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                 Log(fCurrentDetector.Data(), Form("Attempting to store %s", aLocId.ToString().Data()));
 380
 381                 // If local object is valid up to infinity we store it only if it is
 382                 // the first unprocessed run!
 383                 if (aLocId.GetLastRun() == AliCDBRunRange::Infinity() &&
 384                         !fFirstUnprocessed[GetDetPos(fCurrentDetector)])
 385                 {
 386                         Log("SHUTTLE", Form("StoreOCDB - %s: object %s has validity infinite but "
 387                                                 "there are previous unprocessed runs!",
 388                                                 fCurrentDetector.Data(), aLocId.GetPath().Data()));
 389                         Log(fCurrentDetector.Data(), Form("StoreOCDB - %s: object %s has validity infinite but "
 390                                                 "there are previous unprocessed runs!",
 391                                                 fCurrentDetector.Data(), aLocId.GetPath().Data()));
 392                         delayed = kTRUE;
 393                         continue;
 394                 }
 395
 396                 // loop on Grid valid Id's
 397                 Bool_t store = kTRUE;
 398                 TIter gridIter(gridIds);
 399                 AliCDBId* aGridId = 0;
 400                 while ((aGridId = dynamic_cast<AliCDBId*> (gridIter.Next()))) {
 401                         if (aGridId->GetPath() != aLocId.GetPath())
 402                                 continue;
 403                         // skip all objects valid up to infinity
 404                         if (aGridId->GetLastRun() == AliCDBRunRange::Infinity())
 405                                 continue;
 406
 407                         // if we get here, it means there's already some more recent object stored on Grid!
 408                         Log(fCurrentDetector.Data(),
 409                                 Form("StoreOCDB - A more recent object already exists in %s storage: <%s>",
 410                                 type, aGridId->ToString().Data()));
 411
 412                         store = kFALSE;
 413                         break;
 414                 }
 415
 416                 Bool_t storeOk = kFALSE;
 417                 if (store)
 418                 {
 419                         Log(fCurrentDetector.Data(), Form("Prechecks succeeded. Ready to store %s", aLocId.ToString().Data()));
 420                         storeOk = gridSto->Put(aLocEntry);
 421                         if (storeOk) {
 422                                 Log("SHUTTLE",
 423                                 Form("StoreOCDB - Object <%s> successfully put into %s storage",
 424                                         aLocId.ToString().Data(), type));
 425                                 Log(fCurrentDetector.Data(),
 426                                         Form("StoreOCDB - Object <%s> successfully put into %s storage",
 427                                         aLocId.ToString().Data(), type));
 428                         } else  {
 429                                 Log("SHUTTLE",
 430                                         Form("StoreOCDB - Grid %s storage of object <%s> failed",
 431                                         type, aLocId.ToString().Data()));
 432                                 Log(fCurrentDetector.Data(),
 433                                         Form("StoreOCDB - Grid %s storage of object <%s> failed",
 434                                         type, aLocId.ToString().Data()));
 435                                 result = kFALSE;
 436                         }
 437                 }
 438
 439                 if (!store || storeOk) {
 440                         // removing local file...
 441                         TString filename;
 442                         localSto->IdToFilename(aLocId, filename);
 443                         Log("SHUTTLE", Form("StoreOCDB - Removing local file %s", filename.Data()));
 444                         RemoveFile(filename.Data());
 445                 }
 446         }
 447         localEntries->Clear();
 448
 449         Int_t returnCode = 0;
 450
 451         if (result == kFALSE)
 452                 returnCode = -1;
 453         else if (delayed != kFALSE)
 454                 returnCode =  1;
 455
 456         Log("SHUTTLE", Form("StoreOCDB - Returning with %d (result = %d, delayed = %d)", returnCode, result, delayed));
 457         Log(fCurrentDetector.Data(), Form("StoreOCDB - Returning with %d (result = %d, delayed = %d)", returnCode, result, delayed));
 458
 459         return returnCode;
 460 }
 461
 462 //______________________________________________________________________________________________
 463 Bool_t AliShuttle::CleanReferenceStorage(const char* detector)
 464 {
 465         // clears the directory used to store reference files of a given subdetector
 466
 467         AliCDBManager* man = AliCDBManager::Instance();
 468         AliCDBStorage* sto = man->GetStorage(fgkLocalRefStorage);
 469         TString localBaseFolder = sto->GetBaseFolder();
 470
 471         TString targetDir = GetRefFilePrefix(localBaseFolder.Data(), detector);
 472
 473         Log("SHUTTLE", Form("CleanReferenceStorage - Cleaning %s", targetDir.Data()));
 474
 475         TString begin;
 476         begin.Form("%d_", GetCurrentRun());
 477
 478         TSystemDirectory* baseDir = new TSystemDirectory("/", targetDir);
 479         if (!baseDir)
 480                 return kTRUE;
 481
 482         TList* dirList = baseDir->GetListOfFiles();
 483         delete baseDir;
 484
 485         if (!dirList) return kTRUE;
 486
 487         if (dirList->GetEntries() < 3)
 488         {
 489                 delete dirList;
 490                 return kTRUE;
 491         }
 492
 493         Int_t nDirs = 0, nDel = 0;
 494         TIter dirIter(dirList);
 495         TSystemFile* entry = 0;
 496
 497         Bool_t success = kTRUE;
 498
 499         while ((entry = dynamic_cast<TSystemFile*> (dirIter.Next())))
 500         {
 501                 if (entry->IsDirectory())
 502                         continue;
 503
 504                 TString fileName(entry->GetName());
 505                 if (!fileName.BeginsWith(begin))
 506                         continue;
 507
 508                 nDirs++;
 509
 510                 // delete file
 511                 Int_t result = gSystem->Unlink(fileName.Data());
 512
 513                 if (result)
 514                 {
 515                         Log("SHUTTLE", Form("CleanReferenceStorage - Could not delete file %s!", fileName.Data()));
 516                         success = kFALSE;
 517                 } else {
 518                         nDel++;
 519                 }
 520         }
 521
 522         if(nDirs > 0)
 523                 Log("SHUTTLE", Form("CleanReferenceStorage - %d (over %d) reference files in folder %s were deleted.",
 524                         nDel, nDirs, targetDir.Data()));
 525
 526
 527         delete dirList;
 528         return success;
 529
 530
 531
 532
 533
 534
 535   Int_t result = gSystem->GetPathInfo(targetDir, 0, (Long64_t*) 0, 0, 0);
 536   if (result == 0)
 537   {
 538     // delete directory
 539     result = gSystem->Exec(Form("rm -rf %s", targetDir.Data()));
 540     if (result != 0)
 541     {
 542       Log("SHUTTLE", Form("CleanReferenceStorage - Could not clean directory %s", targetDir.Data()));
 543       return kFALSE;
 544     }
 545   }
 546
 547   result = gSystem->mkdir(targetDir, kTRUE);
 548   if (result != 0)
 549   {
 550     Log("SHUTTLE", Form("CleanReferenceStorage - Error creating base directory %s", targetDir.Data()));
 551     return kFALSE;
 552   }
 553
 554   return kTRUE;
 555 }
 556
 557 //______________________________________________________________________________________________
 558 Bool_t AliShuttle::StoreReferenceFile(const char* detector, const char* localFile, const char* gridFileName)
 559 {
 560         //
 561         // Stores reference file directly (without opening it). This function stores the file locally.
 562         //
 563         // The file is stored under the following location:
 564         // <base folder of local reference storage>/<DET>/<RUN#>_<gridFileName>
 565         // where <gridFileName> is the second parameter given to the function
 566         //
 567
 568         if (fTestMode & kErrorStorage)
 569         {
 570                 Log(fCurrentDetector, "StoreReferenceFile - In TESTMODE - Simulating error while storing locally");
 571                 return kFALSE;
 572         }
 573
 574         AliCDBManager* man = AliCDBManager::Instance();
 575         AliCDBStorage* sto = man->GetStorage(fgkLocalRefStorage);
 576
 577         TString localBaseFolder = sto->GetBaseFolder();
 578
 579         TString target = GetRefFilePrefix(localBaseFolder.Data(), detector);
 580         target.Append(Form("/%d_%s", GetCurrentRun(), gridFileName));
 581
 582         return CopyFileLocally(localFile, target);
 583 }
 584
 585 //______________________________________________________________________________________________
 586 Bool_t AliShuttle::StoreRunMetadataFile(const char* localFile, const char* gridFileName)
 587 {
 588         //
 589         // Stores Run metadata file to the Grid, in the run folder
 590         //
 591         // Only GRP can call this function.
 592
 593         if (fTestMode & kErrorStorage)
 594         {
 595                 Log(fCurrentDetector, "StoreRunMetaDataFile - In TESTMODE - Simulating error while storing locally");
 596                 return kFALSE;
 597         }
 598
 599         AliCDBManager* man = AliCDBManager::Instance();
 600         AliCDBStorage* sto = man->GetStorage(fgkLocalRefStorage);
 601
 602         TString localBaseFolder = sto->GetBaseFolder();
 603
 604         // Build Run level folder
 605         // folder = /alice/data/year/lhcPeriod/runNb/raw
 606
 607
 608         TString lhcPeriod = GetLHCPeriod();
 609         if (lhcPeriod.Length() == 0)
 610         {
 611                 Log("SHUTTLE","StoreRunMetaDataFile - LHCPeriod not found in logbook!");
 612                 return 0;
 613         }
 614
 615         // TODO partitions with one detector only write data into LHCperiod_DET
 616         TString partition = GetRunParameter("detector");
 617
 618         if (partition.Length() > 0 && partition != "ALICE")
 619         {
 620                 lhcPeriod.Append(Form("_%s", partition.Data()));
 621                 Log(fCurrentDetector, Form("Run data tags merged file will be written in %s",
 622                                 lhcPeriod.Data()));
 623         }
 624
 625         TString target = Form("%s/GRP/RunMetadata%s%d/%s/%09d/raw/%s",
 626                                 localBaseFolder.Data(), fConfig->GetAlienPath(), GetCurrentYear(),
 627                                 lhcPeriod.Data(), GetCurrentRun(), gridFileName);
 628
 629         return CopyFileLocally(localFile, target);
 630 }
 631
 632 //______________________________________________________________________________________________
 633 Bool_t AliShuttle::CopyFileLocally(const char* localFile, const TString& target)
 634 {
 635         //
 636         // Stores file locally. Called by StoreReferenceFile and StoreRunMetadataFile
 637         // Files are temporarily stored in the local reference storage. When the preprocessor
 638         // finishes, the Shuttle calls CopyFilesToGrid to transfer the files to AliEn
 639         // (in reference or run level folders)
 640         //
 641
 642         TString targetDir(target(0, target.Last('/')));
 643
 644         //try to open base dir folder, if it does not exist
 645         void* dir = gSystem->OpenDirectory(targetDir.Data());
 646         if (dir == NULL) {
 647                 if (gSystem->mkdir(targetDir.Data(), kTRUE)) {
 648                         Log("SHUTTLE", Form("CopyFileLocally - Can't open directory <%s>", targetDir.Data()));
 649                         return kFALSE;
 650                 }
 651
 652         } else {
 653                 gSystem->FreeDirectory(dir);
 654         }
 655
 656         Int_t result = 0;
 657
 658         result = gSystem->GetPathInfo(localFile, 0, (Long64_t*) 0, 0, 0);
 659         if (result)
 660         {
 661                 Log("SHUTTLE", Form("CopyFileLocally - %s does not exist", localFile));
 662                 return kFALSE;
 663         }
 664
 665         result = gSystem->GetPathInfo(target, 0, (Long64_t*) 0, 0, 0);
 666         if (!result)
 667         {
 668                 Log("SHUTTLE", Form("CopyFileLocally - target file %s already exist, removing...", target.Data()));
 669                 if (gSystem->Unlink(target.Data()))
 670                 {
 671                         Log("SHUTTLE", Form("CopyFileLocally - Could not remove existing target file %s!", target.Data()));
 672                         return kFALSE;
 673                 }
 674         }
 675
 676         result = gSystem->CopyFile(localFile, target);
 677
 678         if (result == 0)
 679         {
 680                 Log("SHUTTLE", Form("CopyFileLocally - File %s stored locally to %s", localFile, target.Data()));
 681                 return kTRUE;
 682         }
 683         else
 684         {
 685                 Log("SHUTTLE", Form("CopyFileLocally - Could not store file %s to %s! Error code = %d",
 686                                 localFile, target.Data(), result));
 687                 return kFALSE;
 688         }
 689
 690
 691
 692 }
 693
 694 //______________________________________________________________________________________________
 695 Bool_t AliShuttle::CopyFilesToGrid(const char* type)
 696 {
 697         //
 698         // Transfers local files to the Grid. Local files can be reference files
 699         // or run metadata file (from GRP only).
 700         //
 701         // According to the type (ref, metadata) the files are stored under the following location:
 702         // ref --> <base folder of reference storage>/<DET>/<RUN#>_<gridFileName>
 703         // metadata --> <run data folder>/<MetadataFileName>
 704         //
 705
 706         AliCDBManager* man = AliCDBManager::Instance();
 707         AliCDBStorage* sto = man->GetStorage(fgkLocalRefStorage);
 708         if (!sto)
 709                 return kFALSE;
 710         TString localBaseFolder = sto->GetBaseFolder();
 711
 712         TString dir;
 713         TString alienDir;
 714         TString begin;
 715
 716         if (strcmp(type, "reference") == 0)
 717         {
 718                 dir = GetRefFilePrefix(localBaseFolder.Data(), fCurrentDetector.Data());
 719                 AliCDBStorage* gridSto = man->GetStorage(fgkMainRefStorage);
 720                 if (!gridSto)
 721                         return kFALSE;
 722                 TString gridBaseFolder = gridSto->GetBaseFolder();
 723                 alienDir = GetRefFilePrefix(gridBaseFolder.Data(), fCurrentDetector.Data());
 724                 begin = Form("%d_", GetCurrentRun());
 725         }
 726         else if (strcmp(type, "metadata") == 0)
 727         {
 728
 729                 TString lhcPeriod = GetLHCPeriod();
 730
 731                 if (lhcPeriod.Length() == 0)
 732                 {
 733                         Log("SHUTTLE","CopyFilesToGrid - LHCPeriod not found in logbook!");
 734                         return 0;
 735                 }
 736
 737                 // TODO partitions with one detector only write data into LHCperiod_DET
 738                 TString partition = GetRunParameter("detector");
 739
 740                 if (partition.Length() > 0 && partition != "ALICE")
 741                 {
 742                         lhcPeriod.Append(Form("_%s", partition.Data()));
 743                 }
 744
 745                 dir = Form("%s/GRP/RunMetadata%s%d/%s/%09d/raw",
 746                                 localBaseFolder.Data(), fConfig->GetAlienPath(), GetCurrentYear(),
 747                                 lhcPeriod.Data(), GetCurrentRun());
 748                 alienDir = dir(dir.Index(fConfig->GetAlienPath()), dir.Length());
 749
 750                 begin = "";
 751         }
 752         else
 753         {
 754                 Log("SHUTTLE", "CopyFilesToGrid - Unexpected: type label must be reference or metadata!");
 755                 return kFALSE;
 756         }
 757
 758         TSystemDirectory* baseDir = new TSystemDirectory("/", dir);
 759         if (!baseDir)
 760                 return kTRUE;
 761
 762         TList* dirList = baseDir->GetListOfFiles();
 763         delete baseDir;
 764
 765         if (!dirList) return kTRUE;
 766
 767         if (dirList->GetEntries() < 3)
 768         {
 769                 delete dirList;
 770                 return kTRUE;
 771         }
 772
 773         if (!gGrid)
 774         {
 775                 Log("SHUTTLE", "CopyFilesToGrid - Connection to Grid failed: Cannot continue!");
 776                 delete dirList;
 777                 return kFALSE;
 778         }
 779
 780         Int_t nDirs = 0, nTransfer = 0;
 781         TIter dirIter(dirList);
 782         TSystemFile* entry = 0;
 783
 784         Bool_t success = kTRUE;
 785         Bool_t first = kTRUE;
 786
 787         while ((entry = dynamic_cast<TSystemFile*> (dirIter.Next())))
 788         {
 789                 if (entry->IsDirectory())
 790                         continue;
 791
 792                 TString fileName(entry->GetName());
 793                 if (!fileName.BeginsWith(begin))
 794                         continue;
 795
 796                 nDirs++;
 797
 798                 if (first)
 799                 {
 800                         first = kFALSE;
 801                         // check that folder exists, otherwise create it
 802                         TGridResult* result = gGrid->Ls(alienDir.Data(), "a");
 803
 804                         if (!result)
 805                         {
 806                                 delete dirList;
 807                                 return kFALSE;
 808                         }
 809
 810                         if (!result->GetFileName(1)) // TODO: It looks like element 0 is always 0!!
 811                         {
 812                                 // TODO It does not work currently! Bug in TAliEn::Mkdir
 813                                 // TODO Manually fixed in local root v5-16-00
 814                                 if (!gGrid->Mkdir(alienDir.Data(),"-p",0))
 815                                 {
 816                                         Log("SHUTTLE", Form("CopyFilesToGrid - Cannot create directory %s",
 817                                                         alienDir.Data()));
 818                                         delete dirList;
 819                                         return kFALSE;
 820                                 } else {
 821                                         Log("SHUTTLE",Form("CopyFilesToGrid - Folder %s created", alienDir.Data()));
 822                                 }
 823
 824                         } else {
 825                                         Log("SHUTTLE",Form("CopyFilesToGrid - Folder %s found", alienDir.Data()));
 826                         }
 827                 }
 828
 829                 TString fullLocalPath;
 830                 fullLocalPath.Form("%s/%s", dir.Data(), fileName.Data());
 831
 832                 TString fullGridPath;
 833                 fullGridPath.Form("alien://%s/%s", alienDir.Data(), fileName.Data());
 834
 835                 Bool_t result = TFile::Cp(fullLocalPath, fullGridPath);
 836
 837                 if (result)
 838                 {
 839                         Log("SHUTTLE", Form("CopyFilesToGrid - Copying local file %s to %s succeeded!",
 840                                                 fullLocalPath.Data(), fullGridPath.Data()));
 841                         RemoveFile(fullLocalPath);
 842                         nTransfer++;
 843                 }
 844                 else
 845                 {
 846                         Log("SHUTTLE", Form("CopyFilesToGrid - Copying local file %s to %s FAILED!",
 847                                                 fullLocalPath.Data(), fullGridPath.Data()));
 848                         success = kFALSE;
 849                 }
 850         }
 851
 852         Log("SHUTTLE", Form("CopyFilesToGrid - %d (over %d) files in folder %s copied to Grid.",
 853                                                 nTransfer, nDirs, dir.Data()));
 854
 855
 856         delete dirList;
 857         return success;
 858 }
 859
 860 //______________________________________________________________________________________________
 861 const char* AliShuttle::GetRefFilePrefix(const char* base, const char* detector)
 862 {
 863         //
 864         // Get folder name of reference files
 865         //
 866
 867         TString offDetStr(GetOfflineDetName(detector));
 868         TString dir;
 869         if (offDetStr == "ITS" || offDetStr == "MUON" || offDetStr == "PHOS")
 870         {
 871                 dir.Form("%s/%s/%s", base, offDetStr.Data(), detector);
 872         } else {
 873                 dir.Form("%s/%s", base, offDetStr.Data());
 874         }
 875
 876         return dir.Data();
 877
 878
 879 }
 880
 881 //______________________________________________________________________________________________
 882 void AliShuttle::CleanLocalStorage(const TString& uri)
 883 {
 884         //
 885         // Called in case the preprocessor is declared failed. Remove remaining objects from the local storages.
 886         //
 887
 888         const char* type = 0;
 889         if(uri == fgkLocalCDB) {
 890                 type = "OCDB";
 891         } else if(uri == fgkLocalRefStorage) {
 892                 type = "Reference";
 893         } else {
 894                 AliError(Form("Invalid storage URI: %s", uri.Data()));
 895                 return;
 896         }
 897
 898         AliCDBManager* man = AliCDBManager::Instance();
 899
 900         // open local storage
 901         AliCDBStorage *localSto = man->GetStorage(uri);
 902         if(!localSto) {
 903                 Log("SHUTTLE",
 904                         Form("CleanLocalStorage - cannot activate local %s storage", type));
 905                 return;
 906         }
 907
 908         TString filename(Form("%s/%s/*/Run*_v%d_s*.root",
 909                 localSto->GetBaseFolder().Data(), GetOfflineDetName(fCurrentDetector.Data()), GetCurrentRun()));
 910
 911         AliDebug(2, Form("filename = %s", filename.Data()));
 912
 913         Log("SHUTTLE", Form("Removing remaining local files for run %d and detector %s ...",
 914                 GetCurrentRun(), fCurrentDetector.Data()));
 915
 916         RemoveFile(filename.Data());
 917
 918 }
 919
 920 //______________________________________________________________________________________________
 921 void AliShuttle::RemoveFile(const char* filename)
 922 {
 923         //
 924         // removes local file
 925         //
 926
 927         TString command(Form("rm -f %s", filename));
 928
 929         Int_t result = gSystem->Exec(command.Data());
 930         if(result != 0)
 931         {
 932                 Log("SHUTTLE", Form("RemoveFile - %s: Cannot remove file %s!",
 933                         fCurrentDetector.Data(), filename));
 934         }
 935 }
 936
 937 //______________________________________________________________________________________________
 938 AliShuttleStatus* AliShuttle::ReadShuttleStatus()
 939 {
 940         //
 941         // Reads the AliShuttleStatus from the CDB
 942         //
 943
 944         if (fStatusEntry){
 945                 delete fStatusEntry;
 946                 fStatusEntry = 0;
 947         }
 948
 949         fStatusEntry = AliCDBManager::Instance()->GetStorage(GetLocalCDB())
 950                 ->Get(Form("/SHUTTLE/STATUS/%s", fCurrentDetector.Data()), GetCurrentRun());
 951
 952         if (!fStatusEntry) return 0;
 953         fStatusEntry->SetOwner(1);
 954
 955         AliShuttleStatus* status = dynamic_cast<AliShuttleStatus*> (fStatusEntry->GetObject());
 956         if (!status) {
 957                 AliError("Invalid object stored to CDB!");
 958                 return 0;
 959         }
 960
 961         return status;
 962 }
 963
 964 //______________________________________________________________________________________________
 965 Bool_t AliShuttle::WriteShuttleStatus(AliShuttleStatus* status)
 966 {
 967         //
 968         // writes the status for one subdetector
 969         //
 970
 971         if (fStatusEntry){
 972                 delete fStatusEntry;
 973                 fStatusEntry = 0;
 974         }
 975
 976         Int_t run = GetCurrentRun();
 977
 978         AliCDBId id(AliCDBPath("SHUTTLE", "STATUS", fCurrentDetector), run, run);
 979
 980         fStatusEntry = new AliCDBEntry(status, id, new AliCDBMetaData);
 981         fStatusEntry->SetOwner(1);
 982
 983         UInt_t result = AliCDBManager::Instance()->GetStorage(fgkLocalCDB)->Put(fStatusEntry);
 984
 985         if (!result) {
 986                 Log("SHUTTLE", Form("WriteShuttleStatus - Failed for %s, run %d",
 987                                                 fCurrentDetector.Data(), run));
 988                 return kFALSE;
 989         }
 990
 991         SendMLDetInfo();
 992
 993         return kTRUE;
 994 }
 995
 996 //______________________________________________________________________________________________
 997 void AliShuttle::UpdateShuttleStatus(AliShuttleStatus::Status newStatus, Bool_t increaseCount)
 998 {
 999         //
1000         // changes the AliShuttleStatus for the given detector and run to the given status
1001         //
1002
1003         if (!fStatusEntry){
1004                 AliError("UNEXPECTED: fStatusEntry empty");
1005                 return;
1006         }
1007
1008         AliShuttleStatus* status = dynamic_cast<AliShuttleStatus*> (fStatusEntry->GetObject());
1009
1010         if (!status){
1011                 Log("SHUTTLE", "UpdateShuttleStatus - UNEXPECTED: status could not be read from current CDB entry");
1012                 return;
1013         }
1014
1015         TString actionStr = Form("UpdateShuttleStatus - %s: Changing state from %s to %s",
1016                                 fCurrentDetector.Data(),
1017                                 status->GetStatusName(),
1018                                 status->GetStatusName(newStatus));
1019         Log("SHUTTLE", actionStr);
1020         SetLastAction(actionStr);
1021
1022         status->SetStatus(newStatus);
1023         if (increaseCount) status->IncreaseCount();
1024
1025         AliCDBManager::Instance()->GetStorage(fgkLocalCDB)->Put(fStatusEntry);
1026
1027         SendMLDetInfo();
1028 }
1029
1030 //______________________________________________________________________________________________
1031 void AliShuttle::SendMLDetInfo()
1032 {
1033         //
1034         // sends ML information about the current status of the current detector being processed
1035         //
1036
1037         AliShuttleStatus* status = dynamic_cast<AliShuttleStatus*> (fStatusEntry->GetObject());
1038
1039         if (!status){
1040                 Log("SHUTTLE", "SendMLDetInfo - UNEXPECTED: status could not be read from current CDB entry");
1041                 return;
1042         }
1043
1044         TMonaLisaText  mlStatus(Form("%s_status", fCurrentDetector.Data()), status->GetStatusName());
1045         TMonaLisaValue mlRetryCount(Form("%s_count", fCurrentDetector.Data()), status->GetCount());
1046
1047         TList mlList;
1048         mlList.Add(&mlStatus);
1049         mlList.Add(&mlRetryCount);
1050
1051         TString mlID;
1052         mlID.Form("%d", GetCurrentRun());
1053         fMonaLisa->SendParameters(&mlList, mlID);
1054 }
1055
1056 //______________________________________________________________________________________________
1057 Bool_t AliShuttle::ContinueProcessing()
1058 {
1059         // this function reads the AliShuttleStatus information from CDB and
1060         // checks if the processing should be continued
1061         // if yes it returns kTRUE and updates the AliShuttleStatus with nextStatus
1062
1063         if (!fConfig->HostProcessDetector(fCurrentDetector))
1064                 return kFALSE;
1065
1066         AliPreprocessor* aPreprocessor =
1067                 dynamic_cast<AliPreprocessor*> (fPreprocessorMap.GetValue(fCurrentDetector));
1068         if (!aPreprocessor)
1069         {
1070                 Log("SHUTTLE", Form("ContinueProcessing - %s: no preprocessor registered", fCurrentDetector.Data()));
1071                 return kFALSE;
1072         }
1073
1074         AliShuttleLogbookEntry::Status entryStatus =
1075                 fLogbookEntry->GetDetectorStatus(fCurrentDetector);
1076
1077         if (entryStatus != AliShuttleLogbookEntry::kUnprocessed) {
1078                 Log("SHUTTLE", Form("ContinueProcessing - %s is %s",
1079                                 fCurrentDetector.Data(),
1080                                 fLogbookEntry->GetDetectorStatusName(entryStatus)));
1081                 return kFALSE;
1082         }
1083
1084         // if we get here, according to Shuttle logbook subdetector is in UNPROCESSED state
1085
1086         // check if current run is first unprocessed run for current detector
1087         if (fConfig->StrictRunOrder(fCurrentDetector) &&
1088                 !fFirstUnprocessed[GetDetPos(fCurrentDetector)])
1089         {
1090                 if (fTestMode == kNone)
1091                 {
1092                         Log("SHUTTLE", Form("ContinueProcessing - %s requires strict run ordering"
1093                                         " but this is not the first unprocessed run!"));
1094                         return kFALSE;
1095                 }
1096                 else
1097                 {
1098                         Log("SHUTTLE", Form("ContinueProcessing - In TESTMODE - "
1099                                         "Although %s requires strict run ordering "
1100                                         "and this is not the first unprocessed run, "
1101                                         "the SHUTTLE continues"));
1102                 }
1103         }
1104
1105         // Is the subdetector processed first time for this run?
1106         fFirstProcessing = kFALSE;
1107
1108         AliShuttleStatus* status = ReadShuttleStatus();
1109         if (!status) {
1110                 // first time
1111                 Log("SHUTTLE", Form("ContinueProcessing - %s: Processing first time",
1112                                 fCurrentDetector.Data()));
1113                 status = new AliShuttleStatus(AliShuttleStatus::kStarted);
1114                 fFirstProcessing = kTRUE;
1115                 return WriteShuttleStatus(status);
1116         }
1117
1118         // The following case shouldn't happen if Shuttle Logbook was correctly updated.
1119         // If it happens it may mean Logbook updating failed... let's do it now!
1120         if (status->GetStatus() == AliShuttleStatus::kDone ||
1121             status->GetStatus() == AliShuttleStatus::kFailed ||
1122             status->GetStatus() == AliShuttleStatus::kSkipped) {
1123                 Log("SHUTTLE", Form("ContinueProcessing - %s is already %s. Updating Shuttle Logbook",
1124                                         fCurrentDetector.Data(),
1125                                         status->GetStatusName(status->GetStatus())));
1126
1127                 if (status->GetStatus() == AliShuttleStatus::kSkipped)
1128                 {
1129                         UpdateShuttleLogbook(fCurrentDetector.Data(), "DONE");
1130                 }
1131                 else
1132                         UpdateShuttleLogbook(fCurrentDetector.Data(), status->GetStatusName(status->GetStatus()));
1133
1134                 return kFALSE;
1135         }
1136
1137         if (status->GetStatus() == AliShuttleStatus::kStoreStarted || status->GetStatus() == AliShuttleStatus::kStoreDelayed ||status->GetStatus() == AliShuttleStatus::kStoreError) {
1138                 Log("SHUTTLE",
1139                         Form("ContinueProcessing - %s: Grid storage of one or more "
1140                                 "objects failed. Trying again now",
1141                                 fCurrentDetector.Data()));
1142                 StoreOCDB();
1143                 return kFALSE;
1144         }
1145
1146         // if we get here, there is a restart
1147         Bool_t cont = kFALSE;
1148
1149         // abort conditions
1150         if (status->GetCount() >= fConfig->GetMaxRetries()) {
1151                 Log("SHUTTLE", Form("ContinueProcessing - %s failed %d times in status %s - "
1152                                 "Updating Shuttle Logbook", fCurrentDetector.Data(),
1153                                 status->GetCount(), status->GetStatusName()));
1154                 UpdateShuttleLogbook(fCurrentDetector.Data(), "FAILED");
1155                 UpdateShuttleStatus(AliShuttleStatus::kFailed);
1156
1157                 // there may still be objects in local OCDB and reference storage
1158                 // and FXS databases may be not updated: do it now!
1159
1160                 // TODO Currently disabled, we want to keep files in case of failure!
1161                 // CleanLocalStorage(fgkLocalCDB);
1162                 // CleanLocalStorage(fgkLocalRefStorage);
1163                 // UpdateTableFailCase();
1164
1165                 // Send mail to detector expert!
1166                 Log("SHUTTLE", Form("ContinueProcessing - Sending mail to %s expert...",
1167                                     fCurrentDetector.Data()));
1168                 if (!SendMail(kPPEMail))
1169                         Log("SHUTTLE", Form("ContinueProcessing - Could not send mail to %s expert",
1170                                             fCurrentDetector.Data()));
1171
1172         } else {
1173                 Log("SHUTTLE", Form("ContinueProcessing - %s: restarting. "
1174                                 "Aborted before with %s. Retry number %d.", fCurrentDetector.Data(),
1175                                 status->GetStatusName(), status->GetCount()));
1176                 Bool_t increaseCount = kTRUE;
1177                 if (status->GetStatus() == AliShuttleStatus::kDCSError ||
1178                     status->GetStatus() == AliShuttleStatus::kDCSStarted ||
1179                     status->GetStatus() == AliShuttleStatus::kFXSError)
1180                                 increaseCount = kFALSE;
1181
1182                 UpdateShuttleStatus(AliShuttleStatus::kStarted, increaseCount);
1183                 cont = kTRUE;
1184         }
1185
1186         return cont;
1187 }
1188
1189 //______________________________________________________________________________________________
1190 void AliShuttle::SendMLRunInfo(const char* status)
1191 {
1192         //
1193         // Send information about this run to ML
1194
1195         TMonaLisaText  mlStatus("SHUTTLE_status", status);
1196         TString runType(fLogbookEntry->GetRunType());
1197         if (strlen(fLogbookEntry->GetRunParameter("log")) > 0){
1198
1199                 runType += "(";
1200                 runType += fLogbookEntry->GetRunParameter("log");
1201                 runType += ")";
1202         }
1203         TMonaLisaText  mlRunType("SHUTTLE_runtype", runType);
1204
1205         TList mlList;
1206         mlList.Add(&mlStatus);
1207         mlList.Add(&mlRunType);
1208
1209         TString mlID;
1210         mlID.Form("%d", GetCurrentRun());
1211         fMonaLisa->SendParameters(&mlList, mlID);
1212 }
1213
1214 //______________________________________________________________________________________________
1215 Int_t AliShuttle::GetMem(Int_t pid)
1216 {
1217         // invokes ps to get the memory consumption of the process <pid>
1218         // returns -1 in case of error
1219
1220         TString checkStr;
1221         checkStr.Form("ps -o vsize --pid %d | tail -n 1", pid);
1222         FILE* pipe = gSystem->OpenPipe(checkStr, "r");
1223         if (!pipe)
1224         {
1225                 Log("SHUTTLE", Form("Process - Error: "
1226                         "Could not open pipe to %s", checkStr.Data()));
1227                 return -1;
1228         }
1229
1230         char buffer[100];
1231         if (!fgets(buffer, 100, pipe))
1232         {
1233                 Log("SHUTTLE", "Process - Error: ps did not return anything");
1234                 gSystem->ClosePipe(pipe);
1235                 return -1;
1236         }
1237         gSystem->ClosePipe(pipe);
1238
1239         //Log("SHUTTLE", Form("ps returned %s", buffer));
1240
1241         Int_t mem = 0;
1242         if ((sscanf(buffer, "%d\n", &mem) != 1) || !mem)
1243         {
1244                 Log("SHUTTLE", "Process - Error: Could not parse output of ps");
1245                 return -1;
1246         }
1247
1248         return mem;
1249 }
1250
1251 //______________________________________________________________________________________________
1252 Bool_t AliShuttle::Process(AliShuttleLogbookEntry* entry)
1253 {
1254         //
1255         // Makes data retrieval for all detectors in the configuration.
1256         // entry: Shuttle logbook entry, contains run paramenters and status of detectors
1257         // (Unprocessed, Inactive, Failed or Done).
1258         // Returns kFALSE in case of error occured and kTRUE otherwise
1259         //
1260
1261         if (!entry) return kFALSE;
1262
1263         fLogbookEntry = entry;
1264
1265         Log("SHUTTLE", Form("\t\t\t^*^*^*^*^*^*^*^*^*^*^*^* run %d: START ^*^*^*^*^*^*^*^*^*^*^*^*",
1266                                         GetCurrentRun()));
1267
1268         CountOpenRuns();
1269
1270         // Send the information to ML
1271         SendMLRunInfo("Processing");
1272
1273         if (fLogbookEntry->IsDone())
1274         {
1275                 Log("SHUTTLE","Process - Shuttle is already DONE. Updating logbook");
1276                 UpdateShuttleLogbook("shuttle_done");
1277                 fLogbookEntry = 0;
1278                 return kTRUE;
1279         }
1280
1281         // read test mode if flag is set
1282         if (fReadTestMode)
1283         {
1284                 fTestMode = kNone;
1285                 TString logEntry(entry->GetRunParameter("log"));
1286                 //printf("log entry = %s\n", logEntry.Data());
1287                 TString searchStr("Testmode: ");
1288                 Int_t pos = logEntry.Index(searchStr.Data());
1289                 //printf("%d\n", pos);
1290                 if (pos >= 0)
1291                 {
1292                         TSubString subStr = logEntry(pos + searchStr.Length(), logEntry.Length());
1293                         //printf("%s\n", subStr.String().Data());
1294                         TString newStr(subStr.Data());
1295                         TObjArray* token = newStr.Tokenize(' ');
1296                         if (token)
1297                         {
1298                                 //token->Print();
1299                                 TObjString* tmpStr = dynamic_cast<TObjString*> (token->First());
1300                                 if (tmpStr)
1301                                 {
1302                                         Int_t testMode = tmpStr->String().Atoi();
1303                                         if (testMode > 0)
1304                                         {
1305                                                 Log("SHUTTLE", Form("Process - Enabling test mode %d", testMode));
1306                                                 SetTestMode((TestMode) testMode);
1307                                         }
1308                                 }
1309                                 delete token;
1310                         }
1311                 }
1312         }
1313
1314         fLogbookEntry->Print("all");
1315
1316         // Initialization
1317         Bool_t hasError = kFALSE;
1318
1319         // Set the CDB and Reference folders according to the year and LHC period
1320         TString lhcPeriod(GetLHCPeriod());
1321         if (lhcPeriod.Length() == 0)
1322         {
1323                 Log("SHUTTLE","Process - LHCPeriod not found in logbook!");
1324                 return 0;
1325         }
1326
1327         // build cdb paths (repeat each time, LHCperiod might have changed)
1328         fgkMainCDB.Form("alien://folder=%s%d/%s/OCDB?user=alidaq?cacheFold=/tmp/OCDBCache",
1329                                         fConfig->GetAlienPath(), GetCurrentYear(), lhcPeriod.Data());
1330
1331         fgkMainRefStorage.Form("alien://folder=%s%d/%s/Reference?user=alidaq?cacheFold=/tmp/OCDBCache",
1332                                         fConfig->GetAlienPath(), GetCurrentYear(), lhcPeriod.Data());
1333
1334         // Loop on detectors in the configuration
1335         TIter iter(fConfig->GetDetectors());
1336         TObjString* aDetector = 0;
1337
1338         Bool_t first = kTRUE;
1339
1340         while ((aDetector = (TObjString*) iter.Next()))
1341         {
1342                 fCurrentDetector = aDetector->String();
1343
1344                 if (ContinueProcessing() == kFALSE)
1345                         continue;
1346
1347                 if (first)
1348                 {
1349                   // only read QueryCDB when needed and only once
1350                   AliCDBStorage *mainCDBSto = AliCDBManager::Instance()->GetStorage(fgkMainCDB);
1351                   if(mainCDBSto) mainCDBSto->QueryCDB(GetCurrentRun());
1352                   AliCDBStorage *mainRefSto = AliCDBManager::Instance()->GetStorage(fgkMainRefStorage);
1353                   if(mainRefSto) mainRefSto->QueryCDB(GetCurrentRun());
1354                   first = kFALSE;
1355                 }
1356
1357                 Log("SHUTTLE", Form("\t\t\t****** run %d - %s: START  ******",
1358                                                 GetCurrentRun(), aDetector->GetName()));
1359
1360                 for(Int_t iSys=0;iSys<3;iSys++) fFXSCalled[iSys]=kFALSE;
1361
1362                 Int_t initialMem = GetMem(getpid());
1363                 Log("SHUTTLE", Form("Memory consumption before forking is %d", initialMem));
1364
1365                 Log(fCurrentDetector.Data(), "Process - Starting processing");
1366
1367                 Int_t pid = fork();
1368
1369                 if (pid < 0)
1370                 {
1371                         Log("SHUTTLE", "Process - ERROR: Forking failed");
1372                 }
1373                 else if (pid > 0)
1374                 {
1375                         // parent
1376                         Log("SHUTTLE", Form("Process - In parent process of %d - %s: Starting monitoring",
1377                                                         GetCurrentRun(), aDetector->GetName()));
1378
1379                         Long_t begin = time(0);
1380
1381                         int status; // to be used with waitpid, on purpose an int (not Int_t)!
1382                         while (waitpid(pid, &status, WNOHANG) == 0)
1383                         {
1384                                 Long_t expiredTime = time(0) - begin;
1385
1386                                 if (expiredTime > fConfig->GetPPTimeOut())
1387                                 {
1388                                         TString logMsg;
1389                                         AliShuttleStatus *currentStatus = ReadShuttleStatus();
1390                                         AliShuttleStatus::Status newStatus = AliShuttleStatus::kInvalid;
1391
1392                                         if (currentStatus->GetStatus() <= AliShuttleStatus::kPPDone)
1393                                         {
1394                                                 // in case pp not yet done set status to kPPTimeOut
1395
1396                                                 logMsg.Form("Process - Process of %s timed out. Run time: %d seconds. Killing...",
1397                                                                 fCurrentDetector.Data(), expiredTime);
1398                                                 newStatus = AliShuttleStatus::kPPTimeOut;
1399                                         }
1400                                         else if (currentStatus->GetStatus() == AliShuttleStatus::kStoreStarted)
1401                                         {
1402                                                 // in case the pp goes in TimeOut while storing the objects in the OCDB
1403                                                 // set status to kStoreError
1404
1405                                                 logMsg.Form("Process - Process of %s timed out while storing the OCDB object. Run time: %d seconds. Killing... and setting status to StoreError.",
1406                                                                 fCurrentDetector.Data(), expiredTime);
1407                                                 newStatus = AliShuttleStatus::kStoreError;
1408                                         }
1409                                         else
1410                                         {
1411                                                 // in other cases don't change the status
1412
1413                                                 logMsg.Form("Process - Process of %s timed out in status = %s. Run time: %d seconds. Killing... without changing the status",
1414                                                                 fCurrentDetector.Data(), currentStatus->GetStatusName(), expiredTime);
1415                                         }
1416
1417                                         Log("SHUTTLE", logMsg);
1418                                         Log(fCurrentDetector, logMsg);
1419
1420                                         kill(pid, 9);
1421
1422                                         if (newStatus != AliShuttleStatus::kInvalid)
1423                                                 UpdateShuttleStatus(newStatus);
1424                                         hasError = kTRUE;
1425
1426                                         gSystem->Sleep(1000);
1427                                 }
1428                                 else
1429                                 {
1430                                         gSystem->Sleep(1000);
1431
1432                                         Int_t mem = GetMem(pid);
1433
1434                                         if (mem < 0)
1435                                                 continue;
1436
1437                                         mem -= initialMem;
1438                                         if (mem < 0)
1439                                                 mem = 0;
1440
1441                                         if (expiredTime % 60 == 0)
1442                                         {
1443                                                 Log("SHUTTLE", Form("Process - %s: Checking process. "
1444                                                         "Run time: %d seconds - Memory consumption: %d KB",
1445                                                         fCurrentDetector.Data(), expiredTime, mem));
1446                                                 SendAlive();
1447                                         }
1448
1449                                         if (mem > fConfig->GetPPMaxMem())
1450                                         {
1451                                                 TString tmp;
1452                                                 tmp.Form("Process - Process exceeds maximum allowed memory "
1453                                                         "(%d KB > %d KB). Killing...",
1454                                                         mem, fConfig->GetPPMaxMem());
1455                                                 Log("SHUTTLE", tmp);
1456                                                 Log(fCurrentDetector, tmp);
1457
1458                                                 kill(pid, 9);
1459
1460                                                 UpdateShuttleStatus(AliShuttleStatus::kPPOutOfMemory);
1461                                                 hasError = kTRUE;
1462
1463                                                 gSystem->Sleep(1000);
1464                                         }
1465                                 }
1466                         }
1467
1468                         Log("SHUTTLE", Form("Process - In parent process of %d - %s: Client has terminated.",
1469                                                                 GetCurrentRun(), aDetector->GetName()));
1470
1471                         if (WIFEXITED(status))
1472                         {
1473                                 Int_t returnCode = WEXITSTATUS(status);
1474
1475                                 Log("SHUTTLE", Form("Process - %s: the return code is %d", fCurrentDetector.Data(),
1476                                                                                 returnCode));
1477
1478                                 if (returnCode == 0) hasError = kTRUE;
1479                         }
1480                 }
1481                 else if (pid == 0)
1482                 {
1483                         // child
1484                         Log("SHUTTLE", Form("Process - In child process of %d - %s", GetCurrentRun(),
1485                                 aDetector->GetName()));
1486
1487                         Log("SHUTTLE", Form("Process - Redirecting output to %s log",fCurrentDetector.Data()));
1488
1489                         if ((freopen(GetLogFileName(fCurrentDetector), "a", stdout)) == 0)
1490                         {
1491                                 Log("SHUTTLE", "Process - Could not freopen stdout");
1492                         }
1493                         else
1494                         {
1495                                 fOutputRedirected = kTRUE;
1496                                 if ((dup2(fileno(stdout), fileno(stderr))) < 0)
1497                                         Log("SHUTTLE", "Process - Could not redirect stderr");
1498
1499                         }
1500
1501                         TString wd = gSystem->WorkingDirectory();
1502                         TString tmpDir = Form("%s/%s_%d_process", GetShuttleTempDir(),
1503                                 fCurrentDetector.Data(), GetCurrentRun());
1504
1505                         Int_t result = gSystem->GetPathInfo(tmpDir.Data(), 0, (Long64_t*) 0, 0, 0);
1506                         if (!result) // temp dir already exists!
1507                         {
1508                                 Log(fCurrentDetector.Data(),
1509                                         Form("Process - %s dir already exists! Removing...", tmpDir.Data()));
1510                                 gSystem->Exec(Form("rm -rf %s",tmpDir.Data()));
1511                         }
1512
1513                         if (gSystem->mkdir(tmpDir.Data(), 1))
1514                         {
1515                                 Log(fCurrentDetector.Data(), "Process - could not make temp directory!!");
1516                                 gSystem->Exit(1);
1517                         }
1518
1519                         if (!gSystem->ChangeDirectory(tmpDir.Data()))
1520                         {
1521                                 Log(fCurrentDetector.Data(), "Process - could not change directory!!");
1522                                 gSystem->Exit(1);
1523                         }
1524
1525                         Int_t success = ProcessCurrentDetector();
1526
1527                         gSystem->ChangeDirectory(wd.Data());
1528
1529                         if (success == 1) // Preprocessor finished successfully!
1530                         {
1531                                 // remove temporary folder or DCS map
1532                                 if (!fConfig->KeepTempFolder())
1533                                 {
1534                                         gSystem->Exec(Form("rm -rf %s",tmpDir.Data()));
1535                                 } else if (!fConfig->KeepDCSMap())
1536                                 {
1537                                         gSystem->Exec(Form("rm -f %s/DCSMap.root",tmpDir.Data()));
1538                                 }
1539
1540                                 // Update time_processed field in FXS DB
1541                                 if (UpdateTable() == kFALSE)
1542                                         Log("SHUTTLE", Form("Process - %s: Could not update FXS databases!",
1543                                                         fCurrentDetector.Data()));
1544
1545                                 // Transfer the data from local storage to main storage (Grid)
1546                                 if (StoreOCDB() == kFALSE)
1547                                         success = kFALSE;
1548                         }
1549                         else if (success == 0)
1550                         {
1551                                 Log("SHUTTLE",
1552                                         Form("\t\t\t****** run %d - %s: PP ERROR ******",
1553                                                 GetCurrentRun(), aDetector->GetName()));
1554                         }
1555
1556                         for (UInt_t iSys=0; iSys<3; iSys++)
1557                         {
1558                                 if (fFXSCalled[iSys]) fFXSlist[iSys].Clear();
1559                         }
1560
1561                         Log("SHUTTLE", Form("Process - Client process of %d - %s is exiting now with %d.",
1562                                                         GetCurrentRun(), aDetector->GetName(), success));
1563
1564                         // the client exits here
1565                         gSystem->Exit(success);
1566
1567                         AliError("We should never get here!!!");
1568                 }
1569         }
1570
1571         Log("SHUTTLE", Form("\t\t\t^*^*^*^*^*^*^*^*^*^*^*^* run %d: FINISH ^*^*^*^*^*^*^*^*^*^*^*^*",
1572                                                         GetCurrentRun()));
1573
1574         //check if shuttle is done for this run, if so update logbook
1575         TObjArray checkEntryArray;
1576         checkEntryArray.SetOwner(1);
1577         TString whereClause = Form("where run=%d", GetCurrentRun());
1578         if (!QueryShuttleLogbook(whereClause.Data(), checkEntryArray) ||
1579                         checkEntryArray.GetEntries() == 0) {
1580                 Log("SHUTTLE", Form("Process - Warning: Cannot check status of run %d on Shuttle logbook!",
1581                                                 GetCurrentRun()));
1582                 return hasError == kFALSE;
1583         }
1584
1585         AliShuttleLogbookEntry* checkEntry = dynamic_cast<AliShuttleLogbookEntry*>
1586                                                 (checkEntryArray.At(0));
1587
1588         if (checkEntry)
1589         {
1590                 if (checkEntry->IsDone())
1591                 {
1592                         Log("SHUTTLE","Process - Shuttle is DONE. Updating logbook");
1593                         UpdateShuttleLogbook("shuttle_done");
1594                 }
1595                 else
1596                 {
1597                         for (UInt_t iDet=0; iDet<NDetectors(); iDet++)
1598                         {
1599                                 if (checkEntry->GetDetectorStatus(iDet) == AliShuttleLogbookEntry::kUnprocessed)
1600                                 {
1601                                         AliDebug(2, Form("Run %d: setting %s as \"not first time unprocessed\"",
1602                                                         checkEntry->GetRun(), GetDetName(iDet)));
1603                                         fFirstUnprocessed[iDet] = kFALSE;
1604                                 }
1605                         }
1606                         SendMLRunInfo("Pending");
1607                 }
1608         }
1609
1610         fLogbookEntry = 0;
1611
1612         return hasError == kFALSE;
1613 }
1614
1615 //______________________________________________________________________________________________
1616 Int_t AliShuttle::ProcessCurrentDetector()
1617 {
1618         //
1619         // Makes data retrieval just for a specific detector (fCurrentDetector).
1620         // Threre should be a configuration for this detector.
1621
1622         Log("SHUTTLE", Form("ProcessCurrentDetector - Retrieving values for %s, run %d",
1623                                                 fCurrentDetector.Data(), GetCurrentRun()));
1624
1625         TString wd = gSystem->WorkingDirectory();
1626
1627         if (!CleanReferenceStorage(fCurrentDetector.Data()))
1628                 return 0;
1629
1630         gSystem->ChangeDirectory(wd.Data());
1631
1632         // call preprocessor
1633         AliPreprocessor* aPreprocessor =
1634                 dynamic_cast<AliPreprocessor*> (fPreprocessorMap.GetValue(fCurrentDetector));
1635
1636         // check if the preprocessor wants to process this run type
1637         if (aPreprocessor->ProcessRunType() == kFALSE)
1638         {
1639                 UpdateShuttleStatus(AliShuttleStatus::kSkipped);
1640                 UpdateShuttleLogbook(fCurrentDetector, "DONE");
1641                 if (!UpdateTableSkippedCase(fCurrentDetector.Data()))
1642                 {
1643                         AliError(Form("Could not update FXS tables for run %d !", GetCurrentRun()));
1644                 }
1645                 Log(fCurrentDetector, Form("ProcessCurrentDetector - %s preprocessor is not interested in this run type", fCurrentDetector.Data()));
1646
1647                 return 2;
1648         }
1649
1650         TMap* dcsMap = new TMap();
1651
1652         aPreprocessor->Initialize(GetCurrentRun(), GetCurrentStartTime(), GetCurrentEndTime());
1653
1654         Bool_t processDCS = aPreprocessor->ProcessDCS();
1655
1656         if (!processDCS)
1657         {
1658                 Log(fCurrentDetector, "ProcessCurrentDetector -"
1659                         " The preprocessor requested to skip the retrieval of DCS values");
1660         }
1661         else if (fTestMode & kSkipDCS)
1662         {
1663                 Log(fCurrentDetector, "ProcessCurrentDetector - In TESTMODE: Skipping DCS processing");
1664         }
1665         else if (fTestMode & kErrorDCS)
1666         {
1667                 Log(fCurrentDetector, "ProcessCurrentDetector - In TESTMODE: Simulating DCS error");
1668                 UpdateShuttleStatus(AliShuttleStatus::kDCSStarted);
1669                 UpdateShuttleStatus(AliShuttleStatus::kDCSError);
1670                 delete dcsMap;
1671                 return 0;
1672         } else {
1673
1674                 UpdateShuttleStatus(AliShuttleStatus::kDCSStarted);
1675
1676                 // Query DCS archive
1677                 Int_t nServers = fConfig->GetNServers(fCurrentDetector);
1678
1679                 for (int iServ=0; iServ<nServers; iServ++)
1680                 {
1681
1682                         TString host(fConfig->GetDCSHost(fCurrentDetector, iServ));
1683                         Int_t port = fConfig->GetDCSPort(fCurrentDetector, iServ);
1684                         Int_t multiSplit = fConfig->GetMultiSplit(fCurrentDetector, iServ);
1685
1686                         Log(fCurrentDetector, Form("ProcessCurrentDetector -"
1687                                         " Querying DCS Amanda server %s:%d (%d of %d)",
1688                                         host.Data(), port, iServ+1, nServers));
1689
1690                         TMap* aliasMap = 0;
1691                         TMap* dpMap = 0;
1692
1693                         if (fConfig->GetDCSAliases(fCurrentDetector, iServ)->GetEntries() > 0)
1694                         {
1695                                 Log(fCurrentDetector, Form("Querying %d DCS aliases", fConfig->GetDCSAliases(fCurrentDetector, iServ)->GetEntries()));
1696                                 aliasMap = GetValueSet(host, port,
1697                                                 fConfig->GetDCSAliases(fCurrentDetector, iServ),
1698                                                 kAlias, multiSplit);
1699                                 if (!aliasMap)
1700                                 {
1701                                         Log(fCurrentDetector,
1702                                                 Form("ProcessCurrentDetector -"
1703                                                         " Error retrieving DCS aliases from server %s."
1704                                                         " Sending mail to DCS experts!", host.Data()));
1705                                         UpdateShuttleStatus(AliShuttleStatus::kDCSError);
1706
1707                                         if (!SendMail(kDCSEMail))
1708                                                 Log("SHUTTLE", Form("ProcessCurrentDetector - "
1709                                                                     "Could not send mail to DCS experts!"));
1710
1711                                         delete dcsMap;
1712                                         return 0;
1713                                 }
1714                         }
1715
1716                         if (fConfig->GetDCSDataPoints(fCurrentDetector, iServ)->GetEntries() > 0)
1717                         {
1718                                 Log(fCurrentDetector, Form("Querying %d DCS data points", fConfig->GetDCSDataPoints(fCurrentDetector, iServ)->GetEntries()));
1719                                 dpMap = GetValueSet(host, port,
1720                                                 fConfig->GetDCSDataPoints(fCurrentDetector, iServ),
1721                                                 kDP, multiSplit);
1722                                 if (!dpMap)
1723                                 {
1724                                         Log(fCurrentDetector,
1725                                                 Form("ProcessCurrentDetector -"
1726                                                         " Error retrieving DCS data points from server %s."
1727                                                         " Sending mail to DCS experts!", host.Data()));
1728                                         UpdateShuttleStatus(AliShuttleStatus::kDCSError);
1729
1730                                         if (!SendMail(kDCSEMail))
1731                                                 Log("SHUTTLE", Form("ProcessCurrentDetector - "
1732                                                                     "Could not send mail to DCS experts!"));
1733
1734                                         if (aliasMap) delete aliasMap;
1735                                         delete dcsMap;
1736                                         return 0;
1737                                 }
1738                         }
1739
1740                         // merge aliasMap and dpMap into dcsMap
1741                         if(aliasMap) {
1742                                 TIter iter(aliasMap);
1743                                 TObjString* key = 0;
1744                                 while ((key = (TObjString*) iter.Next()))
1745                                         dcsMap->Add(key, aliasMap->GetValue(key->String()));
1746
1747                                 aliasMap->SetOwner(kFALSE);
1748                                 delete aliasMap;
1749                         }
1750
1751                         if(dpMap) {
1752                                 TIter iter(dpMap);
1753                                 TObjString* key = 0;
1754                                 while ((key = (TObjString*) iter.Next()))
1755                                         dcsMap->Add(key, dpMap->GetValue(key->String()));
1756
1757                                 dpMap->SetOwner(kFALSE);
1758                                 delete dpMap;
1759                         }
1760                 }
1761         }
1762
1763         // save map into file, to help debugging in case of preprocessor error
1764         TFile* f = TFile::Open("DCSMap.root","recreate");
1765         f->cd();
1766         dcsMap->Write("DCSMap", TObject::kSingleKey);
1767         f->Close();
1768         delete f;
1769
1770         // DCS Archive DB processing successful. Call Preprocessor!
1771         UpdateShuttleStatus(AliShuttleStatus::kPPStarted);
1772
1773         fFXSError = -1; // this variable is kTRUE after ::Process if an FXS error occured
1774
1775         UInt_t returnValue = aPreprocessor->Process(dcsMap);
1776
1777         if (fFXSError!=-1) {
1778                 UpdateShuttleStatus(AliShuttleStatus::kFXSError);
1779                 SendMail(kFXSEMail, fFXSError);
1780                 dcsMap->DeleteAll();
1781                 delete dcsMap;
1782                 return 0;
1783         }
1784
1785         if (returnValue > 0) // Preprocessor error!
1786         {
1787                 Log(fCurrentDetector, Form("ProcessCurrentDetector - "
1788                                 "Preprocessor failed. Process returned %d.", returnValue));
1789                 UpdateShuttleStatus(AliShuttleStatus::kPPError);
1790                 dcsMap->DeleteAll();
1791                 delete dcsMap;
1792                 return 0;
1793         }
1794
1795         // preprocessor ok!
1796         UpdateShuttleStatus(AliShuttleStatus::kPPDone);
1797         Log(fCurrentDetector, Form("ProcessCurrentDetector - %s preprocessor returned success",
1798                                 fCurrentDetector.Data()));
1799
1800         dcsMap->DeleteAll();
1801         delete dcsMap;
1802
1803         return 1;
1804 }
1805
1806 //______________________________________________________________________________________________
1807 void AliShuttle::CountOpenRuns()
1808 {
1809         // Query DAQ's Shuttle logbook and sends the number of open runs to ML
1810
1811         SendAlive();
1812
1813         // check connection, in case connect
1814         if (!Connect(3))
1815                 return;
1816
1817         TString sqlQuery;
1818         sqlQuery = Form("select count(*) from %s where shuttle_done=0", fConfig->GetShuttlelbTable());
1819
1820         TSQLResult* aResult = fServer[3]->Query(sqlQuery);
1821         if (!aResult) {
1822                 AliError(Form("Can't execute query <%s>!", sqlQuery.Data()));
1823                 return;
1824         }
1825
1826         AliDebug(2,Form("Query = %s", sqlQuery.Data()));
1827
1828         if (aResult->GetRowCount() == 0) {
1829                 AliError(Form("No result for query %s received", sqlQuery.Data()));
1830                 return;
1831         }
1832
1833         if (aResult->GetFieldCount() != 1) {
1834                 AliError(Form("Invalid field count for query %s received", sqlQuery.Data()));
1835                 return;
1836         }
1837
1838         TSQLRow* aRow = aResult->Next();
1839         if (!aRow) {
1840                 AliError(Form("Could not receive result of query %s", sqlQuery.Data()));
1841                 return;
1842         }
1843
1844         TString result(aRow->GetField(0), aRow->GetFieldLength(0));
1845         Int_t count = result.Atoi();
1846
1847         Log("SHUTTLE", Form("%d unprocessed runs", count));
1848
1849         delete aRow;
1850         delete aResult;
1851
1852         TMonaLisaValue mlStatus("SHUTTLE_openruns", count);
1853
1854         TList mlList;
1855         mlList.Add(&mlStatus);
1856
1857         fMonaLisa->SendParameters(&mlList, "__PROCESSINGINFO__");
1858 }
1859
1860 //______________________________________________________________________________________________
1861 Bool_t AliShuttle::QueryShuttleLogbook(const char* whereClause,
1862                 TObjArray& entries)
1863 {
1864         // Query DAQ's Shuttle logbook and fills detector status object.
1865         // Call QueryRunParameters to query DAQ logbook for run parameters.
1866         //
1867
1868         entries.SetOwner(1);
1869
1870         // check connection, in case connect
1871         if (!Connect(3)) return kFALSE;
1872
1873         TString sqlQuery;
1874         sqlQuery = Form("select * from %s %s order by run", fConfig->GetShuttlelbTable(), whereClause);
1875
1876         TSQLResult* aResult = fServer[3]->Query(sqlQuery);
1877         if (!aResult) {
1878                 AliError(Form("Can't execute query <%s>!", sqlQuery.Data()));
1879                 return kFALSE;
1880         }
1881
1882         AliDebug(2,Form("Query = %s", sqlQuery.Data()));
1883
1884         if(aResult->GetRowCount() == 0) {
1885                 Log("SHUTTLE", "No entries in Shuttle Logbook match request");
1886                 delete aResult;
1887                 return kTRUE;
1888         }
1889
1890         // TODO Check field count!
1891         const UInt_t nCols = 23;
1892         if (aResult->GetFieldCount() != (Int_t) nCols) {
1893                 Log("SHUTTLE", "Invalid SQL result field number!");
1894                 delete aResult;
1895                 return kFALSE;
1896         }
1897
1898         TSQLRow* aRow;
1899         while ((aRow = aResult->Next())) {
1900                 TString runString(aRow->GetField(0), aRow->GetFieldLength(0));
1901                 Int_t run = runString.Atoi();
1902
1903                 AliShuttleLogbookEntry *entry = QueryRunParameters(run);
1904                 if (!entry)
1905                         continue;
1906
1907                 // loop on detectors
1908                 for(UInt_t ii = 0; ii < nCols; ii++)
1909                         entry->SetDetectorStatus(aResult->GetFieldName(ii), aRow->GetField(ii));
1910
1911                 entries.AddLast(entry);
1912                 delete aRow;
1913         }
1914
1915         delete aResult;
1916         return kTRUE;
1917 }
1918
1919 //______________________________________________________________________________________________
1920 AliShuttleLogbookEntry* AliShuttle::QueryRunParameters(Int_t run)
1921 {
1922         //
1923         // Retrieve run parameters written in the DAQ logbook and sets them into AliShuttleLogbookEntry object
1924         //
1925
1926         // check connection, in case connect
1927         if (!Connect(3))
1928                 return 0;
1929
1930         TString sqlQuery;
1931         sqlQuery.Form("select * from %s where run=%d", fConfig->GetDAQlbTable(), run);
1932
1933         TSQLResult* aResult = fServer[3]->Query(sqlQuery);
1934         if (!aResult) {
1935                 Log("SHUTTLE", Form("Can't execute query <%s>!", sqlQuery.Data()));
1936                 return 0;
1937         }
1938
1939         if (aResult->GetRowCount() == 0) {
1940                 Log("SHUTTLE", Form("QueryRunParameters - No entry in DAQ Logbook for run %d. Skipping", run));
1941                 delete aResult;
1942                 return 0;
1943         }
1944
1945         if (aResult->GetRowCount() > 1) {
1946                 Log("SHUTTLE", Form("QueryRunParameters - UNEXPECTED: "
1947                                 "more than one entry in DAQ Logbook for run %d!", run));
1948                 delete aResult;
1949                 return 0;
1950         }
1951
1952         TSQLRow* aRow = aResult->Next();
1953         if (!aRow)
1954         {
1955                 Log("SHUTTLE", Form("QueryRunParameters - Could not retrieve row for run %d. Skipping", run));
1956                 delete aResult;
1957                 return 0;
1958         }
1959
1960         AliShuttleLogbookEntry* entry = new AliShuttleLogbookEntry(run);
1961
1962         for (Int_t ii = 0; ii < aResult->GetFieldCount(); ii++)
1963                 entry->SetRunParameter(aResult->GetFieldName(ii), aRow->GetField(ii));
1964
1965         delete aRow;
1966         delete aResult;
1967
1968         UInt_t startTime = entry->GetStartTime();
1969         UInt_t endTime = entry->GetEndTime();
1970         Bool_t ecsSuccess = entry->GetECSSuccess();
1971
1972         TString totEventsStr = entry->GetRunParameter("totalEvents");
1973         Int_t totEvents = totEventsStr.Atoi();
1974
1975         UInt_t now = time(0);
1976         // TODO make this a configuration parameter
1977         Int_t dcsDelay = 120+180;
1978
1979         // runs are accepted if they have ecsSuccess set or more than 1 event
1980         if (startTime != 0 && endTime != 0 && endTime > startTime && (totEvents > 1 || ecsSuccess) && (endTime < now - dcsDelay))
1981         {
1982                 if (ecsSuccess == kFALSE)
1983                         Log("SHUTTLE", Form("Processing run %d although in status ECS failure, Reason: %s", run, entry->GetRunParameter("eor_reason")));
1984                 return entry;
1985         }
1986
1987         Bool_t skip = kFALSE;
1988
1989         if (endTime != 0 && endTime >= now - dcsDelay)
1990         {
1991                 Log("SHUTTLE", Form("Skipping run %d for now, because DCS buffer time is not yet expired", run));
1992         }
1993         else if (totEvents <= 1)
1994         {
1995                 Log("SHUTTLE", Form("QueryRunParameters - Run %d has 1 event or less - Skipping!", run));
1996                 skip = kTRUE;
1997         }
1998         else
1999         {
2000                 Log("SHUTTLE", Form("QueryRunParameters - Invalid parameters for Run %d: "
2001                         "startTime = %d, endTime = %d. Skipping (Shuttle won't be marked as DONE)!",
2002                         run, startTime, endTime));
2003         }
2004
2005         if (skip)
2006         {
2007                 Log("SHUTTLE", Form("Marking SHUTTLE skipped for run %d", run));
2008                 fLogbookEntry = entry;
2009                 if (!UpdateShuttleLogbook("shuttle_skipped"))
2010                 {
2011                         AliError(Form("Could not update logbook for run %d !", run));
2012                 }
2013                 if (!UpdateTableSkippedCase("ALL"))
2014                 {
2015                         AliError(Form("Could not update FXS tables for run %d !", run));
2016                 }
2017                 fLogbookEntry = 0;
2018         }
2019
2020         delete entry;
2021         return 0;
2022 }
2023
2024 //______________________________________________________________________________________________
2025 TMap* AliShuttle::GetValueSet(const char* host, Int_t port, const TSeqCollection* entries,
2026                               DCSType type, Int_t multiSplit)
2027 {
2028         // Retrieve all "entry" data points from the DCS server
2029         // host, port: TSocket connection parameters
2030         // entries: list of name of the alias or data point
2031         // type: kAlias or kDP
2032         // returns TMap of values, 0 when failure
2033
2034         AliDCSClient client(host, port, fTimeout, fRetries, multiSplit);
2035         Int_t offset = 180;
2036
2037         TMap* result = 0;
2038         if (type == kAlias)
2039         {
2040                 result = client.GetAliasValues(entries, GetCurrentStartTime()-offset,
2041                         GetCurrentEndTime()+offset);
2042         }
2043         else if (type == kDP)
2044         {
2045                 result = client.GetDPValues(entries, GetCurrentStartTime()-offset,
2046                         GetCurrentEndTime()+offset);
2047         }
2048
2049         if (result == 0)
2050         {
2051                 Log(fCurrentDetector.Data(), Form("GetValueSet - Can't get entries! Reason: %s",
2052                         client.GetErrorString(client.GetResultErrorCode())));
2053                 if (client.GetResultErrorCode() == AliDCSClient::fgkServerError)
2054                         Log(fCurrentDetector.Data(), Form("GetValueSet - Server error code: %s",
2055                                 client.GetServerError().Data()));
2056
2057                 return 0;
2058         }
2059
2060         return result;
2061 }
2062
2063 //______________________________________________________________________________________________
2064 const char* AliShuttle::GetFile(Int_t system, const char* detector,
2065                 const char* id, const char* source)
2066 {
2067         // Get calibration file from file exchange servers
2068         // First queris the FXS database for the file name, using the run, detector, id and source info
2069         // then calls RetrieveFile(filename) for actual copy to local disk
2070         // run: current run being processed (given by Logbook entry fLogbookEntry)
2071         // detector: the Preprocessor name
2072         // id: provided as a parameter by the Preprocessor
2073         // source: provided by the Preprocessor through GetFileSources function
2074
2075         // check if test mode should simulate a FXS error
2076         if (fTestMode & kErrorFXSFiles)
2077         {
2078                 Log(detector, Form("GetFile - In TESTMODE - Simulating error while connecting to %s FXS", GetSystemName(system)));
2079                 return 0;
2080         }
2081
2082         // check connection, in case connect
2083         if (!Connect(system))
2084         {
2085                 Log(detector, Form("GetFile - Couldn't connect to %s FXS database", GetSystemName(system)));
2086                 fFXSError = system;
2087                 return 0;
2088         }
2089
2090         // Query preparation
2091         TString sourceName(source);
2092         Int_t nFields = 3;
2093         TString sqlQueryStart = Form("select filePath,size,fileChecksum from %s where",
2094                                                                 fConfig->GetFXSdbTable(system));
2095         TString whereClause = Form("run=%d and detector=\"%s\" and fileId=\"%s\"",
2096                                                                 GetCurrentRun(), detector, id);
2097
2098         if (system == kDAQ)
2099         {
2100                 whereClause += Form(" and DAQsource=\"%s\"", source);
2101         }
2102         else if (system == kDCS)
2103         {
2104                 sourceName="none";
2105         }
2106         else if (system == kHLT)
2107         {
2108                 whereClause += Form(" and DDLnumbers=\"%s\"", source);
2109         }
2110
2111         TString sqlQuery = Form("%s %s", sqlQueryStart.Data(), whereClause.Data());
2112
2113         AliDebug(2, Form("SQL query: \n%s",sqlQuery.Data()));
2114
2115         // Query execution
2116         TSQLResult* aResult = 0;
2117         aResult = dynamic_cast<TSQLResult*> (fServer[system]->Query(sqlQuery));
2118         if (!aResult) {
2119                 Log(detector, Form("GetFile - Can't execute SQL query to %s database for: id = %s, source = %s",
2120                                 GetSystemName(system), id, sourceName.Data()));
2121                 fFXSError = system;
2122                 return 0;
2123         }
2124
2125         if (aResult->GetRowCount() == 0)
2126         {
2127                 Log(detector,
2128                         Form("GetFile - No entry in %s FXS db for: id = %s, source = %s",
2129                                 GetSystemName(system), id, sourceName.Data()));
2130                 delete aResult;
2131                 return 0;
2132         }
2133
2134         if (aResult->GetRowCount() > 1) {
2135                 Log(detector,
2136                         Form("GetFile - More than one entry in %s FXS db for: id = %s, source = %s",
2137                                 GetSystemName(system), id, sourceName.Data()));
2138                 fFXSError = system;
2139                 delete aResult;
2140                 return 0;
2141         }
2142
2143         if (aResult->GetFieldCount() != nFields) {
2144                 Log(detector,
2145                         Form("GetFileName - Wrong field count in %s FXS db for: id = %s, source = %s",
2146                                 GetSystemName(system), id, sourceName.Data()));
2147                 fFXSError = system;
2148                 delete aResult;
2149                 return 0;
2150         }
2151
2152         TSQLRow* aRow = dynamic_cast<TSQLRow*> (aResult->Next());
2153
2154         if (!aRow){
2155                 Log(detector, Form("GetFile - Empty set result in %s FXS db from query: id = %s, source = %s",
2156                                 GetSystemName(system), id, sourceName.Data()));
2157                 fFXSError = system;
2158                 delete aResult;
2159                 return 0;
2160         }
2161
2162         TString filePath(aRow->GetField(0), aRow->GetFieldLength(0));
2163         TString fileSize(aRow->GetField(1), aRow->GetFieldLength(1));
2164         TString fileChecksum(aRow->GetField(2), aRow->GetFieldLength(2));
2165
2166         delete aResult;
2167         delete aRow;
2168
2169         AliDebug(2, Form("filePath = %s; size = %s, fileChecksum = %s",
2170                                 filePath.Data(), fileSize.Data(), fileChecksum.Data()));
2171
2172         // retrieved file is renamed to make it unique
2173         TString localFileName = Form("%s/%s_%d_process/%s_%s_%d_%s_%s.shuttle",
2174                                         GetShuttleTempDir(), detector, GetCurrentRun(),
2175                                         GetSystemName(system), detector, GetCurrentRun(),
2176                                         id, sourceName.Data());
2177
2178
2179         // file retrieval from FXS
2180         UInt_t nRetries = 0;
2181         UInt_t maxRetries = 3;
2182         Bool_t result = kFALSE;
2183
2184         // copy!! if successful TSystem::Exec returns 0
2185         while (nRetries++ < maxRetries) {
2186                 AliDebug(2, Form("Trying to copy file. Retry # %d", nRetries));
2187                 result = RetrieveFile(system, filePath.Data(), localFileName.Data());
2188                 if (!result)
2189                 {
2190                         Log(detector, Form("GetFile - Copy of file %s from %s FXS failed",
2191                                         filePath.Data(), GetSystemName(system)));
2192                         continue;
2193                 }
2194
2195                 if (fileSize.Length()>0)
2196                 {
2197                         // compare filesize of local file with the one stored in the FXS DB
2198                         Long_t size = -1;
2199                         Int_t sizeComp = gSystem->GetPathInfo(localFileName.Data(), 0, &size, 0, 0);
2200
2201                         if (sizeComp != 0 || size != fileSize.Atoi())
2202                         {
2203                                 Log(detector, Form("GetFile - size of file %s does not match with local copy!",
2204                                                         filePath.Data()));
2205                                 result = kFALSE;
2206                                 continue;
2207                         }
2208
2209                 } else {
2210                         Log(fCurrentDetector, Form("GetFile - size of file %s not set in %s database, skipping comparison",
2211                                                 filePath.Data(), GetSystemName(system)));
2212                 }
2213
2214                 if (fileChecksum.Length()>0)
2215                 {
2216                         // compare md5sum of local file with the one stored in the FXS DB
2217                         if(fileChecksum.Contains(' ')) fileChecksum.Resize(fileChecksum.First(' '));
2218                         Int_t md5Comp = gSystem->Exec(Form("md5sum %s |grep %s > /dev/null 2> /dev/null",
2219                                                 localFileName.Data(), fileChecksum.Data()));
2220
2221                         if (md5Comp != 0)
2222                         {
2223                                 Log(detector, Form("GetFile - md5sum of file %s does not match with local copy!",
2224                                                         filePath.Data()));
2225                                 result = kFALSE;
2226                                 continue;
2227                         }
2228                 } else {
2229                         Log(fCurrentDetector, Form("GetFile - md5sum of file %s not set in %s database, skipping comparison",
2230                                                         filePath.Data(), GetSystemName(system)));
2231                 }
2232                 if (result) break;
2233         }
2234
2235         if (!result)
2236         {
2237                 fFXSError = system;
2238                 return 0;
2239         }
2240
2241         fFXSCalled[system]=kTRUE;
2242         TObjString *fileParams = new TObjString(Form("%s#!?!#%s", id, sourceName.Data()));
2243         fFXSlist[system].Add(fileParams);
2244
2245         static TString staticLocalFileName;
2246         staticLocalFileName.Form("%s", localFileName.Data());
2247
2248         Log(fCurrentDetector, Form("GetFile - Retrieved file with id %s and "
2249                         "source %s from %s to %s", id, source,
2250                         GetSystemName(system), localFileName.Data()));
2251
2252         return staticLocalFileName.Data();
2253 }
2254
2255 //______________________________________________________________________________________________
2256 Bool_t AliShuttle::RetrieveFile(UInt_t system, const char* fxsFileName, const char* localFileName)
2257 {
2258         //
2259         // Copies file from FXS to local Shuttle machine
2260         //
2261
2262         // check temp directory: trying to cd to temp; if it does not exist, create it
2263         AliDebug(2, Form("Copy file %s from %s FXS into %s",
2264                         GetSystemName(system), fxsFileName, localFileName));
2265
2266         TString tmpDir(localFileName);
2267
2268         tmpDir = tmpDir(0,tmpDir.Last('/'));
2269
2270         Int_t noDir = gSystem->GetPathInfo(tmpDir.Data(), 0, (Long64_t*) 0, 0, 0);
2271         if (noDir) // temp dir does not exists!
2272         {
2273                 if (gSystem->mkdir(tmpDir.Data(), 1))
2274                 {
2275                         Log(fCurrentDetector.Data(), "RetrieveFile - could not make temp directory!!");
2276                         return kFALSE;
2277                 }
2278         }
2279
2280         TString command = Form("scp -oPort=%d -2 %s@%s:%s/%s %s",
2281                 fConfig->GetFXSPort(system),
2282                 fConfig->GetFXSUser(system),
2283                 fConfig->GetFXSHost(system),
2284                 fConfig->GetFXSBaseFolder(system),
2285                 fxsFileName,
2286                 localFileName);
2287
2288         AliDebug(2, Form("%s",command.Data()));
2289
2290         Bool_t result = (gSystem->Exec(command.Data()) == 0);
2291
2292         return result;
2293 }
2294
2295 //______________________________________________________________________________________________
2296 TList* AliShuttle::GetFileSources(Int_t system, const char* detector, const char* id)
2297 {
2298         //
2299         // Get sources producing the condition file Id from file exchange servers
2300         // if id is NULL all sources are returned (distinct)
2301         //
2302
2303         if (id)
2304         {
2305                 Log(detector, Form("GetFileSources - Querying %s FXS for files with id %s produced by %s", GetSystemName(system), id, detector));
2306         } else {
2307                 Log(detector, Form("GetFileSources - Querying %s FXS for files produced by %s", GetSystemName(system), detector));
2308         }
2309
2310         // check if test mode should simulate a FXS error
2311         if (fTestMode & kErrorFXSSources)
2312         {
2313                 Log(detector, Form("GetFileSources - In TESTMODE - Simulating error while connecting to %s FXS", GetSystemName(system)));
2314                 return 0;
2315         }
2316
2317         if (system == kDCS)
2318         {
2319                 Log(detector, "GetFileSources - WARNING: DCS system has only one source of data!");
2320                 TList *list = new TList();
2321                 list->SetOwner(1);
2322                 list->Add(new TObjString(" "));
2323                 return list;
2324         }
2325
2326         // check connection, in case connect
2327         if (!Connect(system))
2328         {
2329                 Log(detector, Form("GetFileSources - Couldn't connect to %s FXS database", GetSystemName(system)));
2330                 fFXSError = system;
2331                 return NULL;
2332         }
2333
2334         TString sourceName = 0;
2335         if (system == kDAQ)
2336         {
2337                 sourceName = "DAQsource";
2338         } else if (system == kHLT)
2339         {
2340                 sourceName = "DDLnumbers";
2341         }
2342
2343         TString sqlQueryStart = Form("select distinct %s from %s where", sourceName.Data(), fConfig->GetFXSdbTable(system));
2344         TString whereClause = Form("run=%d and detector=\"%s\"",
2345                                 GetCurrentRun(), detector);
2346         if (id)
2347                 whereClause += Form(" and fileId=\"%s\"", id);
2348         TString sqlQuery = Form("%s %s", sqlQueryStart.Data(), whereClause.Data());
2349
2350         AliDebug(2, Form("SQL query: \n%s",sqlQuery.Data()));
2351
2352         // Query execution
2353         TSQLResult* aResult;
2354         aResult = fServer[system]->Query(sqlQuery);
2355         if (!aResult) {
2356                 Log(detector, Form("GetFileSources - Can't execute SQL query to %s database for id: %s",
2357                                 GetSystemName(system), id));
2358                 fFXSError = system;
2359                 return 0;
2360         }
2361
2362         TList *list = new TList();
2363         list->SetOwner(1);
2364
2365         if (aResult->GetRowCount() == 0)
2366         {
2367                 Log(detector,
2368                         Form("GetFileSources - No entry in %s FXS table for id: %s", GetSystemName(system), id));
2369                 delete aResult;
2370                 return list;
2371         }
2372
2373         Log(detector, Form("GetFileSources - Found %d sources", aResult->GetRowCount()));
2374
2375         TSQLRow* aRow;
2376         while ((aRow = aResult->Next()))
2377         {
2378
2379                 TString source(aRow->GetField(0), aRow->GetFieldLength(0));
2380                 AliDebug(2, Form("%s = %s", sourceName.Data(), source.Data()));
2381                 list->Add(new TObjString(source));
2382                 delete aRow;
2383         }
2384
2385         delete aResult;
2386
2387         return list;
2388 }
2389
2390 //______________________________________________________________________________________________
2391 TList* AliShuttle::GetFileIDs(Int_t system, const char* detector, const char* source)
2392 {
2393         //
2394         // Get all ids of condition files produced by a given source from file exchange servers
2395         //
2396
2397         Log(detector, Form("GetFileIDs - Retrieving ids with source %s with %s", source, GetSystemName(system)));
2398
2399         // check if test mode should simulate a FXS error
2400         if (fTestMode & kErrorFXSSources)
2401         {
2402                 Log(detector, Form("GetFileIDs - In TESTMODE - Simulating error while connecting to %s FXS", GetSystemName(system)));
2403                 return 0;
2404         }
2405
2406         // check connection, in case connect
2407         if (!Connect(system))
2408         {
2409                 Log(detector, Form("GetFileIDs - Couldn't connect to %s FXS database", GetSystemName(system)));
2410                 return NULL;
2411         }
2412
2413         TString sourceName = 0;
2414         if (system == kDAQ)
2415         {
2416                 sourceName = "DAQsource";
2417         } else if (system == kHLT)
2418         {
2419                 sourceName = "DDLnumbers";
2420         }
2421
2422         TString sqlQueryStart = Form("select fileId from %s where", fConfig->GetFXSdbTable(system));
2423         TString whereClause = Form("run=%d and detector=\"%s\"",
2424                                 GetCurrentRun(), detector);
2425         if (sourceName.Length() > 0 && source)
2426                 whereClause += Form(" and %s=\"%s\"", sourceName.Data(), source);
2427         TString sqlQuery = Form("%s %s", sqlQueryStart.Data(), whereClause.Data());
2428
2429         AliDebug(2, Form("SQL query: \n%s",sqlQuery.Data()));
2430
2431         // Query execution
2432         TSQLResult* aResult;
2433         aResult = fServer[system]->Query(sqlQuery);
2434         if (!aResult) {
2435                 Log(detector, Form("GetFileIDs - Can't execute SQL query to %s database for source: %s",
2436                                 GetSystemName(system), source));
2437                 return 0;
2438         }
2439
2440         TList *list = new TList();
2441         list->SetOwner(1);
2442
2443         if (aResult->GetRowCount() == 0)
2444         {
2445                 Log(detector,
2446                         Form("GetFileIDs - No entry in %s FXS table for source: %s", GetSystemName(system), source));
2447                 delete aResult;
2448                 return list;
2449         }
2450
2451         Log(detector, Form("GetFileIDs - Found %d ids", aResult->GetRowCount()));
2452
2453         TSQLRow* aRow;
2454
2455         while ((aRow = aResult->Next()))
2456         {
2457
2458                 TString id(aRow->GetField(0), aRow->GetFieldLength(0));
2459                 AliDebug(2, Form("fileId = %s", id.Data()));
2460                 list->Add(new TObjString(id));
2461                 delete aRow;
2462         }
2463
2464         delete aResult;
2465
2466         return list;
2467 }
2468
2469 //______________________________________________________________________________________________
2470 Bool_t AliShuttle::Connect(Int_t system)
2471 {
2472         // Connect to MySQL Server of the system's FXS MySQL databases
2473         // DAQ Logbook, Shuttle Logbook and DAQ FXS db are on the same host
2474         //
2475
2476         // check connection: if already connected return
2477         if(fServer[system] && fServer[system]->IsConnected()) return kTRUE;
2478
2479         TString dbHost, dbUser, dbPass, dbName;
2480
2481         if (system < 3) // FXS db servers
2482         {
2483                 dbHost = Form("mysql://%s:%d", fConfig->GetFXSdbHost(system), fConfig->GetFXSdbPort(system));
2484                 dbUser = fConfig->GetFXSdbUser(system);
2485                 dbPass = fConfig->GetFXSdbPass(system);
2486                 dbName =   fConfig->GetFXSdbName(system);
2487         } else { // Run & Shuttle logbook servers
2488         // TODO Will the Shuttle logbook server be the same as the Run logbook server ???
2489                 dbHost = Form("mysql://%s:%d", fConfig->GetDAQlbHost(), fConfig->GetDAQlbPort());
2490                 dbUser = fConfig->GetDAQlbUser();
2491                 dbPass = fConfig->GetDAQlbPass();
2492                 dbName =   fConfig->GetDAQlbDB();
2493         }
2494
2495         fServer[system] = TSQLServer::Connect(dbHost.Data(), dbUser.Data(), dbPass.Data());
2496         if (!fServer[system] || !fServer[system]->IsConnected()) {
2497                 if(system < 3)
2498                 {
2499                 AliError(Form("Can't establish connection to FXS database for %s",
2500                                         AliShuttleInterface::GetSystemName(system)));
2501                 } else {
2502                 AliError("Can't establish connection to Run logbook.");
2503                 }
2504                 if(fServer[system]) delete fServer[system];
2505                 return kFALSE;
2506         }
2507
2508         // Get tables
2509         TSQLResult* aResult=0;
2510         switch(system){
2511                 case kDAQ:
2512                         aResult = fServer[kDAQ]->GetTables(dbName.Data());
2513                         break;
2514                 case kDCS:
2515                         aResult = fServer[kDCS]->GetTables(dbName.Data());
2516                         break;
2517                 case kHLT:
2518                         aResult = fServer[kHLT]->GetTables(dbName.Data());
2519                         break;
2520                 default:
2521                         aResult = fServer[3]->GetTables(dbName.Data());
2522                         break;
2523         }
2524
2525         delete aResult;
2526         return kTRUE;
2527 }
2528
2529 //______________________________________________________________________________________________
2530 Bool_t AliShuttle::UpdateTable()
2531 {
2532         //
2533         // Update FXS table filling time_processed field in all rows corresponding to current run and detector
2534         //
2535
2536         Bool_t result = kTRUE;
2537
2538         for (UInt_t system=0; system<3; system++)
2539         {
2540                 if(!fFXSCalled[system]) continue;
2541
2542                 // check connection, in case connect
2543                 if (!Connect(system))
2544                 {
2545                         Log(fCurrentDetector, Form("UpdateTable - Couldn't connect to %s FXS database", GetSystemName(system)));
2546                         result = kFALSE;
2547                         continue;
2548                 }
2549
2550                 TTimeStamp now; // now
2551
2552                 // Loop on FXS list entries
2553                 TIter iter(&fFXSlist[system]);
2554                 TObjString *aFXSentry=0;
2555                 while ((aFXSentry = dynamic_cast<TObjString*> (iter.Next())))
2556                 {
2557                         TString aFXSentrystr = aFXSentry->String();
2558                         TObjArray *aFXSarray = aFXSentrystr.Tokenize("#!?!#");
2559                         if (!aFXSarray || aFXSarray->GetEntries() != 2 )
2560                         {
2561                                 Log(fCurrentDetector, Form("UpdateTable - error updating %s FXS entry. Check string: <%s>",
2562                                         GetSystemName(system), aFXSentrystr.Data()));
2563                                 if(aFXSarray) delete aFXSarray;
2564                                 result = kFALSE;
2565                                 continue;
2566                         }
2567                         const char* fileId = ((TObjString*) aFXSarray->At(0))->GetName();
2568                         const char* source = ((TObjString*) aFXSarray->At(1))->GetName();
2569
2570                         TString whereClause;
2571                         if (system == kDAQ)
2572                         {
2573                                 whereClause = Form("where run=%d and detector=\"%s\" and fileId=\"%s\" and DAQsource=\"%s\";",
2574                                                         GetCurrentRun(), fCurrentDetector.Data(), fileId, source);
2575                         }
2576                         else if (system == kDCS)
2577                         {
2578                                 whereClause = Form("where run=%d and detector=\"%s\" and fileId=\"%s\";",
2579                                                         GetCurrentRun(), fCurrentDetector.Data(), fileId);
2580                         }
2581                         else if (system == kHLT)
2582                         {
2583                                 whereClause = Form("where run=%d and detector=\"%s\" and fileId=\"%s\" and DDLnumbers=\"%s\";",
2584                                                         GetCurrentRun(), fCurrentDetector.Data(), fileId, source);
2585                         }
2586
2587                         delete aFXSarray;
2588
2589                         TString sqlQuery = Form("update %s set time_processed=%d %s", fConfig->GetFXSdbTable(system),
2590                                                                 now.GetSec(), whereClause.Data());
2591
2592                         AliDebug(2, Form("SQL query: \n%s",sqlQuery.Data()));
2593
2594                         // Query execution
2595                         TSQLResult* aResult;
2596                         aResult = dynamic_cast<TSQLResult*> (fServer[system]->Query(sqlQuery));
2597                         if (!aResult)
2598                         {
2599                                 Log(fCurrentDetector, Form("UpdateTable - %s db: can't execute SQL query <%s>",
2600                                                                 GetSystemName(system), sqlQuery.Data()));
2601                                 result = kFALSE;
2602                                 continue;
2603                         }
2604                         delete aResult;
2605                 }
2606         }
2607
2608         return result;
2609 }
2610
2611 //_______________________________________________________________________________
2612 Bool_t AliShuttle::UpdateTableSkippedCase(const char* detector)
2613 {
2614         //
2615         // Update FXS table filling time_processed field in all rows corresponding to current run and detector
2616         // if detector = "ALL" update all detectors
2617         //
2618
2619         Bool_t result = kTRUE;
2620
2621         for (UInt_t system=0; system<3; system++)
2622         {
2623
2624                 // check connection, in case connect
2625                 if (!Connect(system))
2626                 {
2627                         Log(fCurrentDetector, Form("UpdateTableSkippedCase - Couldn't connect to %s FXS database", GetSystemName(system)));
2628                         result = kFALSE;
2629                         continue;
2630                 }
2631
2632                 TTimeStamp now; // now
2633
2634                 // Loop on FXS list entries
2635                 TIter iter(&fFXSlist[system]);
2636
2637                 TString whereClause;
2638                 if (detector == "ALL") whereClause = Form("where run=%d and time_processed IS NULL;",GetCurrentRun());
2639                 else whereClause = Form("where run=%d and detector=\"%s\" and time_processed IS NULL;",GetCurrentRun(), detector);
2640
2641                 Log("SHUTTLE",Form(" whereClause = %s ",whereClause.Data()));
2642
2643                 TString sqlQuery = Form("update %s set time_processed=%d %s", fConfig->GetFXSdbTable(system),
2644                                         now.GetSec(), whereClause.Data());
2645
2646                 AliDebug(2, Form("SQL query: \n%s",sqlQuery.Data()));
2647
2648                 // Query execution
2649                 TSQLResult* aResult;
2650                 aResult = dynamic_cast<TSQLResult*> (fServer[system]->Query(sqlQuery));
2651                 if (!aResult)
2652                 {
2653                         Log("SHUTTLE", Form("UpdateTableSkippedCase - %s db: can't execute SQL query <%s>",
2654                                                         GetSystemName(system), sqlQuery.Data()));
2655                         result = kFALSE;
2656                         continue;
2657                 }
2658                 delete aResult;
2659
2660         }
2661
2662         return result;
2663 }
2664 //______________________________________________________________________________________________
2665 Bool_t AliShuttle::UpdateTableFailCase()
2666 {
2667         // Update FXS table filling time_processed field in all rows corresponding to current run and detector
2668         // this is called in case the preprocessor is declared failed for the current run, because
2669         // the fields are updated only in case of success
2670
2671         Bool_t result = kTRUE;
2672
2673         for (UInt_t system=0; system<3; system++)
2674         {
2675                 // check connection, in case connect
2676                 if (!Connect(system))
2677                 {
2678                         Log(fCurrentDetector, Form("UpdateTableFailCase - Couldn't connect to %s FXS database",
2679                                                         GetSystemName(system)));
2680                         result = kFALSE;
2681                         continue;
2682                 }
2683
2684                 TTimeStamp now; // now
2685
2686                 // Loop on FXS list entries
2687
2688                 TString whereClause = Form("where run=%d and detector=\"%s\";",
2689                                                 GetCurrentRun(), fCurrentDetector.Data());
2690
2691
2692                 TString sqlQuery = Form("update %s set time_processed=%d %s", fConfig->GetFXSdbTable(system),
2693                                                         now.GetSec(), whereClause.Data());
2694
2695                 AliDebug(2, Form("SQL query: \n%s",sqlQuery.Data()));
2696
2697                 // Query execution
2698                 TSQLResult* aResult;
2699                 aResult = dynamic_cast<TSQLResult*> (fServer[system]->Query(sqlQuery));
2700                 if (!aResult)
2701                 {
2702                         Log(fCurrentDetector, Form("UpdateTableFailCase - %s db: can't execute SQL query <%s>",
2703                                                         GetSystemName(system), sqlQuery.Data()));
2704                         result = kFALSE;
2705                         continue;
2706                 }
2707                 delete aResult;
2708         }
2709
2710         return result;
2711 }
2712
2713 //______________________________________________________________________________________________
2714 Bool_t AliShuttle::UpdateShuttleLogbook(const char* detector, const char* status)
2715 {
2716         //
2717         // Update Shuttle logbook filling detector or shuttle_done column
2718         // ex. of usage: UpdateShuttleLogbook("PHOS", "DONE") or UpdateShuttleLogbook("shuttle_done")
2719         //
2720
2721         // check connection, in case connect
2722         if(!Connect(3)){
2723                 Log("SHUTTLE", "UpdateShuttleLogbook - Couldn't connect to DAQ Logbook.");
2724                 return kFALSE;
2725         }
2726
2727         TString detName(detector);
2728         TString setClause;
2729         if (detName == "shuttle_done" || detName == "shuttle_skipped")
2730         {
2731                 setClause = "set shuttle_done=1";
2732
2733                 if (detName == "shuttle_done")
2734                 {
2735                         if (TouchFile() != kTRUE)
2736                         {
2737                                 SendMLRunInfo("Pending");
2738                                 return kFALSE;
2739                         }
2740
2741                         SendMLRunInfo("Done");
2742                 }
2743                 else
2744                         SendMLRunInfo("Skipped");
2745         }
2746         else {
2747                 TString statusStr(status);
2748                 if(statusStr.Contains("done", TString::kIgnoreCase) ||
2749                    statusStr.Contains("failed", TString::kIgnoreCase)){
2750                         setClause = Form("set %s=\"%s\"", detector, status);
2751                 } else {
2752                         Log("SHUTTLE",
2753                                 Form("UpdateShuttleLogbook - Invalid status <%s> for detector %s",
2754                                         status, detector));
2755                         return kFALSE;
2756                 }
2757         }
2758
2759         TString whereClause = Form("where run=%d", GetCurrentRun());
2760
2761         TString sqlQuery = Form("update %s %s %s",
2762                                         fConfig->GetShuttlelbTable(), setClause.Data(), whereClause.Data());
2763
2764         AliDebug(2, Form("SQL query: \n%s",sqlQuery.Data()));
2765
2766         // Query execution
2767         TSQLResult* aResult;
2768         aResult = dynamic_cast<TSQLResult*> (fServer[3]->Query(sqlQuery));
2769         if (!aResult) {
2770                 Log("SHUTTLE", Form("UpdateShuttleLogbook - Can't execute query <%s>", sqlQuery.Data()));
2771                 return kFALSE;
2772         }
2773         delete aResult;
2774
2775         return kTRUE;
2776 }
2777
2778 //______________________________________________________________________________________________
2779 Int_t AliShuttle::GetCurrentRun() const
2780 {
2781         //
2782         // Get current run from logbook entry
2783         //
2784
2785         return fLogbookEntry ? fLogbookEntry->GetRun() : -1;
2786 }
2787
2788 //______________________________________________________________________________________________
2789 UInt_t AliShuttle::GetCurrentStartTime() const
2790 {
2791         //
2792         // get current start time
2793         //
2794
2795         return fLogbookEntry ? fLogbookEntry->GetStartTime() : 0;
2796 }
2797
2798 //______________________________________________________________________________________________
2799 UInt_t AliShuttle::GetCurrentEndTime() const
2800 {
2801         //
2802         // get current end time from logbook entry
2803         //
2804
2805         return fLogbookEntry ? fLogbookEntry->GetEndTime() : 0;
2806 }
2807
2808 //______________________________________________________________________________________________
2809 UInt_t AliShuttle::GetCurrentYear() const
2810 {
2811         //
2812         // Get current year from logbook entry
2813         //
2814
2815         if (!fLogbookEntry) return 0;
2816
2817         TTimeStamp startTime(GetCurrentStartTime());
2818         TString year =  Form("%d",startTime.GetDate());
2819         year = year(0,4);
2820
2821         return year.Atoi();
2822 }
2823
2824 //______________________________________________________________________________________________
2825 const char* AliShuttle::GetLHCPeriod() const
2826 {
2827         //
2828         // Get current LHC period from logbook entry
2829         //
2830
2831         if (!fLogbookEntry) return 0;
2832
2833         return fLogbookEntry->GetRunParameter("LHCperiod");
2834 }
2835
2836 //______________________________________________________________________________________________
2837 void AliShuttle::Log(const char* detector, const char* message)
2838 {
2839         //
2840         // Fill log string with a message
2841         //
2842
2843         TString logRunDir = GetShuttleLogDir();
2844         if (GetCurrentRun() >=0)
2845                 logRunDir += Form("/%d", GetCurrentRun());
2846
2847         void* dir = gSystem->OpenDirectory(logRunDir.Data());
2848         if (dir == NULL) {
2849                 if (gSystem->mkdir(logRunDir.Data(), kTRUE)) {
2850                         AliError(Form("Can't open directory <%s>", GetShuttleLogDir()));
2851                         return;
2852                 }
2853
2854         } else {
2855                 gSystem->FreeDirectory(dir);
2856         }
2857
2858         TString toLog = Form("%s UTC (%d): %s - ", TTimeStamp(time(0)).AsString("s"), getpid(), detector);
2859         if (GetCurrentRun() >= 0)
2860                 toLog += Form("run %d - ", GetCurrentRun());
2861         toLog += Form("%s", message);
2862
2863         AliInfo(toLog.Data());
2864
2865         // if we redirect the log output already to the file, leave here
2866         if (fOutputRedirected && strcmp(detector, "SHUTTLE") != 0)
2867                 return;
2868
2869         TString fileName = GetLogFileName(detector);
2870
2871         gSystem->ExpandPathName(fileName);
2872
2873         ofstream logFile;
2874         logFile.open(fileName, ofstream::out | ofstream::app);
2875
2876         if (!logFile.is_open()) {
2877                 AliError(Form("Could not open file %s", fileName.Data()));
2878                 return;
2879         }
2880
2881         logFile << toLog.Data() << "\n";
2882
2883         logFile.close();
2884 }
2885
2886 //______________________________________________________________________________________________
2887 TString AliShuttle::GetLogFileName(const char* detector) const
2888 {
2889         //
2890         // returns the name of the log file for a given sub detector
2891         //
2892
2893         TString fileName;
2894
2895         if (GetCurrentRun() >= 0)
2896         {
2897                 fileName.Form("%s/%d/%s_%d.log", GetShuttleLogDir(), GetCurrentRun(),
2898                         detector, GetCurrentRun());
2899         } else {
2900                 fileName.Form("%s/%s.log", GetShuttleLogDir(), detector);
2901         }
2902
2903         return fileName;
2904 }
2905
2906 //______________________________________________________________________________________________
2907 void AliShuttle::SendAlive()
2908 {
2909         // sends alive message to ML
2910
2911         TMonaLisaText mlStatus("SHUTTLE_status", "Alive");
2912
2913         TList mlList;
2914         mlList.Add(&mlStatus);
2915
2916         fMonaLisa->SendParameters(&mlList, "__PROCESSINGINFO__");
2917 }
2918
2919 //______________________________________________________________________________________________
2920 Bool_t AliShuttle::Collect(Int_t run)
2921 {
2922         //
2923         // Collects conditions data for all UNPROCESSED run written to DAQ LogBook in case of run = -1 (default)
2924         // If a dedicated run is given this run is processed
2925         //
2926         // In operational mode, this is the Shuttle function triggered by the EOR signal.
2927         //
2928
2929         if (run == -1)
2930                 Log("SHUTTLE","Collect - Shuttle called. Collecting conditions data for unprocessed runs");
2931         else
2932                 Log("SHUTTLE", Form("Collect - Shuttle called. Collecting conditions data for run %d", run));
2933
2934         SetLastAction("Starting");
2935
2936         // create ML instance
2937         if (!fMonaLisa)
2938                 fMonaLisa = new TMonaLisaWriter(fConfig->GetMonitorHost(), fConfig->GetMonitorTable());
2939
2940         CountOpenRuns();
2941
2942         TString whereClause("where shuttle_done=0");
2943         if (run != -1)
2944                 whereClause += Form(" and run=%d", run);
2945
2946         TObjArray shuttleLogbookEntries;
2947         if (!QueryShuttleLogbook(whereClause, shuttleLogbookEntries))
2948         {
2949                 Log("SHUTTLE", "Collect - Can't retrieve entries from Shuttle logbook");
2950                 return kFALSE;
2951         }
2952
2953         if (shuttleLogbookEntries.GetEntries() == 0)
2954         {
2955                 if (run == -1)
2956                         Log("SHUTTLE","Collect - Found no UNPROCESSED runs in Shuttle logbook");
2957                 else
2958                         Log("SHUTTLE", Form("Collect - Run %d is already DONE "
2959                                                 "or it does not exist in Shuttle logbook", run));
2960                 return kTRUE;
2961         }
2962
2963         for (UInt_t iDet=0; iDet<NDetectors(); iDet++)
2964                 fFirstUnprocessed[iDet] = kTRUE;
2965
2966         if (run != -1)
2967         {
2968                 // query Shuttle logbook for earlier runs, check if some detectors are unprocessed,
2969                 // flag them into fFirstUnprocessed array
2970                 TString whereClauseBis(Form("where shuttle_done=0 and run < %d", run));
2971                 TObjArray tmpLogbookEntries;
2972                 if (!QueryShuttleLogbook(whereClauseBis, tmpLogbookEntries))
2973                 {
2974                         Log("SHUTTLE", "Collect - Can't retrieve entries from Shuttle logbook");
2975                         return kFALSE;
2976                 }
2977
2978                 TIter iter(&tmpLogbookEntries);
2979                 AliShuttleLogbookEntry* anEntry = 0;
2980                 while ((anEntry = dynamic_cast<AliShuttleLogbookEntry*> (iter.Next())))
2981                 {
2982                         for (UInt_t iDet=0; iDet<NDetectors(); iDet++)
2983                         {
2984                                 if (anEntry->GetDetectorStatus(iDet) == AliShuttleLogbookEntry::kUnprocessed)
2985                                 {
2986                                         AliDebug(2, Form("Run %d: setting %s as \"not first time unprocessed\"",
2987                                                         anEntry->GetRun(), GetDetName(iDet)));
2988                                         fFirstUnprocessed[iDet] = kFALSE;
2989                                 }
2990                         }
2991
2992                 }
2993
2994         }
2995
2996         if (!RetrieveConditionsData(shuttleLogbookEntries))
2997         {
2998                 Log("SHUTTLE", "Collect - Process of at least one run failed");
2999                 CountOpenRuns();
3000                 return kFALSE;
3001         }
3002
3003         Log("SHUTTLE", "Collect - Requested run(s) successfully processed");
3004         CountOpenRuns();
3005         return kTRUE;
3006 }
3007
3008 //______________________________________________________________________________________________
3009 Bool_t AliShuttle::RetrieveConditionsData(const TObjArray& dateEntries)
3010 {
3011         //
3012         // Retrieve conditions data for all runs that aren't processed yet
3013         //
3014
3015         Bool_t hasError = kFALSE;
3016
3017         TIter iter(&dateEntries);
3018         AliShuttleLogbookEntry* anEntry;
3019
3020         while ((anEntry = (AliShuttleLogbookEntry*) iter.Next())){
3021                 if (!Process(anEntry)){
3022                         hasError = kTRUE;
3023                 }
3024
3025                 // clean SHUTTLE temp directory
3026                 //TString filename = Form("%s/*.shuttle", GetShuttleTempDir());
3027                 //RemoveFile(filename.Data());
3028         }
3029
3030         return hasError == kFALSE;
3031 }
3032
3033 //______________________________________________________________________________________________
3034 ULong_t AliShuttle::GetTimeOfLastAction() const
3035 {
3036         //
3037         // Gets time of last action
3038         //
3039
3040         ULong_t tmp;
3041
3042         fMonitoringMutex->Lock();
3043
3044         tmp = fLastActionTime;
3045
3046         fMonitoringMutex->UnLock();
3047
3048         return tmp;
3049 }
3050
3051 //______________________________________________________________________________________________
3052 const TString AliShuttle::GetLastAction() const
3053 {
3054         //
3055         // returns a string description of the last action
3056         //
3057
3058         TString tmp;
3059
3060         fMonitoringMutex->Lock();
3061
3062         tmp = fLastAction;
3063
3064         fMonitoringMutex->UnLock();
3065
3066         return tmp;
3067 }
3068
3069 //______________________________________________________________________________________________
3070 void AliShuttle::SetLastAction(const char* action)
3071 {
3072         //
3073         // updates the monitoring variables
3074         //
3075
3076         fMonitoringMutex->Lock();
3077
3078         fLastAction = action;
3079         fLastActionTime = time(0);
3080
3081         fMonitoringMutex->UnLock();
3082 }
3083
3084 //______________________________________________________________________________________________
3085 const char* AliShuttle::GetRunParameter(const char* param)
3086 {
3087         //
3088         // returns run parameter read from DAQ logbook
3089         //
3090
3091         if(!fLogbookEntry) {
3092                 AliError("No logbook entry!");
3093                 return 0;
3094         }
3095
3096         return fLogbookEntry->GetRunParameter(param);
3097 }
3098
3099 //______________________________________________________________________________________________
3100 AliCDBEntry* AliShuttle::GetFromOCDB(const char* detector, const AliCDBPath& path)
3101 {
3102         //
3103         // returns object from OCDB valid for current run
3104         //
3105
3106         if (fTestMode & kErrorOCDB)
3107         {
3108                 Log(detector, "GetFromOCDB - In TESTMODE - Simulating error with OCDB");
3109                 return 0;
3110         }
3111
3112         AliCDBStorage *sto = AliCDBManager::Instance()->GetStorage(fgkMainCDB);
3113         if (!sto)
3114         {
3115                 Log(detector, "GetFromOCDB - Cannot activate main OCDB for query!");
3116                 return 0;
3117         }
3118
3119         return dynamic_cast<AliCDBEntry*> (sto->Get(path, GetCurrentRun()));
3120 }
3121
3122 //______________________________________________________________________________________________
3123 Bool_t AliShuttle::SendMail(EMailTarget target, Int_t system)
3124 {
3125         //
3126         // sends a mail to the subdetector expert in case of preprocessor error
3127         //
3128
3129         if (fTestMode != kNone)
3130                 return kTRUE;
3131
3132         if (!fConfig->SendMail())
3133                 return kTRUE;
3134
3135         if (target == kDCSEMail || target == kFXSEMail) {
3136                 if (!fFirstProcessing)
3137                 return kTRUE;
3138         }
3139
3140         Int_t runMode = (Int_t)fConfig->GetRunMode();
3141         TString tmpStr;
3142         if (runMode == 0) tmpStr = " Nightly Test:";
3143         else tmpStr = " Data Taking:";
3144         void* dir = gSystem->OpenDirectory(GetShuttleLogDir());
3145         if (dir == NULL)
3146         {
3147                 if (gSystem->mkdir(GetShuttleLogDir(), kTRUE))
3148                 {
3149                         Log("SHUTTLE", Form("SendMail - Can't open directory <%s>", GetShuttleLogDir()));
3150                         return kFALSE;
3151                 }
3152
3153         } else {
3154                 gSystem->FreeDirectory(dir);
3155         }
3156
3157         // det experts in to
3158         TString to="";
3159         TIter *iterExperts = 0;
3160         if (target == kDCSEMail) {
3161                 iterExperts = new TIter(fConfig->GetAdmins(AliShuttleConfig::kAmanda));
3162         }
3163         else if (target == kFXSEMail) {
3164                 iterExperts = new TIter(fConfig->GetAdmins(system));
3165         }
3166         if (iterExperts) {
3167                 TObjString *anExpert=0;
3168                 while ((anExpert = (TObjString*) iterExperts->Next()))
3169                 {
3170                         to += Form("%s,", anExpert->GetName());
3171                 }
3172                 delete iterExperts;
3173         }
3174
3175         // add subdetector experts
3176         iterExperts = new TIter(fConfig->GetResponsibles(fCurrentDetector));
3177         TObjString *anExpert=0;
3178         while ((anExpert = (TObjString*) iterExperts->Next()))
3179         {
3180                 to += Form("%s,", anExpert->GetName());
3181         }
3182         delete iterExperts;
3183
3184         if (to.Length() > 0)
3185           to.Remove(to.Length()-1);
3186         AliDebug(2, Form("to: %s",to.Data()));
3187
3188         if (to.IsNull()) {
3189                 Log("SHUTTLE", Form("List of %d responsibles not set!", (Int_t) target));
3190                 return kFALSE;
3191         }
3192
3193         // SHUTTLE responsibles in cc
3194         TString cc="";
3195         TIter iterAdmins(fConfig->GetAdmins(AliShuttleConfig::kGlobal));
3196         TObjString *anAdmin=0;
3197         while ((anAdmin = (TObjString*) iterAdmins.Next()))
3198         {
3199                 cc += Form("%s,", anAdmin->GetName());
3200         }
3201         if (cc.Length() > 0)
3202           cc.Remove(cc.Length()-1);
3203         AliDebug(2, Form("cc: %s",to.Data()));
3204
3205         // mail body
3206         TString bodyFileName;
3207         bodyFileName.Form("%s/mail.body", GetShuttleLogDir());
3208         gSystem->ExpandPathName(bodyFileName);
3209
3210         ofstream mailBody;
3211         mailBody.open(bodyFileName, ofstream::out);
3212
3213         if (!mailBody.is_open())
3214         {
3215                 Log("SHUTTLE", Form("Could not open mail body file %s", bodyFileName.Data()));
3216                 return kFALSE;
3217         }
3218
3219
3220         TString subject;
3221         TString body;
3222
3223         if (target == kDCSEMail){
3224                 subject = Form("%s CRITICAL Retrieval of data points for %s FAILED in run %d !",
3225                                 tmpStr.Data(), fCurrentDetector.Data(), GetCurrentRun());
3226                 AliDebug(2, Form("subject: %s", subject.Data()));
3227
3228                 body = Form("Dear DCS experts, \n\n");
3229                 body += Form("SHUTTLE couldn\'t retrieve the data points for detector %s "
3230                              "in run %d!!\n\n", fCurrentDetector.Data(), GetCurrentRun());
3231         }
3232         else if (target == kFXSEMail){
3233                 subject = Form("%s CRITICAL FXS communication for %s FAILED in run %d !",
3234                                 tmpStr.Data(), fCurrentDetector.Data(), GetCurrentRun());
3235                 AliDebug(2, Form("subject: %s", subject.Data()));
3236                 TString sys;
3237                 if (system == kDAQ) sys="DAQ";
3238                 else if (system == kDCS) sys="DCS";
3239                 else if (system == kHLT) sys="HLT";
3240                 else return kFALSE;
3241                 body = Form("Dear  %s FXS experts, \n\n",sys.Data());
3242                 body += Form("SHUTTLE couldn\'t retrieve data from the FXS for detector %s "
3243                              "in run %d!!\n\n", fCurrentDetector.Data(), GetCurrentRun());
3244         }
3245         else {
3246                 subject = Form("%s %s Shuttle preprocessor FAILED in run %d (run type = %s)!",
3247                                        tmpStr.Data(), fCurrentDetector.Data(), GetCurrentRun(), GetRunType());
3248                 AliDebug(2, Form("subject: %s", subject.Data()));
3249
3250                 body = Form("Dear %s expert(s), \n\n", fCurrentDetector.Data());
3251                 body += Form("SHUTTLE just detected that your preprocessor "
3252                              "failed processing run %d (run type = %s)!!\n\n",
3253                              GetCurrentRun(), GetRunType());
3254         }
3255
3256         body += Form("Please check %s status on the SHUTTLE monitoring page: \n\n",
3257                                 fCurrentDetector.Data());
3258         if (fConfig->GetRunMode() == AliShuttleConfig::kTest)
3259         {
3260                 body += Form("\thttp://pcalimonitor.cern.ch/shuttle.jsp?time=24 \n\n");
3261         } else {
3262                 body += Form("\thttp://pcalimonitor.cern.ch/shuttle.jsp?instance=PROD&time=24 \n\n");
3263         }
3264
3265
3266         TString logFolder = "logs";
3267         if (fConfig->GetRunMode() == AliShuttleConfig::kProd)
3268                 logFolder += "_PROD";
3269
3270
3271         body += Form("Find the %s log for the current run on \n\n"
3272                 "\thttp://pcalishuttle01.cern.ch:8880/%s/%d/%s_%d.log \n\n",
3273                 fCurrentDetector.Data(), logFolder.Data(), GetCurrentRun(),
3274                                 fCurrentDetector.Data(), GetCurrentRun());
3275         body += Form("The last 10 lines of %s log file are following:\n\n", fCurrentDetector.Data());
3276
3277         AliDebug(2, Form("Body begin: %s", body.Data()));
3278
3279         mailBody << body.Data();
3280         mailBody.close();
3281         mailBody.open(bodyFileName, ofstream::out | ofstream::app);
3282
3283         TString logFileName = Form("%s/%d/%s_%d.log", GetShuttleLogDir(),
3284                 GetCurrentRun(), fCurrentDetector.Data(), GetCurrentRun());
3285         TString tailCommand = Form("tail -n 10 %s >> %s", logFileName.Data(), bodyFileName.Data());
3286         if (gSystem->Exec(tailCommand.Data()))
3287         {
3288                 mailBody << Form("%s log file not found ...\n\n", fCurrentDetector.Data());
3289         }
3290
3291         TString endBody = Form("------------------------------------------------------\n\n");
3292         endBody += Form("In case of problems please contact the SHUTTLE core team.\n\n");
3293         endBody += "Please do not answer this message directly, it is automatically generated.\n\n";
3294         endBody += "Greetings,\n\n \t\t\tthe SHUTTLE\n";
3295
3296         AliDebug(2, Form("Body end: %s", endBody.Data()));
3297
3298         mailBody << endBody.Data();
3299
3300         mailBody.close();
3301
3302         // send mail!
3303         TString mailCommand = Form("mail -s \"%s\" -c %s %s < %s",
3304                                                 subject.Data(),
3305                                                 cc.Data(),
3306                                                 to.Data(),
3307                                                 bodyFileName.Data());
3308         AliDebug(2, Form("mail command: %s", mailCommand.Data()));
3309
3310         Bool_t result = gSystem->Exec(mailCommand.Data());
3311
3312         return result == 0;
3313 }
3314
3315 //______________________________________________________________________________________________
3316 const char* AliShuttle::GetRunType()
3317 {
3318         //
3319         // returns run type read from "run type" logbook
3320         //
3321
3322         if(!fLogbookEntry) {
3323                 AliError("No logbook entry!");
3324                 return 0;
3325         }
3326
3327         return fLogbookEntry->GetRunType();
3328 }
3329
3330 //______________________________________________________________________________________________
3331 Bool_t AliShuttle::GetHLTStatus()
3332 {
3333         // Return HLT status (ON=1 OFF=0)
3334         // Converts the HLT status from the mode string read in the run logbook (not just a bool)
3335
3336         if(!fLogbookEntry) {
3337                 AliError("No logbook entry!");
3338                 return 0;
3339         }
3340
3341         // TODO implement when HLTMode is inserted in run logbook
3342         TString hltMode = fLogbookEntry->GetRunParameter("HLTmode");
3343         TSubString firstChar = hltMode(0,1);
3344         AliDebug(2,Form("First char = %s ",firstChar.Data()));
3345         if (firstChar == "A") {
3346                 return kFALSE;
3347         }
3348         else if ((firstChar == "B") || (firstChar == "C") || (firstChar == "D") || (firstChar == "E")) {
3349                 return kTRUE;
3350         }
3351         else {
3352                 Log("SHUTTLE","Unexpected HLT mode! Returning 0....");
3353                 return kFALSE;
3354         }
3355 }
3356
3357 //______________________________________________________________________________________________
3358 const char* AliShuttle::GetTriggerConfiguration()
3359 {
3360         // Receives the trigger configuration from the DAQ logbook for the current run
3361
3362         // check connection, if needed reconnect
3363         if (!Connect(3))
3364                 return 0;
3365
3366         TString sqlQuery;
3367         sqlQuery.Form("SELECT configFile FROM logbook_trigger_config WHERE run = %d", GetCurrentRun());
3368         TSQLResult* result = fServer[3]->Query(sqlQuery);
3369         if (!result)
3370         {
3371                 Log("SHUTTLE", Form("ERROR: Can't execute query <%s>!", sqlQuery.Data()));
3372                 return 0;
3373         }
3374
3375         if (result->GetRowCount() == 0)
3376         {
3377                 Log("SHUTTLE", "ERROR: Trigger configuration not found in logbook_trigger_config");
3378                 delete result;
3379                 return 0;
3380         }
3381
3382         TSQLRow* row = result->Next();
3383         if (!row)
3384         {
3385                 Log("SHUTTLE", "ERROR: Could not receive logbook_trigger_config data");
3386                 delete result;
3387                 return 0;
3388         }
3389
3390         // static, so that pointer remains valid when it is returned to the calling class
3391         static TString triggerConfig(row->GetField(0));
3392
3393         delete row;
3394         row = 0;
3395
3396         delete result;
3397         result = 0;
3398
3399         Log("SHUTTLE", Form("Found trigger configuration: %s", triggerConfig.Data()));
3400
3401         return triggerConfig;
3402 }
3403
3404 //______________________________________________________________________________________________
3405 void AliShuttle::SetShuttleTempDir(const char* tmpDir)
3406 {
3407         //
3408         // sets Shuttle temp directory
3409         //
3410
3411         fgkShuttleTempDir = gSystem->ExpandPathName(tmpDir);
3412 }
3413
3414 //______________________________________________________________________________________________
3415 void AliShuttle::SetShuttleLogDir(const char* logDir)
3416 {
3417         //
3418         // sets Shuttle log directory
3419         //
3420
3421         fgkShuttleLogDir = gSystem->ExpandPathName(logDir);
3422 }
3423 //______________________________________________________________________________________________
3424 Bool_t AliShuttle::TouchFile()
3425 {
3426         //
3427         // touching a file on the grid if run has been DONE
3428         //
3429
3430         if (!gGrid)
3431         {
3432                 Log("SHUTTLE",Form("No TGrid connection estabilished!"));
3433                 Log("SHUTTLE",Form("Could not touch file for run %i",GetCurrentRun()));
3434                 return kFALSE;
3435         }
3436
3437         TString dir;
3438         dir.Form("%s%d/%s/SHUTTLE_DONE", fConfig->GetAlienPath(), GetCurrentYear(), GetLHCPeriod());
3439         // checking whether directory for touch command exists
3440         TString commandLs;
3441         commandLs.Form("ls %s",dir.Data());
3442         TGridResult *resultLs = dynamic_cast<TGridResult*>(gGrid->Command(commandLs));
3443         if (!resultLs){
3444                 Log("SHUTTLE",Form("No result for %s command, returning without touching",commandLs.Data()));
3445                 return kFALSE;
3446         }
3447         TMap *mapLs = dynamic_cast<TMap*>(resultLs->At(0));
3448         if (!mapLs){
3449                 Log("SHUTTLE",Form("No map for %s command, returning without touching",commandLs.Data()));
3450                 delete resultLs;
3451                 resultLs = 0x0;
3452                 return kFALSE;
3453         }
3454         TObjString *valueLsPath = dynamic_cast<TObjString*>(mapLs->GetValue("path"));
3455         if (!valueLsPath || (TString)(valueLsPath->GetString()).CompareTo(dir)!=1){
3456                 Log("SHUTTLE",Form("No directory %s found, creating it",dir.Data()));
3457
3458                 // creating the directory
3459
3460                 Bool_t boolMkdir = gGrid->Mkdir(dir.Data());
3461                 if (!boolMkdir) {
3462                         Log("SHUTTLE",Form("Impossible to create dir %s in alien catalogue for run %i!",dir.Data(),GetCurrentRun()));
3463                         delete resultLs;
3464                         resultLs = 0x0;
3465                         return kFALSE;
3466                 }
3467                 Log("SHUTTLE",Form("Directory %s successfully created in alien catalogue for run %i",dir.Data(),GetCurrentRun()));
3468         }
3469         else {
3470                 Log("SHUTTLE",Form("Directory %s correctly found for run %i",dir.Data(),GetCurrentRun()));
3471         }
3472
3473         delete resultLs;
3474         resultLs = 0x0;
3475
3476         TString command;
3477         command.Form("touch %s/%i", dir.Data(), GetCurrentRun());
3478         Log("SHUTTLE", Form("Creating entry in file catalog: %s", command.Data()));
3479         TGridResult *resultTouch = dynamic_cast<TGridResult*>(gGrid->Command(command));
3480         if (!resultTouch){
3481                 Log("SHUTTLE",Form("No result for touching command, returning without touching for run %i",GetCurrentRun()));
3482                 return kFALSE;
3483         }
3484         TMap *mapTouch = dynamic_cast<TMap*>(resultTouch->At(0));
3485         if (!mapTouch){
3486                 Log("SHUTTLE",Form("No map for touching command, returning without touching for run %i",GetCurrentRun()));
3487                 delete resultTouch;
3488                 resultTouch = 0x0;
3489                 return kFALSE;
3490         }
3491         TObjString *valueTouch = dynamic_cast<TObjString*>(mapTouch->GetValue("__result__"));
3492         if (!valueTouch){
3493                 Log("SHUTTLE",Form("No value for \"__result__\" key set in the map for touching command, returning without touching for run %i",GetCurrentRun()));
3494                 delete resultTouch;
3495                 resultTouch = 0x0;
3496                 return kFALSE;
3497         }
3498         if (valueTouch->GetString()!="1"){
3499                 Log("SHUTTLE",Form("Failing the touching command, returning without touching for run %i",GetCurrentRun()));
3500                 delete resultTouch;
3501                 resultTouch = 0x0;
3502                 return kFALSE;
3503         }
3504         delete resultTouch;
3505         resultTouch = 0x0;
3506         Log("SHUTTLE", "Sucessfully touched the file");
3507         return kTRUE;
3508 }
3509
3510