a804d77acaec0abd32bdc330a75730790ed72698
[u/mrichter/AliRoot.git] / SHUTTLE / AliShuttle.cxx
1 /**************************************************************************
2  * Copyright(c) 1998-1999, ALICE Experiment at CERN, All rights reserved. *
3  *                                                                        *
4  * Author: The ALICE Off-line Project.                                    *
5  * Contributors are mentioned in the code where appropriate.              *
6  *                                                                        *
7  * Permission to use, copy, modify and distribute this software and its   *
8  * documentation strictly for non-commercial purposes is hereby granted   *
9  * without fee, provided that the above copyright notice appears in all   *
10  * copies and that both the copyright notice and this permission notice   *
11  * appear in the supporting documentation. The authors make no claims     *
12  * about the suitability of this software for any purpose. It is          *
13  * provided "as is" without express or implied warranty.                  *
14  **************************************************************************/
15
16 /* $Id$ */
17
18 //
19 // This class is the main manager for AliShuttle. 
20 // It organizes the data retrieval from DCS and call the 
21 // interface methods of AliPreprocessor.
22 // For every detector in AliShuttleConfgi (see AliShuttleConfig),
23 // data for its set of aliases is retrieved. If there is registered
24 // AliPreprocessor for this detector then it will be used
25 // accroding to the schema (see AliPreprocessor).
26 // If there isn't registered AliPreprocessor than the retrieved
27 // data is stored automatically to the undelying AliCDBStorage.
28 // For detSpec is used the alias name.
29 //
30
31 #include "AliShuttle.h"
32
33 #include "AliCDBManager.h"
34 #include "AliCDBStorage.h"
35 #include "AliCDBId.h"
36 #include "AliCDBRunRange.h"
37 #include "AliCDBPath.h"
38 #include "AliCDBEntry.h"
39 #include "AliShuttleConfig.h"
40 #include "DCSClient/AliDCSClient.h"
41 #include "AliLog.h"
42 #include "AliPreprocessor.h"
43 #include "AliShuttleStatus.h"
44 #include "AliShuttleLogbookEntry.h"
45
46 #include <TSystem.h>
47 #include <TObject.h>
48 #include <TString.h>
49 #include <TTimeStamp.h>
50 #include <TObjString.h>
51 #include <TSQLServer.h>
52 #include <TSQLResult.h>
53 #include <TSQLRow.h>
54 #include <TMutex.h>
55 #include <TSystemDirectory.h>
56 #include <TSystemFile.h>
57 #include <TFile.h>
58 #include <TGrid.h>
59 #include <TGridResult.h>
60 #include <TMap.h>
61
62 #include <TMonaLisaWriter.h>
63
64 #include <fstream>
65
66 #include <sys/types.h>
67 #include <sys/wait.h>
68
69 #include <signal.h>
70
71 ClassImp(AliShuttle)
72
73 //______________________________________________________________________________________________
74 AliShuttle::AliShuttle(const AliShuttleConfig* config,
75                 UInt_t timeout, Int_t retries):
76 fConfig(config),
77 fTimeout(timeout), fRetries(retries),
78 fPreprocessorMap(),
79 fLogbookEntry(0),
80 fCurrentDetector(),
81 fFirstProcessing(0),
82 fFXSError(-1),
83 fStatusEntry(0),
84 fMonitoringMutex(0),
85 fLastActionTime(0),
86 fLastAction(),
87 fMonaLisa(0),
88 fTestMode(kNone),
89 fReadTestMode(kFALSE),
90 fOutputRedirected(kFALSE)
91 {
92         //
93         // config: AliShuttleConfig used
94         // timeout: timeout used for AliDCSClient connection
95         // retries: the number of retries in case of connection error.
96         //
97
98         if (!fConfig->IsValid()) AliFatal("********** !!!!! Invalid configuration !!!!! **********");
99         for(int iSys=0;iSys<4;iSys++) {
100                 fServer[iSys]=0;
101                 if (iSys < 3)
102                         fFXSlist[iSys].SetOwner(kTRUE);
103         }
104         fPreprocessorMap.SetOwner(kTRUE);
105
106         for (UInt_t iDet=0; iDet<NDetectors(); iDet++)
107                 fFirstUnprocessed[iDet] = kFALSE;
108
109         fMonitoringMutex = new TMutex();
110 }
111
112 //______________________________________________________________________________________________
113 AliShuttle::~AliShuttle()
114 {
115         //
116         // destructor
117         //
118
119         fPreprocessorMap.DeleteAll();
120         for(int iSys=0;iSys<4;iSys++)
121                 if(fServer[iSys]) {
122                         fServer[iSys]->Close();
123                         delete fServer[iSys];
124                         fServer[iSys] = 0;
125                 }
126
127         if (fStatusEntry){
128                 delete fStatusEntry;
129                 fStatusEntry = 0;
130         }
131         
132         if (fMonitoringMutex) 
133         {
134                 delete fMonitoringMutex;
135                 fMonitoringMutex = 0;
136         }
137 }
138
139 //______________________________________________________________________________________________
140 void AliShuttle::RegisterPreprocessor(AliPreprocessor* preprocessor)
141 {
142         //
143         // Registers new AliPreprocessor.
144         // It uses GetName() for indentificator of the pre processor.
145         // The pre processor is registered it there isn't any other
146         // with the same identificator (GetName()).
147         //
148
149         const char* detName = preprocessor->GetName();
150         if(GetDetPos(detName) < 0)
151                 AliFatal(Form("********** !!!!! Invalid detector name: %s !!!!! **********", detName));
152
153         if (fPreprocessorMap.GetValue(detName)) {
154                 AliWarning(Form("AliPreprocessor %s is already registered!", detName));
155                 return;
156         }
157
158         fPreprocessorMap.Add(new TObjString(detName), preprocessor);
159 }
160 //______________________________________________________________________________________________
161 Bool_t AliShuttle::Store(const AliCDBPath& path, TObject* object,
162                 AliCDBMetaData* metaData, Int_t validityStart, Bool_t validityInfinite)
163 {
164         // Stores a CDB object in the storage for offline reconstruction. Objects that are not needed for
165         // offline reconstruction, but should be stored anyway (e.g. for debugging) should NOT be stored
166         // using this function. Use StoreReferenceData instead!
167         // It calls StoreLocally function which temporarily stores the data locally; when the preprocessor
168         // finishes the data are transferred to the main storage (Grid).
169
170         return StoreLocally(fgkLocalCDB, path, object, metaData, validityStart, validityInfinite);
171 }
172
173 //______________________________________________________________________________________________
174 Bool_t AliShuttle::StoreReferenceData(const AliCDBPath& path, TObject* object, AliCDBMetaData* metaData)
175 {
176         // Stores a CDB object in the storage for reference data. This objects will not be available during
177         // offline reconstrunction. Use this function for reference data only!
178         // It calls StoreLocally function which temporarily stores the data locally; when the preprocessor
179         // finishes the data are transferred to the main storage (Grid).
180
181         return StoreLocally(fgkLocalRefStorage, path, object, metaData);
182 }
183
184 //______________________________________________________________________________________________
185 Bool_t AliShuttle::StoreLocally(const TString& localUri,
186                         const AliCDBPath& path, TObject* object, AliCDBMetaData* metaData,
187                         Int_t validityStart, Bool_t validityInfinite)
188 {
189         // Store object temporarily in local storage. Parameters are passed by Store and StoreReferenceData functions.
190         // when the preprocessor finishes the data are transferred to the main storage (Grid).
191         // The parameters are:
192         //   1) Uri of the backup storage (Local)
193         //   2) the object's path.
194         //   3) the object to be stored
195         //   4) the metaData to be associated with the object
196         //   5) the validity start run number w.r.t. the current run,
197         //      if the data is valid only for this run leave the default 0
198         //   6) specifies if the calibration data is valid for infinity (this means until updated),
199         //      typical for calibration runs, the default is kFALSE
200         //
201         // returns 0 if fail, 1 otherwise
202
203
204         if (fTestMode & kErrorStorage)
205         {
206                 Log(fCurrentDetector, "StoreLocally - In TESTMODE - Simulating error while storing locally");
207                 return kFALSE;
208         }
209         
210         const char* cdbType = (localUri == fgkLocalCDB) ? "CDB" : "Reference";
211
212         Int_t firstRun = GetCurrentRun() - validityStart;
213         if(firstRun < 0) {
214                 AliWarning("First valid run happens to be less than 0! Setting it to 0.");
215                 firstRun=0;
216         }
217
218         Int_t lastRun = -1;
219         if(validityInfinite) {
220                 lastRun = AliCDBRunRange::Infinity();
221         } else {
222                 lastRun = GetCurrentRun();
223         }
224
225         // Version is set to current run, it will be used later to transfer data to Grid
226         AliCDBId id(path, firstRun, lastRun, GetCurrentRun(), -1);
227
228         if(! dynamic_cast<TObjString*> (metaData->GetProperty("RunUsed(TObjString)"))){
229                 TObjString runUsed = Form("%d", GetCurrentRun());
230                 metaData->SetProperty("RunUsed(TObjString)", runUsed.Clone());
231         }
232
233         Bool_t result = kFALSE;
234
235         if (!(AliCDBManager::Instance()->GetStorage(localUri))) {
236                 Log("SHUTTLE", Form("StoreLocally - Cannot activate local %s storage", cdbType));
237         } else {
238                 Int_t logLevel = AliLog::GetGlobalLogLevel();
239                 AliLog::SetGlobalLogLevel(AliLog::kError);
240                 result = AliCDBManager::Instance()->GetStorage(localUri)
241                                         ->Put(object, id, metaData);
242                 AliLog::SetGlobalLogLevel((AliLog::EType_t)logLevel);
243         }
244
245         if(!result) {
246
247                 Log(fCurrentDetector, Form("StoreLocally - Can't store object <%s>!", id.ToString().Data()));
248         }
249
250
251         return result;
252 }
253
254 //______________________________________________________________________________________________
255 Bool_t AliShuttle::StoreOCDB()
256 {
257         //
258         // Called when preprocessor ends successfully or when previous storage attempt failed (kStoreError status)
259         // Calls underlying StoreOCDB(const char*) function twice, for OCDB and Reference storage.
260         // Then calls StoreRefFilesToGrid to store reference files. 
261         //
262         
263         UpdateShuttleStatus(AliShuttleStatus::kStoreStarted);
264                         
265         if (fTestMode & kErrorGrid)
266         {
267                 Log("SHUTTLE", "StoreOCDB - In TESTMODE - Simulating error while storing in the Grid");
268                 Log(fCurrentDetector, "StoreOCDB - In TESTMODE - Simulating error while storing in the Grid");
269                 return kFALSE;
270         }
271         
272         Log("SHUTTLE","StoreOCDB - Storing OCDB data ...");
273         Int_t resultCDB = StoreOCDB(fgkMainCDB);
274
275         Log("SHUTTLE","StoreOCDB - Storing reference data ...");
276         Int_t resultRef = StoreOCDB(fgkMainRefStorage);
277         
278         Log("SHUTTLE","StoreOCDB - Storing reference files ...");
279         Bool_t resultRefFiles = CopyFilesToGrid("reference");
280         
281         Bool_t resultMetadata = kTRUE;
282         if(fCurrentDetector == "GRP") 
283         {
284                 Log("SHUTTLE","StoreOCDB - Storing Run Metadata file ...");
285                 resultMetadata = CopyFilesToGrid("metadata");
286         }
287         
288         Int_t storeResult = 0;
289         
290         if (resultCDB < 0 || resultRef < 0 || resultRefFiles == kFALSE || resultMetadata == kFALSE)
291                 storeResult = -1;
292         else if (resultCDB > 0 || resultRef > 0)
293                 storeResult = 1;
294                 
295         if (storeResult < 0)
296         {
297                 Log("SHUTTLE", 
298                         Form("\t\t\t****** run %d - %s: STORAGE ERROR ******",
299                                 GetCurrentRun(), fCurrentDetector.Data()));
300                 UpdateShuttleStatus(AliShuttleStatus::kStoreError);
301         } 
302         else if (storeResult > 0)
303         {
304                 Log("SHUTTLE", 
305                         Form("\t\t\t****** run %d - %s: STORAGE DELAYED ******",
306                                 GetCurrentRun(), fCurrentDetector.Data()));
307                 UpdateShuttleStatus(AliShuttleStatus::kStoreDelayed);
308         }
309         else if (storeResult == 0) 
310         {
311                 Log("SHUTTLE", 
312                         Form("\t\t\t****** run %d - %s: DONE ******",
313                                 GetCurrentRun(), fCurrentDetector.Data()));
314                 UpdateShuttleStatus(AliShuttleStatus::kDone);
315                 UpdateShuttleLogbook(fCurrentDetector, "DONE");
316         }
317
318         return (storeResult == 0);
319 }
320
321 //______________________________________________________________________________________________
322 Int_t AliShuttle::StoreOCDB(const TString& gridURI)
323 {
324         //
325         // Called by StoreOCDB(), performs actual storage to the main OCDB and reference storages (Grid)
326         //
327         // Return code:
328         //   -2 initialization error
329         //   -1 storage error
330         //   0  success
331         //   1  storage delayed (e.g. previous unprocessed runs)
332         //
333
334         TObjArray* gridIds=0;
335
336         Bool_t result = kTRUE;
337         Bool_t delayed = kFALSE;
338
339         const char* type = 0;
340         TString localURI;
341         if(gridURI == fgkMainCDB) {
342                 type = "OCDB";
343                 localURI = fgkLocalCDB;
344         } else if(gridURI == fgkMainRefStorage) {
345                 type = "reference";
346                 localURI = fgkLocalRefStorage;
347         } else {
348                 AliError(Form("Invalid storage URI: %s", gridURI.Data()));
349                 return -2;
350         }
351
352         AliCDBManager* man = AliCDBManager::Instance();
353
354         AliCDBStorage *gridSto = man->GetStorage(gridURI);
355         if(!gridSto) {
356                 Log("SHUTTLE",
357                         Form("StoreOCDB - cannot activate main %s storage", type));
358                 return -2;
359         }
360
361         gridIds = gridSto->GetQueryCDBList();
362
363         // get objects previously stored in local CDB
364         AliCDBStorage *localSto = man->GetStorage(localURI);
365         if(!localSto) {
366                 Log("SHUTTLE",
367                         Form("StoreOCDB - cannot activate local %s storage", type));
368                 return -2;
369         }
370         AliCDBPath aPath(GetOfflineDetName(fCurrentDetector.Data()),"*","*");
371         // Local objects were stored with current run as Grid version!
372         TList* localEntries = localSto->GetAll(aPath.GetPath(), GetCurrentRun(), GetCurrentRun());
373         localEntries->SetOwner(1);
374
375         // loop on local stored objects
376         TIter localIter(localEntries);
377         AliCDBEntry *aLocEntry = 0;
378         while((aLocEntry = dynamic_cast<AliCDBEntry*> (localIter.Next()))){
379                 aLocEntry->SetOwner(1);
380                 AliCDBId aLocId = aLocEntry->GetId();
381                 aLocEntry->SetVersion(-1);
382                 aLocEntry->SetSubVersion(-1);
383                                                 
384                 Log(fCurrentDetector.Data(), Form("Attempting to store %s", aLocId.ToString().Data()));
385
386                 // If local object is valid up to infinity we store it only if it is
387                 // the first unprocessed run!
388                 if (aLocId.GetLastRun() == AliCDBRunRange::Infinity() &&
389                         !fFirstUnprocessed[GetDetPos(fCurrentDetector)])
390                 {
391                         Log("SHUTTLE", Form("StoreOCDB - %s: object %s has validity infinite but "
392                                                 "there are previous unprocessed runs!",
393                                                 fCurrentDetector.Data(), aLocId.GetPath().Data()));
394                         Log(fCurrentDetector.Data(), Form("StoreOCDB - %s: object %s has validity infinite but "
395                                                 "there are previous unprocessed runs!",
396                                                 fCurrentDetector.Data(), aLocId.GetPath().Data()));
397                         delayed = kTRUE;
398                         continue;
399                 }
400
401                 // loop on Grid valid Id's
402                 Bool_t store = kTRUE;
403                 TIter gridIter(gridIds);
404                 AliCDBId* aGridId = 0;
405                 while ((aGridId = dynamic_cast<AliCDBId*> (gridIter.Next()))) {
406                         if (aGridId->GetPath() != aLocId.GetPath()) 
407                                 continue;
408                         // skip all objects valid up to infinity
409                         if (aGridId->GetLastRun() == AliCDBRunRange::Infinity()) 
410                                 continue;
411                         
412                         // if we get here, it means there's already some more recent object stored on Grid!
413                         Log(fCurrentDetector.Data(),
414                                 Form("StoreOCDB - A more recent object already exists in %s storage: <%s>",
415                                 type, aGridId->ToString().Data()));
416                         
417                         store = kFALSE;
418                         break;
419                 }
420
421                 Bool_t storeOk = kFALSE;
422                 if (store)
423                 {
424                         Log(fCurrentDetector.Data(), Form("Prechecks succeeded. Ready to store %s", aLocId.ToString().Data()));
425                         storeOk = gridSto->Put(aLocEntry);
426                         if (storeOk) {
427                                 Log("SHUTTLE",
428                                 Form("StoreOCDB - Object <%s> successfully put into %s storage",
429                                         aLocId.ToString().Data(), type));
430                                 Log(fCurrentDetector.Data(),
431                                         Form("StoreOCDB - Object <%s> successfully put into %s storage",
432                                         aLocId.ToString().Data(), type));
433                         } else  {
434                                 Log("SHUTTLE",
435                                         Form("StoreOCDB - Grid %s storage of object <%s> failed",
436                                         type, aLocId.ToString().Data()));
437                                 Log(fCurrentDetector.Data(),
438                                         Form("StoreOCDB - Grid %s storage of object <%s> failed",
439                                         type, aLocId.ToString().Data()));
440                                 result = kFALSE;
441                         }
442                 }
443                 
444                 if (!store || storeOk) {
445                         // removing local file...
446                         TString filename;
447                         localSto->IdToFilename(aLocId, filename);
448                         Log("SHUTTLE", Form("StoreOCDB - Removing local file %s", filename.Data()));
449                         RemoveFile(filename.Data());
450                 }
451         }
452         localEntries->Clear();
453
454         Int_t returnCode = 0;
455         
456         if (result == kFALSE)
457                 returnCode = -1;
458         else if (delayed != kFALSE)
459                 returnCode =  1;
460
461         Log("SHUTTLE", Form("StoreOCDB - Returning with %d (result = %d, delayed = %d)", returnCode, result, delayed));
462         Log(fCurrentDetector.Data(), Form("StoreOCDB - Returning with %d (result = %d, delayed = %d)", returnCode, result, delayed));
463         
464         return returnCode;
465 }
466
467 //______________________________________________________________________________________________
468 Bool_t AliShuttle::CleanReferenceStorage(const char* detector)
469 {
470         // clears the directory used to store reference files of a given subdetector
471   
472         AliCDBManager* man = AliCDBManager::Instance();
473         AliCDBStorage* sto = man->GetStorage(fgkLocalRefStorage);
474         TString localBaseFolder = sto->GetBaseFolder();
475
476         TString targetDir = GetRefFilePrefix(localBaseFolder.Data(), detector);
477         
478         Log("SHUTTLE", Form("CleanReferenceStorage - Cleaning %s", targetDir.Data()));
479
480         TString begin;
481         begin.Form("%d_", GetCurrentRun());
482         
483         TSystemDirectory* baseDir = new TSystemDirectory("/", targetDir);
484         if (!baseDir)
485                 return kTRUE;
486                 
487         TList* dirList = baseDir->GetListOfFiles();
488         delete baseDir;
489         
490         if (!dirList) return kTRUE;
491                         
492         if (dirList->GetEntries() < 3) 
493         {
494                 delete dirList;
495                 return kTRUE;
496         }
497                                 
498         Int_t nDirs = 0, nDel = 0;
499         TIter dirIter(dirList);
500         TSystemFile* entry = 0;
501
502         Bool_t success = kTRUE;
503         
504         while ((entry = dynamic_cast<TSystemFile*> (dirIter.Next())))
505         {                                       
506                 if (entry->IsDirectory())
507                         continue;
508                 
509                 TString fileName(entry->GetName());
510                 if (!fileName.BeginsWith(begin))
511                         continue;
512                         
513                 nDirs++;
514                                                 
515                 // delete file
516                 Int_t result = gSystem->Unlink(fileName.Data());
517                 
518                 if (result)
519                 {
520                         Log("SHUTTLE", Form("CleanReferenceStorage - Could not delete file %s!", fileName.Data()));
521                         success = kFALSE;
522                 } else {
523                         nDel++;
524                 }
525         }
526
527         if(nDirs > 0)
528                 Log("SHUTTLE", Form("CleanReferenceStorage - %d (over %d) reference files in folder %s were deleted.", 
529                         nDel, nDirs, targetDir.Data()));
530
531                 
532         delete dirList;
533         return success;
534
535
536
537
538
539
540   Int_t result = gSystem->GetPathInfo(targetDir, 0, (Long64_t*) 0, 0, 0);
541   if (result == 0)
542   {
543     // delete directory
544     result = gSystem->Exec(Form("rm -rf %s", targetDir.Data()));
545     if (result != 0)
546     {  
547       Log("SHUTTLE", Form("CleanReferenceStorage - Could not clean directory %s", targetDir.Data()));
548       return kFALSE;
549     }
550   }
551
552   result = gSystem->mkdir(targetDir, kTRUE);
553   if (result != 0)
554   {
555     Log("SHUTTLE", Form("CleanReferenceStorage - Error creating base directory %s", targetDir.Data()));
556     return kFALSE;
557   }
558         
559   return kTRUE;
560 }
561
562 //______________________________________________________________________________________________
563 Bool_t AliShuttle::StoreReferenceFile(const char* detector, const char* localFile, const char* gridFileName)
564 {
565         //
566         // Stores reference file directly (without opening it). This function stores the file locally.
567         //
568         // The file is stored under the following location: 
569         // <base folder of local reference storage>/<DET>/<RUN#>_<gridFileName>
570         // where <gridFileName> is the second parameter given to the function
571         // 
572         
573         if (fTestMode & kErrorStorage)
574         {
575                 Log(fCurrentDetector, "StoreReferenceFile - In TESTMODE - Simulating error while storing locally");
576                 return kFALSE;
577         }
578         
579         AliCDBManager* man = AliCDBManager::Instance();
580         AliCDBStorage* sto = man->GetStorage(fgkLocalRefStorage);
581         
582         TString localBaseFolder = sto->GetBaseFolder();
583         
584         TString target = GetRefFilePrefix(localBaseFolder.Data(), detector);    
585         target.Append(Form("/%d_%s", GetCurrentRun(), gridFileName));
586         
587         return CopyFileLocally(localFile, target);
588 }
589
590 //______________________________________________________________________________________________
591 Bool_t AliShuttle::StoreRunMetadataFile(const char* localFile, const char* gridFileName)
592 {
593         //
594         // Stores Run metadata file to the Grid, in the run folder
595         //
596         // Only GRP can call this function.
597         
598         if (fTestMode & kErrorStorage)
599         {
600                 Log(fCurrentDetector, "StoreRunMetaDataFile - In TESTMODE - Simulating error while storing locally");
601                 return kFALSE;
602         }
603         
604         AliCDBManager* man = AliCDBManager::Instance();
605         AliCDBStorage* sto = man->GetStorage(fgkLocalRefStorage);
606         
607         TString localBaseFolder = sto->GetBaseFolder();
608         
609         // Build Run level folder
610         // folder = /alice/data/year/lhcPeriod/runNb/raw
611         
612                 
613         TString lhcPeriod = GetLHCPeriod();     
614         if (lhcPeriod.Length() == 0) 
615         {
616                 Log("SHUTTLE","StoreRunMetaDataFile - LHCPeriod not found in logbook!");
617                 return 0;
618         }
619         
620         // TODO partitions with one detector only write data into LHCperiod_DET
621         TString partition = GetRunParameter("detector");
622         
623         if (partition.Length() > 0 && partition != "ALICE")
624         {
625                 lhcPeriod.Append(Form("_%s", partition.Data()));
626                 Log(fCurrentDetector, Form("Run data tags merged file will be written in %s", 
627                                 lhcPeriod.Data()));
628         }
629                 
630         TString target = Form("%s/GRP/RunMetadata%s%d/%s/%09d/raw/%s", 
631                                 localBaseFolder.Data(), fConfig->GetAlienPath(), GetCurrentYear(), 
632                                 lhcPeriod.Data(), GetCurrentRun(), gridFileName);
633                                         
634         return CopyFileLocally(localFile, target);
635 }
636
637 //______________________________________________________________________________________________
638 Bool_t AliShuttle::CopyFileLocally(const char* localFile, const TString& target)
639 {
640         //
641         // Stores file locally. Called by StoreReferenceFile and StoreRunMetadataFile
642         // Files are temporarily stored in the local reference storage. When the preprocessor 
643         // finishes, the Shuttle calls CopyFilesToGrid to transfer the files to AliEn 
644         // (in reference or run level folders)
645         //
646         
647         TString targetDir(target(0, target.Last('/')));
648         
649         //try to open base dir folder, if it does not exist
650         void* dir = gSystem->OpenDirectory(targetDir.Data());
651         if (dir == NULL) {
652                 if (gSystem->mkdir(targetDir.Data(), kTRUE)) {
653                         Log("SHUTTLE", Form("CopyFileLocally - Can't open directory <%s>", targetDir.Data()));
654                         return kFALSE;
655                 }
656
657         } else {
658                 gSystem->FreeDirectory(dir);
659         }
660         
661         Int_t result = 0;
662         
663         result = gSystem->GetPathInfo(localFile, 0, (Long64_t*) 0, 0, 0);
664         if (result)
665         {
666                 Log("SHUTTLE", Form("CopyFileLocally - %s does not exist", localFile));
667                 return kFALSE;
668         }
669
670         result = gSystem->GetPathInfo(target, 0, (Long64_t*) 0, 0, 0);
671         if (!result)
672         {
673                 Log("SHUTTLE", Form("CopyFileLocally - target file %s already exist, removing...", target.Data()));
674                 if (gSystem->Unlink(target.Data()))
675                 {
676                         Log("SHUTTLE", Form("CopyFileLocally - Could not remove existing target file %s!", target.Data()));
677                         return kFALSE;
678                 }
679         }       
680         
681         result = gSystem->CopyFile(localFile, target);
682
683         if (result == 0)
684         {
685                 Log("SHUTTLE", Form("CopyFileLocally - File %s stored locally to %s", localFile, target.Data()));
686                 return kTRUE;
687         }
688         else
689         {
690                 Log("SHUTTLE", Form("CopyFileLocally - Could not store file %s to %s! Error code = %d", 
691                                 localFile, target.Data(), result));
692                 return kFALSE;
693         }       
694
695
696
697 }
698
699 //______________________________________________________________________________________________
700 Bool_t AliShuttle::CopyFilesToGrid(const char* type)
701 {
702         //
703         // Transfers local files to the Grid. Local files can be reference files 
704         // or run metadata file (from GRP only).
705         //
706         // According to the type (ref, metadata) the files are stored under the following location: 
707         // ref --> <base folder of reference storage>/<DET>/<RUN#>_<gridFileName>
708         // metadata --> <run data folder>/<MetadataFileName>
709         //
710                 
711         AliCDBManager* man = AliCDBManager::Instance();
712         AliCDBStorage* sto = man->GetStorage(fgkLocalRefStorage);
713         if (!sto)
714                 return kFALSE;
715         TString localBaseFolder = sto->GetBaseFolder();
716         
717         TString dir;
718         TString alienDir;
719         TString begin;
720         
721         if (strcmp(type, "reference") == 0) 
722         {
723                 dir = GetRefFilePrefix(localBaseFolder.Data(), fCurrentDetector.Data());
724                 AliCDBStorage* gridSto = man->GetStorage(fgkMainRefStorage);
725                 if (!gridSto)
726                         return kFALSE;
727                 TString gridBaseFolder = gridSto->GetBaseFolder();
728                 alienDir = GetRefFilePrefix(gridBaseFolder.Data(), fCurrentDetector.Data());
729                 begin = Form("%d_", GetCurrentRun());
730         } 
731         else if (strcmp(type, "metadata") == 0)
732         {
733                         
734                 TString lhcPeriod = GetLHCPeriod();
735         
736                 if (lhcPeriod.Length() == 0) 
737                 {
738                         Log("SHUTTLE","CopyFilesToGrid - LHCPeriod not found in logbook!");
739                         return 0;
740                 }
741                 
742                 // TODO partitions with one detector only write data into LHCperiod_DET
743                 TString partition = GetRunParameter("detector");
744         
745                 if (partition.Length() > 0 && partition != "ALICE")
746                 {
747                         lhcPeriod.Append(Form("_%s", partition.Data()));
748                 }
749                 
750                 dir = Form("%s/GRP/RunMetadata%s%d/%s/%09d/raw", 
751                                 localBaseFolder.Data(), fConfig->GetAlienPath(), GetCurrentYear(), 
752                                 lhcPeriod.Data(), GetCurrentRun());
753                 alienDir = dir(dir.Index(fConfig->GetAlienPath()), dir.Length());
754                 
755                 begin = "";
756         }
757         else 
758         {
759                 Log("SHUTTLE", "CopyFilesToGrid - Unexpected: type label must be reference or metadata!");
760                 return kFALSE;
761         }
762                 
763         TSystemDirectory* baseDir = new TSystemDirectory("/", dir);
764         if (!baseDir)
765                 return kTRUE;
766                 
767         TList* dirList = baseDir->GetListOfFiles();
768         delete baseDir;
769         
770         if (!dirList) return kTRUE;
771                 
772         if (dirList->GetEntries() < 3) 
773         {
774                 delete dirList;
775                 return kTRUE;
776         }
777                         
778         if (!gGrid)
779         { 
780                 Log("SHUTTLE", "CopyFilesToGrid - Connection to Grid failed: Cannot continue!");
781                 delete dirList;
782                 return kFALSE;
783         }
784         
785         Int_t nDirs = 0, nTransfer = 0;
786         TIter dirIter(dirList);
787         TSystemFile* entry = 0;
788
789         Bool_t success = kTRUE;
790         Bool_t first = kTRUE;
791         
792         while ((entry = dynamic_cast<TSystemFile*> (dirIter.Next())))
793         {                       
794                 if (entry->IsDirectory())
795                         continue;
796                         
797                 TString fileName(entry->GetName());
798                 if (!fileName.BeginsWith(begin))
799                         continue;
800                         
801                 nDirs++;
802                         
803                 if (first)
804                 {
805                         first = kFALSE;
806                         // check that folder exists, otherwise create it
807                         TGridResult* result = gGrid->Ls(alienDir.Data(), "a");
808                         
809                         if (!result)
810                         {
811                                 delete dirList;
812                                 return kFALSE;
813                         }
814                         
815                         if (!result->GetFileName(1)) // TODO: It looks like element 0 is always 0!!
816                         {
817                                 // TODO It does not work currently! Bug in TAliEn::Mkdir
818                                 // TODO Manually fixed in local root v5-16-00
819                                 if (!gGrid->Mkdir(alienDir.Data(),"-p",0))
820                                 {
821                                         Log("SHUTTLE", Form("CopyFilesToGrid - Cannot create directory %s",
822                                                         alienDir.Data()));
823                                         delete dirList;
824                                         return kFALSE;
825                                 } else {
826                                         Log("SHUTTLE",Form("CopyFilesToGrid - Folder %s created", alienDir.Data()));
827                                 }
828                                 
829                         } else {
830                                         Log("SHUTTLE",Form("CopyFilesToGrid - Folder %s found", alienDir.Data()));
831                         }
832                 }
833                         
834                 TString fullLocalPath;
835                 fullLocalPath.Form("%s/%s", dir.Data(), fileName.Data());
836                 
837                 TString fullGridPath;
838                 fullGridPath.Form("alien://%s/%s", alienDir.Data(), fileName.Data());
839
840                 Bool_t result = TFile::Cp(fullLocalPath, fullGridPath);
841                 
842                 if (result)
843                 {
844                         Log("SHUTTLE", Form("CopyFilesToGrid - Copying local file %s to %s succeeded!", 
845                                                 fullLocalPath.Data(), fullGridPath.Data()));
846                         RemoveFile(fullLocalPath);
847                         nTransfer++;
848                 }
849                 else
850                 {
851                         Log("SHUTTLE", Form("CopyFilesToGrid - Copying local file %s to %s FAILED!", 
852                                                 fullLocalPath.Data(), fullGridPath.Data()));
853                         success = kFALSE;
854                 }
855         }
856
857         Log("SHUTTLE", Form("CopyFilesToGrid - %d (over %d) files in folder %s copied to Grid.", 
858                                                 nTransfer, nDirs, dir.Data()));
859
860                 
861         delete dirList;
862         return success;
863 }
864
865 //______________________________________________________________________________________________
866 const char* AliShuttle::GetRefFilePrefix(const char* base, const char* detector)
867 {
868         //
869         // Get folder name of reference files 
870         //
871
872         TString offDetStr(GetOfflineDetName(detector));
873         static TString dir;
874         if (offDetStr == "ITS" || offDetStr == "MUON" || offDetStr == "PHOS")
875         {
876                 dir.Form("%s/%s/%s", base, offDetStr.Data(), detector);
877         } else {
878                 dir.Form("%s/%s", base, offDetStr.Data());
879         }
880         
881         return dir.Data();
882 }
883
884 //______________________________________________________________________________________________
885 void AliShuttle::CleanLocalStorage(const TString& uri)
886 {
887         //
888         // Called in case the preprocessor is declared failed. Remove remaining objects from the local storages.
889         //
890
891         const char* type = 0;
892         if(uri == fgkLocalCDB) {
893                 type = "OCDB";
894         } else if(uri == fgkLocalRefStorage) {
895                 type = "Reference";
896         } else {
897                 AliError(Form("Invalid storage URI: %s", uri.Data()));
898                 return;
899         }
900
901         AliCDBManager* man = AliCDBManager::Instance();
902
903         // open local storage
904         AliCDBStorage *localSto = man->GetStorage(uri);
905         if(!localSto) {
906                 Log("SHUTTLE",
907                         Form("CleanLocalStorage - cannot activate local %s storage", type));
908                 return;
909         }
910
911         TString filename(Form("%s/%s/*/Run*_v%d_s*.root",
912                 localSto->GetBaseFolder().Data(), GetOfflineDetName(fCurrentDetector.Data()), GetCurrentRun()));
913
914         AliDebug(2, Form("filename = %s", filename.Data()));
915
916         Log("SHUTTLE", Form("Removing remaining local files for run %d and detector %s ...",
917                 GetCurrentRun(), fCurrentDetector.Data()));
918
919         RemoveFile(filename.Data());
920
921 }
922
923 //______________________________________________________________________________________________
924 void AliShuttle::RemoveFile(const char* filename)
925 {
926         //
927         // removes local file
928         //
929
930         TString command(Form("rm -f %s", filename));
931
932         Int_t result = gSystem->Exec(command.Data());
933         if(result != 0)
934         {
935                 Log("SHUTTLE", Form("RemoveFile - %s: Cannot remove file %s!",
936                         fCurrentDetector.Data(), filename));
937         }
938 }
939
940 //______________________________________________________________________________________________
941 AliShuttleStatus* AliShuttle::ReadShuttleStatus()
942 {
943         //
944         // Reads the AliShuttleStatus from the CDB
945         //
946
947         if (fStatusEntry){
948                 delete fStatusEntry;
949                 fStatusEntry = 0;
950         }
951
952         Int_t path1 = GetCurrentRun()/10000;
953         fStatusEntry = AliCDBManager::Instance()->GetStorage(GetLocalCDB())
954                 ->Get(Form("/SHUTTLE/%s/%d", fCurrentDetector.Data(), path1), GetCurrentRun());
955
956         if (!fStatusEntry) return 0;
957         fStatusEntry->SetOwner(1);
958
959         AliShuttleStatus* status = dynamic_cast<AliShuttleStatus*> (fStatusEntry->GetObject());
960         if (!status) {
961                 AliError("Invalid object stored to CDB!");
962                 return 0;
963         }
964
965         return status;
966 }
967
968 //______________________________________________________________________________________________
969 Bool_t AliShuttle::WriteShuttleStatus(AliShuttleStatus* status)
970 {
971         //
972         // writes the status for one subdetector
973         //
974
975         if (fStatusEntry){
976                 delete fStatusEntry;
977                 fStatusEntry = 0;
978         }
979
980         Int_t run = GetCurrentRun();
981         Int_t path1 = run/10000;
982         TString path1_string = Form("%d",path1);
983
984         AliCDBId id(AliCDBPath("SHUTTLE", fCurrentDetector, path1_string), run, run);
985
986         fStatusEntry = new AliCDBEntry(status, id, new AliCDBMetaData);
987         fStatusEntry->SetOwner(1);
988
989         Int_t logLevel = AliLog::GetGlobalLogLevel();
990         AliLog::SetGlobalLogLevel(AliLog::kError);
991
992         UInt_t result = AliCDBManager::Instance()->GetStorage(fgkLocalCDB)->Put(fStatusEntry);
993
994         AliLog::SetGlobalLogLevel((AliLog::EType_t)logLevel);
995
996         if (!result) {
997                 Log("SHUTTLE", Form("WriteShuttleStatus - Failed for %s, run %d",
998                                                 fCurrentDetector.Data(), run));
999                 return kFALSE;
1000         }
1001         
1002         SendMLDetInfo();
1003
1004         return kTRUE;
1005 }
1006
1007 //______________________________________________________________________________________________
1008 void AliShuttle::UpdateShuttleStatus(AliShuttleStatus::Status newStatus, Bool_t increaseCount)
1009 {
1010         //
1011         // changes the AliShuttleStatus for the given detector and run to the given status
1012         //
1013
1014         if (!fStatusEntry){
1015                 AliError("UNEXPECTED: fStatusEntry empty");
1016                 return;
1017         }
1018
1019         AliShuttleStatus* status = dynamic_cast<AliShuttleStatus*> (fStatusEntry->GetObject());
1020
1021         if (!status){
1022                 Log("SHUTTLE", "UpdateShuttleStatus - UNEXPECTED: status could not be read from current CDB entry");
1023                 return;
1024         }
1025
1026         TString actionStr = Form("UpdateShuttleStatus - %s: Changing state from %s to %s",
1027                                 fCurrentDetector.Data(),
1028                                 status->GetStatusName(),
1029                                 status->GetStatusName(newStatus));
1030         Log("SHUTTLE", actionStr);
1031         SetLastAction(actionStr);
1032
1033         status->SetStatus(newStatus);
1034         if (increaseCount) status->IncreaseCount();
1035
1036         Int_t logLevel = AliLog::GetGlobalLogLevel();
1037         AliLog::SetGlobalLogLevel(AliLog::kError);
1038
1039         AliCDBManager::Instance()->GetStorage(fgkLocalCDB)->Put(fStatusEntry);
1040
1041         AliLog::SetGlobalLogLevel((AliLog::EType_t)logLevel);
1042
1043         SendMLDetInfo();
1044 }
1045
1046 //______________________________________________________________________________________________
1047 void AliShuttle::SendMLDetInfo()
1048 {
1049         //
1050         // sends ML information about the current status of the current detector being processed
1051         //
1052         
1053         AliShuttleStatus* status = dynamic_cast<AliShuttleStatus*> (fStatusEntry->GetObject());
1054         
1055         if (!status){
1056                 Log("SHUTTLE", "SendMLDetInfo - UNEXPECTED: status could not be read from current CDB entry");
1057                 return;
1058         }
1059         
1060         TMonaLisaText  mlStatus(Form("%s_status", fCurrentDetector.Data()), status->GetStatusName());
1061         TMonaLisaValue mlRetryCount(Form("%s_count", fCurrentDetector.Data()), status->GetCount());
1062
1063         TList mlList;
1064         mlList.Add(&mlStatus);
1065         mlList.Add(&mlRetryCount);
1066
1067         TString mlID;
1068         mlID.Form("%d", GetCurrentRun());
1069         fMonaLisa->SendParameters(&mlList, mlID);
1070 }
1071
1072 //______________________________________________________________________________________________
1073 Bool_t AliShuttle::ContinueProcessing()
1074 {
1075         // this function reads the AliShuttleStatus information from CDB and
1076         // checks if the processing should be continued
1077         // if yes it returns kTRUE and updates the AliShuttleStatus with nextStatus
1078
1079         if (!fConfig->HostProcessDetector(fCurrentDetector))
1080                 return kFALSE;
1081
1082         AliPreprocessor* aPreprocessor =
1083                 dynamic_cast<AliPreprocessor*> (fPreprocessorMap.GetValue(fCurrentDetector));
1084         if (!aPreprocessor)
1085         {
1086                 Log("SHUTTLE", Form("ContinueProcessing - %s: no preprocessor registered", fCurrentDetector.Data()));
1087                 return kFALSE;
1088         }
1089
1090         AliShuttleLogbookEntry::Status entryStatus =
1091                 fLogbookEntry->GetDetectorStatus(fCurrentDetector);
1092
1093         if (entryStatus != AliShuttleLogbookEntry::kUnprocessed) {
1094                 Log("SHUTTLE", Form("ContinueProcessing - %s is %s",
1095                                 fCurrentDetector.Data(),
1096                                 fLogbookEntry->GetDetectorStatusName(entryStatus)));
1097                 return kFALSE;
1098         }
1099
1100         // if we get here, according to Shuttle logbook subdetector is in UNPROCESSED state
1101
1102         // check if current run is first unprocessed run for current detector
1103         if (fConfig->StrictRunOrder(fCurrentDetector) &&
1104                 !fFirstUnprocessed[GetDetPos(fCurrentDetector)])
1105         {
1106                 if (fTestMode == kNone)
1107                 {
1108                         Log("SHUTTLE", Form("ContinueProcessing - %s requires strict run ordering"
1109                                         " but this is not the first unprocessed run!"));
1110                         return kFALSE;
1111                 }
1112                 else
1113                 {
1114                         Log("SHUTTLE", Form("ContinueProcessing - In TESTMODE - "
1115                                         "Although %s requires strict run ordering "
1116                                         "and this is not the first unprocessed run, "
1117                                         "the SHUTTLE continues"));
1118                 }
1119         }
1120
1121         // Is the subdetector processed first time for this run?
1122         fFirstProcessing = kFALSE;
1123
1124         AliShuttleStatus* status = ReadShuttleStatus();
1125         if (!status) {
1126                 // first time
1127                 Log("SHUTTLE", Form("ContinueProcessing - %s: Processing first time",
1128                                 fCurrentDetector.Data()));
1129                 status = new AliShuttleStatus(AliShuttleStatus::kStarted);
1130                 fFirstProcessing = kTRUE;
1131                 return WriteShuttleStatus(status);
1132         }
1133
1134         // The following case shouldn't happen if Shuttle Logbook was correctly updated.
1135         // If it happens it may mean Logbook updating failed... let's do it now!
1136         if (status->GetStatus() == AliShuttleStatus::kDone ||
1137             status->GetStatus() == AliShuttleStatus::kFailed ||
1138             status->GetStatus() == AliShuttleStatus::kSkipped) {
1139                 Log("SHUTTLE", Form("ContinueProcessing - %s is already %s. Updating Shuttle Logbook",
1140                                         fCurrentDetector.Data(),
1141                                         status->GetStatusName(status->GetStatus())));
1142                 
1143                 if (status->GetStatus() == AliShuttleStatus::kSkipped)
1144                 {
1145                         UpdateShuttleLogbook(fCurrentDetector.Data(), "DONE");
1146                 }
1147                 else
1148                         UpdateShuttleLogbook(fCurrentDetector.Data(), status->GetStatusName(status->GetStatus()));
1149                         
1150                 return kFALSE;
1151         }
1152
1153         if (status->GetStatus() == AliShuttleStatus::kStoreStarted || status->GetStatus() == AliShuttleStatus::kStoreDelayed ||status->GetStatus() == AliShuttleStatus::kStoreError) {
1154                 Log("SHUTTLE",
1155                         Form("ContinueProcessing - %s: Grid storage of one or more "
1156                                 "objects failed. Trying again now",
1157                                 fCurrentDetector.Data()));
1158                 StoreOCDB();
1159                 return kFALSE;
1160         }
1161
1162         // if we get here, there is a restart
1163         Bool_t cont = kFALSE;
1164
1165         // abort conditions
1166         if (status->GetCount() >= fConfig->GetMaxRetries()) {
1167                 Log("SHUTTLE", Form("ContinueProcessing - %s failed %d times in status %s - "
1168                                 "Updating Shuttle Logbook", fCurrentDetector.Data(),
1169                                 status->GetCount(), status->GetStatusName()));
1170                 UpdateShuttleLogbook(fCurrentDetector.Data(), "FAILED");
1171                 UpdateShuttleStatus(AliShuttleStatus::kFailed);
1172
1173                 // there may still be objects in local OCDB and reference storage
1174                 // and FXS databases may be not updated: do it now!
1175                 
1176                 // TODO Currently disabled, we want to keep files in case of failure!
1177                 // CleanLocalStorage(fgkLocalCDB);
1178                 // CleanLocalStorage(fgkLocalRefStorage);
1179                 // UpdateTableFailCase();
1180                 
1181                 // Send mail to detector expert!
1182                 Log("SHUTTLE", Form("ContinueProcessing - Sending mail to %s expert...", 
1183                                     fCurrentDetector.Data()));
1184                 // det experts in to
1185                 TString to="";
1186                 TIter *iterExperts = 0;
1187                 iterExperts = new TIter(fConfig->GetResponsibles(fCurrentDetector));
1188                 TObjString *anExpert=0;
1189                 while ((anExpert = (TObjString*) iterExperts->Next()))
1190                         {
1191                                 to += Form("%s, \n", anExpert->GetName());
1192                         }
1193                 delete iterExperts;
1194                 
1195                 if (to.Length() > 0)
1196                         to.Remove(to.Length()-3);
1197                 AliDebug(2, Form("to: %s",to.Data()));
1198
1199                 if (to.IsNull()) {
1200                         Log("SHUTTLE", Form("List of %s responsibles not set!", fCurrentDetector.Data()));
1201                         return kFALSE;
1202                 }
1203
1204                 Log(fCurrentDetector.Data(), Form("ContinueProcessing - Sending mail to %s expert(s):", 
1205                                     fCurrentDetector.Data()));
1206                 Log(fCurrentDetector.Data(), Form("\n%s", to.Data()));
1207                 if (!SendMail(kPPEMail))
1208                         Log("SHUTTLE", Form("ContinueProcessing - Could not send mail to %s expert",
1209                                             fCurrentDetector.Data()));
1210
1211         } else {
1212                 Log("SHUTTLE", Form("ContinueProcessing - %s: restarting. "
1213                                 "Aborted before with %s. Retry number %d.", fCurrentDetector.Data(),
1214                                 status->GetStatusName(), status->GetCount()));
1215                 Bool_t increaseCount = kTRUE;
1216                 if (status->GetStatus() == AliShuttleStatus::kDCSError || 
1217                     status->GetStatus() == AliShuttleStatus::kDCSStarted ||
1218                     status->GetStatus() == AliShuttleStatus::kFXSError ||
1219                     status->GetStatus() == AliShuttleStatus::kOCDBError)
1220                                 increaseCount = kFALSE;
1221                                 
1222                 UpdateShuttleStatus(AliShuttleStatus::kStarted, increaseCount);
1223                 cont = kTRUE;
1224         }
1225
1226         return cont;
1227 }
1228
1229 //______________________________________________________________________________________________
1230 void AliShuttle::SendMLRunInfo(const char* status)
1231 {
1232         // 
1233         // Send information about this run to ML
1234         
1235         TMonaLisaText  mlStatus("SHUTTLE_status", status);
1236         TString runType(fLogbookEntry->GetRunType());
1237         if (strlen(fLogbookEntry->GetRunParameter("log")) > 0){
1238
1239                 runType += "(";
1240                 runType += fLogbookEntry->GetRunParameter("log");
1241                 runType += ")";
1242         }
1243         if (fLogbookEntry->GetDATestMode()){
1244                 runType += " (DATest)";
1245         }
1246         TMonaLisaText  mlRunType("SHUTTLE_runtype", runType);
1247
1248         TList mlList;
1249         mlList.Add(&mlStatus);
1250         mlList.Add(&mlRunType);
1251
1252         TString mlID;
1253         mlID.Form("%d", GetCurrentRun());
1254         fMonaLisa->SendParameters(&mlList, mlID);       
1255 }
1256
1257 //______________________________________________________________________________________________
1258 Int_t AliShuttle::GetMem(Int_t pid)
1259 {
1260         // invokes ps to get the memory consumption of the process <pid>
1261         // returns -1 in case of error
1262         
1263         TString checkStr;
1264         checkStr.Form("ps -o vsize --pid %d | tail -n 1", pid);
1265         FILE* pipe = gSystem->OpenPipe(checkStr, "r");
1266         if (!pipe)
1267         {
1268                 Log("SHUTTLE", Form("Process - Error: "
1269                         "Could not open pipe to %s", checkStr.Data()));
1270                 return -1;
1271         }
1272                 
1273         char buffer[100];
1274         if (!fgets(buffer, 100, pipe))
1275         {
1276                 Log("SHUTTLE", "Process - Error: ps did not return anything");
1277                 gSystem->ClosePipe(pipe);
1278                 return -1;
1279         }
1280         gSystem->ClosePipe(pipe);
1281         
1282         //Log("SHUTTLE", Form("ps returned %s", buffer));
1283         
1284         Int_t mem = 0;
1285         if ((sscanf(buffer, "%d\n", &mem) != 1) || !mem)
1286         {
1287                 Log("SHUTTLE", "Process - Error: Could not parse output of ps");
1288                 return -1;
1289         }
1290         
1291         return mem;
1292 }
1293
1294 //______________________________________________________________________________________________
1295 Bool_t AliShuttle::Process(AliShuttleLogbookEntry* entry)
1296 {
1297         //
1298         // Makes data retrieval for all detectors in the configuration.
1299         // entry: Shuttle logbook entry, contains run paramenters and status of detectors
1300         // (Unprocessed, Inactive, Failed or Done).
1301         // Returns kFALSE in case of error occured and kTRUE otherwise
1302         //
1303
1304         if (!entry) return kFALSE;
1305
1306         fLogbookEntry = entry;
1307
1308         Log("SHUTTLE", Form("\t\t\t^*^*^*^*^*^*^*^*^*^*^*^* run %d: START ^*^*^*^*^*^*^*^*^*^*^*^*",
1309                                         GetCurrentRun()));
1310
1311         CountOpenRuns();
1312         
1313         // Send the information to ML
1314         SendMLRunInfo("Processing");
1315
1316         if (fLogbookEntry->IsDone())
1317         {
1318                 Log("SHUTTLE","Process - Shuttle is already DONE. Updating logbook");
1319                 UpdateShuttleLogbook("shuttle_done");
1320                 fLogbookEntry = 0;
1321                 return kTRUE;
1322         }
1323
1324         // read test mode if flag is set
1325         if (fReadTestMode)
1326         {
1327                 fTestMode = kNone;
1328                 TString logEntry(entry->GetRunParameter("log"));
1329                 //printf("log entry = %s\n", logEntry.Data());
1330                 TString searchStr("Testmode: ");
1331                 Int_t pos = logEntry.Index(searchStr.Data());
1332                 //printf("%d\n", pos);
1333                 if (pos >= 0)
1334                 {
1335                         TSubString subStr = logEntry(pos + searchStr.Length(), logEntry.Length());
1336                         //printf("%s\n", subStr.String().Data());
1337                         TString newStr(subStr.Data());
1338                         TObjArray* token = newStr.Tokenize(' ');
1339                         if (token)
1340                         {
1341                                 //token->Print();
1342                                 TObjString* tmpStr = dynamic_cast<TObjString*> (token->First());
1343                                 if (tmpStr)
1344                                 {
1345                                         Int_t testMode = tmpStr->String().Atoi();
1346                                         if (testMode > 0)
1347                                         {
1348                                                 Log("SHUTTLE", Form("Process - Enabling test mode %d", testMode));
1349                                                 SetTestMode((TestMode) testMode);
1350                                         }
1351                                 }
1352                                 delete token;          
1353                         }
1354                 }
1355         }
1356                 
1357         fLogbookEntry->Print("all");
1358
1359         // Initialization
1360         Bool_t hasError = kFALSE;
1361
1362         // Set the CDB and Reference folders according to the year
1363
1364         // build cdb paths (repeat each time, run might be a DATest run)
1365         if (!fLogbookEntry->GetDATestMode()){
1366                 fgkMainCDB.Form("alien://folder=%s%d/OCDB?user=alidaq?cacheFold=/tmp/OCDBCache", 
1367                                 fConfig->GetAlienPath(), GetCurrentYear());
1368                 
1369                 fgkMainRefStorage.Form("alien://folder=%s%d/Reference?user=alidaq?cacheFold=/tmp/OCDBCache", 
1370                                        fConfig->GetAlienPath(), GetCurrentYear());
1371         }
1372         else {
1373                 fgkMainCDB.Form("alien://folder=%s%d/DATest/OCDB?user=alidaq?cacheFold=/tmp/OCDBCache",
1374                                 fConfig->GetAlienPath(), GetCurrentYear());
1375                 
1376                 fgkMainRefStorage.Form("alien://folder=%s%d/DATest/Reference?user=alidaq?cacheFold=/tmp/OCDBCache",
1377                                        fConfig->GetAlienPath(), GetCurrentYear());
1378         }
1379
1380         AliDebug(2,Form("Main CDB storage = %s",fgkMainCDB.Data()));
1381         AliDebug(2,Form("Main Reference storage = %s",fgkMainRefStorage.Data()));
1382
1383         // Loop on detectors in the configuration
1384         TIter iter(fConfig->GetDetectors());
1385         TObjString* aDetector = 0;
1386
1387         Bool_t first = kTRUE;
1388
1389         while ((aDetector = (TObjString*) iter.Next()))
1390         {
1391                 fCurrentDetector = aDetector->String();
1392
1393                 if (ContinueProcessing() == kFALSE) 
1394                         continue;
1395                 
1396                 if (first)
1397                 {
1398                   // only read QueryCDB when needed and only once
1399                   AliCDBStorage *mainCDBSto = AliCDBManager::Instance()->GetStorage(fgkMainCDB);
1400                   if(mainCDBSto) mainCDBSto->QueryCDB(GetCurrentRun());
1401                   AliCDBStorage *mainRefSto = AliCDBManager::Instance()->GetStorage(fgkMainRefStorage);
1402                   if(mainRefSto) mainRefSto->QueryCDB(GetCurrentRun());
1403                   first = kFALSE;
1404                 }
1405
1406                 Log("SHUTTLE", Form("\t\t\t****** run %d - %s: START  ******",
1407                                                 GetCurrentRun(), aDetector->GetName()));
1408
1409                 for(Int_t iSys=0;iSys<3;iSys++) fFXSCalled[iSys]=kFALSE;
1410                 
1411                 Int_t initialMem = GetMem(getpid());
1412                 Log("SHUTTLE", Form("Memory consumption before forking is %d", initialMem));
1413
1414                 Log(fCurrentDetector.Data(), "Process - Starting processing");
1415
1416                 Int_t pid = fork();
1417
1418                 if (pid < 0)
1419                 {
1420                         Log("SHUTTLE", "Process - ERROR: Forking failed");
1421                 }
1422                 else if (pid > 0)
1423                 {
1424                         // parent
1425                         Log("SHUTTLE", Form("Process - In parent process of %d - %s: Starting monitoring",
1426                                                         GetCurrentRun(), aDetector->GetName()));
1427
1428                         Long_t begin = time(0);
1429
1430                         int status; // to be used with waitpid, on purpose an int (not Int_t)!
1431                         while (waitpid(pid, &status, WNOHANG) == 0)
1432                         {
1433                                 Long_t expiredTime = time(0) - begin;
1434
1435                                 if (expiredTime > fConfig->GetPPTimeOut())
1436                                 {
1437                                         TString logMsg;
1438                                         AliShuttleStatus *currentStatus = ReadShuttleStatus();
1439                                         AliShuttleStatus::Status newStatus = AliShuttleStatus::kInvalid;
1440                                         
1441                                         if (currentStatus->GetStatus() == AliShuttleStatus::kDCSStarted)
1442                                         {
1443                                                 // in case the pp goes in TimeOut while retrieving the DCS DPs
1444                                                 // set status to kDCSError
1445                                                 
1446                                                 logMsg.Form("Process - Process of %s timed out while retrieving the DCS DataPoints. Run time: %d seconds. Killing... and setting status to DCSError.",
1447                                                                 fCurrentDetector.Data(), expiredTime);
1448                                                 newStatus = AliShuttleStatus::kDCSError;
1449                                         }
1450                                         else if (currentStatus->GetStatus() <= AliShuttleStatus::kPPDone)
1451                                         {
1452                                                 // in case pp not yet done set status to kPPTimeOut
1453                                         
1454                                                 logMsg.Form("Process - Process of %s timed out. Run time: %d seconds. Killing...",
1455                                                                 fCurrentDetector.Data(), expiredTime);
1456                                                 newStatus = AliShuttleStatus::kPPTimeOut;
1457                                         }
1458                                         else if (currentStatus->GetStatus() == AliShuttleStatus::kStoreStarted)
1459                                         {
1460                                                 // in case the pp goes in TimeOut while storing the objects in the OCDB
1461                                                 // set status to kStoreError
1462                                                 
1463                                                 logMsg.Form("Process - Process of %s timed out while storing the OCDB object. Run time: %d seconds. Killing... and setting status to StoreError.",
1464                                                                 fCurrentDetector.Data(), expiredTime);
1465                                                 newStatus = AliShuttleStatus::kStoreError;
1466                                         }
1467                                         else 
1468                                         {
1469                                                 // in other cases don't change the status
1470                                                 
1471                                                 logMsg.Form("Process - Process of %s timed out in status = %s. Run time: %d seconds. Killing... without changing the status",
1472                                                                 fCurrentDetector.Data(), currentStatus->GetStatusName(), expiredTime);
1473                                         }
1474                                 
1475                                         Log("SHUTTLE", logMsg);
1476                                         Log(fCurrentDetector, logMsg);
1477
1478                                         kill(pid, 9);
1479
1480                                         if (newStatus != AliShuttleStatus::kInvalid)
1481                                                 UpdateShuttleStatus(newStatus);
1482                                         hasError = kTRUE;
1483
1484                                         gSystem->Sleep(1000);
1485                                 }
1486                                 else
1487                                 {
1488                                         gSystem->Sleep(1000);
1489                                         
1490                                         Int_t mem = GetMem(pid);
1491
1492                                         if (mem < 0)
1493                                                 continue;
1494                                                 
1495                                         mem -= initialMem;
1496                                         if (mem < 0)
1497                                                 mem = 0;
1498                                         
1499                                         if (expiredTime % 60 == 0)
1500                                         {
1501                                                 Log("SHUTTLE", Form("Process - %s: Checking process. "
1502                                                         "Run time: %d seconds - Memory consumption: %d KB",
1503                                                         fCurrentDetector.Data(), expiredTime, mem));
1504                                                 SendAlive();
1505                                         }
1506                                         
1507                                         if (mem > fConfig->GetPPMaxMem())
1508                                         {
1509                                                 TString tmp;
1510                                                 tmp.Form("Process - Process exceeds maximum allowed memory "
1511                                                         "(%d KB > %d KB). Killing...",
1512                                                         mem, fConfig->GetPPMaxMem());
1513                                                 Log("SHUTTLE", tmp);
1514                                                 Log(fCurrentDetector, tmp);
1515         
1516                                                 kill(pid, 9);
1517         
1518                                                 UpdateShuttleStatus(AliShuttleStatus::kPPOutOfMemory);
1519                                                 hasError = kTRUE;
1520         
1521                                                 gSystem->Sleep(1000);
1522                                         }
1523                                 }
1524                         }
1525
1526                         Log("SHUTTLE", Form("Process - In parent process of %d - %s: Client has terminated.",
1527                                                                 GetCurrentRun(), aDetector->GetName()));
1528
1529                         if (WIFEXITED(status))
1530                         {
1531                                 Int_t returnCode = WEXITSTATUS(status);
1532
1533                                 Log("SHUTTLE", Form("Process - %s: the return code is %d", fCurrentDetector.Data(),
1534                                                                                 returnCode));
1535
1536                                 if (returnCode == 0) hasError = kTRUE;
1537                         }
1538                 }
1539                 else if (pid == 0)
1540                 {
1541                         // child
1542                         Log("SHUTTLE", Form("Process - In child process of %d - %s", GetCurrentRun(),
1543                                 aDetector->GetName()));
1544
1545                         Log("SHUTTLE", Form("Process - Redirecting output to %s log",fCurrentDetector.Data()));
1546
1547                         if ((freopen(GetLogFileName(fCurrentDetector), "a", stdout)) == 0)
1548                         {
1549                                 Log("SHUTTLE", "Process - Could not freopen stdout");
1550                         }
1551                         else
1552                         {
1553                                 fOutputRedirected = kTRUE;
1554                                 if ((dup2(fileno(stdout), fileno(stderr))) < 0)
1555                                         Log("SHUTTLE", "Process - Could not redirect stderr");
1556                                 
1557                         }
1558                         
1559                         TString wd = gSystem->WorkingDirectory();
1560                         Int_t dir_lev1 = GetCurrentRun()/10000;
1561                         TString tmpDir = Form("%s/%d/%d/%s_process", GetShuttleTempDir(), 
1562                                 dir_lev1, GetCurrentRun(), fCurrentDetector.Data());
1563                         
1564                         Int_t result = gSystem->GetPathInfo(tmpDir.Data(), 0, (Long64_t*) 0, 0, 0);
1565                         if (!result) // temp dir already exists!
1566                         {
1567                                 Log(fCurrentDetector.Data(), 
1568                                         Form("Process - %s dir already exists! Removing...", tmpDir.Data()));
1569                                 gSystem->Exec(Form("rm -rf %s",tmpDir.Data()));         
1570                         } 
1571                         
1572                         if (gSystem->mkdir(tmpDir.Data(), 1))
1573                         {
1574                                 Log(fCurrentDetector.Data(), "Process - could not make temp directory!!");
1575                                 gSystem->Exit(1);
1576                         }
1577                         
1578                         if (!gSystem->ChangeDirectory(tmpDir.Data())) 
1579                         {
1580                                 Log(fCurrentDetector.Data(), "Process - could not change directory!!");
1581                                 gSystem->Exit(1);                       
1582                         }
1583                         
1584                         Int_t success = ProcessCurrentDetector();
1585
1586                         gSystem->ChangeDirectory(wd.Data());
1587                                                 
1588                         if (success == 1) // Preprocessor finished successfully!
1589                         { 
1590                                 // remove temporary folder or DCS map
1591                                 if (!fConfig->KeepTempFolder())
1592                                 {
1593                                         gSystem->Exec(Form("rm -rf %s",tmpDir.Data()));
1594                                 } else if (!fConfig->KeepDCSMap())
1595                                 {
1596                                         gSystem->Exec(Form("rm -f %s/DCSMap.root",tmpDir.Data()));
1597                                 }
1598                                 
1599                                 // Update time_processed field in FXS DB
1600                                 if (UpdateTable() == kFALSE)
1601                                         Log("SHUTTLE", Form("Process - %s: Could not update FXS databases!", 
1602                                                         fCurrentDetector.Data()));
1603
1604                                 // Transfer the data from local storage to main storage (Grid)
1605                                 if (StoreOCDB() == kFALSE)
1606                                         success = kFALSE;
1607                         } 
1608                         else if (success == 0)
1609                         {
1610                                 Log("SHUTTLE", 
1611                                         Form("\t\t\t****** run %d - %s: ERROR ******",
1612                                                 GetCurrentRun(), aDetector->GetName()));
1613                         }
1614
1615                         for (UInt_t iSys=0; iSys<3; iSys++)
1616                         {
1617                                 if (fFXSCalled[iSys]) fFXSlist[iSys].Clear();
1618                         }
1619
1620                         Log("SHUTTLE", Form("Process - Client process of %d - %s is exiting now with %d.",
1621                                                         GetCurrentRun(), aDetector->GetName(), success));
1622
1623                         // the client exits here
1624                         gSystem->Exit(success);
1625
1626                         AliError("We should never get here!!!");
1627                 }
1628         }
1629
1630         Log("SHUTTLE", Form("\t\t\t^*^*^*^*^*^*^*^*^*^*^*^* run %d: FINISH ^*^*^*^*^*^*^*^*^*^*^*^*",
1631                                                         GetCurrentRun()));
1632
1633         //check if shuttle is done for this run, if so update logbook
1634         TObjArray checkEntryArray;
1635         checkEntryArray.SetOwner(1);
1636         TString whereClause = Form("where run=%d", GetCurrentRun());
1637         if (!QueryShuttleLogbook(whereClause.Data(), checkEntryArray) || 
1638                         checkEntryArray.GetEntries() == 0) {
1639                 Log("SHUTTLE", Form("Process - Warning: Cannot check status of run %d on Shuttle logbook!",
1640                                                 GetCurrentRun()));
1641                 return hasError == kFALSE;
1642         }
1643
1644         AliShuttleLogbookEntry* checkEntry = dynamic_cast<AliShuttleLogbookEntry*>
1645                                                 (checkEntryArray.At(0));
1646
1647         if (checkEntry)
1648         {
1649                 if (checkEntry->IsDone())
1650                 {
1651                         Log("SHUTTLE","Process - Shuttle is DONE. Updating logbook");
1652                         UpdateShuttleLogbook("shuttle_done");
1653                 }
1654                 else
1655                 {
1656                         for (UInt_t iDet=0; iDet<NDetectors(); iDet++)
1657                         {
1658                                 if (checkEntry->GetDetectorStatus(iDet) == AliShuttleLogbookEntry::kUnprocessed)
1659                                 {
1660                                         AliDebug(2, Form("Run %d: setting %s as \"not first time unprocessed\"",
1661                                                         checkEntry->GetRun(), GetDetName(iDet)));
1662                                         fFirstUnprocessed[iDet] = kFALSE;
1663                                 }
1664                         }
1665                         SendMLRunInfo("Pending");
1666                 }
1667         }
1668
1669         fLogbookEntry = 0;
1670
1671         return hasError == kFALSE;
1672 }
1673
1674 //______________________________________________________________________________________________
1675 Int_t AliShuttle::ProcessCurrentDetector()
1676 {
1677         //
1678         // Makes data retrieval just for a specific detector (fCurrentDetector).
1679         // Threre should be a configuration for this detector.
1680
1681         Log("SHUTTLE", Form("ProcessCurrentDetector - Retrieving values for %s, run %d", 
1682                                                 fCurrentDetector.Data(), GetCurrentRun()));
1683
1684         TString wd = gSystem->WorkingDirectory();
1685         
1686         if (!CleanReferenceStorage(fCurrentDetector.Data()))
1687                 return 0;
1688         
1689         gSystem->ChangeDirectory(wd.Data());
1690         
1691         // call preprocessor
1692         AliPreprocessor* aPreprocessor =
1693                 dynamic_cast<AliPreprocessor*> (fPreprocessorMap.GetValue(fCurrentDetector));
1694
1695         // check if the preprocessor wants to process this run type
1696         if (aPreprocessor->ProcessRunType() == kFALSE)
1697         {
1698                 UpdateShuttleStatus(AliShuttleStatus::kSkipped);
1699                 UpdateShuttleLogbook(fCurrentDetector, "DONE");
1700                 if (!UpdateTableSkippedCase(fCurrentDetector.Data()))
1701                 {
1702                         AliError(Form("Could not update FXS tables for run %d !", GetCurrentRun()));
1703                 }
1704                 Log(fCurrentDetector, Form("ProcessCurrentDetector - %s preprocessor is not interested in this run type", fCurrentDetector.Data()));
1705         
1706                 return 2;
1707         }
1708         
1709         // checking if OCDB is reachable
1710         AliCDBEntry* testEntry = GetFromOCDB("SHUTTLE","GRP/CTP/DummyConfig");
1711         if (!testEntry){
1712                 // OCDB is not accessible, going in OCDBError for current detector
1713                 AliError("OCDB Test entry not accessible");
1714                 UpdateShuttleStatus(AliShuttleStatus::kOCDBError);
1715                 return 0;
1716         }
1717
1718         TMap* dcsMap = new TMap();
1719         
1720         aPreprocessor->Initialize(GetCurrentRun(), GetCurrentStartTime(), GetCurrentEndTime());
1721
1722         Bool_t processDCS = aPreprocessor->ProcessDCS();
1723
1724         if (!processDCS)
1725         {
1726                 Log(fCurrentDetector, "ProcessCurrentDetector -"
1727                         " The preprocessor requested to skip the retrieval of DCS values");
1728         }
1729         else if (fTestMode & kSkipDCS)
1730         {
1731                 Log(fCurrentDetector, "ProcessCurrentDetector - In TESTMODE: Skipping DCS processing");
1732         } 
1733         else if (fTestMode & kErrorDCS)
1734         {
1735                 Log(fCurrentDetector, "ProcessCurrentDetector - In TESTMODE: Simulating DCS error");
1736                 UpdateShuttleStatus(AliShuttleStatus::kDCSStarted);
1737                 UpdateShuttleStatus(AliShuttleStatus::kDCSError);
1738                 delete dcsMap;
1739                 return 0;
1740         } else {
1741
1742                 UpdateShuttleStatus(AliShuttleStatus::kDCSStarted);
1743
1744                 // Query DCS archive
1745                 Int_t nServers = fConfig->GetNServers(fCurrentDetector);
1746                 
1747                 for (int iServ=0; iServ<nServers; iServ++)
1748                 {
1749                 
1750                         TString host(fConfig->GetDCSHost(fCurrentDetector, iServ));
1751                         Int_t port = fConfig->GetDCSPort(fCurrentDetector, iServ);
1752                         Int_t multiSplit = fConfig->GetMultiSplit(fCurrentDetector, iServ);
1753
1754                         Log(fCurrentDetector, Form("ProcessCurrentDetector -"
1755                                         " Querying DCS Amanda server %s:%d (%d of %d)", 
1756                                         host.Data(), port, iServ+1, nServers));
1757                         
1758                         TMap* aliasMap = 0;
1759                         TMap* dpMap = 0;
1760
1761                         if (fConfig->GetDCSAliases(fCurrentDetector, iServ)->GetEntries() > 0)
1762                         {
1763                                 Log(fCurrentDetector, Form("Querying %d DCS aliases", fConfig->GetDCSAliases(fCurrentDetector, iServ)->GetEntries()));
1764                                 aliasMap = GetValueSet(host, port, 
1765                                                 fConfig->GetDCSAliases(fCurrentDetector, iServ), 
1766                                                 kAlias, multiSplit);
1767                                 if (!aliasMap)
1768                                 {
1769                                         Log(fCurrentDetector, 
1770                                                 Form("ProcessCurrentDetector -"
1771                                                         " Error retrieving DCS aliases from server %s."
1772                                                         " Sending mail to DCS experts!", host.Data()));
1773                                         UpdateShuttleStatus(AliShuttleStatus::kDCSError);
1774                                         
1775                                         if (!SendMail(kDCSEMail))
1776                                                 Log("SHUTTLE", Form("ProcessCurrentDetector - "
1777                                                                     "Could not send mail to DCS experts!"));
1778
1779                                         delete dcsMap;
1780                                         return 0;
1781                                 }
1782                         }
1783                         
1784                         if (fConfig->GetDCSDataPoints(fCurrentDetector, iServ)->GetEntries() > 0)
1785                         {
1786                                 Log(fCurrentDetector, Form("Querying %d DCS data points", fConfig->GetDCSDataPoints(fCurrentDetector, iServ)->GetEntries()));
1787                                 dpMap = GetValueSet(host, port, 
1788                                                 fConfig->GetDCSDataPoints(fCurrentDetector, iServ), 
1789                                                 kDP, multiSplit);
1790                                 if (!dpMap)
1791                                 {
1792                                         Log(fCurrentDetector, 
1793                                                 Form("ProcessCurrentDetector -"
1794                                                         " Error retrieving DCS data points from server %s."
1795                                                         " Sending mail to DCS experts!", host.Data()));
1796                                         UpdateShuttleStatus(AliShuttleStatus::kDCSError);
1797                                         
1798                                         if (!SendMail(kDCSEMail))
1799                                                 Log("SHUTTLE", Form("ProcessCurrentDetector - "
1800                                                                     "Could not send mail to DCS experts!"));
1801                                         
1802                                         if (aliasMap) delete aliasMap;
1803                                         delete dcsMap;
1804                                         return 0;
1805                                 }                               
1806                         }
1807                         
1808                         // merge aliasMap and dpMap into dcsMap
1809                         if(aliasMap) {
1810                                 TIter iter(aliasMap);
1811                                 TObjString* key = 0;
1812                                 while ((key = (TObjString*) iter.Next()))
1813                                         dcsMap->Add(key, aliasMap->GetValue(key->String()));
1814                                 
1815                                 aliasMap->SetOwner(kFALSE);
1816                                 delete aliasMap;
1817                         }       
1818                         
1819                         if(dpMap) {
1820                                 TIter iter(dpMap);
1821                                 TObjString* key = 0;
1822                                 while ((key = (TObjString*) iter.Next()))
1823                                         dcsMap->Add(key, dpMap->GetValue(key->String()));
1824                                 
1825                                 dpMap->SetOwner(kFALSE);
1826                                 delete dpMap;
1827                         }
1828                 }
1829         }
1830         
1831         // save map into file, to help debugging in case of preprocessor error
1832         TFile* f = TFile::Open("DCSMap.root","recreate");
1833         f->cd();
1834         dcsMap->Write("DCSMap", TObject::kSingleKey);
1835         f->Close();
1836         delete f;
1837         
1838         // DCS Archive DB processing successful. Call Preprocessor!
1839         UpdateShuttleStatus(AliShuttleStatus::kPPStarted);
1840
1841         fFXSError = -1; // this variable is kTRUE after ::Process if an FXS error occured
1842         
1843         UInt_t returnValue = aPreprocessor->Process(dcsMap);
1844         
1845         if (fFXSError!=-1) {
1846                 UpdateShuttleStatus(AliShuttleStatus::kFXSError);
1847                 SendMail(kFXSEMail, fFXSError);
1848                 dcsMap->DeleteAll();
1849                 delete dcsMap;
1850                 return 0;
1851         }
1852
1853         if (returnValue > 0) // Preprocessor error!
1854         {
1855                 Log(fCurrentDetector, Form("ProcessCurrentDetector - "
1856                                 "Preprocessor failed. Process returned %d.", returnValue));
1857                 UpdateShuttleStatus(AliShuttleStatus::kPPError);
1858                 dcsMap->DeleteAll();
1859                 delete dcsMap;
1860                 return 0;
1861         }
1862         
1863         // preprocessor ok!
1864         UpdateShuttleStatus(AliShuttleStatus::kPPDone);
1865         Log(fCurrentDetector, Form("ProcessCurrentDetector - %s preprocessor returned success",
1866                                 fCurrentDetector.Data()));
1867
1868         dcsMap->DeleteAll();
1869         delete dcsMap;
1870
1871         return 1;
1872 }
1873
1874 //______________________________________________________________________________________________
1875 void AliShuttle::CountOpenRuns()
1876 {
1877         // Query DAQ's Shuttle logbook and sends the number of open runs to ML
1878         
1879         SendAlive();
1880         
1881         // check connection, in case connect
1882         if (!Connect(3)) 
1883                 return;
1884
1885         TString sqlQuery;
1886         sqlQuery = Form("select count(*) from %s where shuttle_done=0", fConfig->GetShuttlelbTable());
1887         
1888         TSQLResult* aResult = fServer[3]->Query(sqlQuery);
1889         if (!aResult) {
1890                 AliError(Form("Can't execute query <%s>!", sqlQuery.Data()));
1891                 return;
1892         }
1893
1894         AliDebug(2,Form("Query = %s", sqlQuery.Data()));
1895         
1896         if (aResult->GetRowCount() == 0) {
1897                 AliError(Form("No result for query %s received", sqlQuery.Data()));
1898                 return;
1899         }
1900
1901         if (aResult->GetFieldCount() != 1) {
1902                 AliError(Form("Invalid field count for query %s received", sqlQuery.Data()));
1903                 return;
1904         }
1905
1906         TSQLRow* aRow = aResult->Next();
1907         if (!aRow) {
1908                 AliError(Form("Could not receive result of query %s", sqlQuery.Data()));
1909                 return;
1910         }
1911         
1912         TString result(aRow->GetField(0), aRow->GetFieldLength(0));
1913         Int_t count = result.Atoi();
1914         
1915         Log("SHUTTLE", Form("%d unprocessed runs", count));
1916         
1917         delete aRow;
1918         delete aResult;
1919
1920         TMonaLisaValue mlStatus("SHUTTLE_openruns", count);
1921
1922         TList mlList;
1923         mlList.Add(&mlStatus);
1924
1925         fMonaLisa->SendParameters(&mlList, "__PROCESSINGINFO__");
1926 }
1927
1928 //______________________________________________________________________________________________
1929 Bool_t AliShuttle::QueryShuttleLogbook(const char* whereClause,
1930                 TObjArray& entries)
1931 {
1932         // Query DAQ's Shuttle logbook and fills detector status object.
1933         // Call QueryRunParameters to query DAQ logbook for run parameters.
1934         //
1935
1936         entries.SetOwner(1);
1937
1938         // check connection, in case connect
1939         if (!Connect(3)) return kFALSE;
1940
1941         TString sqlQuery;
1942         sqlQuery = Form("select * from %s %s order by run", fConfig->GetShuttlelbTable(), whereClause);
1943
1944         TSQLResult* aResult = fServer[3]->Query(sqlQuery);
1945         if (!aResult) {
1946                 AliError(Form("Can't execute query <%s>!", sqlQuery.Data()));
1947                 return kFALSE;
1948         }
1949
1950         AliDebug(2,Form("Query = %s", sqlQuery.Data()));
1951
1952         if(aResult->GetRowCount() == 0) {
1953                 Log("SHUTTLE", "No entries in Shuttle Logbook match request");
1954                 delete aResult;
1955                 return kTRUE;
1956         }
1957
1958         // TODO Check field count!
1959         const UInt_t nCols = 26;
1960         if (aResult->GetFieldCount() != (Int_t) nCols) {
1961                 Log("SHUTTLE", "Invalid SQL result field number!");
1962                 delete aResult;
1963                 return kFALSE;
1964         }
1965
1966         TSQLRow* aRow;
1967         while ((aRow = aResult->Next())) {
1968                 TString runString(aRow->GetField(0), aRow->GetFieldLength(0));
1969                 Int_t run = runString.Atoi();
1970
1971                 AliShuttleLogbookEntry *entry = QueryRunParameters(run);
1972                 if (!entry)
1973                         continue;
1974
1975                 // DA test mode flag
1976                 TString daTestModeString(aRow->GetField(2), aRow->GetFieldLength(2)); // field 2 = DA test mode flag 
1977                 Bool_t daTestMode = (Bool_t)daTestModeString.Atoi();
1978                 entry->SetDATestMode(daTestMode);
1979
1980                 // loop on detectors
1981                 for(UInt_t ii = 0; ii < nCols; ii++)
1982                         entry->SetDetectorStatus(aResult->GetFieldName(ii), aRow->GetField(ii));
1983
1984                 entries.AddLast(entry);
1985                 delete aRow;
1986         }
1987
1988         delete aResult;
1989         return kTRUE;
1990 }
1991
1992 //______________________________________________________________________________________________
1993 AliShuttleLogbookEntry* AliShuttle::QueryRunParameters(Int_t run)
1994 {
1995         //
1996         // Retrieve run parameters written in the DAQ logbook and sets them into AliShuttleLogbookEntry object
1997         //
1998
1999         // check connection, in case connect
2000         if (!Connect(3))
2001                 return 0;
2002
2003         TString sqlQuery;
2004         sqlQuery.Form("select * from %s where run=%d", fConfig->GetDAQlbTable(), run);
2005
2006         TSQLResult* aResult = fServer[3]->Query(sqlQuery);
2007         if (!aResult) {
2008                 Log("SHUTTLE", Form("Can't execute query <%s>!", sqlQuery.Data()));
2009                 return 0;
2010         }
2011
2012         if (aResult->GetRowCount() == 0) {
2013                 Log("SHUTTLE", Form("QueryRunParameters - No entry in DAQ Logbook for run %d. Skipping", run));
2014                 delete aResult;
2015                 return 0;
2016         }
2017
2018         if (aResult->GetRowCount() > 1) {
2019                 Log("SHUTTLE", Form("QueryRunParameters - UNEXPECTED: "
2020                                 "more than one entry in DAQ Logbook for run %d!", run));
2021                 delete aResult;
2022                 return 0;
2023         }
2024
2025         TSQLRow* aRow = aResult->Next();
2026         if (!aRow)
2027         {
2028                 Log("SHUTTLE", Form("QueryRunParameters - Could not retrieve row for run %d. Skipping", run));
2029                 delete aResult;
2030                 return 0;
2031         }
2032
2033         AliShuttleLogbookEntry* entry = new AliShuttleLogbookEntry(run);
2034
2035         for (Int_t ii = 0; ii < aResult->GetFieldCount(); ii++)
2036                 entry->SetRunParameter(aResult->GetFieldName(ii), aRow->GetField(ii));
2037
2038         delete aRow;
2039         delete aResult;
2040         
2041         UInt_t startTime = entry->GetStartTime();
2042         UInt_t endTime = entry->GetEndTime();
2043         Bool_t ecsSuccess = entry->GetECSSuccess();
2044         TString runType = entry->GetRunType();
2045         TString tmpdaqstartTime = entry->GetRunParameter("DAQ_time_start");
2046         UInt_t daqstartTime = tmpdaqstartTime.Atoi();
2047         
2048         UInt_t now = time(0);
2049         Int_t dcsDelay = fConfig->GetDCSDelay()+fConfig->GetDCSQueryOffset();
2050
2051         Bool_t skip = kFALSE;
2052                                 
2053         // runs are processed if
2054         //   a) runType is PHYSICS and ecsSuccess is set
2055         //   b) runType is not PHYSICS and (ecsSuccess is set or DAQ_time_start is non-0)
2056         // effectively this means that all runs are processed that started properly (ecsSucess behaviour is different for PHYSICS and non-PHYSICS runs (check with ECS!)
2057         if (startTime != 0 && endTime != 0) {
2058                 if (endTime > startTime) { 
2059                         if (endTime >= now - dcsDelay) {
2060                                 Log("SHUTTLE", Form("Skipping run %d for now, because DCS buffer time is not yet expired", run));
2061                         } else {
2062                                 if (runType == "PHYSICS") {
2063                                         if (ecsSuccess) {
2064                                                 return entry;
2065                                         } else {
2066                                                 Log("SHUTTLE", Form("QueryRunParameters - Run type for run %d is PHYSICS but ECS success flag not set (Reason = %s) - Skipping!", run, entry->GetRunParameter("eor_reason")));
2067                                                 skip = kTRUE;
2068                                         } 
2069                                 } else {
2070                                         if (ecsSuccess || daqstartTime > 0) {
2071                                                 if (ecsSuccess == kFALSE)
2072                                                         Log("SHUTTLE", Form("Processing run %d although in status ECS failure (Reason: %s), since run type != PHYSICS and DAQ_time_start != 0", run, entry->GetRunParameter("eor_reason")));
2073                                                 return entry;
2074                                         } else {
2075                                                 Log("SHUTTLE", Form("QueryRunParameters - Run type for run %d is %s, ECS success flag was not set (Reason = %s) and DAQ_time_start was NULL - Skipping!", run, runType.Data(), entry->GetRunParameter("eor_reason")));
2076                                                 skip = kTRUE;
2077                                         }
2078                                 }
2079                         }
2080                 } else {
2081                         Log("SHUTTLE", Form("QueryRunParameters - Invalid parameters for run %d: startTime equal to endTime: %d %d - Skipping!", run, startTime, endTime));
2082                         skip = kTRUE;
2083                 }
2084         } else {
2085           Log("SHUTTLE", Form("QueryRunParameters - Invalid parameters for Run %d: "
2086                               "startTime = %d, endTime = %d. Skipping (Shuttle won't be marked as DONE)!",
2087                               run, startTime, endTime));
2088         }
2089         
2090         if (skip)
2091         {
2092                 Log("SHUTTLE", Form("Marking SHUTTLE skipped for run %d", run));
2093                 fLogbookEntry = entry;
2094                 if (!UpdateShuttleLogbook("shuttle_skipped"))
2095                 {
2096                         AliError(Form("Could not update logbook for run %d !", run));
2097                 }
2098                 if (!UpdateTableSkippedCase("ALL"))
2099                 {
2100                         AliError(Form("Could not update FXS tables for run %d !", run));
2101                 }
2102                 fLogbookEntry = 0;
2103         }
2104                         
2105         delete entry;
2106         return 0;
2107 }
2108
2109 //______________________________________________________________________________________________
2110 TMap* AliShuttle::GetValueSet(const char* host, Int_t port, const TSeqCollection* entries,
2111                               DCSType type, Int_t multiSplit)
2112 {
2113         // Retrieve all "entry" data points from the DCS server
2114         // host, port: TSocket connection parameters
2115         // entries: list of name of the alias or data point
2116         // type: kAlias or kDP
2117         // returns TMap of values, 0 when failure
2118         
2119         AliDCSClient client(host, port, fTimeout, fRetries, multiSplit);
2120
2121         TMap* result = 0;
2122         if (type == kAlias)
2123         {
2124                 //result = client.GetAliasValues(entries, GetCurrentStartTime()-offset, 
2125                 //      GetCurrentEndTime()+offset);
2126                 result = client.GetAliasValues(entries, GetStartTimeDCSQuery(), 
2127                         GetEndTimeDCSQuery());
2128         } 
2129         else if (type == kDP)
2130         {
2131                 //result = client.GetDPValues(entries, GetCurrentStartTime()-offset, 
2132                 //      GetCurrentEndTime()+offset);
2133                 result = client.GetDPValues(entries, GetStartTimeDCSQuery(), 
2134                         GetEndTimeDCSQuery());
2135         }
2136
2137         if (result == 0)
2138         {
2139                 Log(fCurrentDetector.Data(), Form("GetValueSet - Can't get entries! Reason: %s",
2140                         client.GetErrorString(client.GetResultErrorCode())));
2141                 if (client.GetResultErrorCode() == AliDCSClient::fgkServerError)        
2142                         Log(fCurrentDetector.Data(), Form("GetValueSet - Server error code: %s",
2143                                 client.GetServerError().Data()));
2144
2145                 return 0;
2146         }
2147                 
2148         return result;
2149 }
2150
2151 //______________________________________________________________________________________________
2152 const char* AliShuttle::GetFile(Int_t system, const char* detector,
2153                 const char* id, const char* source)
2154 {
2155         // Get calibration file from file exchange servers
2156         // First queris the FXS database for the file name, using the run, detector, id and source info
2157         // then calls RetrieveFile(filename) for actual copy to local disk
2158         // run: current run being processed (given by Logbook entry fLogbookEntry)
2159         // detector: the Preprocessor name
2160         // id: provided as a parameter by the Preprocessor
2161         // source: provided by the Preprocessor through GetFileSources function
2162
2163         // check if test mode should simulate a FXS error
2164         if (fTestMode & kErrorFXSFiles)
2165         {
2166                 Log(detector, Form("GetFile - In TESTMODE - Simulating error while connecting to %s FXS", GetSystemName(system)));
2167                 return 0;
2168         }
2169         
2170         // check connection, in case connect
2171         if (!Connect(system))
2172         {
2173                 Log(detector, Form("GetFile - Couldn't connect to %s FXS database", GetSystemName(system)));
2174                 fFXSError = system;
2175                 return 0;
2176         }
2177
2178         // Query preparation
2179         TString sourceName(source);
2180         Int_t nFields = 3;
2181         TString sqlQueryStart = Form("select filePath,size,fileChecksum from %s where",
2182                                                                 fConfig->GetFXSdbTable(system));
2183         TString whereClause = Form("run=%d and detector=\"%s\" and fileId=\"%s\"",
2184                                                                 GetCurrentRun(), detector, id);
2185
2186         if (system == kDAQ)
2187         {
2188                 whereClause += Form(" and DAQsource=\"%s\"", source);
2189         }
2190         else if (system == kDCS)
2191         {
2192                 sourceName="none";
2193         }
2194         else if (system == kHLT)
2195         {
2196                 whereClause += Form(" and DDLnumbers=\"%s\"", source);
2197         }
2198
2199         TString sqlQuery = Form("%s %s", sqlQueryStart.Data(), whereClause.Data());
2200
2201         AliDebug(2, Form("SQL query: \n%s",sqlQuery.Data()));
2202
2203         // Query execution
2204         TSQLResult* aResult = 0;
2205         aResult = dynamic_cast<TSQLResult*> (fServer[system]->Query(sqlQuery));
2206         if (!aResult) {
2207                 Log(detector, Form("GetFile - Can't execute SQL query to %s database for: id = %s, source = %s",
2208                                 GetSystemName(system), id, sourceName.Data()));
2209                 fFXSError = system;
2210                 return 0;
2211         }
2212
2213         if (aResult->GetRowCount() == 0)
2214         {
2215                 Log(detector,
2216                         Form("GetFile - No entry in %s FXS db for: id = %s, source = %s",
2217                                 GetSystemName(system), id, sourceName.Data()));
2218                 delete aResult;
2219                 return 0;
2220         }
2221
2222         if (aResult->GetRowCount() > 1) {
2223                 Log(detector,
2224                         Form("GetFile - More than one entry in %s FXS db for: id = %s, source = %s",
2225                                 GetSystemName(system), id, sourceName.Data()));
2226                 fFXSError = system;
2227                 delete aResult;
2228                 return 0;
2229         }
2230
2231         if (aResult->GetFieldCount() != nFields) {
2232                 Log(detector,
2233                         Form("GetFileName - Wrong field count in %s FXS db for: id = %s, source = %s",
2234                                 GetSystemName(system), id, sourceName.Data()));
2235                 fFXSError = system;
2236                 delete aResult;
2237                 return 0;
2238         }
2239
2240         TSQLRow* aRow = dynamic_cast<TSQLRow*> (aResult->Next());
2241
2242         if (!aRow){
2243                 Log(detector, Form("GetFile - Empty set result in %s FXS db from query: id = %s, source = %s",
2244                                 GetSystemName(system), id, sourceName.Data()));
2245                 fFXSError = system;
2246                 delete aResult;
2247                 return 0;
2248         }
2249
2250         TString filePath(aRow->GetField(0), aRow->GetFieldLength(0));
2251         TString fileSize(aRow->GetField(1), aRow->GetFieldLength(1));
2252         TString fileChecksum(aRow->GetField(2), aRow->GetFieldLength(2));
2253
2254         delete aResult;
2255         delete aRow;
2256
2257         AliDebug(2, Form("filePath = %s; size = %s, fileChecksum = %s",
2258                                 filePath.Data(), fileSize.Data(), fileChecksum.Data()));
2259
2260         // retrieved file is renamed to make it unique
2261         Int_t dir_lev1 = GetCurrentRun()/10000;
2262         TString localFileName = Form("%s/%d/%d/%s_process/%s_%s_%d_%s_%s.shuttle",
2263                                      GetShuttleTempDir(), dir_lev1, GetCurrentRun(), detector,
2264                                         GetSystemName(system), detector, GetCurrentRun(), 
2265                                         id, sourceName.Data());
2266         Log("SHUTTLE",Form("file from FXS = %s",localFileName.Data())); 
2267
2268
2269         // file retrieval from FXS
2270         UInt_t nRetries = 0;
2271         UInt_t maxRetries = 3;
2272         Bool_t result = kFALSE;
2273
2274         // copy!! if successful TSystem::Exec returns 0
2275         while (nRetries++ < maxRetries) {
2276                 AliDebug(2, Form("Trying to copy file. Retry # %d", nRetries));
2277                 result = RetrieveFile(system, filePath.Data(), localFileName.Data());
2278                 if (!result)
2279                 {
2280                         Log(detector, Form("GetFile - Copy of file %s from %s FXS failed",
2281                                         filePath.Data(), GetSystemName(system)));
2282                         continue;
2283                 } 
2284
2285                 if (fileSize.Length()>0)
2286                 {
2287                         // compare filesize of local file with the one stored in the FXS DB
2288                         Long_t size = -1;
2289                         Int_t sizeComp = gSystem->GetPathInfo(localFileName.Data(), 0, &size, 0, 0);
2290
2291                         if (sizeComp != 0 || size != fileSize.Atoi())
2292                         {
2293                                 Log(detector, Form("GetFile - size of file %s does not match with local copy!",
2294                                                         filePath.Data()));
2295                                 result = kFALSE;
2296                                 continue;
2297                         }
2298
2299                 } else {
2300                         Log(fCurrentDetector, Form("GetFile - size of file %s not set in %s database, skipping comparison",
2301                                                 filePath.Data(), GetSystemName(system)));
2302                 }
2303
2304                 if (fileChecksum.Length()>0)
2305                 {
2306                         // compare md5sum of local file with the one stored in the FXS DB
2307                         if(fileChecksum.Contains(' ')) fileChecksum.Resize(fileChecksum.First(' '));
2308                         Int_t md5Comp = gSystem->Exec(Form("md5sum %s |grep %s > /dev/null 2> /dev/null",
2309                                                 localFileName.Data(), fileChecksum.Data()));
2310
2311                         if (md5Comp != 0)
2312                         {
2313                                 Log(detector, Form("GetFile - md5sum of file %s does not match with local copy!",
2314                                                         filePath.Data()));
2315                                 result = kFALSE;
2316                                 continue;
2317                         }
2318                 } else {
2319                         Log(fCurrentDetector, Form("GetFile - md5sum of file %s not set in %s database, skipping comparison",
2320                                                         filePath.Data(), GetSystemName(system)));
2321                 }
2322                 if (result) break;
2323         }
2324
2325         if (!result) 
2326         {
2327                 fFXSError = system;
2328                 return 0;
2329         }
2330
2331         fFXSCalled[system]=kTRUE;
2332         TObjString *fileParams = new TObjString(Form("%s#!?!#%s", id, sourceName.Data()));
2333         fFXSlist[system].Add(fileParams);
2334
2335         static TString staticLocalFileName;
2336         staticLocalFileName.Form("%s", localFileName.Data());
2337         
2338         Log(fCurrentDetector, Form("GetFile - Retrieved file with id %s and "
2339                         "source %s from %s to %s", id, source, 
2340                         GetSystemName(system), localFileName.Data()));
2341                         
2342         return staticLocalFileName.Data();
2343 }
2344
2345 //______________________________________________________________________________________________
2346 Bool_t AliShuttle::RetrieveFile(UInt_t system, const char* fxsFileName, const char* localFileName)
2347 {
2348         //
2349         // Copies file from FXS to local Shuttle machine
2350         //
2351
2352         // check temp directory: trying to cd to temp; if it does not exist, create it
2353         AliDebug(2, Form("Copy file %s from %s FXS into %s",
2354                         GetSystemName(system), fxsFileName, localFileName));
2355                         
2356         TString tmpDir(localFileName);
2357         
2358         tmpDir = tmpDir(0,tmpDir.Last('/'));
2359
2360         Int_t noDir = gSystem->GetPathInfo(tmpDir.Data(), 0, (Long64_t*) 0, 0, 0);
2361         if (noDir) // temp dir does not exists!
2362         {
2363                 if (gSystem->mkdir(tmpDir.Data(), 1))
2364                 {
2365                         Log(fCurrentDetector.Data(), "RetrieveFile - could not make temp directory!!");
2366                         return kFALSE;
2367                 }
2368         }
2369
2370         TString command = Form("scp -oPort=%d -2 %s@%s:%s/%s %s",
2371                 fConfig->GetFXSPort(system),
2372                 fConfig->GetFXSUser(system),
2373                 fConfig->GetFXSHost(system),
2374                 fConfig->GetFXSBaseFolder(system),
2375                 fxsFileName,
2376                 localFileName);
2377
2378         AliDebug(2, Form("%s",command.Data()));
2379
2380         Bool_t result = (gSystem->Exec(command.Data()) == 0);
2381
2382         return result;
2383 }
2384
2385 //______________________________________________________________________________________________
2386 TList* AliShuttle::GetFileSources(Int_t system, const char* detector, const char* id)
2387 {
2388         //
2389         // Get sources producing the condition file Id from file exchange servers
2390         // if id is NULL all sources are returned (distinct)
2391         //
2392
2393         if (id)
2394         {
2395                 Log(detector, Form("GetFileSources - Querying %s FXS for files with id %s produced by %s", GetSystemName(system), id, detector));
2396         } else {
2397                 Log(detector, Form("GetFileSources - Querying %s FXS for files produced by %s", GetSystemName(system), detector));
2398         }
2399         
2400         // check if test mode should simulate a FXS error
2401         if (fTestMode & kErrorFXSSources)
2402         {
2403                 Log(detector, Form("GetFileSources - In TESTMODE - Simulating error while connecting to %s FXS", GetSystemName(system)));
2404                 return 0;
2405         }
2406
2407         if (system == kDCS)
2408         {
2409                 Log(detector, "GetFileSources - WARNING: DCS system has only one source of data!");
2410                 TList *list = new TList();
2411                 list->SetOwner(1);
2412                 list->Add(new TObjString(" "));
2413                 return list;
2414         }
2415
2416         // check connection, in case connect
2417         if (!Connect(system))
2418         {
2419                 Log(detector, Form("GetFileSources - Couldn't connect to %s FXS database", GetSystemName(system)));
2420                 fFXSError = system;
2421                 return NULL;
2422         }
2423
2424         TString sourceName = "";
2425         if (system == kDAQ)
2426         {
2427                 sourceName = "DAQsource";
2428         } else if (system == kHLT)
2429         {
2430                 sourceName = "DDLnumbers";
2431         }
2432
2433         TString sqlQueryStart = Form("select distinct %s from %s where", sourceName.Data(), fConfig->GetFXSdbTable(system));
2434         TString whereClause = Form("run=%d and detector=\"%s\"",
2435                                 GetCurrentRun(), detector);
2436         if (id)
2437                 whereClause += Form(" and fileId=\"%s\"", id);
2438         TString sqlQuery = Form("%s %s", sqlQueryStart.Data(), whereClause.Data());
2439
2440         AliDebug(2, Form("SQL query: \n%s",sqlQuery.Data()));
2441
2442         // Query execution
2443         TSQLResult* aResult;
2444         aResult = fServer[system]->Query(sqlQuery);
2445         if (!aResult) {
2446                 Log(detector, Form("GetFileSources - Can't execute SQL query to %s database for id: %s",
2447                                 GetSystemName(system), id));
2448                 fFXSError = system;
2449                 return 0;
2450         }
2451
2452         TList *list = new TList();
2453         list->SetOwner(1);
2454         
2455         if (aResult->GetRowCount() == 0)
2456         {
2457                 Log(detector,
2458                         Form("GetFileSources - No entry in %s FXS table for id: %s", GetSystemName(system), id));
2459                 delete aResult;
2460                 return list;
2461         }
2462
2463         Log(detector, Form("GetFileSources - Found %d sources", aResult->GetRowCount()));
2464
2465         TSQLRow* aRow;
2466         while ((aRow = aResult->Next()))
2467         {
2468
2469                 TString source(aRow->GetField(0), aRow->GetFieldLength(0));
2470                 AliDebug(2, Form("%s = %s", sourceName.Data(), source.Data()));
2471                 list->Add(new TObjString(source));
2472                 delete aRow;
2473         }
2474
2475         delete aResult;
2476
2477         return list;
2478 }
2479
2480 //______________________________________________________________________________________________
2481 TList* AliShuttle::GetFileIDs(Int_t system, const char* detector, const char* source)
2482 {
2483         //
2484         // Get all ids of condition files produced by a given source from file exchange servers
2485         //
2486         
2487         Log(detector, Form("GetFileIDs - Retrieving ids with source %s with %s", source, GetSystemName(system)));
2488
2489         // check if test mode should simulate a FXS error
2490         if (fTestMode & kErrorFXSSources)
2491         {
2492                 Log(detector, Form("GetFileIDs - In TESTMODE - Simulating error while connecting to %s FXS", GetSystemName(system)));
2493                 return 0;
2494         }
2495
2496         // check connection, in case connect
2497         if (!Connect(system))
2498         {
2499                 Log(detector, Form("GetFileIDs - Couldn't connect to %s FXS database", GetSystemName(system)));
2500                 return NULL;
2501         }
2502
2503         TString sourceName = "";
2504         if (system == kDAQ)
2505         {
2506                 sourceName = "DAQsource";
2507         } else if (system == kHLT)
2508         {
2509                 sourceName = "DDLnumbers";
2510         }
2511
2512         TString sqlQueryStart = Form("select fileId from %s where", fConfig->GetFXSdbTable(system));
2513         TString whereClause = Form("run=%d and detector=\"%s\"",
2514                                 GetCurrentRun(), detector);
2515         if (sourceName.Length() > 0 && source)
2516                 whereClause += Form(" and %s=\"%s\"", sourceName.Data(), source);
2517         TString sqlQuery = Form("%s %s", sqlQueryStart.Data(), whereClause.Data());
2518
2519         AliDebug(2, Form("SQL query: \n%s",sqlQuery.Data()));
2520
2521         // Query execution
2522         TSQLResult* aResult;
2523         aResult = fServer[system]->Query(sqlQuery);
2524         if (!aResult) {
2525                 Log(detector, Form("GetFileIDs - Can't execute SQL query to %s database for source: %s",
2526                                 GetSystemName(system), source));
2527                 return 0;
2528         }
2529
2530         TList *list = new TList();
2531         list->SetOwner(1);
2532         
2533         if (aResult->GetRowCount() == 0)
2534         {
2535                 Log(detector,
2536                         Form("GetFileIDs - No entry in %s FXS table for source: %s", GetSystemName(system), source));
2537                 delete aResult;
2538                 return list;
2539         }
2540
2541         Log(detector, Form("GetFileIDs - Found %d ids", aResult->GetRowCount()));
2542
2543         TSQLRow* aRow;
2544
2545         while ((aRow = aResult->Next()))
2546         {
2547
2548                 TString id(aRow->GetField(0), aRow->GetFieldLength(0));
2549                 AliDebug(2, Form("fileId = %s", id.Data()));
2550                 list->Add(new TObjString(id));
2551                 delete aRow;
2552         }
2553
2554         delete aResult;
2555
2556         return list;
2557 }
2558
2559 //______________________________________________________________________________________________
2560 Bool_t AliShuttle::Connect(Int_t system)
2561 {
2562         // Connect to MySQL Server of the system's FXS MySQL databases
2563         // DAQ Logbook, Shuttle Logbook and DAQ FXS db are on the same host
2564         //
2565
2566         // check connection: if already connected return
2567
2568         if(fServer[system] && fServer[system]->IsConnected()) {
2569                 // ping the server              
2570                 if (fServer[system]->PingVerify()==kTRUE){ // connection is still alive
2571                         return kTRUE;
2572                 }               
2573                 else{
2574                         AliWarning(Form("Connection got lost to FXS database for %s. Closing and reconnecting.",
2575                                         AliShuttleInterface::GetSystemName(system)));
2576                         fServer[system]->Close();
2577                         delete fServer[system];
2578                         fServer[system] = 0x0;
2579                 }
2580         }
2581
2582         TString dbHost, dbUser, dbPass, dbName;
2583
2584         if (system < 3) // FXS db servers
2585         {
2586                 dbHost = Form("mysql://%s:%d", fConfig->GetFXSdbHost(system), fConfig->GetFXSdbPort(system));
2587                 dbUser = fConfig->GetFXSdbUser(system);
2588                 dbPass = fConfig->GetFXSdbPass(system);
2589                 dbName =   fConfig->GetFXSdbName(system);
2590         } else { // Run & Shuttle logbook servers
2591         // TODO Will the Shuttle logbook server be the same as the Run logbook server ???
2592                 dbHost = Form("mysql://%s:%d", fConfig->GetDAQlbHost(), fConfig->GetDAQlbPort());
2593                 dbUser = fConfig->GetDAQlbUser();
2594                 dbPass = fConfig->GetDAQlbPass();
2595                 dbName =   fConfig->GetDAQlbDB();
2596         }
2597
2598         fServer[system] = TSQLServer::Connect(dbHost.Data(), dbUser.Data(), dbPass.Data());
2599         if (!fServer[system] || !fServer[system]->IsConnected()) {
2600                 if(system < 3)
2601                 {
2602                 AliError(Form("Can't establish connection to FXS database for %s",
2603                                         AliShuttleInterface::GetSystemName(system)));
2604                 } else {
2605                 AliError("Can't establish connection to Run logbook.");
2606                 }
2607                 if(fServer[system]) delete fServer[system];
2608                 return kFALSE;
2609         }
2610
2611         // Get tables
2612         TSQLResult* aResult=0;
2613         switch(system){
2614                 case kDAQ:
2615                         aResult = fServer[kDAQ]->GetTables(dbName.Data());
2616                         break;
2617                 case kDCS:
2618                         aResult = fServer[kDCS]->GetTables(dbName.Data());
2619                         break;
2620                 case kHLT:
2621                         aResult = fServer[kHLT]->GetTables(dbName.Data());
2622                         break;
2623                 default:
2624                         aResult = fServer[3]->GetTables(dbName.Data());
2625                         break;
2626         }
2627
2628         delete aResult;
2629         return kTRUE;
2630 }
2631
2632 //______________________________________________________________________________________________
2633 Bool_t AliShuttle::UpdateTable()
2634 {
2635         //
2636         // Update FXS table filling time_processed field in all rows corresponding to current run and detector
2637         //
2638
2639         Bool_t result = kTRUE;
2640
2641         for (UInt_t system=0; system<3; system++)
2642         {
2643                 if(!fFXSCalled[system]) continue;
2644
2645                 // check connection, in case connect
2646                 if (!Connect(system))
2647                 {
2648                         Log(fCurrentDetector, Form("UpdateTable - Couldn't connect to %s FXS database", GetSystemName(system)));
2649                         result = kFALSE;
2650                         continue;
2651                 }
2652
2653                 TTimeStamp now; // now
2654
2655                 // Loop on FXS list entries
2656                 TIter iter(&fFXSlist[system]);
2657                 TObjString *aFXSentry=0;
2658                 while ((aFXSentry = dynamic_cast<TObjString*> (iter.Next())))
2659                 {
2660                         TString aFXSentrystr = aFXSentry->String();
2661                         TObjArray *aFXSarray = aFXSentrystr.Tokenize("#!?!#");
2662                         if (!aFXSarray || aFXSarray->GetEntries() != 2 )
2663                         {
2664                                 Log(fCurrentDetector, Form("UpdateTable - error updating %s FXS entry. Check string: <%s>",
2665                                         GetSystemName(system), aFXSentrystr.Data()));
2666                                 if(aFXSarray) delete aFXSarray;
2667                                 result = kFALSE;
2668                                 continue;
2669                         }
2670                         const char* fileId = ((TObjString*) aFXSarray->At(0))->GetName();
2671                         const char* source = ((TObjString*) aFXSarray->At(1))->GetName();
2672
2673                         TString whereClause;
2674                         if (system == kDAQ)
2675                         {
2676                                 whereClause = Form("where run=%d and detector=\"%s\" and fileId=\"%s\" and DAQsource=\"%s\";",
2677                                                         GetCurrentRun(), fCurrentDetector.Data(), fileId, source);
2678                         }
2679                         else if (system == kDCS)
2680                         {
2681                                 whereClause = Form("where run=%d and detector=\"%s\" and fileId=\"%s\";",
2682                                                         GetCurrentRun(), fCurrentDetector.Data(), fileId);
2683                         }
2684                         else if (system == kHLT)
2685                         {
2686                                 whereClause = Form("where run=%d and detector=\"%s\" and fileId=\"%s\" and DDLnumbers=\"%s\";",
2687                                                         GetCurrentRun(), fCurrentDetector.Data(), fileId, source);
2688                         }
2689
2690                         delete aFXSarray;
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("UpdateTable - %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
2711         return result;
2712 }
2713
2714 //_______________________________________________________________________________
2715 Bool_t AliShuttle::UpdateTableSkippedCase(const char* detector)
2716 {
2717         //
2718         // Update FXS table filling time_processed field in all rows corresponding to current run and detector
2719         // if detector = "ALL" update all detectors
2720         //
2721
2722         Bool_t result = kTRUE;
2723
2724         TString detName(detector);
2725
2726         for (UInt_t system=0; system<3; system++)
2727         {
2728
2729                 // check connection, in case connect
2730                 if (!Connect(system))
2731                 {
2732                         Log(fCurrentDetector, Form("UpdateTableSkippedCase - Couldn't connect to %s FXS database", GetSystemName(system)));
2733                         result = kFALSE;
2734                         continue;
2735                 }
2736
2737                 TTimeStamp now; // now
2738
2739                 // Loop on FXS list entries
2740                 TIter iter(&fFXSlist[system]);
2741                         
2742                 TString whereClause;
2743                 if (detName == "ALL") whereClause = Form("where run=%d and time_processed IS NULL;",GetCurrentRun());
2744                 else whereClause = Form("where run=%d and detector=\"%s\" and time_processed IS NULL;",GetCurrentRun(), detector);
2745
2746                 //Log("SHUTTLE",Form(" whereClause = %s ",whereClause.Data()));
2747
2748                 TString sqlQuery = Form("update %s set time_processed=%d %s", fConfig->GetFXSdbTable(system),
2749                                         now.GetSec(), whereClause.Data());
2750
2751                 AliDebug(2, Form("SQL query: \n%s",sqlQuery.Data()));
2752
2753                 // Query execution
2754                 TSQLResult* aResult;
2755                 aResult = dynamic_cast<TSQLResult*> (fServer[system]->Query(sqlQuery));
2756                 if (!aResult)
2757                 {
2758                         Log("SHUTTLE", Form("UpdateTableSkippedCase - %s db: can't execute SQL query <%s>",
2759                                                         GetSystemName(system), sqlQuery.Data()));
2760                         result = kFALSE;
2761                         continue;
2762                 }
2763                 delete aResult;
2764                 
2765         }
2766
2767         return result;
2768 }
2769 //______________________________________________________________________________________________
2770 Bool_t AliShuttle::UpdateTableFailCase()
2771 {
2772         // Update FXS table filling time_processed field in all rows corresponding to current run and detector
2773         // this is called in case the preprocessor is declared failed for the current run, because
2774         // the fields are updated only in case of success
2775
2776         Bool_t result = kTRUE;
2777
2778         for (UInt_t system=0; system<3; system++)
2779         {
2780                 // check connection, in case connect
2781                 if (!Connect(system))
2782                 {
2783                         Log(fCurrentDetector, Form("UpdateTableFailCase - Couldn't connect to %s FXS database",
2784                                                         GetSystemName(system)));
2785                         result = kFALSE;
2786                         continue;
2787                 }
2788
2789                 TTimeStamp now; // now
2790
2791                 // Loop on FXS list entries
2792
2793                 TString whereClause = Form("where run=%d and detector=\"%s\";",
2794                                                 GetCurrentRun(), fCurrentDetector.Data());
2795
2796
2797                 TString sqlQuery = Form("update %s set time_processed=%d %s", fConfig->GetFXSdbTable(system),
2798                                                         now.GetSec(), whereClause.Data());
2799
2800                 AliDebug(2, Form("SQL query: \n%s",sqlQuery.Data()));
2801
2802                 // Query execution
2803                 TSQLResult* aResult;
2804                 aResult = dynamic_cast<TSQLResult*> (fServer[system]->Query(sqlQuery));
2805                 if (!aResult)
2806                 {
2807                         Log(fCurrentDetector, Form("UpdateTableFailCase - %s db: can't execute SQL query <%s>",
2808                                                         GetSystemName(system), sqlQuery.Data()));
2809                         result = kFALSE;
2810                         continue;
2811                 }
2812                 delete aResult;
2813         }
2814
2815         return result;
2816 }
2817
2818 //______________________________________________________________________________________________
2819 Bool_t AliShuttle::UpdateShuttleLogbook(const char* detector, const char* status)
2820 {
2821         //
2822         // Update Shuttle logbook filling detector or shuttle_done column
2823         // ex. of usage: UpdateShuttleLogbook("PHOS", "DONE") or UpdateShuttleLogbook("shuttle_done")
2824         //
2825
2826         // check connection, in case connect
2827         if(!Connect(3)){
2828                 Log("SHUTTLE", "UpdateShuttleLogbook - Couldn't connect to DAQ Logbook.");
2829                 return kFALSE;
2830         }
2831
2832         TString detName(detector);
2833         TString setClause;
2834         if (detName == "shuttle_done" || detName == "shuttle_skipped")
2835         {
2836                 setClause = "set shuttle_done=1";
2837                 
2838                 if (detName == "shuttle_done")
2839                 {
2840                         if (TouchFile() != kTRUE)
2841                         {
2842                                 SendMLRunInfo("Pending");
2843                                 return kFALSE;
2844                         }
2845                         
2846                         SendMLRunInfo("Done");
2847                 }
2848                 else 
2849                         SendMLRunInfo("Skipped");
2850         } 
2851         else {
2852                 TString statusStr(status);
2853                 if(statusStr.Contains("done", TString::kIgnoreCase) ||
2854                    statusStr.Contains("failed", TString::kIgnoreCase)){
2855                         setClause = Form("set %s=\"%s\"", detector, status);
2856                 } else {
2857                         Log("SHUTTLE",
2858                                 Form("UpdateShuttleLogbook - Invalid status <%s> for detector %s",
2859                                         status, detector));
2860                         return kFALSE;
2861                 }
2862         }
2863
2864         TString whereClause = Form("where run=%d", GetCurrentRun());
2865
2866         TString sqlQuery = Form("update %s %s %s",
2867                                         fConfig->GetShuttlelbTable(), setClause.Data(), whereClause.Data());
2868
2869         AliDebug(2, Form("SQL query: \n%s",sqlQuery.Data()));
2870
2871         // Query execution
2872         TSQLResult* aResult;
2873         aResult = dynamic_cast<TSQLResult*> (fServer[3]->Query(sqlQuery));
2874         if (!aResult) {
2875                 Log("SHUTTLE", Form("UpdateShuttleLogbook - Can't execute query <%s>", sqlQuery.Data()));
2876                 return kFALSE;
2877         }
2878         delete aResult;
2879
2880         return kTRUE;
2881 }
2882
2883 //______________________________________________________________________________________________
2884 Int_t AliShuttle::GetCurrentRun() const
2885 {
2886         //
2887         // Get current run from logbook entry
2888         //
2889
2890         return fLogbookEntry ? fLogbookEntry->GetRun() : -1;
2891 }
2892
2893 //______________________________________________________________________________________________
2894 UInt_t AliShuttle::GetCurrentStartTime() const
2895 {
2896         //
2897         // get current start time
2898         //
2899
2900         return fLogbookEntry ? fLogbookEntry->GetStartTime() : 0;
2901 }
2902
2903 //______________________________________________________________________________________________
2904 UInt_t AliShuttle::GetCurrentEndTime() const
2905 {
2906         //
2907         // get current end time from logbook entry
2908         //
2909
2910         return fLogbookEntry ? fLogbookEntry->GetEndTime() : 0;
2911 }
2912 //______________________________________________________________________________________________
2913 UInt_t AliShuttle::GetCurrentYear() const
2914 {
2915         //
2916         // Get current year from logbook entry
2917         //
2918
2919         if (!fLogbookEntry) return 0;
2920         
2921         TTimeStamp startTime(GetCurrentStartTime());
2922         TString year =  Form("%d",startTime.GetDate());
2923         year = year(0,4);
2924         
2925         return year.Atoi();
2926 }
2927
2928 //______________________________________________________________________________________________
2929 const char* AliShuttle::GetLHCPeriod() const
2930 {
2931         //
2932         // Get current LHC period from logbook entry
2933         //
2934
2935         if (!fLogbookEntry) return 0;
2936                 
2937         return fLogbookEntry->GetRunParameter("LHCperiod");
2938 }
2939
2940 //______________________________________________________________________________________________
2941 void AliShuttle::Log(const char* detector, const char* message)
2942 {
2943         //
2944         // Fill log string with a message
2945         //
2946         
2947         TString logRunDir = GetShuttleLogDir();
2948         if (GetCurrentRun() >=0) {
2949                 Int_t logDir_lev1 = GetCurrentRun()/10000;
2950                 logRunDir += Form("/%d/%d", logDir_lev1, GetCurrentRun());
2951         }               
2952         void* dir = gSystem->OpenDirectory(logRunDir.Data());
2953         if (dir == NULL) {
2954                 if (gSystem->mkdir(logRunDir.Data(), kTRUE)) {
2955                         AliError(Form("Can't open directory <%s>", GetShuttleLogDir()));
2956                         return;
2957                 }
2958
2959         } else {
2960                 gSystem->FreeDirectory(dir);
2961         }
2962
2963         TString toLog = Form("%s UTC (%d): %s - ", TTimeStamp(time(0)).AsString("s"), getpid(), detector);
2964         if (GetCurrentRun() >= 0) 
2965                 toLog += Form("run %d - ", GetCurrentRun());
2966         toLog += Form("%s", message);
2967
2968         AliInfo(toLog.Data());
2969         
2970         // if we redirect the log output already to the file, leave here
2971         if (fOutputRedirected && strcmp(detector, "SHUTTLE") != 0)
2972                 return;
2973
2974         TString fileName = GetLogFileName(detector);
2975         
2976         gSystem->ExpandPathName(fileName);
2977
2978         ofstream logFile;
2979         logFile.open(fileName, ofstream::out | ofstream::app);
2980
2981         if (!logFile.is_open()) {
2982                 AliError(Form("Could not open file %s", fileName.Data()));
2983                 return;
2984         }
2985
2986         logFile << toLog.Data() << "\n";
2987
2988         logFile.close();
2989 }
2990
2991 //______________________________________________________________________________________________
2992 TString AliShuttle::GetLogFileName(const char* detector) const
2993 {
2994         // 
2995         // returns the name of the log file for a given sub detector
2996         //
2997         
2998         TString fileName;
2999         
3000         if (GetCurrentRun() >= 0) 
3001         {
3002                 Int_t logDir_lev1 = GetCurrentRun()/10000;
3003                 fileName.Form("%s/%d/%d/%s.log", GetShuttleLogDir(), logDir_lev1, GetCurrentRun(), 
3004                         detector);
3005         } else {
3006                 fileName.Form("%s/%s.log", GetShuttleLogDir(), detector);
3007         }
3008
3009         return fileName;
3010 }
3011
3012 //______________________________________________________________________________________________
3013 void AliShuttle::SendAlive()
3014 {
3015         // sends alive message to ML
3016         
3017         TMonaLisaText mlStatus("SHUTTLE_status", "Alive");
3018
3019         TList mlList;