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("SHUTTLE","StoreOCDB - 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                 Log(fCurrentDetector.Data(), Form("Attempting to store %s", aLocId.ToString().Data()));
 380
 381                 // If local object is valid up to infinity we store it only if it is
 382                 // the first unprocessed run!
 383                 if (aLocId.GetLastRun() == AliCDBRunRange::Infinity() &&
 384                         !fFirstUnprocessed[GetDetPos(fCurrentDetector)])
 385                 {
 386                         Log("SHUTTLE", Form("StoreOCDB - %s: object %s has validity infinite but "
 387                                                 "there are previous unprocessed runs!",
 388                                                 fCurrentDetector.Data(), aLocId.GetPath().Data()));
 389                         Log(fCurrentDetector.Data(), Form("StoreOCDB - %s: object %s has validity infinite but "
 390                                                 "there are previous unprocessed runs!",
 391                                                 fCurrentDetector.Data(), aLocId.GetPath().Data()));
 392                         delayed = kTRUE;
 393                         continue;
 394                 }
 395
 396                 // loop on Grid valid Id's
 397                 Bool_t store = kTRUE;
 398                 TIter gridIter(gridIds);
 399                 AliCDBId* aGridId = 0;
 400                 while ((aGridId = dynamic_cast<AliCDBId*> (gridIter.Next()))) {
 401                         if (aGridId->GetPath() != aLocId.GetPath())
 402                                 continue;
 403                         // skip all objects valid up to infinity
 404                         if (aGridId->GetLastRun() == AliCDBRunRange::Infinity())
 405                                 continue;
 406
 407                         // if we get here, it means there's already some more recent object stored on Grid!
 408                         Log(fCurrentDetector.Data(),
 409                                 Form("StoreOCDB - A more recent object already exists in %s storage: <%s>",
 410                                 type, aGridId->ToString().Data()));
 411
 412                         store = kFALSE;
 413                         break;
 414                 }
 415
 416                 Bool_t storeOk = kFALSE;
 417                 if (store)
 418                 {
 419                         Log(fCurrentDetector.Data(), Form("Prechecks succeeded. Ready to store %s", aLocId.ToString().Data()));
 420                         storeOk = gridSto->Put(aLocEntry);
 421                         if (storeOk) {
 422                                 Log("SHUTTLE",
 423                                 Form("StoreOCDB - Object <%s> successfully put into %s storage",
 424                                         aLocId.ToString().Data(), type));
 425                                 Log(fCurrentDetector.Data(),
 426                                         Form("StoreOCDB - Object <%s> successfully put into %s storage",
 427                                         aLocId.ToString().Data(), type));
 428                         } else  {
 429                                 Log("SHUTTLE",
 430                                         Form("StoreOCDB - Grid %s storage of object <%s> failed",
 431                                         type, aLocId.ToString().Data()));
 432                                 Log(fCurrentDetector.Data(),
 433                                         Form("StoreOCDB - Grid %s storage of object <%s> failed",
 434                                         type, aLocId.ToString().Data()));
 435                                 result = kFALSE;
 436                         }
 437                 }
 438
 439                 if (!store || storeOk) {
 440                         // removing local file...
 441                         TString filename;
 442                         localSto->IdToFilename(aLocId, filename);
 443                         Log("SHUTTLE", Form("StoreOCDB - Removing local file %s", filename.Data()));
 444                         RemoveFile(filename.Data());
 445                 }
 446         }
 447         localEntries->Clear();
 448
 449         Int_t returnCode = 0;
 450
 451         if (result == kFALSE)
 452                 returnCode = -1;
 453         else if (delayed != kFALSE)
 454                 returnCode =  1;
 455
 456         Log("SHUTTLE", Form("StoreOCDB - Returning with %d (result = %d, delayed = %d)", returnCode, result, delayed));
 457         Log(fCurrentDetector.Data(), Form("StoreOCDB - Returning with %d (result = %d, delayed = %d)", returnCode, result, delayed));
 458
 459         return returnCode;
 460 }
 461
 462 //______________________________________________________________________________________________
 463 Bool_t AliShuttle::CleanReferenceStorage(const char* detector)
 464 {
 465         // clears the directory used to store reference files of a given subdetector
 466
 467         AliCDBManager* man = AliCDBManager::Instance();
 468         AliCDBStorage* sto = man->GetStorage(fgkLocalRefStorage);
 469         TString localBaseFolder = sto->GetBaseFolder();
 470
 471         TString targetDir = GetRefFilePrefix(localBaseFolder.Data(), detector);
 472
 473         Log("SHUTTLE", Form("CleanReferenceStorage - Cleaning %s", targetDir.Data()));
 474
 475         TString begin;
 476         begin.Form("%d_", GetCurrentRun());
 477
 478         TSystemDirectory* baseDir = new TSystemDirectory("/", targetDir);
 479         if (!baseDir)
 480                 return kTRUE;
 481
 482         TList* dirList = baseDir->GetListOfFiles();
 483         delete baseDir;
 484
 485         if (!dirList) return kTRUE;
 486
 487         if (dirList->GetEntries() < 3)
 488         {
 489                 delete dirList;
 490                 return kTRUE;
 491         }
 492
 493         Int_t nDirs = 0, nDel = 0;
 494         TIter dirIter(dirList);
 495         TSystemFile* entry = 0;
 496
 497         Bool_t success = kTRUE;
 498
 499         while ((entry = dynamic_cast<TSystemFile*> (dirIter.Next())))
 500         {
 501                 if (entry->IsDirectory())
 502                         continue;
 503
 504                 TString fileName(entry->GetName());
 505                 if (!fileName.BeginsWith(begin))
 506                         continue;
 507
 508                 nDirs++;
 509
 510                 // delete file
 511                 Int_t result = gSystem->Unlink(fileName.Data());
 512
 513                 if (result)
 514                 {
 515                         Log("SHUTTLE", Form("CleanReferenceStorage - Could not delete file %s!", fileName.Data()));
 516                         success = kFALSE;
 517                 } else {
 518                         nDel++;
 519                 }
 520         }
 521
 522         if(nDirs > 0)
 523                 Log("SHUTTLE", Form("CleanReferenceStorage - %d (over %d) reference files in folder %s were deleted.",
 524                         nDel, nDirs, targetDir.Data()));
 525
 526
 527         delete dirList;
 528         return success;
 529
 530
 531
 532
 533
 534
 535   Int_t result = gSystem->GetPathInfo(targetDir, 0, (Long64_t*) 0, 0, 0);
 536   if (result == 0)
 537   {
 538     // delete directory
 539     result = gSystem->Exec(Form("rm -rf %s", targetDir.Data()));
 540     if (result != 0)
 541     {
 542       Log("SHUTTLE", Form("CleanReferenceStorage - Could not clean directory %s", targetDir.Data()));
 543       return kFALSE;
 544     }
 545   }
 546
 547   result = gSystem->mkdir(targetDir, kTRUE);
 548   if (result != 0)
 549   {
 550     Log("SHUTTLE", Form("CleanReferenceStorage - Error creating base directory %s", targetDir.Data()));
 551     return kFALSE;
 552   }
 553
 554   return kTRUE;
 555 }
 556
 557 //______________________________________________________________________________________________
 558 Bool_t AliShuttle::StoreReferenceFile(const char* detector, const char* localFile, const char* gridFileName)
 559 {
 560         //
 561         // Stores reference file directly (without opening it). This function stores the file locally.
 562         //
 563         // The file is stored under the following location:
 564         // <base folder of local reference storage>/<DET>/<RUN#>_<gridFileName>
 565         // where <gridFileName> is the second parameter given to the function
 566         //
 567
 568         if (fTestMode & kErrorStorage)
 569         {
 570                 Log(fCurrentDetector, "StoreReferenceFile - In TESTMODE - Simulating error while storing locally");
 571                 return kFALSE;
 572         }
 573
 574         AliCDBManager* man = AliCDBManager::Instance();
 575         AliCDBStorage* sto = man->GetStorage(fgkLocalRefStorage);
 576
 577         TString localBaseFolder = sto->GetBaseFolder();
 578
 579         TString target = GetRefFilePrefix(localBaseFolder.Data(), detector);
 580         target.Append(Form("/%d_%s", GetCurrentRun(), gridFileName));
 581
 582         return CopyFileLocally(localFile, target);
 583 }
 584
 585 //______________________________________________________________________________________________
 586 Bool_t AliShuttle::StoreRunMetadataFile(const char* localFile, const char* gridFileName)
 587 {
 588         //
 589         // Stores Run metadata file to the Grid, in the run folder
 590         //
 591         // Only GRP can call this function.
 592
 593         if (fTestMode & kErrorStorage)
 594         {
 595                 Log(fCurrentDetector, "StoreRunMetaDataFile - In TESTMODE - Simulating error while storing locally");
 596                 return kFALSE;
 597         }
 598
 599         AliCDBManager* man = AliCDBManager::Instance();
 600         AliCDBStorage* sto = man->GetStorage(fgkLocalRefStorage);
 601
 602         TString localBaseFolder = sto->GetBaseFolder();
 603
 604         // Build Run level folder
 605         // folder = /alice/data/year/lhcPeriod/runNb/raw
 606
 607
 608         TString lhcPeriod = GetLHCPeriod();
 609         if (lhcPeriod.Length() == 0)
 610         {
 611                 Log("SHUTTLE","StoreRunMetaDataFile - LHCPeriod not found in logbook!");
 612                 return 0;
 613         }
 614
 615         // TODO partitions with one detector only write data into LHCperiod_DET
 616         TString partition = GetRunParameter("detector");
 617
 618         if (partition.Length() > 0 && partition != "ALICE")
 619         {
 620                 lhcPeriod.Append(Form("_%s", partition.Data()));
 621                 Log(fCurrentDetector, Form("Run data tags merged file will be written in %s",
 622                                 lhcPeriod.Data()));
 623         }
 624
 625         TString target = Form("%s/GRP/RunMetadata%s%d/%s/%09d/raw/%s",
 626                                 localBaseFolder.Data(), fConfig->GetAlienPath(), GetCurrentYear(),
 627                                 lhcPeriod.Data(), GetCurrentRun(), gridFileName);
 628
 629         return CopyFileLocally(localFile, target);
 630 }
 631
 632 //______________________________________________________________________________________________
 633 Bool_t AliShuttle::CopyFileLocally(const char* localFile, const TString& target)
 634 {
 635         //
 636         // Stores file locally. Called by StoreReferenceFile and StoreRunMetadataFile
 637         // Files are temporarily stored in the local reference storage. When the preprocessor
 638         // finishes, the Shuttle calls CopyFilesToGrid to transfer the files to AliEn
 639         // (in reference or run level folders)
 640         //
 641
 642         TString targetDir(target(0, target.Last('/')));
 643
 644         //try to open base dir folder, if it does not exist
 645         void* dir = gSystem->OpenDirectory(targetDir.Data());
 646         if (dir == NULL) {
 647                 if (gSystem->mkdir(targetDir.Data(), kTRUE)) {
 648                         Log("SHUTTLE", Form("CopyFileLocally - Can't open directory <%s>", targetDir.Data()));
 649                         return kFALSE;
 650                 }
 651
 652         } else {
 653                 gSystem->FreeDirectory(dir);
 654         }
 655
 656         Int_t result = 0;
 657
 658         result = gSystem->GetPathInfo(localFile, 0, (Long64_t*) 0, 0, 0);
 659         if (result)
 660         {
 661                 Log("SHUTTLE", Form("CopyFileLocally - %s does not exist", localFile));
 662                 return kFALSE;
 663         }
 664
 665         result = gSystem->GetPathInfo(target, 0, (Long64_t*) 0, 0, 0);
 666         if (!result)
 667         {
 668                 Log("SHUTTLE", Form("CopyFileLocally - target file %s already exist, removing...", target.Data()));
 669                 if (gSystem->Unlink(target.Data()))
 670                 {
 671                         Log("SHUTTLE", Form("CopyFileLocally - Could not remove existing target file %s!", target.Data()));
 672                         return kFALSE;
 673                 }
 674         }
 675
 676         result = gSystem->CopyFile(localFile, target);
 677
 678         if (result == 0)
 679         {
 680                 Log("SHUTTLE", Form("CopyFileLocally - File %s stored locally to %s", localFile, target.Data()));
 681                 return kTRUE;
 682         }
 683         else
 684         {
 685                 Log("SHUTTLE", Form("CopyFileLocally - Could not store file %s to %s! Error code = %d",
 686                                 localFile, target.Data(), result));
 687                 return kFALSE;
 688         }
 689
 690
 691
 692 }
 693
 694 //______________________________________________________________________________________________
 695 Bool_t AliShuttle::CopyFilesToGrid(const char* type)
 696 {
 697         //
 698         // Transfers local files to the Grid. Local files can be reference files
 699         // or run metadata file (from GRP only).
 700         //
 701         // According to the type (ref, metadata) the files are stored under the following location:
 702         // ref --> <base folder of reference storage>/<DET>/<RUN#>_<gridFileName>
 703         // metadata --> <run data folder>/<MetadataFileName>
 704         //
 705
 706         AliCDBManager* man = AliCDBManager::Instance();
 707         AliCDBStorage* sto = man->GetStorage(fgkLocalRefStorage);
 708         if (!sto)
 709                 return kFALSE;
 710         TString localBaseFolder = sto->GetBaseFolder();
 711
 712         TString dir;
 713         TString alienDir;
 714         TString begin;
 715
 716         if (strcmp(type, "reference") == 0)
 717         {
 718                 dir = GetRefFilePrefix(localBaseFolder.Data(), fCurrentDetector.Data());
 719                 AliCDBStorage* gridSto = man->GetStorage(fgkMainRefStorage);
 720                 if (!gridSto)
 721                         return kFALSE;
 722                 TString gridBaseFolder = gridSto->GetBaseFolder();
 723                 alienDir = GetRefFilePrefix(gridBaseFolder.Data(), fCurrentDetector.Data());
 724                 begin = Form("%d_", GetCurrentRun());
 725         }
 726         else if (strcmp(type, "metadata") == 0)
 727         {
 728
 729                 TString lhcPeriod = GetLHCPeriod();
 730
 731                 if (lhcPeriod.Length() == 0)
 732                 {
 733                         Log("SHUTTLE","CopyFilesToGrid - LHCPeriod not found in logbook!");
 734                         return 0;
 735                 }
 736
 737                 // TODO partitions with one detector only write data into LHCperiod_DET
 738                 TString partition = GetRunParameter("detector");
 739
 740                 if (partition.Length() > 0 && partition != "ALICE")
 741                 {
 742                         lhcPeriod.Append(Form("_%s", partition.Data()));
 743                 }
 744
 745                 dir = Form("%s/GRP/RunMetadata%s%d/%s/%09d/raw",
 746                                 localBaseFolder.Data(), fConfig->GetAlienPath(), GetCurrentYear(),
 747                                 lhcPeriod.Data(), GetCurrentRun());
 748                 alienDir = dir(dir.Index(fConfig->GetAlienPath()), dir.Length());
 749
 750                 begin = "";
 751         }
 752         else
 753         {
 754                 Log("SHUTTLE", "CopyFilesToGrid - Unexpected: type label must be reference or metadata!");
 755                 return kFALSE;
 756         }
 757
 758         TSystemDirectory* baseDir = new TSystemDirectory("/", dir);
 759         if (!baseDir)
 760                 return kTRUE;
 761
 762         TList* dirList = baseDir->GetListOfFiles();
 763         delete baseDir;
 764
 765         if (!dirList) return kTRUE;
 766
 767         if (dirList->GetEntries() < 3)
 768         {
 769                 delete dirList;
 770                 return kTRUE;
 771         }
 772
 773         if (!gGrid)
 774         {
 775                 Log("SHUTTLE", "CopyFilesToGrid - Connection to Grid failed: Cannot continue!");
 776                 delete dirList;
 777                 return kFALSE;
 778         }
 779
 780         Int_t nDirs = 0, nTransfer = 0;
 781         TIter dirIter(dirList);
 782         TSystemFile* entry = 0;
 783
 784         Bool_t success = kTRUE;
 785         Bool_t first = kTRUE;
 786
 787         while ((entry = dynamic_cast<TSystemFile*> (dirIter.Next())))
 788         {
 789                 if (entry->IsDirectory())
 790                         continue;
 791
 792                 TString fileName(entry->GetName());
 793                 if (!fileName.BeginsWith(begin))
 794                         continue;
 795
 796                 nDirs++;
 797
 798                 if (first)
 799                 {
 800                         first = kFALSE;
 801                         // check that folder exists, otherwise create it
 802                         TGridResult* result = gGrid->Ls(alienDir.Data(), "a");
 803
 804                         if (!result)
 805                         {
 806                                 delete dirList;
 807                                 return kFALSE;
 808                         }
 809
 810                         if (!result->GetFileName(1)) // TODO: It looks like element 0 is always 0!!
 811                         {
 812                                 // TODO It does not work currently! Bug in TAliEn::Mkdir
 813                                 // TODO Manually fixed in local root v5-16-00
 814                                 if (!gGrid->Mkdir(alienDir.Data(),"-p",0))
 815                                 {
 816                                         Log("SHUTTLE", Form("CopyFilesToGrid - Cannot create directory %s",
 817                                                         alienDir.Data()));
 818                                         delete dirList;
 819                                         return kFALSE;
 820                                 } else {
 821                                         Log("SHUTTLE",Form("CopyFilesToGrid - Folder %s created", alienDir.Data()));
 822                                 }
 823
 824                         } else {
 825                                         Log("SHUTTLE",Form("CopyFilesToGrid - Folder %s found", alienDir.Data()));
 826                         }
 827                 }
 828
 829                 TString fullLocalPath;
 830                 fullLocalPath.Form("%s/%s", dir.Data(), fileName.Data());
 831
 832                 TString fullGridPath;
 833                 fullGridPath.Form("alien://%s/%s", alienDir.Data(), fileName.Data());
 834
 835                 Bool_t result = TFile::Cp(fullLocalPath, fullGridPath);
 836
 837                 if (result)
 838                 {
 839                         Log("SHUTTLE", Form("CopyFilesToGrid - Copying local file %s to %s succeeded!",
 840                                                 fullLocalPath.Data(), fullGridPath.Data()));
 841                         RemoveFile(fullLocalPath);
 842                         nTransfer++;
 843                 }
 844                 else
 845                 {
 846                         Log("SHUTTLE", Form("CopyFilesToGrid - Copying local file %s to %s FAILED!",
 847                                                 fullLocalPath.Data(), fullGridPath.Data()));
 848                         success = kFALSE;
 849                 }
 850         }
 851
 852         Log("SHUTTLE", Form("CopyFilesToGrid - %d (over %d) files in folder %s copied to Grid.",
 853                                                 nTransfer, nDirs, dir.Data()));
 854
 855
 856         delete dirList;
 857         return success;
 858 }
 859
 860 //______________________________________________________________________________________________
 861 const char* AliShuttle::GetRefFilePrefix(const char* base, const char* detector)
 862 {
 863         //
 864         // Get folder name of reference files
 865         //
 866
 867         TString offDetStr(GetOfflineDetName(detector));
 868         TString dir;
 869         if (offDetStr == "ITS" || offDetStr == "MUON" || offDetStr == "PHOS")
 870         {
 871                 dir.Form("%s/%s/%s", base, offDetStr.Data(), detector);
 872         } else {
 873                 dir.Form("%s/%s", base, offDetStr.Data());
 874         }
 875
 876         return dir.Data();
 877
 878
 879 }
 880
 881 //______________________________________________________________________________________________
 882 void AliShuttle::CleanLocalStorage(const TString& uri)
 883 {
 884         //
 885         // Called in case the preprocessor is declared failed. Remove remaining objects from the local storages.
 886         //
 887
 888         const char* type = 0;
 889         if(uri == fgkLocalCDB) {
 890                 type = "OCDB";
 891         } else if(uri == fgkLocalRefStorage) {
 892                 type = "Reference";
 893         } else {
 894                 AliError(Form("Invalid storage URI: %s", uri.Data()));
 895                 return;
 896         }
 897
 898         AliCDBManager* man = AliCDBManager::Instance();
 899
 900         // open local storage
 901         AliCDBStorage *localSto = man->GetStorage(uri);
 902         if(!localSto) {
 903                 Log("SHUTTLE",
 904                         Form("CleanLocalStorage - cannot activate local %s storage", type));
 905                 return;
 906         }
 907
 908         TString filename(Form("%s/%s/*/Run*_v%d_s*.root",
 909                 localSto->GetBaseFolder().Data(), GetOfflineDetName(fCurrentDetector.Data()), GetCurrentRun()));
 910
 911         AliDebug(2, Form("filename = %s", filename.Data()));
 912
 913         Log("SHUTTLE", Form("Removing remaining local files for run %d and detector %s ...",
 914                 GetCurrentRun(), fCurrentDetector.Data()));
 915
 916         RemoveFile(filename.Data());
 917
 918 }
 919
 920 //______________________________________________________________________________________________
 921 void AliShuttle::RemoveFile(const char* filename)
 922 {
 923         //
 924         // removes local file
 925         //
 926
 927         TString command(Form("rm -f %s", filename));
 928
 929         Int_t result = gSystem->Exec(command.Data());
 930         if(result != 0)
 931         {
 932                 Log("SHUTTLE", Form("RemoveFile - %s: Cannot remove file %s!",
 933                         fCurrentDetector.Data(), filename));
 934         }
 935 }
 936
 937 //______________________________________________________________________________________________
 938 AliShuttleStatus* AliShuttle::ReadShuttleStatus()
 939 {
 940         //
 941         // Reads the AliShuttleStatus from the CDB
 942         //
 943
 944         if (fStatusEntry){
 945                 delete fStatusEntry;
 946                 fStatusEntry = 0;
 947         }
 948
 949         Int_t path1 = GetCurrentRun()/10000;
 950         fStatusEntry = AliCDBManager::Instance()->GetStorage(GetLocalCDB())
 951                 ->Get(Form("/SHUTTLE/%s/%d", fCurrentDetector.Data(), path1), GetCurrentRun());
 952
 953         if (!fStatusEntry) return 0;
 954         fStatusEntry->SetOwner(1);
 955
 956         AliShuttleStatus* status = dynamic_cast<AliShuttleStatus*> (fStatusEntry->GetObject());
 957         if (!status) {
 958                 AliError("Invalid object stored to CDB!");
 959                 return 0;
 960         }
 961
 962         return status;
 963 }
 964
 965 //______________________________________________________________________________________________
 966 Bool_t AliShuttle::WriteShuttleStatus(AliShuttleStatus* status)
 967 {
 968         //
 969         // writes the status for one subdetector
 970         //
 971
 972         if (fStatusEntry){
 973                 delete fStatusEntry;
 974                 fStatusEntry = 0;
 975         }
 976
 977         Int_t run = GetCurrentRun();
 978         Int_t path1 = run/10000;
 979         TString path1_string = Form("%d",path1);
 980
 981         AliCDBId id(AliCDBPath("SHUTTLE", fCurrentDetector, path1_string), run, run);
 982
 983         fStatusEntry = new AliCDBEntry(status, id, new AliCDBMetaData);
 984         fStatusEntry->SetOwner(1);
 985
 986         UInt_t result = AliCDBManager::Instance()->GetStorage(fgkLocalCDB)->Put(fStatusEntry);
 987
 988         if (!result) {
 989                 Log("SHUTTLE", Form("WriteShuttleStatus - Failed for %s, run %d",
 990                                                 fCurrentDetector.Data(), run));
 991                 return kFALSE;
 992         }
 993
 994         SendMLDetInfo();
 995
 996         return kTRUE;
 997 }
 998
 999 //______________________________________________________________________________________________
1000 void AliShuttle::UpdateShuttleStatus(AliShuttleStatus::Status newStatus, Bool_t increaseCount)
1001 {
1002         //
1003         // changes the AliShuttleStatus for the given detector and run to the given status
1004         //
1005
1006         if (!fStatusEntry){
1007                 AliError("UNEXPECTED: fStatusEntry empty");
1008                 return;
1009         }
1010
1011         AliShuttleStatus* status = dynamic_cast<AliShuttleStatus*> (fStatusEntry->GetObject());
1012
1013         if (!status){
1014                 Log("SHUTTLE", "UpdateShuttleStatus - UNEXPECTED: status could not be read from current CDB entry");
1015                 return;
1016         }
1017
1018         TString actionStr = Form("UpdateShuttleStatus - %s: Changing state from %s to %s",
1019                                 fCurrentDetector.Data(),
1020                                 status->GetStatusName(),
1021                                 status->GetStatusName(newStatus));
1022         Log("SHUTTLE", actionStr);
1023         SetLastAction(actionStr);
1024
1025         status->SetStatus(newStatus);
1026         if (increaseCount) status->IncreaseCount();
1027
1028         AliCDBManager::Instance()->GetStorage(fgkLocalCDB)->Put(fStatusEntry);
1029
1030         SendMLDetInfo();
1031 }
1032
1033 //______________________________________________________________________________________________
1034 void AliShuttle::SendMLDetInfo()
1035 {
1036         //
1037         // sends ML information about the current status of the current detector being processed
1038         //
1039
1040         AliShuttleStatus* status = dynamic_cast<AliShuttleStatus*> (fStatusEntry->GetObject());
1041
1042         if (!status){
1043                 Log("SHUTTLE", "SendMLDetInfo - UNEXPECTED: status could not be read from current CDB entry");
1044                 return;
1045         }
1046
1047         TMonaLisaText  mlStatus(Form("%s_status", fCurrentDetector.Data()), status->GetStatusName());
1048         TMonaLisaValue mlRetryCount(Form("%s_count", fCurrentDetector.Data()), status->GetCount());
1049
1050         TList mlList;
1051         mlList.Add(&mlStatus);
1052         mlList.Add(&mlRetryCount);
1053
1054         TString mlID;
1055         mlID.Form("%d", GetCurrentRun());
1056         fMonaLisa->SendParameters(&mlList, mlID);
1057 }
1058
1059 //______________________________________________________________________________________________
1060 Bool_t AliShuttle::ContinueProcessing()
1061 {
1062         // this function reads the AliShuttleStatus information from CDB and
1063         // checks if the processing should be continued
1064         // if yes it returns kTRUE and updates the AliShuttleStatus with nextStatus
1065
1066         if (!fConfig->HostProcessDetector(fCurrentDetector))
1067                 return kFALSE;
1068
1069         AliPreprocessor* aPreprocessor =
1070                 dynamic_cast<AliPreprocessor*> (fPreprocessorMap.GetValue(fCurrentDetector));
1071         if (!aPreprocessor)
1072         {
1073                 Log("SHUTTLE", Form("ContinueProcessing - %s: no preprocessor registered", fCurrentDetector.Data()));
1074                 return kFALSE;
1075         }
1076
1077         AliShuttleLogbookEntry::Status entryStatus =
1078                 fLogbookEntry->GetDetectorStatus(fCurrentDetector);
1079
1080         if (entryStatus != AliShuttleLogbookEntry::kUnprocessed) {
1081                 Log("SHUTTLE", Form("ContinueProcessing - %s is %s",
1082                                 fCurrentDetector.Data(),
1083                                 fLogbookEntry->GetDetectorStatusName(entryStatus)));
1084                 return kFALSE;
1085         }
1086
1087         // if we get here, according to Shuttle logbook subdetector is in UNPROCESSED state
1088
1089         // check if current run is first unprocessed run for current detector
1090         if (fConfig->StrictRunOrder(fCurrentDetector) &&
1091                 !fFirstUnprocessed[GetDetPos(fCurrentDetector)])
1092         {
1093                 if (fTestMode == kNone)
1094                 {
1095                         Log("SHUTTLE", Form("ContinueProcessing - %s requires strict run ordering"
1096                                         " but this is not the first unprocessed run!"));
1097                         return kFALSE;
1098                 }
1099                 else
1100                 {
1101                         Log("SHUTTLE", Form("ContinueProcessing - In TESTMODE - "
1102                                         "Although %s requires strict run ordering "
1103                                         "and this is not the first unprocessed run, "
1104                                         "the SHUTTLE continues"));
1105                 }
1106         }
1107
1108         // Is the subdetector processed first time for this run?
1109         fFirstProcessing = kFALSE;
1110
1111         AliShuttleStatus* status = ReadShuttleStatus();
1112         if (!status) {
1113                 // first time
1114                 Log("SHUTTLE", Form("ContinueProcessing - %s: Processing first time",
1115                                 fCurrentDetector.Data()));
1116                 status = new AliShuttleStatus(AliShuttleStatus::kStarted);
1117                 fFirstProcessing = kTRUE;
1118                 return WriteShuttleStatus(status);
1119         }
1120
1121         // The following case shouldn't happen if Shuttle Logbook was correctly updated.
1122         // If it happens it may mean Logbook updating failed... let's do it now!
1123         if (status->GetStatus() == AliShuttleStatus::kDone ||
1124             status->GetStatus() == AliShuttleStatus::kFailed ||
1125             status->GetStatus() == AliShuttleStatus::kSkipped) {
1126                 Log("SHUTTLE", Form("ContinueProcessing - %s is already %s. Updating Shuttle Logbook",
1127                                         fCurrentDetector.Data(),
1128                                         status->GetStatusName(status->GetStatus())));
1129
1130                 if (status->GetStatus() == AliShuttleStatus::kSkipped)
1131                 {
1132                         UpdateShuttleLogbook(fCurrentDetector.Data(), "DONE");
1133                 }
1134                 else
1135                         UpdateShuttleLogbook(fCurrentDetector.Data(), status->GetStatusName(status->GetStatus()));
1136
1137                 return kFALSE;
1138         }
1139
1140         if (status->GetStatus() == AliShuttleStatus::kStoreStarted || status->GetStatus() == AliShuttleStatus::kStoreDelayed ||status->GetStatus() == AliShuttleStatus::kStoreError) {
1141                 Log("SHUTTLE",
1142                         Form("ContinueProcessing - %s: Grid storage of one or more "
1143                                 "objects failed. Trying again now",
1144                                 fCurrentDetector.Data()));
1145                 StoreOCDB();
1146                 return kFALSE;
1147         }
1148
1149         // if we get here, there is a restart
1150         Bool_t cont = kFALSE;
1151
1152         // abort conditions
1153         if (status->GetCount() >= fConfig->GetMaxRetries()) {
1154                 Log("SHUTTLE", Form("ContinueProcessing - %s failed %d times in status %s - "
1155                                 "Updating Shuttle Logbook", fCurrentDetector.Data(),
1156                                 status->GetCount(), status->GetStatusName()));
1157                 UpdateShuttleLogbook(fCurrentDetector.Data(), "FAILED");
1158                 UpdateShuttleStatus(AliShuttleStatus::kFailed);
1159
1160                 // there may still be objects in local OCDB and reference storage
1161                 // and FXS databases may be not updated: do it now!
1162
1163                 // TODO Currently disabled, we want to keep files in case of failure!
1164                 // CleanLocalStorage(fgkLocalCDB);
1165                 // CleanLocalStorage(fgkLocalRefStorage);
1166                 // UpdateTableFailCase();
1167
1168                 // Send mail to detector expert!
1169                 Log("SHUTTLE", Form("ContinueProcessing - Sending mail to %s expert...",
1170                                     fCurrentDetector.Data()));
1171                 if (!SendMail(kPPEMail))
1172                         Log("SHUTTLE", Form("ContinueProcessing - Could not send mail to %s expert",
1173                                             fCurrentDetector.Data()));
1174
1175         } else {
1176                 Log("SHUTTLE", Form("ContinueProcessing - %s: restarting. "
1177                                 "Aborted before with %s. Retry number %d.", fCurrentDetector.Data(),
1178                                 status->GetStatusName(), status->GetCount()));
1179                 Bool_t increaseCount = kTRUE;
1180                 if (status->GetStatus() == AliShuttleStatus::kDCSError ||
1181                     status->GetStatus() == AliShuttleStatus::kDCSStarted ||
1182                     status->GetStatus() == AliShuttleStatus::kFXSError)
1183                                 increaseCount = kFALSE;
1184
1185                 UpdateShuttleStatus(AliShuttleStatus::kStarted, increaseCount);
1186                 cont = kTRUE;
1187         }
1188
1189         return cont;
1190 }
1191
1192 //______________________________________________________________________________________________
1193 void AliShuttle::SendMLRunInfo(const char* status)
1194 {
1195         //
1196         // Send information about this run to ML
1197
1198         TMonaLisaText  mlStatus("SHUTTLE_status", status);
1199         TString runType(fLogbookEntry->GetRunType());
1200         if (strlen(fLogbookEntry->GetRunParameter("log")) > 0){
1201
1202                 runType += "(";
1203                 runType += fLogbookEntry->GetRunParameter("log");
1204                 runType += ")";
1205         }
1206         if (fLogbookEntry->GetDATestMode()){
1207                 runType += " (DATest)";
1208         }
1209         TMonaLisaText  mlRunType("SHUTTLE_runtype", runType);
1210
1211         TList mlList;
1212         mlList.Add(&mlStatus);
1213         mlList.Add(&mlRunType);
1214
1215         TString mlID;
1216         mlID.Form("%d", GetCurrentRun());
1217         fMonaLisa->SendParameters(&mlList, mlID);
1218 }
1219
1220 //______________________________________________________________________________________________
1221 Int_t AliShuttle::GetMem(Int_t pid)
1222 {
1223         // invokes ps to get the memory consumption of the process <pid>
1224         // returns -1 in case of error
1225
1226         TString checkStr;
1227         checkStr.Form("ps -o vsize --pid %d | tail -n 1", pid);
1228         FILE* pipe = gSystem->OpenPipe(checkStr, "r");
1229         if (!pipe)
1230         {
1231                 Log("SHUTTLE", Form("Process - Error: "
1232                         "Could not open pipe to %s", checkStr.Data()));
1233                 return -1;
1234         }
1235
1236         char buffer[100];
1237         if (!fgets(buffer, 100, pipe))
1238         {
1239                 Log("SHUTTLE", "Process - Error: ps did not return anything");
1240                 gSystem->ClosePipe(pipe);
1241                 return -1;
1242         }
1243         gSystem->ClosePipe(pipe);
1244
1245         //Log("SHUTTLE", Form("ps returned %s", buffer));
1246
1247         Int_t mem = 0;
1248         if ((sscanf(buffer, "%d\n", &mem) != 1) || !mem)
1249         {
1250                 Log("SHUTTLE", "Process - Error: Could not parse output of ps");
1251                 return -1;
1252         }
1253
1254         return mem;
1255 }
1256
1257 //______________________________________________________________________________________________
1258 Bool_t AliShuttle::Process(AliShuttleLogbookEntry* entry)
1259 {
1260         //
1261         // Makes data retrieval for all detectors in the configuration.
1262         // entry: Shuttle logbook entry, contains run paramenters and status of detectors
1263         // (Unprocessed, Inactive, Failed or Done).
1264         // Returns kFALSE in case of error occured and kTRUE otherwise
1265         //
1266
1267         if (!entry) return kFALSE;
1268
1269         fLogbookEntry = entry;
1270
1271         Log("SHUTTLE", Form("\t\t\t^*^*^*^*^*^*^*^*^*^*^*^* run %d: START ^*^*^*^*^*^*^*^*^*^*^*^*",
1272                                         GetCurrentRun()));
1273
1274         CountOpenRuns();
1275
1276         // Send the information to ML
1277         SendMLRunInfo("Processing");
1278
1279         if (fLogbookEntry->IsDone())
1280         {
1281                 Log("SHUTTLE","Process - Shuttle is already DONE. Updating logbook");
1282                 UpdateShuttleLogbook("shuttle_done");
1283                 fLogbookEntry = 0;
1284                 return kTRUE;
1285         }
1286
1287         // read test mode if flag is set
1288         if (fReadTestMode)
1289         {
1290                 fTestMode = kNone;
1291                 TString logEntry(entry->GetRunParameter("log"));
1292                 //printf("log entry = %s\n", logEntry.Data());
1293                 TString searchStr("Testmode: ");
1294                 Int_t pos = logEntry.Index(searchStr.Data());
1295                 //printf("%d\n", pos);
1296                 if (pos >= 0)
1297                 {
1298                         TSubString subStr = logEntry(pos + searchStr.Length(), logEntry.Length());
1299                         //printf("%s\n", subStr.String().Data());
1300                         TString newStr(subStr.Data());
1301                         TObjArray* token = newStr.Tokenize(' ');
1302                         if (token)
1303                         {
1304                                 //token->Print();
1305                                 TObjString* tmpStr = dynamic_cast<TObjString*> (token->First());
1306                                 if (tmpStr)
1307                                 {
1308                                         Int_t testMode = tmpStr->String().Atoi();
1309                                         if (testMode > 0)
1310                                         {
1311                                                 Log("SHUTTLE", Form("Process - Enabling test mode %d", testMode));
1312                                                 SetTestMode((TestMode) testMode);
1313                                         }
1314                                 }
1315                                 delete token;
1316                         }
1317                 }
1318         }
1319
1320         fLogbookEntry->Print("all");
1321
1322         // Initialization
1323         Bool_t hasError = kFALSE;
1324
1325         // Set the CDB and Reference folders according to the year
1326
1327         // build cdb paths (repeat each time, run might be a DATest run)
1328         if (!fLogbookEntry->GetDATestMode()){
1329                 fgkMainCDB.Form("alien://folder=%s%d/OCDB?user=alidaq?cacheFold=/tmp/OCDBCache",
1330                                 fConfig->GetAlienPath(), GetCurrentYear());
1331
1332                 fgkMainRefStorage.Form("alien://folder=%s%d/Reference?user=alidaq?cacheFold=/tmp/OCDBCache",
1333                                        fConfig->GetAlienPath(), GetCurrentYear());
1334         }
1335         else {
1336                 fgkMainCDB.Form("alien://folder=%s%d/DATest/OCDB?user=alidaq?cacheFold=/tmp/OCDBCache",
1337                                 fConfig->GetAlienPath(), GetCurrentYear());
1338
1339                 fgkMainRefStorage.Form("alien://folder=%s%d/DATest/Reference?user=alidaq?cacheFold=/tmp/OCDBCache",
1340                                        fConfig->GetAlienPath(), GetCurrentYear());
1341         }
1342
1343         AliDebug(2,Form("Main CDB storage = %s",fgkMainCDB.Data()));
1344         AliDebug(2,Form("Main Reference storage = %s",fgkMainRefStorage.Data()));
1345
1346         // Loop on detectors in the configuration
1347         TIter iter(fConfig->GetDetectors());
1348         TObjString* aDetector = 0;
1349
1350         Bool_t first = kTRUE;
1351
1352         while ((aDetector = (TObjString*) iter.Next()))
1353         {
1354                 fCurrentDetector = aDetector->String();
1355
1356                 if (ContinueProcessing() == kFALSE)
1357                         continue;
1358
1359                 if (first)
1360                 {
1361                   // only read QueryCDB when needed and only once
1362                   AliCDBStorage *mainCDBSto = AliCDBManager::Instance()->GetStorage(fgkMainCDB);
1363                   if(mainCDBSto) mainCDBSto->QueryCDB(GetCurrentRun());
1364                   AliCDBStorage *mainRefSto = AliCDBManager::Instance()->GetStorage(fgkMainRefStorage);
1365                   if(mainRefSto) mainRefSto->QueryCDB(GetCurrentRun());
1366                   first = kFALSE;
1367                 }
1368
1369                 Log("SHUTTLE", Form("\t\t\t****** run %d - %s: START  ******",
1370                                                 GetCurrentRun(), aDetector->GetName()));
1371
1372                 for(Int_t iSys=0;iSys<3;iSys++) fFXSCalled[iSys]=kFALSE;
1373
1374                 Int_t initialMem = GetMem(getpid());
1375                 Log("SHUTTLE", Form("Memory consumption before forking is %d", initialMem));
1376
1377                 Log(fCurrentDetector.Data(), "Process - Starting processing");
1378
1379                 Int_t pid = fork();
1380
1381                 if (pid < 0)
1382                 {
1383                         Log("SHUTTLE", "Process - ERROR: Forking failed");
1384                 }
1385                 else if (pid > 0)
1386                 {
1387                         // parent
1388                         Log("SHUTTLE", Form("Process - In parent process of %d - %s: Starting monitoring",
1389                                                         GetCurrentRun(), aDetector->GetName()));
1390
1391                         Long_t begin = time(0);
1392
1393                         int status; // to be used with waitpid, on purpose an int (not Int_t)!
1394                         while (waitpid(pid, &status, WNOHANG) == 0)
1395                         {
1396                                 Long_t expiredTime = time(0) - begin;
1397
1398                                 if (expiredTime > fConfig->GetPPTimeOut())
1399                                 {
1400                                         TString logMsg;
1401                                         AliShuttleStatus *currentStatus = ReadShuttleStatus();
1402                                         AliShuttleStatus::Status newStatus = AliShuttleStatus::kInvalid;
1403
1404                                         if (currentStatus->GetStatus() <= AliShuttleStatus::kPPDone)
1405                                         {
1406                                                 // in case pp not yet done set status to kPPTimeOut
1407
1408                                                 logMsg.Form("Process - Process of %s timed out. Run time: %d seconds. Killing...",
1409                                                                 fCurrentDetector.Data(), expiredTime);
1410                                                 newStatus = AliShuttleStatus::kPPTimeOut;
1411                                         }
1412                                         else if (currentStatus->GetStatus() == AliShuttleStatus::kStoreStarted)
1413                                         {
1414                                                 // in case the pp goes in TimeOut while storing the objects in the OCDB
1415                                                 // set status to kStoreError
1416
1417                                                 logMsg.Form("Process - Process of %s timed out while storing the OCDB object. Run time: %d seconds. Killing... and setting status to StoreError.",
1418                                                                 fCurrentDetector.Data(), expiredTime);
1419                                                 newStatus = AliShuttleStatus::kStoreError;
1420                                         }
1421                                         else
1422                                         {
1423                                                 // in other cases don't change the status
1424
1425                                                 logMsg.Form("Process - Process of %s timed out in status = %s. Run time: %d seconds. Killing... without changing the status",
1426                                                                 fCurrentDetector.Data(), currentStatus->GetStatusName(), expiredTime);
1427                                         }
1428
1429                                         Log("SHUTTLE", logMsg);
1430                                         Log(fCurrentDetector, logMsg);
1431
1432                                         kill(pid, 9);
1433
1434                                         if (newStatus != AliShuttleStatus::kInvalid)
1435                                                 UpdateShuttleStatus(newStatus);
1436                                         hasError = kTRUE;
1437
1438                                         gSystem->Sleep(1000);
1439                                 }
1440                                 else
1441                                 {
1442                                         gSystem->Sleep(1000);
1443
1444                                         Int_t mem = GetMem(pid);
1445
1446                                         if (mem < 0)
1447                                                 continue;
1448
1449                                         mem -= initialMem;
1450                                         if (mem < 0)
1451                                                 mem = 0;
1452
1453                                         if (expiredTime % 60 == 0)
1454                                         {
1455                                                 Log("SHUTTLE", Form("Process - %s: Checking process. "
1456                                                         "Run time: %d seconds - Memory consumption: %d KB",
1457                                                         fCurrentDetector.Data(), expiredTime, mem));
1458                                                 SendAlive();
1459                                         }
1460
1461                                         if (mem > fConfig->GetPPMaxMem())
1462                                         {
1463                                                 TString tmp;
1464                                                 tmp.Form("Process - Process exceeds maximum allowed memory "
1465                                                         "(%d KB > %d KB). Killing...",
1466                                                         mem, fConfig->GetPPMaxMem());
1467                                                 Log("SHUTTLE", tmp);
1468                                                 Log(fCurrentDetector, tmp);
1469
1470                                                 kill(pid, 9);
1471
1472                                                 UpdateShuttleStatus(AliShuttleStatus::kPPOutOfMemory);
1473                                                 hasError = kTRUE;
1474
1475                                                 gSystem->Sleep(1000);
1476                                         }
1477                                 }
1478                         }
1479
1480                         Log("SHUTTLE", Form("Process - In parent process of %d - %s: Client has terminated.",
1481                                                                 GetCurrentRun(), aDetector->GetName()));
1482
1483                         if (WIFEXITED(status))
1484                         {
1485                                 Int_t returnCode = WEXITSTATUS(status);
1486
1487                                 Log("SHUTTLE", Form("Process - %s: the return code is %d", fCurrentDetector.Data(),
1488                                                                                 returnCode));
1489
1490                                 if (returnCode == 0) hasError = kTRUE;
1491                         }
1492                 }
1493                 else if (pid == 0)
1494                 {
1495                         // child
1496                         Log("SHUTTLE", Form("Process - In child process of %d - %s", GetCurrentRun(),
1497                                 aDetector->GetName()));
1498
1499                         Log("SHUTTLE", Form("Process - Redirecting output to %s log",fCurrentDetector.Data()));
1500
1501                         if ((freopen(GetLogFileName(fCurrentDetector), "a", stdout)) == 0)
1502                         {
1503                                 Log("SHUTTLE", "Process - Could not freopen stdout");
1504                         }
1505                         else
1506                         {
1507                                 fOutputRedirected = kTRUE;
1508                                 if ((dup2(fileno(stdout), fileno(stderr))) < 0)
1509                                         Log("SHUTTLE", "Process - Could not redirect stderr");
1510
1511                         }
1512
1513                         TString wd = gSystem->WorkingDirectory();
1514                         Int_t dir_lev1 = GetCurrentRun()/10000;
1515                         TString tmpDir = Form("%s/%d/%d/%s_process", GetShuttleTempDir(),
1516                                 dir_lev1, GetCurrentRun(), fCurrentDetector.Data());
1517
1518                         Int_t result = gSystem->GetPathInfo(tmpDir.Data(), 0, (Long64_t*) 0, 0, 0);
1519                         if (!result) // temp dir already exists!
1520                         {
1521                                 Log(fCurrentDetector.Data(),
1522                                         Form("Process - %s dir already exists! Removing...", tmpDir.Data()));
1523                                 gSystem->Exec(Form("rm -rf %s",tmpDir.Data()));
1524                         }
1525
1526                         if (gSystem->mkdir(tmpDir.Data(), 1))
1527                         {
1528                                 Log(fCurrentDetector.Data(), "Process - could not make temp directory!!");
1529                                 gSystem->Exit(1);
1530                         }
1531
1532                         if (!gSystem->ChangeDirectory(tmpDir.Data()))
1533                         {
1534                                 Log(fCurrentDetector.Data(), "Process - could not change directory!!");
1535                                 gSystem->Exit(1);
1536                         }
1537
1538                         Int_t success = ProcessCurrentDetector();
1539
1540                         gSystem->ChangeDirectory(wd.Data());
1541
1542                         if (success == 1) // Preprocessor finished successfully!
1543                         {
1544                                 // remove temporary folder or DCS map
1545                                 if (!fConfig->KeepTempFolder())
1546                                 {
1547                                         gSystem->Exec(Form("rm -rf %s",tmpDir.Data()));
1548                                 } else if (!fConfig->KeepDCSMap())
1549                                 {
1550                                         gSystem->Exec(Form("rm -f %s/DCSMap.root",tmpDir.Data()));
1551                                 }
1552
1553                                 // Update time_processed field in FXS DB
1554                                 if (UpdateTable() == kFALSE)
1555                                         Log("SHUTTLE", Form("Process - %s: Could not update FXS databases!",
1556                                                         fCurrentDetector.Data()));
1557
1558                                 // Transfer the data from local storage to main storage (Grid)
1559                                 if (StoreOCDB() == kFALSE)
1560                                         success = kFALSE;
1561                         }
1562                         else if (success == 0)
1563                         {
1564                                 Log("SHUTTLE",
1565                                         Form("\t\t\t****** run %d - %s: PP ERROR ******",
1566                                                 GetCurrentRun(), aDetector->GetName()));
1567                         }
1568
1569                         for (UInt_t iSys=0; iSys<3; iSys++)
1570                         {
1571                                 if (fFXSCalled[iSys]) fFXSlist[iSys].Clear();
1572                         }
1573
1574                         Log("SHUTTLE", Form("Process - Client process of %d - %s is exiting now with %d.",
1575                                                         GetCurrentRun(), aDetector->GetName(), success));
1576
1577                         // the client exits here
1578                         gSystem->Exit(success);
1579
1580                         AliError("We should never get here!!!");
1581                 }
1582         }
1583
1584         Log("SHUTTLE", Form("\t\t\t^*^*^*^*^*^*^*^*^*^*^*^* run %d: FINISH ^*^*^*^*^*^*^*^*^*^*^*^*",
1585                                                         GetCurrentRun()));
1586
1587         //check if shuttle is done for this run, if so update logbook
1588         TObjArray checkEntryArray;
1589         checkEntryArray.SetOwner(1);
1590         TString whereClause = Form("where run=%d", GetCurrentRun());
1591         if (!QueryShuttleLogbook(whereClause.Data(), checkEntryArray) ||
1592                         checkEntryArray.GetEntries() == 0) {
1593                 Log("SHUTTLE", Form("Process - Warning: Cannot check status of run %d on Shuttle logbook!",
1594                                                 GetCurrentRun()));
1595                 return hasError == kFALSE;
1596         }
1597
1598         AliShuttleLogbookEntry* checkEntry = dynamic_cast<AliShuttleLogbookEntry*>
1599                                                 (checkEntryArray.At(0));
1600
1601         if (checkEntry)
1602         {
1603                 if (checkEntry->IsDone())
1604                 {
1605                         Log("SHUTTLE","Process - Shuttle is DONE. Updating logbook");
1606                         UpdateShuttleLogbook("shuttle_done");
1607                 }
1608                 else
1609                 {
1610                         for (UInt_t iDet=0; iDet<NDetectors(); iDet++)
1611                         {
1612                                 if (checkEntry->GetDetectorStatus(iDet) == AliShuttleLogbookEntry::kUnprocessed)
1613                                 {
1614                                         AliDebug(2, Form("Run %d: setting %s as \"not first time unprocessed\"",
1615                                                         checkEntry->GetRun(), GetDetName(iDet)));
1616                                         fFirstUnprocessed[iDet] = kFALSE;
1617                                 }
1618                         }
1619                         SendMLRunInfo("Pending");
1620                 }
1621         }
1622
1623         fLogbookEntry = 0;
1624
1625         return hasError == kFALSE;
1626 }
1627
1628 //______________________________________________________________________________________________
1629 Int_t AliShuttle::ProcessCurrentDetector()
1630 {
1631         //
1632         // Makes data retrieval just for a specific detector (fCurrentDetector).
1633         // Threre should be a configuration for this detector.
1634
1635         Log("SHUTTLE", Form("ProcessCurrentDetector - Retrieving values for %s, run %d",
1636                                                 fCurrentDetector.Data(), GetCurrentRun()));
1637
1638         TString wd = gSystem->WorkingDirectory();
1639
1640         if (!CleanReferenceStorage(fCurrentDetector.Data()))
1641                 return 0;
1642
1643         gSystem->ChangeDirectory(wd.Data());
1644
1645         // call preprocessor
1646         AliPreprocessor* aPreprocessor =
1647                 dynamic_cast<AliPreprocessor*> (fPreprocessorMap.GetValue(fCurrentDetector));
1648
1649         // check if the preprocessor wants to process this run type
1650         if (aPreprocessor->ProcessRunType() == kFALSE)
1651         {
1652                 UpdateShuttleStatus(AliShuttleStatus::kSkipped);
1653                 UpdateShuttleLogbook(fCurrentDetector, "DONE");
1654                 if (!UpdateTableSkippedCase(fCurrentDetector.Data()))
1655                 {
1656                         AliError(Form("Could not update FXS tables for run %d !", GetCurrentRun()));
1657                 }
1658                 Log(fCurrentDetector, Form("ProcessCurrentDetector - %s preprocessor is not interested in this run type", fCurrentDetector.Data()));
1659
1660                 return 2;
1661         }
1662
1663         TMap* dcsMap = new TMap();
1664
1665         aPreprocessor->Initialize(GetCurrentRun(), GetCurrentStartTime(), GetCurrentEndTime());
1666
1667         Bool_t processDCS = aPreprocessor->ProcessDCS();
1668
1669         if (!processDCS)
1670         {
1671                 Log(fCurrentDetector, "ProcessCurrentDetector -"
1672                         " The preprocessor requested to skip the retrieval of DCS values");
1673         }
1674         else if (fTestMode & kSkipDCS)
1675         {
1676                 Log(fCurrentDetector, "ProcessCurrentDetector - In TESTMODE: Skipping DCS processing");
1677         }
1678         else if (fTestMode & kErrorDCS)
1679         {
1680                 Log(fCurrentDetector, "ProcessCurrentDetector - In TESTMODE: Simulating DCS error");
1681                 UpdateShuttleStatus(AliShuttleStatus::kDCSStarted);
1682                 UpdateShuttleStatus(AliShuttleStatus::kDCSError);
1683                 delete dcsMap;
1684                 return 0;
1685         } else {
1686
1687                 UpdateShuttleStatus(AliShuttleStatus::kDCSStarted);
1688
1689                 // Query DCS archive
1690                 Int_t nServers = fConfig->GetNServers(fCurrentDetector);
1691
1692                 for (int iServ=0; iServ<nServers; iServ++)
1693                 {
1694
1695                         TString host(fConfig->GetDCSHost(fCurrentDetector, iServ));
1696                         Int_t port = fConfig->GetDCSPort(fCurrentDetector, iServ);
1697                         Int_t multiSplit = fConfig->GetMultiSplit(fCurrentDetector, iServ);
1698
1699                         Log(fCurrentDetector, Form("ProcessCurrentDetector -"
1700                                         " Querying DCS Amanda server %s:%d (%d of %d)",
1701                                         host.Data(), port, iServ+1, nServers));
1702
1703                         TMap* aliasMap = 0;
1704                         TMap* dpMap = 0;
1705
1706                         if (fConfig->GetDCSAliases(fCurrentDetector, iServ)->GetEntries() > 0)
1707                         {
1708                                 Log(fCurrentDetector, Form("Querying %d DCS aliases", fConfig->GetDCSAliases(fCurrentDetector, iServ)->GetEntries()));
1709                                 aliasMap = GetValueSet(host, port,
1710                                                 fConfig->GetDCSAliases(fCurrentDetector, iServ),
1711                                                 kAlias, multiSplit);
1712                                 if (!aliasMap)
1713                                 {
1714                                         Log(fCurrentDetector,
1715                                                 Form("ProcessCurrentDetector -"
1716                                                         " Error retrieving DCS aliases from server %s."
1717                                                         " Sending mail to DCS experts!", host.Data()));
1718                                         UpdateShuttleStatus(AliShuttleStatus::kDCSError);
1719
1720                                         if (!SendMail(kDCSEMail))
1721                                                 Log("SHUTTLE", Form("ProcessCurrentDetector - "
1722                                                                     "Could not send mail to DCS experts!"));
1723
1724                                         delete dcsMap;
1725                                         return 0;
1726                                 }
1727                         }
1728
1729                         if (fConfig->GetDCSDataPoints(fCurrentDetector, iServ)->GetEntries() > 0)
1730                         {
1731                                 Log(fCurrentDetector, Form("Querying %d DCS data points", fConfig->GetDCSDataPoints(fCurrentDetector, iServ)->GetEntries()));
1732                                 dpMap = GetValueSet(host, port,
1733                                                 fConfig->GetDCSDataPoints(fCurrentDetector, iServ),
1734                                                 kDP, multiSplit);
1735                                 if (!dpMap)
1736                                 {
1737                                         Log(fCurrentDetector,
1738                                                 Form("ProcessCurrentDetector -"
1739                                                         " Error retrieving DCS data points from server %s."
1740                                                         " Sending mail to DCS experts!", host.Data()));
1741                                         UpdateShuttleStatus(AliShuttleStatus::kDCSError);
1742
1743                                         if (!SendMail(kDCSEMail))
1744                                                 Log("SHUTTLE", Form("ProcessCurrentDetector - "
1745                                                                     "Could not send mail to DCS experts!"));
1746
1747                                         if (aliasMap) delete aliasMap;
1748                                         delete dcsMap;
1749                                         return 0;
1750                                 }
1751                         }
1752
1753                         // merge aliasMap and dpMap into dcsMap
1754                         if(aliasMap) {
1755                                 TIter iter(aliasMap);
1756                                 TObjString* key = 0;
1757                                 while ((key = (TObjString*) iter.Next()))
1758                                         dcsMap->Add(key, aliasMap->GetValue(key->String()));
1759
1760                                 aliasMap->SetOwner(kFALSE);
1761                                 delete aliasMap;
1762                         }
1763
1764                         if(dpMap) {
1765                                 TIter iter(dpMap);
1766                                 TObjString* key = 0;
1767                                 while ((key = (TObjString*) iter.Next()))
1768                                         dcsMap->Add(key, dpMap->GetValue(key->String()));
1769
1770                                 dpMap->SetOwner(kFALSE);
1771                                 delete dpMap;
1772                         }
1773                 }
1774         }
1775
1776         // save map into file, to help debugging in case of preprocessor error
1777         TFile* f = TFile::Open("DCSMap.root","recreate");
1778         f->cd();
1779         dcsMap->Write("DCSMap", TObject::kSingleKey);
1780         f->Close();
1781         delete f;
1782
1783         // DCS Archive DB processing successful. Call Preprocessor!
1784         UpdateShuttleStatus(AliShuttleStatus::kPPStarted);
1785
1786         fFXSError = -1; // this variable is kTRUE after ::Process if an FXS error occured
1787
1788         UInt_t returnValue = aPreprocessor->Process(dcsMap);
1789
1790         if (fFXSError!=-1) {
1791                 UpdateShuttleStatus(AliShuttleStatus::kFXSError);
1792                 SendMail(kFXSEMail, fFXSError);
1793                 dcsMap->DeleteAll();
1794                 delete dcsMap;
1795                 return 0;
1796         }
1797
1798         if (returnValue > 0) // Preprocessor error!
1799         {
1800                 Log(fCurrentDetector, Form("ProcessCurrentDetector - "
1801                                 "Preprocessor failed. Process returned %d.", returnValue));
1802                 UpdateShuttleStatus(AliShuttleStatus::kPPError);
1803                 dcsMap->DeleteAll();
1804                 delete dcsMap;
1805                 return 0;
1806         }
1807
1808         // preprocessor ok!
1809         UpdateShuttleStatus(AliShuttleStatus::kPPDone);
1810         Log(fCurrentDetector, Form("ProcessCurrentDetector - %s preprocessor returned success",
1811                                 fCurrentDetector.Data()));
1812
1813         dcsMap->DeleteAll();
1814         delete dcsMap;
1815
1816         return 1;
1817 }
1818
1819 //______________________________________________________________________________________________
1820 void AliShuttle::CountOpenRuns()
1821 {
1822         // Query DAQ's Shuttle logbook and sends the number of open runs to ML
1823
1824         SendAlive();
1825
1826         // check connection, in case connect
1827         if (!Connect(3))
1828                 return;
1829
1830         TString sqlQuery;
1831         sqlQuery = Form("select count(*) from %s where shuttle_done=0", fConfig->GetShuttlelbTable());
1832
1833         TSQLResult* aResult = fServer[3]->Query(sqlQuery);
1834         if (!aResult) {
1835                 AliError(Form("Can't execute query <%s>!", sqlQuery.Data()));
1836                 return;
1837         }
1838
1839         AliDebug(2,Form("Query = %s", sqlQuery.Data()));
1840
1841         if (aResult->GetRowCount() == 0) {
1842                 AliError(Form("No result for query %s received", sqlQuery.Data()));
1843                 return;
1844         }
1845
1846         if (aResult->GetFieldCount() != 1) {
1847                 AliError(Form("Invalid field count for query %s received", sqlQuery.Data()));
1848                 return;
1849         }
1850
1851         TSQLRow* aRow = aResult->Next();
1852         if (!aRow) {
1853                 AliError(Form("Could not receive result of query %s", sqlQuery.Data()));
1854                 return;
1855         }
1856
1857         TString result(aRow->GetField(0), aRow->GetFieldLength(0));
1858         Int_t count = result.Atoi();
1859
1860         Log("SHUTTLE", Form("%d unprocessed runs", count));
1861
1862         delete aRow;
1863         delete aResult;
1864
1865         TMonaLisaValue mlStatus("SHUTTLE_openruns", count);
1866
1867         TList mlList;
1868         mlList.Add(&mlStatus);
1869
1870         fMonaLisa->SendParameters(&mlList, "__PROCESSINGINFO__");
1871 }
1872
1873 //______________________________________________________________________________________________
1874 Bool_t AliShuttle::QueryShuttleLogbook(const char* whereClause,
1875                 TObjArray& entries)
1876 {
1877         // Query DAQ's Shuttle logbook and fills detector status object.
1878         // Call QueryRunParameters to query DAQ logbook for run parameters.
1879         //
1880
1881         entries.SetOwner(1);
1882
1883         // check connection, in case connect
1884         if (!Connect(3)) return kFALSE;
1885
1886         TString sqlQuery;
1887         sqlQuery = Form("select * from %s %s order by run", fConfig->GetShuttlelbTable(), whereClause);
1888
1889         TSQLResult* aResult = fServer[3]->Query(sqlQuery);
1890         if (!aResult) {
1891                 AliError(Form("Can't execute query <%s>!", sqlQuery.Data()));
1892                 return kFALSE;
1893         }
1894
1895         AliDebug(2,Form("Query = %s", sqlQuery.Data()));
1896
1897         if(aResult->GetRowCount() == 0) {
1898                 Log("SHUTTLE", "No entries in Shuttle Logbook match request");
1899                 delete aResult;
1900                 return kTRUE;
1901         }
1902
1903         // TODO Check field count!
1904         const UInt_t nCols = 25;
1905         if (aResult->GetFieldCount() != (Int_t) nCols) {
1906                 Log("SHUTTLE", "Invalid SQL result field number!");
1907                 delete aResult;
1908                 return kFALSE;
1909         }
1910
1911         TSQLRow* aRow;
1912         while ((aRow = aResult->Next())) {
1913                 TString runString(aRow->GetField(0), aRow->GetFieldLength(0));
1914                 Int_t run = runString.Atoi();
1915
1916                 AliShuttleLogbookEntry *entry = QueryRunParameters(run);
1917                 if (!entry)
1918                         continue;
1919
1920                 // DA test mode flag
1921                 TString daTestModeString(aRow->GetField(2), aRow->GetFieldLength(2)); // field 2 = DA test mode flag
1922                 Bool_t daTestMode = (Bool_t)daTestModeString.Atoi();
1923                 entry->SetDATestMode(daTestMode);
1924
1925                 // loop on detectors
1926                 for(UInt_t ii = 0; ii < nCols; ii++)
1927                         entry->SetDetectorStatus(aResult->GetFieldName(ii), aRow->GetField(ii));
1928
1929                 entries.AddLast(entry);
1930                 delete aRow;
1931         }
1932
1933         delete aResult;
1934         return kTRUE;
1935 }
1936
1937 //______________________________________________________________________________________________
1938 AliShuttleLogbookEntry* AliShuttle::QueryRunParameters(Int_t run)
1939 {
1940         //
1941         // Retrieve run parameters written in the DAQ logbook and sets them into AliShuttleLogbookEntry object
1942         //
1943
1944         // check connection, in case connect
1945         if (!Connect(3))
1946                 return 0;
1947
1948         TString sqlQuery;
1949         sqlQuery.Form("select * from %s where run=%d", fConfig->GetDAQlbTable(), run);
1950
1951         TSQLResult* aResult = fServer[3]->Query(sqlQuery);
1952         if (!aResult) {
1953                 Log("SHUTTLE", Form("Can't execute query <%s>!", sqlQuery.Data()));
1954                 return 0;
1955         }
1956
1957         if (aResult->GetRowCount() == 0) {
1958                 Log("SHUTTLE", Form("QueryRunParameters - No entry in DAQ Logbook for run %d. Skipping", run));
1959                 delete aResult;
1960                 return 0;
1961         }
1962
1963         if (aResult->GetRowCount() > 1) {
1964                 Log("SHUTTLE", Form("QueryRunParameters - UNEXPECTED: "
1965                                 "more than one entry in DAQ Logbook for run %d!", run));
1966                 delete aResult;
1967                 return 0;
1968         }
1969
1970         TSQLRow* aRow = aResult->Next();
1971         if (!aRow)
1972         {
1973                 Log("SHUTTLE", Form("QueryRunParameters - Could not retrieve row for run %d. Skipping", run));
1974                 delete aResult;
1975                 return 0;
1976         }
1977
1978         AliShuttleLogbookEntry* entry = new AliShuttleLogbookEntry(run);
1979
1980         for (Int_t ii = 0; ii < aResult->GetFieldCount(); ii++)
1981                 entry->SetRunParameter(aResult->GetFieldName(ii), aRow->GetField(ii));
1982
1983         delete aRow;
1984         delete aResult;
1985
1986         UInt_t startTime = entry->GetStartTime();
1987         UInt_t endTime = entry->GetEndTime();
1988         Bool_t ecsSuccess = entry->GetECSSuccess();
1989
1990         TString totEventsStr = entry->GetRunParameter("totalEvents");
1991         Int_t totEvents = totEventsStr.Atoi();
1992
1993         UInt_t now = time(0);
1994         // TODO make this a configuration parameter
1995         Int_t dcsDelay = fConfig->GetDCSDelay()+fConfig->GetDCSQueryOffset();
1996
1997         // runs are accepted if they have ecsSuccess set or more than 1 event
1998         if (startTime != 0 && endTime != 0 && endTime > startTime && (totEvents > 1 || ecsSuccess) && (endTime < now - dcsDelay))
1999         {
2000                 if (ecsSuccess == kFALSE)
2001                         Log("SHUTTLE", Form("Processing run %d although in status ECS failure, Reason: %s", run, entry->GetRunParameter("eor_reason")));
2002                 return entry;
2003         }
2004
2005         Bool_t skip = kFALSE;
2006
2007         if (endTime != 0 && endTime >= now - dcsDelay)
2008         {
2009                 Log("SHUTTLE", Form("Skipping run %d for now, because DCS buffer time is not yet expired", run));
2010         }
2011         else if (totEvents <= 1)
2012         {
2013                 Log("SHUTTLE", Form("QueryRunParameters - Run %d has 1 event or less - Skipping!", run));
2014                 skip = kTRUE;
2015         }
2016         else
2017         {
2018                 Log("SHUTTLE", Form("QueryRunParameters - Invalid parameters for Run %d: "
2019                         "startTime = %d, endTime = %d. Skipping (Shuttle won't be marked as DONE)!",
2020                         run, startTime, endTime));
2021         }
2022
2023         if (skip)
2024         {
2025                 Log("SHUTTLE", Form("Marking SHUTTLE skipped for run %d", run));
2026                 fLogbookEntry = entry;
2027                 if (!UpdateShuttleLogbook("shuttle_skipped"))
2028                 {
2029                         AliError(Form("Could not update logbook for run %d !", run));
2030                 }
2031                 if (!UpdateTableSkippedCase("ALL"))
2032                 {
2033                         AliError(Form("Could not update FXS tables for run %d !", run));
2034                 }
2035                 fLogbookEntry = 0;
2036         }
2037
2038         delete entry;
2039         return 0;
2040 }
2041
2042 //______________________________________________________________________________________________
2043 TMap* AliShuttle::GetValueSet(const char* host, Int_t port, const TSeqCollection* entries,
2044                               DCSType type, Int_t multiSplit)
2045 {
2046         // Retrieve all "entry" data points from the DCS server
2047         // host, port: TSocket connection parameters
2048         // entries: list of name of the alias or data point
2049         // type: kAlias or kDP
2050         // returns TMap of values, 0 when failure
2051
2052         AliDCSClient client(host, port, fTimeout, fRetries, multiSplit);
2053
2054         TMap* result = 0;
2055         if (type == kAlias)
2056         {
2057                 //result = client.GetAliasValues(entries, GetCurrentStartTime()-offset,
2058                 //      GetCurrentEndTime()+offset);
2059                 result = client.GetAliasValues(entries, GetStartTimeDCSQuery(),
2060                         GetEndTimeDCSQuery());
2061         }
2062         else if (type == kDP)
2063         {
2064                 //result = client.GetDPValues(entries, GetCurrentStartTime()-offset,
2065                 //      GetCurrentEndTime()+offset);
2066                 result = client.GetDPValues(entries, GetStartTimeDCSQuery(),
2067                         GetEndTimeDCSQuery());
2068         }
2069
2070         if (result == 0)
2071         {
2072                 Log(fCurrentDetector.Data(), Form("GetValueSet - Can't get entries! Reason: %s",
2073                         client.GetErrorString(client.GetResultErrorCode())));
2074                 if (client.GetResultErrorCode() == AliDCSClient::fgkServerError)
2075                         Log(fCurrentDetector.Data(), Form("GetValueSet - Server error code: %s",
2076                                 client.GetServerError().Data()));
2077
2078                 return 0;
2079         }
2080
2081         return result;
2082 }
2083
2084 //______________________________________________________________________________________________
2085 const char* AliShuttle::GetFile(Int_t system, const char* detector,
2086                 const char* id, const char* source)
2087 {
2088         // Get calibration file from file exchange servers
2089         // First queris the FXS database for the file name, using the run, detector, id and source info
2090         // then calls RetrieveFile(filename) for actual copy to local disk
2091         // run: current run being processed (given by Logbook entry fLogbookEntry)
2092         // detector: the Preprocessor name
2093         // id: provided as a parameter by the Preprocessor
2094         // source: provided by the Preprocessor through GetFileSources function
2095
2096         // check if test mode should simulate a FXS error
2097         if (fTestMode & kErrorFXSFiles)
2098         {
2099                 Log(detector, Form("GetFile - In TESTMODE - Simulating error while connecting to %s FXS", GetSystemName(system)));
2100                 return 0;
2101         }
2102
2103         // check connection, in case connect
2104         if (!Connect(system))
2105         {
2106                 Log(detector, Form("GetFile - Couldn't connect to %s FXS database", GetSystemName(system)));
2107                 fFXSError = system;
2108                 return 0;
2109         }
2110
2111         // Query preparation
2112         TString sourceName(source);
2113         Int_t nFields = 3;
2114         TString sqlQueryStart = Form("select filePath,size,fileChecksum from %s where",
2115                                                                 fConfig->GetFXSdbTable(system));
2116         TString whereClause = Form("run=%d and detector=\"%s\" and fileId=\"%s\"",
2117                                                                 GetCurrentRun(), detector, id);
2118
2119         if (system == kDAQ)
2120         {
2121                 whereClause += Form(" and DAQsource=\"%s\"", source);
2122         }
2123         else if (system == kDCS)
2124         {
2125                 sourceName="none";
2126         }
2127         else if (system == kHLT)
2128         {
2129                 whereClause += Form(" and DDLnumbers=\"%s\"", source);
2130         }
2131
2132         TString sqlQuery = Form("%s %s", sqlQueryStart.Data(), whereClause.Data());
2133
2134         AliDebug(2, Form("SQL query: \n%s",sqlQuery.Data()));
2135
2136         // Query execution
2137         TSQLResult* aResult = 0;
2138         aResult = dynamic_cast<TSQLResult*> (fServer[system]->Query(sqlQuery));
2139         if (!aResult) {
2140                 Log(detector, Form("GetFile - Can't execute SQL query to %s database for: id = %s, source = %s",
2141                                 GetSystemName(system), id, sourceName.Data()));
2142                 fFXSError = system;
2143                 return 0;
2144         }
2145
2146         if (aResult->GetRowCount() == 0)
2147         {
2148                 Log(detector,
2149                         Form("GetFile - No entry in %s FXS db for: id = %s, source = %s",
2150                                 GetSystemName(system), id, sourceName.Data()));
2151                 delete aResult;
2152                 return 0;
2153         }
2154
2155         if (aResult->GetRowCount() > 1) {
2156                 Log(detector,
2157                         Form("GetFile - More than one entry in %s FXS db for: id = %s, source = %s",
2158                                 GetSystemName(system), id, sourceName.Data()));
2159                 fFXSError = system;
2160                 delete aResult;
2161                 return 0;
2162         }
2163
2164         if (aResult->GetFieldCount() != nFields) {
2165                 Log(detector,
2166                         Form("GetFileName - Wrong field count in %s FXS db for: id = %s, source = %s",
2167                                 GetSystemName(system), id, sourceName.Data()));
2168                 fFXSError = system;
2169                 delete aResult;
2170                 return 0;
2171         }
2172
2173         TSQLRow* aRow = dynamic_cast<TSQLRow*> (aResult->Next());
2174
2175         if (!aRow){
2176                 Log(detector, Form("GetFile - Empty set result in %s FXS db from query: id = %s, source = %s",
2177                                 GetSystemName(system), id, sourceName.Data()));
2178                 fFXSError = system;
2179                 delete aResult;
2180                 return 0;
2181         }
2182
2183         TString filePath(aRow->GetField(0), aRow->GetFieldLength(0));
2184         TString fileSize(aRow->GetField(1), aRow->GetFieldLength(1));
2185         TString fileChecksum(aRow->GetField(2), aRow->GetFieldLength(2));
2186
2187         delete aResult;
2188         delete aRow;
2189
2190         AliDebug(2, Form("filePath = %s; size = %s, fileChecksum = %s",
2191                                 filePath.Data(), fileSize.Data(), fileChecksum.Data()));
2192
2193         // retrieved file is renamed to make it unique
2194         Int_t dir_lev1 = GetCurrentRun()/10000;
2195         TString localFileName = Form("%s/%d/%d/%s_process/%s_%s_%d_%s_%s.shuttle",
2196                                      GetShuttleTempDir(), dir_lev1, GetCurrentRun(), detector,
2197                                         GetSystemName(system), detector, GetCurrentRun(),
2198                                         id, sourceName.Data());
2199         Log("SHUTTLE",Form("file from FXS = %s",localFileName.Data()));
2200
2201
2202         // file retrieval from FXS
2203         UInt_t nRetries = 0;
2204         UInt_t maxRetries = 3;
2205         Bool_t result = kFALSE;
2206
2207         // copy!! if successful TSystem::Exec returns 0
2208         while (nRetries++ < maxRetries) {
2209                 AliDebug(2, Form("Trying to copy file. Retry # %d", nRetries));
2210                 result = RetrieveFile(system, filePath.Data(), localFileName.Data());
2211                 if (!result)
2212                 {
2213                         Log(detector, Form("GetFile - Copy of file %s from %s FXS failed",
2214                                         filePath.Data(), GetSystemName(system)));
2215                         continue;
2216                 }
2217
2218                 if (fileSize.Length()>0)
2219                 {
2220                         // compare filesize of local file with the one stored in the FXS DB
2221                         Long_t size = -1;
2222                         Int_t sizeComp = gSystem->GetPathInfo(localFileName.Data(), 0, &size, 0, 0);
2223
2224                         if (sizeComp != 0 || size != fileSize.Atoi())
2225                         {
2226                                 Log(detector, Form("GetFile - size of file %s does not match with local copy!",
2227                                                         filePath.Data()));
2228                                 result = kFALSE;
2229                                 continue;
2230                         }
2231
2232                 } else {
2233                         Log(fCurrentDetector, Form("GetFile - size of file %s not set in %s database, skipping comparison",
2234                                                 filePath.Data(), GetSystemName(system)));
2235                 }
2236
2237                 if (fileChecksum.Length()>0)
2238                 {
2239                         // compare md5sum of local file with the one stored in the FXS DB
2240                         if(fileChecksum.Contains(' ')) fileChecksum.Resize(fileChecksum.First(' '));
2241                         Int_t md5Comp = gSystem->Exec(Form("md5sum %s |grep %s > /dev/null 2> /dev/null",
2242                                                 localFileName.Data(), fileChecksum.Data()));
2243
2244                         if (md5Comp != 0)
2245                         {
2246                                 Log(detector, Form("GetFile - md5sum of file %s does not match with local copy!",
2247                                                         filePath.Data()));
2248                                 result = kFALSE;
2249                                 continue;
2250                         }
2251                 } else {
2252                         Log(fCurrentDetector, Form("GetFile - md5sum of file %s not set in %s database, skipping comparison",
2253                                                         filePath.Data(), GetSystemName(system)));
2254                 }
2255                 if (result) break;
2256         }
2257
2258         if (!result)
2259         {
2260                 fFXSError = system;
2261                 return 0;
2262         }
2263
2264         fFXSCalled[system]=kTRUE;
2265         TObjString *fileParams = new TObjString(Form("%s#!?!#%s", id, sourceName.Data()));
2266         fFXSlist[system].Add(fileParams);
2267
2268         static TString staticLocalFileName;
2269         staticLocalFileName.Form("%s", localFileName.Data());
2270
2271         Log(fCurrentDetector, Form("GetFile - Retrieved file with id %s and "
2272                         "source %s from %s to %s", id, source,
2273                         GetSystemName(system), localFileName.Data()));
2274
2275         return staticLocalFileName.Data();
2276 }
2277
2278 //______________________________________________________________________________________________
2279 Bool_t AliShuttle::RetrieveFile(UInt_t system, const char* fxsFileName, const char* localFileName)
2280 {
2281         //
2282         // Copies file from FXS to local Shuttle machine
2283         //
2284
2285         // check temp directory: trying to cd to temp; if it does not exist, create it
2286         AliDebug(2, Form("Copy file %s from %s FXS into %s",
2287                         GetSystemName(system), fxsFileName, localFileName));
2288
2289         TString tmpDir(localFileName);
2290
2291         tmpDir = tmpDir(0,tmpDir.Last('/'));
2292
2293         Int_t noDir = gSystem->GetPathInfo(tmpDir.Data(), 0, (Long64_t*) 0, 0, 0);
2294         if (noDir) // temp dir does not exists!
2295         {
2296                 if (gSystem->mkdir(tmpDir.Data(), 1))
2297                 {
2298                         Log(fCurrentDetector.Data(), "RetrieveFile - could not make temp directory!!");
2299                         return kFALSE;
2300                 }
2301         }
2302
2303         TString command = Form("scp -oPort=%d -2 %s@%s:%s/%s %s",
2304                 fConfig->GetFXSPort(system),
2305                 fConfig->GetFXSUser(system),
2306                 fConfig->GetFXSHost(system),
2307                 fConfig->GetFXSBaseFolder(system),
2308                 fxsFileName,
2309                 localFileName);
2310
2311         AliDebug(2, Form("%s",command.Data()));
2312
2313         Bool_t result = (gSystem->Exec(command.Data()) == 0);
2314
2315         return result;
2316 }
2317
2318 //______________________________________________________________________________________________
2319 TList* AliShuttle::GetFileSources(Int_t system, const char* detector, const char* id)
2320 {
2321         //
2322         // Get sources producing the condition file Id from file exchange servers
2323         // if id is NULL all sources are returned (distinct)
2324         //
2325
2326         if (id)
2327         {
2328                 Log(detector, Form("GetFileSources - Querying %s FXS for files with id %s produced by %s", GetSystemName(system), id, detector));
2329         } else {
2330                 Log(detector, Form("GetFileSources - Querying %s FXS for files produced by %s", GetSystemName(system), detector));
2331         }
2332
2333         // check if test mode should simulate a FXS error
2334         if (fTestMode & kErrorFXSSources)
2335         {
2336                 Log(detector, Form("GetFileSources - In TESTMODE - Simulating error while connecting to %s FXS", GetSystemName(system)));
2337                 return 0;
2338         }
2339
2340         if (system == kDCS)
2341         {
2342                 Log(detector, "GetFileSources - WARNING: DCS system has only one source of data!");
2343                 TList *list = new TList();
2344                 list->SetOwner(1);
2345                 list->Add(new TObjString(" "));
2346                 return list;
2347         }
2348
2349         // check connection, in case connect
2350         if (!Connect(system))
2351         {
2352                 Log(detector, Form("GetFileSources - Couldn't connect to %s FXS database", GetSystemName(system)));
2353                 fFXSError = system;
2354                 return NULL;
2355         }
2356
2357         TString sourceName = 0;
2358         if (system == kDAQ)
2359         {
2360                 sourceName = "DAQsource";
2361         } else if (system == kHLT)
2362         {
2363                 sourceName = "DDLnumbers";
2364         }
2365
2366         TString sqlQueryStart = Form("select distinct %s from %s where", sourceName.Data(), fConfig->GetFXSdbTable(system));
2367         TString whereClause = Form("run=%d and detector=\"%s\"",
2368                                 GetCurrentRun(), detector);
2369         if (id)
2370                 whereClause += Form(" and fileId=\"%s\"", id);
2371         TString sqlQuery = Form("%s %s", sqlQueryStart.Data(), whereClause.Data());
2372
2373         AliDebug(2, Form("SQL query: \n%s",sqlQuery.Data()));
2374
2375         // Query execution
2376         TSQLResult* aResult;
2377         aResult = fServer[system]->Query(sqlQuery);
2378         if (!aResult) {
2379                 Log(detector, Form("GetFileSources - Can't execute SQL query to %s database for id: %s",
2380                                 GetSystemName(system), id));
2381                 fFXSError = system;
2382                 return 0;
2383         }
2384
2385         TList *list = new TList();
2386         list->SetOwner(1);
2387
2388         if (aResult->GetRowCount() == 0)
2389         {
2390                 Log(detector,
2391                         Form("GetFileSources - No entry in %s FXS table for id: %s", GetSystemName(system), id));
2392                 delete aResult;
2393                 return list;
2394         }
2395
2396         Log(detector, Form("GetFileSources - Found %d sources", aResult->GetRowCount()));
2397
2398         TSQLRow* aRow;
2399         while ((aRow = aResult->Next()))
2400         {
2401
2402                 TString source(aRow->GetField(0), aRow->GetFieldLength(0));
2403                 AliDebug(2, Form("%s = %s", sourceName.Data(), source.Data()));
2404                 list->Add(new TObjString(source));
2405                 delete aRow;
2406         }
2407
2408         delete aResult;
2409
2410         return list;
2411 }
2412
2413 //______________________________________________________________________________________________
2414 TList* AliShuttle::GetFileIDs(Int_t system, const char* detector, const char* source)
2415 {
2416         //
2417         // Get all ids of condition files produced by a given source from file exchange servers
2418         //
2419
2420         Log(detector, Form("GetFileIDs - Retrieving ids with source %s with %s", source, GetSystemName(system)));
2421
2422         // check if test mode should simulate a FXS error
2423         if (fTestMode & kErrorFXSSources)
2424         {
2425                 Log(detector, Form("GetFileIDs - In TESTMODE - Simulating error while connecting to %s FXS", GetSystemName(system)));
2426                 return 0;
2427         }
2428
2429         // check connection, in case connect
2430         if (!Connect(system))
2431         {
2432                 Log(detector, Form("GetFileIDs - Couldn't connect to %s FXS database", GetSystemName(system)));
2433                 return NULL;
2434         }
2435
2436         TString sourceName = 0;
2437         if (system == kDAQ)
2438         {
2439                 sourceName = "DAQsource";
2440         } else if (system == kHLT)
2441         {
2442                 sourceName = "DDLnumbers";
2443         }
2444
2445         TString sqlQueryStart = Form("select fileId from %s where", fConfig->GetFXSdbTable(system));
2446         TString whereClause = Form("run=%d and detector=\"%s\"",
2447                                 GetCurrentRun(), detector);
2448         if (sourceName.Length() > 0 && source)
2449                 whereClause += Form(" and %s=\"%s\"", sourceName.Data(), source);
2450         TString sqlQuery = Form("%s %s", sqlQueryStart.Data(), whereClause.Data());
2451
2452         AliDebug(2, Form("SQL query: \n%s",sqlQuery.Data()));
2453
2454         // Query execution
2455         TSQLResult* aResult;
2456         aResult = fServer[system]->Query(sqlQuery);
2457         if (!aResult) {
2458                 Log(detector, Form("GetFileIDs - Can't execute SQL query to %s database for source: %s",
2459                                 GetSystemName(system), source));
2460                 return 0;
2461         }
2462
2463         TList *list = new TList();
2464         list->SetOwner(1);
2465
2466         if (aResult->GetRowCount() == 0)
2467         {
2468                 Log(detector,
2469                         Form("GetFileIDs - No entry in %s FXS table for source: %s", GetSystemName(system), source));
2470                 delete aResult;
2471                 return list;
2472         }
2473
2474         Log(detector, Form("GetFileIDs - Found %d ids", aResult->GetRowCount()));
2475
2476         TSQLRow* aRow;
2477
2478         while ((aRow = aResult->Next()))
2479         {
2480
2481                 TString id(aRow->GetField(0), aRow->GetFieldLength(0));
2482                 AliDebug(2, Form("fileId = %s", id.Data()));
2483                 list->Add(new TObjString(id));
2484                 delete aRow;
2485         }
2486
2487         delete aResult;
2488
2489         return list;
2490 }
2491
2492 //______________________________________________________________________________________________
2493 Bool_t AliShuttle::Connect(Int_t system)
2494 {
2495         // Connect to MySQL Server of the system's FXS MySQL databases
2496         // DAQ Logbook, Shuttle Logbook and DAQ FXS db are on the same host
2497         //
2498
2499         // check connection: if already connected return
2500         if(fServer[system] && fServer[system]->IsConnected()) {
2501                 // ping the server --> automatic reconnection should occur if it was broken but the
2502                 // server is still alive
2503                 fServer[system]->Ping();
2504                 return kTRUE;
2505         }
2506
2507         TString dbHost, dbUser, dbPass, dbName;
2508
2509         if (system < 3) // FXS db servers
2510         {
2511                 dbHost = Form("mysql://%s:%d", fConfig->GetFXSdbHost(system), fConfig->GetFXSdbPort(system));
2512                 dbUser = fConfig->GetFXSdbUser(system);
2513                 dbPass = fConfig->GetFXSdbPass(system);
2514                 dbName =   fConfig->GetFXSdbName(system);
2515         } else { // Run & Shuttle logbook servers
2516         // TODO Will the Shuttle logbook server be the same as the Run logbook server ???
2517                 dbHost = Form("mysql://%s:%d", fConfig->GetDAQlbHost(), fConfig->GetDAQlbPort());
2518                 dbUser = fConfig->GetDAQlbUser();
2519                 dbPass = fConfig->GetDAQlbPass();
2520                 dbName =   fConfig->GetDAQlbDB();
2521         }
2522
2523         fServer[system] = TSQLServer::Connect(dbHost.Data(), dbUser.Data(), dbPass.Data());
2524         if (!fServer[system] || !fServer[system]->IsConnected()) {
2525                 if(system < 3)
2526                 {
2527                 AliError(Form("Can't establish connection to FXS database for %s",
2528                                         AliShuttleInterface::GetSystemName(system)));
2529                 } else {
2530                 AliError("Can't establish connection to Run logbook.");
2531                 }
2532                 if(fServer[system]) delete fServer[system];
2533                 return kFALSE;
2534         }
2535
2536         // Get tables
2537         TSQLResult* aResult=0;
2538         switch(system){
2539                 case kDAQ:
2540                         aResult = fServer[kDAQ]->GetTables(dbName.Data());
2541                         break;
2542                 case kDCS:
2543                         aResult = fServer[kDCS]->GetTables(dbName.Data());
2544                         break;
2545                 case kHLT:
2546                         aResult = fServer[kHLT]->GetTables(dbName.Data());
2547                         break;
2548                 default:
2549                         aResult = fServer[3]->GetTables(dbName.Data());
2550                         break;
2551         }
2552
2553         delete aResult;
2554         return kTRUE;
2555 }
2556
2557 //______________________________________________________________________________________________
2558 Bool_t AliShuttle::UpdateTable()
2559 {
2560         //
2561         // Update FXS table filling time_processed field in all rows corresponding to current run and detector
2562         //
2563
2564         Bool_t result = kTRUE;
2565
2566         for (UInt_t system=0; system<3; system++)
2567         {
2568                 if(!fFXSCalled[system]) continue;
2569
2570                 // check connection, in case connect
2571                 if (!Connect(system))
2572                 {
2573                         Log(fCurrentDetector, Form("UpdateTable - Couldn't connect to %s FXS database", GetSystemName(system)));
2574                         result = kFALSE;
2575                         continue;
2576                 }
2577
2578                 TTimeStamp now; // now
2579
2580                 // Loop on FXS list entries
2581                 TIter iter(&fFXSlist[system]);
2582                 TObjString *aFXSentry=0;
2583                 while ((aFXSentry = dynamic_cast<TObjString*> (iter.Next())))
2584                 {
2585                         TString aFXSentrystr = aFXSentry->String();
2586                         TObjArray *aFXSarray = aFXSentrystr.Tokenize("#!?!#");
2587                         if (!aFXSarray || aFXSarray->GetEntries() != 2 )
2588                         {
2589                                 Log(fCurrentDetector, Form("UpdateTable - error updating %s FXS entry. Check string: <%s>",
2590                                         GetSystemName(system), aFXSentrystr.Data()));
2591                                 if(aFXSarray) delete aFXSarray;
2592                                 result = kFALSE;
2593                                 continue;
2594                         }
2595                         const char* fileId = ((TObjString*) aFXSarray->At(0))->GetName();
2596                         const char* source = ((TObjString*) aFXSarray->At(1))->GetName();
2597
2598                         TString whereClause;
2599                         if (system == kDAQ)
2600                         {
2601                                 whereClause = Form("where run=%d and detector=\"%s\" and fileId=\"%s\" and DAQsource=\"%s\";",
2602                                                         GetCurrentRun(), fCurrentDetector.Data(), fileId, source);
2603                         }
2604                         else if (system == kDCS)
2605                         {
2606                                 whereClause = Form("where run=%d and detector=\"%s\" and fileId=\"%s\";",
2607                                                         GetCurrentRun(), fCurrentDetector.Data(), fileId);
2608                         }
2609                         else if (system == kHLT)
2610                         {
2611                                 whereClause = Form("where run=%d and detector=\"%s\" and fileId=\"%s\" and DDLnumbers=\"%s\";",
2612                                                         GetCurrentRun(), fCurrentDetector.Data(), fileId, source);
2613                         }
2614
2615                         delete aFXSarray;
2616
2617                         TString sqlQuery = Form("update %s set time_processed=%d %s", fConfig->GetFXSdbTable(system),
2618                                                                 now.GetSec(), whereClause.Data());
2619
2620                         AliDebug(2, Form("SQL query: \n%s",sqlQuery.Data()));
2621
2622                         // Query execution
2623                         TSQLResult* aResult;
2624                         aResult = dynamic_cast<TSQLResult*> (fServer[system]->Query(sqlQuery));
2625                         if (!aResult)
2626                         {
2627                                 Log(fCurrentDetector, Form("UpdateTable - %s db: can't execute SQL query <%s>",
2628                                                                 GetSystemName(system), sqlQuery.Data()));
2629                                 result = kFALSE;
2630                                 continue;
2631                         }
2632                         delete aResult;
2633                 }
2634         }
2635
2636         return result;
2637 }
2638
2639 //_______________________________________________________________________________
2640 Bool_t AliShuttle::UpdateTableSkippedCase(const char* detector)
2641 {
2642         //
2643         // Update FXS table filling time_processed field in all rows corresponding to current run and detector
2644         // if detector = "ALL" update all detectors
2645         //
2646
2647         Bool_t result = kTRUE;
2648
2649         TString detName(detector);
2650
2651         for (UInt_t system=0; system<3; system++)
2652         {
2653
2654                 // check connection, in case connect
2655                 if (!Connect(system))
2656                 {
2657                         Log(fCurrentDetector, Form("UpdateTableSkippedCase - Couldn't connect to %s FXS database", GetSystemName(system)));
2658                         result = kFALSE;
2659                         continue;
2660                 }
2661
2662                 TTimeStamp now; // now
2663
2664                 // Loop on FXS list entries
2665                 TIter iter(&fFXSlist[system]);
2666
2667                 TString whereClause;
2668                 if (detName == "ALL") whereClause = Form("where run=%d and time_processed IS NULL;",GetCurrentRun());
2669                 else whereClause = Form("where run=%d and detector=\"%s\" and time_processed IS NULL;",GetCurrentRun(), detector);
2670
2671                 Log("SHUTTLE",Form(" whereClause = %s ",whereClause.Data()));
2672
2673                 TString sqlQuery = Form("update %s set time_processed=%d %s", fConfig->GetFXSdbTable(system),
2674                                         now.GetSec(), whereClause.Data());
2675
2676                 AliDebug(2, Form("SQL query: \n%s",sqlQuery.Data()));
2677
2678                 // Query execution
2679                 TSQLResult* aResult;
2680                 aResult = dynamic_cast<TSQLResult*> (fServer[system]->Query(sqlQuery));
2681                 if (!aResult)
2682                 {
2683                         Log("SHUTTLE", Form("UpdateTableSkippedCase - %s db: can't execute SQL query <%s>",
2684                                                         GetSystemName(system), sqlQuery.Data()));
2685                         result = kFALSE;
2686                         continue;
2687                 }
2688                 delete aResult;
2689
2690         }
2691
2692         return result;
2693 }
2694 //______________________________________________________________________________________________
2695 Bool_t AliShuttle::UpdateTableFailCase()
2696 {
2697         // Update FXS table filling time_processed field in all rows corresponding to current run and detector
2698         // this is called in case the preprocessor is declared failed for the current run, because
2699         // the fields are updated only in case of success
2700
2701         Bool_t result = kTRUE;
2702
2703         for (UInt_t system=0; system<3; system++)
2704         {
2705                 // check connection, in case connect
2706                 if (!Connect(system))
2707                 {
2708                         Log(fCurrentDetector, Form("UpdateTableFailCase - Couldn't connect to %s FXS database",
2709                                                         GetSystemName(system)));
2710                         result = kFALSE;
2711                         continue;
2712                 }
2713
2714                 TTimeStamp now; // now
2715
2716                 // Loop on FXS list entries
2717
2718                 TString whereClause = Form("where run=%d and detector=\"%s\";",
2719                                                 GetCurrentRun(), fCurrentDetector.Data());
2720
2721
2722                 TString sqlQuery = Form("update %s set time_processed=%d %s", fConfig->GetFXSdbTable(system),
2723                                                         now.GetSec(), whereClause.Data());
2724
2725                 AliDebug(2, Form("SQL query: \n%s",sqlQuery.Data()));
2726
2727                 // Query execution
2728                 TSQLResult* aResult;
2729                 aResult = dynamic_cast<TSQLResult*> (fServer[system]->Query(sqlQuery));
2730                 if (!aResult)
2731                 {
2732                         Log(fCurrentDetector, Form("UpdateTableFailCase - %s db: can't execute SQL query <%s>",
2733                                                         GetSystemName(system), sqlQuery.Data()));
2734                         result = kFALSE;
2735                         continue;
2736                 }
2737                 delete aResult;
2738         }
2739
2740         return result;
2741 }
2742
2743 //______________________________________________________________________________________________
2744 Bool_t AliShuttle::UpdateShuttleLogbook(const char* detector, const char* status)
2745 {
2746         //
2747         // Update Shuttle logbook filling detector or shuttle_done column
2748         // ex. of usage: UpdateShuttleLogbook("PHOS", "DONE") or UpdateShuttleLogbook("shuttle_done")
2749         //
2750
2751         // check connection, in case connect
2752         if(!Connect(3)){
2753                 Log("SHUTTLE", "UpdateShuttleLogbook - Couldn't connect to DAQ Logbook.");
2754                 return kFALSE;
2755         }
2756
2757         TString detName(detector);
2758         TString setClause;
2759         if (detName == "shuttle_done" || detName == "shuttle_skipped")
2760         {
2761                 setClause = "set shuttle_done=1";
2762
2763                 if (detName == "shuttle_done")
2764                 {
2765                         if (TouchFile() != kTRUE)
2766                         {
2767                                 SendMLRunInfo("Pending");
2768                                 return kFALSE;
2769                         }
2770
2771                         SendMLRunInfo("Done");
2772                 }
2773                 else
2774                         SendMLRunInfo("Skipped");
2775         }
2776         else {
2777                 TString statusStr(status);
2778                 if(statusStr.Contains("done", TString::kIgnoreCase) ||
2779                    statusStr.Contains("failed", TString::kIgnoreCase)){
2780                         setClause = Form("set %s=\"%s\"", detector, status);
2781                 } else {
2782                         Log("SHUTTLE",
2783                                 Form("UpdateShuttleLogbook - Invalid status <%s> for detector %s",
2784                                         status, detector));
2785                         return kFALSE;
2786                 }
2787         }
2788
2789         TString whereClause = Form("where run=%d", GetCurrentRun());
2790
2791         TString sqlQuery = Form("update %s %s %s",
2792                                         fConfig->GetShuttlelbTable(), setClause.Data(), whereClause.Data());
2793
2794         AliDebug(2, Form("SQL query: \n%s",sqlQuery.Data()));
2795
2796         // Query execution
2797         TSQLResult* aResult;
2798         aResult = dynamic_cast<TSQLResult*> (fServer[3]->Query(sqlQuery));
2799         if (!aResult) {
2800                 Log("SHUTTLE", Form("UpdateShuttleLogbook - Can't execute query <%s>", sqlQuery.Data()));
2801                 return kFALSE;
2802         }
2803         delete aResult;
2804
2805         return kTRUE;
2806 }
2807
2808 //______________________________________________________________________________________________
2809 Int_t AliShuttle::GetCurrentRun() const
2810 {
2811         //
2812         // Get current run from logbook entry
2813         //
2814
2815         return fLogbookEntry ? fLogbookEntry->GetRun() : -1;
2816 }
2817
2818 //______________________________________________________________________________________________
2819 UInt_t AliShuttle::GetCurrentStartTime() const
2820 {
2821         //
2822         // get current start time
2823         //
2824
2825         return fLogbookEntry ? fLogbookEntry->GetStartTime() : 0;
2826 }
2827
2828 //______________________________________________________________________________________________
2829 UInt_t AliShuttle::GetCurrentEndTime() const
2830 {
2831         //
2832         // get current end time from logbook entry
2833         //
2834
2835         return fLogbookEntry ? fLogbookEntry->GetEndTime() : 0;
2836 }
2837 //______________________________________________________________________________________________
2838 UInt_t AliShuttle::GetCurrentYear() const
2839 {
2840         //
2841         // Get current year from logbook entry
2842         //
2843
2844         if (!fLogbookEntry) return 0;
2845
2846         TTimeStamp startTime(GetCurrentStartTime());
2847         TString year =  Form("%d",startTime.GetDate());
2848         year = year(0,4);
2849
2850         return year.Atoi();
2851 }
2852
2853 //______________________________________________________________________________________________
2854 const char* AliShuttle::GetLHCPeriod() const
2855 {
2856         //
2857         // Get current LHC period from logbook entry
2858         //
2859
2860         if (!fLogbookEntry) return 0;
2861
2862         return fLogbookEntry->GetRunParameter("LHCperiod");
2863 }
2864
2865 //______________________________________________________________________________________________
2866 void AliShuttle::Log(const char* detector, const char* message)
2867 {
2868         //
2869         // Fill log string with a message
2870         //
2871
2872         TString logRunDir = GetShuttleLogDir();
2873         if (GetCurrentRun() >=0) {
2874                 Int_t logDir_lev1 = GetCurrentRun()/10000;
2875                 logRunDir += Form("/%d/%d", logDir_lev1, GetCurrentRun());
2876         }
2877         void* dir = gSystem->OpenDirectory(logRunDir.Data());
2878         if (dir == NULL) {
2879                 if (gSystem->mkdir(logRunDir.Data(), kTRUE)) {
2880                         AliError(Form("Can't open directory <%s>", GetShuttleLogDir()));
2881                         return;
2882                 }
2883
2884         } else {
2885                 gSystem->FreeDirectory(dir);
2886         }
2887
2888         TString toLog = Form("%s UTC (%d): %s - ", TTimeStamp(time(0)).AsString("s"), getpid(), detector);
2889         if (GetCurrentRun() >= 0)
2890                 toLog += Form("run %d - ", GetCurrentRun());
2891         toLog += Form("%s", message);
2892
2893         AliInfo(toLog.Data());
2894
2895         // if we redirect the log output already to the file, leave here
2896         if (fOutputRedirected && strcmp(detector, "SHUTTLE") != 0)
2897                 return;
2898
2899         TString fileName = GetLogFileName(detector);
2900
2901         gSystem->ExpandPathName(fileName);
2902
2903         ofstream logFile;
2904         logFile.open(fileName, ofstream::out | ofstream::app);
2905
2906         if (!logFile.is_open()) {
2907                 AliError(Form("Could not open file %s", fileName.Data()));
2908                 return;
2909         }
2910
2911         logFile << toLog.Data() << "\n";
2912
2913         logFile.close();
2914 }
2915
2916 //______________________________________________________________________________________________
2917 TString AliShuttle::GetLogFileName(const char* detector) const
2918 {
2919         //
2920         // returns the name of the log file for a given sub detector
2921         //
2922
2923         TString fileName;
2924
2925         if (GetCurrentRun() >= 0)
2926         {
2927                 Int_t logDir_lev1 = GetCurrentRun()/10000;
2928                 fileName.Form("%s/%d/%d/%s.log", GetShuttleLogDir(), logDir_lev1, GetCurrentRun(),
2929                         detector);
2930         } else {
2931                 fileName.Form("%s/%s.log", GetShuttleLogDir(), detector);
2932         }
2933
2934         return fileName;
2935 }
2936
2937 //______________________________________________________________________________________________
2938 void AliShuttle::SendAlive()
2939 {
2940         // sends alive message to ML
2941
2942         TMonaLisaText mlStatus("SHUTTLE_status", "Alive");
2943
2944         TList mlList;
2945         mlList.Add(&mlStatus);
2946
2947         fMonaLisa->SendParameters(&mlList, "__PROCESSINGINFO__");
2948 }
2949
2950 //______________________________________________________________________________________________
2951 Bool_t AliShuttle::Collect(Int_t run)
2952 {
2953         //
2954         // Collects conditions data for all UNPROCESSED run written to DAQ LogBook in case of run = -1 (default)
2955         // If a dedicated run is given this run is processed
2956         //
2957         // In operational mode, this is the Shuttle function triggered by the EOR signal.
2958         //
2959
2960         if (run == -1)
2961                 Log("SHUTTLE","Collect - Shuttle called. Collecting conditions data for unprocessed runs");
2962         else
2963                 Log("SHUTTLE", Form("Collect - Shuttle called. Collecting conditions data for run %d", run));
2964
2965         SetLastAction("Starting");
2966
2967         // create ML instance
2968         if (!fMonaLisa)
2969                 fMonaLisa = new TMonaLisaWriter(fConfig->GetMonitorHost(), fConfig->GetMonitorTable());
2970
2971         CountOpenRuns();
2972
2973         TString whereClause("where shuttle_done=0");
2974         if (run != -1)
2975                 whereClause += Form(" and run=%d", run);
2976
2977         TObjArray shuttleLogbookEntries;
2978         if (!QueryShuttleLogbook(whereClause, shuttleLogbookEntries))
2979         {
2980                 Log("SHUTTLE", "Collect - Can't retrieve entries from Shuttle logbook");
2981                 return kFALSE;
2982         }
2983
2984         if (shuttleLogbookEntries.GetEntries() == 0)
2985         {
2986                 if (run == -1)
2987                         Log("SHUTTLE","Collect - Found no UNPROCESSED runs in Shuttle logbook");
2988                 else
2989                         Log("SHUTTLE", Form("Collect - Run %d is already DONE "
2990                                                 "or it does not exist in Shuttle logbook", run));
2991                 return kTRUE;
2992         }
2993
2994         for (UInt_t iDet=0; iDet<NDetectors(); iDet++)
2995                 fFirstUnprocessed[iDet] = kTRUE;
2996
2997         if (run != -1)
2998         {
2999                 // query Shuttle logbook for earlier runs, check if some detectors are unprocessed,
3000                 // flag them into fFirstUnprocessed array
3001                 TString whereClauseBis(Form("where shuttle_done=0 and run < %d", run));
3002                 TObjArray tmpLogbookEntries;
3003                 if (!QueryShuttleLogbook(whereClauseBis, tmpLogbookEntries))
3004                 {
3005                         Log("SHUTTLE", "Collect - Can't retrieve entries from Shuttle logbook");
3006                         return kFALSE;
3007                 }
3008
3009                 TIter iter(&tmpLogbookEntries);
3010                 AliShuttleLogbookEntry* anEntry = 0;
3011                 while ((anEntry = dynamic_cast<AliShuttleLogbookEntry*> (iter.Next())))
3012                 {
3013                         for (UInt_t iDet=0; iDet<NDetectors(); iDet++)
3014                         {
3015                                 if (anEntry->GetDetectorStatus(iDet) == AliShuttleLogbookEntry::kUnprocessed)
3016                                 {
3017                                         AliDebug(2, Form("Run %d: setting %s as \"not first time unprocessed\"",
3018                                                         anEntry->GetRun(), GetDetName(iDet)));
3019                                         fFirstUnprocessed[iDet] = kFALSE;
3020                                 }
3021                         }
3022
3023                 }
3024
3025         }
3026
3027         if (!RetrieveConditionsData(shuttleLogbookEntries))
3028         {
3029                 Log("SHUTTLE", "Collect - Process of at least one run failed");
3030                 CountOpenRuns();
3031                 return kFALSE;
3032         }
3033
3034         Log("SHUTTLE", "Collect - Requested run(s) successfully processed");
3035         CountOpenRuns();
3036         return kTRUE;
3037 }
3038
3039 //______________________________________________________________________________________________
3040 Bool_t AliShuttle::RetrieveConditionsData(const TObjArray& dateEntries)
3041 {
3042         //
3043         // Retrieve conditions data for all runs that aren't processed yet
3044         //
3045
3046         Bool_t hasError = kFALSE;
3047
3048         TIter iter(&dateEntries);
3049         AliShuttleLogbookEntry* anEntry;
3050
3051         while ((anEntry = (AliShuttleLogbookEntry*) iter.Next())){
3052                 if (!Process(anEntry)){
3053                         hasError = kTRUE;
3054                 }
3055
3056                 // clean SHUTTLE temp directory
3057                 //TString filename = Form("%s/*.shuttle", GetShuttleTempDir());
3058                 //RemoveFile(filename.Data());
3059         }
3060
3061         return hasError == kFALSE;
3062 }
3063
3064 //______________________________________________________________________________________________
3065 ULong_t AliShuttle::GetTimeOfLastAction() const
3066 {
3067         //
3068         // Gets time of last action
3069         //
3070
3071         ULong_t tmp;
3072
3073         fMonitoringMutex->Lock();
3074
3075         tmp = fLastActionTime;
3076
3077         fMonitoringMutex->UnLock();
3078
3079         return tmp;
3080 }
3081
3082 //______________________________________________________________________________________________
3083 const TString AliShuttle::GetLastAction() const
3084 {
3085         //
3086         // returns a string description of the last action
3087         //
3088
3089         TString tmp;
3090
3091         fMonitoringMutex->Lock();
3092
3093         tmp = fLastAction;
3094
3095         fMonitoringMutex->UnLock();
3096
3097         return tmp;
3098 }
3099
3100 //______________________________________________________________________________________________
3101 void AliShuttle::SetLastAction(const char* action)
3102 {
3103         //
3104         // updates the monitoring variables
3105         //
3106
3107         fMonitoringMutex->Lock();
3108
3109         fLastAction = action;
3110         fLastActionTime = time(0);
3111
3112         fMonitoringMutex->UnLock();
3113 }
3114
3115 //______________________________________________________________________________________________
3116 const char* AliShuttle::GetRunParameter(const char* param)
3117 {
3118         //
3119         // returns run parameter read from DAQ logbook
3120         //
3121
3122         if(!fLogbookEntry) {
3123                 AliError("No logbook entry!");
3124                 return 0;
3125         }
3126
3127         return fLogbookEntry->GetRunParameter(param);
3128 }
3129
3130 //______________________________________________________________________________________________
3131 AliCDBEntry* AliShuttle::GetFromOCDB(const char* detector, const AliCDBPath& path)
3132 {
3133         //
3134         // returns object from OCDB valid for current run
3135         //
3136
3137         if (fTestMode & kErrorOCDB)
3138         {
3139                 Log(detector, "GetFromOCDB - In TESTMODE - Simulating error with OCDB");
3140                 return 0;
3141         }
3142
3143         AliCDBStorage *sto = AliCDBManager::Instance()->GetStorage(fgkMainCDB);
3144         if (!sto)
3145         {
3146                 Log(detector, "GetFromOCDB - Cannot activate main OCDB for query!");
3147                 return 0;
3148         }
3149
3150         return dynamic_cast<AliCDBEntry*> (sto->Get(path, GetCurrentRun()));
3151 }
3152
3153 //______________________________________________________________________________________________
3154 Bool_t AliShuttle::SendMail(EMailTarget target, Int_t system)
3155 {
3156         //
3157         // sends a mail to the subdetector expert in case of preprocessor error
3158         //
3159
3160         if (fTestMode != kNone)
3161                 return kTRUE;
3162
3163         if (!fConfig->SendMail())
3164                 return kTRUE;
3165
3166         if (target == kDCSEMail || target == kFXSEMail) {
3167                 if (!fFirstProcessing)
3168                         return kTRUE;
3169         }
3170
3171         Int_t runMode = (Int_t)fConfig->GetRunMode();
3172         TString tmpStr;
3173         if (runMode == 0) tmpStr = " Nightly Test:";
3174         else tmpStr = " Data Taking:";
3175         void* dir = gSystem->OpenDirectory(GetShuttleLogDir());
3176         if (dir == NULL)
3177         {
3178                 if (gSystem->mkdir(GetShuttleLogDir(), kTRUE))
3179                 {
3180                         Log("SHUTTLE", Form("SendMail - Can't open directory <%s>", GetShuttleLogDir()));
3181                         return kFALSE;
3182                 }
3183
3184         } else {
3185                 gSystem->FreeDirectory(dir);
3186         }
3187
3188         // det experts in to
3189         TString to="";
3190         TIter *iterExperts = 0;
3191         if (target == kDCSEMail) {
3192                 iterExperts = new TIter(fConfig->GetAdmins(AliShuttleConfig::kAmanda));
3193         }
3194         else if (target == kFXSEMail) {
3195                 iterExperts = new TIter(fConfig->GetAdmins(system));
3196         }
3197         if (iterExperts) {
3198                 TObjString *anExpert=0;
3199                 while ((anExpert = (TObjString*) iterExperts->Next()))
3200                 {
3201                         to += Form("%s,", anExpert->GetName());
3202                 }
3203                 delete iterExperts;
3204         }
3205
3206         // add subdetector experts
3207         iterExperts = new TIter(fConfig->GetResponsibles(fCurrentDetector));
3208         TObjString *anExpert=0;
3209         while ((anExpert = (TObjString*) iterExperts->Next()))
3210         {
3211                 to += Form("%s,", anExpert->GetName());
3212         }
3213         delete iterExperts;
3214
3215         if (to.Length() > 0)
3216           to.Remove(to.Length()-1);
3217         AliDebug(2, Form("to: %s",to.Data()));
3218
3219         if (to.IsNull()) {
3220                 Log("SHUTTLE", Form("List of %d responsibles not set!", (Int_t) target));
3221                 return kFALSE;
3222         }
3223
3224         // SHUTTLE responsibles in cc
3225         TString cc="";
3226         TIter iterAdmins(fConfig->GetAdmins(AliShuttleConfig::kGlobal));
3227         TObjString *anAdmin=0;
3228         while ((anAdmin = (TObjString*) iterAdmins.Next()))
3229         {
3230                 cc += Form("%s,", anAdmin->GetName());
3231         }
3232         if (cc.Length() > 0)
3233           cc.Remove(cc.Length()-1);
3234         AliDebug(2, Form("cc: %s",to.Data()));
3235
3236         // mail body
3237         TString bodyFileName;
3238         bodyFileName.Form("%s/mail.body", GetShuttleLogDir());
3239         gSystem->ExpandPathName(bodyFileName);
3240
3241         ofstream mailBody;
3242         mailBody.open(bodyFileName, ofstream::out);
3243
3244         if (!mailBody.is_open())
3245         {
3246                 Log("SHUTTLE", Form("Could not open mail body file %s", bodyFileName.Data()));
3247                 return kFALSE;
3248         }
3249
3250
3251         TString subject;
3252         TString body;
3253
3254         if (target == kDCSEMail){
3255                 subject = Form("%s CRITICAL Retrieval of data points for %s FAILED in run %d !",
3256                                 tmpStr.Data(), fCurrentDetector.Data(), GetCurrentRun());
3257                 AliDebug(2, Form("subject: %s", subject.Data()));
3258
3259                 body = Form("Dear DCS experts, \n\n");
3260                 body += Form("SHUTTLE couldn\'t retrieve the data points for detector %s "
3261                              "in run %d!!\n\n", fCurrentDetector.Data(), GetCurrentRun());
3262         }
3263         else if (target == kFXSEMail){
3264                 subject = Form("%s CRITICAL FXS communication for %s FAILED in run %d !",
3265                                 tmpStr.Data(), fCurrentDetector.Data(), GetCurrentRun());
3266                 AliDebug(2, Form("subject: %s", subject.Data()));
3267                 TString sys;
3268                 if (system == kDAQ) sys="DAQ";
3269                 else if (system == kDCS) sys="DCS";
3270                 else if (system == kHLT) sys="HLT";
3271                 else return kFALSE;
3272                 body = Form("Dear  %s FXS experts, \n\n",sys.Data());
3273                 body += Form("SHUTTLE couldn\'t retrieve data from the FXS for detector %s "
3274                              "in run %d!!\n\n", fCurrentDetector.Data(), GetCurrentRun());
3275         }
3276         else {
3277                 subject = Form("%s %s Shuttle preprocessor FAILED in run %d (run type = %s)!",
3278                                        tmpStr.Data(), fCurrentDetector.Data(), GetCurrentRun(), GetRunType());
3279                 AliDebug(2, Form("subject: %s", subject.Data()));
3280
3281                 body = Form("Dear %s expert(s), \n\n", fCurrentDetector.Data());
3282                 body += Form("SHUTTLE just detected that your preprocessor "
3283                              "failed processing run %d (run type = %s)!!\n\n",
3284                              GetCurrentRun(), GetRunType());
3285         }
3286
3287         body += Form("Please check %s status on the SHUTTLE monitoring page: \n\n",
3288                                 fCurrentDetector.Data());
3289         if (fConfig->GetRunMode() == AliShuttleConfig::kTest)
3290         {
3291                 body += Form("\thttp://pcalimonitor.cern.ch/shuttle.jsp?time=24 \n\n");
3292         } else {
3293                 body += Form("\thttp://pcalimonitor.cern.ch/shuttle.jsp?instance=PROD&time=24 \n\n");
3294         }
3295
3296
3297         TString logFolder = "logs";
3298         if (fConfig->GetRunMode() == AliShuttleConfig::kProd)
3299                 logFolder += "_PROD";
3300
3301
3302         body += Form("Find the %s log for the current run on \n\n"
3303                 "\thttp://pcalishuttle01.cern.ch:8880/%s/%d/%d/%s.log \n\n",
3304                      fCurrentDetector.Data(), logFolder.Data(), GetCurrentRun()/10000,
3305                                 GetCurrentRun(), fCurrentDetector.Data());
3306         body += Form("The last 10 lines of %s log file are following:\n\n", fCurrentDetector.Data());
3307
3308         AliDebug(2, Form("Body begin: %s", body.Data()));
3309
3310         mailBody << body.Data();
3311         mailBody.close();
3312         mailBody.open(bodyFileName, ofstream::out | ofstream::app);
3313
3314         TString logFileName = Form("%s/%d/%d/%s.log", GetShuttleLogDir(),
3315                 GetCurrentRun()/10000, GetCurrentRun(), fCurrentDetector.Data());
3316         TString tailCommand = Form("tail -n 10 %s >> %s", logFileName.Data(), bodyFileName.Data());
3317         if (gSystem->Exec(tailCommand.Data()))
3318         {
3319                 mailBody << Form("%s log file not found ...\n\n", fCurrentDetector.Data());
3320         }
3321
3322         TString endBody = Form("------------------------------------------------------\n\n");
3323         endBody += Form("In case of problems please contact the SHUTTLE core team.\n\n");
3324         endBody += "Please do not answer this message directly, it is automatically generated.\n\n";
3325         endBody += "Greetings,\n\n \t\t\tthe SHUTTLE\n";
3326
3327         AliDebug(2, Form("Body end: %s", endBody.Data()));
3328
3329         mailBody << endBody.Data();
3330
3331         mailBody.close();
3332
3333         // send mail!
3334         TString mailCommand = Form("mail -s \"%s\" -c %s %s < %s",
3335                                                 subject.Data(),
3336                                                 cc.Data(),
3337                                                 to.Data(),
3338                                                 bodyFileName.Data());
3339         AliDebug(2, Form("mail command: %s", mailCommand.Data()));
3340
3341         Bool_t result = gSystem->Exec(mailCommand.Data());
3342
3343         return result == 0;
3344 }
3345 //______________________________________________________________________________________________
3346 const char* AliShuttle::GetRunType()
3347 {
3348         //
3349         // returns run type read from "run type" logbook
3350         //
3351
3352         if(!fLogbookEntry) {
3353                 AliError("No logbook entry!");
3354                 return 0;
3355         }
3356
3357         return fLogbookEntry->GetRunType();
3358 }
3359
3360 //______________________________________________________________________________________________
3361 Bool_t AliShuttle::GetHLTStatus()
3362 {
3363         // Return HLT status (ON=1 OFF=0)
3364         // Converts the HLT status from the mode string read in the run logbook (not just a bool)
3365
3366         if(!fLogbookEntry) {
3367                 AliError("No logbook entry!");
3368                 return 0;
3369         }
3370
3371         // TODO implement when HLTMode is inserted in run logbook
3372         TString hltMode = fLogbookEntry->GetRunParameter("HLTmode");
3373         TSubString firstChar = hltMode(0,1);
3374         AliDebug(2,Form("First char = %s ",firstChar.Data()));
3375         if (firstChar == "A") {
3376                 return kFALSE;
3377         }
3378         else if ((firstChar == "B") || (firstChar == "C") || (firstChar == "D") || (firstChar == "E")) {
3379                 return kTRUE;
3380         }
3381         else {
3382                 Log("SHUTTLE","Unexpected HLT mode! Returning 0....");
3383                 return kFALSE;
3384         }
3385 }
3386
3387 //______________________________________________________________________________________________
3388 const char* AliShuttle::GetTriggerConfiguration()
3389 {
3390         // Receives the trigger configuration from the DAQ logbook for the current run
3391
3392         // check connection, if needed reconnect
3393         if (!Connect(3))
3394                 return 0;
3395
3396         TString sqlQuery;
3397         sqlQuery.Form("SELECT configFile FROM logbook_trigger_config WHERE run = %d", GetCurrentRun());
3398         TSQLResult* result = fServer[3]->Query(sqlQuery);
3399         if (!result)
3400         {
3401                 Log("SHUTTLE", Form("ERROR: Can't execute query <%s>!", sqlQuery.Data()));
3402                 return 0;
3403         }
3404
3405         if (result->GetRowCount() == 0)
3406         {
3407                 Log("SHUTTLE", "ERROR: Trigger configuration not found in logbook_trigger_config");
3408                 delete result;
3409                 return 0;
3410         }
3411
3412         TSQLRow* row = result->Next();
3413         if (!row)
3414         {
3415                 Log("SHUTTLE", "ERROR: Could not receive logbook_trigger_config data");
3416                 delete result;
3417                 return 0;
3418         }
3419
3420         // static, so that pointer remains valid when it is returned to the calling class
3421         static TString triggerConfig(row->GetField(0));
3422
3423         delete row;
3424         row = 0;
3425
3426         delete result;
3427         result = 0;
3428
3429         Log("SHUTTLE", Form("Found trigger configuration: %s", triggerConfig.Data()));
3430
3431         return triggerConfig;
3432 }
3433
3434 //______________________________________________________________________________________________
3435 void AliShuttle::SetShuttleTempDir(const char* tmpDir)
3436 {
3437         //
3438         // sets Shuttle temp directory
3439         //
3440
3441         fgkShuttleTempDir = gSystem->ExpandPathName(tmpDir);
3442 }
3443
3444 //______________________________________________________________________________________________
3445 void AliShuttle::SetShuttleLogDir(const char* logDir)
3446 {
3447         //
3448         // sets Shuttle log directory
3449         //
3450
3451         fgkShuttleLogDir = gSystem->ExpandPathName(logDir);
3452 }
3453 //______________________________________________________________________________________________
3454 Bool_t AliShuttle::TouchFile()
3455 {
3456         //
3457         // touching a file on the grid if run has been DONE
3458         //
3459
3460         if (!gGrid)
3461         {
3462                 Log("SHUTTLE",Form("No TGrid connection estabilished!"));
3463                 Log("SHUTTLE",Form("Could not touch file for run %i",GetCurrentRun()));
3464                 return kFALSE;
3465         }
3466
3467         TString dir;
3468         dir.Form("%s%d/SHUTTLE_DONE", fConfig->GetAlienPath(), GetCurrentYear());
3469         // checking whether directory for touch command exists
3470         TString commandLs;
3471         commandLs.Form("ls %s",dir.Data());
3472         TGridResult *resultLs = dynamic_cast<TGridResult*>(gGrid->Command(commandLs));
3473         if (!resultLs){
3474                 Log("SHUTTLE",Form("No result for %s command, returning without touching",commandLs.Data()));
3475                 return kFALSE;
3476         }
3477         TMap *mapLs = dynamic_cast<TMap*>(resultLs->At(0));
3478         if (!mapLs){
3479                 Log("SHUTTLE",Form("No map for %s command, returning without touching",commandLs.Data()));
3480                 delete resultLs;
3481                 resultLs = 0x0;
3482                 return kFALSE;
3483         }
3484         TObjString *valueLsPath = dynamic_cast<TObjString*>(mapLs->GetValue("path"));
3485         if (!valueLsPath || (TString)(valueLsPath->GetString()).CompareTo(dir)!=1){
3486                 Log("SHUTTLE",Form("No directory %s found, creating it",dir.Data()));
3487
3488                 // creating the directory
3489
3490                 Bool_t boolMkdir = gGrid->Mkdir(dir.Data());
3491                 if (!boolMkdir) {
3492                         Log("SHUTTLE",Form("Impossible to create dir %s in alien catalogue for run %i!",dir.Data(),GetCurrentRun()));
3493                         delete resultLs;
3494                         resultLs = 0x0;
3495                         return kFALSE;
3496                 }
3497                 Log("SHUTTLE",Form("Directory %s successfully created in alien catalogue for run %i",dir.Data(),GetCurrentRun()));
3498         }
3499         else {
3500                 Log("SHUTTLE",Form("Directory %s correctly found for run %i",dir.Data(),GetCurrentRun()));
3501         }
3502
3503         delete resultLs;
3504         resultLs = 0x0;
3505
3506         TString command;
3507         command.Form("touch %s/%i", dir.Data(), GetCurrentRun());
3508         Log("SHUTTLE", Form("Creating entry in file catalog: %s", command.Data()));
3509         TGridResult *resultTouch = dynamic_cast<TGridResult*>(gGrid->Command(command));
3510         if (!resultTouch){
3511                 Log("SHUTTLE",Form("No result for touching command, returning without touching for run %i",GetCurrentRun()));
3512                 return kFALSE;
3513         }
3514         TMap *mapTouch = dynamic_cast<TMap*>(resultTouch->At(0));
3515         if (!mapTouch){
3516                 Log("SHUTTLE",Form("No map for touching command, returning without touching for run %i",GetCurrentRun()));
3517                 delete resultTouch;
3518                 resultTouch = 0x0;
3519                 return kFALSE;
3520         }
3521         TObjString *valueTouch = dynamic_cast<TObjString*>(mapTouch->GetValue("__result__"));
3522         if (!valueTouch){
3523                 Log("SHUTTLE",Form("No value for \"__result__\" key set in the map for touching command, returning without touching for run %i",GetCurrentRun()));
3524                 delete resultTouch;
3525                 resultTouch = 0x0;
3526                 return kFALSE;
3527         }
3528         if (valueTouch->GetString()!="1"){
3529                 Log("SHUTTLE",Form("Failing the touching command, returning without touching for run %i",GetCurrentRun()));
3530                 delete resultTouch;
3531                 resultTouch = 0x0;
3532                 return kFALSE;
3533         }
3534         delete resultTouch;
3535         resultTouch = 0x0;
3536         Log("SHUTTLE", "Sucessfully touched the file");
3537         return kTRUE;
3538 }
3539 //______________________________________________________________________________________________
3540 UInt_t AliShuttle::GetStartTimeDCSQuery()
3541 {
3542         // Return Start Time for the DCS query
3543         //
3544         // The call is delegated to AliShuttleInterface
3545
3546         return GetCurrentStartTime()-fConfig->GetDCSQueryOffset();
3547 }
3548 //______________________________________________________________________________________________
3549 UInt_t AliShuttle::GetEndTimeDCSQuery()
3550 {
3551         // Return End Time for the DCS query
3552         //
3553         // The call is delegated to AliShuttleInterface
3554
3555         return GetCurrentEndTime()+fConfig->GetDCSQueryOffset();
3556 }
3557