SHUTTLE/AliShuttle.cxx

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