SHUTTLE/AliShuttle.cxx

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