SHUTTLE/AliShuttle.cxx

   1 /**************************************************************************
   2  * Copyright(c) 1998-1999, ALICE Experiment at CERN, All rights reserved. *
   3  *                                                                        *
   4  * Author: The ALICE Off-line Project.                                    *
   5  * Contributors are mentioned in the code where appropriate.              *
   6  *                                                                        *
   7  * Permission to use, copy, modify and distribute this software and its   *
   8  * documentation strictly for non-commercial purposes is hereby granted   *
   9  * without fee, provided that the above copyright notice appears in all   *
  10  * copies and that both the copyright notice and this permission notice   *
  11  * appear in the supporting documentation. The authors make no claims     *
  12  * about the suitability of this software for any purpose. It is          *
  13  * provided "as is" without express or implied warranty.                  *
  14  **************************************************************************/
  15
  16 /* $Id$ */
  17
  18 //
  19 // This class is the main manager for AliShuttle.
  20 // It organizes the data retrieval from DCS and call the
  21 // interface methods of AliPreprocessor.
  22 // For every detector in AliShuttleConfgi (see AliShuttleConfig),
  23 // data for its set of aliases is retrieved. If there is registered
  24 // AliPreprocessor for this detector then it will be used
  25 // accroding to the schema (see AliPreprocessor).
  26 // If there isn't registered AliPreprocessor than the retrieved
  27 // data is stored automatically to the undelying AliCDBStorage.
  28 // For detSpec is used the alias name.
  29 //
  30
  31 #include "AliShuttle.h"
  32
  33 #include "AliCDBManager.h"
  34 #include "AliCDBStorage.h"
  35 #include "AliCDBId.h"
  36 #include "AliCDBRunRange.h"
  37 #include "AliCDBPath.h"
  38 #include "AliCDBEntry.h"
  39 #include "AliShuttleConfig.h"
  40 #include "DCSClient/AliDCSClient.h"
  41 #include "AliLog.h"
  42 #include "AliPreprocessor.h"
  43 #include "AliShuttleStatus.h"
  44 #include "AliShuttleLogbookEntry.h"
  45
  46 #include <TSystem.h>
  47 #include <TObject.h>
  48 #include <TString.h>
  49 #include <TTimeStamp.h>
  50 #include <TObjString.h>
  51 #include <TSQLServer.h>
  52 #include <TSQLResult.h>
  53 #include <TSQLRow.h>
  54 #include <TMutex.h>
  55 #include <TSystemDirectory.h>
  56 #include <TSystemFile.h>
  57 #include <TFile.h>
  58 #include <TGrid.h>
  59 #include <TGridResult.h>
  60 #include <TMap.h>
  61
  62 #include <TMonaLisaWriter.h>
  63
  64 #include <fstream>
  65
  66 #include <sys/types.h>
  67 #include <sys/wait.h>
  68
  69 #include <signal.h>
  70
  71 ClassImp(AliShuttle)
  72
  73 //______________________________________________________________________________________________
  74 AliShuttle::AliShuttle(const AliShuttleConfig* config,
  75                 UInt_t timeout, Int_t retries):
  76 fConfig(config),
  77 fTimeout(timeout), fRetries(retries),
  78 fPreprocessorMap(),
  79 fLogbookEntry(0),
  80 fCurrentDetector(),
  81 fFirstProcessing(0),
  82 fFXSError(-1),
  83 fStatusEntry(0),
  84 fMonitoringMutex(0),
  85 fLastActionTime(0),
  86 fLastAction(),
  87 fMonaLisa(0),
  88 fTestMode(kNone),
  89 fReadTestMode(kFALSE),
  90 fOutputRedirected(kFALSE)
  91 {
  92         //
  93         // config: AliShuttleConfig used
  94         // timeout: timeout used for AliDCSClient connection
  95         // retries: the number of retries in case of connection error.
  96         //
  97
  98         if (!fConfig->IsValid()) AliFatal("********** !!!!! Invalid configuration !!!!! **********");
  99         for(int iSys=0;iSys<4;iSys++) {
 100                 fServer[iSys]=0;
 101                 if (iSys < 3)
 102                         fFXSlist[iSys].SetOwner(kTRUE);
 103         }
 104         fPreprocessorMap.SetOwner(kTRUE);
 105
 106         for (UInt_t iDet=0; iDet<NDetectors(); iDet++)
 107                 fFirstUnprocessed[iDet] = kFALSE;
 108
 109         fMonitoringMutex = new TMutex();
 110 }
 111
 112 //______________________________________________________________________________________________
 113 AliShuttle::~AliShuttle()
 114 {
 115         //
 116         // destructor
 117         //
 118
 119         fPreprocessorMap.DeleteAll();
 120         for(int iSys=0;iSys<4;iSys++)
 121                 if(fServer[iSys]) {
 122                         fServer[iSys]->Close();
 123                         delete fServer[iSys];
 124                         fServer[iSys] = 0;
 125                 }
 126
 127         if (fStatusEntry){
 128                 delete fStatusEntry;
 129                 fStatusEntry = 0;
 130         }
 131
 132         if (fMonitoringMutex)
 133         {
 134                 delete fMonitoringMutex;
 135                 fMonitoringMutex = 0;
 136         }
 137 }
 138
 139 //______________________________________________________________________________________________
 140 void AliShuttle::RegisterPreprocessor(AliPreprocessor* preprocessor)
 141 {
 142         //
 143         // Registers new AliPreprocessor.
 144         // It uses GetName() for indentificator of the pre processor.
 145         // The pre processor is registered it there isn't any other
 146         // with the same identificator (GetName()).
 147         //
 148
 149         const char* detName = preprocessor->GetName();
 150         if(GetDetPos(detName) < 0)
 151                 AliFatal(Form("********** !!!!! Invalid detector name: %s !!!!! **********", detName));
 152
 153         if (fPreprocessorMap.GetValue(detName)) {
 154                 AliWarning(Form("AliPreprocessor %s is already registered!", detName));
 155                 return;
 156         }
 157
 158         fPreprocessorMap.Add(new TObjString(detName), preprocessor);
 159 }
 160 //______________________________________________________________________________________________
 161 Bool_t AliShuttle::Store(const AliCDBPath& path, TObject* object,
 162                 AliCDBMetaData* metaData, Int_t validityStart, Bool_t validityInfinite)
 163 {
 164         // Stores a CDB object in the storage for offline reconstruction. Objects that are not needed for
 165         // offline reconstruction, but should be stored anyway (e.g. for debugging) should NOT be stored
 166         // using this function. Use StoreReferenceData instead!
 167         // It calls StoreLocally function which temporarily stores the data locally; when the preprocessor
 168         // finishes the data are transferred to the main storage (Grid).
 169
 170         return StoreLocally(fgkLocalCDB, path, object, metaData, validityStart, validityInfinite);
 171 }
 172
 173 //______________________________________________________________________________________________
 174 Bool_t AliShuttle::StoreReferenceData(const AliCDBPath& path, TObject* object, AliCDBMetaData* metaData)
 175 {
 176         // Stores a CDB object in the storage for reference data. This objects will not be available during
 177         // offline reconstrunction. Use this function for reference data only!
 178         // It calls StoreLocally function which temporarily stores the data locally; when the preprocessor
 179         // finishes the data are transferred to the main storage (Grid).
 180
 181         return StoreLocally(fgkLocalRefStorage, path, object, metaData);
 182 }
 183
 184 //______________________________________________________________________________________________
 185 Bool_t AliShuttle::StoreLocally(const TString& localUri,
 186                         const AliCDBPath& path, TObject* object, AliCDBMetaData* metaData,
 187                         Int_t validityStart, Bool_t validityInfinite)
 188 {
 189         // Store object temporarily in local storage. Parameters are passed by Store and StoreReferenceData functions.
 190         // when the preprocessor finishes the data are transferred to the main storage (Grid).
 191         // The parameters are:
 192         //   1) Uri of the backup storage (Local)
 193         //   2) the object's path.
 194         //   3) the object to be stored
 195         //   4) the metaData to be associated with the object
 196         //   5) the validity start run number w.r.t. the current run,
 197         //      if the data is valid only for this run leave the default 0
 198         //   6) specifies if the calibration data is valid for infinity (this means until updated),
 199         //      typical for calibration runs, the default is kFALSE
 200         //
 201         // returns 0 if fail, 1 otherwise
 202
 203
 204         if (fTestMode & kErrorStorage)
 205         {
 206                 Log(fCurrentDetector, "StoreLocally - In TESTMODE - Simulating error while storing locally");
 207                 return kFALSE;
 208         }
 209
 210         const char* cdbType = (localUri == fgkLocalCDB) ? "CDB" : "Reference";
 211
 212         Int_t firstRun = GetCurrentRun() - validityStart;
 213         if(firstRun < 0) {
 214                 AliWarning("First valid run happens to be less than 0! Setting it to 0.");
 215                 firstRun=0;
 216         }
 217
 218         Int_t lastRun = -1;
 219         if(validityInfinite) {
 220                 lastRun = AliCDBRunRange::Infinity();
 221         } else {
 222                 lastRun = GetCurrentRun();
 223         }
 224
 225         // Version is set to current run, it will be used later to transfer data to Grid
 226         AliCDBId id(path, firstRun, lastRun, GetCurrentRun(), -1);
 227
 228         if(! dynamic_cast<TObjString*> (metaData->GetProperty("RunUsed(TObjString)"))){
 229                 TObjString runUsed = Form("%d", GetCurrentRun());
 230                 metaData->SetProperty("RunUsed(TObjString)", runUsed.Clone());
 231         }
 232
 233         Bool_t result = kFALSE;
 234
 235         if (!(AliCDBManager::Instance()->GetStorage(localUri))) {
 236                 Log("SHUTTLE", Form("StoreLocally - Cannot activate local %s storage", cdbType));
 237         } else {
 238                 Int_t logLevel = AliLog::GetGlobalLogLevel();
 239                 AliLog::SetGlobalLogLevel(AliLog::kError);
 240                 result = AliCDBManager::Instance()->GetStorage(localUri)
 241                                         ->Put(object, id, metaData);
 242                 AliLog::SetGlobalLogLevel((AliLog::EType_t)logLevel);
 243         }
 244
 245         if(!result) {
 246
 247                 Log(fCurrentDetector, Form("StoreLocally - Can't store object <%s>!", id.ToString().Data()));
 248         }
 249
 250
 251         return result;
 252 }
 253
 254 //______________________________________________________________________________________________
 255 Bool_t AliShuttle::StoreOCDB()
 256 {
 257         //
 258         // Called when preprocessor ends successfully or when previous storage attempt failed (kStoreError status)
 259         // Calls underlying StoreOCDB(const char*) function twice, for OCDB and Reference storage.
 260         // Then calls StoreRefFilesToGrid to store reference files.
 261         //
 262
 263         UpdateShuttleStatus(AliShuttleStatus::kStoreStarted);
 264
 265         if (fTestMode & kErrorGrid)
 266         {
 267                 Log("SHUTTLE", "StoreOCDB - In TESTMODE - Simulating error while storing in the Grid");
 268                 Log(fCurrentDetector, "StoreOCDB - In TESTMODE - Simulating error while storing in the Grid");
 269                 return kFALSE;
 270         }
 271
 272         Log("SHUTTLE","StoreOCDB - Storing OCDB data ...");
 273         Int_t resultCDB = StoreOCDB(fgkMainCDB);
 274
 275         Log("SHUTTLE","StoreOCDB - Storing reference data ...");
 276         Int_t resultRef = StoreOCDB(fgkMainRefStorage);
 277
 278         Log("SHUTTLE","StoreOCDB - Storing reference files ...");
 279         Bool_t resultRefFiles = CopyFilesToGrid("reference");
 280
 281         Bool_t resultMetadata = kTRUE;
 282         if(fCurrentDetector == "GRP")
 283         {
 284                 Log("SHUTTLE","StoreOCDB - Storing Run Metadata file ...");
 285                 resultMetadata = CopyFilesToGrid("metadata");
 286         }
 287
 288         Int_t storeResult = 0;
 289
 290         if (resultCDB < 0 || resultRef < 0 || resultRefFiles == kFALSE || resultMetadata == kFALSE)
 291                 storeResult = -1;
 292         else if (resultCDB > 0 || resultRef > 0)
 293                 storeResult = 1;
 294
 295         if (storeResult < 0)
 296         {
 297                 Log("SHUTTLE",
 298                         Form("\t\t\t****** run %d - %s: STORAGE ERROR ******",
 299                                 GetCurrentRun(), fCurrentDetector.Data()));
 300                 UpdateShuttleStatus(AliShuttleStatus::kStoreError);
 301         }
 302         else if (storeResult > 0)
 303         {
 304                 Log("SHUTTLE",
 305                         Form("\t\t\t****** run %d - %s: STORAGE DELAYED ******",
 306                                 GetCurrentRun(), fCurrentDetector.Data()));
 307                 UpdateShuttleStatus(AliShuttleStatus::kStoreDelayed);
 308         }
 309         else if (storeResult == 0)
 310         {
 311                 Log("SHUTTLE",
 312                         Form("\t\t\t****** run %d - %s: DONE ******",
 313                                 GetCurrentRun(), fCurrentDetector.Data()));
 314                 UpdateShuttleStatus(AliShuttleStatus::kDone);
 315                 UpdateShuttleLogbook(fCurrentDetector, "DONE");
 316         }
 317
 318         return (storeResult == 0);
 319 }
 320
 321 //______________________________________________________________________________________________
 322 Int_t AliShuttle::StoreOCDB(const TString& gridURI)
 323 {
 324         //
 325         // Called by StoreOCDB(), performs actual storage to the main OCDB and reference storages (Grid)
 326         //
 327         // Return code:
 328         //   -2 initialization error
 329         //   -1 storage error
 330         //   0  success
 331         //   1  storage delayed (e.g. previous unprocessed runs)
 332         //
 333
 334         TObjArray* gridIds=0;
 335
 336         Bool_t result = kTRUE;
 337         Bool_t delayed = kFALSE;
 338
 339         const char* type = 0;
 340         TString localURI;
 341         if(gridURI == fgkMainCDB) {
 342                 type = "OCDB";
 343                 localURI = fgkLocalCDB;
 344         } else if(gridURI == fgkMainRefStorage) {
 345                 type = "reference";
 346                 localURI = fgkLocalRefStorage;
 347         } else {
 348                 AliError(Form("Invalid storage URI: %s", gridURI.Data()));
 349                 return -2;
 350         }
 351
 352         AliCDBManager* man = AliCDBManager::Instance();
 353
 354         AliCDBStorage *gridSto = man->GetStorage(gridURI);
 355         if(!gridSto) {
 356                 Log("SHUTTLE",
 357                         Form("StoreOCDB - cannot activate main %s storage", type));
 358                 return -2;
 359         }
 360
 361         gridIds = gridSto->GetQueryCDBList();
 362
 363         // get objects previously stored in local CDB
 364         AliCDBStorage *localSto = man->GetStorage(localURI);
 365         if(!localSto) {
 366                 Log("SHUTTLE",
 367                         Form("StoreOCDB - cannot activate local %s storage", type));
 368                 return -2;
 369         }
 370         AliCDBPath aPath(GetOfflineDetName(fCurrentDetector.Data()),"*","*");
 371         // Local objects were stored with current run as Grid version!
 372         TList* localEntries = localSto->GetAll(aPath.GetPath(), GetCurrentRun(), GetCurrentRun());
 373         localEntries->SetOwner(1);
 374
 375         // loop on local stored objects
 376         TIter localIter(localEntries);
 377         AliCDBEntry *aLocEntry = 0;
 378         while((aLocEntry = dynamic_cast<AliCDBEntry*> (localIter.Next()))){
 379                 aLocEntry->SetOwner(1);
 380                 AliCDBId aLocId = aLocEntry->GetId();
 381                 aLocEntry->SetVersion(-1);
 382                 aLocEntry->SetSubVersion(-1);
 383
 384                 Log(fCurrentDetector.Data(), Form("Attempting to store %s", aLocId.ToString().Data()));
 385
 386                 // If local object is valid up to infinity we store it only if it is
 387                 // the first unprocessed run!
 388                 if (aLocId.GetLastRun() == AliCDBRunRange::Infinity() &&
 389                         !fFirstUnprocessed[GetDetPos(fCurrentDetector)])
 390                 {
 391                         Log("SHUTTLE", Form("StoreOCDB - %s: object %s has validity infinite but "
 392                                                 "there are previous unprocessed runs!",
 393                                                 fCurrentDetector.Data(), aLocId.GetPath().Data()));
 394                         Log(fCurrentDetector.Data(), Form("StoreOCDB - %s: object %s has validity infinite but "
 395                                                 "there are previous unprocessed runs!",
 396                                                 fCurrentDetector.Data(), aLocId.GetPath().Data()));
 397                         delayed = kTRUE;
 398                         continue;
 399                 }
 400
 401                 // loop on Grid valid Id's
 402                 Bool_t store = kTRUE;
 403                 TIter gridIter(gridIds);
 404                 AliCDBId* aGridId = 0;
 405                 while ((aGridId = dynamic_cast<AliCDBId*> (gridIter.Next()))) {
 406                         if (aGridId->GetPath() != aLocId.GetPath())
 407                                 continue;
 408                         // skip all objects valid up to infinity
 409                         if (aGridId->GetLastRun() == AliCDBRunRange::Infinity())
 410                                 continue;
 411
 412                         // if we get here, it means there's already some more recent object stored on Grid!
 413                         Log(fCurrentDetector.Data(),
 414                                 Form("StoreOCDB - A more recent object already exists in %s storage: <%s>",
 415                                 type, aGridId->ToString().Data()));
 416
 417                         store = kFALSE;
 418                         break;
 419                 }
 420
 421                 Bool_t storeOk = kFALSE;
 422                 if (store)
 423                 {
 424                         Log(fCurrentDetector.Data(), Form("Prechecks succeeded. Ready to store %s", aLocId.ToString().Data()));
 425                         storeOk = gridSto->Put(aLocEntry);
 426                         if (storeOk) {
 427                                 Log("SHUTTLE",
 428                                 Form("StoreOCDB - Object <%s> successfully put into %s storage",
 429                                         aLocId.ToString().Data(), type));
 430                                 Log(fCurrentDetector.Data(),
 431                                         Form("StoreOCDB - Object <%s> successfully put into %s storage",
 432                                         aLocId.ToString().Data(), type));
 433                         } else  {
 434                                 Log("SHUTTLE",
 435                                         Form("StoreOCDB - Grid %s storage of object <%s> failed",
 436                                         type, aLocId.ToString().Data()));
 437                                 Log(fCurrentDetector.Data(),
 438                                         Form("StoreOCDB - Grid %s storage of object <%s> failed",
 439                                         type, aLocId.ToString().Data()));
 440                                 result = kFALSE;
 441                         }
 442                 }
 443
 444                 if (!store || storeOk) {
 445                         // removing local file...
 446                         TString filename;
 447                         localSto->IdToFilename(aLocId, filename);
 448                         Log("SHUTTLE", Form("StoreOCDB - Removing local file %s", filename.Data()));
 449                         RemoveFile(filename.Data());
 450                 }
 451         }
 452         localEntries->Clear();
 453
 454         Int_t returnCode = 0;
 455
 456         if (result == kFALSE)
 457                 returnCode = -1;
 458         else if (delayed != kFALSE)
 459                 returnCode =  1;
 460
 461         Log("SHUTTLE", Form("StoreOCDB - Returning with %d (result = %d, delayed = %d)", returnCode, result, delayed));
 462         Log(fCurrentDetector.Data(), Form("StoreOCDB - Returning with %d (result = %d, delayed = %d)", returnCode, result, delayed));
 463
 464         return returnCode;
 465 }
 466
 467 //______________________________________________________________________________________________
 468 Bool_t AliShuttle::CleanReferenceStorage(const char* detector)
 469 {
 470         // clears the directory used to store reference files of a given subdetector
 471
 472         AliCDBManager* man = AliCDBManager::Instance();
 473         AliCDBStorage* sto = man->GetStorage(fgkLocalRefStorage);
 474         TString localBaseFolder = sto->GetBaseFolder();
 475
 476         TString targetDir = GetRefFilePrefix(localBaseFolder.Data(), detector);
 477
 478         Log("SHUTTLE", Form("CleanReferenceStorage - Cleaning %s", targetDir.Data()));
 479
 480         TString begin;
 481         begin.Form("%d_", GetCurrentRun());
 482
 483         TSystemDirectory* baseDir = new TSystemDirectory("/", targetDir);
 484         if (!baseDir)
 485                 return kTRUE;
 486
 487         TList* dirList = baseDir->GetListOfFiles();
 488         delete baseDir;
 489
 490         if (!dirList) return kTRUE;
 491
 492         if (dirList->GetEntries() < 3)
 493         {
 494                 delete dirList;
 495                 return kTRUE;
 496         }
 497
 498         Int_t nDirs = 0, nDel = 0;
 499         TIter dirIter(dirList);
 500         TSystemFile* entry = 0;
 501
 502         Bool_t success = kTRUE;
 503
 504         while ((entry = dynamic_cast<TSystemFile*> (dirIter.Next())))
 505         {
 506                 if (entry->IsDirectory())
 507                         continue;
 508
 509                 TString fileName(entry->GetName());
 510                 if (!fileName.BeginsWith(begin))
 511                         continue;
 512
 513                 nDirs++;
 514
 515                 // delete file
 516                 Int_t result = gSystem->Unlink(fileName.Data());
 517
 518                 if (result)
 519                 {
 520                         Log("SHUTTLE", Form("CleanReferenceStorage - Could not delete file %s!", fileName.Data()));
 521                         success = kFALSE;
 522                 } else {
 523                         nDel++;
 524                 }
 525         }
 526
 527         if(nDirs > 0)
 528                 Log("SHUTTLE", Form("CleanReferenceStorage - %d (over %d) reference files in folder %s were deleted.",
 529                         nDel, nDirs, targetDir.Data()));
 530
 531
 532         delete dirList;
 533         return success;
 534
 535
 536
 537
 538
 539
 540   Int_t result = gSystem->GetPathInfo(targetDir, 0, (Long64_t*) 0, 0, 0);
 541   if (result == 0)
 542   {
 543     // delete directory
 544     result = gSystem->Exec(Form("rm -rf %s", targetDir.Data()));
 545     if (result != 0)
 546     {
 547       Log("SHUTTLE", Form("CleanReferenceStorage - Could not clean directory %s", targetDir.Data()));
 548       return kFALSE;
 549     }
 550   }
 551
 552   result = gSystem->mkdir(targetDir, kTRUE);
 553   if (result != 0)
 554   {
 555     Log("SHUTTLE", Form("CleanReferenceStorage - Error creating base directory %s", targetDir.Data()));
 556     return kFALSE;
 557   }
 558
 559   return kTRUE;
 560 }
 561
 562 //______________________________________________________________________________________________
 563 Bool_t AliShuttle::StoreReferenceFile(const char* detector, const char* localFile, const char* gridFileName)
 564 {
 565         //
 566         // Stores reference file directly (without opening it). This function stores the file locally.
 567         //
 568         // The file is stored under the following location:
 569         // <base folder of local reference storage>/<DET>/<RUN#>_<gridFileName>
 570         // where <gridFileName> is the second parameter given to the function
 571         //
 572
 573         if (fTestMode & kErrorStorage)
 574         {
 575                 Log(fCurrentDetector, "StoreReferenceFile - In TESTMODE - Simulating error while storing locally");
 576                 return kFALSE;
 577         }
 578
 579         AliCDBManager* man = AliCDBManager::Instance();
 580         AliCDBStorage* sto = man->GetStorage(fgkLocalRefStorage);
 581
 582         TString localBaseFolder = sto->GetBaseFolder();
 583
 584         TString target = GetRefFilePrefix(localBaseFolder.Data(), detector);
 585         target.Append(Form("/%d_%s", GetCurrentRun(), gridFileName));
 586
 587         return CopyFileLocally(localFile, target);
 588 }
 589
 590 //______________________________________________________________________________________________
 591 Bool_t AliShuttle::StoreRunMetadataFile(const char* localFile, const char* gridFileName)
 592 {
 593         //
 594         // Stores Run metadata file to the Grid, in the run folder
 595         //
 596         // Only GRP can call this function.
 597
 598         if (fTestMode & kErrorStorage)
 599         {
 600                 Log(fCurrentDetector, "StoreRunMetaDataFile - In TESTMODE - Simulating error while storing locally");
 601                 return kFALSE;
 602         }
 603
 604         AliCDBManager* man = AliCDBManager::Instance();
 605         AliCDBStorage* sto = man->GetStorage(fgkLocalRefStorage);
 606
 607         TString localBaseFolder = sto->GetBaseFolder();
 608
 609         // Build Run level folder
 610         // folder = /alice/data/year/lhcPeriod/runNb/raw
 611
 612
 613         TString lhcPeriod = GetLHCPeriod();
 614         if (lhcPeriod.Length() == 0)
 615         {
 616                 Log("SHUTTLE","StoreRunMetaDataFile - LHCPeriod not found in logbook!");
 617                 return 0;
 618         }
 619
 620         // TODO partitions with one detector only write data into LHCperiod_DET
 621         TString partition = GetRunParameter("detector");
 622
 623         if (partition.Length() > 0 && partition != "ALICE")
 624         {
 625                 lhcPeriod.Append(Form("_%s", partition.Data()));
 626                 Log(fCurrentDetector, Form("Run data tags merged file will be written in %s",
 627                                 lhcPeriod.Data()));
 628         }
 629
 630         TString target = Form("%s/GRP/RunMetadata%s%d/%s/%09d/raw/%s",
 631                                 localBaseFolder.Data(), fConfig->GetAlienPath(), GetCurrentYear(),
 632                                 lhcPeriod.Data(), GetCurrentRun(), gridFileName);
 633
 634         return CopyFileLocally(localFile, target);
 635 }
 636
 637 //______________________________________________________________________________________________
 638 Bool_t AliShuttle::CopyFileLocally(const char* localFile, const TString& target)
 639 {
 640         //
 641         // Stores file locally. Called by StoreReferenceFile and StoreRunMetadataFile
 642         // Files are temporarily stored in the local reference storage. When the preprocessor
 643         // finishes, the Shuttle calls CopyFilesToGrid to transfer the files to AliEn
 644         // (in reference or run level folders)
 645         //
 646
 647         TString targetDir(target(0, target.Last('/')));
 648
 649         //try to open base dir folder, if it does not exist
 650         void* dir = gSystem->OpenDirectory(targetDir.Data());
 651         if (dir == NULL) {
 652                 if (gSystem->mkdir(targetDir.Data(), kTRUE)) {
 653                         Log("SHUTTLE", Form("CopyFileLocally - Can't open directory <%s>", targetDir.Data()));
 654                         return kFALSE;
 655                 }
 656
 657         } else {
 658                 gSystem->FreeDirectory(dir);
 659         }
 660
 661         Int_t result = 0;
 662
 663         result = gSystem->GetPathInfo(localFile, 0, (Long64_t*) 0, 0, 0);
 664         if (result)
 665         {
 666                 Log("SHUTTLE", Form("CopyFileLocally - %s does not exist", localFile));
 667                 return kFALSE;
 668         }
 669
 670         result = gSystem->GetPathInfo(target, 0, (Long64_t*) 0, 0, 0);
 671         if (!result)
 672         {
 673                 Log("SHUTTLE", Form("CopyFileLocally - target file %s already exist, removing...", target.Data()));
 674                 if (gSystem->Unlink(target.Data()))
 675                 {
 676                         Log("SHUTTLE", Form("CopyFileLocally - Could not remove existing target file %s!", target.Data()));
 677                         return kFALSE;
 678                 }
 679         }
 680
 681         result = gSystem->CopyFile(localFile, target);
 682
 683         if (result == 0)
 684         {
 685                 Log("SHUTTLE", Form("CopyFileLocally - File %s stored locally to %s", localFile, target.Data()));
 686                 return kTRUE;
 687         }
 688         else
 689         {
 690                 Log("SHUTTLE", Form("CopyFileLocally - Could not store file %s to %s! Error code = %d",
 691                                 localFile, target.Data(), result));
 692                 return kFALSE;
 693         }
 694
 695
 696
 697 }
 698
 699 //______________________________________________________________________________________________
 700 Bool_t AliShuttle::CopyFilesToGrid(const char* type)
 701 {
 702         //
 703         // Transfers local files to the Grid. Local files can be reference files
 704         // or run metadata file (from GRP only).
 705         //
 706         // According to the type (ref, metadata) the files are stored under the following location:
 707         // ref --> <base folder of reference storage>/<DET>/<RUN#>_<gridFileName>
 708         // metadata --> <run data folder>/<MetadataFileName>
 709         //
 710
 711         AliCDBManager* man = AliCDBManager::Instance();
 712         AliCDBStorage* sto = man->GetStorage(fgkLocalRefStorage);
 713         if (!sto)
 714                 return kFALSE;
 715         TString localBaseFolder = sto->GetBaseFolder();
 716
 717         TString dir;
 718         TString alienDir;
 719         TString begin;
 720
 721         if (strcmp(type, "reference") == 0)
 722         {
 723                 dir = GetRefFilePrefix(localBaseFolder.Data(), fCurrentDetector.Data());
 724                 AliCDBStorage* gridSto = man->GetStorage(fgkMainRefStorage);
 725                 if (!gridSto)
 726                         return kFALSE;
 727                 TString gridBaseFolder = gridSto->GetBaseFolder();
 728                 alienDir = GetRefFilePrefix(gridBaseFolder.Data(), fCurrentDetector.Data());
 729                 begin = Form("%d_", GetCurrentRun());
 730         }
 731         else if (strcmp(type, "metadata") == 0)
 732         {
 733
 734                 TString lhcPeriod = GetLHCPeriod();
 735
 736                 if (lhcPeriod.Length() == 0)
 737                 {
 738                         Log("SHUTTLE","CopyFilesToGrid - LHCPeriod not found in logbook!");
 739                         return 0;
 740                 }
 741
 742                 // TODO partitions with one detector only write data into LHCperiod_DET
 743                 TString partition = GetRunParameter("detector");
 744
 745                 if (partition.Length() > 0 && partition != "ALICE")
 746                 {
 747                         lhcPeriod.Append(Form("_%s", partition.Data()));
 748                 }
 749
 750                 dir = Form("%s/GRP/RunMetadata%s%d/%s/%09d/raw",
 751                                 localBaseFolder.Data(), fConfig->GetAlienPath(), GetCurrentYear(),
 752                                 lhcPeriod.Data(), GetCurrentRun());
 753                 alienDir = dir(dir.Index(fConfig->GetAlienPath()), dir.Length());
 754
 755                 begin = "";
 756         }
 757         else
 758         {
 759                 Log("SHUTTLE", "CopyFilesToGrid - Unexpected: type label must be reference or metadata!");
 760                 return kFALSE;
 761         }
 762
 763         TSystemDirectory* baseDir = new TSystemDirectory("/", dir);
 764         if (!baseDir)
 765                 return kTRUE;
 766
 767         TList* dirList = baseDir->GetListOfFiles();
 768         delete baseDir;
 769
 770         if (!dirList) return kTRUE;
 771
 772         if (dirList->GetEntries() < 3)
 773         {
 774                 delete dirList;
 775                 return kTRUE;
 776         }
 777
 778         if (!gGrid)
 779         {
 780                 Log("SHUTTLE", "CopyFilesToGrid - Connection to Grid failed: Cannot continue!");
 781                 delete dirList;
 782                 return kFALSE;
 783         }
 784
 785         Int_t nDirs = 0, nTransfer = 0;
 786         TIter dirIter(dirList);
 787         TSystemFile* entry = 0;
 788
 789         Bool_t success = kTRUE;
 790         Bool_t first = kTRUE;
 791
 792         while ((entry = dynamic_cast<TSystemFile*> (dirIter.Next())))
 793         {
 794                 if (entry->IsDirectory())
 795                         continue;
 796
 797                 TString fileName(entry->GetName());
 798                 if (!fileName.BeginsWith(begin))
 799                         continue;
 800
 801                 nDirs++;
 802
 803                 if (first)
 804                 {
 805                         first = kFALSE;
 806                         // check that folder exists, otherwise create it
 807                         TGridResult* result = gGrid->Ls(alienDir.Data(), "a");
 808
 809                         if (!result)
 810                         {
 811                                 delete dirList;
 812                                 return kFALSE;
 813                         }
 814
 815                         if (!result->GetFileName(1)) // TODO: It looks like element 0 is always 0!!
 816                         {
 817                                 // TODO It does not work currently! Bug in TAliEn::Mkdir
 818                                 // TODO Manually fixed in local root v5-16-00
 819                                 if (!gGrid->Mkdir(alienDir.Data(),"-p",0))
 820                                 {
 821                                         Log("SHUTTLE", Form("CopyFilesToGrid - Cannot create directory %s",
 822                                                         alienDir.Data()));
 823                                         delete dirList;
 824                                         return kFALSE;
 825                                 } else {
 826                                         Log("SHUTTLE",Form("CopyFilesToGrid - Folder %s created", alienDir.Data()));
 827                                 }
 828
 829                         } else {
 830                                         Log("SHUTTLE",Form("CopyFilesToGrid - Folder %s found", alienDir.Data()));
 831                         }
 832                 }
 833
 834                 TString fullLocalPath;
 835                 fullLocalPath.Form("%s/%s", dir.Data(), fileName.Data());
 836
 837                 TString fullGridPath;
 838                 fullGridPath.Form("alien://%s/%s", alienDir.Data(), fileName.Data());
 839
 840                 Bool_t result = TFile::Cp(fullLocalPath, fullGridPath);
 841
 842                 if (result)
 843                 {
 844                         Log("SHUTTLE", Form("CopyFilesToGrid - Copying local file %s to %s succeeded!",
 845                                                 fullLocalPath.Data(), fullGridPath.Data()));
 846                         RemoveFile(fullLocalPath);
 847                         nTransfer++;
 848                 }
 849                 else
 850                 {
 851                         Log("SHUTTLE", Form("CopyFilesToGrid - Copying local file %s to %s FAILED!",
 852                                                 fullLocalPath.Data(), fullGridPath.Data()));
 853                         success = kFALSE;
 854                 }
 855         }
 856
 857         Log("SHUTTLE", Form("CopyFilesToGrid - %d (over %d) files in folder %s copied to Grid.",
 858                                                 nTransfer, nDirs, dir.Data()));
 859
 860
 861         delete dirList;
 862         return success;
 863 }
 864
 865 //______________________________________________________________________________________________
 866 const char* AliShuttle::GetRefFilePrefix(const char* base, const char* detector)
 867 {
 868         //
 869         // Get folder name of reference files
 870         //
 871
 872         TString offDetStr(GetOfflineDetName(detector));
 873         static TString dir;
 874         if (offDetStr == "ITS" || offDetStr == "MUON" || offDetStr == "PHOS")
 875         {
 876                 dir.Form("%s/%s/%s", base, offDetStr.Data(), detector);
 877         } else {
 878                 dir.Form("%s/%s", base, offDetStr.Data());
 879         }
 880
 881         return dir.Data();
 882 }
 883
 884 //______________________________________________________________________________________________
 885 void AliShuttle::CleanLocalStorage(const TString& uri)
 886 {
 887         //
 888         // Called in case the preprocessor is declared failed. Remove remaining objects from the local storages.
 889         //
 890
 891         const char* type = 0;
 892         if(uri == fgkLocalCDB) {
 893                 type = "OCDB";
 894         } else if(uri == fgkLocalRefStorage) {
 895                 type = "Reference";
 896         } else {
 897                 AliError(Form("Invalid storage URI: %s", uri.Data()));
 898                 return;
 899         }
 900
 901         AliCDBManager* man = AliCDBManager::Instance();
 902
 903         // open local storage
 904         AliCDBStorage *localSto = man->GetStorage(uri);
 905         if(!localSto) {
 906                 Log("SHUTTLE",
 907                         Form("CleanLocalStorage - cannot activate local %s storage", type));
 908                 return;
 909         }
 910
 911         TString filename(Form("%s/%s/*/Run*_v%d_s*.root",
 912                 localSto->GetBaseFolder().Data(), GetOfflineDetName(fCurrentDetector.Data()), GetCurrentRun()));
 913
 914         AliDebug(2, Form("filename = %s", filename.Data()));
 915
 916         Log("SHUTTLE", Form("Removing remaining local files for run %d and detector %s ...",
 917                 GetCurrentRun(), fCurrentDetector.Data()));
 918
 919         RemoveFile(filename.Data());
 920
 921 }
 922
 923 //______________________________________________________________________________________________
 924 void AliShuttle::RemoveFile(const char* filename)
 925 {
 926         //
 927         // removes local file
 928         //
 929
 930         TString command(Form("rm -f %s", filename));
 931
 932         Int_t result = gSystem->Exec(command.Data());
 933         if(result != 0)
 934         {
 935                 Log("SHUTTLE", Form("RemoveFile - %s: Cannot remove file %s!",
 936                         fCurrentDetector.Data(), filename));
 937         }
 938 }
 939
 940 //______________________________________________________________________________________________
 941 AliShuttleStatus* AliShuttle::ReadShuttleStatus()
 942 {
 943         //
 944         // Reads the AliShuttleStatus from the CDB
 945         //
 946
 947         if (fStatusEntry){
 948                 delete fStatusEntry;
 949                 fStatusEntry = 0;
 950         }
 951
 952         Int_t path1 = GetCurrentRun()/10000;
 953         fStatusEntry = AliCDBManager::Instance()->GetStorage(GetLocalCDB())
 954                 ->Get(Form("/SHUTTLE/%s/%d", fCurrentDetector.Data(), path1), GetCurrentRun());
 955
 956         if (!fStatusEntry) return 0;
 957         fStatusEntry->SetOwner(1);
 958
 959         AliShuttleStatus* status = dynamic_cast<AliShuttleStatus*> (fStatusEntry->GetObject());
 960         if (!status) {
 961                 AliError("Invalid object stored to CDB!");
 962                 return 0;
 963         }
 964
 965         return status;
 966 }
 967
 968 //______________________________________________________________________________________________
 969 Bool_t AliShuttle::WriteShuttleStatus(AliShuttleStatus* status)
 970 {
 971         //
 972         // writes the status for one subdetector
 973         //
 974
 975         if (fStatusEntry){
 976                 delete fStatusEntry;
 977                 fStatusEntry = 0;
 978         }
 979
 980         Int_t run = GetCurrentRun();
 981         Int_t path1 = run/10000;
 982         TString path1_string = Form("%d",path1);
 983
 984         AliCDBId id(AliCDBPath("SHUTTLE", fCurrentDetector, path1_string), run, run);
 985
 986         fStatusEntry = new AliCDBEntry(status, id, new AliCDBMetaData);
 987         fStatusEntry->SetOwner(1);
 988
 989         Int_t logLevel = AliLog::GetGlobalLogLevel();
 990         AliLog::SetGlobalLogLevel(AliLog::kError);
 991
 992         UInt_t result = AliCDBManager::Instance()->GetStorage(fgkLocalCDB)->Put(fStatusEntry);
 993
 994         AliLog::SetGlobalLogLevel((AliLog::EType_t)logLevel);
 995
 996         if (!result) {
 997                 Log("SHUTTLE", Form("WriteShuttleStatus - Failed for %s, run %d",
 998                                                 fCurrentDetector.Data(), run));
 999                 return kFALSE;
1000         }
1001
1002         SendMLDetInfo();
1003
1004         return kTRUE;
1005 }
1006
1007 //______________________________________________________________________________________________
1008 void AliShuttle::UpdateShuttleStatus(AliShuttleStatus::Status newStatus, Bool_t increaseCount)
1009 {
1010         //
1011         // changes the AliShuttleStatus for the given detector and run to the given status
1012         //
1013
1014         if (!fStatusEntry){
1015                 AliError("UNEXPECTED: fStatusEntry empty");
1016                 return;
1017         }
1018
1019         AliShuttleStatus* status = dynamic_cast<AliShuttleStatus*> (fStatusEntry->GetObject());
1020
1021         if (!status){
1022                 Log("SHUTTLE", "UpdateShuttleStatus - UNEXPECTED: status could not be read from current CDB entry");
1023                 return;
1024         }
1025
1026         TString actionStr = Form("UpdateShuttleStatus - %s: Changing state from %s to %s",
1027                                 fCurrentDetector.Data(),
1028                                 status->GetStatusName(),
1029                                 status->GetStatusName(newStatus));
1030         Log("SHUTTLE", actionStr);
1031         SetLastAction(actionStr);
1032
1033         status->SetStatus(newStatus);
1034         if (increaseCount) status->IncreaseCount();
1035
1036         Int_t logLevel = AliLog::GetGlobalLogLevel();
1037         AliLog::SetGlobalLogLevel(AliLog::kError);
1038
1039         AliCDBManager::Instance()->GetStorage(fgkLocalCDB)->Put(fStatusEntry);
1040
1041         AliLog::SetGlobalLogLevel((AliLog::EType_t)logLevel);
1042
1043         SendMLDetInfo();
1044 }
1045
1046 //______________________________________________________________________________________________
1047 void AliShuttle::SendMLDetInfo()
1048 {
1049         //
1050         // sends ML information about the current status of the current detector being processed
1051         //
1052
1053         AliShuttleStatus* status = dynamic_cast<AliShuttleStatus*> (fStatusEntry->GetObject());
1054
1055         if (!status){
1056                 Log("SHUTTLE", "SendMLDetInfo - UNEXPECTED: status could not be read from current CDB entry");
1057                 return;
1058         }
1059
1060         TMonaLisaText  mlStatus(Form("%s_status", fCurrentDetector.Data()), status->GetStatusName());
1061         TMonaLisaValue mlRetryCount(Form("%s_count", fCurrentDetector.Data()), status->GetCount());
1062
1063         TList mlList;
1064         mlList.Add(&mlStatus);
1065         mlList.Add(&mlRetryCount);
1066
1067         TString mlID;
1068         mlID.Form("%d", GetCurrentRun());
1069         fMonaLisa->SendParameters(&mlList, mlID);
1070 }
1071
1072 //______________________________________________________________________________________________
1073 Bool_t AliShuttle::ContinueProcessing()
1074 {
1075         // this function reads the AliShuttleStatus information from CDB and
1076         // checks if the processing should be continued
1077         // if yes it returns kTRUE and updates the AliShuttleStatus with nextStatus
1078
1079         if (!fConfig->HostProcessDetector(fCurrentDetector))
1080                 return kFALSE;
1081
1082         AliPreprocessor* aPreprocessor =
1083                 dynamic_cast<AliPreprocessor*> (fPreprocessorMap.GetValue(fCurrentDetector));
1084         if (!aPreprocessor)
1085         {
1086                 Log("SHUTTLE", Form("ContinueProcessing - %s: no preprocessor registered", fCurrentDetector.Data()));
1087                 return kFALSE;
1088         }
1089
1090         AliShuttleLogbookEntry::Status entryStatus =
1091                 fLogbookEntry->GetDetectorStatus(fCurrentDetector);
1092
1093         if (entryStatus != AliShuttleLogbookEntry::kUnprocessed) {
1094                 Log("SHUTTLE", Form("ContinueProcessing - %s is %s",
1095                                 fCurrentDetector.Data(),
1096                                 fLogbookEntry->GetDetectorStatusName(entryStatus)));
1097                 return kFALSE;
1098         }
1099
1100         // if we get here, according to Shuttle logbook subdetector is in UNPROCESSED state
1101
1102         // check if current run is first unprocessed run for current detector
1103         if (fConfig->StrictRunOrder(fCurrentDetector) &&
1104                 !fFirstUnprocessed[GetDetPos(fCurrentDetector)])
1105         {
1106                 if (fTestMode == kNone)
1107                 {
1108                         Log("SHUTTLE", Form("ContinueProcessing - %s requires strict run ordering"
1109                                         " but this is not the first unprocessed run!"));
1110                         return kFALSE;
1111                 }
1112                 else
1113                 {
1114                         Log("SHUTTLE", Form("ContinueProcessing - In TESTMODE - "
1115                                         "Although %s requires strict run ordering "
1116                                         "and this is not the first unprocessed run, "
1117                                         "the SHUTTLE continues"));
1118                 }
1119         }
1120
1121         // Is the subdetector processed first time for this run?
1122         fFirstProcessing = kFALSE;
1123
1124         AliShuttleStatus* status = ReadShuttleStatus();
1125         if (!status) {
1126                 // first time
1127                 Log("SHUTTLE", Form("ContinueProcessing - %s: Processing first time",
1128                                 fCurrentDetector.Data()));
1129                 status = new AliShuttleStatus(AliShuttleStatus::kStarted);
1130                 fFirstProcessing = kTRUE;
1131                 return WriteShuttleStatus(status);
1132         }
1133
1134         // The following case shouldn't happen if Shuttle Logbook was correctly updated.
1135         // If it happens it may mean Logbook updating failed... let's do it now!
1136         if (status->GetStatus() == AliShuttleStatus::kDone ||
1137             status->GetStatus() == AliShuttleStatus::kFailed ||
1138             status->GetStatus() == AliShuttleStatus::kSkipped) {
1139                 Log("SHUTTLE", Form("ContinueProcessing - %s is already %s. Updating Shuttle Logbook",
1140                                         fCurrentDetector.Data(),
1141                                         status->GetStatusName(status->GetStatus())));
1142
1143                 if (status->GetStatus() == AliShuttleStatus::kSkipped)
1144                 {
1145                         UpdateShuttleLogbook(fCurrentDetector.Data(), "DONE");
1146                 }
1147                 else
1148                         UpdateShuttleLogbook(fCurrentDetector.Data(), status->GetStatusName(status->GetStatus()));
1149
1150                 return kFALSE;
1151         }
1152
1153         if (status->GetStatus() == AliShuttleStatus::kStoreStarted || status->GetStatus() == AliShuttleStatus::kStoreDelayed ||status->GetStatus() == AliShuttleStatus::kStoreError) {
1154                 Log("SHUTTLE",
1155                         Form("ContinueProcessing - %s: Grid storage of one or more "
1156                                 "objects failed. Trying again now",
1157                                 fCurrentDetector.Data()));
1158                 StoreOCDB();
1159                 return kFALSE;
1160         }
1161
1162         // if we get here, there is a restart
1163         Bool_t cont = kFALSE;
1164
1165         // abort conditions
1166         if (status->GetCount() >= fConfig->GetMaxRetries()) {
1167                 Log("SHUTTLE", Form("ContinueProcessing - %s failed %d times in status %s - "
1168                                 "Updating Shuttle Logbook", fCurrentDetector.Data(),
1169                                 status->GetCount(), status->GetStatusName()));
1170                 UpdateShuttleLogbook(fCurrentDetector.Data(), "FAILED");
1171                 UpdateShuttleStatus(AliShuttleStatus::kFailed);
1172
1173                 // there may still be objects in local OCDB and reference storage
1174                 // and FXS databases may be not updated: do it now!
1175
1176                 // TODO Currently disabled, we want to keep files in case of failure!
1177                 // CleanLocalStorage(fgkLocalCDB);
1178                 // CleanLocalStorage(fgkLocalRefStorage);
1179                 // UpdateTableFailCase();
1180
1181                 // Send mail to detector expert!
1182                 Log("SHUTTLE", Form("ContinueProcessing - Sending mail to %s expert...",
1183                                     fCurrentDetector.Data()));
1184                 // det experts in to
1185                 TString to="";
1186                 TIter *iterExperts = 0;
1187                 iterExperts = new TIter(fConfig->GetResponsibles(fCurrentDetector));
1188                 TObjString *anExpert=0;
1189                 while ((anExpert = (TObjString*) iterExperts->Next()))
1190                         {
1191                                 to += Form("%s, \n", anExpert->GetName());
1192                         }
1193                 delete iterExperts;
1194
1195                 if (to.Length() > 0)
1196                         to.Remove(to.Length()-3);
1197                 AliDebug(2, Form("to: %s",to.Data()));
1198
1199                 if (to.IsNull()) {
1200                         Log("SHUTTLE", Form("List of %s responsibles not set!", fCurrentDetector.Data()));
1201                         return kFALSE;
1202                 }
1203
1204                 Log(fCurrentDetector.Data(), Form("ContinueProcessing - Sending mail to %s expert(s):",
1205                                     fCurrentDetector.Data()));
1206                 Log(fCurrentDetector.Data(), Form("\n%s", to.Data()));
1207                 if (!SendMail(kPPEMail))
1208                         Log("SHUTTLE", Form("ContinueProcessing - Could not send mail to %s expert",
1209                                             fCurrentDetector.Data()));
1210
1211         } else {
1212                 Log("SHUTTLE", Form("ContinueProcessing - %s: restarting. "
1213                                 "Aborted before with %s. Retry number %d.", fCurrentDetector.Data(),
1214                                 status->GetStatusName(), status->GetCount()));
1215                 Bool_t increaseCount = kTRUE;
1216                 if (status->GetStatus() == AliShuttleStatus::kDCSError ||
1217                     status->GetStatus() == AliShuttleStatus::kDCSStarted ||
1218                     status->GetStatus() == AliShuttleStatus::kFXSError)
1219                                 increaseCount = kFALSE;
1220
1221                 UpdateShuttleStatus(AliShuttleStatus::kStarted, increaseCount);
1222                 cont = kTRUE;
1223         }
1224
1225         return cont;
1226 }
1227
1228 //______________________________________________________________________________________________
1229 void AliShuttle::SendMLRunInfo(const char* status)
1230 {
1231         //
1232         // Send information about this run to ML
1233
1234         TMonaLisaText  mlStatus("SHUTTLE_status", status);
1235         TString runType(fLogbookEntry->GetRunType());
1236         if (strlen(fLogbookEntry->GetRunParameter("log")) > 0){
1237
1238                 runType += "(";
1239                 runType += fLogbookEntry->GetRunParameter("log");
1240                 runType += ")";
1241         }
1242         if (fLogbookEntry->GetDATestMode()){
1243                 runType += " (DATest)";
1244         }
1245         TMonaLisaText  mlRunType("SHUTTLE_runtype", runType);
1246
1247         TList mlList;
1248         mlList.Add(&mlStatus);
1249         mlList.Add(&mlRunType);
1250
1251         TString mlID;
1252         mlID.Form("%d", GetCurrentRun());
1253         fMonaLisa->SendParameters(&mlList, mlID);
1254 }
1255
1256 //______________________________________________________________________________________________
1257 Int_t AliShuttle::GetMem(Int_t pid)
1258 {
1259         // invokes ps to get the memory consumption of the process <pid>
1260         // returns -1 in case of error
1261
1262         TString checkStr;
1263         checkStr.Form("ps -o vsize --pid %d | tail -n 1", pid);
1264         FILE* pipe = gSystem->OpenPipe(checkStr, "r");
1265         if (!pipe)
1266         {
1267                 Log("SHUTTLE", Form("Process - Error: "
1268                         "Could not open pipe to %s", checkStr.Data()));
1269                 return -1;
1270         }
1271
1272         char buffer[100];
1273         if (!fgets(buffer, 100, pipe))
1274         {
1275                 Log("SHUTTLE", "Process - Error: ps did not return anything");
1276                 gSystem->ClosePipe(pipe);
1277                 return -1;
1278         }
1279         gSystem->ClosePipe(pipe);
1280
1281         //Log("SHUTTLE", Form("ps returned %s", buffer));
1282
1283         Int_t mem = 0;
1284         if ((sscanf(buffer, "%d\n", &mem) != 1) || !mem)
1285         {
1286                 Log("SHUTTLE", "Process - Error: Could not parse output of ps");
1287                 return -1;
1288         }
1289
1290         return mem;
1291 }
1292
1293 //______________________________________________________________________________________________
1294 Bool_t AliShuttle::Process(AliShuttleLogbookEntry* entry)
1295 {
1296         //
1297         // Makes data retrieval for all detectors in the configuration.
1298         // entry: Shuttle logbook entry, contains run paramenters and status of detectors
1299         // (Unprocessed, Inactive, Failed or Done).
1300         // Returns kFALSE in case of error occured and kTRUE otherwise
1301         //
1302
1303         if (!entry) return kFALSE;
1304
1305         fLogbookEntry = entry;
1306
1307         Log("SHUTTLE", Form("\t\t\t^*^*^*^*^*^*^*^*^*^*^*^* run %d: START ^*^*^*^*^*^*^*^*^*^*^*^*",
1308                                         GetCurrentRun()));
1309
1310         CountOpenRuns();
1311
1312         // Send the information to ML
1313         SendMLRunInfo("Processing");
1314
1315         if (fLogbookEntry->IsDone())
1316         {
1317                 Log("SHUTTLE","Process - Shuttle is already DONE. Updating logbook");
1318                 UpdateShuttleLogbook("shuttle_done");
1319                 fLogbookEntry = 0;
1320                 return kTRUE;
1321         }
1322
1323         // read test mode if flag is set
1324         if (fReadTestMode)
1325         {
1326                 fTestMode = kNone;
1327                 TString logEntry(entry->GetRunParameter("log"));
1328                 //printf("log entry = %s\n", logEntry.Data());
1329                 TString searchStr("Testmode: ");
1330                 Int_t pos = logEntry.Index(searchStr.Data());
1331                 //printf("%d\n", pos);
1332                 if (pos >= 0)
1333                 {
1334                         TSubString subStr = logEntry(pos + searchStr.Length(), logEntry.Length());
1335                         //printf("%s\n", subStr.String().Data());
1336                         TString newStr(subStr.Data());
1337                         TObjArray* token = newStr.Tokenize(' ');
1338                         if (token)
1339                         {
1340                                 //token->Print();
1341                                 TObjString* tmpStr = dynamic_cast<TObjString*> (token->First());
1342                                 if (tmpStr)
1343                                 {
1344                                         Int_t testMode = tmpStr->String().Atoi();
1345                                         if (testMode > 0)
1346                                         {
1347                                                 Log("SHUTTLE", Form("Process - Enabling test mode %d", testMode));
1348                                                 SetTestMode((TestMode) testMode);
1349                                         }
1350                                 }
1351                                 delete token;
1352                         }
1353                 }
1354         }
1355
1356         fLogbookEntry->Print("all");
1357
1358         // Initialization
1359         Bool_t hasError = kFALSE;
1360
1361         // Set the CDB and Reference folders according to the year
1362
1363         // build cdb paths (repeat each time, run might be a DATest run)
1364         if (!fLogbookEntry->GetDATestMode()){
1365                 fgkMainCDB.Form("alien://folder=%s%d/OCDB?user=alidaq?cacheFold=/tmp/OCDBCache",
1366                                 fConfig->GetAlienPath(), GetCurrentYear());
1367
1368                 fgkMainRefStorage.Form("alien://folder=%s%d/Reference?user=alidaq?cacheFold=/tmp/OCDBCache",
1369                                        fConfig->GetAlienPath(), GetCurrentYear());
1370         }
1371         else {
1372                 fgkMainCDB.Form("alien://folder=%s%d/DATest/OCDB?user=alidaq?cacheFold=/tmp/OCDBCache",
1373                                 fConfig->GetAlienPath(), GetCurrentYear());
1374
1375                 fgkMainRefStorage.Form("alien://folder=%s%d/DATest/Reference?user=alidaq?cacheFold=/tmp/OCDBCache",
1376                                        fConfig->GetAlienPath(), GetCurrentYear());
1377         }
1378
1379         AliDebug(2,Form("Main CDB storage = %s",fgkMainCDB.Data()));
1380         AliDebug(2,Form("Main Reference storage = %s",fgkMainRefStorage.Data()));
1381
1382         // Loop on detectors in the configuration
1383         TIter iter(fConfig->GetDetectors());
1384         TObjString* aDetector = 0;
1385
1386         Bool_t first = kTRUE;
1387
1388         while ((aDetector = (TObjString*) iter.Next()))
1389         {
1390                 fCurrentDetector = aDetector->String();
1391
1392                 if (ContinueProcessing() == kFALSE)
1393                         continue;
1394
1395                 if (first)
1396                 {
1397                   // only read QueryCDB when needed and only once
1398                   AliCDBStorage *mainCDBSto = AliCDBManager::Instance()->GetStorage(fgkMainCDB);
1399                   if(mainCDBSto) mainCDBSto->QueryCDB(GetCurrentRun());
1400                   AliCDBStorage *mainRefSto = AliCDBManager::Instance()->GetStorage(fgkMainRefStorage);
1401                   if(mainRefSto) mainRefSto->QueryCDB(GetCurrentRun());
1402                   first = kFALSE;
1403                 }
1404
1405                 Log("SHUTTLE", Form("\t\t\t****** run %d - %s: START  ******",
1406                                                 GetCurrentRun(), aDetector->GetName()));
1407
1408                 for(Int_t iSys=0;iSys<3;iSys++) fFXSCalled[iSys]=kFALSE;
1409
1410                 Int_t initialMem = GetMem(getpid());
1411                 Log("SHUTTLE", Form("Memory consumption before forking is %d", initialMem));
1412
1413                 Log(fCurrentDetector.Data(), "Process - Starting processing");
1414
1415                 Int_t pid = fork();
1416
1417                 if (pid < 0)
1418                 {
1419                         Log("SHUTTLE", "Process - ERROR: Forking failed");
1420                 }
1421                 else if (pid > 0)
1422                 {
1423                         // parent
1424                         Log("SHUTTLE", Form("Process - In parent process of %d - %s: Starting monitoring",
1425                                                         GetCurrentRun(), aDetector->GetName()));
1426
1427                         Long_t begin = time(0);
1428
1429                         int status; // to be used with waitpid, on purpose an int (not Int_t)!
1430                         while (waitpid(pid, &status, WNOHANG) == 0)
1431                         {
1432                                 Long_t expiredTime = time(0) - begin;
1433
1434                                 if (expiredTime > fConfig->GetPPTimeOut())
1435                                 {
1436                                         TString logMsg;
1437                                         AliShuttleStatus *currentStatus = ReadShuttleStatus();
1438                                         AliShuttleStatus::Status newStatus = AliShuttleStatus::kInvalid;
1439
1440                                         if (currentStatus->GetStatus() <= AliShuttleStatus::kPPDone)
1441                                         {
1442                                                 // in case pp not yet done set status to kPPTimeOut
1443
1444                                                 logMsg.Form("Process - Process of %s timed out. Run time: %d seconds. Killing...",
1445                                                                 fCurrentDetector.Data(), expiredTime);
1446                                                 newStatus = AliShuttleStatus::kPPTimeOut;
1447                                         }
1448                                         else if (currentStatus->GetStatus() == AliShuttleStatus::kStoreStarted)
1449                                         {
1450                                                 // in case the pp goes in TimeOut while storing the objects in the OCDB
1451                                                 // set status to kStoreError
1452
1453                                                 logMsg.Form("Process - Process of %s timed out while storing the OCDB object. Run time: %d seconds. Killing... and setting status to StoreError.",
1454                                                                 fCurrentDetector.Data(), expiredTime);
1455                                                 newStatus = AliShuttleStatus::kStoreError;
1456                                         }
1457                                         else
1458                                         {
1459                                                 // in other cases don't change the status
1460
1461                                                 logMsg.Form("Process - Process of %s timed out in status = %s. Run time: %d seconds. Killing... without changing the status",
1462                                                                 fCurrentDetector.Data(), currentStatus->GetStatusName(), expiredTime);
1463                                         }
1464
1465                                         Log("SHUTTLE", logMsg);
1466                                         Log(fCurrentDetector, logMsg);
1467
1468                                         kill(pid, 9);
1469
1470                                         if (newStatus != AliShuttleStatus::kInvalid)
1471                                                 UpdateShuttleStatus(newStatus);
1472                                         hasError = kTRUE;
1473
1474                                         gSystem->Sleep(1000);
1475                                 }
1476                                 else
1477                                 {
1478                                         gSystem->Sleep(1000);
1479
1480                                         Int_t mem = GetMem(pid);
1481
1482                                         if (mem < 0)
1483                                                 continue;
1484
1485                                         mem -= initialMem;
1486                                         if (mem < 0)
1487                                                 mem = 0;
1488
1489                                         if (expiredTime % 60 == 0)
1490                                         {
1491                                                 Log("SHUTTLE", Form("Process - %s: Checking process. "
1492                                                         "Run time: %d seconds - Memory consumption: %d KB",
1493                                                         fCurrentDetector.Data(), expiredTime, mem));
1494                                                 SendAlive();
1495                                         }
1496
1497                                         if (mem > fConfig->GetPPMaxMem())
1498                                         {
1499                                                 TString tmp;
1500                                                 tmp.Form("Process - Process exceeds maximum allowed memory "
1501                                                         "(%d KB > %d KB). Killing...",
1502                                                         mem, fConfig->GetPPMaxMem());
1503                                                 Log("SHUTTLE", tmp);
1504                                                 Log(fCurrentDetector, tmp);
1505
1506                                                 kill(pid, 9);
1507
1508                                                 UpdateShuttleStatus(AliShuttleStatus::kPPOutOfMemory);
1509                                                 hasError = kTRUE;
1510
1511                                                 gSystem->Sleep(1000);
1512                                         }
1513                                 }
1514                         }
1515
1516                         Log("SHUTTLE", Form("Process - In parent process of %d - %s: Client has terminated.",
1517                                                                 GetCurrentRun(), aDetector->GetName()));
1518
1519                         if (WIFEXITED(status))
1520                         {
1521                                 Int_t returnCode = WEXITSTATUS(status);
1522
1523                                 Log("SHUTTLE", Form("Process - %s: the return code is %d", fCurrentDetector.Data(),
1524                                                                                 returnCode));
1525
1526                                 if (returnCode == 0) hasError = kTRUE;
1527                         }
1528                 }
1529                 else if (pid == 0)
1530                 {
1531                         // child
1532                         Log("SHUTTLE", Form("Process - In child process of %d - %s", GetCurrentRun(),
1533                                 aDetector->GetName()));
1534
1535                         Log("SHUTTLE", Form("Process - Redirecting output to %s log",fCurrentDetector.Data()));
1536
1537                         if ((freopen(GetLogFileName(fCurrentDetector), "a", stdout)) == 0)
1538                         {
1539                                 Log("SHUTTLE", "Process - Could not freopen stdout");
1540                         }
1541                         else
1542                         {
1543                                 fOutputRedirected = kTRUE;
1544                                 if ((dup2(fileno(stdout), fileno(stderr))) < 0)
1545                                         Log("SHUTTLE", "Process - Could not redirect stderr");
1546
1547                         }
1548
1549                         TString wd = gSystem->WorkingDirectory();
1550                         Int_t dir_lev1 = GetCurrentRun()/10000;
1551                         TString tmpDir = Form("%s/%d/%d/%s_process", GetShuttleTempDir(),
1552                                 dir_lev1, GetCurrentRun(), fCurrentDetector.Data());
1553
1554                         Int_t result = gSystem->GetPathInfo(tmpDir.Data(), 0, (Long64_t*) 0, 0, 0);
1555                         if (!result) // temp dir already exists!
1556                         {
1557                                 Log(fCurrentDetector.Data(),
1558                                         Form("Process - %s dir already exists! Removing...", tmpDir.Data()));
1559                                 gSystem->Exec(Form("rm -rf %s",tmpDir.Data()));
1560                         }
1561
1562                         if (gSystem->mkdir(tmpDir.Data(), 1))
1563                         {
1564                                 Log(fCurrentDetector.Data(), "Process - could not make temp directory!!");
1565                                 gSystem->Exit(1);
1566                         }
1567
1568                         if (!gSystem->ChangeDirectory(tmpDir.Data()))
1569                         {
1570                                 Log(fCurrentDetector.Data(), "Process - could not change directory!!");
1571                                 gSystem->Exit(1);
1572                         }
1573
1574                         Int_t success = ProcessCurrentDetector();
1575
1576                         gSystem->ChangeDirectory(wd.Data());
1577
1578                         if (success == 1) // Preprocessor finished successfully!
1579                         {
1580                                 // remove temporary folder or DCS map
1581                                 if (!fConfig->KeepTempFolder())
1582                                 {
1583                                         gSystem->Exec(Form("rm -rf %s",tmpDir.Data()));
1584                                 } else if (!fConfig->KeepDCSMap())
1585                                 {
1586                                         gSystem->Exec(Form("rm -f %s/DCSMap.root",tmpDir.Data()));
1587                                 }
1588
1589                                 // Update time_processed field in FXS DB
1590                                 if (UpdateTable() == kFALSE)
1591                                         Log("SHUTTLE", Form("Process - %s: Could not update FXS databases!",
1592                                                         fCurrentDetector.Data()));
1593
1594                                 // Transfer the data from local storage to main storage (Grid)
1595                                 if (StoreOCDB() == kFALSE)
1596                                         success = kFALSE;
1597                         }
1598                         else if (success == 0)
1599                         {
1600                                 Log("SHUTTLE",
1601                                         Form("\t\t\t****** run %d - %s: PP ERROR ******",
1602                                                 GetCurrentRun(), aDetector->GetName()));
1603                         }
1604
1605                         for (UInt_t iSys=0; iSys<3; iSys++)
1606                         {
1607                                 if (fFXSCalled[iSys]) fFXSlist[iSys].Clear();
1608                         }
1609
1610                         Log("SHUTTLE", Form("Process - Client process of %d - %s is exiting now with %d.",
1611                                                         GetCurrentRun(), aDetector->GetName(), success));
1612
1613                         // the client exits here
1614                         gSystem->Exit(success);
1615
1616                         AliError("We should never get here!!!");
1617                 }
1618         }
1619
1620         Log("SHUTTLE", Form("\t\t\t^*^*^*^*^*^*^*^*^*^*^*^* run %d: FINISH ^*^*^*^*^*^*^*^*^*^*^*^*",
1621                                                         GetCurrentRun()));
1622
1623         //check if shuttle is done for this run, if so update logbook
1624         TObjArray checkEntryArray;
1625         checkEntryArray.SetOwner(1);
1626         TString whereClause = Form("where run=%d", GetCurrentRun());
1627         if (!QueryShuttleLogbook(whereClause.Data(), checkEntryArray) ||
1628                         checkEntryArray.GetEntries() == 0) {
1629                 Log("SHUTTLE", Form("Process - Warning: Cannot check status of run %d on Shuttle logbook!",
1630                                                 GetCurrentRun()));
1631                 return hasError == kFALSE;
1632         }
1633
1634         AliShuttleLogbookEntry* checkEntry = dynamic_cast<AliShuttleLogbookEntry*>
1635                                                 (checkEntryArray.At(0));
1636
1637         if (checkEntry)
1638         {
1639                 if (checkEntry->IsDone())
1640                 {
1641                         Log("SHUTTLE","Process - Shuttle is DONE. Updating logbook");
1642                         UpdateShuttleLogbook("shuttle_done");
1643                 }
1644                 else
1645                 {
1646                         for (UInt_t iDet=0; iDet<NDetectors(); iDet++)
1647                         {
1648                                 if (checkEntry->GetDetectorStatus(iDet) == AliShuttleLogbookEntry::kUnprocessed)
1649                                 {
1650                                         AliDebug(2, Form("Run %d: setting %s as \"not first time unprocessed\"",
1651                                                         checkEntry->GetRun(), GetDetName(iDet)));
1652                                         fFirstUnprocessed[iDet] = kFALSE;
1653                                 }
1654                         }
1655                         SendMLRunInfo("Pending");
1656                 }
1657         }
1658
1659         fLogbookEntry = 0;
1660
1661         return hasError == kFALSE;
1662 }
1663
1664 //______________________________________________________________________________________________
1665 Int_t AliShuttle::ProcessCurrentDetector()
1666 {
1667         //
1668         // Makes data retrieval just for a specific detector (fCurrentDetector).
1669         // Threre should be a configuration for this detector.
1670
1671         Log("SHUTTLE", Form("ProcessCurrentDetector - Retrieving values for %s, run %d",
1672                                                 fCurrentDetector.Data(), GetCurrentRun()));
1673
1674         TString wd = gSystem->WorkingDirectory();
1675
1676         if (!CleanReferenceStorage(fCurrentDetector.Data()))
1677                 return 0;
1678
1679         gSystem->ChangeDirectory(wd.Data());
1680
1681         // call preprocessor
1682         AliPreprocessor* aPreprocessor =
1683                 dynamic_cast<AliPreprocessor*> (fPreprocessorMap.GetValue(fCurrentDetector));
1684
1685         // check if the preprocessor wants to process this run type
1686         if (aPreprocessor->ProcessRunType() == kFALSE)
1687         {
1688                 UpdateShuttleStatus(AliShuttleStatus::kSkipped);
1689                 UpdateShuttleLogbook(fCurrentDetector, "DONE");
1690                 if (!UpdateTableSkippedCase(fCurrentDetector.Data()))
1691                 {
1692                         AliError(Form("Could not update FXS tables for run %d !", GetCurrentRun()));
1693                 }
1694                 Log(fCurrentDetector, Form("ProcessCurrentDetector - %s preprocessor is not interested in this run type", fCurrentDetector.Data()));
1695
1696                 return 2;
1697         }
1698
1699         TMap* dcsMap = new TMap();
1700
1701         aPreprocessor->Initialize(GetCurrentRun(), GetCurrentStartTime(), GetCurrentEndTime());
1702
1703         Bool_t processDCS = aPreprocessor->ProcessDCS();
1704
1705         if (!processDCS)
1706         {
1707                 Log(fCurrentDetector, "ProcessCurrentDetector -"
1708                         " The preprocessor requested to skip the retrieval of DCS values");
1709         }
1710         else if (fTestMode & kSkipDCS)
1711         {
1712                 Log(fCurrentDetector, "ProcessCurrentDetector - In TESTMODE: Skipping DCS processing");
1713         }
1714         else if (fTestMode & kErrorDCS)
1715         {
1716                 Log(fCurrentDetector, "ProcessCurrentDetector - In TESTMODE: Simulating DCS error");
1717                 UpdateShuttleStatus(AliShuttleStatus::kDCSStarted);
1718                 UpdateShuttleStatus(AliShuttleStatus::kDCSError);
1719                 delete dcsMap;
1720                 return 0;
1721         } else {
1722
1723                 UpdateShuttleStatus(AliShuttleStatus::kDCSStarted);
1724
1725                 // Query DCS archive
1726                 Int_t nServers = fConfig->GetNServers(fCurrentDetector);
1727
1728                 for (int iServ=0; iServ<nServers; iServ++)
1729                 {
1730
1731                         TString host(fConfig->GetDCSHost(fCurrentDetector, iServ));
1732                         Int_t port = fConfig->GetDCSPort(fCurrentDetector, iServ);
1733                         Int_t multiSplit = fConfig->GetMultiSplit(fCurrentDetector, iServ);
1734
1735                         Log(fCurrentDetector, Form("ProcessCurrentDetector -"
1736                                         " Querying DCS Amanda server %s:%d (%d of %d)",
1737                                         host.Data(), port, iServ+1, nServers));
1738
1739                         TMap* aliasMap = 0;
1740                         TMap* dpMap = 0;
1741
1742                         if (fConfig->GetDCSAliases(fCurrentDetector, iServ)->GetEntries() > 0)
1743                         {
1744                                 Log(fCurrentDetector, Form("Querying %d DCS aliases", fConfig->GetDCSAliases(fCurrentDetector, iServ)->GetEntries()));
1745                                 aliasMap = GetValueSet(host, port,
1746                                                 fConfig->GetDCSAliases(fCurrentDetector, iServ),
1747                                                 kAlias, multiSplit);
1748                                 if (!aliasMap)
1749                                 {
1750                                         Log(fCurrentDetector,
1751                                                 Form("ProcessCurrentDetector -"
1752                                                         " Error retrieving DCS aliases from server %s."
1753                                                         " Sending mail to DCS experts!", host.Data()));
1754                                         UpdateShuttleStatus(AliShuttleStatus::kDCSError);
1755
1756                                         if (!SendMail(kDCSEMail))
1757                                                 Log("SHUTTLE", Form("ProcessCurrentDetector - "
1758                                                                     "Could not send mail to DCS experts!"));
1759
1760                                         delete dcsMap;
1761                                         return 0;
1762                                 }
1763                         }
1764
1765                         if (fConfig->GetDCSDataPoints(fCurrentDetector, iServ)->GetEntries() > 0)
1766                         {
1767                                 Log(fCurrentDetector, Form("Querying %d DCS data points", fConfig->GetDCSDataPoints(fCurrentDetector, iServ)->GetEntries()));
1768                                 dpMap = GetValueSet(host, port,
1769                                                 fConfig->GetDCSDataPoints(fCurrentDetector, iServ),
1770                                                 kDP, multiSplit);
1771                                 if (!dpMap)
1772                                 {
1773                                         Log(fCurrentDetector,
1774                                                 Form("ProcessCurrentDetector -"
1775                                                         " Error retrieving DCS data points from server %s."
1776                                                         " Sending mail to DCS experts!", host.Data()));
1777                                         UpdateShuttleStatus(AliShuttleStatus::kDCSError);
1778
1779                                         if (!SendMail(kDCSEMail))
1780                                                 Log("SHUTTLE", Form("ProcessCurrentDetector - "
1781                                                                     "Could not send mail to DCS experts!"));
1782
1783                                         if (aliasMap) delete aliasMap;
1784                                         delete dcsMap;
1785                                         return 0;
1786                                 }
1787                         }
1788
1789                         // merge aliasMap and dpMap into dcsMap
1790                         if(aliasMap) {
1791                                 TIter iter(aliasMap);
1792                                 TObjString* key = 0;
1793                                 while ((key = (TObjString*) iter.Next()))
1794                                         dcsMap->Add(key, aliasMap->GetValue(key->String()));
1795
1796                                 aliasMap->SetOwner(kFALSE);
1797                                 delete aliasMap;
1798                         }
1799
1800                         if(dpMap) {
1801                                 TIter iter(dpMap);
1802                                 TObjString* key = 0;
1803                                 while ((key = (TObjString*) iter.Next()))
1804                                         dcsMap->Add(key, dpMap->GetValue(key->String()));
1805
1806                                 dpMap->SetOwner(kFALSE);
1807                                 delete dpMap;
1808                         }
1809                 }
1810         }
1811
1812         // save map into file, to help debugging in case of preprocessor error
1813         TFile* f = TFile::Open("DCSMap.root","recreate");
1814         f->cd();
1815         dcsMap->Write("DCSMap", TObject::kSingleKey);
1816         f->Close();
1817         delete f;
1818
1819         // DCS Archive DB processing successful. Call Preprocessor!
1820         UpdateShuttleStatus(AliShuttleStatus::kPPStarted);
1821
1822         fFXSError = -1; // this variable is kTRUE after ::Process if an FXS error occured
1823
1824         UInt_t returnValue = aPreprocessor->Process(dcsMap);
1825
1826         if (fFXSError!=-1) {
1827                 UpdateShuttleStatus(AliShuttleStatus::kFXSError);
1828                 SendMail(kFXSEMail, fFXSError);
1829                 dcsMap->DeleteAll();
1830                 delete dcsMap;
1831                 return 0;
1832         }
1833
1834         if (returnValue > 0) // Preprocessor error!
1835         {
1836                 Log(fCurrentDetector, Form("ProcessCurrentDetector - "
1837                                 "Preprocessor failed. Process returned %d.", returnValue));
1838                 UpdateShuttleStatus(AliShuttleStatus::kPPError);
1839                 dcsMap->DeleteAll();
1840                 delete dcsMap;
1841                 return 0;
1842         }
1843
1844         // preprocessor ok!
1845         UpdateShuttleStatus(AliShuttleStatus::kPPDone);
1846         Log(fCurrentDetector, Form("ProcessCurrentDetector - %s preprocessor returned success",
1847                                 fCurrentDetector.Data()));
1848
1849         dcsMap->DeleteAll();
1850         delete dcsMap;
1851
1852         return 1;
1853 }
1854
1855 //______________________________________________________________________________________________
1856 void AliShuttle::CountOpenRuns()
1857 {
1858         // Query DAQ's Shuttle logbook and sends the number of open runs to ML
1859
1860         SendAlive();
1861
1862         // check connection, in case connect
1863         if (!Connect(3))
1864                 return;
1865
1866         TString sqlQuery;
1867         sqlQuery = Form("select count(*) from %s where shuttle_done=0", fConfig->GetShuttlelbTable());
1868
1869         TSQLResult* aResult = fServer[3]->Query(sqlQuery);
1870         if (!aResult) {
1871                 AliError(Form("Can't execute query <%s>!", sqlQuery.Data()));
1872                 return;
1873         }
1874
1875         AliDebug(2,Form("Query = %s", sqlQuery.Data()));
1876
1877         if (aResult->GetRowCount() == 0) {
1878                 AliError(Form("No result for query %s received", sqlQuery.Data()));
1879                 return;
1880         }
1881
1882         if (aResult->GetFieldCount() != 1) {
1883                 AliError(Form("Invalid field count for query %s received", sqlQuery.Data()));
1884                 return;
1885         }
1886
1887         TSQLRow* aRow = aResult->Next();
1888         if (!aRow) {
1889                 AliError(Form("Could not receive result of query %s", sqlQuery.Data()));
1890                 return;
1891         }
1892
1893         TString result(aRow->GetField(0), aRow->GetFieldLength(0));
1894         Int_t count = result.Atoi();
1895
1896         Log("SHUTTLE", Form("%d unprocessed runs", count));
1897
1898         delete aRow;
1899         delete aResult;
1900
1901         TMonaLisaValue mlStatus("SHUTTLE_openruns", count);
1902
1903         TList mlList;
1904         mlList.Add(&mlStatus);
1905
1906         fMonaLisa->SendParameters(&mlList, "__PROCESSINGINFO__");
1907 }
1908
1909 //______________________________________________________________________________________________
1910 Bool_t AliShuttle::QueryShuttleLogbook(const char* whereClause,
1911                 TObjArray& entries)
1912 {
1913         // Query DAQ's Shuttle logbook and fills detector status object.
1914         // Call QueryRunParameters to query DAQ logbook for run parameters.
1915         //
1916
1917         entries.SetOwner(1);
1918
1919         // check connection, in case connect
1920         if (!Connect(3)) return kFALSE;
1921
1922         TString sqlQuery;
1923         sqlQuery = Form("select * from %s %s order by run", fConfig->GetShuttlelbTable(), whereClause);
1924
1925         TSQLResult* aResult = fServer[3]->Query(sqlQuery);
1926         if (!aResult) {
1927                 AliError(Form("Can't execute query <%s>!", sqlQuery.Data()));
1928                 return kFALSE;
1929         }
1930
1931         AliDebug(2,Form("Query = %s", sqlQuery.Data()));
1932
1933         if(aResult->GetRowCount() == 0) {
1934                 Log("SHUTTLE", "No entries in Shuttle Logbook match request");
1935                 delete aResult;
1936                 return kTRUE;
1937         }
1938
1939         // TODO Check field count!
1940         const UInt_t nCols = 26;
1941         if (aResult->GetFieldCount() != (Int_t) nCols) {
1942                 Log("SHUTTLE", "Invalid SQL result field number!");
1943                 delete aResult;
1944                 return kFALSE;
1945         }
1946
1947         TSQLRow* aRow;
1948         while ((aRow = aResult->Next())) {
1949                 TString runString(aRow->GetField(0), aRow->GetFieldLength(0));
1950                 Int_t run = runString.Atoi();
1951
1952                 AliShuttleLogbookEntry *entry = QueryRunParameters(run);
1953                 if (!entry)
1954                         continue;
1955
1956                 // DA test mode flag
1957                 TString daTestModeString(aRow->GetField(2), aRow->GetFieldLength(2)); // field 2 = DA test mode flag
1958                 Bool_t daTestMode = (Bool_t)daTestModeString.Atoi();
1959                 entry->SetDATestMode(daTestMode);
1960
1961                 // loop on detectors
1962                 for(UInt_t ii = 0; ii < nCols; ii++)
1963                         entry->SetDetectorStatus(aResult->GetFieldName(ii), aRow->GetField(ii));
1964
1965                 entries.AddLast(entry);
1966                 delete aRow;
1967         }
1968
1969         delete aResult;
1970         return kTRUE;
1971 }
1972
1973 //______________________________________________________________________________________________
1974 AliShuttleLogbookEntry* AliShuttle::QueryRunParameters(Int_t run)
1975 {
1976         //
1977         // Retrieve run parameters written in the DAQ logbook and sets them into AliShuttleLogbookEntry object
1978         //
1979
1980         // check connection, in case connect
1981         if (!Connect(3))
1982                 return 0;
1983
1984         TString sqlQuery;
1985         sqlQuery.Form("select * from %s where run=%d", fConfig->GetDAQlbTable(), run);
1986
1987         TSQLResult* aResult = fServer[3]->Query(sqlQuery);
1988         if (!aResult) {
1989                 Log("SHUTTLE", Form("Can't execute query <%s>!", sqlQuery.Data()));
1990                 return 0;
1991         }
1992
1993         if (aResult->GetRowCount() == 0) {
1994                 Log("SHUTTLE", Form("QueryRunParameters - No entry in DAQ Logbook for run %d. Skipping", run));
1995                 delete aResult;
1996                 return 0;
1997         }
1998
1999         if (aResult->GetRowCount() > 1) {
2000                 Log("SHUTTLE", Form("QueryRunParameters - UNEXPECTED: "
2001                                 "more than one entry in DAQ Logbook for run %d!", run));
2002                 delete aResult;
2003                 return 0;
2004         }
2005
2006         TSQLRow* aRow = aResult->Next();
2007         if (!aRow)
2008         {
2009                 Log("SHUTTLE", Form("QueryRunParameters - Could not retrieve row for run %d. Skipping", run));
2010                 delete aResult;
2011                 return 0;
2012         }
2013
2014         AliShuttleLogbookEntry* entry = new AliShuttleLogbookEntry(run);
2015
2016         for (Int_t ii = 0; ii < aResult->GetFieldCount(); ii++)
2017                 entry->SetRunParameter(aResult->GetFieldName(ii), aRow->GetField(ii));
2018
2019         delete aRow;
2020         delete aResult;
2021
2022         UInt_t startTime = entry->GetStartTime();
2023         UInt_t endTime = entry->GetEndTime();
2024         Bool_t ecsSuccess = entry->GetECSSuccess();
2025         TString runType = entry->GetRunType();
2026         TString tmpdaqstartTime = entry->GetRunParameter("DAQ_time_start");
2027         UInt_t daqstartTime = tmpdaqstartTime.Atoi();
2028
2029         UInt_t now = time(0);
2030         Int_t dcsDelay = fConfig->GetDCSDelay()+fConfig->GetDCSQueryOffset();
2031
2032         Bool_t skip = kFALSE;
2033
2034         // runs are processed if
2035         //   a) runType is PHYSICS and ecsSuccess is set
2036         //   b) runType is not PHYSICS and (ecsSuccess is set or DAQ_time_start is non-0)
2037         // effectively this means that all runs are processed that started properly (ecsSucess behaviour is different for PHYSICS and non-PHYSICS runs (check with ECS!)
2038         if (startTime != 0 && endTime != 0 && endTime > startTime) {
2039           if (endTime >= now - dcsDelay) {
2040             Log("SHUTTLE", Form("Skipping run %d for now, because DCS buffer time is not yet expired", run));
2041           } else {
2042             if (runType == "PHYSICS") {
2043               if (ecsSuccess) {
2044                 return entry;
2045               } else {
2046                 Log("SHUTTLE", Form("QueryRunParameters - Run type for run %d is PHYSICS but ECS success flag not set (Reason = %s) - Skipping!", run, entry->GetRunParameter("eor_reason")));
2047                 skip = kTRUE;
2048               }
2049             } else {
2050               if (ecsSuccess || daqstartTime > 0) {
2051                 if (ecsSuccess == kFALSE)
2052                   Log("SHUTTLE", Form("Processing run %d although in status ECS failure (Reason: %s), since run type != PHYSICS and DAQ_time_start != 0", run, entry->GetRunParameter("eor_reason")));
2053                 return entry;
2054               } else {
2055                 Log("SHUTTLE", Form("QueryRunParameters - Run type for run %d is %s, ECS success flag was not set (Reason = %s) and DAQ_time_start was NULL - Skipping!", run, runType.Data(), entry->GetRunParameter("eor_reason")));
2056                 skip = kTRUE;
2057               }
2058             }
2059           }
2060         } else {
2061           Log("SHUTTLE", Form("QueryRunParameters - Invalid parameters for Run %d: "
2062                               "startTime = %d, endTime = %d. Skipping (Shuttle won't be marked as DONE)!",
2063                               run, startTime, endTime));
2064         }
2065
2066         if (skip)
2067         {
2068                 Log("SHUTTLE", Form("Marking SHUTTLE skipped for run %d", run));
2069                 fLogbookEntry = entry;
2070                 if (!UpdateShuttleLogbook("shuttle_skipped"))
2071                 {
2072                         AliError(Form("Could not update logbook for run %d !", run));
2073                 }
2074                 if (!UpdateTableSkippedCase("ALL"))
2075                 {
2076                         AliError(Form("Could not update FXS tables for run %d !", run));
2077                 }
2078                 fLogbookEntry = 0;
2079         }
2080
2081         delete entry;
2082         return 0;
2083 }
2084
2085 //______________________________________________________________________________________________
2086 TMap* AliShuttle::GetValueSet(const char* host, Int_t port, const TSeqCollection* entries,
2087                               DCSType type, Int_t multiSplit)
2088 {
2089         // Retrieve all "entry" data points from the DCS server
2090         // host, port: TSocket connection parameters
2091         // entries: list of name of the alias or data point
2092         // type: kAlias or kDP
2093         // returns TMap of values, 0 when failure
2094
2095         AliDCSClient client(host, port, fTimeout, fRetries, multiSplit);
2096
2097         TMap* result = 0;
2098         if (type == kAlias)
2099         {
2100                 //result = client.GetAliasValues(entries, GetCurrentStartTime()-offset,
2101                 //      GetCurrentEndTime()+offset);
2102                 result = client.GetAliasValues(entries, GetStartTimeDCSQuery(),
2103                         GetEndTimeDCSQuery());
2104         }
2105         else if (type == kDP)
2106         {
2107                 //result = client.GetDPValues(entries, GetCurrentStartTime()-offset,
2108                 //      GetCurrentEndTime()+offset);
2109                 result = client.GetDPValues(entries, GetStartTimeDCSQuery(),
2110                         GetEndTimeDCSQuery());
2111         }
2112
2113         if (result == 0)
2114         {
2115                 Log(fCurrentDetector.Data(), Form("GetValueSet - Can't get entries! Reason: %s",
2116                         client.GetErrorString(client.GetResultErrorCode())));
2117                 if (client.GetResultErrorCode() == AliDCSClient::fgkServerError)
2118                         Log(fCurrentDetector.Data(), Form("GetValueSet - Server error code: %s",
2119                                 client.GetServerError().Data()));
2120
2121                 return 0;
2122         }
2123
2124         return result;
2125 }
2126
2127 //______________________________________________________________________________________________
2128 const char* AliShuttle::GetFile(Int_t system, const char* detector,
2129                 const char* id, const char* source)
2130 {
2131         // Get calibration file from file exchange servers
2132         // First queris the FXS database for the file name, using the run, detector, id and source info
2133         // then calls RetrieveFile(filename) for actual copy to local disk
2134         // run: current run being processed (given by Logbook entry fLogbookEntry)
2135         // detector: the Preprocessor name
2136         // id: provided as a parameter by the Preprocessor
2137         // source: provided by the Preprocessor through GetFileSources function
2138
2139         // check if test mode should simulate a FXS error
2140         if (fTestMode & kErrorFXSFiles)
2141         {
2142                 Log(detector, Form("GetFile - In TESTMODE - Simulating error while connecting to %s FXS", GetSystemName(system)));
2143                 return 0;
2144         }
2145
2146         // check connection, in case connect
2147         if (!Connect(system))
2148         {
2149                 Log(detector, Form("GetFile - Couldn't connect to %s FXS database", GetSystemName(system)));
2150                 fFXSError = system;
2151                 return 0;
2152         }
2153
2154         // Query preparation
2155         TString sourceName(source);
2156         Int_t nFields = 3;
2157         TString sqlQueryStart = Form("select filePath,size,fileChecksum from %s where",
2158                                                                 fConfig->GetFXSdbTable(system));
2159         TString whereClause = Form("run=%d and detector=\"%s\" and fileId=\"%s\"",
2160                                                                 GetCurrentRun(), detector, id);
2161
2162         if (system == kDAQ)
2163         {
2164                 whereClause += Form(" and DAQsource=\"%s\"", source);
2165         }
2166         else if (system == kDCS)
2167         {
2168                 sourceName="none";
2169         }
2170         else if (system == kHLT)
2171         {
2172                 whereClause += Form(" and DDLnumbers=\"%s\"", source);
2173         }
2174
2175         TString sqlQuery = Form("%s %s", sqlQueryStart.Data(), whereClause.Data());
2176
2177         AliDebug(2, Form("SQL query: \n%s",sqlQuery.Data()));
2178
2179         // Query execution
2180         TSQLResult* aResult = 0;
2181         aResult = dynamic_cast<TSQLResult*> (fServer[system]->Query(sqlQuery));
2182         if (!aResult) {
2183                 Log(detector, Form("GetFile - Can't execute SQL query to %s database for: id = %s, source = %s",
2184                                 GetSystemName(system), id, sourceName.Data()));
2185                 fFXSError = system;
2186                 return 0;
2187         }
2188
2189         if (aResult->GetRowCount() == 0)
2190         {
2191                 Log(detector,
2192                         Form("GetFile - No entry in %s FXS db for: id = %s, source = %s",
2193                                 GetSystemName(system), id, sourceName.Data()));
2194                 delete aResult;
2195                 return 0;
2196         }
2197
2198         if (aResult->GetRowCount() > 1) {
2199                 Log(detector,
2200                         Form("GetFile - More than one entry in %s FXS db for: id = %s, source = %s",
2201                                 GetSystemName(system), id, sourceName.Data()));
2202                 fFXSError = system;
2203                 delete aResult;
2204                 return 0;
2205         }
2206
2207         if (aResult->GetFieldCount() != nFields) {
2208                 Log(detector,
2209                         Form("GetFileName - Wrong field count in %s FXS db for: id = %s, source = %s",
2210                                 GetSystemName(system), id, sourceName.Data()));
2211                 fFXSError = system;
2212                 delete aResult;
2213                 return 0;
2214         }
2215
2216         TSQLRow* aRow = dynamic_cast<TSQLRow*> (aResult->Next());
2217
2218         if (!aRow){
2219                 Log(detector, Form("GetFile - Empty set result in %s FXS db from query: id = %s, source = %s",
2220                                 GetSystemName(system), id, sourceName.Data()));
2221                 fFXSError = system;
2222                 delete aResult;
2223                 return 0;
2224         }
2225
2226         TString filePath(aRow->GetField(0), aRow->GetFieldLength(0));
2227         TString fileSize(aRow->GetField(1), aRow->GetFieldLength(1));
2228         TString fileChecksum(aRow->GetField(2), aRow->GetFieldLength(2));
2229
2230         delete aResult;
2231         delete aRow;
2232
2233         AliDebug(2, Form("filePath = %s; size = %s, fileChecksum = %s",
2234                                 filePath.Data(), fileSize.Data(), fileChecksum.Data()));
2235
2236         // retrieved file is renamed to make it unique
2237         Int_t dir_lev1 = GetCurrentRun()/10000;
2238         TString localFileName = Form("%s/%d/%d/%s_process/%s_%s_%d_%s_%s.shuttle",
2239                                      GetShuttleTempDir(), dir_lev1, GetCurrentRun(), detector,
2240                                         GetSystemName(system), detector, GetCurrentRun(),
2241                                         id, sourceName.Data());
2242         Log("SHUTTLE",Form("file from FXS = %s",localFileName.Data()));
2243
2244
2245         // file retrieval from FXS
2246         UInt_t nRetries = 0;
2247         UInt_t maxRetries = 3;
2248         Bool_t result = kFALSE;
2249
2250         // copy!! if successful TSystem::Exec returns 0
2251         while (nRetries++ < maxRetries) {
2252                 AliDebug(2, Form("Trying to copy file. Retry # %d", nRetries));
2253                 result = RetrieveFile(system, filePath.Data(), localFileName.Data());
2254                 if (!result)
2255                 {
2256                         Log(detector, Form("GetFile - Copy of file %s from %s FXS failed",
2257                                         filePath.Data(), GetSystemName(system)));
2258                         continue;
2259                 }
2260
2261                 if (fileSize.Length()>0)
2262                 {
2263                         // compare filesize of local file with the one stored in the FXS DB
2264                         Long_t size = -1;
2265                         Int_t sizeComp = gSystem->GetPathInfo(localFileName.Data(), 0, &size, 0, 0);
2266
2267                         if (sizeComp != 0 || size != fileSize.Atoi())
2268                         {
2269                                 Log(detector, Form("GetFile - size of file %s does not match with local copy!",
2270                                                         filePath.Data()));
2271                                 result = kFALSE;
2272                                 continue;
2273                         }
2274
2275                 } else {
2276                         Log(fCurrentDetector, Form("GetFile - size of file %s not set in %s database, skipping comparison",
2277                                                 filePath.Data(), GetSystemName(system)));
2278                 }
2279
2280                 if (fileChecksum.Length()>0)
2281                 {
2282                         // compare md5sum of local file with the one stored in the FXS DB
2283                         if(fileChecksum.Contains(' ')) fileChecksum.Resize(fileChecksum.First(' '));
2284                         Int_t md5Comp = gSystem->Exec(Form("md5sum %s |grep %s > /dev/null 2> /dev/null",
2285                                                 localFileName.Data(), fileChecksum.Data()));
2286
2287                         if (md5Comp != 0)
2288                         {
2289                                 Log(detector, Form("GetFile - md5sum of file %s does not match with local copy!",
2290                                                         filePath.Data()));
2291                                 result = kFALSE;
2292                                 continue;
2293                         }
2294                 } else {
2295                         Log(fCurrentDetector, Form("GetFile - md5sum of file %s not set in %s database, skipping comparison",
2296                                                         filePath.Data(), GetSystemName(system)));
2297                 }
2298                 if (result) break;
2299         }
2300
2301         if (!result)
2302         {
2303                 fFXSError = system;
2304                 return 0;
2305         }
2306
2307         fFXSCalled[system]=kTRUE;
2308         TObjString *fileParams = new TObjString(Form("%s#!?!#%s", id, sourceName.Data()));
2309         fFXSlist[system].Add(fileParams);
2310
2311         static TString staticLocalFileName;
2312         staticLocalFileName.Form("%s", localFileName.Data());
2313
2314         Log(fCurrentDetector, Form("GetFile - Retrieved file with id %s and "
2315                         "source %s from %s to %s", id, source,
2316                         GetSystemName(system), localFileName.Data()));
2317
2318         return staticLocalFileName.Data();
2319 }
2320
2321 //______________________________________________________________________________________________
2322 Bool_t AliShuttle::RetrieveFile(UInt_t system, const char* fxsFileName, const char* localFileName)
2323 {
2324         //
2325         // Copies file from FXS to local Shuttle machine
2326         //
2327
2328         // check temp directory: trying to cd to temp; if it does not exist, create it
2329         AliDebug(2, Form("Copy file %s from %s FXS into %s",
2330                         GetSystemName(system), fxsFileName, localFileName));
2331
2332         TString tmpDir(localFileName);
2333
2334         tmpDir = tmpDir(0,tmpDir.Last('/'));
2335
2336         Int_t noDir = gSystem->GetPathInfo(tmpDir.Data(), 0, (Long64_t*) 0, 0, 0);
2337         if (noDir) // temp dir does not exists!
2338         {
2339                 if (gSystem->mkdir(tmpDir.Data(), 1))
2340                 {
2341                         Log(fCurrentDetector.Data(), "RetrieveFile - could not make temp directory!!");
2342                         return kFALSE;
2343                 }
2344         }
2345
2346         TString command = Form("scp -oPort=%d -2 %s@%s:%s/%s %s",
2347                 fConfig->GetFXSPort(system),
2348                 fConfig->GetFXSUser(system),
2349                 fConfig->GetFXSHost(system),
2350                 fConfig->GetFXSBaseFolder(system),
2351                 fxsFileName,
2352                 localFileName);
2353
2354         AliDebug(2, Form("%s",command.Data()));
2355
2356         Bool_t result = (gSystem->Exec(command.Data()) == 0);
2357
2358         return result;
2359 }
2360
2361 //______________________________________________________________________________________________
2362 TList* AliShuttle::GetFileSources(Int_t system, const char* detector, const char* id)
2363 {
2364         //
2365         // Get sources producing the condition file Id from file exchange servers
2366         // if id is NULL all sources are returned (distinct)
2367         //
2368
2369         if (id)
2370         {
2371                 Log(detector, Form("GetFileSources - Querying %s FXS for files with id %s produced by %s", GetSystemName(system), id, detector));
2372         } else {
2373                 Log(detector, Form("GetFileSources - Querying %s FXS for files produced by %s", GetSystemName(system), detector));
2374         }
2375
2376         // check if test mode should simulate a FXS error
2377         if (fTestMode & kErrorFXSSources)
2378         {
2379                 Log(detector, Form("GetFileSources - In TESTMODE - Simulating error while connecting to %s FXS", GetSystemName(system)));
2380                 return 0;
2381         }
2382
2383         if (system == kDCS)
2384         {
2385                 Log(detector, "GetFileSources - WARNING: DCS system has only one source of data!");
2386                 TList *list = new TList();
2387                 list->SetOwner(1);
2388                 list->Add(new TObjString(" "));
2389                 return list;
2390         }
2391
2392         // check connection, in case connect
2393         if (!Connect(system))
2394         {
2395                 Log(detector, Form("GetFileSources - Couldn't connect to %s FXS database", GetSystemName(system)));
2396                 fFXSError = system;
2397                 return NULL;
2398         }
2399
2400         TString sourceName = "";
2401         if (system == kDAQ)
2402         {
2403                 sourceName = "DAQsource";
2404         } else if (system == kHLT)
2405         {
2406                 sourceName = "DDLnumbers";
2407         }
2408
2409         TString sqlQueryStart = Form("select distinct %s from %s where", sourceName.Data(), fConfig->GetFXSdbTable(system));
2410         TString whereClause = Form("run=%d and detector=\"%s\"",
2411                                 GetCurrentRun(), detector);
2412         if (id)
2413                 whereClause += Form(" and fileId=\"%s\"", id);
2414         TString sqlQuery = Form("%s %s", sqlQueryStart.Data(), whereClause.Data());
2415
2416         AliDebug(2, Form("SQL query: \n%s",sqlQuery.Data()));
2417
2418         // Query execution
2419         TSQLResult* aResult;
2420         aResult = fServer[system]->Query(sqlQuery);
2421         if (!aResult) {
2422                 Log(detector, Form("GetFileSources - Can't execute SQL query to %s database for id: %s",
2423                                 GetSystemName(system), id));
2424                 fFXSError = system;
2425                 return 0;
2426         }
2427
2428         TList *list = new TList();
2429         list->SetOwner(1);
2430
2431         if (aResult->GetRowCount() == 0)
2432         {
2433                 Log(detector,
2434                         Form("GetFileSources - No entry in %s FXS table for id: %s", GetSystemName(system), id));
2435                 delete aResult;
2436                 return list;
2437         }
2438
2439         Log(detector, Form("GetFileSources - Found %d sources", aResult->GetRowCount()));
2440
2441         TSQLRow* aRow;
2442         while ((aRow = aResult->Next()))
2443         {
2444
2445                 TString source(aRow->GetField(0), aRow->GetFieldLength(0));
2446                 AliDebug(2, Form("%s = %s", sourceName.Data(), source.Data()));
2447                 list->Add(new TObjString(source));
2448                 delete aRow;
2449         }
2450
2451         delete aResult;
2452
2453         return list;
2454 }
2455
2456 //______________________________________________________________________________________________
2457 TList* AliShuttle::GetFileIDs(Int_t system, const char* detector, const char* source)
2458 {
2459         //
2460         // Get all ids of condition files produced by a given source from file exchange servers
2461         //
2462
2463         Log(detector, Form("GetFileIDs - Retrieving ids with source %s with %s", source, GetSystemName(system)));
2464
2465         // check if test mode should simulate a FXS error
2466         if (fTestMode & kErrorFXSSources)
2467         {
2468                 Log(detector, Form("GetFileIDs - In TESTMODE - Simulating error while connecting to %s FXS", GetSystemName(system)));
2469                 return 0;
2470         }
2471
2472         // check connection, in case connect
2473         if (!Connect(system))
2474         {
2475                 Log(detector, Form("GetFileIDs - Couldn't connect to %s FXS database", GetSystemName(system)));
2476                 return NULL;
2477         }
2478
2479         TString sourceName = "";
2480         if (system == kDAQ)
2481         {
2482                 sourceName = "DAQsource";
2483         } else if (system == kHLT)
2484         {
2485                 sourceName = "DDLnumbers";
2486         }
2487
2488         TString sqlQueryStart = Form("select fileId from %s where", fConfig->GetFXSdbTable(system));
2489         TString whereClause = Form("run=%d and detector=\"%s\"",
2490                                 GetCurrentRun(), detector);
2491         if (sourceName.Length() > 0 && source)
2492                 whereClause += Form(" and %s=\"%s\"", sourceName.Data(), source);
2493         TString sqlQuery = Form("%s %s", sqlQueryStart.Data(), whereClause.Data());
2494
2495         AliDebug(2, Form("SQL query: \n%s",sqlQuery.Data()));
2496
2497         // Query execution
2498         TSQLResult* aResult;
2499         aResult = fServer[system]->Query(sqlQuery);
2500         if (!aResult) {
2501                 Log(detector, Form("GetFileIDs - Can't execute SQL query to %s database for source: %s",
2502                                 GetSystemName(system), source));
2503                 return 0;
2504         }
2505
2506         TList *list = new TList();
2507         list->SetOwner(1);
2508
2509         if (aResult->GetRowCount() == 0)
2510         {
2511                 Log(detector,
2512                         Form("GetFileIDs - No entry in %s FXS table for source: %s", GetSystemName(system), source));
2513                 delete aResult;
2514                 return list;
2515         }
2516
2517         Log(detector, Form("GetFileIDs - Found %d ids", aResult->GetRowCount()));
2518
2519         TSQLRow* aRow;
2520
2521         while ((aRow = aResult->Next()))
2522         {
2523
2524                 TString id(aRow->GetField(0), aRow->GetFieldLength(0));
2525                 AliDebug(2, Form("fileId = %s", id.Data()));
2526                 list->Add(new TObjString(id));
2527                 delete aRow;
2528         }
2529
2530         delete aResult;
2531
2532         return list;
2533 }
2534
2535 //______________________________________________________________________________________________
2536 Bool_t AliShuttle::Connect(Int_t system)
2537 {
2538         // Connect to MySQL Server of the system's FXS MySQL databases
2539         // DAQ Logbook, Shuttle Logbook and DAQ FXS db are on the same host
2540         //
2541
2542         // check connection: if already connected return
2543
2544         if(fServer[system] && fServer[system]->IsConnected()) {
2545                 // ping the server
2546                 if (fServer[system]->PingVerify()==kTRUE){ // connection is still alive
2547                         return kTRUE;
2548                 }
2549                 else{
2550                         AliWarning(Form("Connection got lost to FXS database for %s. Closing and reconnecting.",
2551                                         AliShuttleInterface::GetSystemName(system)));
2552                         fServer[system]->Close();
2553                         delete fServer[system];
2554                         fServer[system] = 0x0;
2555                 }
2556         }
2557
2558         TString dbHost, dbUser, dbPass, dbName;
2559
2560         if (system < 3) // FXS db servers
2561         {
2562                 dbHost = Form("mysql://%s:%d", fConfig->GetFXSdbHost(system), fConfig->GetFXSdbPort(system));
2563                 dbUser = fConfig->GetFXSdbUser(system);
2564                 dbPass = fConfig->GetFXSdbPass(system);
2565                 dbName =   fConfig->GetFXSdbName(system);
2566         } else { // Run & Shuttle logbook servers
2567         // TODO Will the Shuttle logbook server be the same as the Run logbook server ???
2568                 dbHost = Form("mysql://%s:%d", fConfig->GetDAQlbHost(), fConfig->GetDAQlbPort());
2569                 dbUser = fConfig->GetDAQlbUser();
2570                 dbPass = fConfig->GetDAQlbPass();
2571                 dbName =   fConfig->GetDAQlbDB();
2572         }
2573
2574         fServer[system] = TSQLServer::Connect(dbHost.Data(), dbUser.Data(), dbPass.Data());
2575         if (!fServer[system] || !fServer[system]->IsConnected()) {
2576                 if(system < 3)
2577                 {
2578                 AliError(Form("Can't establish connection to FXS database for %s",
2579                                         AliShuttleInterface::GetSystemName(system)));
2580                 } else {
2581                 AliError("Can't establish connection to Run logbook.");
2582                 }
2583                 if(fServer[system]) delete fServer[system];
2584                 return kFALSE;
2585         }
2586
2587         // Get tables
2588         TSQLResult* aResult=0;
2589         switch(system){
2590                 case kDAQ:
2591                         aResult = fServer[kDAQ]->GetTables(dbName.Data());
2592                         break;
2593                 case kDCS:
2594                         aResult = fServer[kDCS]->GetTables(dbName.Data());
2595                         break;
2596                 case kHLT:
2597                         aResult = fServer[kHLT]->GetTables(dbName.Data());
2598                         break;
2599                 default:
2600                         aResult = fServer[3]->GetTables(dbName.Data());
2601                         break;
2602         }
2603
2604         delete aResult;
2605         return kTRUE;
2606 }
2607
2608 //______________________________________________________________________________________________
2609 Bool_t AliShuttle::UpdateTable()
2610 {
2611         //
2612         // Update FXS table filling time_processed field in all rows corresponding to current run and detector
2613         //
2614
2615         Bool_t result = kTRUE;
2616
2617         for (UInt_t system=0; system<3; system++)
2618         {
2619                 if(!fFXSCalled[system]) continue;
2620
2621                 // check connection, in case connect
2622                 if (!Connect(system))
2623                 {
2624                         Log(fCurrentDetector, Form("UpdateTable - Couldn't connect to %s FXS database", GetSystemName(system)));
2625                         result = kFALSE;
2626                         continue;
2627                 }
2628
2629                 TTimeStamp now; // now
2630
2631                 // Loop on FXS list entries
2632                 TIter iter(&fFXSlist[system]);
2633                 TObjString *aFXSentry=0;
2634                 while ((aFXSentry = dynamic_cast<TObjString*> (iter.Next())))
2635                 {
2636                         TString aFXSentrystr = aFXSentry->String();
2637                         TObjArray *aFXSarray = aFXSentrystr.Tokenize("#!?!#");
2638                         if (!aFXSarray || aFXSarray->GetEntries() != 2 )
2639                         {
2640                                 Log(fCurrentDetector, Form("UpdateTable - error updating %s FXS entry. Check string: <%s>",
2641                                         GetSystemName(system), aFXSentrystr.Data()));
2642                                 if(aFXSarray) delete aFXSarray;
2643                                 result = kFALSE;
2644                                 continue;
2645                         }
2646                         const char* fileId = ((TObjString*) aFXSarray->At(0))->GetName();
2647                         const char* source = ((TObjString*) aFXSarray->At(1))->GetName();
2648
2649                         TString whereClause;
2650                         if (system == kDAQ)
2651                         {
2652                                 whereClause = Form("where run=%d and detector=\"%s\" and fileId=\"%s\" and DAQsource=\"%s\";",
2653                                                         GetCurrentRun(), fCurrentDetector.Data(), fileId, source);
2654                         }
2655                         else if (system == kDCS)
2656                         {
2657                                 whereClause = Form("where run=%d and detector=\"%s\" and fileId=\"%s\";",
2658                                                         GetCurrentRun(), fCurrentDetector.Data(), fileId);
2659                         }
2660                         else if (system == kHLT)
2661                         {
2662                                 whereClause = Form("where run=%d and detector=\"%s\" and fileId=\"%s\" and DDLnumbers=\"%s\";",
2663                                                         GetCurrentRun(), fCurrentDetector.Data(), fileId, source);
2664                         }
2665
2666                         delete aFXSarray;
2667
2668                         TString sqlQuery = Form("update %s set time_processed=%d %s", fConfig->GetFXSdbTable(system),
2669                                                                 now.GetSec(), whereClause.Data());
2670
2671                         AliDebug(2, Form("SQL query: \n%s",sqlQuery.Data()));
2672
2673                         // Query execution
2674                         TSQLResult* aResult;
2675                         aResult = dynamic_cast<TSQLResult*> (fServer[system]->Query(sqlQuery));
2676                         if (!aResult)
2677                         {
2678                                 Log(fCurrentDetector, Form("UpdateTable - %s db: can't execute SQL query <%s>",
2679                                                                 GetSystemName(system), sqlQuery.Data()));
2680                                 result = kFALSE;
2681                                 continue;
2682                         }
2683                         delete aResult;
2684                 }
2685         }
2686
2687         return result;
2688 }
2689
2690 //_______________________________________________________________________________
2691 Bool_t AliShuttle::UpdateTableSkippedCase(const char* detector)
2692 {
2693         //
2694         // Update FXS table filling time_processed field in all rows corresponding to current run and detector
2695         // if detector = "ALL" update all detectors
2696         //
2697
2698         Bool_t result = kTRUE;
2699
2700         TString detName(detector);
2701
2702         for (UInt_t system=0; system<3; system++)
2703         {
2704
2705                 // check connection, in case connect
2706                 if (!Connect(system))
2707                 {
2708                         Log(fCurrentDetector, Form("UpdateTableSkippedCase - Couldn't connect to %s FXS database", GetSystemName(system)));
2709                         result = kFALSE;
2710                         continue;
2711                 }
2712
2713                 TTimeStamp now; // now
2714
2715                 // Loop on FXS list entries
2716                 TIter iter(&fFXSlist[system]);
2717
2718                 TString whereClause;
2719                 if (detName == "ALL") whereClause = Form("where run=%d and time_processed IS NULL;",GetCurrentRun());
2720                 else whereClause = Form("where run=%d and detector=\"%s\" and time_processed IS NULL;",GetCurrentRun(), detector);
2721
2722                 //Log("SHUTTLE",Form(" whereClause = %s ",whereClause.Data()));
2723
2724                 TString sqlQuery = Form("update %s set time_processed=%d %s", fConfig->GetFXSdbTable(system),
2725                                         now.GetSec(), whereClause.Data());
2726
2727                 AliDebug(2, Form("SQL query: \n%s",sqlQuery.Data()));
2728
2729                 // Query execution
2730                 TSQLResult* aResult;
2731                 aResult = dynamic_cast<TSQLResult*> (fServer[system]->Query(sqlQuery));
2732                 if (!aResult)
2733                 {
2734                         Log("SHUTTLE", Form("UpdateTableSkippedCase - %s db: can't execute SQL query <%s>",
2735                                                         GetSystemName(system), sqlQuery.Data()));
2736                         result = kFALSE;
2737                         continue;
2738                 }
2739                 delete aResult;
2740
2741         }
2742
2743         return result;
2744 }
2745 //______________________________________________________________________________________________
2746 Bool_t AliShuttle::UpdateTableFailCase()
2747 {
2748         // Update FXS table filling time_processed field in all rows corresponding to current run and detector
2749         // this is called in case the preprocessor is declared failed for the current run, because
2750         // the fields are updated only in case of success
2751
2752         Bool_t result = kTRUE;
2753
2754         for (UInt_t system=0; system<3; system++)
2755         {
2756                 // check connection, in case connect
2757                 if (!Connect(system))
2758                 {
2759                         Log(fCurrentDetector, Form("UpdateTableFailCase - Couldn't connect to %s FXS database",
2760                                                         GetSystemName(system)));
2761                         result = kFALSE;
2762                         continue;
2763                 }
2764
2765                 TTimeStamp now; // now
2766
2767                 // Loop on FXS list entries
2768
2769                 TString whereClause = Form("where run=%d and detector=\"%s\";",
2770                                                 GetCurrentRun(), fCurrentDetector.Data());
2771
2772
2773                 TString sqlQuery = Form("update %s set time_processed=%d %s", fConfig->GetFXSdbTable(system),
2774                                                         now.GetSec(), whereClause.Data());
2775
2776                 AliDebug(2, Form("SQL query: \n%s",sqlQuery.Data()));
2777
2778                 // Query execution
2779                 TSQLResult* aResult;
2780                 aResult = dynamic_cast<TSQLResult*> (fServer[system]->Query(sqlQuery));
2781                 if (!aResult)
2782                 {
2783                         Log(fCurrentDetector, Form("UpdateTableFailCase - %s db: can't execute SQL query <%s>",
2784                                                         GetSystemName(system), sqlQuery.Data()));
2785                         result = kFALSE;
2786                         continue;
2787                 }
2788                 delete aResult;
2789         }
2790
2791         return result;
2792 }
2793
2794 //______________________________________________________________________________________________
2795 Bool_t AliShuttle::UpdateShuttleLogbook(const char* detector, const char* status)
2796 {
2797         //
2798         // Update Shuttle logbook filling detector or shuttle_done column
2799         // ex. of usage: UpdateShuttleLogbook("PHOS", "DONE") or UpdateShuttleLogbook("shuttle_done")
2800         //
2801
2802         // check connection, in case connect
2803         if(!Connect(3)){
2804                 Log("SHUTTLE", "UpdateShuttleLogbook - Couldn't connect to DAQ Logbook.");
2805                 return kFALSE;
2806         }
2807
2808         TString detName(detector);
2809         TString setClause;
2810         if (detName == "shuttle_done" || detName == "shuttle_skipped")
2811         {
2812                 setClause = "set shuttle_done=1";
2813
2814                 if (detName == "shuttle_done")
2815                 {
2816                         if (TouchFile() != kTRUE)
2817                         {
2818                                 SendMLRunInfo("Pending");
2819                                 return kFALSE;
2820                         }
2821
2822                         SendMLRunInfo("Done");
2823                 }
2824                 else
2825                         SendMLRunInfo("Skipped");
2826         }
2827         else {
2828                 TString statusStr(status);
2829                 if(statusStr.Contains("done", TString::kIgnoreCase) ||
2830                    statusStr.Contains("failed", TString::kIgnoreCase)){
2831                         setClause = Form("set %s=\"%s\"", detector, status);
2832                 } else {
2833                         Log("SHUTTLE",
2834                                 Form("UpdateShuttleLogbook - Invalid status <%s> for detector %s",
2835                                         status, detector));
2836                         return kFALSE;
2837                 }
2838         }
2839
2840         TString whereClause = Form("where run=%d", GetCurrentRun());
2841
2842         TString sqlQuery = Form("update %s %s %s",
2843                                         fConfig->GetShuttlelbTable(), setClause.Data(), whereClause.Data());
2844
2845         AliDebug(2, Form("SQL query: \n%s",sqlQuery.Data()));
2846
2847         // Query execution
2848         TSQLResult* aResult;
2849         aResult = dynamic_cast<TSQLResult*> (fServer[3]->Query(sqlQuery));
2850         if (!aResult) {
2851                 Log("SHUTTLE", Form("UpdateShuttleLogbook - Can't execute query <%s>", sqlQuery.Data()));
2852                 return kFALSE;
2853         }
2854         delete aResult;
2855
2856         return kTRUE;
2857 }
2858
2859 //______________________________________________________________________________________________
2860 Int_t AliShuttle::GetCurrentRun() const
2861 {
2862         //
2863         // Get current run from logbook entry
2864         //
2865
2866         return fLogbookEntry ? fLogbookEntry->GetRun() : -1;
2867 }
2868
2869 //______________________________________________________________________________________________
2870 UInt_t AliShuttle::GetCurrentStartTime() const
2871 {
2872         //
2873         // get current start time
2874         //
2875
2876         return fLogbookEntry ? fLogbookEntry->GetStartTime() : 0;
2877 }
2878
2879 //______________________________________________________________________________________________
2880 UInt_t AliShuttle::GetCurrentEndTime() const
2881 {
2882         //
2883         // get current end time from logbook entry
2884         //
2885
2886         return fLogbookEntry ? fLogbookEntry->GetEndTime() : 0;
2887 }
2888 //______________________________________________________________________________________________
2889 UInt_t AliShuttle::GetCurrentYear() const
2890 {
2891         //
2892         // Get current year from logbook entry
2893         //
2894
2895         if (!fLogbookEntry) return 0;
2896
2897         TTimeStamp startTime(GetCurrentStartTime());
2898         TString year =  Form("%d",startTime.GetDate());
2899         year = year(0,4);
2900
2901         return year.Atoi();
2902 }
2903
2904 //______________________________________________________________________________________________
2905 const char* AliShuttle::GetLHCPeriod() const
2906 {
2907         //
2908         // Get current LHC period from logbook entry
2909         //
2910
2911         if (!fLogbookEntry) return 0;
2912
2913         return fLogbookEntry->GetRunParameter("LHCperiod");
2914 }
2915
2916 //______________________________________________________________________________________________
2917 void AliShuttle::Log(const char* detector, const char* message)
2918 {
2919         //
2920         // Fill log string with a message
2921         //
2922
2923         TString logRunDir = GetShuttleLogDir();
2924         if (GetCurrentRun() >=0) {
2925                 Int_t logDir_lev1 = GetCurrentRun()/10000;
2926                 logRunDir += Form("/%d/%d", logDir_lev1, GetCurrentRun());
2927         }
2928         void* dir = gSystem->OpenDirectory(logRunDir.Data());
2929         if (dir == NULL) {
2930                 if (gSystem->mkdir(logRunDir.Data(), kTRUE)) {
2931                         AliError(Form("Can't open directory <%s>", GetShuttleLogDir()));
2932                         return;
2933                 }
2934
2935         } else {
2936                 gSystem->FreeDirectory(dir);
2937         }
2938
2939         TString toLog = Form("%s UTC (%d): %s - ", TTimeStamp(time(0)).AsString("s"), getpid(), detector);
2940         if (GetCurrentRun() >= 0)
2941                 toLog += Form("run %d - ", GetCurrentRun());
2942         toLog += Form("%s", message);
2943
2944         AliInfo(toLog.Data());
2945
2946         // if we redirect the log output already to the file, leave here
2947         if (fOutputRedirected && strcmp(detector, "SHUTTLE") != 0)
2948                 return;
2949
2950         TString fileName = GetLogFileName(detector);
2951
2952         gSystem->ExpandPathName(fileName);
2953
2954         ofstream logFile;
2955         logFile.open(fileName, ofstream::out | ofstream::app);
2956
2957         if (!logFile.is_open()) {
2958                 AliError(Form("Could not open file %s", fileName.Data()));
2959                 return;
2960         }
2961
2962         logFile << toLog.Data() << "\n";
2963
2964         logFile.close();
2965 }
2966
2967 //______________________________________________________________________________________________
2968 TString AliShuttle::GetLogFileName(const char* detector) const
2969 {
2970         //
2971         // returns the name of the log file for a given sub detector
2972         //
2973
2974         TString fileName;
2975
2976         if (GetCurrentRun() >= 0)
2977         {
2978                 Int_t logDir_lev1 = GetCurrentRun()/10000;
2979                 fileName.Form("%s/%d/%d/%s.log", GetShuttleLogDir(), logDir_lev1, GetCurrentRun(),
2980                         detector);
2981         } else {
2982                 fileName.Form("%s/%s.log", GetShuttleLogDir(), detector);
2983         }
2984
2985         return fileName;
2986 }
2987
2988 //______________________________________________________________________________________________
2989 void AliShuttle::SendAlive()
2990 {
2991         // sends alive message to ML
2992
2993         TMonaLisaText mlStatus("SHUTTLE_status", "Alive");
2994
2995         TList mlList;
2996         mlList.Add(&mlStatus);
2997
2998         fMonaLisa->SendParameters(&mlList, "__PROCESSINGINFO__");
2999 }
3000
3001 //______________________________________________________________________________________________
3002 Bool_t AliShuttle::Collect(Int_t run)
3003 {
3004         //
3005         // Collects conditions data for all UNPROCESSED run written to DAQ LogBook in case of run = -1 (default)
3006         // If a dedicated run is given this run is processed
3007         //
3008         // In operational mode, this is the Shuttle function triggered by the EOR signal.
3009         //
3010
3011         if (run == -1)
3012                 Log("SHUTTLE","Collect - Shuttle called. Collecting conditions data for unprocessed runs");
3013         else
3014                 Log("SHUTTLE", Form("Collect - Shuttle called. Collecting conditions data for run %d", run));
3015
3016         SetLastAction("Starting");
3017
3018         // create ML instance
3019         if (!fMonaLisa)
3020                 fMonaLisa = new TMonaLisaWriter(fConfig->GetMonitorHost(), fConfig->GetMonitorTable());
3021
3022         CountOpenRuns();
3023
3024         TString whereClause("where shuttle_done=0");
3025         if (run != -1)
3026                 whereClause += Form(" and run=%d", run);
3027
3028         TObjArray shuttleLogbookEntries;
3029         if (!QueryShuttleLogbook(whereClause, shuttleLogbookEntries))
3030         {
3031                 Log("SHUTTLE", "Collect - Can't retrieve entries from Shuttle logbook");
3032                 return kFALSE;
3033         }
3034
3035         if (shuttleLogbookEntries.GetEntries() == 0)
3036         {
3037                 if (run == -1)
3038                         Log("SHUTTLE","Collect - Found no UNPROCESSED runs in Shuttle logbook");
3039                 else
3040                         Log("SHUTTLE", Form("Collect - Run %d is already DONE "
3041                                                 "or it does not exist in Shuttle logbook", run));
3042                 return kTRUE;
3043         }
3044
3045         for (UInt_t iDet=0; iDet<NDetectors(); iDet++)
3046                 fFirstUnprocessed[iDet] = kTRUE;
3047
3048         if (run != -1)
3049         {
3050                 // query Shuttle logbook for earlier runs, check if some detectors are unprocessed,
3051                 // flag them into fFirstUnprocessed array
3052                 TString whereClauseBis(Form("where shuttle_done=0 and run < %d", run));
3053                 TObjArray tmpLogbookEntries;
3054                 if (!QueryShuttleLogbook(whereClauseBis, tmpLogbookEntries))
3055                 {
3056                         Log("SHUTTLE", "Collect - Can't retrieve entries from Shuttle logbook");
3057                         return kFALSE;
3058                 }
3059
3060                 TIter iter(&tmpLogbookEntries);
3061                 AliShuttleLogbookEntry* anEntry = 0;
3062                 while ((anEntry = dynamic_cast<AliShuttleLogbookEntry*> (iter.Next())))
3063                 {
3064                         for (UInt_t iDet=0; iDet<NDetectors(); iDet++)
3065                         {
3066                                 if (anEntry->GetDetectorStatus(iDet) == AliShuttleLogbookEntry::kUnprocessed)
3067                                 {
3068                                         AliDebug(2, Form("Run %d: setting %s as \"not first time unprocessed\"",
3069                                                         anEntry->GetRun(), GetDetName(iDet)));
3070                                         fFirstUnprocessed[iDet] = kFALSE;
3071                                 }
3072                         }
3073
3074                 }
3075
3076         }
3077
3078         if (!RetrieveConditionsData(shuttleLogbookEntries))
3079         {
3080                 Log("SHUTTLE", "Collect - Process of at least one run failed");
3081                 CountOpenRuns();
3082                 return kFALSE;
3083         }
3084
3085         Log("SHUTTLE", "Collect - Requested run(s) successfully processed");
3086         CountOpenRuns();
3087         return kTRUE;
3088 }
3089
3090 //______________________________________________________________________________________________
3091 Bool_t AliShuttle::RetrieveConditionsData(const TObjArray& dateEntries)
3092 {
3093         //
3094         // Retrieve conditions data for all runs that aren't processed yet
3095         //
3096
3097         Bool_t hasError = kFALSE;
3098
3099         TIter iter(&dateEntries);
3100         AliShuttleLogbookEntry* anEntry;
3101
3102         while ((anEntry = (AliShuttleLogbookEntry*) iter.Next())){
3103                 if (!Process(anEntry)){
3104                         hasError = kTRUE;
3105                 }
3106
3107                 // clean SHUTTLE temp directory
3108                 //TString filename = Form("%s/*.shuttle", GetShuttleTempDir());
3109                 //RemoveFile(filename.Data());
3110         }
3111
3112         return hasError == kFALSE;
3113 }
3114
3115 //______________________________________________________________________________________________
3116 ULong_t AliShuttle::GetTimeOfLastAction() const
3117 {
3118         //
3119         // Gets time of last action
3120         //
3121
3122         ULong_t tmp;
3123
3124         fMonitoringMutex->Lock();
3125
3126         tmp = fLastActionTime;
3127
3128         fMonitoringMutex->UnLock();
3129
3130         return tmp;
3131 }
3132
3133 //______________________________________________________________________________________________
3134 const TString AliShuttle::GetLastAction() const
3135 {
3136         //
3137         // returns a string description of the last action
3138         //
3139
3140         TString tmp;
3141
3142         fMonitoringMutex->Lock();
3143
3144         tmp = fLastAction;
3145
3146         fMonitoringMutex->UnLock();
3147
3148         return tmp;
3149 }
3150
3151 //______________________________________________________________________________________________
3152 void AliShuttle::SetLastAction(const char* action)
3153 {
3154         //
3155         // updates the monitoring variables
3156         //
3157
3158         fMonitoringMutex->Lock();
3159
3160         fLastAction = action;
3161         fLastActionTime = time(0);
3162
3163         fMonitoringMutex->UnLock();
3164 }
3165
3166 //______________________________________________________________________________________________
3167 const char* AliShuttle::GetRunParameter(const char* param)
3168 {
3169         //
3170         // returns run parameter read from DAQ logbook
3171         //
3172
3173         if(!fLogbookEntry) {
3174                 AliError("No logbook entry!");
3175                 return 0;
3176         }
3177
3178         return fLogbookEntry->GetRunParameter(param);
3179 }
3180
3181 //______________________________________________________________________________________________
3182 AliCDBEntry* AliShuttle::GetFromOCDB(const char* detector, const AliCDBPath& path)
3183 {
3184         //
3185         // returns object from OCDB valid for current run
3186         //
3187
3188         if (fTestMode & kErrorOCDB)
3189         {
3190                 Log(detector, "GetFromOCDB - In TESTMODE - Simulating error with OCDB");
3191                 return 0;
3192         }
3193
3194         AliCDBStorage *sto = AliCDBManager::Instance()->GetStorage(fgkMainCDB);
3195         if (!sto)
3196         {
3197                 Log(detector, "GetFromOCDB - Cannot activate main OCDB for query!");
3198                 return 0;
3199         }
3200
3201         return dynamic_cast<AliCDBEntry*> (sto->Get(path, GetCurrentRun()));
3202 }
3203
3204 //______________________________________________________________________________________________
3205 Bool_t AliShuttle::SendMail(EMailTarget target, Int_t system)
3206 {
3207         //
3208         // sends a mail to the subdetector expert in case of preprocessor error
3209         //
3210
3211         if (fTestMode != kNone)
3212                 return kTRUE;
3213
3214         if (!fConfig->SendMail())
3215                 return kTRUE;
3216
3217         if (target == kDCSEMail || target == kFXSEMail) {
3218                 if (!fFirstProcessing)
3219                         return kTRUE;
3220         }
3221
3222         Int_t runMode = (Int_t)fConfig->GetRunMode();
3223         TString tmpStr;
3224         if (runMode == 0) tmpStr = " Nightly Test:";
3225         else tmpStr = " Data Taking:";
3226         void* dir = gSystem->OpenDirectory(GetShuttleLogDir());
3227         if (dir == NULL)
3228         {
3229                 if (gSystem->mkdir(GetShuttleLogDir(), kTRUE))
3230                 {
3231                         Log("SHUTTLE", Form("SendMail - Can't open directory <%s>", GetShuttleLogDir()));
3232                         return kFALSE;
3233                 }
3234
3235         } else {
3236                 gSystem->FreeDirectory(dir);
3237         }
3238
3239         // det experts in to
3240         TString to="";
3241         TIter *iterExperts = 0;
3242         if (target == kDCSEMail) {
3243                 iterExperts = new TIter(fConfig->GetAdmins(AliShuttleConfig::kAmanda));
3244         }
3245         else if (target == kFXSEMail) {
3246                 iterExperts = new TIter(fConfig->GetAdmins(system));
3247         }
3248         if (iterExperts) {
3249                 TObjString *anExpert=0;
3250                 while ((anExpert = (TObjString*) iterExperts->Next()))
3251                 {
3252                         to += Form("%s,", anExpert->GetName());
3253                 }
3254                 delete iterExperts;
3255         }
3256
3257         // add subdetector experts
3258         iterExperts = new TIter(fConfig->GetResponsibles(fCurrentDetector));
3259         TObjString *anExpert=0;
3260         while ((anExpert = (TObjString*) iterExperts->Next()))
3261         {
3262                 to += Form("%s,", anExpert->GetName());
3263         }
3264         delete iterExperts;
3265
3266         if (to.Length() > 0)
3267           to.Remove(to.Length()-1);
3268         AliDebug(2, Form("to: %s",to.Data()));
3269
3270         if (to.IsNull()) {
3271                 Log("SHUTTLE", Form("List of %d responsibles not set!", (Int_t) target));
3272                 return kFALSE;
3273         }
3274
3275         // SHUTTLE responsibles in cc
3276         TString cc="";
3277         TIter iterAdmins(fConfig->GetAdmins(AliShuttleConfig::kGlobal));
3278         TObjString *anAdmin=0;
3279         while ((anAdmin = (TObjString*) iterAdmins.Next()))
3280         {
3281                 cc += Form("%s,", anAdmin->GetName());
3282         }
3283         if (cc.Length() > 0)
3284           cc.Remove(cc.Length()-1);
3285         AliDebug(2, Form("cc: %s",to.Data()));
3286
3287         // mail body
3288         TString bodyFileName;
3289         bodyFileName.Form("%s/mail.body", GetShuttleLogDir());
3290         gSystem->ExpandPathName(bodyFileName);
3291
3292         ofstream mailBody;
3293         mailBody.open(bodyFileName, ofstream::out);
3294
3295         if (!mailBody.is_open())
3296         {
3297                 Log("SHUTTLE", Form("Could not open mail body file %s", bodyFileName.Data()));
3298                 return kFALSE;
3299         }
3300
3301
3302         TString subject;
3303         TString body;
3304
3305         if (target == kDCSEMail){
3306                 subject = Form("%s CRITICAL Retrieval of data points for %s FAILED in run %d !",
3307                                 tmpStr.Data(), fCurrentDetector.Data(), GetCurrentRun());
3308                 AliDebug(2, Form("subject: %s", subject.Data()));
3309
3310                 body = Form("Dear DCS experts, \n\n");
3311                 body += Form("SHUTTLE couldn\'t retrieve the data points for detector %s "
3312                              "in run %d!!\n\n", fCurrentDetector.Data(), GetCurrentRun());
3313         }
3314         else if (target == kFXSEMail){
3315                 subject = Form("%s CRITICAL FXS communication for %s FAILED in run %d !",
3316                                 tmpStr.Data(), fCurrentDetector.Data(), GetCurrentRun());
3317                 AliDebug(2, Form("subject: %s", subject.Data()));
3318                 TString sys;
3319                 if (system == kDAQ) sys="DAQ";
3320                 else if (system == kDCS) sys="DCS";
3321                 else if (system == kHLT) sys="HLT";
3322                 else return kFALSE;
3323                 body = Form("Dear  %s FXS experts, \n\n",sys.Data());
3324                 body += Form("SHUTTLE couldn\'t retrieve data from the FXS for detector %s "
3325                              "in run %d!!\n\n", fCurrentDetector.Data(), GetCurrentRun());
3326                 body += Form("The contacted server was:\nDB: %s\nFXS:%s\n\n", fConfig->GetFXSdbHost(system), fConfig->GetFXSHost(system));
3327         }
3328         else {
3329                 subject = Form("%s %s Shuttle preprocessor FAILED in run %d (run type = %s)!",
3330                                        tmpStr.Data(), fCurrentDetector.Data(), GetCurrentRun(), GetRunType());
3331                 AliDebug(2, Form("subject: %s", subject.Data()));
3332
3333                 body = Form("Dear %s expert(s), \n\n", fCurrentDetector.Data());
3334                 body += Form("SHUTTLE just detected that your preprocessor "
3335                              "failed processing run %d (run type = %s)!!\n\n",
3336                              GetCurrentRun(), GetRunType());
3337         }
3338
3339         body += Form("Please check %s status on the SHUTTLE monitoring page: \n\n",
3340                                 fCurrentDetector.Data());
3341         if (fConfig->GetRunMode() == AliShuttleConfig::kTest)
3342         {
3343                 body += Form("\thttp://pcalimonitor.cern.ch/shuttle.jsp?time=24 \n\n");
3344         } else {
3345                 body += Form("\thttp://pcalimonitor.cern.ch/shuttle.jsp?instance=PROD&time=24 \n\n");
3346         }
3347
3348
3349         TString logFolder = "logs";
3350         if (fConfig->GetRunMode() == AliShuttleConfig::kProd)
3351                 logFolder += "_PROD";
3352
3353
3354         body += Form("Find the %s log for the current run on \n\n"
3355                 "\thttp://pcalishuttle02.cern.ch/%s/%d/%d/%s.log \n\n",
3356                      fCurrentDetector.Data(), logFolder.Data(), GetCurrentRun()/10000,
3357                                 GetCurrentRun(), fCurrentDetector.Data());
3358         body += Form("The last 15 lines of %s log file are following:\n\n", fCurrentDetector.Data());
3359
3360         AliDebug(2, Form("Body begin: %s", body.Data()));
3361
3362         mailBody << body.Data();
3363         mailBody.close();
3364         mailBody.open(bodyFileName, ofstream::out | ofstream::app);
3365
3366         TString logFileName = Form("%s/%d/%d/%s.log", GetShuttleLogDir(),
3367                 GetCurrentRun()/10000, GetCurrentRun(), fCurrentDetector.Data());
3368         TString tailCommand = Form("tail -n 15 %s >> %s", logFileName.Data(), bodyFileName.Data());
3369         if (gSystem->Exec(tailCommand.Data()))
3370         {
3371                 mailBody << Form("%s log file not found ...\n\n", fCurrentDetector.Data());
3372         }
3373
3374         TString endBody = Form("------------------------------------------------------\n\n");
3375         endBody += Form("In case of problems please contact the SHUTTLE core team.\n\n");
3376         endBody += "Please do not answer this message directly, it is automatically generated.\n\n";
3377         endBody += "Greetings,\n\n \t\t\tthe SHUTTLE\n";
3378
3379         AliDebug(2, Form("Body end: %s", endBody.Data()));
3380
3381         mailBody << endBody.Data();
3382
3383         mailBody.close();
3384
3385         // send mail!
3386         TString mailCommand = Form("mail -s \"%s\" -c %s %s < %s",
3387                                                 subject.Data(),
3388                                                 cc.Data(),
3389                                                 to.Data(),
3390                                                 bodyFileName.Data());
3391         AliDebug(2, Form("mail command: %s", mailCommand.Data()));
3392
3393         Bool_t result = gSystem->Exec(mailCommand.Data());
3394
3395         return result == 0;
3396 }
3397 //______________________________________________________________________________________________
3398 const char* AliShuttle::GetRunType()
3399 {
3400         //
3401         // returns run type read from "run type" logbook
3402         //
3403
3404         if(!fLogbookEntry) {
3405                 AliError("No logbook entry!");
3406                 return 0;
3407         }
3408
3409         return fLogbookEntry->GetRunType();
3410 }
3411
3412 //______________________________________________________________________________________________
3413 Bool_t AliShuttle::GetHLTStatus()
3414 {
3415         // Return HLT status (ON=1 OFF=0)
3416         // Converts the HLT status from the mode string read in the run logbook (not just a bool)
3417
3418         if(!fLogbookEntry) {
3419                 AliError("No logbook entry!");
3420                 return 0;
3421         }
3422
3423         // TODO implement when HLTMode is inserted in run logbook
3424         TString hltMode = fLogbookEntry->GetRunParameter("HLTmode");
3425         TSubString firstChar = hltMode(0,1);
3426         AliDebug(2,Form("First char = %s ",firstChar.Data()));
3427         if (firstChar == "A") {
3428                 return kFALSE;
3429         }
3430         else if ((firstChar == "B") || (firstChar == "C") || (firstChar == "D") || (firstChar == "E")) {
3431                 return kTRUE;
3432         }
3433         else {
3434                 Log("SHUTTLE","Unexpected HLT mode! Returning 0....");
3435                 return kFALSE;
3436         }
3437 }
3438
3439 //______________________________________________________________________________________________
3440 const char* AliShuttle::GetTriggerConfiguration()
3441 {
3442         // Receives the trigger configuration from the DAQ logbook for the current run
3443
3444         // check connection, if needed reconnect
3445         if (!Connect(3))
3446                 return 0;
3447
3448         TString sqlQuery;
3449         sqlQuery.Form("SELECT configFile FROM logbook_trigger_config WHERE run = %d", GetCurrentRun());
3450         TSQLResult* result = fServer[3]->Query(sqlQuery);
3451         if (!result)
3452         {
3453                 Log("SHUTTLE", Form("ERROR: Can't execute query <%s>!", sqlQuery.Data()));
3454                 return 0;
3455         }
3456
3457         if (result->GetRowCount() == 0)
3458         {
3459                 Log("SHUTTLE", "WARNING: Trigger configuration not found in logbook_trigger_config");
3460                 delete result;
3461                 return 0;
3462         }
3463
3464         TSQLRow* row = result->Next();
3465         if (!row)
3466         {
3467                 Log("SHUTTLE", "ERROR: Could not receive logbook_trigger_config data");
3468                 delete result;
3469                 return 0;
3470         }
3471
3472         // static, so that pointer remains valid when it is returned to the calling class
3473         static TString triggerConfig(row->GetField(0));
3474
3475         delete row;
3476         row = 0;
3477
3478         delete result;
3479         result = 0;
3480
3481         Log("SHUTTLE", Form("Found trigger configuration: %s", triggerConfig.Data()));
3482
3483         return triggerConfig;
3484 }
3485
3486 //______________________________________________________________________________________________
3487 const char* AliShuttle::GetCTPTimeParams()
3488 {
3489         // Receives the CTP time parameters from the DAQ logbook for the current run
3490
3491         // check connection, if needed reconnect
3492         if (!Connect(3))
3493                 return 0;
3494
3495         TString sqlQuery;
3496         sqlQuery.Form("SELECT alignmentFile FROM logbook_trigger_config WHERE run = %d", GetCurrentRun());
3497         TSQLResult* result = fServer[3]->Query(sqlQuery);
3498         if (!result)
3499         {
3500                 Log("SHUTTLE", Form("ERROR: Can't execute query <%s>!", sqlQuery.Data()));
3501                 return 0;
3502         }
3503
3504         if (result->GetRowCount() == 0)
3505         {
3506                 Log("SHUTTLE", "WARNING: CTP time params not found in logbook_trigger_config");
3507                 delete result;
3508                 return 0;
3509         }
3510
3511         TSQLRow* row = result->Next();
3512         if (!row)
3513         {
3514                 Log("SHUTTLE", "ERROR: Could not receive logbook_trigger_config data");
3515                 delete result;
3516                 return 0;
3517         }
3518
3519         // static, so that pointer remains valid when it is returned to the calling class
3520         static TString triggerTimeParams(row->GetField(0));
3521
3522         delete row;
3523         row = 0;
3524
3525         delete result;
3526         result = 0;
3527
3528         Log("SHUTTLE", Form("Found trigger time parameters: %s", triggerTimeParams.Data()));
3529
3530         return triggerTimeParams;
3531 }
3532
3533 //______________________________________________________________________________________________
3534 const char* AliShuttle::GetTriggerDetectorMask()
3535 {
3536         // Receives the trigger detector mask from DAQ logbook
3537
3538         // check connection, if needed reconnect
3539         if (!Connect(3))
3540                 return 0;
3541
3542         TString sqlQuery;
3543         sqlQuery.Form("SELECT BIN(BIT_OR(inputDetectorMask)) from logbook_trigger_clusters WHERE run = %d;", GetCurrentRun());
3544         TSQLResult* result = fServer[3]->Query(sqlQuery);
3545         if (!result)
3546         {
3547                 Log("SHUTTLE", Form("ERROR: Can't execute query <%s>!", sqlQuery.Data()));
3548                 return 0;
3549         }
3550
3551         if (result->GetRowCount() == 0)
3552         {
3553                 Log("SHUTTLE", "ERROR: Trigger Detector Mask not found in logbook_trigger_clusters");
3554                 delete result;
3555                 return 0;
3556         }
3557
3558         TSQLRow* row = result->Next();
3559         if (!row)
3560         {
3561                 Log("SHUTTLE", "ERROR: Could not receive logbook_trigger_clusters data");
3562                 delete result;
3563                 return 0;
3564         }
3565
3566         // static, so that pointer remains valid when it is returned to the calling class
3567         static TString triggerDetectorMask(row->GetField(0));
3568
3569         delete row;
3570         row = 0;
3571
3572         delete result;
3573         result = 0;
3574
3575         Log("SHUTTLE", Form("Found Trigger Detector Mask: %s", triggerDetectorMask.Data()));
3576
3577         return triggerDetectorMask;
3578 }
3579
3580 //______________________________________________________________________________________________
3581 void AliShuttle::SetShuttleTempDir(const char* tmpDir)
3582 {
3583         //
3584         // sets Shuttle temp directory
3585         //
3586
3587         fgkShuttleTempDir = gSystem->ExpandPathName(tmpDir);
3588 }
3589
3590 //______________________________________________________________________________________________
3591 void AliShuttle::SetShuttleLogDir(const char* logDir)
3592 {
3593         //
3594         // sets Shuttle log directory
3595         //
3596
3597         fgkShuttleLogDir = gSystem->ExpandPathName(logDir);
3598 }
3599 //______________________________________________________________________________________________
3600 Bool_t AliShuttle::TouchFile()
3601 {
3602         //
3603         // touching a file on the grid if run has been DONE
3604         //
3605
3606         if (!gGrid)
3607         {
3608                 Log("SHUTTLE",Form("No TGrid connection estabilished!"));
3609                 Log("SHUTTLE",Form("Could not touch file for run %i",GetCurrentRun()));
3610                 return kFALSE;
3611         }
3612
3613         TString dir;
3614         dir.Form("%s%d/SHUTTLE_DONE", fConfig->GetAlienPath(), GetCurrentYear());
3615         // checking whether directory for touch command exists
3616         TString commandLs;
3617         commandLs.Form("ls %s",dir.Data());
3618         TGridResult *resultLs = dynamic_cast<TGridResult*>(gGrid->Command(commandLs));
3619         if (!resultLs){
3620                 Log("SHUTTLE",Form("No result for %s command, returning without touching",commandLs.Data()));
3621                 return kFALSE;
3622         }
3623         TMap *mapLs = dynamic_cast<TMap*>(resultLs->At(0));
3624         if (!mapLs){
3625                 Log("SHUTTLE",Form("No map for %s command, returning without touching",commandLs.Data()));
3626                 delete resultLs;
3627                 resultLs = 0x0;
3628                 return kFALSE;
3629         }
3630         TObjString *valueLsPath = dynamic_cast<TObjString*>(mapLs->GetValue("path"));
3631         if (!valueLsPath || (valueLsPath->GetString()).CompareTo(dir)!=1){
3632                 Log("SHUTTLE",Form("No directory %s found, creating it",dir.Data()));
3633
3634                 // creating the directory
3635
3636                 Bool_t boolMkdir = gGrid->Mkdir(dir.Data());
3637                 if (!boolMkdir) {
3638                         Log("SHUTTLE",Form("Impossible to create dir %s in alien catalogue for run %i!",dir.Data(),GetCurrentRun()));
3639                         delete resultLs;
3640                         resultLs = 0x0;
3641                         return kFALSE;
3642                 }
3643                 Log("SHUTTLE",Form("Directory %s successfully created in alien catalogue for run %i",dir.Data(),GetCurrentRun()));
3644         }
3645         else {
3646                 Log("SHUTTLE",Form("Directory %s correctly found for run %i",dir.Data(),GetCurrentRun()));
3647         }
3648
3649         delete resultLs;
3650         resultLs = 0x0;
3651
3652         TString command;
3653         command.Form("touch %s/%i", dir.Data(), GetCurrentRun());
3654         Log("SHUTTLE", Form("Creating entry in file catalog: %s", command.Data()));
3655         TGridResult *resultTouch = dynamic_cast<TGridResult*>(gGrid->Command(command));
3656         if (!resultTouch){
3657                 Log("SHUTTLE",Form("No result for touching command, returning without touching for run %i",GetCurrentRun()));
3658                 return kFALSE;
3659         }
3660         TMap *mapTouch = dynamic_cast<TMap*>(resultTouch->At(0));
3661         if (!mapTouch){
3662                 Log("SHUTTLE",Form("No map for touching command, returning without touching for run %i",GetCurrentRun()));
3663                 delete resultTouch;
3664                 resultTouch = 0x0;
3665                 return kFALSE;
3666         }
3667         TObjString *valueTouch = dynamic_cast<TObjString*>(mapTouch->GetValue("__result__"));
3668         if (!valueTouch){
3669                 Log("SHUTTLE",Form("No value for \"__result__\" key set in the map for touching command, returning without touching for run %i",GetCurrentRun()));
3670                 delete resultTouch;
3671                 resultTouch = 0x0;
3672                 return kFALSE;
3673         }
3674         if (valueTouch->GetString()!="1"){
3675                 Log("SHUTTLE",Form("Failing the touching command, returning without touching for run %i",GetCurrentRun()));
3676                 delete resultTouch;
3677                 resultTouch = 0x0;
3678                 return kFALSE;
3679         }
3680         delete resultTouch;
3681         resultTouch = 0x0;
3682         Log("SHUTTLE", "Sucessfully touched the file");
3683         return kTRUE;
3684 }
3685 //______________________________________________________________________________________________
3686 UInt_t AliShuttle::GetStartTimeDCSQuery()
3687 {
3688         // Return Start Time for the DCS query
3689         //
3690         // The call is delegated to AliShuttleInterface
3691
3692         return GetCurrentStartTime()-fConfig->GetDCSQueryOffset();
3693 }
3694 //______________________________________________________________________________________________
3695 UInt_t AliShuttle::GetEndTimeDCSQuery()
3696 {
3697         // Return End Time for the DCS query
3698         //
3699         // The call is delegated to AliShuttleInterface
3700
3701         return GetCurrentEndTime()+fConfig->GetDCSQueryOffset();
3702 }
3703