile - size of file %s does not match with local copy!",
c98417a1 2321 filePath.Data()));
2322 result = kFALSE;
2323 continue;
2324 }
2325
2326 } else {
2327 Log(fCurrentDetector, Form("GetFile - size of file %s not set in %s database, skipping comparison",
2328 filePath.Data(), GetSystemName(system)));
2329 }
2330
d386d623 2331 if (fileChecksum.Length()>0)
4b95672b 2332 {
2333 // compare md5sum of local file with the one stored in the FXS DB
95d4aaef 2334 if(fileChecksum.Contains(' ')) fileChecksum.Resize(fileChecksum.First(' '));
339adafa 2335 Int_t md5Comp = gSystem->Exec(Form("md5sum %s |grep %s > /dev/null 2> /dev/null",
d524ade6 2336 localFileName.Data(), fileChecksum.Data()));
9e080f92 2337
4b95672b 2338 if (md5Comp != 0)
2339 {
8da81210 2340 Log(detector, Form("GetFile - md5sum of file %s does not match with local copy!",
4b95672b 2341 filePath.Data()));
2342 result = kFALSE;
2343 continue;
2344 }
d386d623 2345 } else {
2346 Log(fCurrentDetector, Form("GetFile - md5sum of file %s not set in %s database, skipping comparison",
2347 filePath.Data(), GetSystemName(system)));
9d733021 2348 }
4b95672b 2349 if (result) break;
9e080f92 2350 }
2351
8da81210 2352 if (!result)
2353 {
339adafa 2354 fFXSError = system;
8da81210 2355 return 0;
2356 }
4b95672b 2357
9d733021 2358 fFXSCalled[system]=kTRUE;
2359 TObjString *fileParams = new TObjString(Form("%s#!?!#%s", id, sourceName.Data()));
2360 fFXSlist[system].Add(fileParams);
9e080f92 2361
675f64cd 2362 static TString staticLocalFileName;
2363 staticLocalFileName.Form("%s", localFileName.Data());
2364
c88ad5db 2365 Log(fCurrentDetector, Form("GetFile - Retrieved file with id %s and "
2366 "source %s from %s to %s", id, source,
d524ade6 2367 GetSystemName(system), localFileName.Data()));
675f64cd 2368
d524ade6 2369 return staticLocalFileName.Data();
2bb7b766 2370}
2371
2372//______________________________________________________________________________________________
9d733021 2373Bool_t AliShuttle::RetrieveFile(UInt_t system, const char* fxsFileName, const char* localFileName)
9e080f92 2374{
9827400b 2375 //
2376 // Copies file from FXS to local Shuttle machine
2377 //
2bb7b766 2378
9e080f92 2379 // check temp directory: trying to cd to temp; if it does not exist, create it
d524ade6 2380 AliDebug(2, Form("Copy file %s from %s FXS into %s",
2381 GetSystemName(system), fxsFileName, localFileName));
2382
2383 TString tmpDir(localFileName);
2384
2385 tmpDir = tmpDir(0,tmpDir.Last('/'));
9e080f92 2386
d524ade6 2387 Int_t noDir = gSystem->GetPathInfo(tmpDir.Data(), 0, (Long64_t*) 0, 0, 0);
2388 if (noDir) // temp dir does not exists!
2389 {
2390 if (gSystem->mkdir(tmpDir.Data(), 1))
2391 {
2392 Log(fCurrentDetector.Data(), "RetrieveFile - could not make temp directory!!");
9e080f92 2393 return kFALSE;
2394 }
9e080f92 2395 }
2396
339adafa 2397 TString command = Form("scp -oPort=%d -2 %s@%s:%s/%s %s",
9d733021 2398 fConfig->GetFXSPort(system),
2399 fConfig->GetFXSUser(system),
2400 fConfig->GetFXSHost(system),
339adafa 2401 fConfig->GetFXSBaseFolder(system),
9d733021 2402 fxsFileName,
9e080f92 2403 localFileName);
2404
2405 AliDebug(2, Form("%s",command.Data()));
2406
4b95672b 2407 Bool_t result = (gSystem->Exec(command.Data()) == 0);
9e080f92 2408
4b95672b 2409 return result;
9e080f92 2410}
2411
2412//______________________________________________________________________________________________
9d733021 2413TList* AliShuttle::GetFileSources(Int_t system, const char* detector, const char* id)
2414{
9827400b 2415 //
2416 // Get sources producing the condition file Id from file exchange servers
4a33bdd9 2417 // if id is NULL all sources are returned (distinct)
9827400b 2418 //
1bcd28db 2419
8884d71e 2420 if (id)
2421 {
2422 Log(detector, Form("GetFileSources - Querying %s FXS for files with id %s produced by %s", GetSystemName(system), id, detector));
2423 } else {
2424 Log(detector, Form("GetFileSources - Querying %s FXS for files produced by %s", GetSystemName(system), detector));
2425 }
9827400b 2426
2427 // check if test mode should simulate a FXS error
2428 if (fTestMode & kErrorFXSSources)
2429 {
2430 Log(detector, Form("GetFileSources - In TESTMODE - Simulating error while connecting to %s FXS", GetSystemName(system)));
2431 return 0;
2432 }
2433
9d733021 2434 if (system == kDCS)
2435 {
c88ad5db 2436 Log(detector, "GetFileSources - WARNING: DCS system has only one source of data!");
6297b37d 2437 TList *list = new TList();
2438 list->SetOwner(1);
2439 list->Add(new TObjString(" "));
2440 return list;
9d733021 2441 }
9e080f92 2442
2443 // check connection, in case connect
9d733021 2444 if (!Connect(system))
2445 {
4a33bdd9 2446 Log(detector, Form("GetFileSources - Couldn't connect to %s FXS database", GetSystemName(system)));
339adafa 2447 fFXSError = system;
9d733021 2448 return NULL;
9e080f92 2449 }
2450
b827f957 2451 TString sourceName = "";
efec19bd 2452 if (system == kDAQ || system == kDQM)
9d733021 2453 {
2454 sourceName = "DAQsource";
2455 } else if (system == kHLT)
2456 {
2457 sourceName = "DDLnumbers";
2458 }
2459
4a33bdd9 2460 TString sqlQueryStart = Form("select distinct %s from %s where", sourceName.Data(), fConfig->GetFXSdbTable(system));
2461 TString whereClause = Form("run=%d and detector=\"%s\"",
2462 GetCurrentRun(), detector);
2463 if (id)
2464 whereClause += Form(" and fileId=\"%s\"", id);
9e080f92 2465 TString sqlQuery = Form("%s %s", sqlQueryStart.Data(), whereClause.Data());
2466
2467 AliDebug(2, Form("SQL query: \n%s",sqlQuery.Data()));
2468
2469 // Query execution
2470 TSQLResult* aResult;
9d733021 2471 aResult = fServer[system]->Query(sqlQuery);
9e080f92 2472 if (!aResult) {
9d733021 2473 Log(detector, Form("GetFileSources - Can't execute SQL query to %s database for id: %s",
2474 GetSystemName(system), id));
339adafa 2475 fFXSError = system;
9e080f92 2476 return 0;
2477 }
2478
86aa42c3 2479 TList *list = new TList();
2480 list->SetOwner(1);
2481
9d733021 2482 if (aResult->GetRowCount() == 0)
2483 {
9e080f92 2484 Log(detector,
9d733021 2485 Form("GetFileSources - No entry in %s FXS table for id: %s", GetSystemName(system), id));
9e080f92 2486 delete aResult;
86aa42c3 2487 return list;
9e080f92 2488 }
2489
1bcd28db 2490 Log(detector, Form("GetFileSources - Found %d sources", aResult->GetRowCount()));
9e080f92 2491
1bcd28db 2492 TSQLRow* aRow;
9d733021 2493 while ((aRow = aResult->Next()))
2494 {
9e080f92 2495
9d733021 2496 TString source(aRow->GetField(0), aRow->GetFieldLength(0));
2497 AliDebug(2, Form("%s = %s", sourceName.Data(), source.Data()));
2498 list->Add(new TObjString(source));
9e080f92 2499 delete aRow;
2500 }
9d733021 2501
9e080f92 2502 delete aResult;
2503
2504 return list;
2bb7b766 2505}
2506
4a33bdd9 2507//______________________________________________________________________________________________
2508TList* AliShuttle::GetFileIDs(Int_t system, const char* detector, const char* source)
2509{
2510 //
2511 // Get all ids of condition files produced by a given source from file exchange servers
2512 //
2513
1bcd28db 2514 Log(detector, Form("GetFileIDs - Retrieving ids with source %s with %s", source, GetSystemName(system)));
2515
4a33bdd9 2516 // check if test mode should simulate a FXS error
2517 if (fTestMode & kErrorFXSSources)
2518 {
2519 Log(detector, Form("GetFileIDs - In TESTMODE - Simulating error while connecting to %s FXS", GetSystemName(system)));
2520 return 0;
2521 }
2522
2523 // check connection, in case connect
2524 if (!Connect(system))
2525 {
2526 Log(detector, Form("GetFileIDs - Couldn't connect to %s FXS database", GetSystemName(system)));
2527 return NULL;
2528 }
2529
b827f957 2530 TString sourceName = "";
4a33bdd9 2531 if (system == kDAQ)
2532 {
2533 sourceName = "DAQsource";
2534 } else if (system == kHLT)
2535 {
2536 sourceName = "DDLnumbers";
2537 }
2538
2539 TString sqlQueryStart = Form("select fileId from %s where", fConfig->GetFXSdbTable(system));
2540 TString whereClause = Form("run=%d and detector=\"%s\"",
2541 GetCurrentRun(), detector);
2542 if (sourceName.Length() > 0 && source)
2543 whereClause += Form(" and %s=\"%s\"", sourceName.Data(), source);
2544 TString sqlQuery = Form("%s %s", sqlQueryStart.Data(), whereClause.Data());
2545
2546 AliDebug(2, Form("SQL query: \n%s",sqlQuery.Data()));
2547
2548 // Query execution
2549 TSQLResult* aResult;
2550 aResult = fServer[system]->Query(sqlQuery);
2551 if (!aResult) {
2552 Log(detector, Form("GetFileIDs - Can't execute SQL query to %s database for source: %s",
2553 GetSystemName(system), source));
2554 return 0;
2555 }
2556
2557 TList *list = new TList();
2558 list->SetOwner(1);
2559
2560 if (aResult->GetRowCount() == 0)
2561 {
2562 Log(detector,
2563 Form("GetFileIDs - No entry in %s FXS table for source: %s", GetSystemName(system), source));
2564 delete aResult;
2565 return list;
2566 }
2567
1bcd28db 2568 Log(detector, Form("GetFileIDs - Found %d ids", aResult->GetRowCount()));
2569
4a33bdd9 2570 TSQLRow* aRow;
2571
2572 while ((aRow = aResult->Next()))
2573 {
2574
2575 TString id(aRow->GetField(0), aRow->GetFieldLength(0));
2576 AliDebug(2, Form("fileId = %s", id.Data()));
2577 list->Add(new TObjString(id));
2578 delete aRow;
2579 }
2580
2581 delete aResult;
2582
2583 return list;
2584}
2585
2bb7b766 2586//______________________________________________________________________________________________
9d733021 2587Bool_t AliShuttle::Connect(Int_t system)
2bb7b766 2588{
9827400b 2589 // Connect to MySQL Server of the system's FXS MySQL databases
2590 // DAQ Logbook, Shuttle Logbook and DAQ FXS db are on the same host
2591 //
57f50b3c 2592
9d733021 2593 // check connection: if already connected return
dd84c76d 2594
3a4ddf0e 2595 if(fServer[system] && fServer[system]->IsConnected()) {
dd84c76d 2596 // ping the server
2597 if (fServer[system]->PingVerify()==kTRUE){ // connection is still alive
2598 return kTRUE;
2599 }
2600 else{
2601 AliWarning(Form("Connection got lost to FXS database for %s. Closing and reconnecting.",
2602 AliShuttleInterface::GetSystemName(system)));
2603 fServer[system]->Close();
2604 delete fServer[system];
2605 fServer[system] = 0x0;
2606 }
3a4ddf0e 2607 }
57f50b3c 2608
9d733021 2609 TString dbHost, dbUser, dbPass, dbName;
57f50b3c 2610
efec19bd 2611 if (system < 4) // FXS db servers
9d733021 2612 {
2613 dbHost = Form("mysql://%s:%d", fConfig->GetFXSdbHost(system), fConfig->GetFXSdbPort(system));
2614 dbUser = fConfig->GetFXSdbUser(system);
2615 dbPass = fConfig->GetFXSdbPass(system);
2616 dbName = fConfig->GetFXSdbName(system);
2617 } else { // Run & Shuttle logbook servers
2618 // TODO Will the Shuttle logbook server be the same as the Run logbook server ???
2619 dbHost = Form("mysql://%s:%d", fConfig->GetDAQlbHost(), fConfig->GetDAQlbPort());
2620 dbUser = fConfig->GetDAQlbUser();
2621 dbPass = fConfig->GetDAQlbPass();
2622 dbName = fConfig->GetDAQlbDB();
2623 }
57f50b3c 2624
9d733021 2625 fServer[system] = TSQLServer::Connect(dbHost.Data(), dbUser.Data(), dbPass.Data());
efec19bd 2626 if (!fServer[system] || !fServer[system]->IsConnected()) {
2627 if(system < 4)
9d733021 2628 {
2629 AliError(Form("Can't establish connection to FXS database for %s",
2630 AliShuttleInterface::GetSystemName(system)));
2631 } else {
2632 AliError("Can't establish connection to Run logbook.");
57f50b3c 2633 }
9d733021 2634 if(fServer[system]) delete fServer[system];
2635 return kFALSE;
2bb7b766 2636 }
57f50b3c 2637
9d733021 2638 // Get tables
2639 TSQLResult* aResult=0;
2640 switch(system){
2641 case kDAQ:
2642 aResult = fServer[kDAQ]->GetTables(dbName.Data());
2643 break;
2644 case kDCS:
2645 aResult = fServer[kDCS]->GetTables(dbName.Data());
2646 break;
2647 case kHLT:
2648 aResult = fServer[kHLT]->GetTables(dbName.Data());
2649 break;
efec19bd 2650 case kDQM:
2651 aResult = fServer[kDQM]->GetTables(dbName.Data());
2652 break;
9d733021 2653 default:
efec19bd 2654 aResult = fServer[4]->GetTables(dbName.Data());
9d733021 2655 break;
2656 }
2657
2658 delete aResult;
2bb7b766 2659 return kTRUE;
2660}
57f50b3c 2661
9e080f92 2662//______________________________________________________________________________________________
9d733021 2663Bool_t AliShuttle::UpdateTable()
9e080f92 2664{
9827400b 2665 //
2666 // Update FXS table filling time_processed field in all rows corresponding to current run and detector
2667 //
9e080f92 2668
9d733021 2669 Bool_t result = kTRUE;
9e080f92 2670
6e25f828 2671 for (UInt_t system=0; system<4; system++)
9d733021 2672 {
2673 if(!fFXSCalled[system]) continue;
9e080f92 2674
9d733021 2675 // check connection, in case connect
2676 if (!Connect(system))
2677 {
2678 Log(fCurrentDetector, Form("UpdateTable - Couldn't connect to %s FXS database", GetSystemName(system)));
2679 result = kFALSE;
2680 continue;
9e080f92 2681 }
9e080f92 2682
9d733021 2683 TTimeStamp now; // now
2684
2685 // Loop on FXS list entries
2686 TIter iter(&fFXSlist[system]);
2687 TObjString *aFXSentry=0;
2688 while ((aFXSentry = dynamic_cast<TObjString*> (iter.Next())))
2689 {
2690 TString aFXSentrystr = aFXSentry->String();
2691 TObjArray *aFXSarray = aFXSentrystr.Tokenize("#!?!#");
2692 if (!aFXSarray || aFXSarray->GetEntries() != 2 )
2693 {
2694 Log(fCurrentDetector, Form("UpdateTable - error updating %s FXS entry. Check string: <%s>",
2695 GetSystemName(system), aFXSentrystr.Data()));
2696 if(aFXSarray) delete aFXSarray;
2697 result = kFALSE;
2698 continue;
2699 }
2700 const char* fileId = ((TObjString*) aFXSarray->At(0))->GetName();
2701 const char* source = ((TObjString*) aFXSarray->At(1))->GetName();
2702
2703 TString whereClause;
efec19bd 2704 if (system == kDAQ || system == kDQM)
9d733021 2705 {
2706 whereClause = Form("where run=%d and detector=\"%s\" and fileId=\"%s\" and DAQsource=\"%s\";",
2707 GetCurrentRun(), fCurrentDetector.Data(), fileId, source);
2708 }
2709 else if (system == kDCS)
2710 {
2711 whereClause = Form("where run=%d and detector=\"%s\" and fileId=\"%s\";",
2712 GetCurrentRun(), fCurrentDetector.Data(), fileId);
2713 }
2714 else if (system == kHLT)
2715 {
2716 whereClause = Form("where run=%d and detector=\"%s\" and fileId=\"%s\" and DDLnumbers=\"%s\";",
2717 GetCurrentRun(), fCurrentDetector.Data(), fileId, source);
2718 }
2719
2720 delete aFXSarray;
9e080f92 2721
0a0b087a 2722 TString sqlQuery = Form("update %s set time_processed=%ld %s", fConfig->GetFXSdbTable(system),
2723 (ULong_t)now.GetSec(), whereClause.Data());
9e080f92 2724
9d733021 2725 AliDebug(2, Form("SQL query: \n%s",sqlQuery.Data()));
9e080f92 2726
9d733021 2727 // Query execution
2728 TSQLResult* aResult;
2729 aResult = dynamic_cast<TSQLResult*> (fServer[system]->Query(sqlQuery));
2730 if (!aResult)
2731 {
2732 Log(fCurrentDetector, Form("UpdateTable - %s db: can't execute SQL query <%s>",
2733 GetSystemName(system), sqlQuery.Data()));
2734 result = kFALSE;
2735 continue;
2736 }
2737 delete aResult;
9e080f92 2738 }
9e080f92 2739 }
2740
9d733021 2741 return result;
9e080f92 2742}
57f50b3c 2743
bd23d48b 2744//_______________________________________________________________________________
2745Bool_t AliShuttle::UpdateTableSkippedCase(const char* detector)
2746{
2747 //
2748 // Update FXS table filling time_processed field in all rows corresponding to current run and detector
2749 // if detector = "ALL" update all detectors
2750 //
2751
2752 Bool_t result = kTRUE;
2753
7a782903 2754 TString detName(detector);
2755
6e25f828 2756 for (UInt_t system=0; system<4; system++)
bd23d48b 2757 {
2758
2759 // check connection, in case connect
2760 if (!Connect(system))
2761 {
2762 Log(fCurrentDetector, Form("UpdateTableSkippedCase - Couldn't connect to %s FXS database", GetSystemName(system)));
2763 result = kFALSE;
2764 continue;
2765 }
2766
2767 TTimeStamp now; // now
2768
2769 // Loop on FXS list entries
2770 TIter iter(&fFXSlist[system]);
2771
2772 TString whereClause;
7a782903 2773 if (detName == "ALL") whereClause = Form("where run=%d and time_processed IS NULL;",GetCurrentRun());
bd23d48b 2774 else whereClause = Form("where run=%d and detector=\"%s\" and time_processed IS NULL;",GetCurrentRun(), detector);
2775
4be2f262 2776 //Log("SHUTTLE",Form(" whereClause = %s ",whereClause.Data()));
bd23d48b 2777
0a0b087a 2778 TString sqlQuery = Form("update %s set time_processed=%ld %s", fConfig->GetFXSdbTable(system),
2779 (ULong_t)now.GetSec(), whereClause.Data());
bd23d48b 2780
2781 AliDebug(2, Form("SQL query: \n%s",sqlQuery.Data()));
2782
2783 // Query execution
2784 TSQLResult* aResult;
2785 aResult = dynamic_cast<TSQLResult*> (fServer[system]->Query(sqlQuery));
2786 if (!aResult)
2787 {
2788 Log("SHUTTLE", Form("UpdateTableSkippedCase - %s db: can't execute SQL query <%s>",
2789 GetSystemName(system), sqlQuery.Data()));
2790 result = kFALSE;
2791 continue;
2792 }
2793 delete aResult;
2794
2795 }
2796
2797 return result;
2798}
3301427a 2799//______________________________________________________________________________________________
2800Bool_t AliShuttle::UpdateTableFailCase()
2801{
9827400b 2802 // Update FXS table filling time_processed field in all rows corresponding to current run and detector
2803 // this is called in case the preprocessor is declared failed for the current run, because
2804 // the fields are updated only in case of success
3301427a 2805
2806 Bool_t result = kTRUE;
2807
6e25f828 2808 for (UInt_t system=0; system<4; system++)
3301427a 2809 {
2810 // check connection, in case connect
2811 if (!Connect(system))
2812 {
2813 Log(fCurrentDetector, Form("UpdateTableFailCase - Couldn't connect to %s FXS database",
2814 GetSystemName(system)));
2815 result = kFALSE;
2816 continue;
2817 }
2818
2819 TTimeStamp now; // now
2820
2821 // Loop on FXS list entries
2822
2823 TString whereClause = Form("where run=%d and detector=\"%s\";",
2824 GetCurrentRun(), fCurrentDetector.Data());
2825
2826
0a0b087a 2827 TString sqlQuery = Form("update %s set time_processed=%ld %s", fConfig->GetFXSdbTable(system),
2828 (ULong_t)now.GetSec(), whereClause.Data());
3301427a 2829
2830 AliDebug(2, Form("SQL query: \n%s",sqlQuery.Data()));
2831
2832 // Query execution
2833 TSQLResult* aResult;
2834 aResult = dynamic_cast<TSQLResult*> (fServer[system]->Query(sqlQuery));
2835 if (!aResult)
2836 {
2837 Log(fCurrentDetector, Form("UpdateTableFailCase - %s db: can't execute SQL query <%s>",
2838 GetSystemName(system), sqlQuery.Data()));
2839 result = kFALSE;
2840 continue;
2841 }
2842 delete aResult;
2843 }
2844
2845 return result;
2846}
2847
2bb7b766 2848//______________________________________________________________________________________________
2849Bool_t AliShuttle::UpdateShuttleLogbook(const char* detector, const char* status)
2850{
e7f62f16 2851 //
2852 // Update Shuttle logbook filling detector or shuttle_done column
2853 // ex. of usage: UpdateShuttleLogbook("PHOS", "DONE") or UpdateShuttleLogbook("shuttle_done")
2854 //
57f50b3c 2855
2bb7b766 2856 // check connection, in case connect
efec19bd 2857 if(!Connect(4)){
2bb7b766 2858 Log("SHUTTLE", "UpdateShuttleLogbook - Couldn't connect to DAQ Logbook.");
2859 return kFALSE;
57f50b3c 2860 }
2861
2bb7b766 2862 TString detName(detector);
2863 TString setClause;
8e828d39 2864 if (detName == "shuttle_done" || detName == "shuttle_skipped")
e7f62f16 2865 {
2bb7b766 2866 setClause = "set shuttle_done=1";
8e828d39 2867
ee6f7523 2868 if (detName == "shuttle_done")
b0e53b15 2869 {
8e828d39 2870 if (TouchFile() != kTRUE)
2156c755 2871 {
2872 SendMLRunInfo("Pending");
503ff24f 2873 return kFALSE;
2156c755 2874 }
8e828d39 2875
2876 SendMLRunInfo("Done");
b0e53b15 2877 }
8e828d39 2878 else
2879 SendMLRunInfo("Skipped");
2880 }
2881 else {
2bb7b766 2882 TString statusStr(status);
2883 if(statusStr.Contains("done", TString::kIgnoreCase) ||
2884 statusStr.Contains("failed", TString::kIgnoreCase)){
eba76848 2885 setClause = Form("set %s=\"%s\"", detector, status);
2bb7b766 2886 } else {
2887 Log("SHUTTLE",
2888 Form("UpdateShuttleLogbook - Invalid status <%s> for detector %s",
2889 status, detector));
2890 return kFALSE;
2891 }
2892 }
57f50b3c 2893
2bb7b766 2894 TString whereClause = Form("where run=%d", GetCurrentRun());
2895
441b0e9c 2896 TString sqlQuery = Form("update %s %s %s",
2897 fConfig->GetShuttlelbTable(), setClause.Data(), whereClause.Data());
57f50b3c 2898
2bb7b766 2899 AliDebug(2, Form("SQL query: \n%s",sqlQuery.Data()));
2900
2901 // Query execution
2902 TSQLResult* aResult;
efec19bd 2903 aResult = dynamic_cast<TSQLResult*> (fServer[4]->Query(sqlQuery));
2bb7b766 2904 if (!aResult) {
2905 Log("SHUTTLE", Form("UpdateShuttleLogbook - Can't execute query <%s>", sqlQuery.Data()));
2906 return kFALSE;
57f50b3c 2907 }
2bb7b766 2908 delete aResult;
57f50b3c 2909
2910 return kTRUE;
2911}
2912
2913//______________________________________________________________________________________________
2bb7b766 2914Int_t AliShuttle::GetCurrentRun() const
2915{
9827400b 2916 //
2917 // Get current run from logbook entry
2918 //
57f50b3c 2919
2bb7b766 2920 return fLogbookEntry ? fLogbookEntry->GetRun() : -1;
57f50b3c 2921}
2922
2923//______________________________________________________________________________________________
2bb7b766 2924UInt_t AliShuttle::GetCurrentStartTime() const
2925{
9827400b 2926 //
2927 // get current start time
2928 //
57f50b3c 2929
2bb7b766 2930 return fLogbookEntry ? fLogbookEntry->GetStartTime() : 0;
57f50b3c 2931}
2932
2933//______________________________________________________________________________________________
2bb7b766 2934UInt_t AliShuttle::GetCurrentEndTime() const
2935{
9827400b 2936 //
2937 // get current end time from logbook entry
2938 //
57f50b3c 2939
2bb7b766 2940 return fLogbookEntry ? fLogbookEntry->GetEndTime() : 0;
57f50b3c 2941}
675f64cd 2942//______________________________________________________________________________________________
2943UInt_t AliShuttle::GetCurrentYear() const
2944{
2945 //
2946 // Get current year from logbook entry
2947 //
2948
2949 if (!fLogbookEntry) return 0;
2950
2951 TTimeStamp startTime(GetCurrentStartTime());
2952 TString year = Form("%d",startTime.GetDate());
2953 year = year(0,4);
2954
2955 return year.Atoi();
2956}
2957
2958//______________________________________________________________________________________________
2959const char* AliShuttle::GetLHCPeriod() const
2960{
2961 //
2962 // Get current LHC period from logbook entry
2963 //
2964
2965 if (!fLogbookEntry) return 0;
2966
2967 return fLogbookEntry->GetRunParameter("LHCperiod");
2968}
2969
b948db8d 2970//______________________________________________________________________________________________
6a150f83 2971void AliShuttle::Log(const char* detector, const char* message, UInt_t level)
b948db8d 2972{
9827400b 2973 //
2974 // Fill log string with a message
2975 //
7d4cf768 2976
3ebbf3c4 2977 TString logRunDir = GetShuttleLogDir();
2978 if (GetCurrentRun() >=0) {
2979 Int_t logDir_lev1 = GetCurrentRun()/10000;
2980 logRunDir += Form("/%d/%d", logDir_lev1, GetCurrentRun());
2981 }
7d4cf768 2982 void* dir = gSystem->OpenDirectory(logRunDir.Data());
84090f85 2983 if (dir == NULL) {
7d4cf768 2984 if (gSystem->mkdir(logRunDir.Data(), kTRUE)) {
36c99a6a 2985 AliError(Form("Can't open directory <%s>", GetShuttleLogDir()));
84090f85 2986 return;
2987 }
b948db8d 2988
84090f85 2989 } else {
2990 gSystem->FreeDirectory(dir);
2991 }
b948db8d 2992
0b025a52 2993 TString toLog = Form("%s UTC (%d): %s - ", TTimeStamp(time(0)).AsString("s"), getpid(), detector);
e7f62f16 2994 if (GetCurrentRun() >= 0)
2995 toLog += Form("run %d - ", GetCurrentRun());
2bb7b766 2996 toLog += Form("%s", message);
2997
6a150f83 2998 AliLog::Message(level, toLog, MODULENAME(), ClassName(), FUNCTIONNAME(), __FILE__, __LINE__);
ffa29e93 2999
3000 // if we redirect the log output already to the file, leave here
3001 if (fOutputRedirected && strcmp(detector, "SHUTTLE") != 0)
3002 return;
b948db8d 3003
ffa29e93 3004 TString fileName = GetLogFileName(detector);
e7f62f16 3005
84090f85 3006 gSystem->ExpandPathName(fileName);
3007
3008 ofstream logFile;
3009 logFile.open(fileName, ofstream::out | ofstream::app);
3010
3011 if (!logFile.is_open()) {
3012 AliError(Form("Could not open file %s", fileName.Data()));
3013 return;
3014 }
7bfb2090 3015
84090f85 3016 logFile << toLog.Data() << "\n";
b948db8d 3017
84090f85 3018 logFile.close();
b948db8d 3019}
2bb7b766 3020
ffa29e93 3021//______________________________________________________________________________________________
3022TString AliShuttle::GetLogFileName(const char* detector) const
3023{
3024 //
3025 // returns the name of the log file for a given sub detector
3026 //
3027
3028 TString fileName;
3029
3030 if (GetCurrentRun() >= 0)
7d4cf768 3031 {
3ebbf3c4 3032 Int_t logDir_lev1 = GetCurrentRun()/10000;
3033 fileName.Form("%s/%d/%d/%s.log", GetShuttleLogDir(), logDir_lev1, GetCurrentRun(),
3034 detector);
7d4cf768 3035 } else {
ffa29e93 3036 fileName.Form("%s/%s.log", GetShuttleLogDir(), detector);
7d4cf768 3037 }
ffa29e93 3038
3039 return fileName;
3040}
3041
ee6f7523 3042//______________________________________________________________________________________________
3043void AliShuttle::SendAlive()
3044{
3045 // sends alive message to ML
3046
3047 TMonaLisaText mlStatus("SHUTTLE_status", "Alive");
3048
3049 TList mlList;
3050 mlList.Add(&mlStatus);
3051
3052 fMonaLisa->SendParameters(&mlList, "__PROCESSINGINFO__");
3053}
3054
2bb7b766 3055//______________________________________________________________________________________________
3056Bool_t AliShuttle::Collect(Int_t run)
3057{
9827400b 3058 //
3059 // Collects conditions data for all UNPROCESSED run written to DAQ LogBook in case of run = -1 (default)
3060 // If a dedicated run is given this run is processed
3061 //
3062 // In operational mode, this is the Shuttle function triggered by the EOR signal.
3063 //
2bb7b766 3064
eba76848 3065 if (run == -1)
3066 Log("SHUTTLE","Collect - Shuttle called. Collecting conditions data for unprocessed runs");
3067 else
3068 Log("SHUTTLE", Form("Collect - Shuttle called. Collecting conditions data for run %d", run));
cb343cfd 3069
3070 SetLastAction("Starting");
2bb7b766 3071
ee6f7523 3072 // create ML instance
3073 if (!fMonaLisa)
3074 fMonaLisa = new TMonaLisaWriter(fConfig->GetMonitorHost(), fConfig->GetMonitorTable());
3075
fbc112e3 3076 CountOpenRuns();
ee6f7523 3077
2bb7b766 3078 TString whereClause("where shuttle_done=0");
eba76848 3079 if (run != -1)
3080 whereClause += Form(" and run=%d", run);
2bb7b766 3081
3082 TObjArray shuttleLogbookEntries;
be48e3ea 3083 if (!QueryShuttleLogbook(whereClause, shuttleLogbookEntries))
3084 {
cb343cfd 3085 Log("SHUTTLE", "Collect - Can't retrieve entries from Shuttle logbook");
2bb7b766 3086 return kFALSE;
3087 }
3088
9e080f92 3089 if (shuttleLogbookEntries.GetEntries() == 0)
3090 {
3091 if (run == -1)
3092 Log("SHUTTLE","Collect - Found no UNPROCESSED runs in Shuttle logbook");
3093 else
3094 Log("SHUTTLE", Form("Collect - Run %d is already DONE "
3095 "or it does not exist in Shuttle logbook", run));
3096 return kTRUE;
3097 }
3098
be48e3ea 3099 for (UInt_t iDet=0; iDet<NDetectors(); iDet++)
3100 fFirstUnprocessed[iDet] = kTRUE;
3101
fc5a4708 3102 if (run != -1)
be48e3ea 3103 {
3104 // query Shuttle logbook for earlier runs, check if some detectors are unprocessed,
3105 // flag them into fFirstUnprocessed array
e3ae8d33 3106 TString whereClauseBis(Form("where shuttle_done=0 and run < %d", run));
be48e3ea 3107 TObjArray tmpLogbookEntries;
e3ae8d33 3108 if (!QueryShuttleLogbook(whereClauseBis, tmpLogbookEntries))
be48e3ea 3109 {
3110 Log("SHUTTLE", "Collect - Can't retrieve entries from Shuttle logbook");
3111 return kFALSE;
3112 }
3113
3114 TIter iter(&tmpLogbookEntries);
3115 AliShuttleLogbookEntry* anEntry = 0;
3116 while ((anEntry = dynamic_cast<AliShuttleLogbookEntry*> (iter.Next())))
3117 {
3118 for (UInt_t iDet=0; iDet<NDetectors(); iDet++)
3119 {
3120 if (anEntry->GetDetectorStatus(iDet) == AliShuttleLogbookEntry::kUnprocessed)
3121 {
3122 AliDebug(2, Form("Run %d: setting %s as \"not first time unprocessed\"",
3123 anEntry->GetRun(), GetDetName(iDet)));
3124 fFirstUnprocessed[iDet] = kFALSE;
3125 }
3126 }
3127
3128 }
3129
3130 }
3131
3132 if (!RetrieveConditionsData(shuttleLogbookEntries))
3133 {
cb343cfd 3134 Log("SHUTTLE", "Collect - Process of at least one run failed");
a3ca69fe 3135 CountOpenRuns();
2bb7b766 3136 return kFALSE;
3137 }
3138
36c99a6a 3139 Log("SHUTTLE", "Collect - Requested run(s) successfully processed");
a3ca69fe 3140 CountOpenRuns();
eba76848 3141 return kTRUE;
2bb7b766 3142}
3143
2bb7b766 3144//______________________________________________________________________________________________
3145Bool_t AliShuttle::RetrieveConditionsData(const TObjArray& dateEntries)
3146{
9827400b 3147 //
3148 // Retrieve conditions data for all runs that aren't processed yet
3149 //
2bb7b766 3150
3151 Bool_t hasError = kFALSE;
3152
3153 TIter iter(&dateEntries);
3154 AliShuttleLogbookEntry* anEntry;
3155
3156 while ((anEntry = (AliShuttleLogbookEntry*) iter.Next())){
3157 if (!Process(anEntry)){
3158 hasError = kTRUE;
3159 }
4b95672b 3160
3161 // clean SHUTTLE temp directory
d524ade6 3162 //TString filename = Form("%s/*.shuttle", GetShuttleTempDir());
3163 //RemoveFile(filename.Data());
2bb7b766 3164 }
3165
3166 return hasError == kFALSE;
3167}
cb343cfd 3168
3169//______________________________________________________________________________________________
3170ULong_t AliShuttle::GetTimeOfLastAction() const
3171{
9827400b 3172 //
3173 // Gets time of last action
3174 //
3175
cb343cfd 3176 ULong_t tmp;
36c99a6a 3177
cb343cfd 3178 fMonitoringMutex->Lock();
be48e3ea 3179
cb343cfd 3180 tmp = fLastActionTime;
36c99a6a 3181
cb343cfd 3182 fMonitoringMutex->UnLock();
36c99a6a 3183
cb343cfd 3184 return tmp;
3185}
3186
3187//______________________________________________________________________________________________
3188const TString AliShuttle::GetLastAction() const
3189{
9827400b 3190 //
cb343cfd 3191 // returns a string description of the last action
9827400b 3192 //
cb343cfd 3193
3194 TString tmp;
36c99a6a 3195
cb343cfd 3196 fMonitoringMutex->Lock();
3197
3198 tmp = fLastAction;
3199
3200 fMonitoringMutex->UnLock();
3201
36c99a6a 3202 return tmp;
cb343cfd 3203}
3204
3205//______________________________________________________________________________________________
3206void AliShuttle::SetLastAction(const char* action)
3207{
9827400b 3208 //
cb343cfd 3209 // updates the monitoring variables
9827400b 3210 //
36c99a6a 3211
cb343cfd 3212 fMonitoringMutex->Lock();
36c99a6a 3213
cb343cfd 3214 fLastAction = action;
3215 fLastActionTime = time(0);
3216
3217 fMonitoringMutex->UnLock();
3218}
eba76848 3219
3220//______________________________________________________________________________________________
3221const char* AliShuttle::GetRunParameter(const char* param)
3222{
9827400b 3223 //
3224 // returns run parameter read from DAQ logbook
3225 //
eba76848 3226
3227 if(!fLogbookEntry) {
3228 AliError("No logbook entry!");
3229 return 0;
3230 }
3231
3232 return fLogbookEntry->GetRunParameter(param);
3233}
57c1a579 3234
d386d623 3235//______________________________________________________________________________________________
9827400b 3236AliCDBEntry* AliShuttle::GetFromOCDB(const char* detector, const AliCDBPath& path)
d386d623 3237{
9827400b 3238 //
3239 // returns object from OCDB valid for current run
3240 //
d386d623 3241
9827400b 3242 if (fTestMode & kErrorOCDB)
3243 {
3244 Log(detector, "GetFromOCDB - In TESTMODE - Simulating error with OCDB");
3245 return 0;
3246 }
3247
d386d623 3248 AliCDBStorage *sto = AliCDBManager::Instance()->GetStorage(fgkMainCDB);
3249 if (!sto)
3250 {
9827400b 3251 Log(detector, "GetFromOCDB - Cannot activate main OCDB for query!");
d386d623 3252 return 0;
3253 }
3254
3255 return dynamic_cast<AliCDBEntry*> (sto->Get(path, GetCurrentRun()));
3256}
3257
57c1a579 3258//______________________________________________________________________________________________
339adafa 3259Bool_t AliShuttle::SendMail(EMailTarget target, Int_t system)
57c1a579 3260{
9827400b 3261 //
3262 // sends a mail to the subdetector expert in case of preprocessor error
3263 //
3264
3265 if (fTestMode != kNone)
3266 return kTRUE;
fb40ead4 3267
926cbc0e 3268 if (!fConfig->SendMail())
3269 return kTRUE;
57c1a579 3270
339adafa 3271 if (target == kDCSEMail || target == kFXSEMail) {
3272 if (!fFirstProcessing)
3a4ddf0e 3273 return kTRUE;
36c99a6a 3274 }
3275
7bdfe1fa 3276 Int_t runMode = (Int_t)fConfig->GetRunMode();
3277 TString tmpStr;
3278 if (runMode == 0) tmpStr = " Nightly Test:";
3279 else tmpStr = " Data Taking:";
fbc112e3 3280 void* dir = gSystem->OpenDirectory(GetShuttleLogDir());
3281 if (dir == NULL)
3282 {
3283 if (gSystem->mkdir(GetShuttleLogDir(), kTRUE))
3284 {
3285 Log("SHUTTLE", Form("SendMail - Can't open directory <%s>", GetShuttleLogDir()));
3286 return kFALSE;
3287 }
3288
3289 } else {
3290 gSystem->FreeDirectory(dir);
3291 }
3292
339adafa 3293 // det experts in to
3294 TString to="";
3835a613 3295 TIter *iterExperts = 0;
339adafa 3296 if (target == kDCSEMail) {
3297 iterExperts = new TIter(fConfig->GetAdmins(AliShuttleConfig::kAmanda));
3298 }
3299 else if (target == kFXSEMail) {
3300 iterExperts = new TIter(fConfig->GetAdmins(system));
3301 }
3835a613 3302 if (iterExperts) {
3303 TObjString *anExpert=0;
3304 while ((anExpert = (TObjString*) iterExperts->Next()))
3305 {
3306 to += Form("%s,", anExpert->GetName());
3307 }
3308 delete iterExperts;
339adafa 3309 }
3835a613 3310
3311 // add subdetector experts
3312 iterExperts = new TIter(fConfig->GetResponsibles(fCurrentDetector));
339adafa 3313 TObjString *anExpert=0;
3314 while ((anExpert = (TObjString*) iterExperts->Next()))
4508ab8b 3315 {
339adafa 3316 to += Form("%s,", anExpert->GetName());
3317 }
3318 delete iterExperts;
3835a613 3319
339adafa 3320 if (to.Length() > 0)
3321 to.Remove(to.Length()-1);
3322 AliDebug(2, Form("to: %s",to.Data()));
4508ab8b 3323
339adafa 3324 if (to.IsNull()) {
3325 Log("SHUTTLE", Form("List of %d responsibles not set!", (Int_t) target));
3326 return kFALSE;
3327 }
3328
3329 // SHUTTLE responsibles in cc
fb40ead4 3330 TString cc="";
3331 TIter iterAdmins(fConfig->GetAdmins(AliShuttleConfig::kGlobal));
3332 TObjString *anAdmin=0;
3333 while ((anAdmin = (TObjString*) iterAdmins.Next()))
3334 {
3335 cc += Form("%s,", anAdmin->GetName());
3336 }
3337 if (cc.Length() > 0)
c19963db 3338 cc.Remove(cc.Length()-1);
fb40ead4 3339 AliDebug(2, Form("cc: %s",to.Data()));
57c1a579 3340
339adafa 3341 // mail body
675f64cd 3342 TString bodyFileName;
3343 bodyFileName.Form("%s/mail.body", GetShuttleLogDir());
3344 gSystem->ExpandPathName(bodyFileName);
3345
3346 ofstream mailBody;
3347 mailBody.open(bodyFileName, ofstream::out);
3348
3349 if (!mailBody.is_open())
3350 {
339adafa 3351 Log("SHUTTLE", Form("Could not open mail body file %s", bodyFileName.Data()));
675f64cd 3352 return kFALSE;
3353 }
3354
675f64cd 3355
339adafa 3356 TString subject;
3357 TString body;
675f64cd 3358
339adafa 3359 if (target == kDCSEMail){
af93e5d7 3360 subject = Form("%s CRITICAL Retrieval of data points for %s FAILED in run %d !",
7bdfe1fa 3361 tmpStr.Data(), fCurrentDetector.Data(), GetCurrentRun());
339adafa 3362 AliDebug(2, Form("subject: %s", subject.Data()));
3363
3364 body = Form("Dear DCS experts, \n\n");
3365 body += Form("SHUTTLE couldn\'t retrieve the data points for detector %s "
3366 "in run %d!!\n\n", fCurrentDetector.Data(), GetCurrentRun());
fb40ead4 3367 }
339adafa 3368 else if (target == kFXSEMail){
af93e5d7 3369 subject = Form("%s CRITICAL FXS communication for %s FAILED in run %d !",
7bdfe1fa 3370 tmpStr.Data(), fCurrentDetector.Data(), GetCurrentRun());
339adafa 3371 AliDebug(2, Form("subject: %s", subject.Data()));
3372 TString sys;
3373 if (system == kDAQ) sys="DAQ";
3374 else if (system == kDCS) sys="DCS";
3375 else if (system == kHLT) sys="HLT";
3d2b6cdd 3376 else if (system == kDQM) sys="DQM";
339adafa 3377 else return kFALSE;
9c0e2d64 3378 body = Form("Dear %s FXS experts, \n\n",sys.Data());
339adafa 3379 body += Form("SHUTTLE couldn\'t retrieve data from the FXS for detector %s "
9c0e2d64 3380 "in run %d!!\n\n", fCurrentDetector.Data(), GetCurrentRun());
8662f0a8 3381 body += Form("The contacted server was:\nDB: %s\nFXS:%s\n\n", fConfig->GetFXSdbHost(system), fConfig->GetFXSHost(system));
339adafa 3382 }
3383 else {
7bdfe1fa 3384 subject = Form("%s %s Shuttle preprocessor FAILED in run %d (run type = %s)!",
3385 tmpStr.Data(), fCurrentDetector.Data(), GetCurrentRun(), GetRunType());
339adafa 3386 AliDebug(2, Form("subject: %s", subject.Data()));
3387
3388 body = Form("Dear %s expert(s), \n\n", fCurrentDetector.Data());
3389 body += Form("SHUTTLE just detected that your preprocessor "
3390 "failed processing run %d (run type = %s)!!\n\n",
3391 GetCurrentRun(), GetRunType());
3392 }
675f64cd 3393
7d4cf768 3394 body += Form("Please check %s status on the SHUTTLE monitoring page: \n\n",
3395 fCurrentDetector.Data());
b0e53b15 3396 if (fConfig->GetRunMode() == AliShuttleConfig::kTest)
3397 {
8e828d39 3398 body += Form("\thttp://pcalimonitor.cern.ch/shuttle.jsp?time=24 \n\n");
b0e53b15 3399 } else {
8e828d39 3400 body += Form("\thttp://pcalimonitor.cern.ch/shuttle.jsp?instance=PROD&time=24 \n\n");
b0e53b15 3401 }
339adafa 3402
3403
7d4cf768 3404 TString logFolder = "logs";
3405 if (fConfig->GetRunMode() == AliShuttleConfig::kProd)
3406 logFolder += "_PROD";
3407
3408
675f64cd 3409 body += Form("Find the %s log for the current run on \n\n"
29fb87f4 3410 "\thttp://pcalishuttle02.cern.ch/%s/%d/%d/%s.log \n\n",
3ebbf3c4 3411 fCurrentDetector.Data(), logFolder.Data(), GetCurrentRun()/10000,
3412 GetCurrentRun(), fCurrentDetector.Data());
fe6a3257 3413 body += Form("The last 15 lines of %s log file are following:\n\n", fCurrentDetector.Data());
675f64cd 3414
3415 AliDebug(2, Form("Body begin: %s", body.Data()));
3416
3417 mailBody << body.Data();
3418 mailBody.close();
3419 mailBody.open(bodyFileName, ofstream::out | ofstream::app);
3420
3ebbf3c4 3421 TString logFileName = Form("%s/%d/%d/%s.log", GetShuttleLogDir(),
3422 GetCurrentRun()/10000, GetCurrentRun(), fCurrentDetector.Data());
fe6a3257 3423 TString tailCommand = Form("tail -n 15 %s >> %s", logFileName.Data(), bodyFileName.Data());
675f64cd 3424 if (gSystem->Exec(tailCommand.Data()))
3425 {
3426 mailBody << Form("%s log file not found ...\n\n", fCurrentDetector.Data());
3427 }
3428
3429 TString endBody = Form("------------------------------------------------------\n\n");
36c99a6a 3430 endBody += Form("In case of problems please contact the SHUTTLE core team.\n\n");
3431 endBody += "Please do not answer this message directly, it is automatically generated.\n\n";
546242fb 3432 endBody += "Greetings,\n\n \t\t\tthe SHUTTLE\n";
57c1a579 3433
909732f7 3434 AliDebug(2, Form("Body end: %s", endBody.Data()));
57c1a579 3435
3436 mailBody << endBody.Data();
3437
3438 mailBody.close();
3439
3440 // send mail!
3441 TString mailCommand = Form("mail -s \"%s\" -c %s %s < %s",
3442 subject.Data(),
3443 cc.Data(),
3444 to.Data(),
3445 bodyFileName.Data());
909732f7 3446 AliDebug(2, Form("mail command: %s", mailCommand.Data()));
57c1a579 3447
3448 Bool_t result = gSystem->Exec(mailCommand.Data());
3449
3450 return result == 0;
3451}
441b0e9c 3452//______________________________________________________________________________________________
9827400b 3453const char* AliShuttle::GetRunType()
441b0e9c 3454{
9827400b 3455 //
3456 // returns run type read from "run type" logbook
3457 //
441b0e9c 3458
3459 if(!fLogbookEntry) {
3460 AliError("No logbook entry!");
3461 return 0;
3462 }
3463
9827400b 3464 return fLogbookEntry->GetRunType();
441b0e9c 3465}
3466
4859271b 3467//______________________________________________________________________________________________
3468Bool_t AliShuttle::GetHLTStatus()
3469{
3470 // Return HLT status (ON=1 OFF=0)
af76b20d 3471 // Converts the HLT status from the mode string read in the run logbook (not just a bool)
4859271b 3472
3473 if(!fLogbookEntry) {
3474 AliError("No logbook entry!");
3475 return 0;
3476 }
3477
af76b20d 3478 // TODO implement when HLTMode is inserted in run logbook
b4ebc2d5 3479 TString hltMode = fLogbookEntry->GetRunParameter("HLTmode");
af76b20d 3480 TSubString firstChar = hltMode(0,1);
3481 AliDebug(2,Form("First char = %s ",firstChar.Data()));
3482 if (firstChar == "A") {
3483 return kFALSE;
3484 }
3485 else if ((firstChar == "B") || (firstChar == "C") || (firstChar == "D") || (firstChar == "E")) {
3486 return kTRUE;
3487 }
3488 else {
3489 Log("SHUTTLE","Unexpected HLT mode! Returning 0....");
3490 return kFALSE;
3491 }
4859271b 3492}
ca8ea066 3493
3494//______________________________________________________________________________________________
3495const char* AliShuttle::GetTriggerConfiguration()
3496{
3497 // Receives the trigger configuration from the DAQ logbook for the current run
3498
3499 // check connection, if needed reconnect
efec19bd 3500 if (!Connect(4))
ca8ea066 3501 return 0;
3502
3503 TString sqlQuery;
3504 sqlQuery.Form("SELECT configFile FROM logbook_trigger_config WHERE run = %d", GetCurrentRun());
efec19bd 3505 TSQLResult* result = fServer[4]->Query(sqlQuery);
ca8ea066 3506 if (!result)
3507 {
3508 Log("SHUTTLE", Form("ERROR: Can't execute query <%s>!", sqlQuery.Data()));
3509 return 0;
3510 }
3511
3512 if (result->GetRowCount() == 0)
3513 {
ff0191c7 3514 Log("SHUTTLE", "WARNING: Trigger configuration not found in logbook_trigger_config");
ca8ea066 3515 delete result;
3516 return 0;
3517 }
3518
3519 TSQLRow* row = result->Next();
3520 if (!row)
3521 {
3522 Log("SHUTTLE", "ERROR: Could not receive logbook_trigger_config data");
3523 delete result;
3524 return 0;
3525 }
3526
3527 // static, so that pointer remains valid when it is returned to the calling class
3528 static TString triggerConfig(row->GetField(0));
3529
3530 delete row;
3531 row = 0;
3532
3533 delete result;
3534 result = 0;
3535
3536 Log("SHUTTLE", Form("Found trigger configuration: %s", triggerConfig.Data()));
3537
3538 return triggerConfig;
3539}
3540
2c5f9d06 3541//______________________________________________________________________________________________
3542const char* AliShuttle::GetCTPTimeParams()
3543{
3544 // Receives the CTP time parameters from the DAQ logbook for the current run
3545
3546 // check connection, if needed reconnect
efec19bd 3547 if (!Connect(4))
2c5f9d06 3548 return 0;
3549
3550 TString sqlQuery;
3551 sqlQuery.Form("SELECT alignmentFile FROM logbook_trigger_config WHERE run = %d", GetCurrentRun());
efec19bd 3552 TSQLResult* result = fServer[4]->Query(sqlQuery);
2c5f9d06 3553 if (!result)
3554 {
3555 Log("SHUTTLE", Form("ERROR: Can't execute query <%s>!", sqlQuery.Data()));
3556 return 0;
3557 }
3558
3559 if (result->GetRowCount() == 0)
3560 {
ff0191c7 3561 Log("SHUTTLE", "WARNING: CTP time params not found in logbook_trigger_config");
2c5f9d06 3562 delete result;
3563 return 0;
3564 }
3565
3566 TSQLRow* row = result->Next();
3567 if (!row)
3568 {
3569 Log("SHUTTLE", "ERROR: Could not receive logbook_trigger_config data");
3570 delete result;
3571 return 0;
3572 }
3573
3574 // static, so that pointer remains valid when it is returned to the calling class
3575 static TString triggerTimeParams(row->GetField(0));
3576
3577 delete row;
3578 row = 0;
3579
3580 delete result;
3581 result = 0;
3582
a191c286 3583 Log("SHUTTLE", Form("Found trigger time parameters: %s", triggerTimeParams.Data()));
2c5f9d06 3584
3585 return triggerTimeParams;
3586}
3587
d386d623 3588//______________________________________________________________________________________________
b48d2542 3589const char* AliShuttle::GetTriggerDetectorMask()
3590{
3591 // Receives the trigger detector mask from DAQ logbook
3592
3593 // check connection, if needed reconnect
efec19bd 3594 if (!Connect(4))
b48d2542 3595 return 0;
3596
3597 TString sqlQuery;
3598 sqlQuery.Form("SELECT BIN(BIT_OR(inputDetectorMask)) from logbook_trigger_clusters WHERE run = %d;", GetCurrentRun());
efec19bd 3599 TSQLResult* result = fServer[4]->Query(sqlQuery);
b48d2542 3600 if (!result)
3601 {
3602 Log("SHUTTLE", Form("ERROR: Can't execute query <%s>!", sqlQuery.Data()));
3603 return 0;
3604 }
3605
3606 if (result->GetRowCount() == 0)
3607 {
3608 Log("SHUTTLE", "ERROR: Trigger Detector Mask not found in logbook_trigger_clusters");
3609 delete result;
3610 return 0;
3611 }
3612
3613 TSQLRow* row = result->Next();
3614 if (!row)
3615 {
3616 Log("SHUTTLE", "ERROR: Could not receive logbook_trigger_clusters data");
3617 delete result;
3618 return 0;
3619 }
3620
3621 // static, so that pointer remains valid when it is returned to the calling class
3622 static TString triggerDetectorMask(row->GetField(0));
3623
3624 delete row;
3625 row = 0;
3626
3627 delete result;
3628 result = 0;
3629
2e7d9963 3630 Log("SHUTTLE", Form("Found Trigger Detector Mask: %s", triggerDetectorMask.Data()));
b48d2542 3631
3632 return triggerDetectorMask;
3633}
3634
3635//______________________________________________________________________________________________
d386d623 3636void AliShuttle::SetShuttleTempDir(const char* tmpDir)
3637{
9827400b 3638 //
3639 // sets Shuttle temp directory
3640 //
d386d623 3641
3642 fgkShuttleTempDir = gSystem->ExpandPathName(tmpDir);
3643}
3644
3645//______________________________________________________________________________________________
3646void AliShuttle::SetShuttleLogDir(const char* logDir)
3647{
9827400b 3648 //
3649 // sets Shuttle log directory
3650 //
d386d623 3651
3652 fgkShuttleLogDir = gSystem->ExpandPathName(logDir);
3653}
503ff24f 3654//______________________________________________________________________________________________
3655Bool_t AliShuttle::TouchFile()
3656{
3657 //
3658 // touching a file on the grid if run has been DONE
3659 //
3660
3661 if (!gGrid)
3662 {
3663 Log("SHUTTLE",Form("No TGrid connection estabilished!"));
3664 Log("SHUTTLE",Form("Could not touch file for run %i",GetCurrentRun()));
3665 return kFALSE;
3666 }
3667
b8cf5e0f 3668 TString dir;
728290a4 3669 dir.Form("%s%d/SHUTTLE_DONE", fConfig->GetAlienPath(), GetCurrentYear());
b8cf5e0f 3670 // checking whether directory for touch command exists
aca77fbf 3671 TGridResult* resultLs = gGrid->Ls(dir.Data());
3672 if (!resultLs){ // unfortunately we don't get this for ls of a non existing dir
29f58d16 3673 Log("SHUTTLE",Form("No result for \"Ls(\"%s\")\", returning without touching", dir.Data()));
b8cf5e0f 3674 return kFALSE;
3675 }
aca77fbf 3676
3677 if ( resultLs->GetEntries() == 1 && !resultLs->GetFileName(0) ) {
3678 // this is what we currently get for ls of a non existing dir
b8cf5e0f 3679 Log("SHUTTLE",Form("No directory %s found, creating it",dir.Data()));
3680
3681 // creating the directory
3682
3683 Bool_t boolMkdir = gGrid->Mkdir(dir.Data());
3684 if (!boolMkdir) {
3685 Log("SHUTTLE",Form("Impossible to create dir %s in alien catalogue for run %i!",dir.Data(),GetCurrentRun()));
97d38a5a 3686 delete resultLs;
3687 resultLs = 0x0;
b8cf5e0f 3688 return kFALSE;
3689 }
3690 Log("SHUTTLE",Form("Directory %s successfully created in alien catalogue for run %i",dir.Data(),GetCurrentRun()));
3691 }
3692 else {
3693 Log("SHUTTLE",Form("Directory %s correctly found for run %i",dir.Data(),GetCurrentRun()));
3694 }
3695
97d38a5a 3696 delete resultLs;
3697 resultLs = 0x0;
3698
aaf11974 3699 // Before trying to touch, check that the file is not already there (the touch would fail forever, leaving the run in pending)
3700 TString lsFileCommand;
3701 lsFileCommand.Form("ls %s/%i", dir.Data(), GetCurrentRun());
3702 TGridResult *resultLsFile = dynamic_cast<TGridResult*>(gGrid->Command(lsFileCommand));
3703 if (!resultLsFile){
3704 Log("SHUTTLE",Form("No result for file ls command, returning without touching for run %i",GetCurrentRun()));
b8cf5e0f 3705 return kFALSE;
503ff24f 3706 }
aaf11974 3707 TMap *mapLsFile = dynamic_cast<TMap*>(resultLsFile->At(0));
3708 if (!mapLsFile){
3709 Log("SHUTTLE",Form("No map for file ls command, returning without touching for run %i",GetCurrentRun()));
3710 delete resultLsFile;
3711 resultLsFile = 0x0;
b8cf5e0f 3712 return kFALSE;
3713 }
aaf11974 3714 TObjString *valueLsFile = dynamic_cast<TObjString*>(mapLsFile->GetValue("name"));
3715 if (valueLsFile){
3716 Log("SHUTTLE",Form("\"name\" key set in the map for file ls command. Touchfile for run %i already there.",GetCurrentRun()));
3717 Log("SHUTTLE", "The file was already there, did not touch it.");
3718 }else{
3719 TString command;
3720 command.Form("touch %s/%i", dir.Data(), GetCurrentRun());
3721 Log("SHUTTLE", Form("Creating entry in file catalog: %s", command.Data()));
3722 TGridResult *resultTouch = dynamic_cast<TGridResult*>(gGrid->Command(command));
3723 if (!resultTouch){
3724 Log("SHUTTLE",Form("No result for touching command, returning without touching for run %i",GetCurrentRun()));
3725 return kFALSE;
3726 }
3727 TMap *mapTouch = dynamic_cast<TMap*>(resultTouch->At(0));
3728 if (!mapTouch){
3729 Log("SHUTTLE",Form("No map for touching command, returning without touching for run %i",GetCurrentRun()));
3730 delete resultTouch;
3731 resultTouch = 0x0;
3732 return kFALSE;
3733 }
3734 TObjString *valueTouch = dynamic_cast<TObjString*>(mapTouch->GetValue("__result__"));
3735 if (!valueTouch){
3736 Log("SHUTTLE",Form("No value for \"__result__\" key set in the map for touching command, returning without touching for run %i",GetCurrentRun()));
3737 delete resultTouch;
3738 resultTouch = 0x0;
3739 return kFALSE;
3740 }
3741 if (valueTouch->GetString()!="1"){
3742 Log("SHUTTLE",Form("Failing the touching command, returning without touching for run %i",GetCurrentRun()));
3743 delete resultTouch;
3744 resultTouch = 0x0;
3745 return kFALSE;
3746 }
3747 delete resultLsFile;
3748 resultLsFile = 0x0;
3749 delete resultTouch;
3750 resultTouch = 0x0;
3751 Log("SHUTTLE", "Sucessfully touched the file");
3752 }
3753 return kTRUE;
503ff24f 3754}
aaf11974 3755
2eb98462 3756//______________________________________________________________________________________________
f100adeb 3757UInt_t AliShuttle::GetStartTimeDCSQuery()
2eb98462 3758{
3759 // Return Start Time for the DCS query
3760 //
3761 // The call is delegated to AliShuttleInterface
3762
3d24ad5a 3763 return GetCurrentStartTime()-fConfig->GetDCSQueryOffset();
2eb98462 3764}
3765//______________________________________________________________________________________________
f100adeb 3766UInt_t AliShuttle::GetEndTimeDCSQuery()
2eb98462 3767{
3768 // Return End Time for the DCS query
3769 //
3770 // The call is delegated to AliShuttleInterface
3771
3772 return GetCurrentEndTime()+fConfig->GetDCSQueryOffset();
3773}
1f0f7c2e 3774//______________________________________________________________________________________________
3775void AliShuttle::SendMLFromDet(const char* value)
3776{
3777 //
3778 // Sending an information coming from the current detector to ML
3779 //
3780
3781 TMonaLisaText mlText(Form("%s_RunCondition", fCurrentDetector.Data()), value);
3782
3783 TList mlList;
3784 mlList.Add(&mlText);
3785
3786 TString mlID;
3787 mlID.Form("%d", GetCurrentRun());
3788 fMonaLisa->SendParameters(&mlList, mlID);
3789
3790 return;
3791}
6ecb073a 3792//______________________________________________________________________________________________
3793TString* AliShuttle::GetLTUConfig(const char* det)
3794{
3795 //
3796 // Getting ltuFineDelay1, ltuFineDelay2, ltuBCDelay for detector det from logbook_detectors table in logbook
3797 //
3798
efec19bd 3799 if (!Connect(4))
6ecb073a 3800 return 0;
3801
3802 TString sqlQuery;
3803 sqlQuery.Form("select LTUFineDelay1, LTUFineDelay2, LTUBCDelayAdd from logbook_detectors WHERE run_number = %d and detector = \"%s\";", GetCurrentRun(),det);
3804
efec19bd 3805 TSQLResult* result = fServer[4]->Query(sqlQuery);
6ecb073a 3806 if (!result){
3807 Log("SHUTTLE","ERROR: No result found for the LTU configuration query");
3808 return 0x0;
3809 }
3810 if (result->GetRowCount() == 0){
3811 Log("SHUTTLE",Form("ERROR: LTU configuration not found in logbook_detectors for detector %s, returning null pointer",det));
3812 delete result;
3813 return 0x0;
3814 }
3815 if (result->GetFieldCount() != 3){
3816 Log("SHUTTLE",Form("ERROR: not all the required fields are there for the LTU configuration for detector %s (only %d found), returning a null pointer",det, result->GetFieldCount()));
3817 delete result;
3818 return 0x0;
3819 }
3820 TSQLRow* row = result->Next();
3821 if (!row){
3822 Printf("ERROR: Could not receive logbook_detectors data, returning null pointer");
3823 delete result;
3824 return 0x0;
3825 }
3826 TString* ltuConfigString = new TString[3];
3827
3828 ltuConfigString[0] = row->GetField(0);
3829 ltuConfigString[1] = row->GetField(1);
3830 ltuConfigString[2] = row->GetField(2);
503ff24f 3831
6ecb073a 3832 return ltuConfigString;
3833
3834}