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         TMonaLisaText  mlRunType("SHUTTLE_runtype", runType);
1207
1208         TList mlList;
1209         mlList.Add(&mlStatus);
1210         mlList.Add(&mlRunType);
1211
1212         TString mlID;
1213         mlID.Form("%d", GetCurrentRun());
1214         fMonaLisa->SendParameters(&mlList, mlID);
1215 }
1216
1217 //______________________________________________________________________________________________
1218 Int_t AliShuttle::GetMem(Int_t pid)
1219 {
1220         // invokes ps to get the memory consumption of the process <pid>
1221         // returns -1 in case of error
1222
1223         TString checkStr;
1224         checkStr.Form("ps -o vsize --pid %d | tail -n 1", pid);
1225         FILE* pipe = gSystem->OpenPipe(checkStr, "r");
1226         if (!pipe)
1227         {
1228                 Log("SHUTTLE", Form("Process - Error: "
1229                         "Could not open pipe to %s", checkStr.Data()));
1230                 return -1;
1231         }
1232
1233         char buffer[100];
1234         if (!fgets(buffer, 100, pipe))
1235         {
1236                 Log("SHUTTLE", "Process - Error: ps did not return anything");
1237                 gSystem->ClosePipe(pipe);
1238                 return -1;
1239         }
1240         gSystem->ClosePipe(pipe);
1241
1242         //Log("SHUTTLE", Form("ps returned %s", buffer));
1243
1244         Int_t mem = 0;
1245         if ((sscanf(buffer, "%d\n", &mem) != 1) || !mem)
1246         {
1247                 Log("SHUTTLE", "Process - Error: Could not parse output of ps");
1248                 return -1;
1249         }
1250
1251         return mem;
1252 }
1253
1254 //______________________________________________________________________________________________
1255 Bool_t AliShuttle::Process(AliShuttleLogbookEntry* entry)
1256 {
1257         //
1258         // Makes data retrieval for all detectors in the configuration.
1259         // entry: Shuttle logbook entry, contains run paramenters and status of detectors
1260         // (Unprocessed, Inactive, Failed or Done).
1261         // Returns kFALSE in case of error occured and kTRUE otherwise
1262         //
1263
1264         if (!entry) return kFALSE;
1265
1266         fLogbookEntry = entry;
1267
1268         Log("SHUTTLE", Form("\t\t\t^*^*^*^*^*^*^*^*^*^*^*^* run %d: START ^*^*^*^*^*^*^*^*^*^*^*^*",
1269                                         GetCurrentRun()));
1270
1271         CountOpenRuns();
1272
1273         // Send the information to ML
1274         SendMLRunInfo("Processing");
1275
1276         if (fLogbookEntry->IsDone())
1277         {
1278                 Log("SHUTTLE","Process - Shuttle is already DONE. Updating logbook");
1279                 UpdateShuttleLogbook("shuttle_done");
1280                 fLogbookEntry = 0;
1281                 return kTRUE;
1282         }
1283
1284         // read test mode if flag is set
1285         if (fReadTestMode)
1286         {
1287                 fTestMode = kNone;
1288                 TString logEntry(entry->GetRunParameter("log"));
1289                 //printf("log entry = %s\n", logEntry.Data());
1290                 TString searchStr("Testmode: ");
1291                 Int_t pos = logEntry.Index(searchStr.Data());
1292                 //printf("%d\n", pos);
1293                 if (pos >= 0)
1294                 {
1295                         TSubString subStr = logEntry(pos + searchStr.Length(), logEntry.Length());
1296                         //printf("%s\n", subStr.String().Data());
1297                         TString newStr(subStr.Data());
1298                         TObjArray* token = newStr.Tokenize(' ');
1299                         if (token)
1300                         {
1301                                 //token->Print();
1302                                 TObjString* tmpStr = dynamic_cast<TObjString*> (token->First());
1303                                 if (tmpStr)
1304                                 {
1305                                         Int_t testMode = tmpStr->String().Atoi();
1306                                         if (testMode > 0)
1307                                         {
1308                                                 Log("SHUTTLE", Form("Process - Enabling test mode %d", testMode));
1309                                                 SetTestMode((TestMode) testMode);
1310                                         }
1311                                 }
1312                                 delete token;
1313                         }
1314                 }
1315         }
1316
1317         fLogbookEntry->Print("all");
1318
1319         // Initialization
1320         Bool_t hasError = kFALSE;
1321
1322         // Set the CDB and Reference folders according to the year and LHC period
1323         TString lhcPeriod(GetLHCPeriod());
1324         if (lhcPeriod.Length() == 0)
1325         {
1326                 Log("SHUTTLE","Process - LHCPeriod not found in logbook!");
1327                 return 0;
1328         }
1329
1330         // build cdb paths (repeat each time, LHCperiod might have changed)
1331         fgkMainCDB.Form("alien://folder=%s%d/%s/OCDB?user=alidaq?cacheFold=/tmp/OCDBCache",
1332                                         fConfig->GetAlienPath(), GetCurrentYear(), lhcPeriod.Data());
1333
1334         fgkMainRefStorage.Form("alien://folder=%s%d/%s/Reference?user=alidaq?cacheFold=/tmp/OCDBCache",
1335                                         fConfig->GetAlienPath(), GetCurrentYear(), lhcPeriod.Data());
1336
1337         // Loop on detectors in the configuration
1338         TIter iter(fConfig->GetDetectors());
1339         TObjString* aDetector = 0;
1340
1341         Bool_t first = kTRUE;
1342
1343         while ((aDetector = (TObjString*) iter.Next()))
1344         {
1345                 fCurrentDetector = aDetector->String();
1346
1347                 if (ContinueProcessing() == kFALSE)
1348                         continue;
1349
1350                 if (first)
1351                 {
1352                   // only read QueryCDB when needed and only once
1353                   AliCDBStorage *mainCDBSto = AliCDBManager::Instance()->GetStorage(fgkMainCDB);
1354                   if(mainCDBSto) mainCDBSto->QueryCDB(GetCurrentRun());
1355                   AliCDBStorage *mainRefSto = AliCDBManager::Instance()->GetStorage(fgkMainRefStorage);
1356                   if(mainRefSto) mainRefSto->QueryCDB(GetCurrentRun());
1357                   first = kFALSE;
1358                 }
1359
1360                 Log("SHUTTLE", Form("\t\t\t****** run %d - %s: START  ******",
1361                                                 GetCurrentRun(), aDetector->GetName()));
1362
1363                 for(Int_t iSys=0;iSys<3;iSys++) fFXSCalled[iSys]=kFALSE;
1364
1365                 Int_t initialMem = GetMem(getpid());
1366                 Log("SHUTTLE", Form("Memory consumption before forking is %d", initialMem));
1367
1368                 Log(fCurrentDetector.Data(), "Process - Starting processing");
1369
1370                 Int_t pid = fork();
1371
1372                 if (pid < 0)
1373                 {
1374                         Log("SHUTTLE", "Process - ERROR: Forking failed");
1375                 }
1376                 else if (pid > 0)
1377                 {
1378                         // parent
1379                         Log("SHUTTLE", Form("Process - In parent process of %d - %s: Starting monitoring",
1380                                                         GetCurrentRun(), aDetector->GetName()));
1381
1382                         Long_t begin = time(0);
1383
1384                         int status; // to be used with waitpid, on purpose an int (not Int_t)!
1385                         while (waitpid(pid, &status, WNOHANG) == 0)
1386                         {
1387                                 Long_t expiredTime = time(0) - begin;
1388
1389                                 if (expiredTime > fConfig->GetPPTimeOut())
1390                                 {
1391                                         TString logMsg;
1392                                         AliShuttleStatus *currentStatus = ReadShuttleStatus();
1393                                         AliShuttleStatus::Status newStatus = AliShuttleStatus::kInvalid;
1394
1395                                         if (currentStatus->GetStatus() <= AliShuttleStatus::kPPDone)
1396                                         {
1397                                                 // in case pp not yet done set status to kPPTimeOut
1398
1399                                                 logMsg.Form("Process - Process of %s timed out. Run time: %d seconds. Killing...",
1400                                                                 fCurrentDetector.Data(), expiredTime);
1401                                                 newStatus = AliShuttleStatus::kPPTimeOut;
1402                                         }
1403                                         else if (currentStatus->GetStatus() == AliShuttleStatus::kStoreStarted)
1404                                         {
1405                                                 // in case the pp goes in TimeOut while storing the objects in the OCDB
1406                                                 // set status to kStoreError
1407
1408                                                 logMsg.Form("Process - Process of %s timed out while storing the OCDB object. Run time: %d seconds. Killing... and setting status to StoreError.",
1409                                                                 fCurrentDetector.Data(), expiredTime);
1410                                                 newStatus = AliShuttleStatus::kStoreError;
1411                                         }
1412                                         else
1413                                         {
1414                                                 // in other cases don't change the status
1415
1416                                                 logMsg.Form("Process - Process of %s timed out in status = %s. Run time: %d seconds. Killing... without changing the status",
1417                                                                 fCurrentDetector.Data(), currentStatus->GetStatusName(), expiredTime);
1418                                         }
1419
1420                                         Log("SHUTTLE", logMsg);
1421                                         Log(fCurrentDetector, logMsg);
1422
1423                                         kill(pid, 9);
1424
1425                                         if (newStatus != AliShuttleStatus::kInvalid)
1426                                                 UpdateShuttleStatus(newStatus);
1427                                         hasError = kTRUE;
1428
1429                                         gSystem->Sleep(1000);
1430                                 }
1431                                 else
1432                                 {
1433                                         gSystem->Sleep(1000);
1434
1435                                         Int_t mem = GetMem(pid);
1436
1437                                         if (mem < 0)
1438                                                 continue;
1439
1440                                         mem -= initialMem;
1441                                         if (mem < 0)
1442                                                 mem = 0;
1443
1444                                         if (expiredTime % 60 == 0)
1445                                         {
1446                                                 Log("SHUTTLE", Form("Process - %s: Checking process. "
1447                                                         "Run time: %d seconds - Memory consumption: %d KB",
1448                                                         fCurrentDetector.Data(), expiredTime, mem));
1449                                                 SendAlive();
1450                                         }
1451
1452                                         if (mem > fConfig->GetPPMaxMem())
1453                                         {
1454                                                 TString tmp;
1455                                                 tmp.Form("Process - Process exceeds maximum allowed memory "
1456                                                         "(%d KB > %d KB). Killing...",
1457                                                         mem, fConfig->GetPPMaxMem());
1458                                                 Log("SHUTTLE", tmp);
1459                                                 Log(fCurrentDetector, tmp);
1460
1461                                                 kill(pid, 9);
1462
1463                                                 UpdateShuttleStatus(AliShuttleStatus::kPPOutOfMemory);
1464                                                 hasError = kTRUE;
1465
1466                                                 gSystem->Sleep(1000);
1467                                         }
1468                                 }
1469                         }
1470
1471                         Log("SHUTTLE", Form("Process - In parent process of %d - %s: Client has terminated.",
1472                                                                 GetCurrentRun(), aDetector->GetName()));
1473
1474                         if (WIFEXITED(status))
1475                         {
1476                                 Int_t returnCode = WEXITSTATUS(status);
1477
1478                                 Log("SHUTTLE", Form("Process - %s: the return code is %d", fCurrentDetector.Data(),
1479                                                                                 returnCode));
1480
1481                                 if (returnCode == 0) hasError = kTRUE;
1482                         }
1483                 }
1484                 else if (pid == 0)
1485                 {
1486                         // child
1487                         Log("SHUTTLE", Form("Process - In child process of %d - %s", GetCurrentRun(),
1488                                 aDetector->GetName()));
1489
1490                         Log("SHUTTLE", Form("Process - Redirecting output to %s log",fCurrentDetector.Data()));
1491
1492                         if ((freopen(GetLogFileName(fCurrentDetector), "a", stdout)) == 0)
1493                         {
1494                                 Log("SHUTTLE", "Process - Could not freopen stdout");
1495                         }
1496                         else
1497                         {
1498                                 fOutputRedirected = kTRUE;
1499                                 if ((dup2(fileno(stdout), fileno(stderr))) < 0)
1500                                         Log("SHUTTLE", "Process - Could not redirect stderr");
1501
1502                         }
1503
1504                         TString wd = gSystem->WorkingDirectory();
1505                         Int_t dir_lev1 = GetCurrentRun()/10000;
1506                         TString tmpDir = Form("%s/%d/%d/%s_process", GetShuttleTempDir(),
1507                                 dir_lev1, GetCurrentRun(), fCurrentDetector.Data());
1508
1509                         Int_t result = gSystem->GetPathInfo(tmpDir.Data(), 0, (Long64_t*) 0, 0, 0);
1510                         if (!result) // temp dir already exists!
1511                         {
1512                                 Log(fCurrentDetector.Data(),
1513                                         Form("Process - %s dir already exists! Removing...", tmpDir.Data()));
1514                                 gSystem->Exec(Form("rm -rf %s",tmpDir.Data()));
1515                         }
1516
1517                         if (gSystem->mkdir(tmpDir.Data(), 1))
1518                         {
1519                                 Log(fCurrentDetector.Data(), "Process - could not make temp directory!!");
1520                                 gSystem->Exit(1);
1521                         }
1522
1523                         if (!gSystem->ChangeDirectory(tmpDir.Data()))
1524                         {
1525                                 Log(fCurrentDetector.Data(), "Process - could not change directory!!");
1526                                 gSystem->Exit(1);
1527                         }
1528
1529                         Int_t success = ProcessCurrentDetector();
1530
1531                         gSystem->ChangeDirectory(wd.Data());
1532
1533                         if (success == 1) // Preprocessor finished successfully!
1534                         {
1535                                 // remove temporary folder or DCS map
1536                                 if (!fConfig->KeepTempFolder())
1537                                 {
1538                                         gSystem->Exec(Form("rm -rf %s",tmpDir.Data()));
1539                                 } else if (!fConfig->KeepDCSMap())
1540                                 {
1541                                         gSystem->Exec(Form("rm -f %s/DCSMap.root",tmpDir.Data()));
1542                                 }
1543
1544                                 // Update time_processed field in FXS DB
1545                                 if (UpdateTable() == kFALSE)
1546                                         Log("SHUTTLE", Form("Process - %s: Could not update FXS databases!",
1547                                                         fCurrentDetector.Data()));
1548
1549                                 // Transfer the data from local storage to main storage (Grid)
1550                                 if (StoreOCDB() == kFALSE)
1551                                         success = kFALSE;
1552                         }
1553                         else if (success == 0)
1554                         {
1555                                 Log("SHUTTLE",
1556                                         Form("\t\t\t****** run %d - %s: PP ERROR ******",
1557                                                 GetCurrentRun(), aDetector->GetName()));
1558                         }
1559
1560                         for (UInt_t iSys=0; iSys<3; iSys++)
1561                         {
1562                                 if (fFXSCalled[iSys]) fFXSlist[iSys].Clear();
1563                         }
1564
1565                         Log("SHUTTLE", Form("Process - Client process of %d - %s is exiting now with %d.",
1566                                                         GetCurrentRun(), aDetector->GetName(), success));
1567
1568                         // the client exits here
1569                         gSystem->Exit(success);
1570
1571                         AliError("We should never get here!!!");
1572                 }
1573         }
1574
1575         Log("SHUTTLE", Form("\t\t\t^*^*^*^*^*^*^*^*^*^*^*^* run %d: FINISH ^*^*^*^*^*^*^*^*^*^*^*^*",
1576                                                         GetCurrentRun()));
1577
1578         //check if shuttle is done for this run, if so update logbook
1579         TObjArray checkEntryArray;
1580         checkEntryArray.SetOwner(1);
1581         TString whereClause = Form("where run=%d", GetCurrentRun());
1582         if (!QueryShuttleLogbook(whereClause.Data(), checkEntryArray) ||
1583                         checkEntryArray.GetEntries() == 0) {
1584                 Log("SHUTTLE", Form("Process - Warning: Cannot check status of run %d on Shuttle logbook!",
1585                                                 GetCurrentRun()));
1586                 return hasError == kFALSE;
1587         }
1588
1589         AliShuttleLogbookEntry* checkEntry = dynamic_cast<AliShuttleLogbookEntry*>
1590                                                 (checkEntryArray.At(0));
1591
1592         if (checkEntry)
1593         {
1594                 if (checkEntry->IsDone())
1595                 {
1596                         Log("SHUTTLE","Process - Shuttle is DONE. Updating logbook");
1597                         UpdateShuttleLogbook("shuttle_done");
1598                 }
1599                 else
1600                 {
1601                         for (UInt_t iDet=0; iDet<NDetectors(); iDet++)
1602                         {
1603                                 if (checkEntry->GetDetectorStatus(iDet) == AliShuttleLogbookEntry::kUnprocessed)
1604                                 {
1605                                         AliDebug(2, Form("Run %d: setting %s as \"not first time unprocessed\"",
1606                                                         checkEntry->GetRun(), GetDetName(iDet)));
1607                                         fFirstUnprocessed[iDet] = kFALSE;
1608                                 }
1609                         }
1610                         SendMLRunInfo("Pending");
1611                 }
1612         }
1613
1614         fLogbookEntry = 0;
1615
1616         return hasError == kFALSE;
1617 }
1618
1619 //______________________________________________________________________________________________
1620 Int_t AliShuttle::ProcessCurrentDetector()
1621 {
1622         //
1623         // Makes data retrieval just for a specific detector (fCurrentDetector).
1624         // Threre should be a configuration for this detector.
1625
1626         Log("SHUTTLE", Form("ProcessCurrentDetector - Retrieving values for %s, run %d",
1627                                                 fCurrentDetector.Data(), GetCurrentRun()));
1628
1629         TString wd = gSystem->WorkingDirectory();
1630
1631         if (!CleanReferenceStorage(fCurrentDetector.Data()))
1632                 return 0;
1633
1634         gSystem->ChangeDirectory(wd.Data());
1635
1636         // call preprocessor
1637         AliPreprocessor* aPreprocessor =
1638                 dynamic_cast<AliPreprocessor*> (fPreprocessorMap.GetValue(fCurrentDetector));
1639
1640         // check if the preprocessor wants to process this run type
1641         if (aPreprocessor->ProcessRunType() == kFALSE)
1642         {
1643                 UpdateShuttleStatus(AliShuttleStatus::kSkipped);
1644                 UpdateShuttleLogbook(fCurrentDetector, "DONE");
1645                 if (!UpdateTableSkippedCase(fCurrentDetector.Data()))
1646                 {
1647                         AliError(Form("Could not update FXS tables for run %d !", GetCurrentRun()));
1648                 }
1649                 Log(fCurrentDetector, Form("ProcessCurrentDetector - %s preprocessor is not interested in this run type", fCurrentDetector.Data()));
1650
1651                 return 2;
1652         }
1653
1654         TMap* dcsMap = new TMap();
1655
1656         aPreprocessor->Initialize(GetCurrentRun(), GetCurrentStartTime(), GetCurrentEndTime());
1657
1658         Bool_t processDCS = aPreprocessor->ProcessDCS();
1659
1660         if (!processDCS)
1661         {
1662                 Log(fCurrentDetector, "ProcessCurrentDetector -"
1663                         " The preprocessor requested to skip the retrieval of DCS values");
1664         }
1665         else if (fTestMode & kSkipDCS)
1666         {
1667                 Log(fCurrentDetector, "ProcessCurrentDetector - In TESTMODE: Skipping DCS processing");
1668         }
1669         else if (fTestMode & kErrorDCS)
1670         {
1671                 Log(fCurrentDetector, "ProcessCurrentDetector - In TESTMODE: Simulating DCS error");
1672                 UpdateShuttleStatus(AliShuttleStatus::kDCSStarted);
1673                 UpdateShuttleStatus(AliShuttleStatus::kDCSError);
1674                 delete dcsMap;
1675                 return 0;
1676         } else {
1677
1678                 UpdateShuttleStatus(AliShuttleStatus::kDCSStarted);
1679
1680                 // Query DCS archive
1681                 Int_t nServers = fConfig->GetNServers(fCurrentDetector);
1682
1683                 for (int iServ=0; iServ<nServers; iServ++)
1684                 {
1685
1686                         TString host(fConfig->GetDCSHost(fCurrentDetector, iServ));
1687                         Int_t port = fConfig->GetDCSPort(fCurrentDetector, iServ);
1688                         Int_t multiSplit = fConfig->GetMultiSplit(fCurrentDetector, iServ);
1689
1690                         Log(fCurrentDetector, Form("ProcessCurrentDetector -"
1691                                         " Querying DCS Amanda server %s:%d (%d of %d)",
1692                                         host.Data(), port, iServ+1, nServers));
1693
1694                         TMap* aliasMap = 0;
1695                         TMap* dpMap = 0;
1696
1697                         if (fConfig->GetDCSAliases(fCurrentDetector, iServ)->GetEntries() > 0)
1698                         {
1699                                 Log(fCurrentDetector, Form("Querying %d DCS aliases", fConfig->GetDCSAliases(fCurrentDetector, iServ)->GetEntries()));
1700                                 aliasMap = GetValueSet(host, port,
1701                                                 fConfig->GetDCSAliases(fCurrentDetector, iServ),
1702                                                 kAlias, multiSplit);
1703                                 if (!aliasMap)
1704                                 {
1705                                         Log(fCurrentDetector,
1706                                                 Form("ProcessCurrentDetector -"
1707                                                         " Error retrieving DCS aliases from server %s."
1708                                                         " Sending mail to DCS experts!", host.Data()));
1709                                         UpdateShuttleStatus(AliShuttleStatus::kDCSError);
1710
1711                                         if (!SendMail(kDCSEMail))
1712                                                 Log("SHUTTLE", Form("ProcessCurrentDetector - "
1713                                                                     "Could not send mail to DCS experts!"));
1714
1715                                         delete dcsMap;
1716                                         return 0;
1717                                 }
1718                         }
1719
1720                         if (fConfig->GetDCSDataPoints(fCurrentDetector, iServ)->GetEntries() > 0)
1721                         {
1722                                 Log(fCurrentDetector, Form("Querying %d DCS data points", fConfig->GetDCSDataPoints(fCurrentDetector, iServ)->GetEntries()));
1723                                 dpMap = GetValueSet(host, port,
1724                                                 fConfig->GetDCSDataPoints(fCurrentDetector, iServ),
1725                                                 kDP, multiSplit);
1726                                 if (!dpMap)
1727                                 {
1728                                         Log(fCurrentDetector,
1729                                                 Form("ProcessCurrentDetector -"
1730                                                         " Error retrieving DCS data points from server %s."
1731                                                         " Sending mail to DCS experts!", host.Data()));
1732                                         UpdateShuttleStatus(AliShuttleStatus::kDCSError);
1733
1734                                         if (!SendMail(kDCSEMail))
1735                                                 Log("SHUTTLE", Form("ProcessCurrentDetector - "
1736                                                                     "Could not send mail to DCS experts!"));
1737
1738                                         if (aliasMap) delete aliasMap;
1739                                         delete dcsMap;
1740                                         return 0;
1741                                 }
1742                         }
1743
1744                         // merge aliasMap and dpMap into dcsMap
1745                         if(aliasMap) {
1746                                 TIter iter(aliasMap);
1747                                 TObjString* key = 0;
1748                                 while ((key = (TObjString*) iter.Next()))
1749                                         dcsMap->Add(key, aliasMap->GetValue(key->String()));
1750
1751                                 aliasMap->SetOwner(kFALSE);
1752                                 delete aliasMap;
1753                         }
1754
1755                         if(dpMap) {
1756                                 TIter iter(dpMap);
1757                                 TObjString* key = 0;
1758                                 while ((key = (TObjString*) iter.Next()))
1759                                         dcsMap->Add(key, dpMap->GetValue(key->String()));
1760
1761                                 dpMap->SetOwner(kFALSE);
1762                                 delete dpMap;
1763                         }
1764                 }
1765         }
1766
1767         // save map into file, to help debugging in case of preprocessor error
1768         TFile* f = TFile::Open("DCSMap.root","recreate");
1769         f->cd();
1770         dcsMap->Write("DCSMap", TObject::kSingleKey);
1771         f->Close();
1772         delete f;
1773
1774         // DCS Archive DB processing successful. Call Preprocessor!
1775         UpdateShuttleStatus(AliShuttleStatus::kPPStarted);
1776
1777         fFXSError = -1; // this variable is kTRUE after ::Process if an FXS error occured
1778
1779         UInt_t returnValue = aPreprocessor->Process(dcsMap);
1780
1781         if (fFXSError!=-1) {
1782                 UpdateShuttleStatus(AliShuttleStatus::kFXSError);
1783                 SendMail(kFXSEMail, fFXSError);
1784                 dcsMap->DeleteAll();
1785                 delete dcsMap;
1786                 return 0;
1787         }
1788
1789         if (returnValue > 0) // Preprocessor error!
1790         {
1791                 Log(fCurrentDetector, Form("ProcessCurrentDetector - "
1792                                 "Preprocessor failed. Process returned %d.", returnValue));
1793                 UpdateShuttleStatus(AliShuttleStatus::kPPError);
1794                 dcsMap->DeleteAll();
1795                 delete dcsMap;
1796                 return 0;
1797         }
1798
1799         // preprocessor ok!
1800         UpdateShuttleStatus(AliShuttleStatus::kPPDone);
1801         Log(fCurrentDetector, Form("ProcessCurrentDetector - %s preprocessor returned success",
1802                                 fCurrentDetector.Data()));
1803
1804         dcsMap->DeleteAll();
1805         delete dcsMap;
1806
1807         return 1;
1808 }
1809
1810 //______________________________________________________________________________________________
1811 void AliShuttle::CountOpenRuns()
1812 {
1813         // Query DAQ's Shuttle logbook and sends the number of open runs to ML
1814
1815         SendAlive();
1816
1817         // check connection, in case connect
1818         if (!Connect(3))
1819                 return;
1820
1821         TString sqlQuery;
1822         sqlQuery = Form("select count(*) from %s where shuttle_done=0", fConfig->GetShuttlelbTable());
1823
1824         TSQLResult* aResult = fServer[3]->Query(sqlQuery);
1825         if (!aResult) {
1826                 AliError(Form("Can't execute query <%s>!", sqlQuery.Data()));
1827                 return;
1828         }
1829
1830         AliDebug(2,Form("Query = %s", sqlQuery.Data()));
1831
1832         if (aResult->GetRowCount() == 0) {
1833                 AliError(Form("No result for query %s received", sqlQuery.Data()));
1834                 return;
1835         }
1836
1837         if (aResult->GetFieldCount() != 1) {
1838                 AliError(Form("Invalid field count for query %s received", sqlQuery.Data()));
1839                 return;
1840         }
1841
1842         TSQLRow* aRow = aResult->Next();
1843         if (!aRow) {
1844                 AliError(Form("Could not receive result of query %s", sqlQuery.Data()));
1845                 return;
1846         }
1847
1848         TString result(aRow->GetField(0), aRow->GetFieldLength(0));
1849         Int_t count = result.Atoi();
1850
1851         Log("SHUTTLE", Form("%d unprocessed runs", count));
1852
1853         delete aRow;
1854         delete aResult;
1855
1856         TMonaLisaValue mlStatus("SHUTTLE_openruns", count);
1857
1858         TList mlList;
1859         mlList.Add(&mlStatus);
1860
1861         fMonaLisa->SendParameters(&mlList, "__PROCESSINGINFO__");
1862 }
1863
1864 //______________________________________________________________________________________________
1865 Bool_t AliShuttle::QueryShuttleLogbook(const char* whereClause,
1866                 TObjArray& entries)
1867 {
1868         // Query DAQ's Shuttle logbook and fills detector status object.
1869         // Call QueryRunParameters to query DAQ logbook for run parameters.
1870         //
1871
1872         entries.SetOwner(1);
1873
1874         // check connection, in case connect
1875         if (!Connect(3)) return kFALSE;
1876
1877         TString sqlQuery;
1878         sqlQuery = Form("select * from %s %s order by run", fConfig->GetShuttlelbTable(), whereClause);
1879
1880         TSQLResult* aResult = fServer[3]->Query(sqlQuery);
1881         if (!aResult) {
1882                 AliError(Form("Can't execute query <%s>!", sqlQuery.Data()));
1883                 return kFALSE;
1884         }
1885
1886         AliDebug(2,Form("Query = %s", sqlQuery.Data()));
1887
1888         if(aResult->GetRowCount() == 0) {
1889                 Log("SHUTTLE", "No entries in Shuttle Logbook match request");
1890                 delete aResult;
1891                 return kTRUE;
1892         }
1893
1894         // TODO Check field count!
1895         const UInt_t nCols = 23;
1896         if (aResult->GetFieldCount() != (Int_t) nCols) {
1897                 Log("SHUTTLE", "Invalid SQL result field number!");
1898                 delete aResult;
1899                 return kFALSE;
1900         }
1901
1902         TSQLRow* aRow;
1903         while ((aRow = aResult->Next())) {
1904                 TString runString(aRow->GetField(0), aRow->GetFieldLength(0));
1905                 Int_t run = runString.Atoi();
1906
1907                 AliShuttleLogbookEntry *entry = QueryRunParameters(run);
1908                 if (!entry)
1909                         continue;
1910
1911                 // loop on detectors
1912                 for(UInt_t ii = 0; ii < nCols; ii++)
1913                         entry->SetDetectorStatus(aResult->GetFieldName(ii), aRow->GetField(ii));
1914
1915                 entries.AddLast(entry);
1916                 delete aRow;
1917         }
1918
1919         delete aResult;
1920         return kTRUE;
1921 }
1922
1923 //______________________________________________________________________________________________
1924 AliShuttleLogbookEntry* AliShuttle::QueryRunParameters(Int_t run)
1925 {
1926         //
1927         // Retrieve run parameters written in the DAQ logbook and sets them into AliShuttleLogbookEntry object
1928         //
1929
1930         // check connection, in case connect
1931         if (!Connect(3))
1932                 return 0;
1933
1934         TString sqlQuery;
1935         sqlQuery.Form("select * from %s where run=%d", fConfig->GetDAQlbTable(), run);
1936
1937         TSQLResult* aResult = fServer[3]->Query(sqlQuery);
1938         if (!aResult) {
1939                 Log("SHUTTLE", Form("Can't execute query <%s>!", sqlQuery.Data()));
1940                 return 0;
1941         }
1942
1943         if (aResult->GetRowCount() == 0) {
1944                 Log("SHUTTLE", Form("QueryRunParameters - No entry in DAQ Logbook for run %d. Skipping", run));
1945                 delete aResult;
1946                 return 0;
1947         }
1948
1949         if (aResult->GetRowCount() > 1) {
1950                 Log("SHUTTLE", Form("QueryRunParameters - UNEXPECTED: "
1951                                 "more than one entry in DAQ Logbook for run %d!", run));
1952                 delete aResult;
1953                 return 0;
1954         }
1955
1956         TSQLRow* aRow = aResult->Next();
1957         if (!aRow)
1958         {
1959                 Log("SHUTTLE", Form("QueryRunParameters - Could not retrieve row for run %d. Skipping", run));
1960                 delete aResult;
1961                 return 0;
1962         }
1963
1964         AliShuttleLogbookEntry* entry = new AliShuttleLogbookEntry(run);
1965
1966         for (Int_t ii = 0; ii < aResult->GetFieldCount(); ii++)
1967                 entry->SetRunParameter(aResult->GetFieldName(ii), aRow->GetField(ii));
1968
1969         delete aRow;
1970         delete aResult;
1971
1972         UInt_t startTime = entry->GetStartTime();
1973         UInt_t endTime = entry->GetEndTime();
1974         Bool_t ecsSuccess = entry->GetECSSuccess();
1975
1976         TString totEventsStr = entry->GetRunParameter("totalEvents");
1977         Int_t totEvents = totEventsStr.Atoi();
1978
1979         UInt_t now = time(0);
1980         // TODO make this a configuration parameter
1981         Int_t dcsDelay = fConfig->GetDCSDelay()+fConfig->GetDCSQueryOffset();
1982
1983         // runs are accepted if they have ecsSuccess set or more than 1 event
1984         if (startTime != 0 && endTime != 0 && endTime > startTime && (totEvents > 1 || ecsSuccess) && (endTime < now - dcsDelay))
1985         {
1986                 if (ecsSuccess == kFALSE)
1987                         Log("SHUTTLE", Form("Processing run %d although in status ECS failure, Reason: %s", run, entry->GetRunParameter("eor_reason")));
1988                 return entry;
1989         }
1990
1991         Bool_t skip = kFALSE;
1992
1993         if (endTime != 0 && endTime >= now - dcsDelay)
1994         {
1995                 Log("SHUTTLE", Form("Skipping run %d for now, because DCS buffer time is not yet expired", run));
1996         }
1997         else if (totEvents <= 1)
1998         {
1999                 Log("SHUTTLE", Form("QueryRunParameters - Run %d has 1 event or less - Skipping!", run));
2000                 skip = kTRUE;
2001         }
2002         else
2003         {
2004                 Log("SHUTTLE", Form("QueryRunParameters - Invalid parameters for Run %d: "
2005                         "startTime = %d, endTime = %d. Skipping (Shuttle won't be marked as DONE)!",
2006                         run, startTime, endTime));
2007         }
2008
2009         if (skip)
2010         {
2011                 Log("SHUTTLE", Form("Marking SHUTTLE skipped for run %d", run));
2012                 fLogbookEntry = entry;
2013                 if (!UpdateShuttleLogbook("shuttle_skipped"))
2014                 {
2015                         AliError(Form("Could not update logbook for run %d !", run));
2016                 }
2017                 if (!UpdateTableSkippedCase("ALL"))
2018                 {
2019                         AliError(Form("Could not update FXS tables for run %d !", run));
2020                 }
2021                 fLogbookEntry = 0;
2022         }
2023
2024         delete entry;
2025         return 0;
2026 }
2027
2028 //______________________________________________________________________________________________
2029 TMap* AliShuttle::GetValueSet(const char* host, Int_t port, const TSeqCollection* entries,
2030                               DCSType type, Int_t multiSplit)
2031 {
2032         // Retrieve all "entry" data points from the DCS server
2033         // host, port: TSocket connection parameters
2034         // entries: list of name of the alias or data point
2035         // type: kAlias or kDP
2036         // returns TMap of values, 0 when failure
2037
2038         AliDCSClient client(host, port, fTimeout, fRetries, multiSplit);
2039
2040         TMap* result = 0;
2041         if (type == kAlias)
2042         {
2043                 //result = client.GetAliasValues(entries, GetCurrentStartTime()-offset,
2044                 //      GetCurrentEndTime()+offset);
2045                 result = client.GetAliasValues(entries, GetStartTimeDCSQuery(),
2046                         GetEndTimeDCSQuery());
2047         }
2048         else if (type == kDP)
2049         {
2050                 //result = client.GetDPValues(entries, GetCurrentStartTime()-offset,
2051                 //      GetCurrentEndTime()+offset);
2052                 result = client.GetDPValues(entries, GetStartTimeDCSQuery(),
2053                         GetEndTimeDCSQuery());
2054         }
2055
2056         if (result == 0)
2057         {
2058                 Log(fCurrentDetector.Data(), Form("GetValueSet - Can't get entries! Reason: %s",
2059                         client.GetErrorString(client.GetResultErrorCode())));
2060                 if (client.GetResultErrorCode() == AliDCSClient::fgkServerError)
2061                         Log(fCurrentDetector.Data(), Form("GetValueSet - Server error code: %s",
2062                                 client.GetServerError().Data()));
2063
2064                 return 0;
2065         }
2066
2067         return result;
2068 }
2069
2070 //______________________________________________________________________________________________
2071 const char* AliShuttle::GetFile(Int_t system, const char* detector,
2072                 const char* id, const char* source)
2073 {
2074         // Get calibration file from file exchange servers
2075         // First queris the FXS database for the file name, using the run, detector, id and source info
2076         // then calls RetrieveFile(filename) for actual copy to local disk
2077         // run: current run being processed (given by Logbook entry fLogbookEntry)
2078         // detector: the Preprocessor name
2079         // id: provided as a parameter by the Preprocessor
2080         // source: provided by the Preprocessor through GetFileSources function
2081
2082         // check if test mode should simulate a FXS error
2083         if (fTestMode & kErrorFXSFiles)
2084         {
2085                 Log(detector, Form("GetFile - In TESTMODE - Simulating error while connecting to %s FXS", GetSystemName(system)));
2086                 return 0;
2087         }
2088
2089         // check connection, in case connect
2090         if (!Connect(system))
2091         {
2092                 Log(detector, Form("GetFile - Couldn't connect to %s FXS database", GetSystemName(system)));
2093                 fFXSError = system;
2094                 return 0;
2095         }
2096
2097         // Query preparation
2098         TString sourceName(source);
2099         Int_t nFields = 3;
2100         TString sqlQueryStart = Form("select filePath,size,fileChecksum from %s where",
2101                                                                 fConfig->GetFXSdbTable(system));
2102         TString whereClause = Form("run=%d and detector=\"%s\" and fileId=\"%s\"",
2103                                                                 GetCurrentRun(), detector, id);
2104
2105         if (system == kDAQ)
2106         {
2107                 whereClause += Form(" and DAQsource=\"%s\"", source);
2108         }
2109         else if (system == kDCS)
2110         {
2111                 sourceName="none";
2112         }
2113         else if (system == kHLT)
2114         {
2115                 whereClause += Form(" and DDLnumbers=\"%s\"", source);
2116         }
2117
2118         TString sqlQuery = Form("%s %s", sqlQueryStart.Data(), whereClause.Data());
2119
2120         AliDebug(2, Form("SQL query: \n%s",sqlQuery.Data()));
2121
2122         // Query execution
2123         TSQLResult* aResult = 0;
2124         aResult = dynamic_cast<TSQLResult*> (fServer[system]->Query(sqlQuery));
2125         if (!aResult) {
2126                 Log(detector, Form("GetFile - Can't execute SQL query to %s database for: id = %s, source = %s",
2127                                 GetSystemName(system), id, sourceName.Data()));
2128                 fFXSError = system;
2129                 return 0;
2130         }
2131
2132         if (aResult->GetRowCount() == 0)
2133         {
2134                 Log(detector,
2135                         Form("GetFile - No entry in %s FXS db for: id = %s, source = %s",
2136                                 GetSystemName(system), id, sourceName.Data()));
2137                 delete aResult;
2138                 return 0;
2139         }
2140
2141         if (aResult->GetRowCount() > 1) {
2142                 Log(detector,
2143                         Form("GetFile - More than one entry in %s FXS db for: id = %s, source = %s",
2144                                 GetSystemName(system), id, sourceName.Data()));
2145                 fFXSError = system;
2146                 delete aResult;
2147                 return 0;
2148         }
2149
2150         if (aResult->GetFieldCount() != nFields) {
2151                 Log(detector,
2152                         Form("GetFileName - Wrong field count in %s FXS db for: id = %s, source = %s",
2153                                 GetSystemName(system), id, sourceName.Data()));
2154                 fFXSError = system;
2155                 delete aResult;
2156                 return 0;
2157         }
2158
2159         TSQLRow* aRow = dynamic_cast<TSQLRow*> (aResult->Next());
2160
2161         if (!aRow){
2162                 Log(detector, Form("GetFile - Empty set result in %s FXS db from query: id = %s, source = %s",
2163                                 GetSystemName(system), id, sourceName.Data()));
2164                 fFXSError = system;
2165                 delete aResult;
2166                 return 0;
2167         }
2168
2169         TString filePath(aRow->GetField(0), aRow->GetFieldLength(0));
2170         TString fileSize(aRow->GetField(1), aRow->GetFieldLength(1));
2171         TString fileChecksum(aRow->GetField(2), aRow->GetFieldLength(2));
2172
2173         delete aResult;
2174         delete aRow;
2175
2176         AliDebug(2, Form("filePath = %s; size = %s, fileChecksum = %s",
2177                                 filePath.Data(), fileSize.Data(), fileChecksum.Data()));
2178
2179         // retrieved file is renamed to make it unique
2180         Int_t dir_lev1 = GetCurrentRun()/10000;
2181         TString localFileName = Form("%s/%d/%d/%s_process/%s_%s_%d_%s_%s.shuttle",
2182                                      GetShuttleTempDir(), dir_lev1, GetCurrentRun(), detector,
2183                                         GetSystemName(system), detector, GetCurrentRun(),
2184                                         id, sourceName.Data());
2185         Log("SHUTTLE",Form("file from FXS = %s",localFileName.Data()));
2186
2187
2188         // file retrieval from FXS
2189         UInt_t nRetries = 0;
2190         UInt_t maxRetries = 3;
2191         Bool_t result = kFALSE;
2192
2193         // copy!! if successful TSystem::Exec returns 0
2194         while (nRetries++ < maxRetries) {
2195                 AliDebug(2, Form("Trying to copy file. Retry # %d", nRetries));
2196                 result = RetrieveFile(system, filePath.Data(), localFileName.Data());
2197                 if (!result)
2198                 {
2199                         Log(detector, Form("GetFile - Copy of file %s from %s FXS failed",
2200                                         filePath.Data(), GetSystemName(system)));
2201                         continue;
2202                 }
2203
2204                 if (fileSize.Length()>0)
2205                 {
2206                         // compare filesize of local file with the one stored in the FXS DB
2207                         Long_t size = -1;
2208                         Int_t sizeComp = gSystem->GetPathInfo(localFileName.Data(), 0, &size, 0, 0);
2209
2210                         if (sizeComp != 0 || size != fileSize.Atoi())
2211                         {
2212                                 Log(detector, Form("GetFile - size of file %s does not match with local copy!",
2213                                                         filePath.Data()));
2214                                 result = kFALSE;
2215                                 continue;
2216                         }
2217
2218                 } else {
2219                         Log(fCurrentDetector, Form("GetFile - size of file %s not set in %s database, skipping comparison",
2220                                                 filePath.Data(), GetSystemName(system)));
2221                 }
2222
2223                 if (fileChecksum.Length()>0)
2224                 {
2225                         // compare md5sum of local file with the one stored in the FXS DB
2226                         if(fileChecksum.Contains(' ')) fileChecksum.Resize(fileChecksum.First(' '));
2227                         Int_t md5Comp = gSystem->Exec(Form("md5sum %s |grep %s > /dev/null 2> /dev/null",
2228                                                 localFileName.Data(), fileChecksum.Data()));
2229
2230                         if (md5Comp != 0)
2231                         {
2232                                 Log(detector, Form("GetFile - md5sum of file %s does not match with local copy!",
2233                                                         filePath.Data()));
2234                                 result = kFALSE;
2235                                 continue;
2236                         }
2237                 } else {
2238                         Log(fCurrentDetector, Form("GetFile - md5sum of file %s not set in %s database, skipping comparison",
2239                                                         filePath.Data(), GetSystemName(system)));
2240                 }
2241                 if (result) break;
2242         }
2243
2244         if (!result)
2245         {
2246                 fFXSError = system;
2247                 return 0;
2248         }
2249
2250         fFXSCalled[system]=kTRUE;
2251         TObjString *fileParams = new TObjString(Form("%s#!?!#%s", id, sourceName.Data()));
2252         fFXSlist[system].Add(fileParams);
2253
2254         static TString staticLocalFileName;
2255         staticLocalFileName.Form("%s", localFileName.Data());
2256
2257         Log(fCurrentDetector, Form("GetFile - Retrieved file with id %s and "
2258                         "source %s from %s to %s", id, source,
2259                         GetSystemName(system), localFileName.Data()));
2260
2261         return staticLocalFileName.Data();
2262 }
2263
2264 //______________________________________________________________________________________________
2265 Bool_t AliShuttle::RetrieveFile(UInt_t system, const char* fxsFileName, const char* localFileName)
2266 {
2267         //
2268         // Copies file from FXS to local Shuttle machine
2269         //
2270
2271         // check temp directory: trying to cd to temp; if it does not exist, create it
2272         AliDebug(2, Form("Copy file %s from %s FXS into %s",
2273                         GetSystemName(system), fxsFileName, localFileName));
2274
2275         TString tmpDir(localFileName);
2276
2277         tmpDir = tmpDir(0,tmpDir.Last('/'));
2278
2279         Int_t noDir = gSystem->GetPathInfo(tmpDir.Data(), 0, (Long64_t*) 0, 0, 0);
2280         if (noDir) // temp dir does not exists!
2281         {
2282                 if (gSystem->mkdir(tmpDir.Data(), 1))
2283                 {
2284                         Log(fCurrentDetector.Data(), "RetrieveFile - could not make temp directory!!");
2285                         return kFALSE;
2286                 }
2287         }
2288
2289         TString command = Form("scp -oPort=%d -2 %s@%s:%s/%s %s",
2290                 fConfig->GetFXSPort(system),
2291                 fConfig->GetFXSUser(system),
2292                 fConfig->GetFXSHost(system),
2293                 fConfig->GetFXSBaseFolder(system),
2294                 fxsFileName,
2295                 localFileName);
2296
2297         AliDebug(2, Form("%s",command.Data()));
2298
2299         Bool_t result = (gSystem->Exec(command.Data()) == 0);
2300
2301         return result;
2302 }
2303
2304 //______________________________________________________________________________________________
2305 TList* AliShuttle::GetFileSources(Int_t system, const char* detector, const char* id)
2306 {
2307         //
2308         // Get sources producing the condition file Id from file exchange servers
2309         // if id is NULL all sources are returned (distinct)
2310         //
2311
2312         if (id)
2313         {
2314                 Log(detector, Form("GetFileSources - Querying %s FXS for files with id %s produced by %s", GetSystemName(system), id, detector));
2315         } else {
2316                 Log(detector, Form("GetFileSources - Querying %s FXS for files produced by %s", GetSystemName(system), detector));
2317         }
2318
2319         // check if test mode should simulate a FXS error
2320         if (fTestMode & kErrorFXSSources)
2321         {
2322                 Log(detector, Form("GetFileSources - In TESTMODE - Simulating error while connecting to %s FXS", GetSystemName(system)));
2323                 return 0;
2324         }
2325
2326         if (system == kDCS)
2327         {
2328                 Log(detector, "GetFileSources - WARNING: DCS system has only one source of data!");
2329                 TList *list = new TList();
2330                 list->SetOwner(1);
2331                 list->Add(new TObjString(" "));
2332                 return list;
2333         }
2334
2335         // check connection, in case connect
2336         if (!Connect(system))
2337         {
2338                 Log(detector, Form("GetFileSources - Couldn't connect to %s FXS database", GetSystemName(system)));
2339                 fFXSError = system;
2340                 return NULL;
2341         }
2342
2343         TString sourceName = 0;
2344         if (system == kDAQ)
2345         {
2346                 sourceName = "DAQsource";
2347         } else if (system == kHLT)
2348         {
2349                 sourceName = "DDLnumbers";
2350         }
2351
2352         TString sqlQueryStart = Form("select distinct %s from %s where", sourceName.Data(), fConfig->GetFXSdbTable(system));
2353         TString whereClause = Form("run=%d and detector=\"%s\"",
2354                                 GetCurrentRun(), detector);
2355         if (id)
2356                 whereClause += Form(" and fileId=\"%s\"", id);
2357         TString sqlQuery = Form("%s %s", sqlQueryStart.Data(), whereClause.Data());
2358
2359         AliDebug(2, Form("SQL query: \n%s",sqlQuery.Data()));
2360
2361         // Query execution
2362         TSQLResult* aResult;
2363         aResult = fServer[system]->Query(sqlQuery);
2364         if (!aResult) {
2365                 Log(detector, Form("GetFileSources - Can't execute SQL query to %s database for id: %s",
2366                                 GetSystemName(system), id));
2367                 fFXSError = system;
2368                 return 0;
2369         }
2370
2371         TList *list = new TList();
2372         list->SetOwner(1);
2373
2374         if (aResult->GetRowCount() == 0)
2375         {
2376                 Log(detector,
2377                         Form("GetFileSources - No entry in %s FXS table for id: %s", GetSystemName(system), id));
2378                 delete aResult;
2379                 return list;
2380         }
2381
2382         Log(detector, Form("GetFileSources - Found %d sources", aResult->GetRowCount()));
2383
2384         TSQLRow* aRow;
2385         while ((aRow = aResult->Next()))
2386         {
2387
2388                 TString source(aRow->GetField(0), aRow->GetFieldLength(0));
2389                 AliDebug(2, Form("%s = %s", sourceName.Data(), source.Data()));
2390                 list->Add(new TObjString(source));
2391                 delete aRow;
2392         }
2393
2394         delete aResult;
2395
2396         return list;
2397 }
2398
2399 //______________________________________________________________________________________________
2400 TList* AliShuttle::GetFileIDs(Int_t system, const char* detector, const char* source)
2401 {
2402         //
2403         // Get all ids of condition files produced by a given source from file exchange servers
2404         //
2405
2406         Log(detector, Form("GetFileIDs - Retrieving ids with source %s with %s", source, GetSystemName(system)));
2407
2408         // check if test mode should simulate a FXS error
2409         if (fTestMode & kErrorFXSSources)
2410         {
2411                 Log(detector, Form("GetFileIDs - In TESTMODE - Simulating error while connecting to %s FXS", GetSystemName(system)));
2412                 return 0;
2413         }
2414
2415         // check connection, in case connect
2416         if (!Connect(system))
2417         {
2418                 Log(detector, Form("GetFileIDs - Couldn't connect to %s FXS database", GetSystemName(system)));
2419                 return NULL;
2420         }
2421
2422         TString sourceName = 0;
2423         if (system == kDAQ)
2424         {
2425                 sourceName = "DAQsource";
2426         } else if (system == kHLT)
2427         {
2428                 sourceName = "DDLnumbers";
2429         }
2430
2431         TString sqlQueryStart = Form("select fileId from %s where", fConfig->GetFXSdbTable(system));
2432         TString whereClause = Form("run=%d and detector=\"%s\"",
2433                                 GetCurrentRun(), detector);
2434         if (sourceName.Length() > 0 && source)
2435                 whereClause += Form(" and %s=\"%s\"", sourceName.Data(), source);
2436         TString sqlQuery = Form("%s %s", sqlQueryStart.Data(), whereClause.Data());
2437
2438         AliDebug(2, Form("SQL query: \n%s",sqlQuery.Data()));
2439
2440         // Query execution
2441         TSQLResult* aResult;
2442         aResult = fServer[system]->Query(sqlQuery);
2443         if (!aResult) {
2444                 Log(detector, Form("GetFileIDs - Can't execute SQL query to %s database for source: %s",
2445                                 GetSystemName(system), source));
2446                 return 0;
2447         }
2448
2449         TList *list = new TList();
2450         list->SetOwner(1);
2451
2452         if (aResult->GetRowCount() == 0)
2453         {
2454                 Log(detector,
2455                         Form("GetFileIDs - No entry in %s FXS table for source: %s", GetSystemName(system), source));
2456                 delete aResult;
2457                 return list;
2458         }
2459
2460         Log(detector, Form("GetFileIDs - Found %d ids", aResult->GetRowCount()));
2461
2462         TSQLRow* aRow;
2463
2464         while ((aRow = aResult->Next()))
2465         {
2466
2467                 TString id(aRow->GetField(0), aRow->GetFieldLength(0));
2468                 AliDebug(2, Form("fileId = %s", id.Data()));
2469                 list->Add(new TObjString(id));
2470                 delete aRow;
2471         }
2472
2473         delete aResult;
2474
2475         return list;
2476 }
2477
2478 //______________________________________________________________________________________________
2479 Bool_t AliShuttle::Connect(Int_t system)
2480 {
2481         // Connect to MySQL Server of the system's FXS MySQL databases
2482         // DAQ Logbook, Shuttle Logbook and DAQ FXS db are on the same host
2483         //
2484
2485         // check connection: if already connected return
2486         if(fServer[system] && fServer[system]->IsConnected()) {
2487                 // ping the server --> automatic reconnection should occur if it was broken but the
2488                 // server is still alive
2489                 fServer[system]->Ping();
2490                 return kTRUE;
2491         }
2492
2493         TString dbHost, dbUser, dbPass, dbName;
2494
2495         if (system < 3) // FXS db servers
2496         {
2497                 dbHost = Form("mysql://%s:%d", fConfig->GetFXSdbHost(system), fConfig->GetFXSdbPort(system));
2498                 dbUser = fConfig->GetFXSdbUser(system);
2499                 dbPass = fConfig->GetFXSdbPass(system);
2500                 dbName =   fConfig->GetFXSdbName(system);
2501         } else { // Run & Shuttle logbook servers
2502         // TODO Will the Shuttle logbook server be the same as the Run logbook server ???
2503                 dbHost = Form("mysql://%s:%d", fConfig->GetDAQlbHost(), fConfig->GetDAQlbPort());
2504                 dbUser = fConfig->GetDAQlbUser();
2505                 dbPass = fConfig->GetDAQlbPass();
2506                 dbName =   fConfig->GetDAQlbDB();
2507         }
2508
2509         fServer[system] = TSQLServer::Connect(dbHost.Data(), dbUser.Data(), dbPass.Data());
2510         if (!fServer[system] || !fServer[system]->IsConnected()) {
2511                 if(system < 3)
2512                 {
2513                 AliError(Form("Can't establish connection to FXS database for %s",
2514                                         AliShuttleInterface::GetSystemName(system)));
2515                 } else {
2516                 AliError("Can't establish connection to Run logbook.");
2517                 }
2518                 if(fServer[system]) delete fServer[system];
2519                 return kFALSE;
2520         }
2521
2522         // Get tables
2523         TSQLResult* aResult=0;
2524         switch(system){
2525                 case kDAQ:
2526                         aResult = fServer[kDAQ]->GetTables(dbName.Data());
2527                         break;
2528                 case kDCS:
2529                         aResult = fServer[kDCS]->GetTables(dbName.Data());
2530                         break;
2531                 case kHLT:
2532                         aResult = fServer[kHLT]->GetTables(dbName.Data());
2533                         break;
2534                 default:
2535                         aResult = fServer[3]->GetTables(dbName.Data());
2536                         break;
2537         }
2538
2539         delete aResult;
2540         return kTRUE;
2541 }
2542
2543 //______________________________________________________________________________________________
2544 Bool_t AliShuttle::UpdateTable()
2545 {
2546         //
2547         // Update FXS table filling time_processed field in all rows corresponding to current run and detector
2548         //
2549
2550         Bool_t result = kTRUE;
2551
2552         for (UInt_t system=0; system<3; system++)
2553         {
2554                 if(!fFXSCalled[system]) continue;
2555
2556                 // check connection, in case connect
2557                 if (!Connect(system))
2558                 {
2559                         Log(fCurrentDetector, Form("UpdateTable - Couldn't connect to %s FXS database", GetSystemName(system)));
2560                         result = kFALSE;
2561                         continue;
2562                 }
2563
2564                 TTimeStamp now; // now
2565
2566                 // Loop on FXS list entries
2567                 TIter iter(&fFXSlist[system]);
2568                 TObjString *aFXSentry=0;
2569                 while ((aFXSentry = dynamic_cast<TObjString*> (iter.Next())))
2570                 {
2571                         TString aFXSentrystr = aFXSentry->String();
2572                         TObjArray *aFXSarray = aFXSentrystr.Tokenize("#!?!#");
2573                         if (!aFXSarray || aFXSarray->GetEntries() != 2 )
2574                         {
2575                                 Log(fCurrentDetector, Form("UpdateTable - error updating %s FXS entry. Check string: <%s>",
2576                                         GetSystemName(system), aFXSentrystr.Data()));
2577                                 if(aFXSarray) delete aFXSarray;
2578                                 result = kFALSE;
2579                                 continue;
2580                         }
2581                         const char* fileId = ((TObjString*) aFXSarray->At(0))->GetName();
2582                         const char* source = ((TObjString*) aFXSarray->At(1))->GetName();
2583
2584                         TString whereClause;
2585                         if (system == kDAQ)
2586                         {
2587                                 whereClause = Form("where run=%d and detector=\"%s\" and fileId=\"%s\" and DAQsource=\"%s\";",
2588                                                         GetCurrentRun(), fCurrentDetector.Data(), fileId, source);
2589                         }
2590                         else if (system == kDCS)
2591                         {
2592                                 whereClause = Form("where run=%d and detector=\"%s\" and fileId=\"%s\";",
2593                                                         GetCurrentRun(), fCurrentDetector.Data(), fileId);
2594                         }
2595                         else if (system == kHLT)
2596                         {
2597                                 whereClause = Form("where run=%d and detector=\"%s\" and fileId=\"%s\" and DDLnumbers=\"%s\";",
2598                                                         GetCurrentRun(), fCurrentDetector.Data(), fileId, source);
2599                         }
2600
2601                         delete aFXSarray;
2602
2603                         TString sqlQuery = Form("update %s set time_processed=%d %s", fConfig->GetFXSdbTable(system),
2604                                                                 now.GetSec(), whereClause.Data());
2605
2606                         AliDebug(2, Form("SQL query: \n%s",sqlQuery.Data()));
2607
2608                         // Query execution
2609                         TSQLResult* aResult;
2610                         aResult = dynamic_cast<TSQLResult*> (fServer[system]->Query(sqlQuery));
2611                         if (!aResult)
2612                         {
2613                                 Log(fCurrentDetector, Form("UpdateTable - %s db: can't execute SQL query <%s>",
2614                                                                 GetSystemName(system), sqlQuery.Data()));
2615                                 result = kFALSE;
2616                                 continue;
2617                         }
2618                         delete aResult;
2619                 }
2620         }
2621
2622         return result;
2623 }
2624
2625 //_______________________________________________________________________________
2626 Bool_t AliShuttle::UpdateTableSkippedCase(const char* detector)
2627 {
2628         //
2629         // Update FXS table filling time_processed field in all rows corresponding to current run and detector
2630         // if detector = "ALL" update all detectors
2631         //
2632
2633         Bool_t result = kTRUE;
2634
2635         TString detName(detector);
2636
2637         for (UInt_t system=0; system<3; system++)
2638         {
2639
2640                 // check connection, in case connect
2641                 if (!Connect(system))
2642                 {
2643                         Log(fCurrentDetector, Form("UpdateTableSkippedCase - Couldn't connect to %s FXS database", GetSystemName(system)));
2644                         result = kFALSE;
2645                         continue;
2646                 }
2647
2648                 TTimeStamp now; // now
2649
2650                 // Loop on FXS list entries
2651                 TIter iter(&fFXSlist[system]);
2652
2653                 TString whereClause;
2654                 if (detName == "ALL") whereClause = Form("where run=%d and time_processed IS NULL;",GetCurrentRun());
2655                 else whereClause = Form("where run=%d and detector=\"%s\" and time_processed IS NULL;",GetCurrentRun(), detector);
2656
2657                 Log("SHUTTLE",Form(" whereClause = %s ",whereClause.Data()));
2658
2659                 TString sqlQuery = Form("update %s set time_processed=%d %s", fConfig->GetFXSdbTable(system),
2660                                         now.GetSec(), whereClause.Data());
2661
2662                 AliDebug(2, Form("SQL query: \n%s",sqlQuery.Data()));
2663
2664                 // Query execution
2665                 TSQLResult* aResult;
2666                 aResult = dynamic_cast<TSQLResult*> (fServer[system]->Query(sqlQuery));
2667                 if (!aResult)
2668                 {
2669                         Log("SHUTTLE", Form("UpdateTableSkippedCase - %s db: can't execute SQL query <%s>",
2670                                                         GetSystemName(system), sqlQuery.Data()));
2671                         result = kFALSE;
2672                         continue;
2673                 }
2674                 delete aResult;
2675
2676         }
2677
2678         return result;
2679 }
2680 //______________________________________________________________________________________________
2681 Bool_t AliShuttle::UpdateTableFailCase()
2682 {
2683         // Update FXS table filling time_processed field in all rows corresponding to current run and detector
2684         // this is called in case the preprocessor is declared failed for the current run, because
2685         // the fields are updated only in case of success
2686
2687         Bool_t result = kTRUE;
2688
2689         for (UInt_t system=0; system<3; system++)
2690         {
2691                 // check connection, in case connect
2692                 if (!Connect(system))
2693                 {
2694                         Log(fCurrentDetector, Form("UpdateTableFailCase - Couldn't connect to %s FXS database",
2695                                                         GetSystemName(system)));
2696                         result = kFALSE;
2697                         continue;
2698                 }
2699
2700                 TTimeStamp now; // now
2701
2702                 // Loop on FXS list entries
2703
2704                 TString whereClause = Form("where run=%d and detector=\"%s\";",
2705                                                 GetCurrentRun(), fCurrentDetector.Data());
2706
2707
2708                 TString sqlQuery = Form("update %s set time_processed=%d %s", fConfig->GetFXSdbTable(system),
2709                                                         now.GetSec(), whereClause.Data());
2710
2711                 AliDebug(2, Form("SQL query: \n%s",sqlQuery.Data()));
2712
2713                 // Query execution
2714                 TSQLResult* aResult;
2715                 aResult = dynamic_cast<TSQLResult*> (fServer[system]->Query(sqlQuery));
2716                 if (!aResult)
2717                 {
2718                         Log(fCurrentDetector, Form("UpdateTableFailCase - %s db: can't execute SQL query <%s>",
2719                                                         GetSystemName(system), sqlQuery.Data()));
2720                         result = kFALSE;
2721                         continue;
2722                 }
2723                 delete aResult;
2724         }
2725
2726         return result;
2727 }
2728
2729 //______________________________________________________________________________________________
2730 Bool_t AliShuttle::UpdateShuttleLogbook(const char* detector, const char* status)
2731 {
2732         //
2733         // Update Shuttle logbook filling detector or shuttle_done column
2734         // ex. of usage: UpdateShuttleLogbook("PHOS", "DONE") or UpdateShuttleLogbook("shuttle_done")
2735         //
2736
2737         // check connection, in case connect
2738         if(!Connect(3)){
2739                 Log("SHUTTLE", "UpdateShuttleLogbook - Couldn't connect to DAQ Logbook.");
2740                 return kFALSE;
2741         }
2742
2743         TString detName(detector);
2744         TString setClause;
2745         if (detName == "shuttle_done" || detName == "shuttle_skipped")
2746         {
2747                 setClause = "set shuttle_done=1";
2748
2749                 if (detName == "shuttle_done")
2750                 {
2751                         if (TouchFile() != kTRUE)
2752                         {
2753                                 SendMLRunInfo("Pending");
2754                                 return kFALSE;
2755                         }
2756
2757                         SendMLRunInfo("Done");
2758                 }
2759                 else
2760                         SendMLRunInfo("Skipped");
2761         }
2762         else {
2763                 TString statusStr(status);
2764                 if(statusStr.Contains("done", TString::kIgnoreCase) ||
2765                    statusStr.Contains("failed", TString::kIgnoreCase)){
2766                         setClause = Form("set %s=\"%s\"", detector, status);
2767                 } else {
2768                         Log("SHUTTLE",
2769                                 Form("UpdateShuttleLogbook - Invalid status <%s> for detector %s",
2770                                         status, detector));
2771                         return kFALSE;
2772                 }
2773         }
2774
2775         TString whereClause = Form("where run=%d", GetCurrentRun());
2776
2777         TString sqlQuery = Form("update %s %s %s",
2778                                         fConfig->GetShuttlelbTable(), setClause.Data(), whereClause.Data());
2779
2780         AliDebug(2, Form("SQL query: \n%s",sqlQuery.Data()));
2781
2782         // Query execution
2783         TSQLResult* aResult;
2784         aResult = dynamic_cast<TSQLResult*> (fServer[3]->Query(sqlQuery));
2785         if (!aResult) {
2786                 Log("SHUTTLE", Form("UpdateShuttleLogbook - Can't execute query <%s>", sqlQuery.Data()));
2787                 return kFALSE;
2788         }
2789         delete aResult;
2790
2791         return kTRUE;
2792 }
2793
2794 //______________________________________________________________________________________________
2795 Int_t AliShuttle::GetCurrentRun() const
2796 {
2797         //
2798         // Get current run from logbook entry
2799         //
2800
2801         return fLogbookEntry ? fLogbookEntry->GetRun() : -1;
2802 }
2803
2804 //______________________________________________________________________________________________
2805 UInt_t AliShuttle::GetCurrentStartTime() const
2806 {
2807         //
2808         // get current start time
2809         //
2810
2811         return fLogbookEntry ? fLogbookEntry->GetStartTime() : 0;
2812 }
2813
2814 //______________________________________________________________________________________________
2815 UInt_t AliShuttle::GetCurrentEndTime() const
2816 {
2817         //
2818         // get current end time from logbook entry
2819         //
2820
2821         return fLogbookEntry ? fLogbookEntry->GetEndTime() : 0;
2822 }
2823 //______________________________________________________________________________________________
2824 UInt_t AliShuttle::GetCurrentTimeCreated() const
2825 {
2826         //
2827         // get current end time from logbook entry
2828         //
2829
2830         return fLogbookEntry ? fLogbookEntry->GetTimeCreated() : 0;
2831 }
2832
2833 //______________________________________________________________________________________________
2834 UInt_t AliShuttle::GetCurrentYear() const
2835 {
2836         //
2837         // Get current year from logbook entry
2838         //
2839
2840         if (!fLogbookEntry) return 0;
2841
2842         TTimeStamp startTime(GetCurrentStartTime());
2843         TString year =  Form("%d",startTime.GetDate());
2844         year = year(0,4);
2845
2846         return year.Atoi();
2847 }
2848
2849 //______________________________________________________________________________________________
2850 const char* AliShuttle::GetLHCPeriod() const
2851 {
2852         //
2853         // Get current LHC period from logbook entry
2854         //
2855
2856         if (!fLogbookEntry) return 0;
2857
2858         return fLogbookEntry->GetRunParameter("LHCperiod");
2859 }
2860
2861 //______________________________________________________________________________________________
2862 void AliShuttle::Log(const char* detector, const char* message)
2863 {
2864         //
2865         // Fill log string with a message
2866         //
2867
2868         TString logRunDir = GetShuttleLogDir();
2869         if (GetCurrentRun() >=0) {
2870                 Int_t logDir_lev1 = GetCurrentRun()/10000;
2871                 logRunDir += Form("/%d/%d", logDir_lev1, GetCurrentRun());
2872         }
2873         void* dir = gSystem->OpenDirectory(logRunDir.Data());
2874         if (dir == NULL) {
2875                 if (gSystem->mkdir(logRunDir.Data(), kTRUE)) {
2876                         AliError(Form("Can't open directory <%s>", GetShuttleLogDir()));
2877                         return;
2878                 }
2879
2880         } else {
2881                 gSystem->FreeDirectory(dir);
2882         }
2883
2884         TString toLog = Form("%s UTC (%d): %s - ", TTimeStamp(time(0)).AsString("s"), getpid(), detector);
2885         if (GetCurrentRun() >= 0)
2886                 toLog += Form("run %d - ", GetCurrentRun());
2887         toLog += Form("%s", message);
2888
2889         AliInfo(toLog.Data());
2890
2891         // if we redirect the log output already to the file, leave here
2892         if (fOutputRedirected && strcmp(detector, "SHUTTLE") != 0)
2893                 return;
2894
2895         TString fileName = GetLogFileName(detector);
2896
2897         gSystem->ExpandPathName(fileName);
2898
2899         ofstream logFile;
2900         logFile.open(fileName, ofstream::out | ofstream::app);
2901
2902         if (!logFile.is_open()) {
2903                 AliError(Form("Could not open file %s", fileName.Data()));
2904                 return;
2905         }
2906
2907         logFile << toLog.Data() << "\n";
2908
2909         logFile.close();
2910 }
2911
2912 //______________________________________________________________________________________________
2913 TString AliShuttle::GetLogFileName(const char* detector) const
2914 {
2915         //
2916         // returns the name of the log file for a given sub detector
2917         //
2918
2919         TString fileName;
2920
2921         if (GetCurrentRun() >= 0)
2922         {
2923                 Int_t logDir_lev1 = GetCurrentRun()/10000;
2924                 fileName.Form("%s/%d/%d/%s.log", GetShuttleLogDir(), logDir_lev1, GetCurrentRun(),
2925                         detector);
2926         } else {
2927                 fileName.Form("%s/%s.log", GetShuttleLogDir(), detector);
2928         }
2929
2930         return fileName;
2931 }
2932
2933 //______________________________________________________________________________________________
2934 void AliShuttle::SendAlive()
2935 {
2936         // sends alive message to ML
2937
2938         TMonaLisaText mlStatus("SHUTTLE_status", "Alive");
2939
2940         TList mlList;
2941         mlList.Add(&mlStatus);
2942
2943         fMonaLisa->SendParameters(&mlList, "__PROCESSINGINFO__");
2944 }
2945
2946 //______________________________________________________________________________________________
2947 Bool_t AliShuttle::Collect(Int_t run)
2948 {
2949         //
2950         // Collects conditions data for all UNPROCESSED run written to DAQ LogBook in case of run = -1 (default)
2951         // If a dedicated run is given this run is processed
2952         //
2953         // In operational mode, this is the Shuttle function triggered by the EOR signal.
2954         //
2955
2956         if (run == -1)
2957                 Log("SHUTTLE","Collect - Shuttle called. Collecting conditions data for unprocessed runs");
2958         else
2959                 Log("SHUTTLE", Form("Collect - Shuttle called. Collecting conditions data for run %d", run));
2960
2961         SetLastAction("Starting");
2962
2963         // create ML instance
2964         if (!fMonaLisa)
2965                 fMonaLisa = new TMonaLisaWriter(fConfig->GetMonitorHost(), fConfig->GetMonitorTable());
2966
2967         CountOpenRuns();
2968
2969         TString whereClause("where shuttle_done=0");
2970         if (run != -1)
2971                 whereClause += Form(" and run=%d", run);
2972
2973         TObjArray shuttleLogbookEntries;
2974         if (!QueryShuttleLogbook(whereClause, shuttleLogbookEntries))
2975         {
2976                 Log("SHUTTLE", "Collect - Can't retrieve entries from Shuttle logbook");
2977                 return kFALSE;
2978         }
2979
2980         if (shuttleLogbookEntries.GetEntries() == 0)
2981         {
2982                 if (run == -1)
2983                         Log("SHUTTLE","Collect - Found no UNPROCESSED runs in Shuttle logbook");
2984                 else
2985                         Log("SHUTTLE", Form("Collect - Run %d is already DONE "
2986                                                 "or it does not exist in Shuttle logbook", run));
2987                 return kTRUE;
2988         }
2989
2990         for (UInt_t iDet=0; iDet<NDetectors(); iDet++)
2991                 fFirstUnprocessed[iDet] = kTRUE;
2992
2993         if (run != -1)
2994         {
2995                 // query Shuttle logbook for earlier runs, check if some detectors are unprocessed,
2996                 // flag them into fFirstUnprocessed array
2997                 TString whereClauseBis(Form("where shuttle_done=0 and run < %d", run));
2998                 TObjArray tmpLogbookEntries;
2999                 if (!QueryShuttleLogbook(whereClauseBis, tmpLogbookEntries))
3000                 {
3001                         Log("SHUTTLE", "Collect - Can't retrieve entries from Shuttle logbook");
3002                         return kFALSE;
3003                 }
3004
3005                 TIter iter(&tmpLogbookEntries);
3006                 AliShuttleLogbookEntry* anEntry = 0;
3007                 while ((anEntry = dynamic_cast<AliShuttleLogbookEntry*> (iter.Next())))
3008                 {
3009                         for (UInt_t iDet=0; iDet<NDetectors(); iDet++)
3010                         {
3011                                 if (anEntry->GetDetectorStatus(iDet) == AliShuttleLogbookEntry::kUnprocessed)
3012                                 {
3013                                         AliDebug(2, Form("Run %d: setting %s as \"not first time unprocessed\"",
3014                                                         anEntry->GetRun(), GetDetName(iDet)));
3015                                         fFirstUnprocessed[iDet] = kFALSE;
3016                                 }
3017                         }
3018
3019                 }
3020
3021         }
3022
3023         if (!RetrieveConditionsData(shuttleLogbookEntries))
3024         {
3025                 Log("SHUTTLE", "Collect - Process of at least one run failed");
3026                 CountOpenRuns();
3027                 return kFALSE;
3028         }
3029
3030         Log("SHUTTLE", "Collect - Requested run(s) successfully processed");
3031         CountOpenRuns();
3032         return kTRUE;
3033 }
3034
3035 //______________________________________________________________________________________________
3036 Bool_t AliShuttle::RetrieveConditionsData(const TObjArray& dateEntries)
3037 {
3038         //
3039         // Retrieve conditions data for all runs that aren't processed yet
3040         //
3041
3042         Bool_t hasError = kFALSE;
3043
3044         TIter iter(&dateEntries);
3045         AliShuttleLogbookEntry* anEntry;
3046
3047         while ((anEntry = (AliShuttleLogbookEntry*) iter.Next())){
3048                 if (!Process(anEntry)){
3049                         hasError = kTRUE;
3050                 }
3051
3052                 // clean SHUTTLE temp directory
3053                 //TString filename = Form("%s/*.shuttle", GetShuttleTempDir());
3054                 //RemoveFile(filename.Data());
3055         }
3056
3057         return hasError == kFALSE;
3058 }
3059
3060 //______________________________________________________________________________________________
3061 ULong_t AliShuttle::GetTimeOfLastAction() const
3062 {
3063         //
3064         // Gets time of last action
3065         //
3066
3067         ULong_t tmp;
3068
3069         fMonitoringMutex->Lock();
3070
3071         tmp = fLastActionTime;
3072
3073         fMonitoringMutex->UnLock();
3074
3075         return tmp;
3076 }
3077
3078 //______________________________________________________________________________________________
3079 const TString AliShuttle::GetLastAction() const
3080 {
3081         //
3082         // returns a string description of the last action
3083         //
3084
3085         TString tmp;
3086
3087         fMonitoringMutex->Lock();
3088
3089         tmp = fLastAction;
3090
3091         fMonitoringMutex->UnLock();
3092
3093         return tmp;
3094 }
3095
3096 //______________________________________________________________________________________________
3097 void AliShuttle::SetLastAction(const char* action)
3098 {
3099         //
3100         // updates the monitoring variables
3101         //
3102
3103         fMonitoringMutex->Lock();
3104
3105         fLastAction = action;
3106         fLastActionTime = time(0);
3107
3108         fMonitoringMutex->UnLock();
3109 }
3110
3111 //______________________________________________________________________________________________
3112 const char* AliShuttle::GetRunParameter(const char* param)
3113 {
3114         //
3115         // returns run parameter read from DAQ logbook
3116         //
3117
3118         if(!fLogbookEntry) {
3119                 AliError("No logbook entry!");
3120                 return 0;
3121         }
3122
3123         return fLogbookEntry->GetRunParameter(param);
3124 }
3125
3126 //______________________________________________________________________________________________
3127 AliCDBEntry* AliShuttle::GetFromOCDB(const char* detector, const AliCDBPath& path)
3128 {
3129         //
3130         // returns object from OCDB valid for current run
3131         //
3132
3133         if (fTestMode & kErrorOCDB)
3134         {
3135                 Log(detector, "GetFromOCDB - In TESTMODE - Simulating error with OCDB");
3136                 return 0;
3137         }
3138
3139         AliCDBStorage *sto = AliCDBManager::Instance()->GetStorage(fgkMainCDB);
3140         if (!sto)
3141         {
3142                 Log(detector, "GetFromOCDB - Cannot activate main OCDB for query!");
3143                 return 0;
3144         }
3145
3146         return dynamic_cast<AliCDBEntry*> (sto->Get(path, GetCurrentRun()));
3147 }
3148
3149 //______________________________________________________________________________________________
3150 Bool_t AliShuttle::SendMail(EMailTarget target, Int_t system)
3151 {
3152         //
3153         // sends a mail to the subdetector expert in case of preprocessor error
3154         //
3155
3156         if (fTestMode != kNone)
3157                 return kTRUE;
3158
3159         if (!fConfig->SendMail())
3160                 return kTRUE;
3161
3162         if (target == kDCSEMail || target == kFXSEMail) {
3163                 if (!fFirstProcessing)
3164                         return kTRUE;
3165         }
3166
3167         Int_t runMode = (Int_t)fConfig->GetRunMode();
3168         TString tmpStr;
3169         if (runMode == 0) tmpStr = " Nightly Test:";
3170         else tmpStr = " Data Taking:";
3171         void* dir = gSystem->OpenDirectory(GetShuttleLogDir());
3172         if (dir == NULL)
3173         {
3174                 if (gSystem->mkdir(GetShuttleLogDir(), kTRUE))
3175                 {
3176                         Log("SHUTTLE", Form("SendMail - Can't open directory <%s>", GetShuttleLogDir()));
3177                         return kFALSE;
3178                 }
3179
3180         } else {
3181                 gSystem->FreeDirectory(dir);
3182         }
3183
3184         // det experts in to
3185         TString to="";
3186         TIter *iterExperts = 0;
3187         if (target == kDCSEMail) {
3188                 iterExperts = new TIter(fConfig->GetAdmins(AliShuttleConfig::kAmanda));
3189         }
3190         else if (target == kFXSEMail) {
3191                 iterExperts = new TIter(fConfig->GetAdmins(system));
3192         }
3193         if (iterExperts) {
3194                 TObjString *anExpert=0;
3195                 while ((anExpert = (TObjString*) iterExperts->Next()))
3196                 {
3197                         to += Form("%s,", anExpert->GetName());
3198                 }
3199                 delete iterExperts;
3200         }
3201
3202         // add subdetector experts
3203         iterExperts = new TIter(fConfig->GetResponsibles(fCurrentDetector));
3204         TObjString *anExpert=0;
3205         while ((anExpert = (TObjString*) iterExperts->Next()))
3206         {
3207                 to += Form("%s,", anExpert->GetName());
3208         }
3209         delete iterExperts;
3210
3211         if (to.Length() > 0)
3212           to.Remove(to.Length()-1);
3213         AliDebug(2, Form("to: %s",to.Data()));
3214
3215         if (to.IsNull()) {
3216                 Log("SHUTTLE", Form("List of %d responsibles not set!", (Int_t) target));
3217                 return kFALSE;
3218         }
3219
3220         // SHUTTLE responsibles in cc
3221         TString cc="";
3222         TIter iterAdmins(fConfig->GetAdmins(AliShuttleConfig::kGlobal));
3223         TObjString *anAdmin=0;
3224         while ((anAdmin = (TObjString*) iterAdmins.Next()))
3225         {
3226                 cc += Form("%s,", anAdmin->GetName());
3227         }
3228         if (cc.Length() > 0)
3229           cc.Remove(cc.Length()-1);
3230         AliDebug(2, Form("cc: %s",to.Data()));
3231
3232         // mail body
3233         TString bodyFileName;
3234         bodyFileName.Form("%s/mail.body", GetShuttleLogDir());
3235         gSystem->ExpandPathName(bodyFileName);
3236
3237         ofstream mailBody;
3238         mailBody.open(bodyFileName, ofstream::out);
3239
3240         if (!mailBody.is_open())
3241         {
3242                 Log("SHUTTLE", Form("Could not open mail body file %s", bodyFileName.Data()));
3243                 return kFALSE;
3244         }
3245
3246
3247         TString subject;
3248         TString body;
3249
3250         if (target == kDCSEMail){
3251                 subject = Form("%s CRITICAL Retrieval of data points for %s FAILED in run %d !",
3252                                 tmpStr.Data(), fCurrentDetector.Data(), GetCurrentRun());
3253                 AliDebug(2, Form("subject: %s", subject.Data()));
3254
3255                 body = Form("Dear DCS experts, \n\n");
3256                 body += Form("SHUTTLE couldn\'t retrieve the data points for detector %s "
3257                              "in run %d!!\n\n", fCurrentDetector.Data(), GetCurrentRun());
3258         }
3259         else if (target == kFXSEMail){
3260                 subject = Form("%s CRITICAL FXS communication for %s FAILED in run %d !",
3261                                 tmpStr.Data(), fCurrentDetector.Data(), GetCurrentRun());
3262                 AliDebug(2, Form("subject: %s", subject.Data()));
3263                 TString sys;
3264                 if (system == kDAQ) sys="DAQ";
3265                 else if (system == kDCS) sys="DCS";
3266                 else if (system == kHLT) sys="HLT";
3267                 else return kFALSE;
3268                 body = Form("Dear  %s FXS experts, \n\n",sys.Data());
3269                 body += Form("SHUTTLE couldn\'t retrieve data from the FXS for detector %s "
3270                              "in run %d!!\n\n", fCurrentDetector.Data(), GetCurrentRun());
3271         }
3272         else {
3273                 subject = Form("%s %s Shuttle preprocessor FAILED in run %d (run type = %s)!",
3274                                        tmpStr.Data(), fCurrentDetector.Data(), GetCurrentRun(), GetRunType());
3275                 AliDebug(2, Form("subject: %s", subject.Data()));
3276
3277                 body = Form("Dear %s expert(s), \n\n", fCurrentDetector.Data());
3278                 body += Form("SHUTTLE just detected that your preprocessor "
3279                              "failed processing run %d (run type = %s)!!\n\n",
3280                              GetCurrentRun(), GetRunType());
3281         }
3282
3283         body += Form("Please check %s status on the SHUTTLE monitoring page: \n\n",
3284                                 fCurrentDetector.Data());
3285         if (fConfig->GetRunMode() == AliShuttleConfig::kTest)
3286         {
3287                 body += Form("\thttp://pcalimonitor.cern.ch/shuttle.jsp?time=24 \n\n");
3288         } else {
3289                 body += Form("\thttp://pcalimonitor.cern.ch/shuttle.jsp?instance=PROD&time=24 \n\n");
3290         }
3291
3292
3293         TString logFolder = "logs";
3294         if (fConfig->GetRunMode() == AliShuttleConfig::kProd)
3295                 logFolder += "_PROD";
3296
3297
3298         body += Form("Find the %s log for the current run on \n\n"
3299                 "\thttp://pcalishuttle01.cern.ch:8880/%s/%d/%d/%s.log \n\n",
3300                      fCurrentDetector.Data(), logFolder.Data(), GetCurrentRun()/10000,
3301                                 GetCurrentRun(), fCurrentDetector.Data());
3302         body += Form("The last 10 lines of %s log file are following:\n\n", fCurrentDetector.Data());
3303
3304         AliDebug(2, Form("Body begin: %s", body.Data()));
3305
3306         mailBody << body.Data();
3307         mailBody.close();
3308         mailBody.open(bodyFileName, ofstream::out | ofstream::app);
3309
3310         TString logFileName = Form("%s/%d/%d/%s.log", GetShuttleLogDir(),
3311                 GetCurrentRun()/10000, GetCurrentRun(), fCurrentDetector.Data());
3312         TString tailCommand = Form("tail -n 10 %s >> %s", logFileName.Data(), bodyFileName.Data());
3313         if (gSystem->Exec(tailCommand.Data()))
3314         {
3315                 mailBody << Form("%s log file not found ...\n\n", fCurrentDetector.Data());
3316         }
3317
3318         TString endBody = Form("------------------------------------------------------\n\n");
3319         endBody += Form("In case of problems please contact the SHUTTLE core team.\n\n");
3320         endBody += "Please do not answer this message directly, it is automatically generated.\n\n";
3321         endBody += "Greetings,\n\n \t\t\tthe SHUTTLE\n";
3322
3323         AliDebug(2, Form("Body end: %s", endBody.Data()));
3324
3325         mailBody << endBody.Data();
3326
3327         mailBody.close();
3328
3329         // send mail!
3330         TString mailCommand = Form("mail -s \"%s\" -c %s %s < %s",
3331                                                 subject.Data(),
3332                                                 cc.Data(),
3333                                                 to.Data(),
3334                                                 bodyFileName.Data());
3335         AliDebug(2, Form("mail command: %s", mailCommand.Data()));
3336
3337         Bool_t result = gSystem->Exec(mailCommand.Data());
3338
3339         return result == 0;
3340 }
3341 //______________________________________________________________________________________________
3342 const char* AliShuttle::GetRunType()
3343 {
3344         //
3345         // returns run type read from "run type" logbook
3346         //
3347
3348         if(!fLogbookEntry) {
3349                 AliError("No logbook entry!");
3350                 return 0;
3351         }
3352
3353         return fLogbookEntry->GetRunType();
3354 }
3355
3356 //______________________________________________________________________________________________
3357 Bool_t AliShuttle::GetHLTStatus()
3358 {
3359         // Return HLT status (ON=1 OFF=0)
3360         // Converts the HLT status from the mode string read in the run logbook (not just a bool)
3361
3362         if(!fLogbookEntry) {
3363                 AliError("No logbook entry!");
3364                 return 0;
3365         }
3366
3367         // TODO implement when HLTMode is inserted in run logbook
3368         TString hltMode = fLogbookEntry->GetRunParameter("HLTmode");
3369         TSubString firstChar = hltMode(0,1);
3370         AliDebug(2,Form("First char = %s ",firstChar.Data()));
3371         if (firstChar == "A") {
3372                 return kFALSE;
3373         }
3374         else if ((firstChar == "B") || (firstChar == "C") || (firstChar == "D") || (firstChar == "E")) {
3375                 return kTRUE;
3376         }
3377         else {
3378                 Log("SHUTTLE","Unexpected HLT mode! Returning 0....");
3379                 return kFALSE;
3380         }
3381 }
3382
3383 //______________________________________________________________________________________________
3384 const char* AliShuttle::GetTriggerConfiguration()
3385 {
3386         // Receives the trigger configuration from the DAQ logbook for the current run
3387
3388         // check connection, if needed reconnect
3389         if (!Connect(3))
3390                 return 0;
3391
3392         TString sqlQuery;
3393         sqlQuery.Form("SELECT configFile FROM logbook_trigger_config WHERE run = %d", GetCurrentRun());
3394         TSQLResult* result = fServer[3]->Query(sqlQuery);
3395         if (!result)
3396         {
3397                 Log("SHUTTLE", Form("ERROR: Can't execute query <%s>!", sqlQuery.Data()));
3398                 return 0;
3399         }
3400
3401         if (result->GetRowCount() == 0)
3402         {
3403                 Log("SHUTTLE", "ERROR: Trigger configuration not found in logbook_trigger_config");
3404                 delete result;
3405                 return 0;
3406         }
3407
3408         TSQLRow* row = result->Next();
3409         if (!row)
3410         {
3411                 Log("SHUTTLE", "ERROR: Could not receive logbook_trigger_config data");
3412                 delete result;
3413                 return 0;
3414         }
3415
3416         // static, so that pointer remains valid when it is returned to the calling class
3417         static TString triggerConfig(row->GetField(0));
3418
3419         delete row;
3420         row = 0;
3421
3422         delete result;
3423         result = 0;
3424
3425         Log("SHUTTLE", Form("Found trigger configuration: %s", triggerConfig.Data()));
3426
3427         return triggerConfig;
3428 }
3429
3430 //______________________________________________________________________________________________
3431 void AliShuttle::SetShuttleTempDir(const char* tmpDir)
3432 {
3433         //
3434         // sets Shuttle temp directory
3435         //
3436
3437         fgkShuttleTempDir = gSystem->ExpandPathName(tmpDir);
3438 }
3439
3440 //______________________________________________________________________________________________
3441 void AliShuttle::SetShuttleLogDir(const char* logDir)
3442 {
3443         //
3444         // sets Shuttle log directory
3445         //
3446
3447         fgkShuttleLogDir = gSystem->ExpandPathName(logDir);
3448 }
3449 //______________________________________________________________________________________________
3450 Bool_t AliShuttle::TouchFile()
3451 {
3452         //
3453         // touching a file on the grid if run has been DONE
3454         //
3455
3456         if (!gGrid)
3457         {
3458                 Log("SHUTTLE",Form("No TGrid connection estabilished!"));
3459                 Log("SHUTTLE",Form("Could not touch file for run %i",GetCurrentRun()));
3460                 return kFALSE;
3461         }
3462
3463         TString dir;
3464         dir.Form("%s%d/%s/SHUTTLE_DONE", fConfig->GetAlienPath(), GetCurrentYear(), GetLHCPeriod());
3465         // checking whether directory for touch command exists
3466         TString commandLs;
3467         commandLs.Form("ls %s",dir.Data());
3468         TGridResult *resultLs = dynamic_cast<TGridResult*>(gGrid->Command(commandLs));
3469         if (!resultLs){
3470                 Log("SHUTTLE",Form("No result for %s command, returning without touching",commandLs.Data()));
3471                 return kFALSE;
3472         }
3473         TMap *mapLs = dynamic_cast<TMap*>(resultLs->At(0));
3474         if (!mapLs){
3475                 Log("SHUTTLE",Form("No map for %s command, returning without touching",commandLs.Data()));
3476                 delete resultLs;
3477                 resultLs = 0x0;
3478                 return kFALSE;
3479         }
3480         TObjString *valueLsPath = dynamic_cast<TObjString*>(mapLs->GetValue("path"));
3481         if (!valueLsPath || (TString)(valueLsPath->GetString()).CompareTo(dir)!=1){
3482                 Log("SHUTTLE",Form("No directory %s found, creating it",dir.Data()));
3483
3484                 // creating the directory
3485
3486                 Bool_t boolMkdir = gGrid->Mkdir(dir.Data());
3487                 if (!boolMkdir) {
3488                         Log("SHUTTLE",Form("Impossible to create dir %s in alien catalogue for run %i!",dir.Data(),GetCurrentRun()));
3489                         delete resultLs;
3490                         resultLs = 0x0;
3491                         return kFALSE;
3492                 }
3493                 Log("SHUTTLE",Form("Directory %s successfully created in alien catalogue for run %i",dir.Data(),GetCurrentRun()));
3494         }
3495         else {
3496                 Log("SHUTTLE",Form("Directory %s correctly found for run %i",dir.Data(),GetCurrentRun()));
3497         }
3498
3499         delete resultLs;
3500         resultLs = 0x0;
3501
3502         TString command;
3503         command.Form("touch %s/%i", dir.Data(), GetCurrentRun());
3504         Log("SHUTTLE", Form("Creating entry in file catalog: %s", command.Data()));
3505         TGridResult *resultTouch = dynamic_cast<TGridResult*>(gGrid->Command(command));
3506         if (!resultTouch){
3507                 Log("SHUTTLE",Form("No result for touching command, returning without touching for run %i",GetCurrentRun()));
3508                 return kFALSE;
3509         }
3510         TMap *mapTouch = dynamic_cast<TMap*>(resultTouch->At(0));
3511         if (!mapTouch){
3512                 Log("SHUTTLE",Form("No map for touching command, returning without touching for run %i",GetCurrentRun()));
3513                 delete resultTouch;
3514                 resultTouch = 0x0;
3515                 return kFALSE;
3516         }
3517         TObjString *valueTouch = dynamic_cast<TObjString*>(mapTouch->GetValue("__result__"));
3518         if (!valueTouch){
3519                 Log("SHUTTLE",Form("No value for \"__result__\" key set in the map for touching command, returning without touching for run %i",GetCurrentRun()));
3520                 delete resultTouch;
3521                 resultTouch = 0x0;
3522                 return kFALSE;
3523         }
3524         if (valueTouch->GetString()!="1"){
3525                 Log("SHUTTLE",Form("Failing the touching command, returning without touching for run %i",GetCurrentRun()));
3526                 delete resultTouch;
3527                 resultTouch = 0x0;
3528                 return kFALSE;
3529         }
3530         delete resultTouch;
3531         resultTouch = 0x0;
3532         Log("SHUTTLE", "Sucessfully touched the file");
3533         return kTRUE;
3534 }
3535 //______________________________________________________________________________________________
3536 UInt_t AliShuttle::GetStartTimeDCSQuery()
3537 {
3538         // Return Start Time for the DCS query
3539         //
3540         // The call is delegated to AliShuttleInterface
3541
3542         return GetCurrentTimeCreated()-fConfig->GetDCSQueryOffset();
3543 }
3544 //______________________________________________________________________________________________
3545 UInt_t AliShuttle::GetEndTimeDCSQuery()
3546 {
3547         // Return End Time for the DCS query
3548         //
3549         // The call is delegated to AliShuttleInterface
3550
3551         return GetCurrentEndTime()+fConfig->GetDCSQueryOffset();
3552 }
3553