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
  61 #include <TMonaLisaWriter.h>
  62
  63 #include <fstream>
  64
  65 #include <sys/types.h>
  66 #include <sys/wait.h>
  67
  68 #include <signal.h>
  69
  70 ClassImp(AliShuttle)
  71
  72 //______________________________________________________________________________________________
  73 AliShuttle::AliShuttle(const AliShuttleConfig* config,
  74                 UInt_t timeout, Int_t retries):
  75 fConfig(config),
  76 fTimeout(timeout), fRetries(retries),
  77 fPreprocessorMap(),
  78 fLogbookEntry(0),
  79 fCurrentDetector(),
  80 fStatusEntry(0),
  81 fMonitoringMutex(0),
  82 fLastActionTime(0),
  83 fLastAction(),
  84 fMonaLisa(0),
  85 fTestMode(kNone),
  86 fReadTestMode(kFALSE),
  87 fOutputRedirected(kFALSE)
  88 {
  89         //
  90         // config: AliShuttleConfig used
  91         // timeout: timeout used for AliDCSClient connection
  92         // retries: the number of retries in case of connection error.
  93         //
  94
  95         if (!fConfig->IsValid()) AliFatal("********** !!!!! Invalid configuration !!!!! **********");
  96         for(int iSys=0;iSys<4;iSys++) {
  97                 fServer[iSys]=0;
  98                 if (iSys < 3)
  99                         fFXSlist[iSys].SetOwner(kTRUE);
 100         }
 101         fPreprocessorMap.SetOwner(kTRUE);
 102
 103         for (UInt_t iDet=0; iDet<NDetectors(); iDet++)
 104                 fFirstUnprocessed[iDet] = kFALSE;
 105
 106         fMonitoringMutex = new TMutex();
 107 }
 108
 109 //______________________________________________________________________________________________
 110 AliShuttle::~AliShuttle()
 111 {
 112         //
 113         // destructor
 114         //
 115
 116         fPreprocessorMap.DeleteAll();
 117         for(int iSys=0;iSys<4;iSys++)
 118                 if(fServer[iSys]) {
 119                         fServer[iSys]->Close();
 120                         delete fServer[iSys];
 121                         fServer[iSys] = 0;
 122                 }
 123
 124         if (fStatusEntry){
 125                 delete fStatusEntry;
 126                 fStatusEntry = 0;
 127         }
 128
 129         if (fMonitoringMutex)
 130         {
 131                 delete fMonitoringMutex;
 132                 fMonitoringMutex = 0;
 133         }
 134 }
 135
 136 //______________________________________________________________________________________________
 137 void AliShuttle::RegisterPreprocessor(AliPreprocessor* preprocessor)
 138 {
 139         //
 140         // Registers new AliPreprocessor.
 141         // It uses GetName() for indentificator of the pre processor.
 142         // The pre processor is registered it there isn't any other
 143         // with the same identificator (GetName()).
 144         //
 145
 146         const char* detName = preprocessor->GetName();
 147         if(GetDetPos(detName) < 0)
 148                 AliFatal(Form("********** !!!!! Invalid detector name: %s !!!!! **********", detName));
 149
 150         if (fPreprocessorMap.GetValue(detName)) {
 151                 AliWarning(Form("AliPreprocessor %s is already registered!", detName));
 152                 return;
 153         }
 154
 155         fPreprocessorMap.Add(new TObjString(detName), preprocessor);
 156 }
 157 //______________________________________________________________________________________________
 158 Bool_t AliShuttle::Store(const AliCDBPath& path, TObject* object,
 159                 AliCDBMetaData* metaData, Int_t validityStart, Bool_t validityInfinite)
 160 {
 161         // Stores a CDB object in the storage for offline reconstruction. Objects that are not needed for
 162         // offline reconstruction, but should be stored anyway (e.g. for debugging) should NOT be stored
 163         // using this function. Use StoreReferenceData instead!
 164         // It calls StoreLocally function which temporarily stores the data locally; when the preprocessor
 165         // finishes the data are transferred to the main storage (Grid).
 166
 167         return StoreLocally(fgkLocalCDB, path, object, metaData, validityStart, validityInfinite);
 168 }
 169
 170 //______________________________________________________________________________________________
 171 Bool_t AliShuttle::StoreReferenceData(const AliCDBPath& path, TObject* object, AliCDBMetaData* metaData)
 172 {
 173         // Stores a CDB object in the storage for reference data. This objects will not be available during
 174         // offline reconstrunction. Use this function for reference data only!
 175         // It calls StoreLocally function which temporarily stores the data locally; when the preprocessor
 176         // finishes the data are transferred to the main storage (Grid).
 177
 178         return StoreLocally(fgkLocalRefStorage, path, object, metaData);
 179 }
 180
 181 //______________________________________________________________________________________________
 182 Bool_t AliShuttle::StoreLocally(const TString& localUri,
 183                         const AliCDBPath& path, TObject* object, AliCDBMetaData* metaData,
 184                         Int_t validityStart, Bool_t validityInfinite)
 185 {
 186         // Store object temporarily in local storage. Parameters are passed by Store and StoreReferenceData functions.
 187         // when the preprocessor finishes the data are transferred to the main storage (Grid).
 188         // The parameters are:
 189         //   1) Uri of the backup storage (Local)
 190         //   2) the object's path.
 191         //   3) the object to be stored
 192         //   4) the metaData to be associated with the object
 193         //   5) the validity start run number w.r.t. the current run,
 194         //      if the data is valid only for this run leave the default 0
 195         //   6) specifies if the calibration data is valid for infinity (this means until updated),
 196         //      typical for calibration runs, the default is kFALSE
 197         //
 198         // returns 0 if fail, 1 otherwise
 199
 200         if (fTestMode & kErrorStorage)
 201         {
 202                 Log(fCurrentDetector, "StoreLocally - In TESTMODE - Simulating error while storing locally");
 203                 return kFALSE;
 204         }
 205
 206         const char* cdbType = (localUri == fgkLocalCDB) ? "CDB" : "Reference";
 207
 208         Int_t firstRun = GetCurrentRun() - validityStart;
 209         if(firstRun < 0) {
 210                 AliWarning("First valid run happens to be less than 0! Setting it to 0.");
 211                 firstRun=0;
 212         }
 213
 214         Int_t lastRun = -1;
 215         if(validityInfinite) {
 216                 lastRun = AliCDBRunRange::Infinity();
 217         } else {
 218                 lastRun = GetCurrentRun();
 219         }
 220
 221         // Version is set to current run, it will be used later to transfer data to Grid
 222         AliCDBId id(path, firstRun, lastRun, GetCurrentRun(), -1);
 223
 224         if(! dynamic_cast<TObjString*> (metaData->GetProperty("RunUsed(TObjString)"))){
 225                 TObjString runUsed = Form("%d", GetCurrentRun());
 226                 metaData->SetProperty("RunUsed(TObjString)", runUsed.Clone());
 227         }
 228
 229         Bool_t result = kFALSE;
 230
 231         if (!(AliCDBManager::Instance()->GetStorage(localUri))) {
 232                 Log("SHUTTLE", Form("StoreLocally - Cannot activate local %s storage", cdbType));
 233         } else {
 234                 result = AliCDBManager::Instance()->GetStorage(localUri)
 235                                         ->Put(object, id, metaData);
 236         }
 237
 238         if(!result) {
 239
 240                 Log(fCurrentDetector, Form("StoreLocally - Can't store object <%s>!", id.ToString().Data()));
 241         }
 242
 243         return result;
 244 }
 245
 246 //______________________________________________________________________________________________
 247 Bool_t AliShuttle::StoreOCDB()
 248 {
 249         //
 250         // Called when preprocessor ends successfully or when previous storage attempt failed (kStoreError status)
 251         // Calls underlying StoreOCDB(const char*) function twice, for OCDB and Reference storage.
 252         // Then calls StoreRefFilesToGrid to store reference files.
 253         //
 254
 255         if (fTestMode & kErrorGrid)
 256         {
 257                 Log("SHUTTLE", "StoreOCDB - In TESTMODE - Simulating error while storing in the Grid");
 258                 Log(fCurrentDetector, "StoreOCDB - In TESTMODE - Simulating error while storing in the Grid");
 259                 return kFALSE;
 260         }
 261
 262         Log("SHUTTLE","StoreOCDB - Storing OCDB data ...");
 263         Bool_t resultCDB = StoreOCDB(fgkMainCDB);
 264
 265         Log("SHUTTLE","StoreOCDB - Storing reference data ...");
 266         Bool_t resultRef = StoreOCDB(fgkMainRefStorage);
 267
 268         Log("SHUTTLE","StoreOCDB - Storing reference files ...");
 269         Bool_t resultRefFiles = CopyFilesToGrid("reference");
 270
 271         Bool_t resultMetadata = kTRUE;
 272         if(fCurrentDetector == "GRP")
 273         {
 274                 Log("StoreOCDB - SHUTTLE","Storing Run Metadata file ...");
 275                 resultMetadata = CopyFilesToGrid("metadata");
 276         }
 277
 278         return resultCDB && resultRef && resultRefFiles && resultMetadata;
 279 }
 280
 281 //______________________________________________________________________________________________
 282 Bool_t AliShuttle::StoreOCDB(const TString& gridURI)
 283 {
 284         //
 285         // Called by StoreOCDB(), performs actual storage to the main OCDB and reference storages (Grid)
 286         //
 287
 288         TObjArray* gridIds=0;
 289
 290         Bool_t result = kTRUE;
 291
 292         const char* type = 0;
 293         TString localURI;
 294         if(gridURI == fgkMainCDB) {
 295                 type = "OCDB";
 296                 localURI = fgkLocalCDB;
 297         } else if(gridURI == fgkMainRefStorage) {
 298                 type = "reference";
 299                 localURI = fgkLocalRefStorage;
 300         } else {
 301                 AliError(Form("Invalid storage URI: %s", gridURI.Data()));
 302                 return kFALSE;
 303         }
 304
 305         AliCDBManager* man = AliCDBManager::Instance();
 306
 307         AliCDBStorage *gridSto = man->GetStorage(gridURI);
 308         if(!gridSto) {
 309                 Log("SHUTTLE",
 310                         Form("StoreOCDB - cannot activate main %s storage", type));
 311                 return kFALSE;
 312         }
 313
 314         gridIds = gridSto->GetQueryCDBList();
 315
 316         // get objects previously stored in local CDB
 317         AliCDBStorage *localSto = man->GetStorage(localURI);
 318         if(!localSto) {
 319                 Log("SHUTTLE",
 320                         Form("StoreOCDB - cannot activate local %s storage", type));
 321                 return kFALSE;
 322         }
 323         AliCDBPath aPath(GetOfflineDetName(fCurrentDetector.Data()),"*","*");
 324         // Local objects were stored with current run as Grid version!
 325         TList* localEntries = localSto->GetAll(aPath.GetPath(), GetCurrentRun(), GetCurrentRun());
 326         localEntries->SetOwner(1);
 327
 328         // loop on local stored objects
 329         TIter localIter(localEntries);
 330         AliCDBEntry *aLocEntry = 0;
 331         while((aLocEntry = dynamic_cast<AliCDBEntry*> (localIter.Next()))){
 332                 aLocEntry->SetOwner(1);
 333                 AliCDBId aLocId = aLocEntry->GetId();
 334                 aLocEntry->SetVersion(-1);
 335                 aLocEntry->SetSubVersion(-1);
 336
 337                 // If local object is valid up to infinity we store it only if it is
 338                 // the first unprocessed run!
 339                 if (aLocId.GetLastRun() == AliCDBRunRange::Infinity() &&
 340                         !fFirstUnprocessed[GetDetPos(fCurrentDetector)])
 341                 {
 342                         Log("SHUTTLE", Form("StoreOCDB - %s: object %s has validity infinite but "
 343                                                 "there are previous unprocessed runs!",
 344                                                 fCurrentDetector.Data(), aLocId.GetPath().Data()));
 345                         result = kFALSE;
 346                         continue;
 347                 }
 348
 349                 // loop on Grid valid Id's
 350                 Bool_t store = kTRUE;
 351                 TIter gridIter(gridIds);
 352                 AliCDBId* aGridId = 0;
 353                 while((aGridId = dynamic_cast<AliCDBId*> (gridIter.Next()))){
 354                         if(aGridId->GetPath() != aLocId.GetPath()) continue;
 355                         // skip all objects valid up to infinity
 356                         if(aGridId->GetLastRun() == AliCDBRunRange::Infinity()) continue;
 357                         // if we get here, it means there's already some more recent object stored on Grid!
 358                         store = kFALSE;
 359                         break;
 360                 }
 361
 362                 // If we get here, the file can be stored!
 363                 Bool_t storeOk = gridSto->Put(aLocEntry);
 364                 if(!store || storeOk){
 365
 366                         if (!store)
 367                         {
 368                                 Log(fCurrentDetector.Data(),
 369                                         Form("StoreOCDB - A more recent object already exists in %s storage: <%s>",
 370                                                 type, aGridId->ToString().Data()));
 371                         } else {
 372                                 Log("SHUTTLE",
 373                                         Form("StoreOCDB - Object <%s> successfully put into %s storage",
 374                                                 aLocId.ToString().Data(), type));
 375                                 Log(fCurrentDetector.Data(),
 376                                         Form("StoreOCDB - Object <%s> successfully put into %s storage",
 377                                                 aLocId.ToString().Data(), type));
 378                         }
 379
 380                         // removing local filename...
 381                         TString filename;
 382                         localSto->IdToFilename(aLocId, filename);
 383                         Log("SHUTTLE", Form("StoreOCDB - Removing local file %s", filename.Data()));
 384                         RemoveFile(filename.Data());
 385                         continue;
 386                 } else  {
 387                         Log("SHUTTLE",
 388                                 Form("StoreOCDB - Grid %s storage of object <%s> failed",
 389                                         type, aLocId.ToString().Data()));
 390                         Log(fCurrentDetector.Data(),
 391                                 Form("StoreOCDB - Grid %s storage of object <%s> failed",
 392                                         type, aLocId.ToString().Data()));
 393                         result = kFALSE;
 394                 }
 395         }
 396         localEntries->Clear();
 397
 398         return result;
 399 }
 400
 401 //______________________________________________________________________________________________
 402 Bool_t AliShuttle::CleanReferenceStorage(const char* detector)
 403 {
 404         // clears the directory used to store reference files of a given subdetector
 405
 406         AliCDBManager* man = AliCDBManager::Instance();
 407         AliCDBStorage* sto = man->GetStorage(fgkLocalRefStorage);
 408         TString localBaseFolder = sto->GetBaseFolder();
 409
 410         TString targetDir = GetRefFilePrefix(localBaseFolder.Data(), detector);
 411
 412         Log("SHUTTLE", Form("CleanReferenceStorage - Cleaning %s", targetDir.Data()));
 413
 414         TString begin;
 415         begin.Form("%d_", GetCurrentRun());
 416
 417         TSystemDirectory* baseDir = new TSystemDirectory("/", targetDir);
 418         if (!baseDir)
 419                 return kTRUE;
 420
 421         TList* dirList = baseDir->GetListOfFiles();
 422         delete baseDir;
 423
 424         if (!dirList) return kTRUE;
 425
 426         if (dirList->GetEntries() < 3)
 427         {
 428                 delete dirList;
 429                 return kTRUE;
 430         }
 431
 432         Int_t nDirs = 0, nDel = 0;
 433         TIter dirIter(dirList);
 434         TSystemFile* entry = 0;
 435
 436         Bool_t success = kTRUE;
 437
 438         while ((entry = dynamic_cast<TSystemFile*> (dirIter.Next())))
 439         {
 440                 if (entry->IsDirectory())
 441                         continue;
 442
 443                 TString fileName(entry->GetName());
 444                 if (!fileName.BeginsWith(begin))
 445                         continue;
 446
 447                 nDirs++;
 448
 449                 // delete file
 450                 Int_t result = gSystem->Unlink(fileName.Data());
 451
 452                 if (result)
 453                 {
 454                         Log("SHUTTLE", Form("CleanReferenceStorage - Could not delete file %s!", fileName.Data()));
 455                         success = kFALSE;
 456                 } else {
 457                         nDel++;
 458                 }
 459         }
 460
 461         if(nDirs > 0)
 462                 Log("SHUTTLE", Form("CleanReferenceStorage - %d (over %d) reference files in folder %s were deleted.",
 463                         nDel, nDirs, targetDir.Data()));
 464
 465
 466         delete dirList;
 467         return success;
 468
 469
 470
 471
 472
 473
 474   Int_t result = gSystem->GetPathInfo(targetDir, 0, (Long64_t*) 0, 0, 0);
 475   if (result == 0)
 476   {
 477     // delete directory
 478     result = gSystem->Exec(Form("rm -rf %s", targetDir.Data()));
 479     if (result != 0)
 480     {
 481       Log("SHUTTLE", Form("CleanReferenceStorage - Could not clean directory %s", targetDir.Data()));
 482       return kFALSE;
 483     }
 484   }
 485
 486   result = gSystem->mkdir(targetDir, kTRUE);
 487   if (result != 0)
 488   {
 489     Log("SHUTTLE", Form("CleanReferenceStorage - Error creating base directory %s", targetDir.Data()));
 490     return kFALSE;
 491   }
 492
 493   return kTRUE;
 494 }
 495
 496 //______________________________________________________________________________________________
 497 Bool_t AliShuttle::StoreReferenceFile(const char* detector, const char* localFile, const char* gridFileName)
 498 {
 499         //
 500         // Stores reference file directly (without opening it). This function stores the file locally.
 501         //
 502         // The file is stored under the following location:
 503         // <base folder of local reference storage>/<DET>/<RUN#>_<gridFileName>
 504         // where <gridFileName> is the second parameter given to the function
 505         //
 506
 507         if (fTestMode & kErrorStorage)
 508         {
 509                 Log(fCurrentDetector, "StoreReferenceFile - In TESTMODE - Simulating error while storing locally");
 510                 return kFALSE;
 511         }
 512
 513         AliCDBManager* man = AliCDBManager::Instance();
 514         AliCDBStorage* sto = man->GetStorage(fgkLocalRefStorage);
 515
 516         TString localBaseFolder = sto->GetBaseFolder();
 517
 518         TString target = GetRefFilePrefix(localBaseFolder.Data(), detector);
 519         target.Append(Form("/%d_%s", GetCurrentRun(), gridFileName));
 520
 521         return CopyFileLocally(localFile, target);
 522 }
 523
 524 //______________________________________________________________________________________________
 525 Bool_t AliShuttle::StoreRunMetadataFile(const char* localFile, const char* gridFileName)
 526 {
 527         //
 528         // Stores Run metadata file to the Grid, in the run folder
 529         //
 530         // Only GRP can call this function.
 531
 532         if (fTestMode & kErrorStorage)
 533         {
 534                 Log(fCurrentDetector, "StoreRunMetaDataFile - In TESTMODE - Simulating error while storing locally");
 535                 return kFALSE;
 536         }
 537
 538         AliCDBManager* man = AliCDBManager::Instance();
 539         AliCDBStorage* sto = man->GetStorage(fgkLocalRefStorage);
 540
 541         TString localBaseFolder = sto->GetBaseFolder();
 542
 543         // Build Run level folder
 544         // folder = /alice/data/year/lhcPeriod/runNb/raw
 545
 546
 547         TString lhcPeriod = GetLHCPeriod();
 548         if (lhcPeriod.Length() == 0)
 549         {
 550                 Log("SHUTTLE","StoreRunMetaDataFile - LHCPeriod not found in logbook!");
 551                 return 0;
 552         }
 553
 554         // TODO partitions with one detector only write data into LHCperiod_DET
 555         TString partition = GetRunParameter("detector");
 556
 557         if (partition.Length() > 0 && partition != "ALICE")
 558         {
 559                 lhcPeriod.Append(Form("_%s", partition.Data()));
 560                 Log(fCurrentDetector, Form("Run data tags merged file will be written in %s",
 561                                 lhcPeriod.Data()));
 562         }
 563
 564         TString target = Form("%s/GRP/RunMetadata/alice/data/%d/%s/%09d/raw/%s",
 565                                 localBaseFolder.Data(), GetCurrentYear(),
 566                                 lhcPeriod.Data(), GetCurrentRun(), gridFileName);
 567
 568         return CopyFileLocally(localFile, target);
 569 }
 570
 571 //______________________________________________________________________________________________
 572 Bool_t AliShuttle::CopyFileLocally(const char* localFile, const TString& target)
 573 {
 574         //
 575         // Stores file locally. Called by StoreReferenceFile and StoreRunMetadataFile
 576         // Files are temporarily stored in the local reference storage. When the preprocessor
 577         // finishes, the Shuttle calls CopyFilesToGrid to transfer the files to AliEn
 578         // (in reference or run level folders)
 579         //
 580
 581         TString targetDir(target(0, target.Last('/')));
 582
 583         //try to open base dir folder, if it does not exist
 584         void* dir = gSystem->OpenDirectory(targetDir.Data());
 585         if (dir == NULL) {
 586                 if (gSystem->mkdir(targetDir.Data(), kTRUE)) {
 587                         Log("SHUTTLE", Form("CopyFileLocally - Can't open directory <%s>", targetDir.Data()));
 588                         return kFALSE;
 589                 }
 590
 591         } else {
 592                 gSystem->FreeDirectory(dir);
 593         }
 594
 595         Int_t result = 0;
 596
 597         result = gSystem->GetPathInfo(localFile, 0, (Long64_t*) 0, 0, 0);
 598         if (result)
 599         {
 600                 Log("SHUTTLE", Form("CopyFileLocally - %s does not exist", localFile));
 601                 return kFALSE;
 602         }
 603
 604         result = gSystem->GetPathInfo(target, 0, (Long64_t*) 0, 0, 0);
 605         if (!result)
 606         {
 607                 Log("SHUTTLE", Form("CopyFileLocally - target file %s already exist, removing...", target.Data()));
 608                 if (gSystem->Unlink(target.Data()))
 609                 {
 610                         Log("SHUTTLE", Form("CopyFileLocally - Could not remove existing target file %s!", target.Data()));
 611                         return kFALSE;
 612                 }
 613         }
 614
 615         result = gSystem->CopyFile(localFile, target);
 616
 617         if (result == 0)
 618         {
 619                 Log("SHUTTLE", Form("CopyFileLocally - File %s stored locally to %s", localFile, target.Data()));
 620                 return kTRUE;
 621         }
 622         else
 623         {
 624                 Log("SHUTTLE", Form("CopyFileLocally - Could not store file %s to %s! Error code = %d",
 625                                 localFile, target.Data(), result));
 626                 return kFALSE;
 627         }
 628
 629
 630
 631 }
 632
 633 //______________________________________________________________________________________________
 634 Bool_t AliShuttle::CopyFilesToGrid(const char* type)
 635 {
 636         //
 637         // Transfers local files to the Grid. Local files can be reference files
 638         // or run metadata file (from GRP only).
 639         //
 640         // According to the type (ref, metadata) the files are stored under the following location:
 641         // ref --> <base folder of reference storage>/<DET>/<RUN#>_<gridFileName>
 642         // metadata --> <run data folder>/<MetadataFileName>
 643         //
 644
 645         AliCDBManager* man = AliCDBManager::Instance();
 646         AliCDBStorage* sto = man->GetStorage(fgkLocalRefStorage);
 647         if (!sto)
 648                 return kFALSE;
 649         TString localBaseFolder = sto->GetBaseFolder();
 650
 651         TString dir;
 652         TString alienDir;
 653         TString begin;
 654
 655         if (strcmp(type, "reference") == 0)
 656         {
 657                 dir = GetRefFilePrefix(localBaseFolder.Data(), fCurrentDetector.Data());
 658                 AliCDBStorage* gridSto = man->GetStorage(fgkMainRefStorage);
 659                 if (!gridSto)
 660                         return kFALSE;
 661                 TString gridBaseFolder = gridSto->GetBaseFolder();
 662                 alienDir = GetRefFilePrefix(gridBaseFolder.Data(), fCurrentDetector.Data());
 663                 begin = Form("%d_", GetCurrentRun());
 664         }
 665         else if (strcmp(type, "metadata") == 0)
 666         {
 667
 668                 TString lhcPeriod = GetLHCPeriod();
 669
 670                 if (lhcPeriod.Length() == 0)
 671                 {
 672                         Log("SHUTTLE","CopyFilesToGrid - LHCPeriod not found in logbook!");
 673                         return 0;
 674                 }
 675
 676                 // TODO partitions with one detector only write data into LHCperiod_DET
 677                 TString partition = GetRunParameter("detector");
 678
 679                 if (partition.Length() > 0 && partition != "ALICE")
 680                 {
 681                         lhcPeriod.Append(Form("_%s", partition.Data()));
 682                 }
 683
 684                 dir = Form("%s/GRP/RunMetadata/alice/data/%d/%s/%09d/raw",
 685                                 localBaseFolder.Data(), GetCurrentYear(),
 686                                 lhcPeriod.Data(), GetCurrentRun());
 687                 alienDir = dir(dir.Index("/alice/data/"), dir.Length());
 688
 689                 begin = "";
 690         }
 691         else
 692         {
 693                 Log("SHUTTLE", "CopyFilesToGrid - Unexpected: type label must be reference or metadata!");
 694                 return kFALSE;
 695         }
 696
 697         TSystemDirectory* baseDir = new TSystemDirectory("/", dir);
 698         if (!baseDir)
 699                 return kTRUE;
 700
 701         TList* dirList = baseDir->GetListOfFiles();
 702         delete baseDir;
 703
 704         if (!dirList) return kTRUE;
 705
 706         if (dirList->GetEntries() < 3)
 707         {
 708                 delete dirList;
 709                 return kTRUE;
 710         }
 711
 712         if (!gGrid)
 713         {
 714                 Log("SHUTTLE", "CopyFilesToGrid - Connection to Grid failed: Cannot continue!");
 715                 delete dirList;
 716                 return kFALSE;
 717         }
 718
 719         Int_t nDirs = 0, nTransfer = 0;
 720         TIter dirIter(dirList);
 721         TSystemFile* entry = 0;
 722
 723         Bool_t success = kTRUE;
 724         Bool_t first = kTRUE;
 725
 726         while ((entry = dynamic_cast<TSystemFile*> (dirIter.Next())))
 727         {
 728                 if (entry->IsDirectory())
 729                         continue;
 730
 731                 TString fileName(entry->GetName());
 732                 if (!fileName.BeginsWith(begin))
 733                         continue;
 734
 735                 nDirs++;
 736
 737                 if (first)
 738                 {
 739                         first = kFALSE;
 740                         // check that folder exists, otherwise create it
 741                         TGridResult* result = gGrid->Ls(alienDir.Data(), "a");
 742
 743                         if (!result)
 744                         {
 745                                 delete dirList;
 746                                 return kFALSE;
 747                         }
 748
 749                         if (!result->GetFileName(1)) // TODO: It looks like element 0 is always 0!!
 750                         {
 751                                 // TODO It does not work currently! Bug in TAliEn::Mkdir
 752                                 // TODO Manually fixed in local root v5-16-00
 753                                 if (!gGrid->Mkdir(alienDir.Data(),"-p",0))
 754                                 {
 755                                         Log("SHUTTLE", Form("CopyFilesToGrid - Cannot create directory %s",
 756                                                         alienDir.Data()));
 757                                         delete dirList;
 758                                         return kFALSE;
 759                                 } else {
 760                                         Log("SHUTTLE",Form("CopyFilesToGrid - Folder %s created", alienDir.Data()));
 761                                 }
 762
 763                         } else {
 764                                         Log("SHUTTLE",Form("CopyFilesToGrid - Folder %s found", alienDir.Data()));
 765                         }
 766                 }
 767
 768                 TString fullLocalPath;
 769                 fullLocalPath.Form("%s/%s", dir.Data(), fileName.Data());
 770
 771                 TString fullGridPath;
 772                 fullGridPath.Form("alien://%s/%s", alienDir.Data(), fileName.Data());
 773
 774                 Bool_t result = TFile::Cp(fullLocalPath, fullGridPath);
 775
 776                 if (result)
 777                 {
 778                         Log("SHUTTLE", Form("CopyFilesToGrid - Copying local file %s to %s succeeded!",
 779                                                 fullLocalPath.Data(), fullGridPath.Data()));
 780                         RemoveFile(fullLocalPath);
 781                         nTransfer++;
 782                 }
 783                 else
 784                 {
 785                         Log("SHUTTLE", Form("CopyFilesToGrid - Copying local file %s to %s FAILED!",
 786                                                 fullLocalPath.Data(), fullGridPath.Data()));
 787                         success = kFALSE;
 788                 }
 789         }
 790
 791         Log("SHUTTLE", Form("CopyFilesToGrid - %d (over %d) files in folder %s copied to Grid.",
 792                                                 nTransfer, nDirs, dir.Data()));
 793
 794
 795         delete dirList;
 796         return success;
 797 }
 798
 799 //______________________________________________________________________________________________
 800 const char* AliShuttle::GetRefFilePrefix(const char* base, const char* detector)
 801 {
 802         //
 803         // Get folder name of reference files
 804         //
 805
 806         TString offDetStr(GetOfflineDetName(detector));
 807         TString dir;
 808         if (offDetStr == "ITS" || offDetStr == "MUON" || offDetStr == "PHOS")
 809         {
 810                 dir.Form("%s/%s/%s", base, offDetStr.Data(), detector);
 811         } else {
 812                 dir.Form("%s/%s", base, offDetStr.Data());
 813         }
 814
 815         return dir.Data();
 816
 817
 818 }
 819
 820 //______________________________________________________________________________________________
 821 void AliShuttle::CleanLocalStorage(const TString& uri)
 822 {
 823         //
 824         // Called in case the preprocessor is declared failed. Remove remaining objects from the local storages.
 825         //
 826
 827         const char* type = 0;
 828         if(uri == fgkLocalCDB) {
 829                 type = "OCDB";
 830         } else if(uri == fgkLocalRefStorage) {
 831                 type = "Reference";
 832         } else {
 833                 AliError(Form("Invalid storage URI: %s", uri.Data()));
 834                 return;
 835         }
 836
 837         AliCDBManager* man = AliCDBManager::Instance();
 838
 839         // open local storage
 840         AliCDBStorage *localSto = man->GetStorage(uri);
 841         if(!localSto) {
 842                 Log("SHUTTLE",
 843                         Form("CleanLocalStorage - cannot activate local %s storage", type));
 844                 return;
 845         }
 846
 847         TString filename(Form("%s/%s/*/Run*_v%d_s*.root",
 848                 localSto->GetBaseFolder().Data(), GetOfflineDetName(fCurrentDetector.Data()), GetCurrentRun()));
 849
 850         AliDebug(2, Form("filename = %s", filename.Data()));
 851
 852         Log("SHUTTLE", Form("Removing remaining local files for run %d and detector %s ...",
 853                 GetCurrentRun(), fCurrentDetector.Data()));
 854
 855         RemoveFile(filename.Data());
 856
 857 }
 858
 859 //______________________________________________________________________________________________
 860 void AliShuttle::RemoveFile(const char* filename)
 861 {
 862         //
 863         // removes local file
 864         //
 865
 866         TString command(Form("rm -f %s", filename));
 867
 868         Int_t result = gSystem->Exec(command.Data());
 869         if(result != 0)
 870         {
 871                 Log("SHUTTLE", Form("RemoveFile - %s: Cannot remove file %s!",
 872                         fCurrentDetector.Data(), filename));
 873         }
 874 }
 875
 876 //______________________________________________________________________________________________
 877 AliShuttleStatus* AliShuttle::ReadShuttleStatus()
 878 {
 879         //
 880         // Reads the AliShuttleStatus from the CDB
 881         //
 882
 883         if (fStatusEntry){
 884                 delete fStatusEntry;
 885                 fStatusEntry = 0;
 886         }
 887
 888         fStatusEntry = AliCDBManager::Instance()->GetStorage(GetLocalCDB())
 889                 ->Get(Form("/SHUTTLE/STATUS/%s", fCurrentDetector.Data()), GetCurrentRun());
 890
 891         if (!fStatusEntry) return 0;
 892         fStatusEntry->SetOwner(1);
 893
 894         AliShuttleStatus* status = dynamic_cast<AliShuttleStatus*> (fStatusEntry->GetObject());
 895         if (!status) {
 896                 AliError("Invalid object stored to CDB!");
 897                 return 0;
 898         }
 899
 900         return status;
 901 }
 902
 903 //______________________________________________________________________________________________
 904 Bool_t AliShuttle::WriteShuttleStatus(AliShuttleStatus* status)
 905 {
 906         //
 907         // writes the status for one subdetector
 908         //
 909
 910         if (fStatusEntry){
 911                 delete fStatusEntry;
 912                 fStatusEntry = 0;
 913         }
 914
 915         Int_t run = GetCurrentRun();
 916
 917         AliCDBId id(AliCDBPath("SHUTTLE", "STATUS", fCurrentDetector), run, run);
 918
 919         fStatusEntry = new AliCDBEntry(status, id, new AliCDBMetaData);
 920         fStatusEntry->SetOwner(1);
 921
 922         UInt_t result = AliCDBManager::Instance()->GetStorage(fgkLocalCDB)->Put(fStatusEntry);
 923
 924         if (!result) {
 925                 Log("SHUTTLE", Form("WriteShuttleStatus - Failed for %s, run %d",
 926                                                 fCurrentDetector.Data(), run));
 927                 return kFALSE;
 928         }
 929
 930         SendMLInfo();
 931
 932         return kTRUE;
 933 }
 934
 935 //______________________________________________________________________________________________
 936 void AliShuttle::UpdateShuttleStatus(AliShuttleStatus::Status newStatus, Bool_t increaseCount)
 937 {
 938         //
 939         // changes the AliShuttleStatus for the given detector and run to the given status
 940         //
 941
 942         if (!fStatusEntry){
 943                 AliError("UNEXPECTED: fStatusEntry empty");
 944                 return;
 945         }
 946
 947         AliShuttleStatus* status = dynamic_cast<AliShuttleStatus*> (fStatusEntry->GetObject());
 948
 949         if (!status){
 950                 Log("SHUTTLE", "UpdateShuttleStatus - UNEXPECTED: status could not be read from current CDB entry");
 951                 return;
 952         }
 953
 954         TString actionStr = Form("UpdateShuttleStatus - %s: Changing state from %s to %s",
 955                                 fCurrentDetector.Data(),
 956                                 status->GetStatusName(),
 957                                 status->GetStatusName(newStatus));
 958         Log("SHUTTLE", actionStr);
 959         SetLastAction(actionStr);
 960
 961         status->SetStatus(newStatus);
 962         if (increaseCount) status->IncreaseCount();
 963
 964         AliCDBManager::Instance()->GetStorage(fgkLocalCDB)->Put(fStatusEntry);
 965
 966         SendMLInfo();
 967 }
 968
 969 //______________________________________________________________________________________________
 970 void AliShuttle::SendMLInfo()
 971 {
 972         //
 973         // sends ML information about the current status of the current detector being processed
 974         //
 975
 976         AliShuttleStatus* status = dynamic_cast<AliShuttleStatus*> (fStatusEntry->GetObject());
 977
 978         if (!status){
 979                 Log("SHUTTLE", "SendMLInfo - UNEXPECTED: status could not be read from current CDB entry");
 980                 return;
 981         }
 982
 983         TMonaLisaText  mlStatus(Form("%s_status", fCurrentDetector.Data()), status->GetStatusName());
 984         TMonaLisaValue mlRetryCount(Form("%s_count", fCurrentDetector.Data()), status->GetCount());
 985
 986         TList mlList;
 987         mlList.Add(&mlStatus);
 988         mlList.Add(&mlRetryCount);
 989
 990         TString mlID;
 991         mlID.Form("%d", GetCurrentRun());
 992         fMonaLisa->SendParameters(&mlList, mlID);
 993 }
 994
 995 //______________________________________________________________________________________________
 996 Bool_t AliShuttle::ContinueProcessing()
 997 {
 998         // this function reads the AliShuttleStatus information from CDB and
 999         // checks if the processing should be continued
1000         // if yes it returns kTRUE and updates the AliShuttleStatus with nextStatus
1001
1002         if (!fConfig->HostProcessDetector(fCurrentDetector)) return kFALSE;
1003
1004         AliPreprocessor* aPreprocessor =
1005                 dynamic_cast<AliPreprocessor*> (fPreprocessorMap.GetValue(fCurrentDetector));
1006         if (!aPreprocessor)
1007         {
1008                 Log("SHUTTLE", Form("ContinueProcessing - %s: no preprocessor registered", fCurrentDetector.Data()));
1009                 return kFALSE;
1010         }
1011
1012         AliShuttleLogbookEntry::Status entryStatus =
1013                 fLogbookEntry->GetDetectorStatus(fCurrentDetector);
1014
1015         if(entryStatus != AliShuttleLogbookEntry::kUnprocessed) {
1016                 Log("SHUTTLE", Form("ContinueProcessing - %s is %s",
1017                                 fCurrentDetector.Data(),
1018                                 fLogbookEntry->GetDetectorStatusName(entryStatus)));
1019                 return kFALSE;
1020         }
1021
1022         // if we get here, according to Shuttle logbook subdetector is in UNPROCESSED state
1023
1024         // check if current run is first unprocessed run for current detector
1025         if (fConfig->StrictRunOrder(fCurrentDetector) &&
1026                 !fFirstUnprocessed[GetDetPos(fCurrentDetector)])
1027         {
1028                 if (fTestMode == kNone)
1029                 {
1030                         Log("SHUTTLE", Form("ContinueProcessing - %s requires strict run ordering"
1031                                         " but this is not the first unprocessed run!"));
1032                         return kFALSE;
1033                 }
1034                 else
1035                 {
1036                         Log("SHUTTLE", Form("ContinueProcessing - In TESTMODE - "
1037                                         "Although %s requires strict run ordering "
1038                                         "and this is not the first unprocessed run, "
1039                                         "the SHUTTLE continues"));
1040                 }
1041         }
1042
1043         AliShuttleStatus* status = ReadShuttleStatus();
1044         if (!status) {
1045                 // first time
1046                 Log("SHUTTLE", Form("ContinueProcessing - %s: Processing first time",
1047                                 fCurrentDetector.Data()));
1048                 status = new AliShuttleStatus(AliShuttleStatus::kStarted);
1049                 return WriteShuttleStatus(status);
1050         }
1051
1052         // The following two cases shouldn't happen if Shuttle Logbook was correctly updated.
1053         // If it happens it may mean Logbook updating failed... let's do it now!
1054         if (status->GetStatus() == AliShuttleStatus::kDone ||
1055             status->GetStatus() == AliShuttleStatus::kFailed){
1056                 Log("SHUTTLE", Form("ContinueProcessing - %s is already %s. Updating Shuttle Logbook",
1057                                         fCurrentDetector.Data(),
1058                                         status->GetStatusName(status->GetStatus())));
1059                 UpdateShuttleLogbook(fCurrentDetector.Data(),
1060                                         status->GetStatusName(status->GetStatus()));
1061                 return kFALSE;
1062         }
1063
1064         if (status->GetStatus() == AliShuttleStatus::kStoreStarted || status->GetStatus() == AliShuttleStatus::kStoreError) {
1065                 Log("SHUTTLE",
1066                         Form("ContinueProcessing - %s: Grid storage of one or more "
1067                                 "objects failed. Trying again now",
1068                                 fCurrentDetector.Data()));
1069                 UpdateShuttleStatus(AliShuttleStatus::kStoreStarted);
1070                 if (StoreOCDB()){
1071                         Log("SHUTTLE", Form("ContinueProcessing - %s: all objects "
1072                                 "successfully stored into main storage",
1073                                 fCurrentDetector.Data()));
1074                         UpdateShuttleStatus(AliShuttleStatus::kDone);
1075                         UpdateShuttleLogbook(fCurrentDetector, "DONE");
1076                 } else {
1077                         Log("SHUTTLE",
1078                                 Form("ContinueProcessing - %s: Grid storage failed again",
1079                                         fCurrentDetector.Data()));
1080                         UpdateShuttleStatus(AliShuttleStatus::kStoreError);
1081                 }
1082                 return kFALSE;
1083         }
1084
1085         // if we get here, there is a restart
1086         Bool_t cont = kFALSE;
1087
1088         // abort conditions
1089         if (status->GetCount() >= fConfig->GetMaxRetries()) {
1090                 Log("SHUTTLE", Form("ContinueProcessing - %s failed %d times in status %s - "
1091                                 "Updating Shuttle Logbook", fCurrentDetector.Data(),
1092                                 status->GetCount(), status->GetStatusName()));
1093                 UpdateShuttleLogbook(fCurrentDetector.Data(), "FAILED");
1094                 UpdateShuttleStatus(AliShuttleStatus::kFailed);
1095
1096                 // there may still be objects in local OCDB and reference storage
1097                 // and FXS databases may be not updated: do it now!
1098
1099                 // TODO Currently disabled, we want to keep files in case of failure!
1100                 // CleanLocalStorage(fgkLocalCDB);
1101                 // CleanLocalStorage(fgkLocalRefStorage);
1102                 // UpdateTableFailCase();
1103
1104                 // Send mail to detector expert!
1105                 Log("SHUTTLE", Form("ContinueProcessing - Sending mail to %s expert...",
1106                                         fCurrentDetector.Data()));
1107                 if (!SendMail())
1108                         Log("SHUTTLE", Form("ContinueProcessing - Could not send mail to %s expert",
1109                                         fCurrentDetector.Data()));
1110
1111         } else {
1112                 Log("SHUTTLE", Form("ContinueProcessing - %s: restarting. "
1113                                 "Aborted before with %s. Retry number %d.", fCurrentDetector.Data(),
1114                                 status->GetStatusName(), status->GetCount()));
1115                 Bool_t increaseCount = kTRUE;
1116                 if (status->GetStatus() == AliShuttleStatus::kDCSError ||
1117                         status->GetStatus() == AliShuttleStatus::kDCSStarted)
1118                                 increaseCount = kFALSE;
1119
1120                 UpdateShuttleStatus(AliShuttleStatus::kStarted, increaseCount);
1121                 cont = kTRUE;
1122         }
1123
1124         return cont;
1125 }
1126
1127 //______________________________________________________________________________________________
1128 Bool_t AliShuttle::Process(AliShuttleLogbookEntry* entry)
1129 {
1130         //
1131         // Makes data retrieval for all detectors in the configuration.
1132         // entry: Shuttle logbook entry, contains run paramenters and status of detectors
1133         // (Unprocessed, Inactive, Failed or Done).
1134         // Returns kFALSE in case of error occured and kTRUE otherwise
1135         //
1136
1137         if (!entry) return kFALSE;
1138
1139         fLogbookEntry = entry;
1140
1141         Log("SHUTTLE", Form("\t\t\t^*^*^*^*^*^*^*^*^*^*^*^* run %d: START ^*^*^*^*^*^*^*^*^*^*^*^*",
1142                                         GetCurrentRun()));
1143
1144         // Send the information to ML
1145         TMonaLisaText  mlStatus("SHUTTLE_status", "Processing");
1146         TMonaLisaText  mlRunType("SHUTTLE_runtype", Form("%s (%s)", entry->GetRunType(), entry->GetRunParameter("log")));
1147
1148         TList mlList;
1149         mlList.Add(&mlStatus);
1150         mlList.Add(&mlRunType);
1151
1152         TString mlID;
1153         mlID.Form("%d", GetCurrentRun());
1154         fMonaLisa->SendParameters(&mlList, mlID);
1155
1156         if (fLogbookEntry->IsDone())
1157         {
1158                 Log("SHUTTLE","Process - Shuttle is already DONE. Updating logbook");
1159                 UpdateShuttleLogbook("shuttle_done");
1160                 fLogbookEntry = 0;
1161                 return kTRUE;
1162         }
1163
1164         // read test mode if flag is set
1165         if (fReadTestMode)
1166         {
1167                 fTestMode = kNone;
1168                 TString logEntry(entry->GetRunParameter("log"));
1169                 //printf("log entry = %s\n", logEntry.Data());
1170                 TString searchStr("Testmode: ");
1171                 Int_t pos = logEntry.Index(searchStr.Data());
1172                 //printf("%d\n", pos);
1173                 if (pos >= 0)
1174                 {
1175                         TSubString subStr = logEntry(pos + searchStr.Length(), logEntry.Length());
1176                         //printf("%s\n", subStr.String().Data());
1177                         TString newStr(subStr.Data());
1178                         TObjArray* token = newStr.Tokenize(' ');
1179                         if (token)
1180                         {
1181                                 //token->Print();
1182                                 TObjString* tmpStr = dynamic_cast<TObjString*> (token->First());
1183                                 if (tmpStr)
1184                                 {
1185                                         Int_t testMode = tmpStr->String().Atoi();
1186                                         if (testMode > 0)
1187                                         {
1188                                                 Log("SHUTTLE", Form("Process - Enabling test mode %d", testMode));
1189                                                 SetTestMode((TestMode) testMode);
1190                                         }
1191                                 }
1192                                 delete token;
1193                         }
1194                 }
1195         }
1196
1197         fLogbookEntry->Print("all");
1198
1199         // Initialization
1200         Bool_t hasError = kFALSE;
1201
1202         // Set the CDB and Reference folders according to the year and LHC period
1203         TString lhcPeriod(GetLHCPeriod());
1204         if (lhcPeriod.Length() == 0)
1205         {
1206                 Log("SHUTTLE","Process - LHCPeriod not found in logbook!");
1207                 return 0;
1208         }
1209
1210         if (fgkMainCDB.Length() == 0)
1211                 fgkMainCDB = Form("alien://folder=/alice/data/%d/%s/OCDB?user=alidaq?cacheFold=/tmp/OCDBCache",
1212                                         GetCurrentYear(), lhcPeriod.Data());
1213
1214         if (fgkMainRefStorage.Length() == 0)
1215                 fgkMainRefStorage = Form("alien://folder=/alice/data/%d/%s/Reference?user=alidaq?cacheFold=/tmp/OCDBCache",
1216                                         GetCurrentYear(), lhcPeriod.Data());
1217
1218         // Loop on detectors in the configuration
1219         TIter iter(fConfig->GetDetectors());
1220         TObjString* aDetector = 0;
1221
1222         Bool_t first = kTRUE;
1223
1224         while ((aDetector = (TObjString*) iter.Next()))
1225         {
1226                 fCurrentDetector = aDetector->String();
1227
1228                 if (ContinueProcessing() == kFALSE) continue;
1229
1230                 if (first)
1231                 {
1232                   // only read QueryCDB when needed and only once
1233                   AliCDBStorage *mainCDBSto = AliCDBManager::Instance()->GetStorage(fgkMainCDB);
1234                   if(mainCDBSto) mainCDBSto->QueryCDB(GetCurrentRun());
1235                   AliCDBStorage *mainRefSto = AliCDBManager::Instance()->GetStorage(fgkMainRefStorage);
1236                   if(mainRefSto) mainRefSto->QueryCDB(GetCurrentRun());
1237                   first = kFALSE;
1238                 }
1239
1240                 Log("SHUTTLE", Form("\t\t\t****** run %d - %s: START  ******",
1241                                                 GetCurrentRun(), aDetector->GetName()));
1242
1243                 for(Int_t iSys=0;iSys<3;iSys++) fFXSCalled[iSys]=kFALSE;
1244
1245                 Log(fCurrentDetector.Data(), "Process - Starting processing");
1246
1247                 Int_t pid = fork();
1248
1249                 if (pid < 0)
1250                 {
1251                         Log("SHUTTLE", "Process - ERROR: Forking failed");
1252                 }
1253                 else if (pid > 0)
1254                 {
1255                         // parent
1256                         Log("SHUTTLE", Form("Process - In parent process of %d - %s: Starting monitoring",
1257                                                         GetCurrentRun(), aDetector->GetName()));
1258
1259                         Long_t begin = time(0);
1260
1261                         int status; // to be used with waitpid, on purpose an int (not Int_t)!
1262                         while (waitpid(pid, &status, WNOHANG) == 0)
1263                         {
1264                                 Long_t expiredTime = time(0) - begin;
1265
1266                                 if (expiredTime > fConfig->GetPPTimeOut())
1267                                 {
1268                                         TString tmp;
1269                                         tmp.Form("Process - Process of %s time out. "
1270                                                         "Run time: %d seconds. Killing...",
1271                                                         fCurrentDetector.Data(), expiredTime);
1272                                         Log("SHUTTLE", tmp);
1273                                         Log(fCurrentDetector, tmp);
1274
1275                                         kill(pid, 9);
1276
1277                                         UpdateShuttleStatus(AliShuttleStatus::kPPTimeOut);
1278                                         hasError = kTRUE;
1279
1280                                         gSystem->Sleep(1000);
1281                                 }
1282                                 else
1283                                 {
1284                                         gSystem->Sleep(1000);
1285
1286                                         TString checkStr;
1287                                         checkStr.Form("ps -o vsize --pid %d | tail -n 1", pid);
1288                                         FILE* pipe = gSystem->OpenPipe(checkStr, "r");
1289                                         if (!pipe)
1290                                         {
1291                                                 Log("SHUTTLE", Form("Process - Error: "
1292                                                         "Could not open pipe to %s", checkStr.Data()));
1293                                                 continue;
1294                                         }
1295
1296                                         char buffer[100];
1297                                         if (!fgets(buffer, 100, pipe))
1298                                         {
1299                                                 Log("SHUTTLE", "Process - Error: ps did not return anything");
1300                                                 gSystem->ClosePipe(pipe);
1301                                                 continue;
1302                                         }
1303                                         gSystem->ClosePipe(pipe);
1304
1305                                         //Log("SHUTTLE", Form("ps returned %s", buffer));
1306
1307                                         Int_t mem = 0;
1308                                         if ((sscanf(buffer, "%d\n", &mem) != 1) || !mem)
1309                                         {
1310                                                 Log("SHUTTLE", "Process - Error: Could not parse output of ps");
1311                                                 continue;
1312                                         }
1313
1314                                         if (expiredTime % 60 == 0)
1315                                         {
1316                                                 Log("SHUTTLE", Form("Process - %s: Checking process. "
1317                                                         "Run time: %d seconds - Memory consumption: %d KB",
1318                                                         fCurrentDetector.Data(), expiredTime, mem));
1319                                                 SendAlive();
1320                                         }
1321
1322                                         if (mem > fConfig->GetPPMaxMem())
1323                                         {
1324                                                 TString tmp;
1325                                                 tmp.Form("Process - Process exceeds maximum allowed memory "
1326                                                         "(%d KB > %d KB). Killing...",
1327                                                         mem, fConfig->GetPPMaxMem());
1328                                                 Log("SHUTTLE", tmp);
1329                                                 Log(fCurrentDetector, tmp);
1330
1331                                                 kill(pid, 9);
1332
1333                                                 UpdateShuttleStatus(AliShuttleStatus::kPPOutOfMemory);
1334                                                 hasError = kTRUE;
1335
1336                                                 gSystem->Sleep(1000);
1337                                         }
1338                                 }
1339                         }
1340
1341                         Log("SHUTTLE", Form("Process - In parent process of %d - %s: Client has terminated.",
1342                                                                 GetCurrentRun(), aDetector->GetName()));
1343
1344                         if (WIFEXITED(status))
1345                         {
1346                                 Int_t returnCode = WEXITSTATUS(status);
1347
1348                                 Log("SHUTTLE", Form("Process - %s: the return code is %d", fCurrentDetector.Data(),
1349                                                                                 returnCode));
1350
1351                                 if (returnCode == 0) hasError = kTRUE;
1352                         }
1353                 }
1354                 else if (pid == 0)
1355                 {
1356                         // client
1357                         Log("SHUTTLE", Form("Process - In client process of %d - %s", GetCurrentRun(),
1358                                 aDetector->GetName()));
1359
1360                         Log("SHUTTLE", Form("Process - Redirecting output to %s log",fCurrentDetector.Data()));
1361
1362                         if ((freopen(GetLogFileName(fCurrentDetector), "a", stdout)) == 0)
1363                         {
1364                                 Log("SHUTTLE", "Process - Could not freopen stdout");
1365                         }
1366                         else
1367                         {
1368                                 fOutputRedirected = kTRUE;
1369                                 if ((dup2(fileno(stdout), fileno(stderr))) < 0)
1370                                         Log("SHUTTLE", "Process - Could not redirect stderr");
1371
1372                         }
1373
1374                         TString wd = gSystem->WorkingDirectory();
1375                         TString tmpDir = Form("%s/%s_%d_process", GetShuttleTempDir(),
1376                                 fCurrentDetector.Data(), GetCurrentRun());
1377
1378                         Int_t result = gSystem->GetPathInfo(tmpDir.Data(), 0, (Long64_t*) 0, 0, 0);
1379                         if (!result) // temp dir already exists!
1380                         {
1381                                 Log(fCurrentDetector.Data(),
1382                                         Form("Process - %s dir already exists! Removing...", tmpDir.Data()));
1383                                 gSystem->Exec(Form("rm -rf %s",tmpDir.Data()));
1384                         }
1385
1386                         if (gSystem->mkdir(tmpDir.Data(), 1))
1387                         {
1388                                 Log(fCurrentDetector.Data(), "Process - could not make temp directory!!");
1389                                 gSystem->Exit(1);
1390                         }
1391
1392                         if (!gSystem->ChangeDirectory(tmpDir.Data()))
1393                         {
1394                                 Log(fCurrentDetector.Data(), "Process - could not change directory!!");
1395                                 gSystem->Exit(1);
1396                         }
1397
1398                         Bool_t success = ProcessCurrentDetector();
1399
1400                         gSystem->ChangeDirectory(wd.Data());
1401
1402                         if (success) // Preprocessor finished successfully!
1403                         {
1404                                 // remove temporary folder or DCS map
1405                                 if (!fConfig->KeepTempFolder())
1406                                 {
1407                                         gSystem->Exec(Form("rm -rf %s",tmpDir.Data()));
1408                                 } else if (!fConfig->KeepDCSMap())
1409                                 {
1410                                         gSystem->Exec(Form("rm -f %s/DCSMap.root",tmpDir.Data()));
1411                                 }
1412
1413                                 // Update time_processed field in FXS DB
1414                                 if (UpdateTable() == kFALSE)
1415                                         Log("SHUTTLE", Form("Process - %s: Could not update FXS databases!",
1416                                                         fCurrentDetector.Data()));
1417
1418                                 // Transfer the data from local storage to main storage (Grid)
1419                                 UpdateShuttleStatus(AliShuttleStatus::kStoreStarted);
1420                                 if (StoreOCDB() == kFALSE)
1421                                 {
1422                                         Log("SHUTTLE",
1423                                                 Form("\t\t\t****** run %d - %s: STORAGE ERROR ******",
1424                                                         GetCurrentRun(), aDetector->GetName()));
1425                                         UpdateShuttleStatus(AliShuttleStatus::kStoreError);
1426                                         success = kFALSE;
1427                                 } else {
1428                                         Log("SHUTTLE",
1429                                                 Form("\t\t\t****** run %d - %s: DONE ******",
1430                                                         GetCurrentRun(), aDetector->GetName()));
1431                                         UpdateShuttleStatus(AliShuttleStatus::kDone);
1432                                         UpdateShuttleLogbook(fCurrentDetector, "DONE");
1433                                 }
1434                         } else
1435                         {
1436                                 Log("SHUTTLE",
1437                                         Form("\t\t\t****** run %d - %s: PP ERROR ******",
1438                                                 GetCurrentRun(), aDetector->GetName()));
1439                         }
1440
1441                         for (UInt_t iSys=0; iSys<3; iSys++)
1442                         {
1443                                 if (fFXSCalled[iSys]) fFXSlist[iSys].Clear();
1444                         }
1445
1446                         Log("SHUTTLE", Form("Process - Client process of %d - %s is exiting now with %d.",
1447                                                         GetCurrentRun(), aDetector->GetName(), success));
1448
1449                         // the client exits here
1450                         gSystem->Exit(success);
1451
1452                         AliError("We should never get here!!!");
1453                 }
1454         }
1455
1456         Log("SHUTTLE", Form("\t\t\t^*^*^*^*^*^*^*^*^*^*^*^* run %d: FINISH ^*^*^*^*^*^*^*^*^*^*^*^*",
1457                                                         GetCurrentRun()));
1458
1459         //check if shuttle is done for this run, if so update logbook
1460         TObjArray checkEntryArray;
1461         checkEntryArray.SetOwner(1);
1462         TString whereClause = Form("where run=%d", GetCurrentRun());
1463         if (!QueryShuttleLogbook(whereClause.Data(), checkEntryArray) ||
1464                         checkEntryArray.GetEntries() == 0) {
1465                 Log("SHUTTLE", Form("Process - Warning: Cannot check status of run %d on Shuttle logbook!",
1466                                                 GetCurrentRun()));
1467                 return hasError == kFALSE;
1468         }
1469
1470         AliShuttleLogbookEntry* checkEntry = dynamic_cast<AliShuttleLogbookEntry*>
1471                                                 (checkEntryArray.At(0));
1472
1473         if (checkEntry)
1474         {
1475                 if (checkEntry->IsDone())
1476                 {
1477                         Log("SHUTTLE","Process - Shuttle is DONE. Updating logbook");
1478                         UpdateShuttleLogbook("shuttle_done");
1479                 }
1480                 else
1481                 {
1482                         for (UInt_t iDet=0; iDet<NDetectors(); iDet++)
1483                         {
1484                                 if (checkEntry->GetDetectorStatus(iDet) == AliShuttleLogbookEntry::kUnprocessed)
1485                                 {
1486                                         AliDebug(2, Form("Run %d: setting %s as \"not first time unprocessed\"",
1487                                                         checkEntry->GetRun(), GetDetName(iDet)));
1488                                         fFirstUnprocessed[iDet] = kFALSE;
1489                                 }
1490                         }
1491                 }
1492         }
1493
1494         fLogbookEntry = 0;
1495
1496         return hasError == kFALSE;
1497 }
1498
1499 //______________________________________________________________________________________________
1500 Bool_t AliShuttle::ProcessCurrentDetector()
1501 {
1502         //
1503         // Makes data retrieval just for a specific detector (fCurrentDetector).
1504         // Threre should be a configuration for this detector.
1505
1506         Log("SHUTTLE", Form("ProcessCurrentDetector - Retrieving values for %s, run %d",
1507                                                 fCurrentDetector.Data(), GetCurrentRun()));
1508
1509         TString wd = gSystem->WorkingDirectory();
1510
1511         if (!CleanReferenceStorage(fCurrentDetector.Data()))
1512                 return kFALSE;
1513
1514         gSystem->ChangeDirectory(wd.Data());
1515
1516         TMap* dcsMap = new TMap();
1517
1518         // call preprocessor
1519         AliPreprocessor* aPreprocessor =
1520                 dynamic_cast<AliPreprocessor*> (fPreprocessorMap.GetValue(fCurrentDetector));
1521
1522         aPreprocessor->Initialize(GetCurrentRun(), GetCurrentStartTime(), GetCurrentEndTime());
1523
1524         Bool_t processDCS = aPreprocessor->ProcessDCS();
1525
1526         if (!processDCS)
1527         {
1528                 Log(fCurrentDetector, "ProcessCurrentDetector -"
1529                         " The preprocessor requested to skip the retrieval of DCS values");
1530         }
1531         else if (fTestMode & kSkipDCS)
1532         {
1533                 Log(fCurrentDetector, "ProcessCurrentDetector - In TESTMODE: Skipping DCS processing");
1534         }
1535         else if (fTestMode & kErrorDCS)
1536         {
1537                 Log(fCurrentDetector, "ProcessCurrentDetector - In TESTMODE: Simulating DCS error");
1538                 UpdateShuttleStatus(AliShuttleStatus::kDCSStarted);
1539                 UpdateShuttleStatus(AliShuttleStatus::kDCSError);
1540                 delete dcsMap;
1541                 return kFALSE;
1542         } else {
1543
1544                 UpdateShuttleStatus(AliShuttleStatus::kDCSStarted);
1545
1546                 // Query DCS archive
1547                 Int_t nServers = fConfig->GetNServers(fCurrentDetector);
1548
1549                 for (int iServ=0; iServ<nServers; iServ++)
1550                 {
1551
1552                         TString host(fConfig->GetDCSHost(fCurrentDetector, iServ));
1553                         Int_t port = fConfig->GetDCSPort(fCurrentDetector, iServ);
1554                         Int_t multiSplit = fConfig->GetMultiSplit(fCurrentDetector, iServ);
1555
1556                         Log(fCurrentDetector, Form("ProcessCurrentDetector -"
1557                                         " Querying DCS Amanda server %s:%d (%d of %d)",
1558                                         host.Data(), port, iServ+1, nServers));
1559
1560                         TMap* aliasMap = 0;
1561                         TMap* dpMap = 0;
1562
1563                         if (fConfig->GetDCSAliases(fCurrentDetector, iServ)->GetEntries() > 0)
1564                         {
1565                                 aliasMap = GetValueSet(host, port,
1566                                                 fConfig->GetDCSAliases(fCurrentDetector, iServ),
1567                                                 kAlias, multiSplit);
1568                                 if (!aliasMap)
1569                                 {
1570                                         Log(fCurrentDetector,
1571                                                 Form("ProcessCurrentDetector -"
1572                                                         " Error retrieving DCS aliases from server %s."
1573                                                         " Sending mail to DCS experts!", host.Data()));
1574                                         UpdateShuttleStatus(AliShuttleStatus::kDCSError);
1575
1576                                         if (!SendMailToDCS())
1577                                                 Log("SHUTTLE", Form("ProcessCurrentDetector - "
1578                                                         "Could not send mail to DCS experts!"));
1579
1580                                         delete dcsMap;
1581                                         return kFALSE;
1582                                 }
1583                         }
1584
1585                         if (fConfig->GetDCSDataPoints(fCurrentDetector, iServ)->GetEntries() > 0)
1586                         {
1587                                 dpMap = GetValueSet(host, port,
1588                                                 fConfig->GetDCSDataPoints(fCurrentDetector, iServ),
1589                                                 kDP, multiSplit);
1590                                 if (!dpMap)
1591                                 {
1592                                         Log(fCurrentDetector,
1593                                                 Form("ProcessCurrentDetector -"
1594                                                         " Error retrieving DCS data points from server %s."
1595                                                         " Sending mail to DCS experts!", host.Data()));
1596                                         UpdateShuttleStatus(AliShuttleStatus::kDCSError);
1597
1598                                         if (!SendMailToDCS())
1599                                                 Log("SHUTTLE", Form("ProcessCurrentDetector - "
1600                                                         "Could not send mail to DCS experts!"));
1601
1602                                         if (aliasMap) delete aliasMap;
1603                                         delete dcsMap;
1604                                         return kFALSE;
1605                                 }
1606                         }
1607
1608                         // merge aliasMap and dpMap into dcsMap
1609                         if(aliasMap) {
1610                                 TIter iter(aliasMap);
1611                                 TObjString* key = 0;
1612                                 while ((key = (TObjString*) iter.Next()))
1613                                         dcsMap->Add(key, aliasMap->GetValue(key->String()));
1614
1615                                 aliasMap->SetOwner(kFALSE);
1616                                 delete aliasMap;
1617                         }
1618
1619                         if(dpMap) {
1620                                 TIter iter(dpMap);
1621                                 TObjString* key = 0;
1622                                 while ((key = (TObjString*) iter.Next()))
1623                                         dcsMap->Add(key, dpMap->GetValue(key->String()));
1624
1625                                 dpMap->SetOwner(kFALSE);
1626                                 delete dpMap;
1627                         }
1628                 }
1629         }
1630
1631         // save map into file, to help debugging in case of preprocessor error
1632         TFile* f = TFile::Open("DCSMap.root","recreate");
1633         f->cd();
1634         dcsMap->Write("DCSMap", TObject::kSingleKey);
1635         f->Close();
1636         delete f;
1637
1638         // DCS Archive DB processing successful. Call Preprocessor!
1639         UpdateShuttleStatus(AliShuttleStatus::kPPStarted);
1640
1641         UInt_t returnValue = aPreprocessor->Process(dcsMap);
1642
1643         if (returnValue > 0) // Preprocessor error!
1644         {
1645                 Log(fCurrentDetector, Form("ProcessCurrentDetector - "
1646                                 "Preprocessor failed. Process returned %d.", returnValue));
1647                 UpdateShuttleStatus(AliShuttleStatus::kPPError);
1648                 dcsMap->DeleteAll();
1649                 delete dcsMap;
1650                 return kFALSE;
1651         }
1652
1653         // preprocessor ok!
1654         UpdateShuttleStatus(AliShuttleStatus::kPPDone);
1655         Log(fCurrentDetector, Form("ProcessCurrentDetector - %s preprocessor returned success",
1656                                 fCurrentDetector.Data()));
1657
1658         dcsMap->DeleteAll();
1659         delete dcsMap;
1660
1661         return kTRUE;
1662 }
1663
1664 //______________________________________________________________________________________________
1665 void AliShuttle::CountOpenRuns()
1666 {
1667         // Query DAQ's Shuttle logbook and sends the number of open runs to ML
1668
1669         // check connection, in case connect
1670         if (!Connect(3))
1671                 return;
1672
1673         TString sqlQuery;
1674         sqlQuery = Form("select count(*) from %s where shuttle_done=0", fConfig->GetShuttlelbTable());
1675
1676         TSQLResult* aResult = fServer[3]->Query(sqlQuery);
1677         if (!aResult) {
1678                 AliError(Form("Can't execute query <%s>!", sqlQuery.Data()));
1679                 return;
1680         }
1681
1682         AliDebug(2,Form("Query = %s", sqlQuery.Data()));
1683
1684         if (aResult->GetRowCount() == 0) {
1685                 AliError(Form("No result for query %s received", sqlQuery.Data()));
1686                 return;
1687         }
1688
1689         if (aResult->GetFieldCount() != 1) {
1690                 AliError(Form("Invalid field count for query %s received", sqlQuery.Data()));
1691                 return;
1692         }
1693
1694         TSQLRow* aRow = aResult->Next();
1695         if (!aRow) {
1696                 AliError(Form("Could not receive result of query %s", sqlQuery.Data()));
1697                 return;
1698         }
1699
1700         TString result(aRow->GetField(0), aRow->GetFieldLength(0));
1701         Int_t count = result.Atoi();
1702
1703         Log("SHUTTLE", Form("%d unprocessed runs", count));
1704
1705         delete aRow;
1706         delete aResult;
1707
1708         TMonaLisaValue mlStatus("SHUTTLE_openruns", count);
1709
1710         TList mlList;
1711         mlList.Add(&mlStatus);
1712
1713         fMonaLisa->SendParameters(&mlList, "__PROCESSINGINFO__");
1714 }
1715
1716 //______________________________________________________________________________________________
1717 Bool_t AliShuttle::QueryShuttleLogbook(const char* whereClause,
1718                 TObjArray& entries)
1719 {
1720         // Query DAQ's Shuttle logbook and fills detector status object.
1721         // Call QueryRunParameters to query DAQ logbook for run parameters.
1722         //
1723
1724         entries.SetOwner(1);
1725
1726         // check connection, in case connect
1727         if (!Connect(3)) return kFALSE;
1728
1729         TString sqlQuery;
1730         sqlQuery = Form("select * from %s %s order by run", fConfig->GetShuttlelbTable(), whereClause);
1731
1732         TSQLResult* aResult = fServer[3]->Query(sqlQuery);
1733         if (!aResult) {
1734                 AliError(Form("Can't execute query <%s>!", sqlQuery.Data()));
1735                 return kFALSE;
1736         }
1737
1738         AliDebug(2,Form("Query = %s", sqlQuery.Data()));
1739
1740         if(aResult->GetRowCount() == 0) {
1741                 Log("SHUTTLE", "No entries in Shuttle Logbook match request");
1742                 delete aResult;
1743                 return kTRUE;
1744         }
1745
1746         // TODO Check field count!
1747         const UInt_t nCols = 23;
1748         if (aResult->GetFieldCount() != (Int_t) nCols) {
1749                 Log("SHUTTLE", "Invalid SQL result field number!");
1750                 delete aResult;
1751                 return kFALSE;
1752         }
1753
1754         TSQLRow* aRow;
1755         while ((aRow = aResult->Next())) {
1756                 TString runString(aRow->GetField(0), aRow->GetFieldLength(0));
1757                 Int_t run = runString.Atoi();
1758
1759                 AliShuttleLogbookEntry *entry = QueryRunParameters(run);
1760                 if (!entry)
1761                         continue;
1762
1763                 // loop on detectors
1764                 for(UInt_t ii = 0; ii < nCols; ii++)
1765                         entry->SetDetectorStatus(aResult->GetFieldName(ii), aRow->GetField(ii));
1766
1767                 entries.AddLast(entry);
1768                 delete aRow;
1769         }
1770
1771         delete aResult;
1772         return kTRUE;
1773 }
1774
1775 //______________________________________________________________________________________________
1776 AliShuttleLogbookEntry* AliShuttle::QueryRunParameters(Int_t run)
1777 {
1778         //
1779         // Retrieve run parameters written in the DAQ logbook and sets them into AliShuttleLogbookEntry object
1780         //
1781
1782         // check connection, in case connect
1783         if (!Connect(3))
1784                 return 0;
1785
1786         TString sqlQuery;
1787         sqlQuery.Form("select * from %s where run=%d", fConfig->GetDAQlbTable(), run);
1788
1789         TSQLResult* aResult = fServer[3]->Query(sqlQuery);
1790         if (!aResult) {
1791                 Log("SHUTTLE", Form("Can't execute query <%s>!", sqlQuery.Data()));
1792                 return 0;
1793         }
1794
1795         if (aResult->GetRowCount() == 0) {
1796                 Log("SHUTTLE", Form("QueryRunParameters - No entry in DAQ Logbook for run %d. Skipping", run));
1797                 delete aResult;
1798                 return 0;
1799         }
1800
1801         if (aResult->GetRowCount() > 1) {
1802                 Log("SHUTTLE", Form("QueryRunParameters - UNEXPECTED: "
1803                                 "more than one entry in DAQ Logbook for run %d!", run));
1804                 delete aResult;
1805                 return 0;
1806         }
1807
1808         TSQLRow* aRow = aResult->Next();
1809         if (!aRow)
1810         {
1811                 Log("SHUTTLE", Form("QueryRunParameters - Could not retrieve row for run %d. Skipping", run));
1812                 delete aResult;
1813                 return 0;
1814         }
1815
1816         AliShuttleLogbookEntry* entry = new AliShuttleLogbookEntry(run);
1817
1818         for (Int_t ii = 0; ii < aResult->GetFieldCount(); ii++)
1819                 entry->SetRunParameter(aResult->GetFieldName(ii), aRow->GetField(ii));
1820
1821         UInt_t startTime = entry->GetStartTime();
1822         UInt_t endTime = entry->GetEndTime();
1823
1824 //      if (!startTime || !endTime || startTime > endTime)
1825 //      {
1826 //              Log("SHUTTLE",
1827 //                      Form("QueryRunParameters - Invalid parameters for Run %d: startTime = %d, endTime = %d. Skipping!",
1828 //                              run, startTime, endTime));
1829 //
1830 //              Log("SHUTTLE", Form("Marking SHUTTLE done for run %d", run));
1831 //              fLogbookEntry = entry;
1832 //              if (!UpdateShuttleLogbook("shuttle_done"))
1833 //              {
1834 //                      AliError(Form("Could not update logbook for run %d !", run));
1835 //              }
1836 //              fLogbookEntry = 0;
1837 //
1838 //              delete entry;
1839 //              delete aRow;
1840 //              delete aResult;
1841 //              return 0;
1842 //      }
1843
1844         if (!startTime)
1845         {
1846                 Log("SHUTTLE",
1847                         Form("QueryRunParameters - Invalid parameters for Run %d: "
1848                                 "startTime = %d, endTime = %d. Skipping!",
1849                                         run, startTime, endTime));
1850
1851                 Log("SHUTTLE", Form("Marking SHUTTLE done for run %d", run));
1852                 fLogbookEntry = entry;
1853                 if (!UpdateShuttleLogbook("shuttle_ignored"))
1854                 {
1855                         AliError(Form("Could not update logbook for run %d !", run));
1856                 }
1857                 fLogbookEntry = 0;
1858
1859                 delete entry;
1860                 delete aRow;
1861                 delete aResult;
1862                 return 0;
1863         }
1864
1865         if (startTime && !endTime)
1866         {
1867                 // TODO Here we don't mark SHUTTLE done, because this may mean
1868                 //the run is still ongoing!!
1869                 Log("SHUTTLE",
1870                         Form("QueryRunParameters - Invalid parameters for Run %d: "
1871                              "startTime = %d, endTime = %d. Skipping (Shuttle won't be marked as DONE)!",
1872                                         run, startTime, endTime));
1873
1874                 //Log("SHUTTLE", Form("Marking SHUTTLE done for run %d", run));
1875                 //fLogbookEntry = entry;
1876                 //if (!UpdateShuttleLogbook("shuttle_done"))
1877                 //{
1878                 //      AliError(Form("Could not update logbook for run %d !", run));
1879                 //}
1880                 //fLogbookEntry = 0;
1881
1882                 delete entry;
1883                 delete aRow;
1884                 delete aResult;
1885                 return 0;
1886         }
1887
1888         if (startTime && endTime && (startTime > endTime))
1889         {
1890                 Log("SHUTTLE",
1891                         Form("QueryRunParameters - Invalid parameters for Run %d: "
1892                                 "startTime = %d, endTime = %d. Skipping!",
1893                                         run, startTime, endTime));
1894
1895                 Log("SHUTTLE", Form("Marking SHUTTLE done for run %d", run));
1896                 fLogbookEntry = entry;
1897                 if (!UpdateShuttleLogbook("shuttle_ignored"))
1898                 {
1899                         AliError(Form("Could not update logbook for run %d !", run));
1900                 }
1901                 fLogbookEntry = 0;
1902
1903                 delete entry;
1904                 delete aRow;
1905                 delete aResult;
1906                 return 0;
1907         }
1908
1909         TString totEventsStr = entry->GetRunParameter("totalEvents");
1910         Int_t totEvents = totEventsStr.Atoi();
1911         if (totEvents < 1)
1912         {
1913                 Log("SHUTTLE",
1914                         Form("QueryRunParameters - Run %d has 0 events - Skipping!", run));
1915
1916                 Log("SHUTTLE", Form("Marking SHUTTLE done for run %d", run));
1917                 fLogbookEntry = entry;
1918                 if (!UpdateShuttleLogbook("shuttle_ignored"))
1919                 {
1920                         AliError(Form("Could not update logbook for run %d !", run));
1921                 }
1922                 fLogbookEntry = 0;
1923
1924                 delete entry;
1925                 delete aRow;
1926                 delete aResult;
1927                 return 0;
1928         }
1929
1930         delete aRow;
1931         delete aResult;
1932
1933         return entry;
1934 }
1935
1936 //______________________________________________________________________________________________
1937 TMap* AliShuttle::GetValueSet(const char* host, Int_t port, const TSeqCollection* entries,
1938                               DCSType type, Int_t multiSplit)
1939 {
1940         // Retrieve all "entry" data points from the DCS server
1941         // host, port: TSocket connection parameters
1942         // entries: list of name of the alias or data point
1943         // type: kAlias or kDP
1944         // returns TMap of values, 0 when failure
1945
1946         AliDCSClient client(host, port, fTimeout, fRetries, multiSplit);
1947
1948         TMap* result = 0;
1949         if (type == kAlias)
1950         {
1951                 result = client.GetAliasValues(entries, GetCurrentStartTime(),
1952                         GetCurrentEndTime());
1953         }
1954         else if (type == kDP)
1955         {
1956                 result = client.GetDPValues(entries, GetCurrentStartTime(),
1957                         GetCurrentEndTime());
1958         }
1959
1960         if (result == 0)
1961         {
1962                 Log(fCurrentDetector.Data(), Form("GetValueSet - Can't get entries! Reason: %s",
1963                         client.GetErrorString(client.GetResultErrorCode())));
1964                 if (client.GetResultErrorCode() == AliDCSClient::fgkServerError)
1965                         Log(fCurrentDetector.Data(), Form("GetValueSet - Server error code: %s",
1966                                 client.GetServerError().Data()));
1967
1968                 return 0;
1969         }
1970
1971         return result;
1972 }
1973
1974 //______________________________________________________________________________________________
1975 const char* AliShuttle::GetFile(Int_t system, const char* detector,
1976                 const char* id, const char* source)
1977 {
1978         // Get calibration file from file exchange servers
1979         // First queris the FXS database for the file name, using the run, detector, id and source info
1980         // then calls RetrieveFile(filename) for actual copy to local disk
1981         // run: current run being processed (given by Logbook entry fLogbookEntry)
1982         // detector: the Preprocessor name
1983         // id: provided as a parameter by the Preprocessor
1984         // source: provided by the Preprocessor through GetFileSources function
1985
1986         // check if test mode should simulate a FXS error
1987         if (fTestMode & kErrorFXSFiles)
1988         {
1989                 Log(detector, Form("GetFile - In TESTMODE - Simulating error while connecting to %s FXS", GetSystemName(system)));
1990                 return 0;
1991         }
1992
1993         // check connection, in case connect
1994         if (!Connect(system))
1995         {
1996                 Log(detector, Form("GetFile - Couldn't connect to %s FXS database", GetSystemName(system)));
1997                 return 0;
1998         }
1999
2000         // Query preparation
2001         TString sourceName(source);
2002         Int_t nFields = 3;
2003         TString sqlQueryStart = Form("select filePath,size,fileChecksum from %s where",
2004                                                                 fConfig->GetFXSdbTable(system));
2005         TString whereClause = Form("run=%d and detector=\"%s\" and fileId=\"%s\"",
2006                                                                 GetCurrentRun(), detector, id);
2007
2008         if (system == kDAQ)
2009         {
2010                 whereClause += Form(" and DAQsource=\"%s\"", source);
2011         }
2012         else if (system == kDCS)
2013         {
2014                 sourceName="none";
2015         }
2016         else if (system == kHLT)
2017         {
2018                 whereClause += Form(" and DDLnumbers=\"%s\"", source);
2019                 nFields = 3;
2020         }
2021
2022         TString sqlQuery = Form("%s %s", sqlQueryStart.Data(), whereClause.Data());
2023
2024         AliDebug(2, Form("SQL query: \n%s",sqlQuery.Data()));
2025
2026         // Query execution
2027         TSQLResult* aResult = 0;
2028         aResult = dynamic_cast<TSQLResult*> (fServer[system]->Query(sqlQuery));
2029         if (!aResult) {
2030                 Log(detector, Form("GetFileName - Can't execute SQL query to %s database for: id = %s, source = %s",
2031                                 GetSystemName(system), id, sourceName.Data()));
2032                 return 0;
2033         }
2034
2035         if(aResult->GetRowCount() == 0)
2036         {
2037                 Log(detector,
2038                         Form("GetFileName - No entry in %s FXS db for: id = %s, source = %s",
2039                                 GetSystemName(system), id, sourceName.Data()));
2040                 delete aResult;
2041                 return 0;
2042         }
2043
2044         if (aResult->GetRowCount() > 1) {
2045                 Log(detector,
2046                         Form("GetFileName - More than one entry in %s FXS db for: id = %s, source = %s",
2047                                 GetSystemName(system), id, sourceName.Data()));
2048                 delete aResult;
2049                 return 0;
2050         }
2051
2052         if (aResult->GetFieldCount() != nFields) {
2053                 Log(detector,
2054                         Form("GetFileName - Wrong field count in %s FXS db for: id = %s, source = %s",
2055                                 GetSystemName(system), id, sourceName.Data()));
2056                 delete aResult;
2057                 return 0;
2058         }
2059
2060         TSQLRow* aRow = dynamic_cast<TSQLRow*> (aResult->Next());
2061
2062         if (!aRow){
2063                 Log(detector, Form("GetFileName - Empty set result in %s FXS db from query: id = %s, source = %s",
2064                                 GetSystemName(system), id, sourceName.Data()));
2065                 delete aResult;
2066                 return 0;
2067         }
2068
2069         TString filePath(aRow->GetField(0), aRow->GetFieldLength(0));
2070         TString fileSize(aRow->GetField(1), aRow->GetFieldLength(1));
2071         TString fileChecksum(aRow->GetField(2), aRow->GetFieldLength(2));
2072
2073         delete aResult;
2074         delete aRow;
2075
2076         AliDebug(2, Form("filePath = %s; size = %s, fileChecksum = %s",
2077                                 filePath.Data(), fileSize.Data(), fileChecksum.Data()));
2078
2079         // retrieved file is renamed to make it unique
2080         TString localFileName = Form("%s/%s_%d_process/%s_%s_%d_%s_%s.shuttle",
2081                                         GetShuttleTempDir(), detector, GetCurrentRun(),
2082                                         GetSystemName(system), detector, GetCurrentRun(),
2083                                         id, sourceName.Data());
2084
2085
2086         // file retrieval from FXS
2087         UInt_t nRetries = 0;
2088         UInt_t maxRetries = 3;
2089         Bool_t result = kFALSE;
2090
2091         // copy!! if successful TSystem::Exec returns 0
2092         while(nRetries++ < maxRetries) {
2093                 AliDebug(2, Form("Trying to copy file. Retry # %d", nRetries));
2094                 result = RetrieveFile(system, filePath.Data(), localFileName.Data());
2095                 if(!result)
2096                 {
2097                         Log(detector, Form("GetFileName - Copy of file %s from %s FXS failed",
2098                                         filePath.Data(), GetSystemName(system)));
2099                         continue;
2100                 }
2101
2102                 if (fileSize.Length()>0)
2103                 {
2104                         // compare filesize of local file with the one stored in the FXS DB
2105                         TString command=("stat --format=%s");
2106                         Int_t sizeComp = gSystem->Exec(Form("%s %s |grep %s 2>&1 > /dev/null",
2107                                                 command.Data(), localFileName.Data(),fileSize.Data()));
2108
2109                         if ( sizeComp != 0)
2110                         {
2111                                 Log(detector, Form("GetFileName - size of file %s does not match with local copy!",
2112                                                         filePath.Data()));
2113                                 result = kFALSE;
2114                                 continue;
2115                         }
2116
2117                 } else {
2118                         Log(fCurrentDetector, Form("GetFile - size of file %s not set in %s database, skipping comparison",
2119                                                 filePath.Data(), GetSystemName(system)));
2120                 }
2121
2122                 if (fileChecksum.Length()>0)
2123                 {
2124                         // compare md5sum of local file with the one stored in the FXS DB
2125                         Int_t md5Comp = gSystem->Exec(Form("md5sum %s |grep %s 2>&1 > /dev/null",
2126                                                 localFileName.Data(), fileChecksum.Data()));
2127
2128                         if (md5Comp != 0)
2129                         {
2130                                 Log(detector, Form("GetFileName - md5sum of file %s does not match with local copy!",
2131                                                         filePath.Data()));
2132                                 result = kFALSE;
2133                                 continue;
2134                         }
2135                 } else {
2136                         Log(fCurrentDetector, Form("GetFile - md5sum of file %s not set in %s database, skipping comparison",
2137                                                         filePath.Data(), GetSystemName(system)));
2138                 }
2139                 if (result) break;
2140         }
2141
2142         if(!result) return 0;
2143
2144         fFXSCalled[system]=kTRUE;
2145         TObjString *fileParams = new TObjString(Form("%s#!?!#%s", id, sourceName.Data()));
2146         fFXSlist[system].Add(fileParams);
2147
2148         static TString staticLocalFileName;
2149         staticLocalFileName.Form("%s", localFileName.Data());
2150
2151         Log(fCurrentDetector, Form("GetFile - Retrieved file with id %s and "
2152                         "source %s from %s to %s", id, source,
2153                         GetSystemName(system), localFileName.Data()));
2154
2155         return staticLocalFileName.Data();
2156 }
2157
2158 //______________________________________________________________________________________________
2159 Bool_t AliShuttle::RetrieveFile(UInt_t system, const char* fxsFileName, const char* localFileName)
2160 {
2161         //
2162         // Copies file from FXS to local Shuttle machine
2163         //
2164
2165         // check temp directory: trying to cd to temp; if it does not exist, create it
2166         AliDebug(2, Form("Copy file %s from %s FXS into %s",
2167                         GetSystemName(system), fxsFileName, localFileName));
2168
2169         TString tmpDir(localFileName);
2170
2171         tmpDir = tmpDir(0,tmpDir.Last('/'));
2172
2173         Int_t noDir = gSystem->GetPathInfo(tmpDir.Data(), 0, (Long64_t*) 0, 0, 0);
2174         if (noDir) // temp dir does not exists!
2175         {
2176                 if (gSystem->mkdir(tmpDir.Data(), 1))
2177                 {
2178                         Log(fCurrentDetector.Data(), "RetrieveFile - could not make temp directory!!");
2179                         return kFALSE;
2180                 }
2181         }
2182
2183         TString baseFXSFolder;
2184         if (system == kDAQ)
2185         {
2186                 baseFXSFolder = "FES/";
2187         }
2188         else if (system == kDCS)
2189         {
2190                 baseFXSFolder = "";
2191         }
2192         else if (system == kHLT)
2193         {
2194                 baseFXSFolder = "/opt/FXS/";
2195         }
2196
2197
2198         TString command = Form("scp -oPort=%d -2 %s@%s:%s%s %s",
2199                 fConfig->GetFXSPort(system),
2200                 fConfig->GetFXSUser(system),
2201                 fConfig->GetFXSHost(system),
2202                 baseFXSFolder.Data(),
2203                 fxsFileName,
2204                 localFileName);
2205
2206         AliDebug(2, Form("%s",command.Data()));
2207
2208         Bool_t result = (gSystem->Exec(command.Data()) == 0);
2209
2210         return result;
2211 }
2212
2213 //______________________________________________________________________________________________
2214 TList* AliShuttle::GetFileSources(Int_t system, const char* detector, const char* id)
2215 {
2216         //
2217         // Get sources producing the condition file Id from file exchange servers
2218         // if id is NULL all sources are returned (distinct)
2219         //
2220
2221         if (id)
2222         {
2223                 Log(detector, Form("GetFileSources - Querying %s FXS for files with id %s produced by %s", GetSystemName(system), id, detector));
2224         } else {
2225                 Log(detector, Form("GetFileSources - Querying %s FXS for files produced by %s", GetSystemName(system), detector));
2226         }
2227
2228         // check if test mode should simulate a FXS error
2229         if (fTestMode & kErrorFXSSources)
2230         {
2231                 Log(detector, Form("GetFileSources - In TESTMODE - Simulating error while connecting to %s FXS", GetSystemName(system)));
2232                 return 0;
2233         }
2234
2235         if (system == kDCS)
2236         {
2237                 Log(detector, "GetFileSources - WARNING: DCS system has only one source of data!");
2238                 TList *list = new TList();
2239                 list->SetOwner(1);
2240                 list->Add(new TObjString(" "));
2241                 return list;
2242         }
2243
2244         // check connection, in case connect
2245         if (!Connect(system))
2246         {
2247                 Log(detector, Form("GetFileSources - Couldn't connect to %s FXS database", GetSystemName(system)));
2248                 return NULL;
2249         }
2250
2251         TString sourceName = 0;
2252         if (system == kDAQ)
2253         {
2254                 sourceName = "DAQsource";
2255         } else if (system == kHLT)
2256         {
2257                 sourceName = "DDLnumbers";
2258         }
2259
2260         TString sqlQueryStart = Form("select distinct %s from %s where", sourceName.Data(), fConfig->GetFXSdbTable(system));
2261         TString whereClause = Form("run=%d and detector=\"%s\"",
2262                                 GetCurrentRun(), detector);
2263         if (id)
2264                 whereClause += Form(" and fileId=\"%s\"", id);
2265         TString sqlQuery = Form("%s %s", sqlQueryStart.Data(), whereClause.Data());
2266
2267         AliDebug(2, Form("SQL query: \n%s",sqlQuery.Data()));
2268
2269         // Query execution
2270         TSQLResult* aResult;
2271         aResult = fServer[system]->Query(sqlQuery);
2272         if (!aResult) {
2273                 Log(detector, Form("GetFileSources - Can't execute SQL query to %s database for id: %s",
2274                                 GetSystemName(system), id));
2275                 return 0;
2276         }
2277
2278         TList *list = new TList();
2279         list->SetOwner(1);
2280
2281         if (aResult->GetRowCount() == 0)
2282         {
2283                 Log(detector,
2284                         Form("GetFileSources - No entry in %s FXS table for id: %s", GetSystemName(system), id));
2285                 delete aResult;
2286                 return list;
2287         }
2288
2289         Log(detector, Form("GetFileSources - Found %d sources", aResult->GetRowCount()));
2290
2291         TSQLRow* aRow;
2292         while ((aRow = aResult->Next()))
2293         {
2294
2295                 TString source(aRow->GetField(0), aRow->GetFieldLength(0));
2296                 AliDebug(2, Form("%s = %s", sourceName.Data(), source.Data()));
2297                 list->Add(new TObjString(source));
2298                 delete aRow;
2299         }
2300
2301         delete aResult;
2302
2303         return list;
2304 }
2305
2306 //______________________________________________________________________________________________
2307 TList* AliShuttle::GetFileIDs(Int_t system, const char* detector, const char* source)
2308 {
2309         //
2310         // Get all ids of condition files produced by a given source from file exchange servers
2311         //
2312
2313         Log(detector, Form("GetFileIDs - Retrieving ids with source %s with %s", source, GetSystemName(system)));
2314
2315         // check if test mode should simulate a FXS error
2316         if (fTestMode & kErrorFXSSources)
2317         {
2318                 Log(detector, Form("GetFileIDs - In TESTMODE - Simulating error while connecting to %s FXS", GetSystemName(system)));
2319                 return 0;
2320         }
2321
2322         // check connection, in case connect
2323         if (!Connect(system))
2324         {
2325                 Log(detector, Form("GetFileIDs - Couldn't connect to %s FXS database", GetSystemName(system)));
2326                 return NULL;
2327         }
2328
2329         TString sourceName = 0;
2330         if (system == kDAQ)
2331         {
2332                 sourceName = "DAQsource";
2333         } else if (system == kHLT)
2334         {
2335                 sourceName = "DDLnumbers";
2336         }
2337
2338         TString sqlQueryStart = Form("select fileId from %s where", fConfig->GetFXSdbTable(system));
2339         TString whereClause = Form("run=%d and detector=\"%s\"",
2340                                 GetCurrentRun(), detector);
2341         if (sourceName.Length() > 0 && source)
2342                 whereClause += Form(" and %s=\"%s\"", sourceName.Data(), source);
2343         TString sqlQuery = Form("%s %s", sqlQueryStart.Data(), whereClause.Data());
2344
2345         AliDebug(2, Form("SQL query: \n%s",sqlQuery.Data()));
2346
2347         // Query execution
2348         TSQLResult* aResult;
2349         aResult = fServer[system]->Query(sqlQuery);
2350         if (!aResult) {
2351                 Log(detector, Form("GetFileIDs - Can't execute SQL query to %s database for source: %s",
2352                                 GetSystemName(system), source));
2353                 return 0;
2354         }
2355
2356         TList *list = new TList();
2357         list->SetOwner(1);
2358
2359         if (aResult->GetRowCount() == 0)
2360         {
2361                 Log(detector,
2362                         Form("GetFileIDs - No entry in %s FXS table for source: %s", GetSystemName(system), source));
2363                 delete aResult;
2364                 return list;
2365         }
2366
2367         Log(detector, Form("GetFileIDs - Found %d ids", aResult->GetRowCount()));
2368
2369         TSQLRow* aRow;
2370
2371         while ((aRow = aResult->Next()))
2372         {
2373
2374                 TString id(aRow->GetField(0), aRow->GetFieldLength(0));
2375                 AliDebug(2, Form("fileId = %s", id.Data()));
2376                 list->Add(new TObjString(id));
2377                 delete aRow;
2378         }
2379
2380         delete aResult;
2381
2382         return list;
2383 }
2384
2385 //______________________________________________________________________________________________
2386 Bool_t AliShuttle::Connect(Int_t system)
2387 {
2388         // Connect to MySQL Server of the system's FXS MySQL databases
2389         // DAQ Logbook, Shuttle Logbook and DAQ FXS db are on the same host
2390         //
2391
2392         // check connection: if already connected return
2393         if(fServer[system] && fServer[system]->IsConnected()) return kTRUE;
2394
2395         TString dbHost, dbUser, dbPass, dbName;
2396
2397         if (system < 3) // FXS db servers
2398         {
2399                 dbHost = Form("mysql://%s:%d", fConfig->GetFXSdbHost(system), fConfig->GetFXSdbPort(system));
2400                 dbUser = fConfig->GetFXSdbUser(system);
2401                 dbPass = fConfig->GetFXSdbPass(system);
2402                 dbName =   fConfig->GetFXSdbName(system);
2403         } else { // Run & Shuttle logbook servers
2404         // TODO Will the Shuttle logbook server be the same as the Run logbook server ???
2405                 dbHost = Form("mysql://%s:%d", fConfig->GetDAQlbHost(), fConfig->GetDAQlbPort());
2406                 dbUser = fConfig->GetDAQlbUser();
2407                 dbPass = fConfig->GetDAQlbPass();
2408                 dbName =   fConfig->GetDAQlbDB();
2409         }
2410
2411         fServer[system] = TSQLServer::Connect(dbHost.Data(), dbUser.Data(), dbPass.Data());
2412         if (!fServer[system] || !fServer[system]->IsConnected()) {
2413                 if(system < 3)
2414                 {
2415                 AliError(Form("Can't establish connection to FXS database for %s",
2416                                         AliShuttleInterface::GetSystemName(system)));
2417                 } else {
2418                 AliError("Can't establish connection to Run logbook.");
2419                 }
2420                 if(fServer[system]) delete fServer[system];
2421                 return kFALSE;
2422         }
2423
2424         // Get tables
2425         TSQLResult* aResult=0;
2426         switch(system){
2427                 case kDAQ:
2428                         aResult = fServer[kDAQ]->GetTables(dbName.Data());
2429                         break;
2430                 case kDCS:
2431                         aResult = fServer[kDCS]->GetTables(dbName.Data());
2432                         break;
2433                 case kHLT:
2434                         aResult = fServer[kHLT]->GetTables(dbName.Data());
2435                         break;
2436                 default:
2437                         aResult = fServer[3]->GetTables(dbName.Data());
2438                         break;
2439         }
2440
2441         delete aResult;
2442         return kTRUE;
2443 }
2444
2445 //______________________________________________________________________________________________
2446 Bool_t AliShuttle::UpdateTable()
2447 {
2448         //
2449         // Update FXS table filling time_processed field in all rows corresponding to current run and detector
2450         //
2451
2452         Bool_t result = kTRUE;
2453
2454         for (UInt_t system=0; system<3; system++)
2455         {
2456                 if(!fFXSCalled[system]) continue;
2457
2458                 // check connection, in case connect
2459                 if (!Connect(system))
2460                 {
2461                         Log(fCurrentDetector, Form("UpdateTable - Couldn't connect to %s FXS database", GetSystemName(system)));
2462                         result = kFALSE;
2463                         continue;
2464                 }
2465
2466                 TTimeStamp now; // now
2467
2468                 // Loop on FXS list entries
2469                 TIter iter(&fFXSlist[system]);
2470                 TObjString *aFXSentry=0;
2471                 while ((aFXSentry = dynamic_cast<TObjString*> (iter.Next())))
2472                 {
2473                         TString aFXSentrystr = aFXSentry->String();
2474                         TObjArray *aFXSarray = aFXSentrystr.Tokenize("#!?!#");
2475                         if (!aFXSarray || aFXSarray->GetEntries() != 2 )
2476                         {
2477                                 Log(fCurrentDetector, Form("UpdateTable - error updating %s FXS entry. Check string: <%s>",
2478                                         GetSystemName(system), aFXSentrystr.Data()));
2479                                 if(aFXSarray) delete aFXSarray;
2480                                 result = kFALSE;
2481                                 continue;
2482                         }
2483                         const char* fileId = ((TObjString*) aFXSarray->At(0))->GetName();
2484                         const char* source = ((TObjString*) aFXSarray->At(1))->GetName();
2485
2486                         TString whereClause;
2487                         if (system == kDAQ)
2488                         {
2489                                 whereClause = Form("where run=%d and detector=\"%s\" and fileId=\"%s\" and DAQsource=\"%s\";",
2490                                                         GetCurrentRun(), fCurrentDetector.Data(), fileId, source);
2491                         }
2492                         else if (system == kDCS)
2493                         {
2494                                 whereClause = Form("where run=%d and detector=\"%s\" and fileId=\"%s\";",
2495                                                         GetCurrentRun(), fCurrentDetector.Data(), fileId);
2496                         }
2497                         else if (system == kHLT)
2498                         {
2499                                 whereClause = Form("where run=%d and detector=\"%s\" and fileId=\"%s\" and DDLnumbers=\"%s\";",
2500                                                         GetCurrentRun(), fCurrentDetector.Data(), fileId, source);
2501                         }
2502
2503                         delete aFXSarray;
2504
2505                         TString sqlQuery = Form("update %s set time_processed=%d %s", fConfig->GetFXSdbTable(system),
2506                                                                 now.GetSec(), whereClause.Data());
2507
2508                         AliDebug(2, Form("SQL query: \n%s",sqlQuery.Data()));
2509
2510                         // Query execution
2511                         TSQLResult* aResult;
2512                         aResult = dynamic_cast<TSQLResult*> (fServer[system]->Query(sqlQuery));
2513                         if (!aResult)
2514                         {
2515                                 Log(fCurrentDetector, Form("UpdateTable - %s db: can't execute SQL query <%s>",
2516                                                                 GetSystemName(system), sqlQuery.Data()));
2517                                 result = kFALSE;
2518                                 continue;
2519                         }
2520                         delete aResult;
2521                 }
2522         }
2523
2524         return result;
2525 }
2526
2527 //______________________________________________________________________________________________
2528 Bool_t AliShuttle::UpdateTableFailCase()
2529 {
2530         // Update FXS table filling time_processed field in all rows corresponding to current run and detector
2531         // this is called in case the preprocessor is declared failed for the current run, because
2532         // the fields are updated only in case of success
2533
2534         Bool_t result = kTRUE;
2535
2536         for (UInt_t system=0; system<3; system++)
2537         {
2538                 // check connection, in case connect
2539                 if (!Connect(system))
2540                 {
2541                         Log(fCurrentDetector, Form("UpdateTableFailCase - Couldn't connect to %s FXS database",
2542                                                         GetSystemName(system)));
2543                         result = kFALSE;
2544                         continue;
2545                 }
2546
2547                 TTimeStamp now; // now
2548
2549                 // Loop on FXS list entries
2550
2551                 TString whereClause = Form("where run=%d and detector=\"%s\";",
2552                                                 GetCurrentRun(), fCurrentDetector.Data());
2553
2554
2555                 TString sqlQuery = Form("update %s set time_processed=%d %s", fConfig->GetFXSdbTable(system),
2556                                                         now.GetSec(), whereClause.Data());
2557
2558                 AliDebug(2, Form("SQL query: \n%s",sqlQuery.Data()));
2559
2560                 // Query execution
2561                 TSQLResult* aResult;
2562                 aResult = dynamic_cast<TSQLResult*> (fServer[system]->Query(sqlQuery));
2563                 if (!aResult)
2564                 {
2565                         Log(fCurrentDetector, Form("UpdateTableFailCase - %s db: can't execute SQL query <%s>",
2566                                                         GetSystemName(system), sqlQuery.Data()));
2567                         result = kFALSE;
2568                         continue;
2569                 }
2570                 delete aResult;
2571         }
2572
2573         return result;
2574 }
2575
2576 //______________________________________________________________________________________________
2577 Bool_t AliShuttle::UpdateShuttleLogbook(const char* detector, const char* status)
2578 {
2579         //
2580         // Update Shuttle logbook filling detector or shuttle_done column
2581         // ex. of usage: UpdateShuttleLogbook("PHOS", "DONE") or UpdateShuttleLogbook("shuttle_done")
2582         //
2583
2584         // check connection, in case connect
2585         if(!Connect(3)){
2586                 Log("SHUTTLE", "UpdateShuttleLogbook - Couldn't connect to DAQ Logbook.");
2587                 return kFALSE;
2588         }
2589
2590         TString detName(detector);
2591         TString setClause;
2592         if (detName == "shuttle_done" || detName == "shuttle_ignored")
2593         {
2594                 setClause = "set shuttle_done=1";
2595
2596                 if (detName == "shuttle_done")
2597                 {
2598                         // Send the information to ML
2599                         TMonaLisaText  mlStatus("SHUTTLE_status", "Done");
2600
2601                         TList mlList;
2602                         mlList.Add(&mlStatus);
2603
2604                         TString mlID;
2605                         mlID.Form("%d", GetCurrentRun());
2606                         fMonaLisa->SendParameters(&mlList, mlID);
2607                 }
2608         } else {
2609                 TString statusStr(status);
2610                 if(statusStr.Contains("done", TString::kIgnoreCase) ||
2611                    statusStr.Contains("failed", TString::kIgnoreCase)){
2612                         setClause = Form("set %s=\"%s\"", detector, status);
2613                 } else {
2614                         Log("SHUTTLE",
2615                                 Form("UpdateShuttleLogbook - Invalid status <%s> for detector %s",
2616                                         status, detector));
2617                         return kFALSE;
2618                 }
2619         }
2620
2621         TString whereClause = Form("where run=%d", GetCurrentRun());
2622
2623         TString sqlQuery = Form("update %s %s %s",
2624                                         fConfig->GetShuttlelbTable(), setClause.Data(), whereClause.Data());
2625
2626         AliDebug(2, Form("SQL query: \n%s",sqlQuery.Data()));
2627
2628         // Query execution
2629         TSQLResult* aResult;
2630         aResult = dynamic_cast<TSQLResult*> (fServer[3]->Query(sqlQuery));
2631         if (!aResult) {
2632                 Log("SHUTTLE", Form("UpdateShuttleLogbook - Can't execute query <%s>", sqlQuery.Data()));
2633                 return kFALSE;
2634         }
2635         delete aResult;
2636
2637         return kTRUE;
2638 }
2639
2640 //______________________________________________________________________________________________
2641 Int_t AliShuttle::GetCurrentRun() const
2642 {
2643         //
2644         // Get current run from logbook entry
2645         //
2646
2647         return fLogbookEntry ? fLogbookEntry->GetRun() : -1;
2648 }
2649
2650 //______________________________________________________________________________________________
2651 UInt_t AliShuttle::GetCurrentStartTime() const
2652 {
2653         //
2654         // get current start time
2655         //
2656
2657         return fLogbookEntry ? fLogbookEntry->GetStartTime() : 0;
2658 }
2659
2660 //______________________________________________________________________________________________
2661 UInt_t AliShuttle::GetCurrentEndTime() const
2662 {
2663         //
2664         // get current end time from logbook entry
2665         //
2666
2667         return fLogbookEntry ? fLogbookEntry->GetEndTime() : 0;
2668 }
2669
2670 //______________________________________________________________________________________________
2671 UInt_t AliShuttle::GetCurrentYear() const
2672 {
2673         //
2674         // Get current year from logbook entry
2675         //
2676
2677         if (!fLogbookEntry) return 0;
2678
2679         TTimeStamp startTime(GetCurrentStartTime());
2680         TString year =  Form("%d",startTime.GetDate());
2681         year = year(0,4);
2682
2683         return year.Atoi();
2684 }
2685
2686 //______________________________________________________________________________________________
2687 const char* AliShuttle::GetLHCPeriod() const
2688 {
2689         //
2690         // Get current LHC period from logbook entry
2691         //
2692
2693         if (!fLogbookEntry) return 0;
2694
2695         return fLogbookEntry->GetRunParameter("LHCperiod");
2696 }
2697
2698 //______________________________________________________________________________________________
2699 void AliShuttle::Log(const char* detector, const char* message)
2700 {
2701         //
2702         // Fill log string with a message
2703         //
2704
2705         TString logRunDir = GetShuttleLogDir();
2706         if (GetCurrentRun() >=0)
2707                 logRunDir += Form("/%d", GetCurrentRun());
2708
2709         void* dir = gSystem->OpenDirectory(logRunDir.Data());
2710         if (dir == NULL) {
2711                 if (gSystem->mkdir(logRunDir.Data(), kTRUE)) {
2712                         AliError(Form("Can't open directory <%s>", GetShuttleLogDir()));
2713                         return;
2714                 }
2715
2716         } else {
2717                 gSystem->FreeDirectory(dir);
2718         }
2719
2720         TString toLog = Form("%s (%d): %s - ", TTimeStamp(time(0)).AsString("s"), getpid(), detector);
2721         if (GetCurrentRun() >= 0)
2722                 toLog += Form("run %d - ", GetCurrentRun());
2723         toLog += Form("%s", message);
2724
2725         AliInfo(toLog.Data());
2726
2727         // if we redirect the log output already to the file, leave here
2728         if (fOutputRedirected && strcmp(detector, "SHUTTLE") != 0)
2729                 return;
2730
2731         TString fileName = GetLogFileName(detector);
2732
2733         gSystem->ExpandPathName(fileName);
2734
2735         ofstream logFile;
2736         logFile.open(fileName, ofstream::out | ofstream::app);
2737
2738         if (!logFile.is_open()) {
2739                 AliError(Form("Could not open file %s", fileName.Data()));
2740                 return;
2741         }
2742
2743         logFile << toLog.Data() << "\n";
2744
2745         logFile.close();
2746 }
2747
2748 //______________________________________________________________________________________________
2749 TString AliShuttle::GetLogFileName(const char* detector) const
2750 {
2751         //
2752         // returns the name of the log file for a given sub detector
2753         //
2754
2755         TString fileName;
2756
2757         if (GetCurrentRun() >= 0)
2758         {
2759                 fileName.Form("%s/%d/%s_%d.log", GetShuttleLogDir(), GetCurrentRun(),
2760                         detector, GetCurrentRun());
2761         } else {
2762                 fileName.Form("%s/%s.log", GetShuttleLogDir(), detector);
2763         }
2764
2765         return fileName;
2766 }
2767
2768 //______________________________________________________________________________________________
2769 void AliShuttle::SendAlive()
2770 {
2771         // sends alive message to ML
2772
2773         TMonaLisaText mlStatus("SHUTTLE_status", "Alive");
2774
2775         TList mlList;
2776         mlList.Add(&mlStatus);
2777
2778         fMonaLisa->SendParameters(&mlList, "__PROCESSINGINFO__");
2779 }
2780
2781 //______________________________________________________________________________________________
2782 Bool_t AliShuttle::Collect(Int_t run)
2783 {
2784         //
2785         // Collects conditions data for all UNPROCESSED run written to DAQ LogBook in case of run = -1 (default)
2786         // If a dedicated run is given this run is processed
2787         //
2788         // In operational mode, this is the Shuttle function triggered by the EOR signal.
2789         //
2790
2791         if (run == -1)
2792                 Log("SHUTTLE","Collect - Shuttle called. Collecting conditions data for unprocessed runs");
2793         else
2794                 Log("SHUTTLE", Form("Collect - Shuttle called. Collecting conditions data for run %d", run));
2795
2796         SetLastAction("Starting");
2797
2798         // create ML instance
2799         if (!fMonaLisa)
2800                 fMonaLisa = new TMonaLisaWriter(fConfig->GetMonitorHost(), fConfig->GetMonitorTable());
2801
2802         SendAlive();
2803         CountOpenRuns();
2804
2805         TString whereClause("where shuttle_done=0");
2806         if (run != -1)
2807                 whereClause += Form(" and run=%d", run);
2808
2809         TObjArray shuttleLogbookEntries;
2810         if (!QueryShuttleLogbook(whereClause, shuttleLogbookEntries))
2811         {
2812                 Log("SHUTTLE", "Collect - Can't retrieve entries from Shuttle logbook");
2813                 return kFALSE;
2814         }
2815
2816         if (shuttleLogbookEntries.GetEntries() == 0)
2817         {
2818                 if (run == -1)
2819                         Log("SHUTTLE","Collect - Found no UNPROCESSED runs in Shuttle logbook");
2820                 else
2821                         Log("SHUTTLE", Form("Collect - Run %d is already DONE "
2822                                                 "or it does not exist in Shuttle logbook", run));
2823                 return kTRUE;
2824         }
2825
2826         for (UInt_t iDet=0; iDet<NDetectors(); iDet++)
2827                 fFirstUnprocessed[iDet] = kTRUE;
2828
2829         if (run != -1)
2830         {
2831                 // query Shuttle logbook for earlier runs, check if some detectors are unprocessed,
2832                 // flag them into fFirstUnprocessed array
2833                 TString whereClause(Form("where shuttle_done=0 and run < %d", run));
2834                 TObjArray tmpLogbookEntries;
2835                 if (!QueryShuttleLogbook(whereClause, tmpLogbookEntries))
2836                 {
2837                         Log("SHUTTLE", "Collect - Can't retrieve entries from Shuttle logbook");
2838                         return kFALSE;
2839                 }
2840
2841                 TIter iter(&tmpLogbookEntries);
2842                 AliShuttleLogbookEntry* anEntry = 0;
2843                 while ((anEntry = dynamic_cast<AliShuttleLogbookEntry*> (iter.Next())))
2844                 {
2845                         for (UInt_t iDet=0; iDet<NDetectors(); iDet++)
2846                         {
2847                                 if (anEntry->GetDetectorStatus(iDet) == AliShuttleLogbookEntry::kUnprocessed)
2848                                 {
2849                                         AliDebug(2, Form("Run %d: setting %s as \"not first time unprocessed\"",
2850                                                         anEntry->GetRun(), GetDetName(iDet)));
2851                                         fFirstUnprocessed[iDet] = kFALSE;
2852                                 }
2853                         }
2854
2855                 }
2856
2857         }
2858
2859         if (!RetrieveConditionsData(shuttleLogbookEntries))
2860         {
2861                 Log("SHUTTLE", "Collect - Process of at least one run failed");
2862                 CountOpenRuns();
2863                 return kFALSE;
2864         }
2865
2866         Log("SHUTTLE", "Collect - Requested run(s) successfully processed");
2867         CountOpenRuns();
2868         return kTRUE;
2869 }
2870
2871 //______________________________________________________________________________________________
2872 Bool_t AliShuttle::RetrieveConditionsData(const TObjArray& dateEntries)
2873 {
2874         //
2875         // Retrieve conditions data for all runs that aren't processed yet
2876         //
2877
2878         Bool_t hasError = kFALSE;
2879
2880         TIter iter(&dateEntries);
2881         AliShuttleLogbookEntry* anEntry;
2882
2883         while ((anEntry = (AliShuttleLogbookEntry*) iter.Next())){
2884                 if (!Process(anEntry)){
2885                         hasError = kTRUE;
2886                 }
2887
2888                 // clean SHUTTLE temp directory
2889                 //TString filename = Form("%s/*.shuttle", GetShuttleTempDir());
2890                 //RemoveFile(filename.Data());
2891         }
2892
2893         return hasError == kFALSE;
2894 }
2895
2896 //______________________________________________________________________________________________
2897 ULong_t AliShuttle::GetTimeOfLastAction() const
2898 {
2899         //
2900         // Gets time of last action
2901         //
2902
2903         ULong_t tmp;
2904
2905         fMonitoringMutex->Lock();
2906
2907         tmp = fLastActionTime;
2908
2909         fMonitoringMutex->UnLock();
2910
2911         return tmp;
2912 }
2913
2914 //______________________________________________________________________________________________
2915 const TString AliShuttle::GetLastAction() const
2916 {
2917         //
2918         // returns a string description of the last action
2919         //
2920
2921         TString tmp;
2922
2923         fMonitoringMutex->Lock();
2924
2925         tmp = fLastAction;
2926
2927         fMonitoringMutex->UnLock();
2928
2929         return tmp;
2930 }
2931
2932 //______________________________________________________________________________________________
2933 void AliShuttle::SetLastAction(const char* action)
2934 {
2935         //
2936         // updates the monitoring variables
2937         //
2938
2939         fMonitoringMutex->Lock();
2940
2941         fLastAction = action;
2942         fLastActionTime = time(0);
2943
2944         fMonitoringMutex->UnLock();
2945 }
2946
2947 //______________________________________________________________________________________________
2948 const char* AliShuttle::GetRunParameter(const char* param)
2949 {
2950         //
2951         // returns run parameter read from DAQ logbook
2952         //
2953
2954         if(!fLogbookEntry) {
2955                 AliError("No logbook entry!");
2956                 return 0;
2957         }
2958
2959         return fLogbookEntry->GetRunParameter(param);
2960 }
2961
2962 //______________________________________________________________________________________________
2963 AliCDBEntry* AliShuttle::GetFromOCDB(const char* detector, const AliCDBPath& path)
2964 {
2965         //
2966         // returns object from OCDB valid for current run
2967         //
2968
2969         if (fTestMode & kErrorOCDB)
2970         {
2971                 Log(detector, "GetFromOCDB - In TESTMODE - Simulating error with OCDB");
2972                 return 0;
2973         }
2974
2975         AliCDBStorage *sto = AliCDBManager::Instance()->GetStorage(fgkMainCDB);
2976         if (!sto)
2977         {
2978                 Log(detector, "GetFromOCDB - Cannot activate main OCDB for query!");
2979                 return 0;
2980         }
2981
2982         return dynamic_cast<AliCDBEntry*> (sto->Get(path, GetCurrentRun()));
2983 }
2984
2985 //______________________________________________________________________________________________
2986 Bool_t AliShuttle::SendMail()
2987 {
2988         //
2989         // sends a mail to the subdetector expert in case of preprocessor error
2990         //
2991
2992         if (fTestMode != kNone)
2993                 return kTRUE;
2994
2995         if (!fConfig->SendMail()) return kTRUE;
2996
2997         TString to="";
2998         TIter iterExperts(fConfig->GetResponsibles(fCurrentDetector));
2999         TObjString *anExpert=0;
3000         while ((anExpert = (TObjString*) iterExperts.Next()))
3001         {
3002                 to += Form("%s,", anExpert->GetName());
3003         }
3004         if (to.Length() > 0)
3005           to.Remove(to.Length()-1);
3006         AliDebug(2, Form("to: %s",to.Data()));
3007
3008         if (to.IsNull()) {
3009                 Log("SHUTTLE", "List of detector responsibles not set!");
3010                 return kFALSE;
3011         }
3012
3013         void* dir = gSystem->OpenDirectory(GetShuttleLogDir());
3014         if (dir == NULL)
3015         {
3016                 if (gSystem->mkdir(GetShuttleLogDir(), kTRUE))
3017                 {
3018                         Log("SHUTTLE", Form("SendMail - Can't open directory <%s>", GetShuttleLogDir()));
3019                         return kFALSE;
3020                 }
3021
3022         } else {
3023                 gSystem->FreeDirectory(dir);
3024         }
3025
3026         TString bodyFileName;
3027         bodyFileName.Form("%s/mail.body", GetShuttleLogDir());
3028         gSystem->ExpandPathName(bodyFileName);
3029
3030         ofstream mailBody;
3031         mailBody.open(bodyFileName, ofstream::out);
3032
3033         if (!mailBody.is_open())
3034         {
3035                 Log("SHUTTLE", Form("Could not open mail body file %s", bodyFileName.Data()));
3036                 return kFALSE;
3037         }
3038
3039         TString cc="";
3040         TIter iterAdmins(fConfig->GetAdmins(AliShuttleConfig::kGlobal));
3041         TObjString *anAdmin=0;
3042         while ((anAdmin = (TObjString*) iterAdmins.Next()))
3043         {
3044                 cc += Form("%s,", anAdmin->GetName());
3045         }
3046         if (cc.Length() > 0)
3047           cc.Remove(to.Length()-1);
3048         AliDebug(2, Form("cc: %s",to.Data()));
3049
3050         TString subject = Form("%s Shuttle preprocessor FAILED in run %d (run type = %s)!",
3051                                 fCurrentDetector.Data(), GetCurrentRun(), GetRunType());
3052         AliDebug(2, Form("subject: %s", subject.Data()));
3053
3054         TString body = Form("Dear %s expert(s), \n\n", fCurrentDetector.Data());
3055         body += Form("SHUTTLE just detected that your preprocessor "
3056                         "failed processing run %d (run type = %s)!!\n\n",
3057                                         GetCurrentRun(), GetRunType());
3058         body += Form("Please check %s status on the SHUTTLE monitoring page: \n\n",
3059                                 fCurrentDetector.Data());
3060         if (fConfig->GetRunMode() == AliShuttleConfig::kTest)
3061         {
3062                 body += Form("\thttp://pcalimonitor.cern.ch:8889/shuttle.jsp?time=168 \n\n");
3063         } else {
3064                 body += Form("\thttp://pcalimonitor.cern.ch/shuttle.jsp?instance=PROD&time=168 \n\n");
3065         }
3066
3067
3068         TString logFolder = "logs";
3069         if (fConfig->GetRunMode() == AliShuttleConfig::kProd)
3070                 logFolder += "_PROD";
3071
3072
3073         body += Form("Find the %s log for the current run on \n\n"
3074                 "\thttp://pcalishuttle01.cern.ch:8880/%s/%d/%s_%d.log \n\n",
3075                 fCurrentDetector.Data(), logFolder.Data(), GetCurrentRun(),
3076                                 fCurrentDetector.Data(), GetCurrentRun());
3077         body += Form("The last 10 lines of %s log file are following:\n\n", fCurrentDetector.Data());
3078
3079         AliDebug(2, Form("Body begin: %s", body.Data()));
3080
3081         mailBody << body.Data();
3082         mailBody.close();
3083         mailBody.open(bodyFileName, ofstream::out | ofstream::app);
3084
3085         TString logFileName = Form("%s/%d/%s_%d.log", GetShuttleLogDir(),
3086                 GetCurrentRun(), fCurrentDetector.Data(), GetCurrentRun());
3087         TString tailCommand = Form("tail -n 10 %s >> %s", logFileName.Data(), bodyFileName.Data());
3088         if (gSystem->Exec(tailCommand.Data()))
3089         {
3090                 mailBody << Form("%s log file not found ...\n\n", fCurrentDetector.Data());
3091         }
3092
3093         TString endBody = Form("------------------------------------------------------\n\n");
3094         endBody += Form("In case of problems please contact the SHUTTLE core team.\n\n");
3095         endBody += "Please do not answer this message directly, it is automatically generated.\n\n";
3096         endBody += "Greetings,\n\n \t\t\tthe SHUTTLE\n";
3097
3098         AliDebug(2, Form("Body end: %s", endBody.Data()));
3099
3100         mailBody << endBody.Data();
3101
3102         mailBody.close();
3103
3104         // send mail!
3105         TString mailCommand = Form("mail -s \"%s\" -c %s %s < %s",
3106                                                 subject.Data(),
3107                                                 cc.Data(),
3108                                                 to.Data(),
3109                                                 bodyFileName.Data());
3110         AliDebug(2, Form("mail command: %s", mailCommand.Data()));
3111
3112         Bool_t result = gSystem->Exec(mailCommand.Data());
3113
3114         return result == 0;
3115 }
3116
3117 //______________________________________________________________________________________________
3118 Bool_t AliShuttle::SendMailToDCS()
3119 {
3120         //
3121         // sends a mail to the DCS Amanda experts in case of DCS data point retrieval error
3122         //
3123
3124         if (fTestMode != kNone)
3125                 return kTRUE;
3126
3127         if (!fConfig->SendMail()) return kTRUE;
3128
3129         void* dir = gSystem->OpenDirectory(GetShuttleLogDir());
3130         if (dir == NULL)
3131         {
3132                 if (gSystem->mkdir(GetShuttleLogDir(), kTRUE))
3133                 {
3134                         Log("SHUTTLE", Form("SendMailToDCS - Can't open directory <%s>", GetShuttleLogDir()));
3135                         return kFALSE;
3136                 }
3137
3138         } else {
3139                 gSystem->FreeDirectory(dir);
3140         }
3141
3142         TString bodyFileName;
3143         bodyFileName.Form("%s/mail.body", GetShuttleLogDir());
3144         gSystem->ExpandPathName(bodyFileName);
3145
3146         ofstream mailBody;
3147         mailBody.open(bodyFileName, ofstream::out);
3148
3149         if (!mailBody.is_open())
3150         {
3151                 Log("SHUTTLE", Form("SendMailToDCS - Could not open mail body file %s", bodyFileName.Data()));
3152                 return kFALSE;
3153         }
3154
3155         TString to="";
3156         TIter iterExperts(fConfig->GetAdmins(AliShuttleConfig::kAmanda));
3157         TObjString *anExpert=0;
3158         while ((anExpert = (TObjString*) iterExperts.Next()))
3159         {
3160                 to += Form("%s,", anExpert->GetName());
3161         }
3162         if (to.Length() > 0)
3163           to.Remove(to.Length()-1);
3164         AliDebug(2, Form("to: %s",to.Data()));
3165
3166         if (to.IsNull()) {
3167                 Log("SHUTTLE", "List of Amanda server administrators not set!");
3168                 return kFALSE;
3169         }
3170
3171         TString cc="";
3172         TIter iterAdmins(fConfig->GetAdmins(AliShuttleConfig::kGlobal));
3173         TObjString *anAdmin=0;
3174         while ((anAdmin = (TObjString*) iterAdmins.Next()))
3175         {
3176                 cc += Form("%s,", anAdmin->GetName());
3177         }
3178         if (cc.Length() > 0)
3179           cc.Remove(to.Length()-1);
3180         AliDebug(2, Form("cc: %s",to.Data()));
3181
3182         TString subject = Form("Retrieval of data points for %s FAILED in run %d !",
3183                                 fCurrentDetector.Data(), GetCurrentRun());
3184         AliDebug(2, Form("subject: %s", subject.Data()));
3185
3186         TString body = Form("Dear DCS experts, \n\n");
3187         body += Form("SHUTTLE couldn\'t retrieve the data points for detector %s "
3188                         "in run %d!!\n\n", fCurrentDetector.Data(), GetCurrentRun());
3189         body += Form("Please check %s status on the SHUTTLE monitoring page: \n\n",
3190                                 fCurrentDetector.Data());
3191         if (fConfig->GetRunMode() == AliShuttleConfig::kTest)
3192         {
3193                 body += Form("\thttp://pcalimonitor.cern.ch:8889/shuttle.jsp?time=168 \n\n");
3194         } else {
3195                 body += Form("\thttp://pcalimonitor.cern.ch/shuttle.jsp?instance=PROD?time=168 \n\n");
3196         }
3197
3198         TString logFolder = "logs";
3199         if (fConfig->GetRunMode() == AliShuttleConfig::kProd)
3200                 logFolder += "_PROD";
3201
3202
3203         body += Form("Find the %s log for the current run on \n\n"
3204                 "\thttp://pcalishuttle01.cern.ch:8880/%s/%d/%s_%d.log \n\n",
3205                 fCurrentDetector.Data(), logFolder.Data(), GetCurrentRun(),
3206                                 fCurrentDetector.Data(), GetCurrentRun());
3207         body += Form("The last 10 lines of %s log file are following:\n\n", fCurrentDetector.Data());
3208
3209         AliDebug(2, Form("Body begin: %s", body.Data()));
3210
3211         mailBody << body.Data();
3212         mailBody.close();
3213         mailBody.open(bodyFileName, ofstream::out | ofstream::app);
3214
3215         TString logFileName = Form("%s/%d/%s_%d.log", GetShuttleLogDir(), GetCurrentRun(),
3216                 fCurrentDetector.Data(), GetCurrentRun());
3217         TString tailCommand = Form("tail -n 10 %s >> %s", logFileName.Data(), bodyFileName.Data());
3218         if (gSystem->Exec(tailCommand.Data()))
3219         {
3220                 mailBody << Form("%s log file not found ...\n\n", fCurrentDetector.Data());
3221         }
3222
3223         TString endBody = Form("------------------------------------------------------\n\n");
3224         endBody += Form("In case of problems please contact the SHUTTLE core team.\n\n");
3225         endBody += "Please do not answer this message directly, it is automatically generated.\n\n";
3226         endBody += "Greetings,\n\n \t\t\tthe SHUTTLE\n";
3227
3228         AliDebug(2, Form("Body end: %s", endBody.Data()));
3229
3230         mailBody << endBody.Data();
3231
3232         mailBody.close();
3233
3234         // send mail!
3235         TString mailCommand = Form("mail -s \"%s\" -c %s %s < %s",
3236                                                 subject.Data(),
3237                                                 cc.Data(),
3238                                                 to.Data(),
3239                                                 bodyFileName.Data());
3240         AliDebug(2, Form("mail command: %s", mailCommand.Data()));
3241
3242         Bool_t result = gSystem->Exec(mailCommand.Data());
3243
3244         return result == 0;
3245 }
3246
3247 //______________________________________________________________________________________________
3248 const char* AliShuttle::GetRunType()
3249 {
3250         //
3251         // returns run type read from "run type" logbook
3252         //
3253
3254         if(!fLogbookEntry) {
3255                 AliError("No logbook entry!");
3256                 return 0;
3257         }
3258
3259         return fLogbookEntry->GetRunType();
3260 }
3261
3262 //______________________________________________________________________________________________
3263 Bool_t AliShuttle::GetHLTStatus()
3264 {
3265         // Return HLT status (ON=1 OFF=0)
3266         // Converts the HLT status from the status string read in the run logbook (not just a bool)
3267
3268         if(!fLogbookEntry) {
3269                 AliError("No logbook entry!");
3270                 return 0;
3271         }
3272
3273         // TODO implement when HLTStatus is inserted in run logbook
3274         //TString hltStatus = fLogbookEntry->GetRunParameter("HLTStatus");
3275         //if(hltStatus == "OFF") {return kFALSE};
3276
3277         return kTRUE;
3278 }
3279
3280 //______________________________________________________________________________________________
3281 void AliShuttle::SetShuttleTempDir(const char* tmpDir)
3282 {
3283         //
3284         // sets Shuttle temp directory
3285         //
3286
3287         fgkShuttleTempDir = gSystem->ExpandPathName(tmpDir);
3288 }
3289
3290 //______________________________________________________________________________________________
3291 void AliShuttle::SetShuttleLogDir(const char* logDir)
3292 {
3293         //
3294         // sets Shuttle log directory
3295         //
3296
3297         fgkShuttleLogDir = gSystem->ExpandPathName(logDir);
3298 }