SHUTTLE/AliShuttle.cxx

   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
  16 /* $Id$ */
  17
  18 //
  19 // This class is the main manager for AliShuttle.
  20 // It organizes the data retrieval from DCS and call the
  21 // interface methods of AliPreprocessor.
  22 // For every detector in AliShuttleConfgi (see AliShuttleConfig),
  23 // data for its set of aliases is retrieved. If there is registered
  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
  27 // data is stored automatically to the undelying AliCDBStorage.
  28 // For detSpec is used the alias name.
  29 //
  30
  31 #include "AliShuttle.h"
  32
  33 #include "AliCDBManager.h"
  34 #include "AliCDBStorage.h"
  35 #include "AliCDBId.h"
  36 #include "AliCDBRunRange.h"
  37 #include "AliCDBPath.h"
  38 #include "AliCDBEntry.h"
  39 #include "AliShuttleConfig.h"
  40 #include "DCSClient/AliDCSClient.h"
  41 #include "AliLog.h"
  42 #include "AliPreprocessor.h"
  43 #include "AliShuttleStatus.h"
  44 #include "AliShuttleLogbookEntry.h"
  45
  46 #include <TSystem.h>
  47 #include <TObject.h>
  48 #include <TString.h>
  49 #include <TTimeStamp.h>
  50 #include <TObjString.h>
  51 #include <TSQLServer.h>
  52 #include <TSQLResult.h>
  53 #include <TSQLRow.h>
  54 #include <TMutex.h>
  55 #include <TSystemDirectory.h>
  56 #include <TSystemFile.h>
  57 #include <TFile.h>
  58 #include <TGrid.h>
  59 #include <TGridResult.h>
  60 #include <TMap.h>
  61
  62 #include <TMonaLisaWriter.h>
  63
  64 #include <fstream>
  65
  66 #include <sys/types.h>
  67 #include <sys/wait.h>
  68
  69 #include <signal.h>
  70
  71 ClassImp(AliShuttle)
  72
  73 //______________________________________________________________________________________________
  74 AliShuttle::AliShuttle(const AliShuttleConfig* config,
  75                 UInt_t timeout, Int_t retries):
  76 fConfig(config),
  77 fTimeout(timeout), fRetries(retries),
  78 fPreprocessorMap(),
  79 fLogbookEntry(0),
  80 fCurrentDetector(),
  81 fFirstProcessing(0),
  82 fFXSError(-1),
  83 fStatusEntry(0),
  84 fMonitoringMutex(0),
  85 fLastActionTime(0),
  86 fLastAction(),
  87 fMonaLisa(0),
  88 fTestMode(kNone),
  89 fReadTestMode(kFALSE),
  90 fOutputRedirected(kFALSE)
  91 {
  92         //
  93         // config: AliShuttleConfig used
  94         // timeout: timeout used for AliDCSClient connection
  95         // retries: the number of retries in case of connection error.
  96         //
  97
  98         if (!fConfig->IsValid()) AliFatal("********** !!!!! Invalid configuration !!!!! **********");
  99         for(int iSys=0;iSys<4;iSys++) {
 100                 fServer[iSys]=0;
 101                 if (iSys < 3)
 102                         fFXSlist[iSys].SetOwner(kTRUE);
 103         }
 104         fPreprocessorMap.SetOwner(kTRUE);
 105
 106         for (UInt_t iDet=0; iDet<NDetectors(); iDet++)
 107                 fFirstUnprocessed[iDet] = kFALSE;
 108
 109         fMonitoringMutex = new TMutex();
 110 }
 111
 112 //______________________________________________________________________________________________
 113 AliShuttle::~AliShuttle()
 114 {
 115         //
 116         // destructor
 117         //
 118
 119         fPreprocessorMap.DeleteAll();
 120         for(int iSys=0;iSys<4;iSys++)
 121                 if(fServer[iSys]) {
 122                         fServer[iSys]->Close();
 123                         delete fServer[iSys];
 124                         fServer[iSys] = 0;
 125                 }
 126
 127         if (fStatusEntry){
 128                 delete fStatusEntry;
 129                 fStatusEntry = 0;
 130         }
 131
 132         if (fMonitoringMutex)
 133         {
 134                 delete fMonitoringMutex;
 135                 fMonitoringMutex = 0;
 136         }
 137 }
 138
 139 //______________________________________________________________________________________________
 140 void AliShuttle::RegisterPreprocessor(AliPreprocessor* preprocessor)
 141 {
 142         //
 143         // Registers new AliPreprocessor.
 144         // It uses GetName() for indentificator of the pre processor.
 145         // The pre processor is registered it there isn't any other
 146         // with the same identificator (GetName()).
 147         //
 148
 149         const char* detName = preprocessor->GetName();
 150         if(GetDetPos(detName) < 0)
 151                 AliFatal(Form("********** !!!!! Invalid detector name: %s !!!!! **********", detName));
 152
 153         if (fPreprocessorMap.GetValue(detName)) {
 154                 AliWarning(Form("AliPreprocessor %s is already registered!", detName));
 155                 return;
 156         }
 157
 158         fPreprocessorMap.Add(new TObjString(detName), preprocessor);
 159 }
 160 //______________________________________________________________________________________________
 161 Bool_t AliShuttle::Store(const AliCDBPath& path, TObject* object,
 162                 AliCDBMetaData* metaData, Int_t validityStart, Bool_t validityInfinite)
 163 {
 164         // Stores a CDB object in the storage for offline reconstruction. Objects that are not needed for
 165         // offline reconstruction, but should be stored anyway (e.g. for debugging) should NOT be stored
 166         // using this function. Use StoreReferenceData instead!
 167         // It calls StoreLocally function which temporarily stores the data locally; when the preprocessor
 168         // finishes the data are transferred to the main storage (Grid).
 169
 170         return StoreLocally(fgkLocalCDB, path, object, metaData, validityStart, validityInfinite);
 171 }
 172
 173 //______________________________________________________________________________________________
 174 Bool_t AliShuttle::StoreReferenceData(const AliCDBPath& path, TObject* object, AliCDBMetaData* metaData)
 175 {
 176         // Stores a CDB object in the storage for reference data. This objects will not be available during
 177         // offline reconstrunction. Use this function for reference data only!
 178         // It calls StoreLocally function which temporarily stores the data locally; when the preprocessor
 179         // finishes the data are transferred to the main storage (Grid).
 180
 181         return StoreLocally(fgkLocalRefStorage, path, object, metaData);
 182 }
 183
 184 //______________________________________________________________________________________________
 185 Bool_t AliShuttle::StoreLocally(const TString& localUri,
 186                         const AliCDBPath& path, TObject* object, AliCDBMetaData* metaData,
 187                         Int_t validityStart, Bool_t validityInfinite)
 188 {
 189         // Store object temporarily in local storage. Parameters are passed by Store and StoreReferenceData functions.
 190         // when the preprocessor finishes the data are transferred to the main storage (Grid).
 191         // The parameters are:
 192         //   1) Uri of the backup storage (Local)
 193         //   2) the object's path.
 194         //   3) the object to be stored
 195         //   4) the metaData to be associated with the object
 196         //   5) the validity start run number w.r.t. the current run,
 197         //      if the data is valid only for this run leave the default 0
 198         //   6) specifies if the calibration data is valid for infinity (this means until updated),
 199         //      typical for calibration runs, the default is kFALSE
 200         //
 201         // returns 0 if fail, 1 otherwise
 202
 203
 204         if (fTestMode & kErrorStorage)
 205         {
 206                 Log(fCurrentDetector, "StoreLocally - In TESTMODE - Simulating error while storing locally");
 207                 return kFALSE;
 208         }
 209
 210         const char* cdbType = (localUri == fgkLocalCDB) ? "CDB" : "Reference";
 211
 212         Int_t firstRun = GetCurrentRun() - validityStart;
 213         if(firstRun < 0) {
 214                 AliWarning("First valid run happens to be less than 0! Setting it to 0.");
 215                 firstRun=0;
 216         }
 217
 218         Int_t lastRun = -1;
 219         if(validityInfinite) {
 220                 lastRun = AliCDBRunRange::Infinity();
 221         } else {
 222                 lastRun = GetCurrentRun();
 223         }
 224
 225         // Version is set to current run, it will be used later to transfer data to Grid
 226         AliCDBId id(path, firstRun, lastRun, GetCurrentRun(), -1);
 227
 228         if(! dynamic_cast<TObjString*> (metaData->GetProperty("RunUsed(TObjString)"))){
 229                 TObjString runUsed = Form("%d", GetCurrentRun());
 230                 metaData->SetProperty("RunUsed(TObjString)", runUsed.Clone());
 231         }
 232
 233         Bool_t result = kFALSE;
 234
 235         if (!(AliCDBManager::Instance()->GetStorage(localUri))) {
 236                 Log("SHUTTLE", Form("StoreLocally - Cannot activate local %s storage", cdbType));
 237         } else {
 238                 Int_t logLevel = AliLog::GetGlobalLogLevel();
 239                 AliLog::SetGlobalLogLevel(AliLog::kError);
 240                 result = AliCDBManager::Instance()->GetStorage(localUri)
 241                                         ->Put(object, id, metaData);
 242                 AliLog::SetGlobalLogLevel((AliLog::EType_t)logLevel);
 243         }
 244
 245         if(!result) {
 246
 247                 Log(fCurrentDetector, Form("StoreLocally - Can't store object <%s>!", id.ToString().Data()));
 248         }
 249
 250
 251         return result;
 252 }
 253
 254 //______________________________________________________________________________________________
 255 Bool_t AliShuttle::StoreOCDB()
 256 {
 257         //
 258         // Called when preprocessor ends successfully or when previous storage attempt failed (kStoreError status)
 259         // Calls underlying StoreOCDB(const char*) function twice, for OCDB and Reference storage.
 260         // Then calls StoreRefFilesToGrid to store reference files.
 261         //
 262
 263         UpdateShuttleStatus(AliShuttleStatus::kStoreStarted);
 264
 265         if (fTestMode & kErrorGrid)
 266         {
 267                 Log("SHUTTLE", "StoreOCDB - In TESTMODE - Simulating error while storing in the Grid");
 268                 Log(fCurrentDetector, "StoreOCDB - In TESTMODE - Simulating error while storing in the Grid");
 269                 return kFALSE;
 270         }
 271
 272         Log("SHUTTLE","StoreOCDB - Storing OCDB data ...");
 273         Int_t resultCDB = StoreOCDB(fgkMainCDB);
 274
 275         Log("SHUTTLE","StoreOCDB - Storing reference data ...");
 276         Int_t resultRef = StoreOCDB(fgkMainRefStorage);
 277
 278         Log("SHUTTLE","StoreOCDB - Storing reference files ...");
 279         Bool_t resultRefFiles = CopyFilesToGrid("reference");
 280
 281         Bool_t resultMetadata = kTRUE;
 282         if(fCurrentDetector == "GRP")
 283         {
 284                 Log("SHUTTLE","StoreOCDB - Storing Run Metadata file ...");
 285                 resultMetadata = CopyFilesToGrid("metadata");
 286         }
 287
 288         Int_t storeResult = 0;
 289
 290         if (resultCDB < 0 || resultRef < 0 || resultRefFiles == kFALSE || resultMetadata == kFALSE)
 291                 storeResult = -1;
 292         else if (resultCDB > 0 || resultRef > 0)
 293                 storeResult = 1;
 294
 295         if (storeResult < 0)
 296         {
 297                 Log("SHUTTLE",
 298                         Form("\t\t\t****** run %d - %s: STORAGE ERROR ******",
 299                                 GetCurrentRun(), fCurrentDetector.Data()));
 300                 UpdateShuttleStatus(AliShuttleStatus::kStoreError);
 301         }
 302         else if (storeResult > 0)
 303         {
 304                 Log("SHUTTLE",
 305                         Form("\t\t\t****** run %d - %s: STORAGE DELAYED ******",
 306                                 GetCurrentRun(), fCurrentDetector.Data()));
 307                 UpdateShuttleStatus(AliShuttleStatus::kStoreDelayed);
 308         }
 309         else if (storeResult == 0)
 310         {
 311                 Log("SHUTTLE",
 312                         Form("\t\t\t****** run %d - %s: DONE ******",
 313                                 GetCurrentRun(), fCurrentDetector.Data()));
 314                 UpdateShuttleStatus(AliShuttleStatus::kDone);
 315                 UpdateShuttleLogbook(fCurrentDetector, "DONE");
 316         }
 317
 318         return (storeResult == 0);
 319 }
 320
 321 //______________________________________________________________________________________________
 322 Int_t AliShuttle::StoreOCDB(const TString& gridURI)
 323 {
 324         //
 325         // Called by StoreOCDB(), performs actual storage to the main OCDB and reference storages (Grid)
 326         //
 327         // Return code:
 328         //   -2 initialization error
 329         //   -1 storage error
 330         //   0  success
 331         //   1  storage delayed (e.g. previous unprocessed runs)
 332         //
 333
 334         TObjArray* gridIds=0;
 335
 336         Bool_t result = kTRUE;
 337         Bool_t delayed = kFALSE;
 338
 339         const char* type = 0;
 340         TString localURI;
 341         if(gridURI == fgkMainCDB) {
 342                 type = "OCDB";
 343                 localURI = fgkLocalCDB;
 344         } else if(gridURI == fgkMainRefStorage) {
 345                 type = "reference";
 346                 localURI = fgkLocalRefStorage;
 347         } else {
 348                 AliError(Form("Invalid storage URI: %s", gridURI.Data()));
 349                 return -2;
 350         }
 351
 352         AliCDBManager* man = AliCDBManager::Instance();
 353
 354         AliCDBStorage *gridSto = man->GetStorage(gridURI);
 355         if(!gridSto) {
 356                 Log("SHUTTLE",
 357                         Form("StoreOCDB - cannot activate main %s storage", type));
 358                 return -2;
 359         }
 360
 361         gridIds = gridSto->GetQueryCDBList();
 362
 363         // get objects previously stored in local CDB
 364         AliCDBStorage *localSto = man->GetStorage(localURI);
 365         if(!localSto) {
 366                 Log("SHUTTLE",
 367                         Form("StoreOCDB - cannot activate local %s storage", type));
 368                 return -2;
 369         }
 370         AliCDBPath aPath(GetOfflineDetName(fCurrentDetector.Data()),"*","*");
 371         // Local objects were stored with current run as Grid version!
 372         TList* localEntries = localSto->GetAll(aPath.GetPath(), GetCurrentRun(), GetCurrentRun());
 373         localEntries->SetOwner(1);
 374
 375         // loop on local stored objects
 376         TIter localIter(localEntries);
 377         AliCDBEntry *aLocEntry = 0;
 378         while((aLocEntry = dynamic_cast<AliCDBEntry*> (localIter.Next()))){
 379                 aLocEntry->SetOwner(1);
 380                 AliCDBId aLocId = aLocEntry->GetId();
 381                 aLocEntry->SetVersion(-1);
 382                 aLocEntry->SetSubVersion(-1);
 383
 384                 Log(fCurrentDetector.Data(), Form("Attempting to store %s", aLocId.ToString().Data()));
 385
 386                 // If local object is valid up to infinity we store it only if it is
 387                 // the first unprocessed run!
 388                 if (aLocId.GetLastRun() == AliCDBRunRange::Infinity() &&
 389                         !fFirstUnprocessed[GetDetPos(fCurrentDetector)])
 390                 {
 391                         Log("SHUTTLE", Form("StoreOCDB - %s: object %s has validity infinite but "
 392                                                 "there are previous unprocessed runs!",
 393                                                 fCurrentDetector.Data(), aLocId.GetPath().Data()));
 394                         Log(fCurrentDetector.Data(), Form("StoreOCDB - %s: object %s has validity infinite but "
 395                                                 "there are previous unprocessed runs!",
 396                                                 fCurrentDetector.Data(), aLocId.GetPath().Data()));
 397                         delayed = kTRUE;
 398                         continue;
 399                 }
 400
 401                 // loop on Grid valid Id's
 402                 Bool_t store = kTRUE;
 403                 TIter gridIter(gridIds);
 404                 AliCDBId* aGridId = 0;
 405                 while ((aGridId = dynamic_cast<AliCDBId*> (gridIter.Next()))) {
 406                         if (aGridId->GetPath() != aLocId.GetPath())
 407                                 continue;
 408                         // skip all objects valid up to infinity
 409                         if (aGridId->GetLastRun() == AliCDBRunRange::Infinity())
 410                                 continue;
 411
 412                         // if we get here, it means there's already some more recent object stored on Grid!
 413                         Log(fCurrentDetector.Data(),
 414                                 Form("StoreOCDB - A more recent object already exists in %s storage: <%s>",
 415                                 type, aGridId->ToString().Data()));
 416
 417                         store = kFALSE;
 418                         break;
 419                 }
 420
 421                 Bool_t storeOk = kFALSE;
 422                 if (store)
 423                 {
 424                         Log(fCurrentDetector.Data(), Form("Prechecks succeeded. Ready to store %s", aLocId.ToString().Data()));
 425                         storeOk = gridSto->Put(aLocEntry);
 426                         if (storeOk) {
 427                                 Log("SHUTTLE",
 428                                 Form("StoreOCDB - Object <%s> successfully put into %s storage",
 429                                         aLocId.ToString().Data(), type));
 430                                 Log(fCurrentDetector.Data(),
 431                                         Form("StoreOCDB - Object <%s> successfully put into %s storage",
 432                                         aLocId.ToString().Data(), type));
 433                         } else  {
 434                                 Log("SHUTTLE",
 435                                         Form("StoreOCDB - Grid %s storage of object <%s> failed",
 436                                         type, aLocId.ToString().Data()));
 437                                 Log(fCurrentDetector.Data(),
 438                                         Form("StoreOCDB - Grid %s storage of object <%s> failed",
 439                                         type, aLocId.ToString().Data()));
 440                                 result = kFALSE;
 441                         }
 442                 }
 443
 444                 if (!store || storeOk) {
 445                         // removing local file...
 446                         TString filename;
 447                         localSto->IdToFilename(aLocId, filename);
 448                         Log("SHUTTLE", Form("StoreOCDB - Removing local file %s", filename.Data()));
 449                         RemoveFile(filename.Data());
 450                 }
 451         }
 452         localEntries->Clear();
 453
 454         Int_t returnCode = 0;
 455
 456         if (result == kFALSE)
 457                 returnCode = -1;
 458         else if (delayed != kFALSE)
 459                 returnCode =  1;
 460
 461         Log("SHUTTLE", Form("StoreOCDB - Returning with %d (result = %d, delayed = %d)", returnCode, result, delayed));
 462         Log(fCurrentDetector.Data(), Form("StoreOCDB - Returning with %d (result = %d, delayed = %d)", returnCode, result, delayed));
 463
 464         return returnCode;
 465 }
 466
 467 //______________________________________________________________________________________________
 468 Bool_t AliShuttle::CleanReferenceStorage(const char* detector)
 469 {
 470         // clears the directory used to store reference files of a given subdetector
 471
 472         AliCDBManager* man = AliCDBManager::Instance();
 473         AliCDBStorage* sto = man->GetStorage(fgkLocalRefStorage);
 474         TString localBaseFolder = sto->GetBaseFolder();
 475
 476         TString targetDir = GetRefFilePrefix(localBaseFolder.Data(), detector);
 477
 478         Log("SHUTTLE", Form("CleanReferenceStorage - Cleaning %s", targetDir.Data()));
 479
 480         TString begin;
 481         begin.Form("%d_", GetCurrentRun());
 482
 483         TSystemDirectory* baseDir = new TSystemDirectory("/", targetDir);
 484         if (!baseDir)
 485                 return kTRUE;
 486
 487         TList* dirList = baseDir->GetListOfFiles();
 488         delete baseDir;
 489
 490         if (!dirList) return kTRUE;
 491
 492         if (dirList->GetEntries() < 3)
 493         {
 494                 delete dirList;
 495                 return kTRUE;
 496         }
 497
 498         Int_t nDirs = 0, nDel = 0;
 499         TIter dirIter(dirList);
 500         TSystemFile* entry = 0;
 501
 502         Bool_t success = kTRUE;
 503
 504         while ((entry = dynamic_cast<TSystemFile*> (dirIter.Next())))
 505         {
 506                 if (entry->IsDirectory())
 507                         continue;
 508
 509                 TString fileName(entry->GetName());
 510                 if (!fileName.BeginsWith(begin))
 511                         continue;
 512
 513                 nDirs++;
 514
 515                 // delete file
 516                 Int_t result = gSystem->Unlink(fileName.Data());
 517
 518                 if (result)
 519                 {
 520                         Log("SHUTTLE", Form("CleanReferenceStorage - Could not delete file %s!", fileName.Data()));
 521                         success = kFALSE;
 522                 } else {
 523                         nDel++;
 524                 }
 525         }
 526
 527         if(nDirs > 0)
 528                 Log("SHUTTLE", Form("CleanReferenceStorage - %d (over %d) reference files in folder %s were deleted.",
 529                         nDel, nDirs, targetDir.Data()));
 530
 531
 532         delete dirList;
 533         return success;
 534
 535
 536
 537
 538
 539
 540   Int_t result = gSystem->GetPathInfo(targetDir, 0, (Long64_t*) 0, 0, 0);
 541   if (result == 0)
 542   {
 543     // delete directory
 544     result = gSystem->Exec(Form("rm -rf %s", targetDir.Data()));
 545     if (result != 0)
 546     {
 547       Log("SHUTTLE", Form("CleanReferenceStorage - Could not clean directory %s", targetDir.Data()));
 548       return kFALSE;
 549     }
 550   }
 551
 552   result = gSystem->mkdir(targetDir, kTRUE);
 553   if (result != 0)
 554   {
 555     Log("SHUTTLE", Form("CleanReferenceStorage - Error creating base directory %s", targetDir.Data()));
 556     return kFALSE;
 557   }
 558
 559   return kTRUE;
 560 }
 561
 562 //______________________________________________________________________________________________
 563 Bool_t AliShuttle::StoreReferenceFile(const char* detector, const char* localFile, const char* gridFileName)
 564 {
 565         //
 566         // Stores reference file directly (without opening it). This function stores the file locally.
 567         //
 568         // The file is stored under the following location:
 569         // <base folder of local reference storage>/<DET>/<RUN#>_<gridFileName>
 570         // where <gridFileName> is the second parameter given to the function
 571         //
 572
 573         if (fTestMode & kErrorStorage)
 574         {
 575                 Log(fCurrentDetector, "StoreReferenceFile - In TESTMODE - Simulating error while storing locally");
 576                 return kFALSE;
 577         }
 578
 579         AliCDBManager* man = AliCDBManager::Instance();
 580         AliCDBStorage* sto = man->GetStorage(fgkLocalRefStorage);
 581
 582         TString localBaseFolder = sto->GetBaseFolder();
 583
 584         TString target = GetRefFilePrefix(localBaseFolder.Data(), detector);
 585         target.Append(Form("/%d_%s", GetCurrentRun(), gridFileName));
 586
 587         return CopyFileLocally(localFile, target);
 588 }
 589
 590 //______________________________________________________________________________________________
 591 Bool_t AliShuttle::StoreRunMetadataFile(const char* localFile, const char* gridFileName)
 592 {
 593         //
 594         // Stores Run metadata file to the Grid, in the run folder
 595         //
 596         // Only GRP can call this function.
 597
 598         if (fTestMode & kErrorStorage)
 599         {
 600                 Log(fCurrentDetector, "StoreRunMetaDataFile - In TESTMODE - Simulating error while storing locally");
 601                 return kFALSE;
 602         }
 603
 604         AliCDBManager* man = AliCDBManager::Instance();
 605         AliCDBStorage* sto = man->GetStorage(fgkLocalRefStorage);
 606
 607         TString localBaseFolder = sto->GetBaseFolder();
 608
 609         // Build Run level folder
 610         // folder = /alice/data/year/lhcPeriod/runNb/raw
 611
 612
 613         TString lhcPeriod = GetLHCPeriod();
 614         if (lhcPeriod.Length() == 0)
 615         {
 616                 Log("SHUTTLE","StoreRunMetaDataFile - LHCPeriod not found in logbook!");
 617                 return 0;
 618         }
 619
 620         // TODO partitions with one detector only write data into LHCperiod_DET
 621         TString partition = GetRunParameter("detector");
 622
 623         if (partition.Length() > 0 && partition != "ALICE")
 624         {
 625                 lhcPeriod.Append(Form("_%s", partition.Data()));
 626                 Log(fCurrentDetector, Form("Run data tags merged file will be written in %s",
 627                                 lhcPeriod.Data()));
 628         }
 629
 630         TString target = Form("%s/GRP/RunMetadata%s%d/%s/%09d/raw/%s",
 631                                 localBaseFolder.Data(), fConfig->GetAlienPath(), GetCurrentYear(),
 632                                 lhcPeriod.Data(), GetCurrentRun(), gridFileName);
 633
 634         return CopyFileLocally(localFile, target);
 635 }
 636
 637 //______________________________________________________________________________________________
 638 Bool_t AliShuttle::CopyFileLocally(const char* localFile, const TString& target)
 639 {
 640         //
 641         // Stores file locally. Called by StoreReferenceFile and StoreRunMetadataFile
 642         // Files are temporarily stored in the local reference storage. When the preprocessor
 643         // finishes, the Shuttle calls CopyFilesToGrid to transfer the files to AliEn
 644         // (in reference or run level folders)
 645         //
 646
 647         TString targetDir(target(0, target.Last('/')));
 648
 649         //try to open base dir folder, if it does not exist
 650         void* dir = gSystem->OpenDirectory(targetDir.Data());
 651         if (dir == NULL) {
 652                 if (gSystem->mkdir(targetDir.Data(), kTRUE)) {
 653                         Log("SHUTTLE", Form("CopyFileLocally - Can't open directory <%s>", targetDir.Data()));
 654                         return kFALSE;
 655                 }
 656
 657         } else {
 658                 gSystem->FreeDirectory(dir);
 659         }
 660
 661         Int_t result = 0;
 662
 663         result = gSystem->GetPathInfo(localFile, 0, (Long64_t*) 0, 0, 0);
 664         if (result)
 665         {
 666                 Log("SHUTTLE", Form("CopyFileLocally - %s does not exist", localFile));
 667                 return kFALSE;
 668         }
 669
 670         result = gSystem->GetPathInfo(target, 0, (Long64_t*) 0, 0, 0);
 671         if (!result)
 672         {
 673                 Log("SHUTTLE", Form("CopyFileLocally - target file %s already exist, removing...", target.Data()));
 674                 if (gSystem->Unlink(target.Data()))
 675                 {
 676                         Log("SHUTTLE", Form("CopyFileLocally - Could not remove existing target file %s!", target.Data()));
 677                         return kFALSE;
 678                 }
 679         }
 680
 681         result = gSystem->CopyFile(localFile, target);
 682
 683         if (result == 0)
 684         {
 685                 Log("SHUTTLE", Form("CopyFileLocally - File %s stored locally to %s", localFile, target.Data()));
 686                 return kTRUE;
 687         }
 688         else
 689         {
 690                 Log("SHUTTLE", Form("CopyFileLocally - Could not store file %s to %s! Error code = %d",
 691                                 localFile, target.Data(), result));
 692                 return kFALSE;
 693         }
 694
 695
 696
 697 }
 698
 699 //______________________________________________________________________________________________
 700 Bool_t AliShuttle::CopyFilesToGrid(const char* type)
 701 {
 702         //
 703         // Transfers local files to the Grid. Local files can be reference files
 704         // or run metadata file (from GRP only).
 705         //
 706         // According to the type (ref, metadata) the files are stored under the following location:
 707         // ref --> <base folder of reference storage>/<DET>/<RUN#>_<gridFileName>
 708         // metadata --> <run data folder>/<MetadataFileName>
 709         //
 710
 711         AliCDBManager* man = AliCDBManager::Instance();
 712         AliCDBStorage* sto = man->GetStorage(fgkLocalRefStorage);
 713         if (!sto)
 714                 return kFALSE;
 715         TString localBaseFolder = sto->GetBaseFolder();
 716
 717         TString dir;
 718         TString alienDir;
 719         TString begin;
 720
 721         if (strcmp(type, "reference") == 0)
 722         {
 723                 dir = GetRefFilePrefix(localBaseFolder.Data(), fCurrentDetector.Data());
 724                 AliCDBStorage* gridSto = man->GetStorage(fgkMainRefStorage);
 725                 if (!gridSto)
 726                         return kFALSE;
 727                 TString gridBaseFolder = gridSto->GetBaseFolder();
 728                 alienDir = GetRefFilePrefix(gridBaseFolder.Data(), fCurrentDetector.Data());
 729                 begin = Form("%d_", GetCurrentRun());
 730         }
 731         else if (strcmp(type, "metadata") == 0)
 732         {
 733
 734                 TString lhcPeriod = GetLHCPeriod();
 735
 736                 if (lhcPeriod.Length() == 0)
 737                 {
 738                         Log("SHUTTLE","CopyFilesToGrid - LHCPeriod not found in logbook!");
 739                         return 0;
 740                 }
 741
 742                 // TODO partitions with one detector only write data into LHCperiod_DET
 743                 TString partition = GetRunParameter("detector");
 744
 745                 if (partition.Length() > 0 && partition != "ALICE")
 746                 {
 747                         lhcPeriod.Append(Form("_%s", partition.Data()));
 748                 }
 749
 750                 dir = Form("%s/GRP/RunMetadata%s%d/%s/%09d/raw",
 751                                 localBaseFolder.Data(), fConfig->GetAlienPath(), GetCurrentYear(),
 752                                 lhcPeriod.Data(), GetCurrentRun());
 753                 alienDir = dir(dir.Index(fConfig->GetAlienPath()), dir.Length());
 754
 755                 begin = "";
 756         }
 757         else
 758         {
 759                 Log("SHUTTLE", "CopyFilesToGrid - Unexpected: type label must be reference or metadata!");
 760                 return kFALSE;
 761         }
 762
 763         TSystemDirectory* baseDir = new TSystemDirectory("/", dir);
 764         if (!baseDir)
 765                 return kTRUE;
 766
 767         TList* dirList = baseDir->GetListOfFiles();
 768         delete baseDir;
 769
 770         if (!dirList) return kTRUE;
 771
 772         if (dirList->GetEntries() < 3)
 773         {
 774                 delete dirList;
 775                 return kTRUE;
 776         }
 777
 778         if (!gGrid)
 779         {
 780                 Log("SHUTTLE", "CopyFilesToGrid - Connection to Grid failed: Cannot continue!");
 781                 delete dirList;
 782                 return kFALSE;
 783         }
 784
 785         Int_t nDirs = 0, nTransfer = 0;
 786         TIter dirIter(dirList);
 787         TSystemFile* entry = 0;
 788
 789         Bool_t success = kTRUE;
 790         Bool_t first = kTRUE;
 791
 792         while ((entry = dynamic_cast<TSystemFile*> (dirIter.Next())))
 793         {
 794                 if (entry->IsDirectory())
 795                         continue;
 796
 797                 TString fileName(entry->GetName());
 798                 if (!fileName.BeginsWith(begin))
 799                         continue;
 800
 801                 nDirs++;
 802
 803                 if (first)
 804                 {
 805                         first = kFALSE;
 806                         // check that folder exists, otherwise create it
 807                         TGridResult* result = gGrid->Ls(alienDir.Data(), "a");
 808
 809                         if (!result)
 810                         {
 811                                 delete dirList;
 812                                 return kFALSE;
 813                         }
 814
 815                         if (!result->GetFileName(1)) // TODO: It looks like element 0 is always 0!!
 816                         {
 817                                 // TODO It does not work currently! Bug in TAliEn::Mkdir
 818                                 // TODO Manually fixed in local root v5-16-00
 819                                 if (!gGrid->Mkdir(alienDir.Data(),"-p",0))
 820                                 {
 821                                         Log("SHUTTLE", Form("CopyFilesToGrid - Cannot create directory %s",
 822                                                         alienDir.Data()));
 823                                         delete dirList;
 824                                         return kFALSE;
 825                                 } else {
 826                                         Log("SHUTTLE",Form("CopyFilesToGrid - Folder %s created", alienDir.Data()));
 827                                 }
 828
 829                         } else {
 830                                         Log("SHUTTLE",Form("CopyFilesToGrid - Folder %s found", alienDir.Data()));
 831                         }
 832                 }
 833
 834                 TString fullLocalPath;
 835                 fullLocalPath.Form("%s/%s", dir.Data(), fileName.Data());
 836
 837                 TString fullGridPath;
 838                 fullGridPath.Form("alien://%s/%s", alienDir.Data(), fileName.Data());
 839
 840                 Bool_t result = TFile::Cp(fullLocalPath, fullGridPath);
 841
 842                 if (result)
 843                 {
 844                         Log("SHUTTLE", Form("CopyFilesToGrid - Copying local file %s to %s succeeded!",
 845                                                 fullLocalPath.Data(), fullGridPath.Data()));
 846                         RemoveFile(fullLocalPath);
 847                         nTransfer++;
 848                 }
 849                 else
 850                 {
 851                         Log("SHUTTLE", Form("CopyFilesToGrid - Copying local file %s to %s FAILED!",
 852                                                 fullLocalPath.Data(), fullGridPath.Data()));
 853                         success = kFALSE;
 854                 }
 855         }
 856
 857         Log("SHUTTLE", Form("CopyFilesToGrid - %d (over %d) files in folder %s copied to Grid.",
 858                                                 nTransfer, nDirs, dir.Data()));
 859
 860
 861         delete dirList;
 862         return success;
 863 }
 864
 865 //______________________________________________________________________________________________
 866 const char* AliShuttle::GetRefFilePrefix(const char* base, const char* detector)
 867 {
 868         //
 869         // Get folder name of reference files
 870         //
 871
 872         TString offDetStr(GetOfflineDetName(detector));
 873         static TString dir;
 874         if (offDetStr == "ITS" || offDetStr == "MUON" || offDetStr == "PHOS")
 875         {
 876                 dir.Form("%s/%s/%s", base, offDetStr.Data(), detector);
 877         } else {
 878                 dir.Form("%s/%s", base, offDetStr.Data());
 879         }
 880
 881         return dir.Data();
 882 }
 883
 884 //______________________________________________________________________________________________
 885 void AliShuttle::CleanLocalStorage(const TString& uri)
 886 {
 887         //
 888         // Called in case the preprocessor is declared failed. Remove remaining objects from the local storages.
 889         //
 890
 891         const char* type = 0;
 892         if(uri == fgkLocalCDB) {
 893                 type = "OCDB";
 894         } else if(uri == fgkLocalRefStorage) {
 895                 type = "Reference";
 896         } else {
 897                 AliError(Form("Invalid storage URI: %s", uri.Data()));
 898                 return;
 899         }
 900
 901         AliCDBManager* man = AliCDBManager::Instance();
 902
 903         // open local storage
 904         AliCDBStorage *localSto = man->GetStorage(uri);
 905         if(!localSto) {
 906                 Log("SHUTTLE",
 907                         Form("CleanLocalStorage - cannot activate local %s storage", type));
 908                 return;
 909         }
 910
 911         TString filename(Form("%s/%s/*/Run*_v%d_s*.root",
 912                 localSto->GetBaseFolder().Data(), GetOfflineDetName(fCurrentDetector.Data()), GetCurrentRun()));
 913
 914         AliDebug(2, Form("filename = %s", filename.Data()));
 915
 916         Log("SHUTTLE", Form("Removing remaining local files for run %d and detector %s ...",
 917                 GetCurrentRun(), fCurrentDetector.Data()));
 918
 919         RemoveFile(filename.Data());
 920
 921 }
 922
 923 //______________________________________________________________________________________________
 924 void AliShuttle::RemoveFile(const char* filename)
 925 {
 926         //
 927         // removes local file
 928         //
 929
 930         TString command(Form("rm -f %s", filename));
 931
 932         Int_t result = gSystem->Exec(command.Data());
 933         if(result != 0)
 934         {
 935                 Log("SHUTTLE", Form("RemoveFile - %s: Cannot remove file %s!",
 936                         fCurrentDetector.Data(), filename));
 937         }
 938 }
 939
 940 //______________________________________________________________________________________________
 941 AliShuttleStatus* AliShuttle::ReadShuttleStatus()
 942 {
 943         //
 944         // Reads the AliShuttleStatus from the CDB
 945         //
 946
 947         if (fStatusEntry){
 948                 delete fStatusEntry;
 949                 fStatusEntry = 0;
 950         }
 951
 952         Int_t path1 = GetCurrentRun()/10000;
 953         fStatusEntry = AliCDBManager::Instance()->GetStorage(GetLocalCDB())
 954                 ->Get(Form("/SHUTTLE/%s/%d", fCurrentDetector.Data(), path1), GetCurrentRun());
 955
 956         if (!fStatusEntry) return 0;
 957         fStatusEntry->SetOwner(1);
 958
 959         AliShuttleStatus* status = dynamic_cast<AliShuttleStatus*> (fStatusEntry->GetObject());
 960         if (!status) {
 961                 AliError("Invalid object stored to CDB!");
 962                 return 0;
 963         }
 964
 965         return status;
 966 }
 967
 968 //______________________________________________________________________________________________
 969 Bool_t AliShuttle::WriteShuttleStatus(AliShuttleStatus* status)
 970 {
 971         //
 972         // writes the status for one subdetector
 973         //
 974
 975         if (fStatusEntry){
 976                 delete fStatusEntry;
 977                 fStatusEntry = 0;
 978         }
 979
 980         Int_t run = GetCurrentRun();
 981         Int_t path1 = run/10000;
 982         TString path1_string = Form("%d",path1);
 983
 984         AliCDBId id(AliCDBPath("SHUTTLE", fCurrentDetector, path1_string), run, run);
 985
 986         fStatusEntry = new AliCDBEntry(status, id, new AliCDBMetaData);
 987         fStatusEntry->SetOwner(1);
 988
 989         Int_t logLevel = AliLog::GetGlobalLogLevel();
 990         AliLog::SetGlobalLogLevel(AliLog::kError);
 991
 992         UInt_t result = AliCDBManager::Instance()->GetStorage(fgkLocalCDB)->Put(fStatusEntry);
 993
 994         AliLog::SetGlobalLogLevel((AliLog::EType_t)logLevel);
 995
 996         if (!result) {
 997                 Log("SHUTTLE", Form("WriteShuttleStatus - Failed for %s, run %d",
 998                                                 fCurrentDetector.Data(), run));
 999                 return kFALSE;
1000         }
1001
1002         SendMLDetInfo();
1003
1004         return kTRUE;
1005 }
1006
1007 //______________________________________________________________________________________________
1008 void AliShuttle::UpdateShuttleStatus(AliShuttleStatus::Status newStatus, Bool_t increaseCount)
1009 {
1010         //
1011         // changes the AliShuttleStatus for the given detector and run to the given status
1012         //
1013
1014         if (!fStatusEntry){
1015                 AliError("UNEXPECTED: fStatusEntry empty");
1016                 return;
1017         }
1018
1019         AliShuttleStatus* status = dynamic_cast<AliShuttleStatus*> (fStatusEntry->GetObject());
1020
1021         if (!status){
1022                 Log("SHUTTLE", "UpdateShuttleStatus - UNEXPECTED: status could not be read from current CDB entry");
1023                 return;
1024         }
1025
1026         TString actionStr = Form("UpdateShuttleStatus - %s: Changing state from %s to %s",
1027                                 fCurrentDetector.Data(),
1028                                 status->GetStatusName(),
1029                                 status->GetStatusName(newStatus));
1030         Log("SHUTTLE", actionStr);
1031         SetLastAction(actionStr);
1032
1033         status->SetStatus(newStatus);
1034         if (increaseCount) status->IncreaseCount();
1035
1036         Int_t logLevel = AliLog::GetGlobalLogLevel();
1037         AliLog::SetGlobalLogLevel(AliLog::kError);
1038
1039         AliCDBManager::Instance()->GetStorage(fgkLocalCDB)->Put(fStatusEntry);
1040
1041         AliLog::SetGlobalLogLevel((AliLog::EType_t)logLevel);
1042
1043         SendMLDetInfo();
1044 }
1045
1046 //______________________________________________________________________________________________
1047 void AliShuttle::SendMLDetInfo()
1048 {
1049         //
1050         // sends ML information about the current status of the current detector being processed
1051         //
1052
1053         AliShuttleStatus* status = dynamic_cast<AliShuttleStatus*> (fStatusEntry->GetObject());
1054
1055         if (!status){
1056                 Log("SHUTTLE", "SendMLDetInfo - UNEXPECTED: status could not be read from current CDB entry");
1057                 return;
1058         }
1059
1060         TMonaLisaText  mlStatus(Form("%s_status", fCurrentDetector.Data()), status->GetStatusName());
1061         TMonaLisaValue mlRetryCount(Form("%s_count", fCurrentDetector.Data()), status->GetCount());
1062
1063         TList mlList;
1064         mlList.Add(&mlStatus);
1065         mlList.Add(&mlRetryCount);
1066
1067         TString mlID;
1068         mlID.Form("%d", GetCurrentRun());
1069         fMonaLisa->SendParameters(&mlList, mlID);
1070 }
1071
1072 //______________________________________________________________________________________________
1073 Bool_t AliShuttle::ContinueProcessing()
1074 {
1075         // this function reads the AliShuttleStatus information from CDB and
1076         // checks if the processing should be continued
1077         // if yes it returns kTRUE and updates the AliShuttleStatus with nextStatus
1078
1079         if (!fConfig->HostProcessDetector(fCurrentDetector))
1080                 return kFALSE;
1081
1082         AliPreprocessor* aPreprocessor =
1083                 dynamic_cast<AliPreprocessor*> (fPreprocessorMap.GetValue(fCurrentDetector));
1084         if (!aPreprocessor)
1085         {
1086                 Log("SHUTTLE", Form("ContinueProcessing - %s: no preprocessor registered", fCurrentDetector.Data()));
1087                 return kFALSE;
1088         }
1089
1090         AliShuttleLogbookEntry::Status entryStatus =
1091                 fLogbookEntry->GetDetectorStatus(fCurrentDetector);
1092
1093         if (entryStatus != AliShuttleLogbookEntry::kUnprocessed) {
1094                 Log("SHUTTLE", Form("ContinueProcessing - %s is %s",
1095                                 fCurrentDetector.Data(),
1096                                 fLogbookEntry->GetDetectorStatusName(entryStatus)));
1097                 return kFALSE;
1098         }
1099
1100         // if we get here, according to Shuttle logbook subdetector is in UNPROCESSED state
1101
1102         // check if current run is first unprocessed run for current detector
1103         if (fConfig->StrictRunOrder(fCurrentDetector) &&
1104                 !fFirstUnprocessed[GetDetPos(fCurrentDetector)])
1105         {
1106                 if (fTestMode == kNone)
1107                 {
1108                         Log("SHUTTLE", Form("ContinueProcessing - %s requires strict run ordering"
1109                                         " but this is not the first unprocessed run!"));
1110                         return kFALSE;
1111                 }
1112                 else
1113                 {
1114                         Log("SHUTTLE", Form("ContinueProcessing - In TESTMODE - "
1115                                         "Although %s requires strict run ordering "
1116                                         "and this is not the first unprocessed run, "
1117                                         "the SHUTTLE continues"));
1118                 }
1119         }
1120
1121         // Is the subdetector processed first time for this run?
1122         fFirstProcessing = kFALSE;
1123
1124         AliShuttleStatus* status = ReadShuttleStatus();
1125         if (!status) {
1126                 // first time
1127                 Log("SHUTTLE", Form("ContinueProcessing - %s: Processing first time",
1128                                 fCurrentDetector.Data()));
1129                 status = new AliShuttleStatus(AliShuttleStatus::kStarted);
1130                 fFirstProcessing = kTRUE;
1131                 return WriteShuttleStatus(status);
1132         }
1133
1134         // The following case shouldn't happen if Shuttle Logbook was correctly updated.
1135         // If it happens it may mean Logbook updating failed... let's do it now!
1136         if (status->GetStatus() == AliShuttleStatus::kDone ||
1137             status->GetStatus() == AliShuttleStatus::kFailed ||
1138             status->GetStatus() == AliShuttleStatus::kSkipped) {
1139                 Log("SHUTTLE", Form("ContinueProcessing - %s is already %s. Updating Shuttle Logbook",
1140                                         fCurrentDetector.Data(),
1141                                         status->GetStatusName(status->GetStatus())));
1142
1143                 if (status->GetStatus() == AliShuttleStatus::kSkipped)
1144                 {
1145                         UpdateShuttleLogbook(fCurrentDetector.Data(), "DONE");
1146                 }
1147                 else
1148                         UpdateShuttleLogbook(fCurrentDetector.Data(), status->GetStatusName(status->GetStatus()));
1149
1150                 return kFALSE;
1151         }
1152
1153         if (status->GetStatus() == AliShuttleStatus::kStoreStarted || status->GetStatus() == AliShuttleStatus::kStoreDelayed ||status->GetStatus() == AliShuttleStatus::kStoreError) {
1154                 Log("SHUTTLE",
1155                         Form("ContinueProcessing - %s: Grid storage of one or more "
1156                                 "objects failed. Trying again now",
1157                                 fCurrentDetector.Data()));
1158                 StoreOCDB();
1159                 return kFALSE;
1160         }
1161
1162         // if we get here, there is a restart
1163         Bool_t cont = kFALSE;
1164
1165         // abort conditions
1166         if (status->GetCount() >= fConfig->GetMaxRetries()) {
1167                 Log("SHUTTLE", Form("ContinueProcessing - %s failed %d times in status %s - "
1168                                 "Updating Shuttle Logbook", fCurrentDetector.Data(),
1169                                 status->GetCount(), status->GetStatusName()));
1170                 UpdateShuttleLogbook(fCurrentDetector.Data(), "FAILED");
1171                 UpdateShuttleStatus(AliShuttleStatus::kFailed);
1172
1173                 // there may still be objects in local OCDB and reference storage
1174                 // and FXS databases may be not updated: do it now!
1175
1176                 // TODO Currently disabled, we want to keep files in case of failure!
1177                 // CleanLocalStorage(fgkLocalCDB);
1178                 // CleanLocalStorage(fgkLocalRefStorage);
1179                 // UpdateTableFailCase();
1180
1181                 // Send mail to detector expert!
1182                 Log("SHUTTLE", Form("ContinueProcessing - Sending mail to %s expert...",
1183                                     fCurrentDetector.Data()));
1184                 // det experts in to
1185                 TString to="";
1186                 TIter *iterExperts = 0;
1187                 iterExperts = new TIter(fConfig->GetResponsibles(fCurrentDetector));
1188                 TObjString *anExpert=0;
1189                 while ((anExpert = (TObjString*) iterExperts->Next()))
1190                         {
1191                                 to += Form("%s, \n", anExpert->GetName());
1192                         }
1193                 delete iterExperts;
1194
1195                 if (to.Length() > 0)
1196                         to.Remove(to.Length()-3);
1197                 AliDebug(2, Form("to: %s",to.Data()));
1198
1199                 if (to.IsNull()) {
1200                         Log("SHUTTLE", Form("List of %s responsibles not set!", fCurrentDetector.Data()));
1201                         return kFALSE;
1202                 }
1203
1204                 Log(fCurrentDetector.Data(), Form("ContinueProcessing - Sending mail to %s expert(s):",
1205                                     fCurrentDetector.Data()));
1206                 Log(fCurrentDetector.Data(), Form("\n%s", to.Data()));
1207                 if (!SendMail(kPPEMail))
1208                         Log("SHUTTLE", Form("ContinueProcessing - Could not send mail to %s expert",
1209                                             fCurrentDetector.Data()));
1210
1211         } else {
1212                 Log("SHUTTLE", Form("ContinueProcessing - %s: restarting. "
1213                                 "Aborted before with %s. Retry number %d.", fCurrentDetector.Data(),
1214                                 status->GetStatusName(), status->GetCount()));
1215                 Bool_t increaseCount = kTRUE;
1216                 if (status->GetStatus() == AliShuttleStatus::kDCSError ||
1217                     status->GetStatus() == AliShuttleStatus::kDCSStarted ||
1218                     status->GetStatus() == AliShuttleStatus::kFXSError ||
1219                     status->GetStatus() == AliShuttleStatus::kOCDBError)
1220                                 increaseCount = kFALSE;
1221
1222                 UpdateShuttleStatus(AliShuttleStatus::kStarted, increaseCount);
1223                 cont = kTRUE;
1224         }
1225
1226         return cont;
1227 }
1228
1229 //______________________________________________________________________________________________
1230 void AliShuttle::SendMLRunInfo(const char* status)
1231 {
1232         //
1233         // Send information about this run to ML
1234
1235         TMonaLisaText  mlStatus("SHUTTLE_status", status);
1236         TString runType(fLogbookEntry->GetRunType());
1237         if (strlen(fLogbookEntry->GetRunParameter("log")) > 0){
1238
1239                 runType += "(";
1240                 runType += fLogbookEntry->GetRunParameter("log");
1241                 runType += ")";
1242         }
1243         if (fLogbookEntry->GetDATestMode()){
1244                 runType += " (DATest)";
1245         }
1246         TMonaLisaText  mlRunType("SHUTTLE_runtype", runType);
1247
1248         TList mlList;
1249         mlList.Add(&mlStatus);
1250         mlList.Add(&mlRunType);
1251
1252         TString mlID;
1253         mlID.Form("%d", GetCurrentRun());
1254         fMonaLisa->SendParameters(&mlList, mlID);
1255 }
1256
1257 //______________________________________________________________________________________________
1258 Int_t AliShuttle::GetMem(Int_t pid)
1259 {
1260         // invokes ps to get the memory consumption of the process <pid>
1261         // returns -1 in case of error
1262
1263         TString checkStr;
1264         checkStr.Form("ps -o vsize --pid %d | tail -n 1", pid);
1265         FILE* pipe = gSystem->OpenPipe(checkStr, "r");
1266         if (!pipe)
1267         {
1268                 Log("SHUTTLE", Form("Process - Error: "
1269                         "Could not open pipe to %s", checkStr.Data()));
1270                 return -1;
1271         }
1272
1273         char buffer[100];
1274         if (!fgets(buffer, 100, pipe))
1275         {
1276                 Log("SHUTTLE", "Process - Error: ps did not return anything");
1277                 gSystem->ClosePipe(pipe);
1278                 return -1;
1279         }
1280         gSystem->ClosePipe(pipe);
1281
1282         //Log("SHUTTLE", Form("ps returned %s", buffer));
1283
1284         Int_t mem = 0;
1285         if ((sscanf(buffer, "%d\n", &mem) != 1) || !mem)
1286         {
1287                 Log("SHUTTLE", "Process - Error: Could not parse output of ps");
1288                 return -1;
1289         }
1290
1291         return mem;
1292 }
1293
1294 //______________________________________________________________________________________________
1295 Bool_t AliShuttle::Process(AliShuttleLogbookEntry* entry)
1296 {
1297         //
1298         // Makes data retrieval for all detectors in the configuration.
1299         // entry: Shuttle logbook entry, contains run paramenters and status of detectors
1300         // (Unprocessed, Inactive, Failed or Done).
1301         // Returns kFALSE in case of error occured and kTRUE otherwise
1302         //
1303
1304         if (!entry) return kFALSE;
1305
1306         fLogbookEntry = entry;
1307
1308         Log("SHUTTLE", Form("\t\t\t^*^*^*^*^*^*^*^*^*^*^*^* run %d: START ^*^*^*^*^*^*^*^*^*^*^*^*",
1309                                         GetCurrentRun()));
1310
1311         CountOpenRuns();
1312
1313         // Send the information to ML
1314         SendMLRunInfo("Processing");
1315
1316         if (fLogbookEntry->IsDone())
1317         {
1318                 Log("SHUTTLE","Process - Shuttle is already DONE. Updating logbook");
1319                 UpdateShuttleLogbook("shuttle_done");
1320                 fLogbookEntry = 0;
1321                 return kTRUE;
1322         }
1323
1324         // read test mode if flag is set
1325         if (fReadTestMode)
1326         {
1327                 fTestMode = kNone;
1328                 TString logEntry(entry->GetRunParameter("log"));
1329                 //printf("log entry = %s\n", logEntry.Data());
1330                 TString searchStr("Testmode: ");
1331                 Int_t pos = logEntry.Index(searchStr.Data());
1332                 //printf("%d\n", pos);
1333                 if (pos >= 0)
1334                 {
1335                         TSubString subStr = logEntry(pos + searchStr.Length(), logEntry.Length());
1336                         //printf("%s\n", subStr.String().Data());
1337                         TString newStr(subStr.Data());
1338                         TObjArray* token = newStr.Tokenize(' ');
1339                         if (token)
1340                         {
1341                                 //token->Print();
1342                                 TObjString* tmpStr = dynamic_cast<TObjString*> (token->First());
1343                                 if (tmpStr)
1344                                 {
1345                                         Int_t testMode = tmpStr->String().Atoi();
1346                                         if (testMode > 0)
1347                                         {
1348                                                 Log("SHUTTLE", Form("Process - Enabling test mode %d", testMode));
1349                                                 SetTestMode((TestMode) testMode);
1350                                         }
1351                                 }
1352                                 delete token;
1353                         }
1354                 }
1355         }
1356
1357         fLogbookEntry->Print("all");
1358
1359         // Initialization
1360         Bool_t hasError = kFALSE;
1361
1362         // Set the CDB and Reference folders according to the year
1363
1364         // build cdb paths (repeat each time, run might be a DATest run)
1365         if (!fLogbookEntry->GetDATestMode()){
1366                 fgkMainCDB.Form("alien://folder=%s%d/OCDB?user=alidaq?cacheFold=/tmp/OCDBCache",
1367                                 fConfig->GetAlienPath(), GetCurrentYear());
1368
1369                 fgkMainRefStorage.Form("alien://folder=%s%d/Reference?user=alidaq?cacheFold=/tmp/OCDBCache",
1370                                        fConfig->GetAlienPath(), GetCurrentYear());
1371         }
1372         else {
1373                 fgkMainCDB.Form("alien://folder=%s%d/DATest/OCDB?user=alidaq?cacheFold=/tmp/OCDBCache",
1374                                 fConfig->GetAlienPath(), GetCurrentYear());
1375
1376                 fgkMainRefStorage.Form("alien://folder=%s%d/DATest/Reference?user=alidaq?cacheFold=/tmp/OCDBCache",
1377                                        fConfig->GetAlienPath(), GetCurrentYear());
1378         }
1379
1380         AliDebug(2,Form("Main CDB storage = %s",fgkMainCDB.Data()));
1381         AliDebug(2,Form("Main Reference storage = %s",fgkMainRefStorage.Data()));
1382
1383         // Loop on detectors in the configuration
1384         TIter iter(fConfig->GetDetectors());
1385         TObjString* aDetector = 0;
1386
1387         Bool_t first = kTRUE;
1388
1389         while ((aDetector = (TObjString*) iter.Next()))
1390         {
1391                 fCurrentDetector = aDetector->String();
1392
1393                 if (ContinueProcessing() == kFALSE)
1394                         continue;
1395
1396                 if (first)
1397                 {
1398                   // only read QueryCDB when needed and only once
1399                   AliCDBStorage *mainCDBSto = AliCDBManager::Instance()->GetStorage(fgkMainCDB);
1400                   if(mainCDBSto) mainCDBSto->QueryCDB(GetCurrentRun());
1401                   AliCDBStorage *mainRefSto = AliCDBManager::Instance()->GetStorage(fgkMainRefStorage);
1402                   if(mainRefSto) mainRefSto->QueryCDB(GetCurrentRun());
1403                   first = kFALSE;
1404                 }
1405
1406                 Log("SHUTTLE", Form("\t\t\t****** run %d - %s: START  ******",
1407                                                 GetCurrentRun(), aDetector->GetName()));
1408
1409                 for(Int_t iSys=0;iSys<3;iSys++) fFXSCalled[iSys]=kFALSE;
1410
1411                 Int_t initialMem = GetMem(getpid());
1412                 Log("SHUTTLE", Form("Memory consumption before forking is %d", initialMem));
1413
1414                 Log(fCurrentDetector.Data(), "Process - Starting processing");
1415
1416                 Int_t pid = fork();
1417
1418                 if (pid < 0)
1419                 {
1420                         Log("SHUTTLE", "Process - ERROR: Forking failed");
1421                 }
1422                 else if (pid > 0)
1423                 {
1424                         // parent
1425                         Log("SHUTTLE", Form("Process - In parent process of %d - %s: Starting monitoring",
1426                                                         GetCurrentRun(), aDetector->GetName()));
1427
1428                         Long_t begin = time(0);
1429
1430                         int status; // to be used with waitpid, on purpose an int (not Int_t)!
1431                         while (waitpid(pid, &status, WNOHANG) == 0)
1432                         {
1433                                 Long_t expiredTime = time(0) - begin;
1434
1435                                 if (expiredTime > fConfig->GetPPTimeOut())
1436                                 {
1437                                         TString logMsg;
1438                                         AliShuttleStatus *currentStatus = ReadShuttleStatus();
1439                                         AliShuttleStatus::Status newStatus = AliShuttleStatus::kInvalid;
1440
1441                                         if (currentStatus->GetStatus() == AliShuttleStatus::kDCSStarted)
1442                                         {
1443                                                 // in case the pp goes in TimeOut while retrieving the DCS DPs
1444                                                 // set status to kDCSError
1445
1446                                                 logMsg.Form("Process - Process of %s timed out while retrieving the DCS DataPoints. Run time: %d seconds. Killing... and setting status to DCSError.",
1447                                                                 fCurrentDetector.Data(), expiredTime);
1448                                                 newStatus = AliShuttleStatus::kDCSError;
1449                                         }
1450                                         else if (currentStatus->GetStatus() <= AliShuttleStatus::kPPDone)
1451                                         {
1452                                                 // in case pp not yet done set status to kPPTimeOut
1453
1454                                                 logMsg.Form("Process - Process of %s timed out. Run time: %d seconds. Killing...",
1455                                                                 fCurrentDetector.Data(), expiredTime);
1456                                                 newStatus = AliShuttleStatus::kPPTimeOut;
1457                                         }
1458                                         else if (currentStatus->GetStatus() == AliShuttleStatus::kStoreStarted)
1459                                         {
1460                                                 // in case the pp goes in TimeOut while storing the objects in the OCDB
1461                                                 // set status to kStoreError
1462
1463                                                 logMsg.Form("Process - Process of %s timed out while storing the OCDB object. Run time: %d seconds. Killing... and setting status to StoreError.",
1464                                                                 fCurrentDetector.Data(), expiredTime);
1465                                                 newStatus = AliShuttleStatus::kStoreError;
1466                                         }
1467                                         else
1468                                         {
1469                                                 // in other cases don't change the status
1470
1471                                                 logMsg.Form("Process - Process of %s timed out in status = %s. Run time: %d seconds. Killing... without changing the status",
1472                                                                 fCurrentDetector.Data(), currentStatus->GetStatusName(), expiredTime);
1473                                         }
1474
1475                                         Log("SHUTTLE", logMsg);
1476                                         Log(fCurrentDetector, logMsg);
1477
1478                                         kill(pid, 9);
1479
1480                                         if (newStatus != AliShuttleStatus::kInvalid)
1481                                                 UpdateShuttleStatus(newStatus);
1482                                         hasError = kTRUE;
1483
1484                                         gSystem->Sleep(1000);
1485                                 }
1486                                 else
1487                                 {
1488                                         gSystem->Sleep(1000);
1489
1490                                         Int_t mem = GetMem(pid);
1491
1492                                         if (mem < 0)
1493                                                 continue;
1494
1495                                         mem -= initialMem;
1496                                         if (mem < 0)
1497                                                 mem = 0;
1498
1499                                         if (expiredTime % 60 == 0)
1500                                         {
1501                                                 Log("SHUTTLE", Form("Process - %s: Checking process. "
1502                                                         "Run time: %d seconds - Memory consumption: %d KB",
1503                                                         fCurrentDetector.Data(), expiredTime, mem));
1504                                                 SendAlive();
1505                                         }
1506
1507                                         if (mem > fConfig->GetPPMaxMem())
1508                                         {
1509                                                 TString tmp;
1510                                                 tmp.Form("Process - Process exceeds maximum allowed memory "
1511                                                         "(%d KB > %d KB). Killing...",
1512                                                         mem, fConfig->GetPPMaxMem());
1513                                                 Log("SHUTTLE", tmp);
1514                                                 Log(fCurrentDetector, tmp);
1515
1516                                                 kill(pid, 9);
1517
1518                                                 UpdateShuttleStatus(AliShuttleStatus::kPPOutOfMemory);
1519                                                 hasError = kTRUE;
1520
1521                                                 gSystem->Sleep(1000);
1522                                         }
1523                                 }
1524                         }
1525
1526                         Log("SHUTTLE", Form("Process - In parent process of %d - %s: Client has terminated.",
1527                                                                 GetCurrentRun(), aDetector->GetName()));
1528
1529                         if (WIFEXITED(status))
1530                         {
1531                                 Int_t returnCode = WEXITSTATUS(status);
1532
1533                                 Log("SHUTTLE", Form("Process - %s: the return code is %d", fCurrentDetector.Data(),
1534                                                                                 returnCode));
1535
1536                                 if (returnCode == 0) hasError = kTRUE;
1537                         }
1538                 }
1539                 else if (pid == 0)
1540                 {
1541                         // child
1542                         Log("SHUTTLE", Form("Process - In child process of %d - %s", GetCurrentRun(),
1543                                 aDetector->GetName()));
1544
1545                         Log("SHUTTLE", Form("Process - Redirecting output to %s log",fCurrentDetector.Data()));
1546
1547                         if ((freopen(GetLogFileName(fCurrentDetector), "a", stdout)) == 0)
1548                         {
1549                                 Log("SHUTTLE", "Process - Could not freopen stdout");
1550                         }
1551                         else
1552                         {
1553                                 fOutputRedirected = kTRUE;
1554                                 if ((dup2(fileno(stdout), fileno(stderr))) < 0)
1555                                         Log("SHUTTLE", "Process - Could not redirect stderr");
1556
1557                         }
1558
1559                         Log("SHUTTLE", "Executing TGrid::Connect");
1560                         TGrid::Connect("alien://");
1561
1562                         TString wd = gSystem->WorkingDirectory();
1563                         Int_t dir_lev1 = GetCurrentRun()/10000;
1564                         TString tmpDir = Form("%s/%d/%d/%s_process", GetShuttleTempDir(),
1565                                 dir_lev1, GetCurrentRun(), fCurrentDetector.Data());
1566
1567                         Int_t result = gSystem->GetPathInfo(tmpDir.Data(), 0, (Long64_t*) 0, 0, 0);
1568                         if (!result) // temp dir already exists!
1569                         {
1570                                 Log(fCurrentDetector.Data(),
1571                                         Form("Process - %s dir already exists! Removing...", tmpDir.Data()));
1572                                 gSystem->Exec(Form("rm -rf %s",tmpDir.Data()));
1573                         }
1574
1575                         if (gSystem->mkdir(tmpDir.Data(), 1))
1576                         {
1577                                 Log(fCurrentDetector.Data(), "Process - could not make temp directory!!");
1578                                 gSystem->Exit(1);
1579                         }
1580
1581                         if (!gSystem->ChangeDirectory(tmpDir.Data()))
1582                         {
1583                                 Log(fCurrentDetector.Data(), "Process - could not change directory!!");
1584                                 gSystem->Exit(1);
1585                         }
1586
1587                         Int_t success = ProcessCurrentDetector();
1588
1589                         gSystem->ChangeDirectory(wd.Data());
1590
1591                         if (success == 1) // Preprocessor finished successfully!
1592                         {
1593                                 // remove temporary folder or DCS map
1594                                 if (!fConfig->KeepTempFolder())
1595                                 {
1596                                         gSystem->Exec(Form("rm -rf %s",tmpDir.Data()));
1597                                 } else if (!fConfig->KeepDCSMap())
1598                                 {
1599                                         gSystem->Exec(Form("rm -f %s/DCSMap.root",tmpDir.Data()));
1600                                 }
1601
1602                                 // Update time_processed field in FXS DB
1603                                 if (UpdateTable() == kFALSE)
1604                                         Log("SHUTTLE", Form("Process - %s: Could not update FXS databases!",
1605                                                         fCurrentDetector.Data()));
1606
1607                                 // Transfer the data from local storage to main storage (Grid)
1608                                 if (StoreOCDB() == kFALSE)
1609                                         success = kFALSE;
1610                         }
1611                         else if (success == 0)
1612                         {
1613                                 Log("SHUTTLE",
1614                                         Form("\t\t\t****** run %d - %s: ERROR ******",
1615                                                 GetCurrentRun(), aDetector->GetName()));
1616                         }
1617
1618                         for (UInt_t iSys=0; iSys<3; iSys++)
1619                         {
1620                                 if (fFXSCalled[iSys]) fFXSlist[iSys].Clear();
1621                         }
1622
1623                         Log("SHUTTLE", Form("Process - Client process of %d - %s is exiting now with %d.",
1624                                                         GetCurrentRun(), aDetector->GetName(), success));
1625
1626                         // the client exits here
1627                         gSystem->Exit(success);
1628
1629                         AliError("We should never get here!!!");
1630                 }
1631         }
1632
1633         Log("SHUTTLE", Form("\t\t\t^*^*^*^*^*^*^*^*^*^*^*^* run %d: FINISH ^*^*^*^*^*^*^*^*^*^*^*^*",
1634                                                         GetCurrentRun()));
1635
1636         //check if shuttle is done for this run, if so update logbook
1637         TObjArray checkEntryArray;
1638         checkEntryArray.SetOwner(1);
1639         TString whereClause = Form("where run=%d", GetCurrentRun());
1640         if (!QueryShuttleLogbook(whereClause.Data(), checkEntryArray) ||
1641                         checkEntryArray.GetEntries() == 0) {
1642                 Log("SHUTTLE", Form("Process - Warning: Cannot check status of run %d on Shuttle logbook!",
1643                                                 GetCurrentRun()));
1644                 return hasError == kFALSE;
1645         }
1646
1647         AliShuttleLogbookEntry* checkEntry = dynamic_cast<AliShuttleLogbookEntry*>
1648                                                 (checkEntryArray.At(0));
1649
1650         if (checkEntry)
1651         {
1652                 if (checkEntry->IsDone())
1653                 {
1654                         Log("SHUTTLE","Process - Shuttle is DONE. Updating logbook");
1655                         UpdateShuttleLogbook("shuttle_done");
1656                 }
1657                 else
1658                 {
1659                         for (UInt_t iDet=0; iDet<NDetectors(); iDet++)
1660                         {
1661                                 if (checkEntry->GetDetectorStatus(iDet) == AliShuttleLogbookEntry::kUnprocessed)
1662                                 {
1663                                         AliDebug(2, Form("Run %d: setting %s as \"not first time unprocessed\"",
1664                                                         checkEntry->GetRun(), GetDetName(iDet)));
1665                                         fFirstUnprocessed[iDet] = kFALSE;
1666                                 }
1667                         }
1668                         SendMLRunInfo("Pending");
1669                 }
1670         }
1671
1672         fLogbookEntry = 0;
1673
1674         return hasError == kFALSE;
1675 }
1676
1677 //______________________________________________________________________________________________
1678 Int_t AliShuttle::ProcessCurrentDetector()
1679 {
1680         //
1681         // Makes data retrieval just for a specific detector (fCurrentDetector).
1682         // Threre should be a configuration for this detector.
1683
1684         Log("SHUTTLE", Form("ProcessCurrentDetector - Retrieving values for %s, run %d",
1685                                                 fCurrentDetector.Data(), GetCurrentRun()));
1686
1687         TString wd = gSystem->WorkingDirectory();
1688
1689         if (!CleanReferenceStorage(fCurrentDetector.Data()))
1690                 return 0;
1691
1692         gSystem->ChangeDirectory(wd.Data());
1693
1694         // call preprocessor
1695         AliPreprocessor* aPreprocessor =
1696                 dynamic_cast<AliPreprocessor*> (fPreprocessorMap.GetValue(fCurrentDetector));
1697
1698         // check if the preprocessor wants to process this run type
1699         if (aPreprocessor->ProcessRunType() == kFALSE)
1700         {
1701                 UpdateShuttleStatus(AliShuttleStatus::kSkipped);
1702                 UpdateShuttleLogbook(fCurrentDetector, "DONE");
1703                 if (!UpdateTableSkippedCase(fCurrentDetector.Data()))
1704                 {
1705                         AliError(Form("Could not update FXS tables for run %d !", GetCurrentRun()));
1706                 }
1707                 Log(fCurrentDetector, Form("ProcessCurrentDetector - %s preprocessor is not interested in this run type", fCurrentDetector.Data()));
1708
1709                 return 2;
1710         }
1711
1712         // checking if OCDB is reachable
1713         AliCDBEntry* testEntry = GetFromOCDB("SHUTTLE","GRP/CTP/DummyConfig");
1714         if (!testEntry){
1715                 // OCDB is not accessible, going in OCDBError for current detector
1716                 AliError("OCDB Test entry not accessible");
1717                 UpdateShuttleStatus(AliShuttleStatus::kOCDBError);
1718                 return 0;
1719         }
1720
1721         TMap* dcsMap = new TMap();
1722
1723         aPreprocessor->Initialize(GetCurrentRun(), GetCurrentStartTime(), GetCurrentEndTime());
1724
1725         Bool_t processDCS = aPreprocessor->ProcessDCS();
1726
1727         if (!processDCS)
1728         {
1729                 Log(fCurrentDetector, "ProcessCurrentDetector -"
1730                         " The preprocessor requested to skip the retrieval of DCS values");
1731         }
1732         else if (fTestMode & kSkipDCS)
1733         {
1734                 Log(fCurrentDetector, "ProcessCurrentDetector - In TESTMODE: Skipping DCS processing");
1735         }
1736         else if (fTestMode & kErrorDCS)
1737         {
1738                 Log(fCurrentDetector, "ProcessCurrentDetector - In TESTMODE: Simulating DCS error");
1739                 UpdateShuttleStatus(AliShuttleStatus::kDCSStarted);
1740                 UpdateShuttleStatus(AliShuttleStatus::kDCSError);
1741                 delete dcsMap;
1742                 return 0;
1743         } else {
1744
1745                 UpdateShuttleStatus(AliShuttleStatus::kDCSStarted);
1746
1747                 // Query DCS archive
1748                 Int_t nServers = fConfig->GetNServers(fCurrentDetector);
1749
1750                 for (int iServ=0; iServ<nServers; iServ++)
1751                 {
1752
1753                         TString host(fConfig->GetDCSHost(fCurrentDetector, iServ));
1754                         Int_t port = fConfig->GetDCSPort(fCurrentDetector, iServ);
1755                         Int_t multiSplit = fConfig->GetMultiSplit(fCurrentDetector, iServ);
1756
1757                         Log(fCurrentDetector, Form("ProcessCurrentDetector -"
1758                                         " Querying DCS Amanda server %s:%d (%d of %d)",
1759                                         host.Data(), port, iServ+1, nServers));
1760
1761                         TMap* aliasMap = 0;
1762                         TMap* dpMap = 0;
1763
1764                         if (fConfig->GetDCSAliases(fCurrentDetector, iServ)->GetEntries() > 0)
1765                         {
1766                                 Log(fCurrentDetector, Form("Querying %d DCS aliases", fConfig->GetDCSAliases(fCurrentDetector, iServ)->GetEntries()));
1767                                 aliasMap = GetValueSet(host, port,
1768                                                 fConfig->GetDCSAliases(fCurrentDetector, iServ),
1769                                                 kAlias, multiSplit);
1770                                 if (!aliasMap)
1771                                 {
1772                                         Log(fCurrentDetector,
1773                                                 Form("ProcessCurrentDetector -"
1774                                                         " Error retrieving DCS aliases from server %s."
1775                                                         " Sending mail to DCS experts!", host.Data()));
1776                                         UpdateShuttleStatus(AliShuttleStatus::kDCSError);
1777
1778                                         if (!SendMail(kDCSEMail))
1779                                                 Log("SHUTTLE", Form("ProcessCurrentDetector - "
1780                                                                     "Could not send mail to DCS experts!"));
1781
1782                                         delete dcsMap;
1783                                         return 0;
1784                                 }
1785                         }
1786
1787                         if (fConfig->GetDCSDataPoints(fCurrentDetector, iServ)->GetEntries() > 0)
1788                         {
1789                                 Log(fCurrentDetector, Form("Querying %d DCS data points", fConfig->GetDCSDataPoints(fCurrentDetector, iServ)->GetEntries()));
1790                                 dpMap = GetValueSet(host, port,
1791                                                 fConfig->GetDCSDataPoints(fCurrentDetector, iServ),
1792                                                 kDP, multiSplit);
1793                                 if (!dpMap)
1794                                 {
1795                                         Log(fCurrentDetector,
1796                                                 Form("ProcessCurrentDetector -"
1797                                                         " Error retrieving DCS data points from server %s."
1798                                                         " Sending mail to DCS experts!", host.Data()));
1799                                         UpdateShuttleStatus(AliShuttleStatus::kDCSError);
1800
1801                                         if (!SendMail(kDCSEMail))
1802                                                 Log("SHUTTLE", Form("ProcessCurrentDetector - "
1803                                                                     "Could not send mail to DCS experts!"));
1804
1805                                         if (aliasMap) delete aliasMap;
1806                                         delete dcsMap;
1807                                         return 0;
1808                                 }
1809                         }
1810
1811                         // merge aliasMap and dpMap into dcsMap
1812                         if(aliasMap) {
1813                                 TIter iter(aliasMap);
1814                                 TObjString* key = 0;
1815                                 while ((key = (TObjString*) iter.Next()))
1816                                         dcsMap->Add(key, aliasMap->GetValue(key->String()));
1817
1818                                 aliasMap->SetOwner(kFALSE);
1819                                 delete aliasMap;
1820                         }
1821
1822                         if(dpMap) {
1823                                 TIter iter(dpMap);
1824                                 TObjString* key = 0;
1825                                 while ((key = (TObjString*) iter.Next()))
1826                                         dcsMap->Add(key, dpMap->GetValue(key->String()));
1827
1828                                 dpMap->SetOwner(kFALSE);
1829                                 delete dpMap;
1830                         }
1831                 }
1832         }
1833
1834         // save map into file, to help debugging in case of preprocessor error
1835         TFile* f = TFile::Open("DCSMap.root","recreate");
1836         f->cd();
1837         dcsMap->Write("DCSMap", TObject::kSingleKey);
1838         f->Close();
1839         delete f;
1840
1841         // DCS Archive DB processing successful. Call Preprocessor!
1842         UpdateShuttleStatus(AliShuttleStatus::kPPStarted);
1843
1844         fFXSError = -1; // this variable is kTRUE after ::Process if an FXS error occured
1845
1846         UInt_t returnValue = aPreprocessor->Process(dcsMap);
1847
1848         if (fFXSError!=-1) {
1849                 UpdateShuttleStatus(AliShuttleStatus::kFXSError);
1850                 SendMail(kFXSEMail, fFXSError);
1851                 dcsMap->DeleteAll();
1852                 delete dcsMap;
1853                 return 0;
1854         }
1855
1856         if (returnValue > 0) // Preprocessor error!
1857         {
1858                 Log(fCurrentDetector, Form("ProcessCurrentDetector - "
1859                                 "Preprocessor failed. Process returned %d.", returnValue));
1860                 UpdateShuttleStatus(AliShuttleStatus::kPPError);
1861                 dcsMap->DeleteAll();
1862                 delete dcsMap;
1863                 return 0;
1864         }
1865
1866         // preprocessor ok!
1867         UpdateShuttleStatus(AliShuttleStatus::kPPDone);
1868         Log(fCurrentDetector, Form("ProcessCurrentDetector - %s preprocessor returned success",
1869                                 fCurrentDetector.Data()));
1870
1871         dcsMap->DeleteAll();
1872         delete dcsMap;
1873
1874         return 1;
1875 }
1876
1877 //______________________________________________________________________________________________
1878 void AliShuttle::CountOpenRuns()
1879 {
1880         // Query DAQ's Shuttle logbook and sends the number of open runs to ML
1881
1882         SendAlive();
1883
1884         // check connection, in case connect
1885         if (!Connect(3))
1886                 return;
1887
1888         TString sqlQuery;
1889         sqlQuery = Form("select count(*) from %s where shuttle_done=0", fConfig->GetShuttlelbTable());
1890
1891         TSQLResult* aResult = fServer[3]->Query(sqlQuery);
1892         if (!aResult) {
1893                 AliError(Form("Can't execute query <%s>!", sqlQuery.Data()));
1894                 return;
1895         }
1896
1897         AliDebug(2,Form("Query = %s", sqlQuery.Data()));
1898
1899         if (aResult->GetRowCount() == 0) {
1900                 AliError(Form("No result for query %s received", sqlQuery.Data()));
1901                 return;
1902         }
1903
1904         if (aResult->GetFieldCount() != 1) {
1905                 AliError(Form("Invalid field count for query %s received", sqlQuery.Data()));
1906                 return;
1907         }
1908
1909         TSQLRow* aRow = aResult->Next();
1910         if (!aRow) {
1911                 AliError(Form("Could not receive result of query %s", sqlQuery.Data()));
1912                 return;
1913         }
1914
1915         TString result(aRow->GetField(0), aRow->GetFieldLength(0));
1916         Int_t count = result.Atoi();
1917
1918         Log("SHUTTLE", Form("%d unprocessed runs", count));
1919
1920         delete aRow;
1921         delete aResult;
1922
1923         TMonaLisaValue mlStatus("SHUTTLE_openruns", count);
1924
1925         TList mlList;
1926         mlList.Add(&mlStatus);
1927
1928         fMonaLisa->SendParameters(&mlList, "__PROCESSINGINFO__");
1929 }
1930
1931 //______________________________________________________________________________________________
1932 Bool_t AliShuttle::QueryShuttleLogbook(const char* whereClause,
1933                 TObjArray& entries)
1934 {
1935         // Query DAQ's Shuttle logbook and fills detector status object.
1936         // Call QueryRunParameters to query DAQ logbook for run parameters.
1937         //
1938
1939         entries.SetOwner(1);
1940
1941         // check connection, in case connect
1942         if (!Connect(3)) return kFALSE;
1943
1944         TString sqlQuery;
1945         sqlQuery = Form("select * from %s %s order by run", fConfig->GetShuttlelbTable(), whereClause);
1946
1947         TSQLResult* aResult = fServer[3]->Query(sqlQuery);
1948         if (!aResult) {
1949                 AliError(Form("Can't execute query <%s>!", sqlQuery.Data()));
1950                 return kFALSE;
1951         }
1952
1953         AliDebug(2,Form("Query = %s", sqlQuery.Data()));
1954
1955         if(aResult->GetRowCount() == 0) {
1956                 Log("SHUTTLE", "No entries in Shuttle Logbook match request");
1957                 delete aResult;
1958                 return kTRUE;
1959         }
1960
1961         // TODO Check field count!
1962         const UInt_t nCols = 26;
1963         if (aResult->GetFieldCount() != (Int_t) nCols) {
1964                 Log("SHUTTLE", "Invalid SQL result field number!");
1965                 delete aResult;
1966                 return kFALSE;
1967         }
1968
1969         TSQLRow* aRow;
1970         while ((aRow = aResult->Next())) {
1971                 TString runString(aRow->GetField(0), aRow->GetFieldLength(0));
1972                 Int_t run = runString.Atoi();
1973
1974                 AliShuttleLogbookEntry *entry = QueryRunParameters(run);
1975                 if (!entry)
1976                         continue;
1977
1978                 // DA test mode flag
1979                 TString daTestModeString(aRow->GetField(2), aRow->GetFieldLength(2)); // field 2 = DA test mode flag
1980                 Bool_t daTestMode = (Bool_t)daTestModeString.Atoi();
1981                 entry->SetDATestMode(daTestMode);
1982
1983                 // loop on detectors
1984                 for(UInt_t ii = 0; ii < nCols; ii++)
1985                         entry->SetDetectorStatus(aResult->GetFieldName(ii), aRow->GetField(ii));
1986
1987                 entries.AddLast(entry);
1988                 delete aRow;
1989         }
1990
1991         delete aResult;
1992         return kTRUE;
1993 }
1994
1995 //______________________________________________________________________________________________
1996 AliShuttleLogbookEntry* AliShuttle::QueryRunParameters(Int_t run)
1997 {
1998         //
1999         // Retrieve run parameters written in the DAQ logbook and sets them into AliShuttleLogbookEntry object
2000         //
2001
2002         // check connection, in case connect
2003         if (!Connect(3))
2004                 return 0;
2005
2006         TString sqlQuery;
2007         sqlQuery.Form("select * from %s where run=%d", fConfig->GetDAQlbTable(), run);
2008
2009         TSQLResult* aResult = fServer[3]->Query(sqlQuery);
2010         if (!aResult) {
2011                 Log("SHUTTLE", Form("Can't execute query <%s>!", sqlQuery.Data()));
2012                 return 0;
2013         }
2014
2015         if (aResult->GetRowCount() == 0) {
2016                 Log("SHUTTLE", Form("QueryRunParameters - No entry in DAQ Logbook for run %d. Skipping", run));
2017                 delete aResult;
2018                 return 0;
2019         }
2020
2021         if (aResult->GetRowCount() > 1) {
2022                 Log("SHUTTLE", Form("QueryRunParameters - UNEXPECTED: "
2023                                 "more than one entry in DAQ Logbook for run %d!", run));
2024                 delete aResult;
2025                 return 0;
2026         }
2027
2028         TSQLRow* aRow = aResult->Next();
2029         if (!aRow)
2030         {
2031                 Log("SHUTTLE", Form("QueryRunParameters - Could not retrieve row for run %d. Skipping", run));
2032                 delete aResult;
2033                 return 0;
2034         }
2035
2036         AliShuttleLogbookEntry* entry = new AliShuttleLogbookEntry(run);
2037
2038         for (Int_t ii = 0; ii < aResult->GetFieldCount(); ii++)
2039                 entry->SetRunParameter(aResult->GetFieldName(ii), aRow->GetField(ii));
2040
2041         delete aRow;
2042         delete aResult;
2043
2044         UInt_t startTime = entry->GetStartTime();
2045         UInt_t endTime = entry->GetEndTime();
2046         Bool_t ecsSuccess = entry->GetECSSuccess();
2047         TString runType = entry->GetRunType();
2048         TString tmpdaqstartTime = entry->GetRunParameter("DAQ_time_start");
2049         UInt_t daqstartTime = tmpdaqstartTime.Atoi();
2050
2051         UInt_t now = time(0);
2052         Int_t dcsDelay = fConfig->GetDCSDelay()+fConfig->GetDCSQueryOffset();
2053
2054         Bool_t skip = kFALSE;
2055
2056         // runs are processed if
2057         //   a) runType is PHYSICS and ecsSuccess is set
2058         //   b) runType is not PHYSICS and (ecsSuccess is set or DAQ_time_start is non-0)
2059         // effectively this means that all runs are processed that started properly (ecsSucess behaviour is different for PHYSICS and non-PHYSICS runs (check with ECS!)
2060         if (startTime != 0 && endTime != 0) {
2061                 if (endTime > startTime) {
2062                         if (endTime >= now - dcsDelay) {
2063                                 Log("SHUTTLE", Form("Skipping run %d for now, because DCS buffer time is not yet expired", run));
2064                         } else {
2065                                 if (runType == "PHYSICS") {
2066                                         if (ecsSuccess) {
2067                                                 return entry;
2068                                         } else {
2069                                                 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")));
2070                                                 skip = kTRUE;
2071                                         }
2072                                 } else {
2073                                         if (ecsSuccess || daqstartTime > 0) {
2074                                                 if (ecsSuccess == kFALSE)
2075                                                         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")));
2076                                                 return entry;
2077                                         } else {
2078                                                 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")));
2079                                                 skip = kTRUE;
2080                                         }
2081                                 }
2082                         }
2083                 } else {
2084                         Log("SHUTTLE", Form("QueryRunParameters - Invalid parameters for run %d: startTime equal to endTime: %d %d - Skipping!", run, startTime, endTime));
2085                         skip = kTRUE;
2086                 }
2087         } else {
2088           Log("SHUTTLE", Form("QueryRunParameters - Invalid parameters for Run %d: "
2089                               "startTime = %d, endTime = %d. Skipping (Shuttle won't be marked as DONE)!",
2090                               run, startTime, endTime));
2091         }
2092
2093         if (skip)
2094         {
2095                 Log("SHUTTLE", Form("Marking SHUTTLE skipped for run %d", run));
2096                 fLogbookEntry = entry;
2097                 if (!UpdateShuttleLogbook("shuttle_skipped"))
2098                 {
2099                         AliError(Form("Could not update logbook for run %d !", run));
2100                 }
2101                 if (!UpdateTableSkippedCase("ALL"))
2102                 {
2103                         AliError(Form("Could not update FXS tables for run %d !", run));
2104                 }
2105                 fLogbookEntry = 0;
2106         }
2107
2108         delete entry;
2109         return 0;
2110 }
2111
2112 //______________________________________________________________________________________________
2113 TMap* AliShuttle::GetValueSet(const char* host, Int_t port, const TSeqCollection* entries,
2114                               DCSType type, Int_t multiSplit)
2115 {
2116         // Retrieve all "entry" data points from the DCS server
2117         // host, port: TSocket connection parameters
2118         // entries: list of name of the alias or data point
2119         // type: kAlias or kDP
2120         // returns TMap of values, 0 when failure
2121
2122         AliDCSClient client(host, port, fTimeout, fRetries, multiSplit);
2123
2124         TMap* result = 0;
2125         if (type == kAlias)
2126         {
2127                 //result = client.GetAliasValues(entries, GetCurrentStartTime()-offset,
2128                 //      GetCurrentEndTime()+offset);
2129                 result = client.GetAliasValues(entries, GetStartTimeDCSQuery(),
2130                         GetEndTimeDCSQuery());
2131         }
2132         else if (type == kDP)
2133         {
2134                 //result = client.GetDPValues(entries, GetCurrentStartTime()-offset,
2135                 //      GetCurrentEndTime()+offset);
2136                 result = client.GetDPValues(entries, GetStartTimeDCSQuery(),
2137                         GetEndTimeDCSQuery());
2138         }
2139
2140         if (result == 0)
2141         {
2142                 Log(fCurrentDetector.Data(), Form("GetValueSet - Can't get entries! Reason: %s",
2143                         client.GetErrorString(client.GetResultErrorCode())));
2144                 if (client.GetResultErrorCode() == AliDCSClient::fgkServerError)
2145                         Log(fCurrentDetector.Data(), Form("GetValueSet - Server error code: %s",
2146                                 client.GetServerError().Data()));
2147
2148                 return 0;
2149         }
2150
2151         return result;
2152 }
2153
2154 //______________________________________________________________________________________________
2155 const char* AliShuttle::GetFile(Int_t system, const char* detector,
2156                 const char* id, const char* source)
2157 {
2158         // Get calibration file from file exchange servers
2159         // First queris the FXS database for the file name, using the run, detector, id and source info
2160         // then calls RetrieveFile(filename) for actual copy to local disk
2161         // run: current run being processed (given by Logbook entry fLogbookEntry)
2162         // detector: the Preprocessor name
2163         // id: provided as a parameter by the Preprocessor
2164         // source: provided by the Preprocessor through GetFileSources function
2165
2166         // check if test mode should simulate a FXS error
2167         if (fTestMode & kErrorFXSFiles)
2168         {
2169                 Log(detector, Form("GetFile - In TESTMODE - Simulating error while connecting to %s FXS", GetSystemName(system)));
2170                 return 0;
2171         }
2172
2173         // check connection, in case connect
2174         if (!Connect(system))
2175         {
2176                 Log(detector, Form("GetFile - Couldn't connect to %s FXS database", GetSystemName(system)));
2177                 fFXSError = system;
2178                 return 0;
2179         }
2180
2181         // Query preparation
2182         TString sourceName(source);
2183         Int_t nFields = 3;
2184         TString sqlQueryStart = Form("select filePath,size,fileChecksum from %s where",
2185                                                                 fConfig->GetFXSdbTable(system));
2186         TString whereClause = Form("run=%d and detector=\"%s\" and fileId=\"%s\"",
2187                                                                 GetCurrentRun(), detector, id);
2188
2189         if (system == kDAQ)
2190         {
2191                 whereClause += Form(" and DAQsource=\"%s\"", source);
2192         }
2193         else if (system == kDCS)
2194         {
2195                 sourceName="none";
2196         }
2197         else if (system == kHLT)
2198         {
2199                 whereClause += Form(" and DDLnumbers=\"%s\"", source);
2200         }
2201
2202         TString sqlQuery = Form("%s %s", sqlQueryStart.Data(), whereClause.Data());
2203
2204         AliDebug(2, Form("SQL query: \n%s",sqlQuery.Data()));
2205
2206         // Query execution
2207         TSQLResult* aResult = 0;
2208         aResult = dynamic_cast<TSQLResult*> (fServer[system]->Query(sqlQuery));
2209         if (!aResult) {
2210                 Log(detector, Form("GetFile - Can't execute SQL query to %s database for: id = %s, source = %s",
2211                                 GetSystemName(system), id, sourceName.Data()));
2212                 fFXSError = system;
2213                 return 0;
2214         }
2215
2216         if (aResult->GetRowCount() == 0)
2217         {
2218                 Log(detector,
2219                         Form("GetFile - No entry in %s FXS db for: id = %s, source = %s",
2220                                 GetSystemName(system), id, sourceName.Data()));
2221                 delete aResult;
2222                 return 0;
2223         }
2224
2225         if (aResult->GetRowCount() > 1) {
2226                 Log(detector,
2227                         Form("GetFile - More than one entry in %s FXS db for: id = %s, source = %s",
2228                                 GetSystemName(system), id, sourceName.Data()));
2229                 fFXSError = system;
2230                 delete aResult;
2231                 return 0;
2232         }
2233
2234         if (aResult->GetFieldCount() != nFields) {
2235                 Log(detector,
2236                         Form("GetFileName - Wrong field count in %s FXS db for: id = %s, source = %s",
2237                                 GetSystemName(system), id, sourceName.Data()));
2238                 fFXSError = system;
2239                 delete aResult;
2240                 return 0;
2241         }
2242
2243         TSQLRow* aRow = dynamic_cast<TSQLRow*> (aResult->Next());
2244
2245         if (!aRow){
2246                 Log(detector, Form("GetFile - Empty set result in %s FXS db from query: id = %s, source = %s",
2247                                 GetSystemName(system), id, sourceName.Data()));
2248                 fFXSError = system;
2249                 delete aResult;
2250                 return 0;
2251         }
2252
2253         TString filePath(aRow->GetField(0), aRow->GetFieldLength(0));
2254         TString fileSize(aRow->GetField(1), aRow->GetFieldLength(1));
2255         TString fileChecksum(aRow->GetField(2), aRow->GetFieldLength(2));
2256
2257         delete aResult;
2258         delete aRow;
2259
2260         AliDebug(2, Form("filePath = %s; size = %s, fileChecksum = %s",
2261                                 filePath.Data(), fileSize.Data(), fileChecksum.Data()));
2262
2263         // retrieved file is renamed to make it unique
2264         Int_t dir_lev1 = GetCurrentRun()/10000;
2265         TString localFileName = Form("%s/%d/%d/%s_process/%s_%s_%d_%s_%s.shuttle",
2266                                      GetShuttleTempDir(), dir_lev1, GetCurrentRun(), detector,
2267                                         GetSystemName(system), detector, GetCurrentRun(),
2268                                         id, sourceName.Data());
2269         Log("SHUTTLE",Form("file from FXS = %s",localFileName.Data()));
2270
2271
2272         // file retrieval from FXS
2273         UInt_t nRetries = 0;
2274         UInt_t maxRetries = 3;
2275         Bool_t result = kFALSE;
2276
2277         // copy!! if successful TSystem::Exec returns 0
2278         while (nRetries++ < maxRetries) {
2279                 AliDebug(2, Form("Trying to copy file. Retry # %d", nRetries));
2280                 result = RetrieveFile(system, filePath.Data(), localFileName.Data());
2281                 if (!result)
2282                 {
2283                         Log(detector, Form("GetFile - Copy of file %s from %s FXS failed",
2284                                         filePath.Data(), GetSystemName(system)));
2285                         continue;
2286                 }
2287
2288                 if (fileSize.Length()>0)
2289                 {
2290                         // compare filesize of local file with the one stored in the FXS DB
2291                         Long_t size = -1;
2292                         Int_t sizeComp = gSystem->GetPathInfo(localFileName.Data(), 0, &size, 0, 0);
2293
2294                         if (sizeComp != 0 || size != fileSize.Atoi())
2295                         {
2296                                 Log(detector, Form("GetFile - size of file %s does not match with local copy!",
2297                                                         filePath.Data()));
2298                                 result = kFALSE;
2299                                 continue;
2300                         }
2301
2302                 } else {
2303                         Log(fCurrentDetector, Form("GetFile - size of file %s not set in %s database, skipping comparison",
2304                                                 filePath.Data(), GetSystemName(system)));
2305                 }
2306
2307                 if (fileChecksum.Length()>0)
2308                 {
2309                         // compare md5sum of local file with the one stored in the FXS DB
2310                         if(fileChecksum.Contains(' ')) fileChecksum.Resize(fileChecksum.First(' '));
2311                         Int_t md5Comp = gSystem->Exec(Form("md5sum %s |grep %s > /dev/null 2> /dev/null",
2312                                                 localFileName.Data(), fileChecksum.Data()));
2313
2314                         if (md5Comp != 0)
2315                         {
2316                                 Log(detector, Form("GetFile - md5sum of file %s does not match with local copy!",
2317                                                         filePath.Data()));
2318                                 result = kFALSE;
2319                                 continue;
2320                         }
2321                 } else {
2322                         Log(fCurrentDetector, Form("GetFile - md5sum of file %s not set in %s database, skipping comparison",
2323                                                         filePath.Data(), GetSystemName(system)));
2324                 }
2325                 if (result) break;
2326         }
2327
2328         if (!result)
2329         {
2330                 fFXSError = system;
2331                 return 0;
2332         }
2333
2334         fFXSCalled[system]=kTRUE;
2335         TObjString *fileParams = new TObjString(Form("%s#!?!#%s", id, sourceName.Data()));
2336         fFXSlist[system].Add(fileParams);
2337
2338         static TString staticLocalFileName;
2339         staticLocalFileName.Form("%s", localFileName.Data());
2340
2341         Log(fCurrentDetector, Form("GetFile - Retrieved file with id %s and "
2342                         "source %s from %s to %s", id, source,
2343                         GetSystemName(system), localFileName.Data()));
2344
2345         return staticLocalFileName.Data();
2346 }
2347
2348 //______________________________________________________________________________________________
2349 Bool_t AliShuttle::RetrieveFile(UInt_t system, const char* fxsFileName, const char* localFileName)
2350 {
2351         //
2352         // Copies file from FXS to local Shuttle machine
2353         //
2354
2355         // check temp directory: trying to cd to temp; if it does not exist, create it
2356         AliDebug(2, Form("Copy file %s from %s FXS into %s",
2357                         GetSystemName(system), fxsFileName, localFileName));
2358
2359         TString tmpDir(localFileName);
2360
2361         tmpDir = tmpDir(0,tmpDir.Last('/'));
2362
2363         Int_t noDir = gSystem->GetPathInfo(tmpDir.Data(), 0, (Long64_t*) 0, 0, 0);
2364         if (noDir) // temp dir does not exists!
2365         {
2366                 if (gSystem->mkdir(tmpDir.Data(), 1))
2367                 {
2368                         Log(fCurrentDetector.Data(), "RetrieveFile - could not make temp directory!!");
2369                         return kFALSE;
2370                 }
2371         }
2372
2373         TString command = Form("scp -oPort=%d -2 %s@%s:%s/%s %s",
2374                 fConfig->GetFXSPort(system),
2375                 fConfig->GetFXSUser(system),
2376                 fConfig->GetFXSHost(system),
2377                 fConfig->GetFXSBaseFolder(system),
2378                 fxsFileName,
2379                 localFileName);
2380
2381         AliDebug(2, Form("%s",command.Data()));
2382
2383         Bool_t result = (gSystem->Exec(command.Data()) == 0);
2384
2385         return result;
2386 }
2387
2388 //______________________________________________________________________________________________
2389 TList* AliShuttle::GetFileSources(Int_t system, const char* detector, const char* id)
2390 {
2391         //
2392         // Get sources producing the condition file Id from file exchange servers
2393         // if id is NULL all sources are returned (distinct)
2394         //
2395
2396         if (id)
2397         {
2398                 Log(detector, Form("GetFileSources - Querying %s FXS for files with id %s produced by %s", GetSystemName(system), id, detector));
2399         } else {
2400                 Log(detector, Form("GetFileSources - Querying %s FXS for files produced by %s", GetSystemName(system), detector));
2401         }
2402
2403         // check if test mode should simulate a FXS error
2404         if (fTestMode & kErrorFXSSources)
2405         {
2406                 Log(detector, Form("GetFileSources - In TESTMODE - Simulating error while connecting to %s FXS", GetSystemName(system)));
2407                 return 0;
2408         }
2409
2410         if (system == kDCS)
2411         {
2412                 Log(detector, "GetFileSources - WARNING: DCS system has only one source of data!");
2413                 TList *list = new TList();
2414                 list->SetOwner(1);
2415                 list->Add(new TObjString(" "));
2416                 return list;
2417         }
2418
2419         // check connection, in case connect
2420         if (!Connect(system))
2421         {
2422                 Log(detector, Form("GetFileSources - Couldn't connect to %s FXS database", GetSystemName(system)));
2423                 fFXSError = system;
2424                 return NULL;
2425         }
2426
2427         TString sourceName = "";
2428         if (system == kDAQ)
2429         {
2430                 sourceName = "DAQsource";
2431         } else if (system == kHLT)
2432         {
2433                 sourceName = "DDLnumbers";
2434         }
2435
2436         TString sqlQueryStart = Form("select distinct %s from %s where", sourceName.Data(), fConfig->GetFXSdbTable(system));
2437         TString whereClause = Form("run=%d and detector=\"%s\"",
2438                                 GetCurrentRun(), detector);
2439         if (id)
2440                 whereClause += Form(" and fileId=\"%s\"", id);
2441         TString sqlQuery = Form("%s %s", sqlQueryStart.Data(), whereClause.Data());
2442
2443         AliDebug(2, Form("SQL query: \n%s",sqlQuery.Data()));
2444
2445         // Query execution
2446         TSQLResult* aResult;
2447         aResult = fServer[system]->Query(sqlQuery);
2448         if (!aResult) {
2449                 Log(detector, Form("GetFileSources - Can't execute SQL query to %s database for id: %s",
2450                                 GetSystemName(system), id));
2451                 fFXSError = system;
2452                 return 0;
2453         }
2454
2455         TList *list = new TList();
2456         list->SetOwner(1);
2457
2458         if (aResult->GetRowCount() == 0)
2459         {
2460                 Log(detector,
2461                         Form("GetFileSources - No entry in %s FXS table for id: %s", GetSystemName(system), id));
2462                 delete aResult;
2463                 return list;
2464         }
2465
2466         Log(detector, Form("GetFileSources - Found %d sources", aResult->GetRowCount()));
2467
2468         TSQLRow* aRow;
2469         while ((aRow = aResult->Next()))
2470         {
2471
2472                 TString source(aRow->GetField(0), aRow->GetFieldLength(0));
2473                 AliDebug(2, Form("%s = %s", sourceName.Data(), source.Data()));
2474                 list->Add(new TObjString(source));
2475                 delete aRow;
2476         }
2477
2478         delete aResult;
2479
2480         return list;
2481 }
2482
2483 //______________________________________________________________________________________________
2484 TList* AliShuttle::GetFileIDs(Int_t system, const char* detector, const char* source)
2485 {
2486         //
2487         // Get all ids of condition files produced by a given source from file exchange servers
2488         //
2489
2490         Log(detector, Form("GetFileIDs - Retrieving ids with source %s with %s", source, GetSystemName(system)));
2491
2492         // check if test mode should simulate a FXS error
2493         if (fTestMode & kErrorFXSSources)
2494         {
2495                 Log(detector, Form("GetFileIDs - In TESTMODE - Simulating error while connecting to %s FXS", GetSystemName(system)));
2496                 return 0;
2497         }
2498
2499         // check connection, in case connect
2500         if (!Connect(system))
2501         {
2502                 Log(detector, Form("GetFileIDs - Couldn't connect to %s FXS database", GetSystemName(system)));
2503                 return NULL;
2504         }
2505
2506         TString sourceName = "";
2507         if (system == kDAQ)
2508         {
2509                 sourceName = "DAQsource";
2510         } else if (system == kHLT)
2511         {
2512                 sourceName = "DDLnumbers";
2513         }
2514
2515         TString sqlQueryStart = Form("select fileId from %s where", fConfig->GetFXSdbTable(system));
2516         TString whereClause = Form("run=%d and detector=\"%s\"",
2517                                 GetCurrentRun(), detector);
2518         if (sourceName.Length() > 0 && source)
2519                 whereClause += Form(" and %s=\"%s\"", sourceName.Data(), source);
2520         TString sqlQuery = Form("%s %s", sqlQueryStart.Data(), whereClause.Data());
2521
2522         AliDebug(2, Form("SQL query: \n%s",sqlQuery.Data()));
2523
2524         // Query execution
2525         TSQLResult* aResult;
2526         aResult = fServer[system]->Query(sqlQuery);
2527         if (!aResult) {
2528                 Log(detector, Form("GetFileIDs - Can't execute SQL query to %s database for source: %s",
2529                                 GetSystemName(system), source));
2530                 return 0;
2531         }
2532
2533         TList *list = new TList();
2534         list->SetOwner(1);
2535
2536         if (aResult->GetRowCount() == 0)
2537         {
2538                 Log(detector,
2539                         Form("GetFileIDs - No entry in %s FXS table for source: %s", GetSystemName(system), source));
2540                 delete aResult;
2541                 return list;
2542         }
2543
2544         Log(detector, Form("GetFileIDs - Found %d ids", aResult->GetRowCount()));
2545
2546         TSQLRow* aRow;
2547
2548         while ((aRow = aResult->Next()))
2549         {
2550
2551                 TString id(aRow->GetField(0), aRow->GetFieldLength(0));
2552                 AliDebug(2, Form("fileId = %s", id.Data()));
2553                 list->Add(new TObjString(id));
2554                 delete aRow;
2555         }
2556
2557         delete aResult;
2558
2559         return list;
2560 }
2561
2562 //______________________________________________________________________________________________
2563 Bool_t AliShuttle::Connect(Int_t system)
2564 {
2565         // Connect to MySQL Server of the system's FXS MySQL databases
2566         // DAQ Logbook, Shuttle Logbook and DAQ FXS db are on the same host
2567         //
2568
2569         // check connection: if already connected return
2570
2571         if(fServer[system] && fServer[system]->IsConnected()) {
2572                 // ping the server
2573                 if (fServer[system]->PingVerify()==kTRUE){ // connection is still alive
2574                         return kTRUE;
2575                 }
2576                 else{
2577                         AliWarning(Form("Connection got lost to FXS database for %s. Closing and reconnecting.",
2578                                         AliShuttleInterface::GetSystemName(system)));
2579                         fServer[system]->Close();
2580                         delete fServer[system];
2581                         fServer[system] = 0x0;
2582                 }
2583         }
2584
2585         TString dbHost, dbUser, dbPass, dbName;
2586
2587         if (system < 3) // FXS db servers
2588         {
2589                 dbHost = Form("mysql://%s:%d", fConfig->GetFXSdbHost(system), fConfig->GetFXSdbPort(system));
2590                 dbUser = fConfig->GetFXSdbUser(system);
2591                 dbPass = fConfig->GetFXSdbPass(system);
2592                 dbName =   fConfig->GetFXSdbName(system);
2593         } else { // Run & Shuttle logbook servers
2594         // TODO Will the Shuttle logbook server be the same as the Run logbook server ???
2595                 dbHost = Form("mysql://%s:%d", fConfig->GetDAQlbHost(), fConfig->GetDAQlbPort());
2596                 dbUser = fConfig->GetDAQlbUser();
2597                 dbPass = fConfig->GetDAQlbPass();
2598                 dbName =   fConfig->GetDAQlbDB();
2599         }
2600
2601         fServer[system] = TSQLServer::Connect(dbHost.Data(), dbUser.Data(), dbPass.Data());
2602         if (!fServer[system] || !fServer[system]->IsConnected()) {
2603                 if(system < 3)
2604                 {
2605                 AliError(Form("Can't establish connection to FXS database for %s",
2606                                         AliShuttleInterface::GetSystemName(system)));
2607                 } else {
2608                 AliError("Can't establish connection to Run logbook.");
2609                 }
2610                 if(fServer[system]) delete fServer[system];
2611                 return kFALSE;
2612         }
2613
2614         // Get tables
2615         TSQLResult* aResult=0;
2616         switch(system){
2617                 case kDAQ:
2618                         aResult = fServer[kDAQ]->GetTables(dbName.Data());
2619                         break;
2620                 case kDCS:
2621                         aResult = fServer[kDCS]->GetTables(dbName.Data());
2622                         break;
2623                 case kHLT:
2624                         aResult = fServer[kHLT]->GetTables(dbName.Data());
2625                         break;
2626                 default:
2627                         aResult = fServer[3]->GetTables(dbName.Data());
2628                         break;
2629         }
2630
2631         delete aResult;
2632         return kTRUE;
2633 }
2634
2635 //______________________________________________________________________________________________
2636 Bool_t AliShuttle::UpdateTable()
2637 {
2638         //
2639         // Update FXS table filling time_processed field in all rows corresponding to current run and detector
2640         //
2641
2642         Bool_t result = kTRUE;
2643
2644         for (UInt_t system=0; system<3; system++)
2645         {
2646                 if(!fFXSCalled[system]) continue;
2647
2648                 // check connection, in case connect
2649                 if (!Connect(system))
2650                 {
2651                         Log(fCurrentDetector, Form("UpdateTable - Couldn't connect to %s FXS database", GetSystemName(system)));
2652                         result = kFALSE;
2653                         continue;
2654                 }
2655
2656                 TTimeStamp now; // now
2657
2658                 // Loop on FXS list entries
2659                 TIter iter(&fFXSlist[system]);
2660                 TObjString *aFXSentry=0;
2661                 while ((aFXSentry = dynamic_cast<TObjString*> (iter.Next())))
2662                 {
2663                         TString aFXSentrystr = aFXSentry->String();
2664                         TObjArray *aFXSarray = aFXSentrystr.Tokenize("#!?!#");
2665                         if (!aFXSarray || aFXSarray->GetEntries() != 2 )
2666                         {
2667                                 Log(fCurrentDetector, Form("UpdateTable - error updating %s FXS entry. Check string: <%s>",
2668                                         GetSystemName(system), aFXSentrystr.Data()));
2669                                 if(aFXSarray) delete aFXSarray;
2670                                 result = kFALSE;
2671                                 continue;
2672                         }
2673                         const char* fileId = ((TObjString*) aFXSarray->At(0))->GetName();
2674                         const char* source = ((TObjString*) aFXSarray->At(1))->GetName();
2675
2676                         TString whereClause;
2677                         if (system == kDAQ)
2678                         {
2679                                 whereClause = Form("where run=%d and detector=\"%s\" and fileId=\"%s\" and DAQsource=\"%s\";",
2680                                                         GetCurrentRun(), fCurrentDetector.Data(), fileId, source);
2681                         }
2682                         else if (system == kDCS)
2683                         {
2684                                 whereClause = Form("where run=%d and detector=\"%s\" and fileId=\"%s\";",
2685                                                         GetCurrentRun(), fCurrentDetector.Data(), fileId);
2686                         }
2687                         else if (system == kHLT)
2688                         {
2689                                 whereClause = Form("where run=%d and detector=\"%s\" and fileId=\"%s\" and DDLnumbers=\"%s\";",
2690                                                         GetCurrentRun(), fCurrentDetector.Data(), fileId, source);
2691                         }
2692
2693                         delete aFXSarray;
2694
2695                         TString sqlQuery = Form("update %s set time_processed=%d %s", fConfig->GetFXSdbTable(system),
2696                                                                 now.GetSec(), whereClause.Data());
2697
2698                         AliDebug(2, Form("SQL query: \n%s",sqlQuery.Data()));
2699
2700                         // Query execution
2701                         TSQLResult* aResult;
2702                         aResult = dynamic_cast<TSQLResult*> (fServer[system]->Query(sqlQuery));
2703                         if (!aResult)
2704                         {
2705                                 Log(fCurrentDetector, Form("UpdateTable - %s db: can't execute SQL query <%s>",
2706                                                                 GetSystemName(system), sqlQuery.Data()));
2707                                 result = kFALSE;
2708                                 continue;
2709                         }
2710                         delete aResult;
2711                 }
2712         }
2713
2714         return result;
2715 }
2716
2717 //_______________________________________________________________________________
2718 Bool_t AliShuttle::UpdateTableSkippedCase(const char* detector)
2719 {
2720         //
2721         // Update FXS table filling time_processed field in all rows corresponding to current run and detector
2722         // if detector = "ALL" update all detectors
2723         //
2724
2725         Bool_t result = kTRUE;
2726
2727         TString detName(detector);
2728
2729         for (UInt_t system=0; system<3; system++)
2730         {
2731
2732                 // check connection, in case connect
2733                 if (!Connect(system))
2734                 {
2735                         Log(fCurrentDetector, Form("UpdateTableSkippedCase - Couldn't connect to %s FXS database", GetSystemName(system)));
2736                         result = kFALSE;
2737                         continue;
2738                 }
2739
2740                 TTimeStamp now; // now
2741
2742                 // Loop on FXS list entries
2743                 TIter iter(&fFXSlist[system]);
2744
2745                 TString whereClause;
2746                 if (detName == "ALL") whereClause = Form("where run=%d and time_processed IS NULL;",GetCurrentRun());
2747                 else whereClause = Form("where run=%d and detector=\"%s\" and time_processed IS NULL;",GetCurrentRun(), detector);
2748
2749                 //Log("SHUTTLE",Form(" whereClause = %s ",whereClause.Data()));
2750
2751                 TString sqlQuery = Form("update %s set time_processed=%d %s", fConfig->GetFXSdbTable(system),
2752                                         now.GetSec(), whereClause.Data());
2753
2754                 AliDebug(2, Form("SQL query: \n%s",sqlQuery.Data()));
2755
2756                 // Query execution
2757                 TSQLResult* aResult;
2758                 aResult = dynamic_cast<TSQLResult*> (fServer[system]->Query(sqlQuery));
2759                 if (!aResult)
2760                 {
2761                         Log("SHUTTLE", Form("UpdateTableSkippedCase - %s db: can't execute SQL query <%s>",
2762                                                         GetSystemName(system), sqlQuery.Data()));
2763                         result = kFALSE;
2764                         continue;
2765                 }
2766                 delete aResult;
2767
2768         }
2769
2770         return result;
2771 }
2772 //______________________________________________________________________________________________
2773 Bool_t AliShuttle::UpdateTableFailCase()
2774 {
2775         // Update FXS table filling time_processed field in all rows corresponding to current run and detector
2776         // this is called in case the preprocessor is declared failed for the current run, because
2777         // the fields are updated only in case of success
2778
2779         Bool_t result = kTRUE;
2780
2781         for (UInt_t system=0; system<3; system++)
2782         {
2783                 // check connection, in case connect
2784                 if (!Connect(system))
2785                 {
2786                         Log(fCurrentDetector, Form("UpdateTableFailCase - Couldn't connect to %s FXS database",
2787                                                         GetSystemName(system)));
2788                         result = kFALSE;
2789                         continue;
2790                 }
2791
2792                 TTimeStamp now; // now
2793
2794                 // Loop on FXS list entries
2795
2796                 TString whereClause = Form("where run=%d and detector=\"%s\";",
2797                                                 GetCurrentRun(), fCurrentDetector.Data());
2798
2799
2800                 TString sqlQuery = Form("update %s set time_processed=%d %s", fConfig->GetFXSdbTable(system),
2801                                                         now.GetSec(), whereClause.Data());
2802
2803                 AliDebug(2, Form("SQL query: \n%s",sqlQuery.Data()));
2804
2805                 // Query execution
2806                 TSQLResult* aResult;
2807                 aResult = dynamic_cast<TSQLResult*> (fServer[system]->Query(sqlQuery));
2808                 if (!aResult)
2809                 {
2810                         Log(fCurrentDetector, Form("UpdateTableFailCase - %s db: can't execute SQL query <%s>",
2811                                                         GetSystemName(system), sqlQuery.Data()));
2812                         result = kFALSE;
2813                         continue;
2814                 }
2815                 delete aResult;
2816         }
2817
2818         return result;
2819 }
2820
2821 //______________________________________________________________________________________________
2822 Bool_t AliShuttle::UpdateShuttleLogbook(const char* detector, const char* status)
2823 {
2824         //
2825         // Update Shuttle logbook filling detector or shuttle_done column
2826         // ex. of usage: UpdateShuttleLogbook("PHOS", "DONE") or UpdateShuttleLogbook("shuttle_done")
2827         //
2828
2829         // check connection, in case connect
2830         if(!Connect(3)){
2831                 Log("SHUTTLE", "UpdateShuttleLogbook - Couldn't connect to DAQ Logbook.");
2832                 return kFALSE;
2833         }
2834
2835         TString detName(detector);
2836         TString setClause;
2837         if (detName == "shuttle_done" || detName == "shuttle_skipped")
2838         {
2839                 setClause = "set shuttle_done=1";
2840
2841                 if (detName == "shuttle_done")
2842                 {
2843                         if (TouchFile() != kTRUE)
2844                         {
2845                                 SendMLRunInfo("Pending");
2846                                 return kFALSE;
2847                         }
2848
2849                         SendMLRunInfo("Done");
2850                 }
2851                 else
2852                         SendMLRunInfo("Skipped");
2853         }
2854         else {
2855                 TString statusStr(status);
2856                 if(statusStr.Contains("done", TString::kIgnoreCase) ||
2857                    statusStr.Contains("failed", TString::kIgnoreCase)){
2858                         setClause = Form("set %s=\"%s\"", detector, status);
2859                 } else {
2860                         Log("SHUTTLE",
2861                                 Form("UpdateShuttleLogbook - Invalid status <%s> for detector %s",
2862                                         status, detector));
2863                         return kFALSE;
2864                 }
2865         }
2866
2867         TString whereClause = Form("where run=%d", GetCurrentRun());
2868
2869         TString sqlQuery = Form("update %s %s %s",
2870                                         fConfig->GetShuttlelbTable(), setClause.Data(), whereClause.Data());
2871
2872         AliDebug(2, Form("SQL query: \n%s",sqlQuery.Data()));
2873
2874         // Query execution
2875         TSQLResult* aResult;
2876         aResult = dynamic_cast<TSQLResult*> (fServer[3]->Query(sqlQuery));
2877         if (!aResult) {
2878                 Log("SHUTTLE", Form("UpdateShuttleLogbook - Can't execute query <%s>", sqlQuery.Data()));
2879                 return kFALSE;
2880         }
2881         delete aResult;
2882
2883         return kTRUE;
2884 }
2885
2886 //______________________________________________________________________________________________
2887 Int_t AliShuttle::GetCurrentRun() const
2888 {
2889         //
2890         // Get current run from logbook entry
2891         //
2892
2893         return fLogbookEntry ? fLogbookEntry->GetRun() : -1;
2894 }
2895
2896 //______________________________________________________________________________________________
2897 UInt_t AliShuttle::GetCurrentStartTime() const
2898 {
2899         //
2900         // get current start time
2901         //
2902
2903         return fLogbookEntry ? fLogbookEntry->GetStartTime() : 0;
2904 }
2905
2906 //______________________________________________________________________________________________
2907 UInt_t AliShuttle::GetCurrentEndTime() const
2908 {
2909         //
2910         // get current end time from logbook entry
2911         //
2912
2913         return fLogbookEntry ? fLogbookEntry->GetEndTime() : 0;
2914 }
2915 //______________________________________________________________________________________________
2916 UInt_t AliShuttle::GetCurrentYear() const
2917 {
2918         //
2919         // Get current year from logbook entry
2920         //
2921
2922         if (!fLogbookEntry) return 0;
2923
2924         TTimeStamp startTime(GetCurrentStartTime());
2925         TString year =  Form("%d",startTime.GetDate());
2926         year = year(0,4);
2927
2928         return year.Atoi();
2929 }
2930
2931 //______________________________________________________________________________________________
2932 const char* AliShuttle::GetLHCPeriod() const
2933 {
2934         //
2935         // Get current LHC period from logbook entry
2936         //
2937
2938         if (!fLogbookEntry) return 0;
2939
2940         return fLogbookEntry->GetRunParameter("LHCperiod");
2941 }
2942
2943 //______________________________________________________________________________________________
2944 void AliShuttle::Log(const char* detector, const char* message, UInt_t level)
2945 {
2946         //
2947         // Fill log string with a message
2948         //
2949
2950         TString logRunDir = GetShuttleLogDir();
2951         if (GetCurrentRun() >=0) {
2952                 Int_t logDir_lev1 = GetCurrentRun()/10000;
2953                 logRunDir += Form("/%d/%d", logDir_lev1, GetCurrentRun());
2954         }
2955         void* dir = gSystem->OpenDirectory(logRunDir.Data());
2956         if (dir == NULL) {
2957                 if (gSystem->mkdir(logRunDir.Data(), kTRUE)) {
2958                         AliError(Form("Can't open directory <%s>", GetShuttleLogDir()));
2959                         return;
2960                 }
2961
2962         } else {
2963                 gSystem->FreeDirectory(dir);
2964         }
2965
2966         TString toLog = Form("%s UTC (%d): %s - ", TTimeStamp(time(0)).AsString("s"), getpid(), detector);
2967         if (GetCurrentRun() >= 0)
2968                 toLog += Form("run %d - ", GetCurrentRun());
2969         toLog += Form("%s", message);
2970
2971         AliLog::Message(level, toLog, MODULENAME(), ClassName(), FUNCTIONNAME(), __FILE__, __LINE__);
2972
2973         // if we redirect the log output already to the file, leave here
2974         if (fOutputRedirected && strcmp(detector, "SHUTTLE") != 0)
2975                 return;
2976
2977         TString fileName = GetLogFileName(detector);
2978
2979         gSystem->ExpandPathName(fileName);
2980
2981         ofstream logFile;
2982         logFile.open(fileName, ofstream::out | ofstream::app);
2983
2984         if (!logFile.is_open()) {
2985                 AliError(Form("Could not open file %s", fileName.Data()));
2986                 return;
2987         }
2988
2989         logFile << toLog.Data() << "\n";
2990
2991         logFile.close();
2992 }
2993
2994 //______________________________________________________________________________________________
2995 TString AliShuttle::GetLogFileName(const char* detector) const
2996 {
2997         //
2998         // returns the name of the log file for a given sub detector
2999         //
3000
3001         TString fileName;
3002
3003         if (GetCurrentRun() >= 0)
3004         {
3005                 Int_t logDir_lev1 = GetCurrentRun()/10000;
3006                 fileName.Form("%s/%d/%d/%s.log", GetShuttleLogDir(), logDir_lev1, GetCurrentRun(),
3007                         detector);
3008         } else {
3009                 fileName.Form("%s/%s.log", GetShuttleLogDir(), detector);
3010         }
3011
3012         return fileName;
3013 }
3014
3015 //______________________________________________________________________________________________
3016 void AliShuttle::SendAlive()
3017 {
3018         // sends alive message to ML
3019
3020         TMonaLisaText mlStatus("SHUTTLE_status", "Alive");
3021
3022         TList mlList;
3023         mlList.Add(&mlStatus);
3024
3025         fMonaLisa->SendParameters(&mlList, "__PROCESSINGINFO__");
3026 }
3027
3028 //______________________________________________________________________________________________
3029 Bool_t AliShuttle::Collect(Int_t run)
3030 {
3031         //
3032         // Collects conditions data for all UNPROCESSED run written to DAQ LogBook in case of run = -1 (default)
3033         // If a dedicated run is given this run is processed
3034         //
3035         // In operational mode, this is the Shuttle function triggered by the EOR signal.
3036         //
3037
3038         if (run == -1)
3039                 Log("SHUTTLE","Collect - Shuttle called. Collecting conditions data for unprocessed runs");
3040         else
3041                 Log("SHUTTLE", Form("Collect - Shuttle called. Collecting conditions data for run %d", run));
3042
3043         SetLastAction("Starting");
3044
3045         // create ML instance
3046         if (!fMonaLisa)
3047                 fMonaLisa = new TMonaLisaWriter(fConfig->GetMonitorHost(), fConfig->GetMonitorTable());
3048
3049         CountOpenRuns();
3050
3051         TString whereClause("where shuttle_done=0");
3052         if (run != -1)
3053                 whereClause += Form(" and run=%d", run);
3054
3055         TObjArray shuttleLogbookEntries;
3056         if (!QueryShuttleLogbook(whereClause, shuttleLogbookEntries))
3057         {
3058                 Log("SHUTTLE", "Collect - Can't retrieve entries from Shuttle logbook");
3059                 return kFALSE;
3060         }
3061
3062         if (shuttleLogbookEntries.GetEntries() == 0)
3063         {
3064                 if (run == -1)
3065                         Log("SHUTTLE","Collect - Found no UNPROCESSED runs in Shuttle logbook");
3066                 else
3067                         Log("SHUTTLE", Form("Collect - Run %d is already DONE "
3068                                                 "or it does not exist in Shuttle logbook", run));
3069                 return kTRUE;
3070         }
3071
3072         for (UInt_t iDet=0; iDet<NDetectors(); iDet++)
3073                 fFirstUnprocessed[iDet] = kTRUE;
3074
3075         if (run != -1)
3076         {
3077                 // query Shuttle logbook for earlier runs, check if some detectors are unprocessed,
3078                 // flag them into fFirstUnprocessed array
3079                 TString whereClauseBis(Form("where shuttle_done=0 and run < %d", run));
3080                 TObjArray tmpLogbookEntries;
3081                 if (!QueryShuttleLogbook(whereClauseBis, tmpLogbookEntries))
3082                 {
3083                         Log("SHUTTLE", "Collect - Can't retrieve entries from Shuttle logbook");
3084                         return kFALSE;
3085                 }
3086
3087                 TIter iter(&tmpLogbookEntries);
3088                 AliShuttleLogbookEntry* anEntry = 0;
3089                 while ((anEntry = dynamic_cast<AliShuttleLogbookEntry*> (iter.Next())))
3090                 {
3091                         for (UInt_t iDet=0; iDet<NDetectors(); iDet++)
3092                         {
3093                                 if (anEntry->GetDetectorStatus(iDet) == AliShuttleLogbookEntry::kUnprocessed)
3094                                 {
3095                                         AliDebug(2, Form("Run %d: setting %s as \"not first time unprocessed\"",
3096                                                         anEntry->GetRun(), GetDetName(iDet)));
3097                                         fFirstUnprocessed[iDet] = kFALSE;
3098                                 }
3099                         }
3100
3101                 }
3102
3103         }
3104
3105         if (!RetrieveConditionsData(shuttleLogbookEntries))
3106         {
3107                 Log("SHUTTLE", "Collect - Process of at least one run failed");
3108                 CountOpenRuns();
3109                 return kFALSE;
3110         }
3111
3112         Log("SHUTTLE", "Collect - Requested run(s) successfully processed");
3113         CountOpenRuns();
3114         return kTRUE;
3115 }
3116
3117 //______________________________________________________________________________________________
3118 Bool_t AliShuttle::RetrieveConditionsData(const TObjArray& dateEntries)
3119 {
3120         //
3121         // Retrieve conditions data for all runs that aren't processed yet
3122         //
3123
3124         Bool_t hasError = kFALSE;
3125
3126         TIter iter(&dateEntries);
3127         AliShuttleLogbookEntry* anEntry;
3128
3129         while ((anEntry = (AliShuttleLogbookEntry*) iter.Next())){
3130                 if (!Process(anEntry)){
3131                         hasError = kTRUE;
3132                 }
3133
3134                 // clean SHUTTLE temp directory
3135                 //TString filename = Form("%s/*.shuttle", GetShuttleTempDir());
3136                 //RemoveFile(filename.Data());
3137         }
3138
3139         return hasError == kFALSE;
3140 }
3141
3142 //______________________________________________________________________________________________
3143 ULong_t AliShuttle::GetTimeOfLastAction() const
3144 {
3145         //
3146         // Gets time of last action
3147         //
3148
3149         ULong_t tmp;
3150
3151         fMonitoringMutex->Lock();
3152
3153         tmp = fLastActionTime;
3154
3155         fMonitoringMutex->UnLock();
3156
3157         return tmp;
3158 }
3159
3160 //______________________________________________________________________________________________
3161 const TString AliShuttle::GetLastAction() const
3162 {
3163         //
3164         // returns a string description of the last action
3165         //
3166
3167         TString tmp;
3168
3169         fMonitoringMutex->Lock();
3170
3171         tmp = fLastAction;
3172
3173         fMonitoringMutex->UnLock();
3174
3175         return tmp;
3176 }
3177
3178 //______________________________________________________________________________________________
3179 void AliShuttle::SetLastAction(const char* action)
3180 {
3181         //
3182         // updates the monitoring variables
3183         //
3184
3185         fMonitoringMutex->Lock();
3186
3187         fLastAction = action;
3188         fLastActionTime = time(0);
3189
3190         fMonitoringMutex->UnLock();
3191 }
3192
3193 //______________________________________________________________________________________________
3194 const char* AliShuttle::GetRunParameter(const char* param)
3195 {
3196         //
3197         // returns run parameter read from DAQ logbook
3198         //
3199
3200         if(!fLogbookEntry) {
3201                 AliError("No logbook entry!");
3202                 return 0;
3203         }
3204
3205         return fLogbookEntry->GetRunParameter(param);
3206 }
3207
3208 //______________________________________________________________________________________________
3209 AliCDBEntry* AliShuttle::GetFromOCDB(const char* detector, const AliCDBPath& path)
3210 {
3211         //
3212         // returns object from OCDB valid for current run
3213         //
3214
3215         if (fTestMode & kErrorOCDB)
3216         {
3217                 Log(detector, "GetFromOCDB - In TESTMODE - Simulating error with OCDB");
3218                 return 0;
3219         }
3220
3221         AliCDBStorage *sto = AliCDBManager::Instance()->GetStorage(fgkMainCDB);
3222         if (!sto)
3223         {
3224                 Log(detector, "GetFromOCDB - Cannot activate main OCDB for query!");
3225                 return 0;
3226         }
3227
3228         return dynamic_cast<AliCDBEntry*> (sto->Get(path, GetCurrentRun()));
3229 }
3230
3231 //______________________________________________________________________________________________
3232 Bool_t AliShuttle::SendMail(EMailTarget target, Int_t system)
3233 {
3234         //
3235         // sends a mail to the subdetector expert in case of preprocessor error
3236         //
3237
3238         if (fTestMode != kNone)
3239                 return kTRUE;
3240
3241         if (!fConfig->SendMail())
3242                 return kTRUE;
3243
3244         if (target == kDCSEMail || target == kFXSEMail) {
3245                 if (!fFirstProcessing)
3246                         return kTRUE;
3247         }
3248
3249         Int_t runMode = (Int_t)fConfig->GetRunMode();
3250         TString tmpStr;
3251         if (runMode == 0) tmpStr = " Nightly Test:";
3252         else tmpStr = " Data Taking:";
3253         void* dir = gSystem->OpenDirectory(GetShuttleLogDir());
3254         if (dir == NULL)
3255         {
3256                 if (gSystem->mkdir(GetShuttleLogDir(), kTRUE))
3257                 {
3258                         Log("SHUTTLE", Form("SendMail - Can't open directory <%s>", GetShuttleLogDir()));
3259                         return kFALSE;
3260                 }
3261
3262         } else {
3263                 gSystem->FreeDirectory(dir);
3264         }
3265
3266         // det experts in to
3267         TString to="";
3268         TIter *iterExperts = 0;
3269         if (target == kDCSEMail) {
3270                 iterExperts = new TIter(fConfig->GetAdmins(AliShuttleConfig::kAmanda));
3271         }
3272         else if (target == kFXSEMail) {
3273                 iterExperts = new TIter(fConfig->GetAdmins(system));
3274         }
3275         if (iterExperts) {
3276                 TObjString *anExpert=0;
3277                 while ((anExpert = (TObjString*) iterExperts->Next()))
3278                 {
3279                         to += Form("%s,", anExpert->GetName());
3280                 }
3281                 delete iterExperts;
3282         }
3283
3284         // add subdetector experts
3285         iterExperts = new TIter(fConfig->GetResponsibles(fCurrentDetector));
3286         TObjString *anExpert=0;
3287         while ((anExpert = (TObjString*) iterExperts->Next()))
3288         {
3289                 to += Form("%s,", anExpert->GetName());
3290         }
3291         delete iterExperts;
3292
3293         if (to.Length() > 0)
3294           to.Remove(to.Length()-1);
3295         AliDebug(2, Form("to: %s",to.Data()));
3296
3297         if (to.IsNull()) {
3298                 Log("SHUTTLE", Form("List of %d responsibles not set!", (Int_t) target));
3299                 return kFALSE;
3300         }
3301
3302         // SHUTTLE responsibles in cc
3303         TString cc="";
3304         TIter iterAdmins(fConfig->GetAdmins(AliShuttleConfig::kGlobal));
3305         TObjString *anAdmin=0;
3306         while ((anAdmin = (TObjString*) iterAdmins.Next()))
3307         {
3308                 cc += Form("%s,", anAdmin->GetName());
3309         }
3310         if (cc.Length() > 0)
3311           cc.Remove(cc.Length()-1);
3312         AliDebug(2, Form("cc: %s",to.Data()));
3313
3314         // mail body
3315         TString bodyFileName;
3316         bodyFileName.Form("%s/mail.body", GetShuttleLogDir());
3317         gSystem->ExpandPathName(bodyFileName);
3318
3319         ofstream mailBody;
3320         mailBody.open(bodyFileName, ofstream::out);
3321
3322         if (!mailBody.is_open())
3323         {
3324                 Log("SHUTTLE", Form("Could not open mail body file %s", bodyFileName.Data()));
3325                 return kFALSE;
3326         }
3327
3328
3329         TString subject;
3330         TString body;
3331
3332         if (target == kDCSEMail){
3333                 subject = Form("%s CRITICAL Retrieval of data points for %s FAILED in run %d !",
3334                                 tmpStr.Data(), fCurrentDetector.Data(), GetCurrentRun());
3335                 AliDebug(2, Form("subject: %s", subject.Data()));
3336
3337                 body = Form("Dear DCS experts, \n\n");
3338                 body += Form("SHUTTLE couldn\'t retrieve the data points for detector %s "
3339                              "in run %d!!\n\n", fCurrentDetector.Data(), GetCurrentRun());
3340         }
3341         else if (target == kFXSEMail){
3342                 subject = Form("%s CRITICAL FXS communication for %s FAILED in run %d !",
3343                                 tmpStr.Data(), fCurrentDetector.Data(), GetCurrentRun());
3344                 AliDebug(2, Form("subject: %s", subject.Data()));
3345                 TString sys;
3346                 if (system == kDAQ) sys="DAQ";
3347                 else if (system == kDCS) sys="DCS";
3348                 else if (system == kHLT) sys="HLT";
3349                 else return kFALSE;
3350                 body = Form("Dear  %s FXS experts, \n\n",sys.Data());
3351                 body += Form("SHUTTLE couldn\'t retrieve data from the FXS for detector %s "
3352                              "in run %d!!\n\n", fCurrentDetector.Data(), GetCurrentRun());
3353                 body += Form("The contacted server was:\nDB: %s\nFXS:%s\n\n", fConfig->GetFXSdbHost(system), fConfig->GetFXSHost(system));
3354         }
3355         else {
3356                 subject = Form("%s %s Shuttle preprocessor FAILED in run %d (run type = %s)!",
3357                                        tmpStr.Data(), fCurrentDetector.Data(), GetCurrentRun(), GetRunType());
3358                 AliDebug(2, Form("subject: %s", subject.Data()));
3359
3360                 body = Form("Dear %s expert(s), \n\n", fCurrentDetector.Data());
3361                 body += Form("SHUTTLE just detected that your preprocessor "
3362                              "failed processing run %d (run type = %s)!!\n\n",
3363                              GetCurrentRun(), GetRunType());
3364         }
3365
3366         body += Form("Please check %s status on the SHUTTLE monitoring page: \n\n",
3367                                 fCurrentDetector.Data());
3368         if (fConfig->GetRunMode() == AliShuttleConfig::kTest)
3369         {
3370                 body += Form("\thttp://pcalimonitor.cern.ch/shuttle.jsp?time=24 \n\n");
3371         } else {
3372                 body += Form("\thttp://pcalimonitor.cern.ch/shuttle.jsp?instance=PROD&time=24 \n\n");
3373         }
3374
3375
3376         TString logFolder = "logs";
3377         if (fConfig->GetRunMode() == AliShuttleConfig::kProd)
3378                 logFolder += "_PROD";
3379
3380
3381         body += Form("Find the %s log for the current run on \n\n"
3382                 "\thttp://pcalishuttle02.cern.ch/%s/%d/%d/%s.log \n\n",
3383                      fCurrentDetector.Data(), logFolder.Data(), GetCurrentRun()/10000,
3384                                 GetCurrentRun(), fCurrentDetector.Data());
3385         body += Form("The last 15 lines of %s log file are following:\n\n", fCurrentDetector.Data());
3386
3387         AliDebug(2, Form("Body begin: %s", body.Data()));
3388
3389         mailBody << body.Data();
3390         mailBody.close();
3391         mailBody.open(bodyFileName, ofstream::out | ofstream::app);
3392
3393         TString logFileName = Form("%s/%d/%d/%s.log", GetShuttleLogDir(),
3394                 GetCurrentRun()/10000, GetCurrentRun(), fCurrentDetector.Data());
3395         TString tailCommand = Form("tail -n 15 %s >> %s", logFileName.Data(), bodyFileName.Data());
3396         if (gSystem->Exec(tailCommand.Data()))
3397         {
3398                 mailBody << Form("%s log file not found ...\n\n", fCurrentDetector.Data());
3399         }
3400
3401         TString endBody = Form("------------------------------------------------------\n\n");
3402         endBody += Form("In case of problems please contact the SHUTTLE core team.\n\n");
3403         endBody += "Please do not answer this message directly, it is automatically generated.\n\n";
3404         endBody += "Greetings,\n\n \t\t\tthe SHUTTLE\n";
3405
3406         AliDebug(2, Form("Body end: %s", endBody.Data()));
3407
3408         mailBody << endBody.Data();
3409
3410         mailBody.close();
3411
3412         // send mail!
3413         TString mailCommand = Form("mail -s \"%s\" -c %s %s < %s",
3414                                                 subject.Data(),
3415                                                 cc.Data(),
3416                                                 to.Data(),
3417                                                 bodyFileName.Data());
3418         AliDebug(2, Form("mail command: %s", mailCommand.Data()));
3419
3420         Bool_t result = gSystem->Exec(mailCommand.Data());
3421
3422         return result == 0;
3423 }
3424 //______________________________________________________________________________________________
3425 const char* AliShuttle::GetRunType()
3426 {
3427         //
3428         // returns run type read from "run type" logbook
3429         //
3430
3431         if(!fLogbookEntry) {
3432                 AliError("No logbook entry!");
3433                 return 0;
3434         }
3435
3436         return fLogbookEntry->GetRunType();
3437 }
3438
3439 //______________________________________________________________________________________________
3440 Bool_t AliShuttle::GetHLTStatus()
3441 {
3442         // Return HLT status (ON=1 OFF=0)
3443         // Converts the HLT status from the mode string read in the run logbook (not just a bool)
3444
3445         if(!fLogbookEntry) {
3446                 AliError("No logbook entry!");
3447                 return 0;
3448         }
3449
3450         // TODO implement when HLTMode is inserted in run logbook
3451         TString hltMode = fLogbookEntry->GetRunParameter("HLTmode");
3452         TSubString firstChar = hltMode(0,1);
3453         AliDebug(2,Form("First char = %s ",firstChar.Data()));
3454         if (firstChar == "A") {
3455                 return kFALSE;
3456         }
3457         else if ((firstChar == "B") || (firstChar == "C") || (firstChar == "D") || (firstChar == "E")) {
3458                 return kTRUE;
3459         }
3460         else {
3461                 Log("SHUTTLE","Unexpected HLT mode! Returning 0....");
3462                 return kFALSE;
3463         }
3464 }
3465
3466 //______________________________________________________________________________________________
3467 const char* AliShuttle::GetTriggerConfiguration()
3468 {
3469         // Receives the trigger configuration from the DAQ logbook for the current run
3470
3471         // check connection, if needed reconnect
3472         if (!Connect(3))
3473                 return 0;
3474
3475         TString sqlQuery;
3476         sqlQuery.Form("SELECT configFile FROM logbook_trigger_config WHERE run = %d", GetCurrentRun());
3477         TSQLResult* result = fServer[3]->Query(sqlQuery);
3478         if (!result)
3479         {
3480                 Log("SHUTTLE", Form("ERROR: Can't execute query <%s>!", sqlQuery.Data()));
3481                 return 0;
3482         }
3483
3484         if (result->GetRowCount() == 0)
3485         {
3486                 Log("SHUTTLE", "WARNING: Trigger configuration not found in logbook_trigger_config");
3487                 delete result;
3488                 return 0;
3489         }
3490
3491         TSQLRow* row = result->Next();
3492         if (!row)
3493         {
3494                 Log("SHUTTLE", "ERROR: Could not receive logbook_trigger_config data");
3495                 delete result;
3496                 return 0;
3497         }
3498
3499         // static, so that pointer remains valid when it is returned to the calling class
3500         static TString triggerConfig(row->GetField(0));
3501
3502         delete row;
3503         row = 0;
3504
3505         delete result;
3506         result = 0;
3507
3508         Log("SHUTTLE", Form("Found trigger configuration: %s", triggerConfig.Data()));
3509
3510         return triggerConfig;
3511 }
3512
3513 //______________________________________________________________________________________________
3514 const char* AliShuttle::GetCTPTimeParams()
3515 {
3516         // Receives the CTP time parameters from the DAQ logbook for the current run
3517
3518         // check connection, if needed reconnect
3519         if (!Connect(3))
3520                 return 0;
3521
3522         TString sqlQuery;
3523         sqlQuery.Form("SELECT alignmentFile FROM logbook_trigger_config WHERE run = %d", GetCurrentRun());
3524         TSQLResult* result = fServer[3]->Query(sqlQuery);
3525         if (!result)
3526         {
3527                 Log("SHUTTLE", Form("ERROR: Can't execute query <%s>!", sqlQuery.Data()));
3528                 return 0;
3529         }
3530
3531         if (result->GetRowCount() == 0)
3532         {
3533                 Log("SHUTTLE", "WARNING: CTP time params not found in logbook_trigger_config");
3534                 delete result;
3535                 return 0;
3536         }
3537
3538         TSQLRow* row = result->Next();
3539         if (!row)
3540         {
3541                 Log("SHUTTLE", "ERROR: Could not receive logbook_trigger_config data");
3542                 delete result;
3543                 return 0;
3544         }
3545
3546         // static, so that pointer remains valid when it is returned to the calling class
3547         static TString triggerTimeParams(row->GetField(0));
3548
3549         delete row;
3550         row = 0;
3551
3552         delete result;
3553         result = 0;
3554
3555         Log("SHUTTLE", Form("Found trigger time parameters: %s", triggerTimeParams.Data()));
3556
3557         return triggerTimeParams;
3558 }
3559
3560 //______________________________________________________________________________________________
3561 const char* AliShuttle::GetTriggerDetectorMask()
3562 {
3563         // Receives the trigger detector mask from DAQ logbook
3564
3565         // check connection, if needed reconnect
3566         if (!Connect(3))
3567                 return 0;
3568
3569         TString sqlQuery;
3570         sqlQuery.Form("SELECT BIN(BIT_OR(inputDetectorMask)) from logbook_trigger_clusters WHERE run = %d;", GetCurrentRun());
3571         TSQLResult* result = fServer[3]->Query(sqlQuery);
3572         if (!result)
3573         {
3574                 Log("SHUTTLE", Form("ERROR: Can't execute query <%s>!", sqlQuery.Data()));
3575                 return 0;
3576         }
3577
3578         if (result->GetRowCount() == 0)
3579         {
3580                 Log("SHUTTLE", "ERROR: Trigger Detector Mask not found in logbook_trigger_clusters");
3581                 delete result;
3582                 return 0;
3583         }
3584
3585         TSQLRow* row = result->Next();
3586         if (!row)
3587         {
3588                 Log("SHUTTLE", "ERROR: Could not receive logbook_trigger_clusters data");
3589                 delete result;
3590                 return 0;
3591         }
3592
3593         // static, so that pointer remains valid when it is returned to the calling class
3594         static TString triggerDetectorMask(row->GetField(0));
3595
3596         delete row;
3597         row = 0;
3598
3599         delete result;
3600         result = 0;
3601
3602         Log("SHUTTLE", Form("Found Trigger Detector Mask: %s", triggerDetectorMask.Data()));
3603
3604         return triggerDetectorMask;
3605 }
3606
3607 //______________________________________________________________________________________________
3608 void AliShuttle::SetShuttleTempDir(const char* tmpDir)
3609 {
3610         //
3611         // sets Shuttle temp directory
3612         //
3613
3614         fgkShuttleTempDir = gSystem->ExpandPathName(tmpDir);
3615 }
3616
3617 //______________________________________________________________________________________________
3618 void AliShuttle::SetShuttleLogDir(const char* logDir)
3619 {
3620         //
3621         // sets Shuttle log directory
3622         //
3623
3624         fgkShuttleLogDir = gSystem->ExpandPathName(logDir);
3625 }
3626 //______________________________________________________________________________________________
3627 Bool_t AliShuttle::TouchFile()
3628 {
3629         //
3630         // touching a file on the grid if run has been DONE
3631         //
3632
3633         if (!gGrid)
3634         {
3635                 Log("SHUTTLE",Form("No TGrid connection estabilished!"));
3636                 Log("SHUTTLE",Form("Could not touch file for run %i",GetCurrentRun()));
3637                 return kFALSE;
3638         }
3639
3640         TString dir;
3641         dir.Form("%s%d/SHUTTLE_DONE", fConfig->GetAlienPath(), GetCurrentYear());
3642         // checking whether directory for touch command exists
3643         TString commandLs;
3644         commandLs.Form("ls %s",dir.Data());
3645         TGridResult *resultLs = dynamic_cast<TGridResult*>(gGrid->Command(commandLs));
3646         if (!resultLs){
3647                 Log("SHUTTLE",Form("No result for %s command, returning without touching",commandLs.Data()));
3648                 return kFALSE;
3649         }
3650         TMap *mapLs = dynamic_cast<TMap*>(resultLs->At(0));
3651         if (!mapLs){
3652                 Log("SHUTTLE",Form("No map for %s command, returning without touching",commandLs.Data()));
3653                 delete resultLs;
3654                 resultLs = 0x0;
3655                 return kFALSE;
3656         }
3657         TObjString *valueLsPath = dynamic_cast<TObjString*>(mapLs->GetValue("path"));
3658         if (!valueLsPath || (valueLsPath->GetString()).CompareTo(dir)!=1){
3659                 Log("SHUTTLE",Form("No directory %s found, creating it",dir.Data()));
3660
3661                 // creating the directory
3662
3663                 Bool_t boolMkdir = gGrid->Mkdir(dir.Data());
3664                 if (!boolMkdir) {
3665                         Log("SHUTTLE",Form("Impossible to create dir %s in alien catalogue for run %i!",dir.Data(),GetCurrentRun()));
3666                         delete resultLs;
3667                         resultLs = 0x0;
3668                         return kFALSE;
3669                 }
3670                 Log("SHUTTLE",Form("Directory %s successfully created in alien catalogue for run %i",dir.Data(),GetCurrentRun()));
3671         }
3672         else {
3673                 Log("SHUTTLE",Form("Directory %s correctly found for run %i",dir.Data(),GetCurrentRun()));
3674         }
3675
3676         delete resultLs;
3677         resultLs = 0x0;
3678
3679         TString command;
3680         command.Form("touch %s/%i", dir.Data(), GetCurrentRun());
3681         Log("SHUTTLE", Form("Creating entry in file catalog: %s", command.Data()));
3682         TGridResult *resultTouch = dynamic_cast<TGridResult*>(gGrid->Command(command));
3683         if (!resultTouch){
3684                 Log("SHUTTLE",Form("No result for touching command, returning without touching for run %i",GetCurrentRun()));
3685                 return kFALSE;
3686         }
3687         TMap *mapTouch = dynamic_cast<TMap*>(resultTouch->At(0));
3688         if (!mapTouch){
3689                 Log("SHUTTLE",Form("No map for touching command, returning without touching for run %i",GetCurrentRun()));
3690                 delete resultTouch;
3691                 resultTouch = 0x0;
3692                 return kFALSE;
3693         }
3694         TObjString *valueTouch = dynamic_cast<TObjString*>(mapTouch->GetValue("__result__"));
3695         if (!valueTouch){
3696                 Log("SHUTTLE",Form("No value for \"__result__\" key set in the map for touching command, returning without touching for run %i",GetCurrentRun()));
3697                 delete resultTouch;
3698                 resultTouch = 0x0;
3699                 return kFALSE;
3700         }
3701         if (valueTouch->GetString()!="1"){
3702                 Log("SHUTTLE",Form("Failing the touching command, returning without touching for run %i",GetCurrentRun()));
3703                 delete resultTouch;
3704                 resultTouch = 0x0;
3705                 return kFALSE;
3706         }
3707         delete resultTouch;
3708         resultTouch = 0x0;
3709         Log("SHUTTLE", "Sucessfully touched the file");
3710         return kTRUE;
3711 }
3712 //______________________________________________________________________________________________
3713 UInt_t AliShuttle::GetStartTimeDCSQuery()
3714 {
3715         // Return Start Time for the DCS query
3716         //
3717         // The call is delegated to AliShuttleInterface
3718
3719         return GetCurrentStartTime()-fConfig->GetDCSQueryOffset();
3720 }
3721 //______________________________________________________________________________________________
3722 UInt_t AliShuttle::GetEndTimeDCSQuery()
3723 {
3724         // Return End Time for the DCS query
3725         //
3726         // The call is delegated to AliShuttleInterface
3727
3728         return GetCurrentEndTime()+fConfig->GetDCSQueryOffset();
3729 }
3730