Adding names of detector experts in the log in case of failure.
[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         if (fTestMode & kErrorStorage)
204         {
205                 Log(fCurrentDetector, "StoreLocally - In TESTMODE - Simulating error while storing locally");
206                 return kFALSE;
207         }
208         
209         const char* cdbType = (localUri == fgkLocalCDB) ? "CDB" : "Reference";
210
211         Int_t firstRun = GetCurrentRun() - validityStart;
212         if(firstRun < 0) {
213                 AliWarning("First valid run happens to be less than 0! Setting it to 0.");
214                 firstRun=0;
215         }
216
217         Int_t lastRun = -1;
218         if(validityInfinite) {
219                 lastRun = AliCDBRunRange::Infinity();
220         } else {
221                 lastRun = GetCurrentRun();
222         }
223
224         // Version is set to current run, it will be used later to transfer data to Grid
225         AliCDBId id(path, firstRun, lastRun, GetCurrentRun(), -1);
226
227         if(! dynamic_cast<TObjString*> (metaData->GetProperty("RunUsed(TObjString)"))){
228                 TObjString runUsed = Form("%d", GetCurrentRun());
229                 metaData->SetProperty("RunUsed(TObjString)", runUsed.Clone());
230         }
231
232         Bool_t result = kFALSE;
233
234         if (!(AliCDBManager::Instance()->GetStorage(localUri))) {
235                 Log("SHUTTLE", Form("StoreLocally - Cannot activate local %s storage", cdbType));
236         } else {
237                 result = AliCDBManager::Instance()->GetStorage(localUri)
238                                         ->Put(object, id, metaData);
239         }
240
241         if(!result) {
242
243                 Log(fCurrentDetector, Form("StoreLocally - Can't store object <%s>!", id.ToString().Data()));
244         }
245
246         return result;
247 }
248
249 //______________________________________________________________________________________________
250 Bool_t AliShuttle::StoreOCDB()
251 {
252         //
253         // Called when preprocessor ends successfully or when previous storage attempt failed (kStoreError status)
254         // Calls underlying StoreOCDB(const char*) function twice, for OCDB and Reference storage.
255         // Then calls StoreRefFilesToGrid to store reference files. 
256         //
257         
258         UpdateShuttleStatus(AliShuttleStatus::kStoreStarted);
259                         
260         if (fTestMode & kErrorGrid)
261         {
262                 Log("SHUTTLE", "StoreOCDB - In TESTMODE - Simulating error while storing in the Grid");
263                 Log(fCurrentDetector, "StoreOCDB - In TESTMODE - Simulating error while storing in the Grid");
264                 return kFALSE;
265         }
266         
267         Log("SHUTTLE","StoreOCDB - Storing OCDB data ...");
268         Int_t resultCDB = StoreOCDB(fgkMainCDB);
269
270         Log("SHUTTLE","StoreOCDB - Storing reference data ...");
271         Int_t resultRef = StoreOCDB(fgkMainRefStorage);
272         
273         Log("SHUTTLE","StoreOCDB - Storing reference files ...");
274         Bool_t resultRefFiles = CopyFilesToGrid("reference");
275         
276         Bool_t resultMetadata = kTRUE;
277         if(fCurrentDetector == "GRP") 
278         {
279                 Log("SHUTTLE","StoreOCDB - Storing Run Metadata file ...");
280                 resultMetadata = CopyFilesToGrid("metadata");
281         }
282         
283         Int_t storeResult = 0;
284         
285         if (resultCDB < 0 || resultRef < 0 || resultRefFiles == kFALSE || resultMetadata == kFALSE)
286                 storeResult = -1;
287         else if (resultCDB > 0 || resultRef > 0)
288                 storeResult = 1;
289                 
290         if (storeResult < 0)
291         {
292                 Log("SHUTTLE", 
293                         Form("\t\t\t****** run %d - %s: STORAGE ERROR ******",
294                                 GetCurrentRun(), fCurrentDetector.Data()));
295                 UpdateShuttleStatus(AliShuttleStatus::kStoreError);
296         } 
297         else if (storeResult > 0)
298         {
299                 Log("SHUTTLE", 
300                         Form("\t\t\t****** run %d - %s: STORAGE DELAYED ******",
301                                 GetCurrentRun(), fCurrentDetector.Data()));
302                 UpdateShuttleStatus(AliShuttleStatus::kStoreDelayed);
303         }
304         else if (storeResult == 0) 
305         {
306                 Log("SHUTTLE", 
307                         Form("\t\t\t****** run %d - %s: DONE ******",
308                                 GetCurrentRun(), fCurrentDetector.Data()));
309                 UpdateShuttleStatus(AliShuttleStatus::kDone);
310                 UpdateShuttleLogbook(fCurrentDetector, "DONE");
311         }
312
313         return (storeResult == 0);
314 }
315
316 //______________________________________________________________________________________________
317 Int_t AliShuttle::StoreOCDB(const TString& gridURI)
318 {
319         //
320         // Called by StoreOCDB(), performs actual storage to the main OCDB and reference storages (Grid)
321         //
322         // Return code:
323         //   -2 initialization error
324         //   -1 storage error
325         //   0  success
326         //   1  storage delayed (e.g. previous unprocessed runs)
327         //
328
329         TObjArray* gridIds=0;
330
331         Bool_t result = kTRUE;
332         Bool_t delayed = kFALSE;
333
334         const char* type = 0;
335         TString localURI;
336         if(gridURI == fgkMainCDB) {
337                 type = "OCDB";
338                 localURI = fgkLocalCDB;
339         } else if(gridURI == fgkMainRefStorage) {
340                 type = "reference";
341                 localURI = fgkLocalRefStorage;
342         } else {
343                 AliError(Form("Invalid storage URI: %s", gridURI.Data()));
344                 return -2;
345         }
346
347         AliCDBManager* man = AliCDBManager::Instance();
348
349         AliCDBStorage *gridSto = man->GetStorage(gridURI);
350         if(!gridSto) {
351                 Log("SHUTTLE",
352                         Form("StoreOCDB - cannot activate main %s storage", type));
353                 return -2;
354         }
355
356         gridIds = gridSto->GetQueryCDBList();
357
358         // get objects previously stored in local CDB
359         AliCDBStorage *localSto = man->GetStorage(localURI);
360         if(!localSto) {
361                 Log("SHUTTLE",
362                         Form("StoreOCDB - cannot activate local %s storage", type));
363                 return -2;
364         }
365         AliCDBPath aPath(GetOfflineDetName(fCurrentDetector.Data()),"*","*");
366         // Local objects were stored with current run as Grid version!
367         TList* localEntries = localSto->GetAll(aPath.GetPath(), GetCurrentRun(), GetCurrentRun());
368         localEntries->SetOwner(1);
369
370         // loop on local stored objects
371         TIter localIter(localEntries);
372         AliCDBEntry *aLocEntry = 0;
373         while((aLocEntry = dynamic_cast<AliCDBEntry*> (localIter.Next()))){
374                 aLocEntry->SetOwner(1);
375                 AliCDBId aLocId = aLocEntry->GetId();
376                 aLocEntry->SetVersion(-1);
377                 aLocEntry->SetSubVersion(-1);
378                                                 
379                 Log(fCurrentDetector.Data(), Form("Attempting to store %s", aLocId.ToString().Data()));
380
381                 // If local object is valid up to infinity we store it only if it is
382                 // the first unprocessed run!
383                 if (aLocId.GetLastRun() == AliCDBRunRange::Infinity() &&
384                         !fFirstUnprocessed[GetDetPos(fCurrentDetector)])
385                 {
386                         Log("SHUTTLE", Form("StoreOCDB - %s: object %s has validity infinite but "
387                                                 "there are previous unprocessed runs!",
388                                                 fCurrentDetector.Data(), aLocId.GetPath().Data()));
389                         Log(fCurrentDetector.Data(), Form("StoreOCDB - %s: object %s has validity infinite but "
390                                                 "there are previous unprocessed runs!",
391                                                 fCurrentDetector.Data(), aLocId.GetPath().Data()));
392                         delayed = kTRUE;
393                         continue;
394                 }
395
396                 // loop on Grid valid Id's
397                 Bool_t store = kTRUE;
398                 TIter gridIter(gridIds);
399                 AliCDBId* aGridId = 0;
400                 while ((aGridId = dynamic_cast<AliCDBId*> (gridIter.Next()))) {
401                         if (aGridId->GetPath() != aLocId.GetPath()) 
402                                 continue;
403                         // skip all objects valid up to infinity
404                         if (aGridId->GetLastRun() == AliCDBRunRange::Infinity()) 
405                                 continue;
406                         
407                         // if we get here, it means there's already some more recent object stored on Grid!
408                         Log(fCurrentDetector.Data(),
409                                 Form("StoreOCDB - A more recent object already exists in %s storage: <%s>",
410                                 type, aGridId->ToString().Data()));
411                         
412                         store = kFALSE;
413                         break;
414                 }
415
416                 Bool_t storeOk = kFALSE;
417                 if (store)
418                 {
419                         Log(fCurrentDetector.Data(), Form("Prechecks succeeded. Ready to store %s", aLocId.ToString().Data()));
420                         storeOk = gridSto->Put(aLocEntry);
421                         if (storeOk) {
422                                 Log("SHUTTLE",
423                                 Form("StoreOCDB - Object <%s> successfully put into %s storage",
424                                         aLocId.ToString().Data(), type));
425                                 Log(fCurrentDetector.Data(),
426                                         Form("StoreOCDB - Object <%s> successfully put into %s storage",
427                                         aLocId.ToString().Data(), type));
428                         } else  {
429                                 Log("SHUTTLE",
430                                         Form("StoreOCDB - Grid %s storage of object <%s> failed",
431                                         type, aLocId.ToString().Data()));
432                                 Log(fCurrentDetector.Data(),
433                                         Form("StoreOCDB - Grid %s storage of object <%s> failed",
434                                         type, aLocId.ToString().Data()));
435                                 result = kFALSE;
436                         }
437                 }
438                 
439                 if (!store || storeOk) {
440                         // removing local file...
441                         TString filename;
442                         localSto->IdToFilename(aLocId, filename);
443                         Log("SHUTTLE", Form("StoreOCDB - Removing local file %s", filename.Data()));
444                         RemoveFile(filename.Data());
445                 }
446         }
447         localEntries->Clear();
448
449         Int_t returnCode = 0;
450         
451         if (result == kFALSE)
452                 returnCode = -1;
453         else if (delayed != kFALSE)
454                 returnCode =  1;
455
456         Log("SHUTTLE", Form("StoreOCDB - Returning with %d (result = %d, delayed = %d)", returnCode, result, delayed));
457         Log(fCurrentDetector.Data(), Form("StoreOCDB - Returning with %d (result = %d, delayed = %d)", returnCode, result, delayed));
458         
459         return returnCode;
460 }
461
462 //______________________________________________________________________________________________
463 Bool_t AliShuttle::CleanReferenceStorage(const char* detector)
464 {
465         // clears the directory used to store reference files of a given subdetector
466   
467         AliCDBManager* man = AliCDBManager::Instance();
468         AliCDBStorage* sto = man->GetStorage(fgkLocalRefStorage);
469         TString localBaseFolder = sto->GetBaseFolder();
470
471         TString targetDir = GetRefFilePrefix(localBaseFolder.Data(), detector);
472         
473         Log("SHUTTLE", Form("CleanReferenceStorage - Cleaning %s", targetDir.Data()));
474
475         TString begin;
476         begin.Form("%d_", GetCurrentRun());
477         
478         TSystemDirectory* baseDir = new TSystemDirectory("/", targetDir);
479         if (!baseDir)
480                 return kTRUE;
481                 
482         TList* dirList = baseDir->GetListOfFiles();
483         delete baseDir;
484         
485         if (!dirList) return kTRUE;
486                         
487         if (dirList->GetEntries() < 3) 
488         {
489                 delete dirList;
490                 return kTRUE;
491         }
492                                 
493         Int_t nDirs = 0, nDel = 0;
494         TIter dirIter(dirList);
495         TSystemFile* entry = 0;
496
497         Bool_t success = kTRUE;
498         
499         while ((entry = dynamic_cast<TSystemFile*> (dirIter.Next())))
500         {                                       
501                 if (entry->IsDirectory())
502                         continue;
503                 
504                 TString fileName(entry->GetName());
505                 if (!fileName.BeginsWith(begin))
506                         continue;
507                         
508                 nDirs++;
509                                                 
510                 // delete file
511                 Int_t result = gSystem->Unlink(fileName.Data());
512                 
513                 if (result)
514                 {
515                         Log("SHUTTLE", Form("CleanReferenceStorage - Could not delete file %s!", fileName.Data()));
516                         success = kFALSE;
517                 } else {
518                         nDel++;
519                 }
520         }
521
522         if(nDirs > 0)
523                 Log("SHUTTLE", Form("CleanReferenceStorage - %d (over %d) reference files in folder %s were deleted.", 
524                         nDel, nDirs, targetDir.Data()));
525
526                 
527         delete dirList;
528         return success;
529
530
531
532
533
534
535   Int_t result = gSystem->GetPathInfo(targetDir, 0, (Long64_t*) 0, 0, 0);
536   if (result == 0)
537   {
538     // delete directory
539     result = gSystem->Exec(Form("rm -rf %s", targetDir.Data()));
540     if (result != 0)
541     {  
542       Log("SHUTTLE", Form("CleanReferenceStorage - Could not clean directory %s", targetDir.Data()));
543       return kFALSE;
544     }
545   }
546
547   result = gSystem->mkdir(targetDir, kTRUE);
548   if (result != 0)
549   {
550     Log("SHUTTLE", Form("CleanReferenceStorage - Error creating base directory %s", targetDir.Data()));
551     return kFALSE;
552   }
553         
554   return kTRUE;
555 }
556
557 //______________________________________________________________________________________________
558 Bool_t AliShuttle::StoreReferenceFile(const char* detector, const char* localFile, const char* gridFileName)
559 {
560         //
561         // Stores reference file directly (without opening it). This function stores the file locally.
562         //
563         // The file is stored under the following location: 
564         // <base folder of local reference storage>/<DET>/<RUN#>_<gridFileName>
565         // where <gridFileName> is the second parameter given to the function
566         // 
567         
568         if (fTestMode & kErrorStorage)
569         {
570                 Log(fCurrentDetector, "StoreReferenceFile - In TESTMODE - Simulating error while storing locally");
571                 return kFALSE;
572         }
573         
574         AliCDBManager* man = AliCDBManager::Instance();
575         AliCDBStorage* sto = man->GetStorage(fgkLocalRefStorage);
576         
577         TString localBaseFolder = sto->GetBaseFolder();
578         
579         TString target = GetRefFilePrefix(localBaseFolder.Data(), detector);    
580         target.Append(Form("/%d_%s", GetCurrentRun(), gridFileName));
581         
582         return CopyFileLocally(localFile, target);
583 }
584
585 //______________________________________________________________________________________________
586 Bool_t AliShuttle::StoreRunMetadataFile(const char* localFile, const char* gridFileName)
587 {
588         //
589         // Stores Run metadata file to the Grid, in the run folder
590         //
591         // Only GRP can call this function.
592         
593         if (fTestMode & kErrorStorage)
594         {
595                 Log(fCurrentDetector, "StoreRunMetaDataFile - In TESTMODE - Simulating error while storing locally");
596                 return kFALSE;
597         }
598         
599         AliCDBManager* man = AliCDBManager::Instance();
600         AliCDBStorage* sto = man->GetStorage(fgkLocalRefStorage);
601         
602         TString localBaseFolder = sto->GetBaseFolder();
603         
604         // Build Run level folder
605         // folder = /alice/data/year/lhcPeriod/runNb/raw
606         
607                 
608         TString lhcPeriod = GetLHCPeriod();     
609         if (lhcPeriod.Length() == 0) 
610         {
611                 Log("SHUTTLE","StoreRunMetaDataFile - LHCPeriod not found in logbook!");
612                 return 0;
613         }
614         
615         // TODO partitions with one detector only write data into LHCperiod_DET
616         TString partition = GetRunParameter("detector");
617         
618         if (partition.Length() > 0 && partition != "ALICE")
619         {
620                 lhcPeriod.Append(Form("_%s", partition.Data()));
621                 Log(fCurrentDetector, Form("Run data tags merged file will be written in %s", 
622                                 lhcPeriod.Data()));
623         }
624                 
625         TString target = Form("%s/GRP/RunMetadata%s%d/%s/%09d/raw/%s", 
626                                 localBaseFolder.Data(), fConfig->GetAlienPath(), GetCurrentYear(), 
627                                 lhcPeriod.Data(), GetCurrentRun(), gridFileName);
628                                         
629         return CopyFileLocally(localFile, target);
630 }
631
632 //______________________________________________________________________________________________
633 Bool_t AliShuttle::CopyFileLocally(const char* localFile, const TString& target)
634 {
635         //
636         // Stores file locally. Called by StoreReferenceFile and StoreRunMetadataFile
637         // Files are temporarily stored in the local reference storage. When the preprocessor 
638         // finishes, the Shuttle calls CopyFilesToGrid to transfer the files to AliEn 
639         // (in reference or run level folders)
640         //
641         
642         TString targetDir(target(0, target.Last('/')));
643         
644         //try to open base dir folder, if it does not exist
645         void* dir = gSystem->OpenDirectory(targetDir.Data());
646         if (dir == NULL) {
647                 if (gSystem->mkdir(targetDir.Data(), kTRUE)) {
648                         Log("SHUTTLE", Form("CopyFileLocally - Can't open directory <%s>", targetDir.Data()));
649                         return kFALSE;
650                 }
651
652         } else {
653                 gSystem->FreeDirectory(dir);
654         }
655         
656         Int_t result = 0;
657         
658         result = gSystem->GetPathInfo(localFile, 0, (Long64_t*) 0, 0, 0);
659         if (result)
660         {
661                 Log("SHUTTLE", Form("CopyFileLocally - %s does not exist", localFile));
662                 return kFALSE;
663         }
664
665         result = gSystem->GetPathInfo(target, 0, (Long64_t*) 0, 0, 0);
666         if (!result)
667         {
668                 Log("SHUTTLE", Form("CopyFileLocally - target file %s already exist, removing...", target.Data()));
669                 if (gSystem->Unlink(target.Data()))
670                 {
671                         Log("SHUTTLE", Form("CopyFileLocally - Could not remove existing target file %s!", target.Data()));
672                         return kFALSE;
673                 }
674         }       
675         
676         result = gSystem->CopyFile(localFile, target);
677
678         if (result == 0)
679         {
680                 Log("SHUTTLE", Form("CopyFileLocally - File %s stored locally to %s", localFile, target.Data()));
681                 return kTRUE;
682         }
683         else
684         {
685                 Log("SHUTTLE", Form("CopyFileLocally - Could not store file %s to %s! Error code = %d", 
686                                 localFile, target.Data(), result));
687                 return kFALSE;
688         }       
689
690
691
692 }
693
694 //______________________________________________________________________________________________
695 Bool_t AliShuttle::CopyFilesToGrid(const char* type)
696 {
697         //
698         // Transfers local files to the Grid. Local files can be reference files 
699         // or run metadata file (from GRP only).
700         //
701         // According to the type (ref, metadata) the files are stored under the following location: 
702         // ref --> <base folder of reference storage>/<DET>/<RUN#>_<gridFileName>
703         // metadata --> <run data folder>/<MetadataFileName>
704         //
705                 
706         AliCDBManager* man = AliCDBManager::Instance();
707         AliCDBStorage* sto = man->GetStorage(fgkLocalRefStorage);
708         if (!sto)
709                 return kFALSE;
710         TString localBaseFolder = sto->GetBaseFolder();
711         
712         TString dir;
713         TString alienDir;
714         TString begin;
715         
716         if (strcmp(type, "reference") == 0) 
717         {
718                 dir = GetRefFilePrefix(localBaseFolder.Data(), fCurrentDetector.Data());
719                 AliCDBStorage* gridSto = man->GetStorage(fgkMainRefStorage);
720                 if (!gridSto)
721                         return kFALSE;
722                 TString gridBaseFolder = gridSto->GetBaseFolder();
723                 alienDir = GetRefFilePrefix(gridBaseFolder.Data(), fCurrentDetector.Data());
724                 begin = Form("%d_", GetCurrentRun());
725         } 
726         else if (strcmp(type, "metadata") == 0)
727         {
728                         
729                 TString lhcPeriod = GetLHCPeriod();
730         
731                 if (lhcPeriod.Length() == 0) 
732                 {
733                         Log("SHUTTLE","CopyFilesToGrid - LHCPeriod not found in logbook!");
734                         return 0;
735                 }
736                 
737                 // TODO partitions with one detector only write data into LHCperiod_DET
738                 TString partition = GetRunParameter("detector");
739         
740                 if (partition.Length() > 0 && partition != "ALICE")
741                 {
742                         lhcPeriod.Append(Form("_%s", partition.Data()));
743                 }
744                 
745                 dir = Form("%s/GRP/RunMetadata%s%d/%s/%09d/raw", 
746                                 localBaseFolder.Data(), fConfig->GetAlienPath(), GetCurrentYear(), 
747                                 lhcPeriod.Data(), GetCurrentRun());
748                 alienDir = dir(dir.Index(fConfig->GetAlienPath()), dir.Length());
749                 
750                 begin = "";
751         }
752         else 
753         {
754                 Log("SHUTTLE", "CopyFilesToGrid - Unexpected: type label must be reference or metadata!");
755                 return kFALSE;
756         }
757                 
758         TSystemDirectory* baseDir = new TSystemDirectory("/", dir);
759         if (!baseDir)
760                 return kTRUE;
761                 
762         TList* dirList = baseDir->GetListOfFiles();
763         delete baseDir;
764         
765         if (!dirList) return kTRUE;
766                 
767         if (dirList->GetEntries() < 3) 
768         {
769                 delete dirList;
770                 return kTRUE;
771         }
772                         
773         if (!gGrid)
774         { 
775                 Log("SHUTTLE", "CopyFilesToGrid - Connection to Grid failed: Cannot continue!");
776                 delete dirList;
777                 return kFALSE;
778         }
779         
780         Int_t nDirs = 0, nTransfer = 0;
781         TIter dirIter(dirList);
782         TSystemFile* entry = 0;
783
784         Bool_t success = kTRUE;
785         Bool_t first = kTRUE;
786         
787         while ((entry = dynamic_cast<TSystemFile*> (dirIter.Next())))
788         {                       
789                 if (entry->IsDirectory())
790                         continue;
791                         
792                 TString fileName(entry->GetName());
793                 if (!fileName.BeginsWith(begin))
794                         continue;
795                         
796                 nDirs++;
797                         
798                 if (first)
799                 {
800                         first = kFALSE;
801                         // check that folder exists, otherwise create it
802                         TGridResult* result = gGrid->Ls(alienDir.Data(), "a");
803                         
804                         if (!result)
805                         {
806                                 delete dirList;
807                                 return kFALSE;
808                         }
809                         
810                         if (!result->GetFileName(1)) // TODO: It looks like element 0 is always 0!!
811                         {
812                                 // TODO It does not work currently! Bug in TAliEn::Mkdir
813                                 // TODO Manually fixed in local root v5-16-00
814                                 if (!gGrid->Mkdir(alienDir.Data(),"-p",0))
815                                 {
816                                         Log("SHUTTLE", Form("CopyFilesToGrid - Cannot create directory %s",
817                                                         alienDir.Data()));
818                                         delete dirList;
819                                         return kFALSE;
820                                 } else {
821                                         Log("SHUTTLE",Form("CopyFilesToGrid - Folder %s created", alienDir.Data()));
822                                 }
823                                 
824                         } else {
825                                         Log("SHUTTLE",Form("CopyFilesToGrid - Folder %s found", alienDir.Data()));
826                         }
827                 }
828                         
829                 TString fullLocalPath;
830                 fullLocalPath.Form("%s/%s", dir.Data(), fileName.Data());
831                 
832                 TString fullGridPath;
833                 fullGridPath.Form("alien://%s/%s", alienDir.Data(), fileName.Data());
834
835                 Bool_t result = TFile::Cp(fullLocalPath, fullGridPath);
836                 
837                 if (result)
838                 {
839                         Log("SHUTTLE", Form("CopyFilesToGrid - Copying local file %s to %s succeeded!", 
840                                                 fullLocalPath.Data(), fullGridPath.Data()));
841                         RemoveFile(fullLocalPath);
842                         nTransfer++;
843                 }
844                 else
845                 {
846                         Log("SHUTTLE", Form("CopyFilesToGrid - Copying local file %s to %s FAILED!", 
847                                                 fullLocalPath.Data(), fullGridPath.Data()));
848                         success = kFALSE;
849                 }
850         }
851
852         Log("SHUTTLE", Form("CopyFilesToGrid - %d (over %d) files in folder %s copied to Grid.", 
853                                                 nTransfer, nDirs, dir.Data()));
854
855                 
856         delete dirList;
857         return success;
858 }
859
860 //______________________________________________________________________________________________
861 const char* AliShuttle::GetRefFilePrefix(const char* base, const char* detector)
862 {
863         //
864         // Get folder name of reference files 
865         //
866
867         TString offDetStr(GetOfflineDetName(detector));
868         TString dir;
869         if (offDetStr == "ITS" || offDetStr == "MUON" || offDetStr == "PHOS")
870         {
871                 dir.Form("%s/%s/%s", base, offDetStr.Data(), detector);
872         } else {
873                 dir.Form("%s/%s", base, offDetStr.Data());
874         }
875         
876         return dir.Data();
877         
878
879 }
880
881 //______________________________________________________________________________________________
882 void AliShuttle::CleanLocalStorage(const TString& uri)
883 {
884         //
885         // Called in case the preprocessor is declared failed. Remove remaining objects from the local storages.
886         //
887
888         const char* type = 0;
889         if(uri == fgkLocalCDB) {
890                 type = "OCDB";
891         } else if(uri == fgkLocalRefStorage) {
892                 type = "Reference";
893         } else {
894                 AliError(Form("Invalid storage URI: %s", uri.Data()));
895                 return;
896         }
897
898         AliCDBManager* man = AliCDBManager::Instance();
899
900         // open local storage
901         AliCDBStorage *localSto = man->GetStorage(uri);
902         if(!localSto) {
903                 Log("SHUTTLE",
904                         Form("CleanLocalStorage - cannot activate local %s storage", type));
905                 return;
906         }
907
908         TString filename(Form("%s/%s/*/Run*_v%d_s*.root",
909                 localSto->GetBaseFolder().Data(), GetOfflineDetName(fCurrentDetector.Data()), GetCurrentRun()));
910
911         AliDebug(2, Form("filename = %s", filename.Data()));
912
913         Log("SHUTTLE", Form("Removing remaining local files for run %d and detector %s ...",
914                 GetCurrentRun(), fCurrentDetector.Data()));
915
916         RemoveFile(filename.Data());
917
918 }
919
920 //______________________________________________________________________________________________
921 void AliShuttle::RemoveFile(const char* filename)
922 {
923         //
924         // removes local file
925         //
926
927         TString command(Form("rm -f %s", filename));
928
929         Int_t result = gSystem->Exec(command.Data());
930         if(result != 0)
931         {
932                 Log("SHUTTLE", Form("RemoveFile - %s: Cannot remove file %s!",
933                         fCurrentDetector.Data(), filename));
934         }
935 }
936
937 //______________________________________________________________________________________________
938 AliShuttleStatus* AliShuttle::ReadShuttleStatus()
939 {
940         //
941         // Reads the AliShuttleStatus from the CDB
942         //
943
944         if (fStatusEntry){
945                 delete fStatusEntry;
946                 fStatusEntry = 0;
947         }
948
949         Int_t path1 = GetCurrentRun()/10000;
950         fStatusEntry = AliCDBManager::Instance()->GetStorage(GetLocalCDB())
951                 ->Get(Form("/SHUTTLE/%s/%d", fCurrentDetector.Data(), path1), GetCurrentRun());
952
953         if (!fStatusEntry) return 0;
954         fStatusEntry->SetOwner(1);
955
956         AliShuttleStatus* status = dynamic_cast<AliShuttleStatus*> (fStatusEntry->GetObject());
957         if (!status) {
958                 AliError("Invalid object stored to CDB!");
959                 return 0;
960         }
961
962         return status;
963 }
964
965 //______________________________________________________________________________________________
966 Bool_t AliShuttle::WriteShuttleStatus(AliShuttleStatus* status)
967 {
968         //
969         // writes the status for one subdetector
970         //
971
972         if (fStatusEntry){
973                 delete fStatusEntry;
974                 fStatusEntry = 0;
975         }
976
977         Int_t run = GetCurrentRun();
978         Int_t path1 = run/10000;
979         TString path1_string = Form("%d",path1);
980
981         AliCDBId id(AliCDBPath("SHUTTLE", fCurrentDetector, path1_string), run, run);
982
983         fStatusEntry = new AliCDBEntry(status, id, new AliCDBMetaData);
984         fStatusEntry->SetOwner(1);
985
986         UInt_t result = AliCDBManager::Instance()->GetStorage(fgkLocalCDB)->Put(fStatusEntry);
987
988         if (!result) {
989                 Log("SHUTTLE", Form("WriteShuttleStatus - Failed for %s, run %d",
990                                                 fCurrentDetector.Data(), run));
991                 return kFALSE;
992         }
993         
994         SendMLDetInfo();
995
996         return kTRUE;
997 }
998
999 //______________________________________________________________________________________________
1000 void AliShuttle::UpdateShuttleStatus(AliShuttleStatus::Status newStatus, Bool_t increaseCount)
1001 {
1002         //
1003         // changes the AliShuttleStatus for the given detector and run to the given status
1004         //
1005
1006         if (!fStatusEntry){
1007                 AliError("UNEXPECTED: fStatusEntry empty");
1008                 return;
1009         }
1010
1011         AliShuttleStatus* status = dynamic_cast<AliShuttleStatus*> (fStatusEntry->GetObject());
1012
1013         if (!status){
1014                 Log("SHUTTLE", "UpdateShuttleStatus - UNEXPECTED: status could not be read from current CDB entry");
1015                 return;
1016         }
1017
1018         TString actionStr = Form("UpdateShuttleStatus - %s: Changing state from %s to %s",
1019                                 fCurrentDetector.Data(),
1020                                 status->GetStatusName(),
1021                                 status->GetStatusName(newStatus));
1022         Log("SHUTTLE", actionStr);
1023         SetLastAction(actionStr);
1024
1025         status->SetStatus(newStatus);
1026         if (increaseCount) status->IncreaseCount();
1027
1028         AliCDBManager::Instance()->GetStorage(fgkLocalCDB)->Put(fStatusEntry);
1029
1030         SendMLDetInfo();
1031 }
1032
1033 //______________________________________________________________________________________________
1034 void AliShuttle::SendMLDetInfo()
1035 {
1036         //
1037         // sends ML information about the current status of the current detector being processed
1038         //
1039         
1040         AliShuttleStatus* status = dynamic_cast<AliShuttleStatus*> (fStatusEntry->GetObject());
1041         
1042         if (!status){
1043                 Log("SHUTTLE", "SendMLDetInfo - UNEXPECTED: status could not be read from current CDB entry");
1044                 return;
1045         }
1046         
1047         TMonaLisaText  mlStatus(Form("%s_status", fCurrentDetector.Data()), status->GetStatusName());
1048         TMonaLisaValue mlRetryCount(Form("%s_count", fCurrentDetector.Data()), status->GetCount());
1049
1050         TList mlList;
1051         mlList.Add(&mlStatus);
1052         mlList.Add(&mlRetryCount);
1053
1054         TString mlID;
1055         mlID.Form("%d", GetCurrentRun());
1056         fMonaLisa->SendParameters(&mlList, mlID);
1057 }
1058
1059 //______________________________________________________________________________________________
1060 Bool_t AliShuttle::ContinueProcessing()
1061 {
1062         // this function reads the AliShuttleStatus information from CDB and
1063         // checks if the processing should be continued
1064         // if yes it returns kTRUE and updates the AliShuttleStatus with nextStatus
1065
1066         if (!fConfig->HostProcessDetector(fCurrentDetector))
1067                 return kFALSE;
1068
1069         AliPreprocessor* aPreprocessor =
1070                 dynamic_cast<AliPreprocessor*> (fPreprocessorMap.GetValue(fCurrentDetector));
1071         if (!aPreprocessor)
1072         {
1073                 Log("SHUTTLE", Form("ContinueProcessing - %s: no preprocessor registered", fCurrentDetector.Data()));
1074                 return kFALSE;
1075         }
1076
1077         AliShuttleLogbookEntry::Status entryStatus =
1078                 fLogbookEntry->GetDetectorStatus(fCurrentDetector);
1079
1080         if (entryStatus != AliShuttleLogbookEntry::kUnprocessed) {
1081                 Log("SHUTTLE", Form("ContinueProcessing - %s is %s",
1082                                 fCurrentDetector.Data(),
1083                                 fLogbookEntry->GetDetectorStatusName(entryStatus)));
1084                 return kFALSE;
1085         }
1086
1087         // if we get here, according to Shuttle logbook subdetector is in UNPROCESSED state
1088
1089         // check if current run is first unprocessed run for current detector
1090         if (fConfig->StrictRunOrder(fCurrentDetector) &&
1091                 !fFirstUnprocessed[GetDetPos(fCurrentDetector)])
1092         {
1093                 if (fTestMode == kNone)
1094                 {
1095                         Log("SHUTTLE", Form("ContinueProcessing - %s requires strict run ordering"
1096                                         " but this is not the first unprocessed run!"));
1097                         return kFALSE;
1098                 }
1099                 else
1100                 {
1101                         Log("SHUTTLE", Form("ContinueProcessing - In TESTMODE - "
1102                                         "Although %s requires strict run ordering "
1103                                         "and this is not the first unprocessed run, "
1104                                         "the SHUTTLE continues"));
1105                 }
1106         }
1107
1108         // Is the subdetector processed first time for this run?
1109         fFirstProcessing = kFALSE;
1110
1111         AliShuttleStatus* status = ReadShuttleStatus();
1112         if (!status) {
1113                 // first time
1114                 Log("SHUTTLE", Form("ContinueProcessing - %s: Processing first time",
1115                                 fCurrentDetector.Data()));
1116                 status = new AliShuttleStatus(AliShuttleStatus::kStarted);
1117                 fFirstProcessing = kTRUE;
1118                 return WriteShuttleStatus(status);
1119         }
1120
1121         // The following case shouldn't happen if Shuttle Logbook was correctly updated.
1122         // If it happens it may mean Logbook updating failed... let's do it now!
1123         if (status->GetStatus() == AliShuttleStatus::kDone ||
1124             status->GetStatus() == AliShuttleStatus::kFailed ||
1125             status->GetStatus() == AliShuttleStatus::kSkipped) {
1126                 Log("SHUTTLE", Form("ContinueProcessing - %s is already %s. Updating Shuttle Logbook",
1127                                         fCurrentDetector.Data(),
1128                                         status->GetStatusName(status->GetStatus())));
1129                 
1130                 if (status->GetStatus() == AliShuttleStatus::kSkipped)
1131                 {
1132                         UpdateShuttleLogbook(fCurrentDetector.Data(), "DONE");
1133                 }
1134                 else
1135                         UpdateShuttleLogbook(fCurrentDetector.Data(), status->GetStatusName(status->GetStatus()));
1136                         
1137                 return kFALSE;
1138         }
1139
1140         if (status->GetStatus() == AliShuttleStatus::kStoreStarted || status->GetStatus() == AliShuttleStatus::kStoreDelayed ||status->GetStatus() == AliShuttleStatus::kStoreError) {
1141                 Log("SHUTTLE",
1142                         Form("ContinueProcessing - %s: Grid storage of one or more "
1143                                 "objects failed. Trying again now",
1144                                 fCurrentDetector.Data()));
1145                 StoreOCDB();
1146                 return kFALSE;
1147         }
1148
1149         // if we get here, there is a restart
1150         Bool_t cont = kFALSE;
1151
1152         // abort conditions
1153         if (status->GetCount() >= fConfig->GetMaxRetries()) {
1154                 Log("SHUTTLE", Form("ContinueProcessing - %s failed %d times in status %s - "
1155                                 "Updating Shuttle Logbook", fCurrentDetector.Data(),
1156                                 status->GetCount(), status->GetStatusName()));
1157                 UpdateShuttleLogbook(fCurrentDetector.Data(), "FAILED");
1158                 UpdateShuttleStatus(AliShuttleStatus::kFailed);
1159
1160                 // there may still be objects in local OCDB and reference storage
1161                 // and FXS databases may be not updated: do it now!
1162                 
1163                 // TODO Currently disabled, we want to keep files in case of failure!
1164                 // CleanLocalStorage(fgkLocalCDB);
1165                 // CleanLocalStorage(fgkLocalRefStorage);
1166                 // UpdateTableFailCase();
1167                 
1168                 // Send mail to detector expert!
1169                 Log("SHUTTLE", Form("ContinueProcessing - Sending mail to %s expert...", 
1170                                     fCurrentDetector.Data()));
1171                 // det experts in to
1172                 TString to="";
1173                 TIter *iterExperts = 0;
1174                 iterExperts = new TIter(fConfig->GetResponsibles(fCurrentDetector));
1175                 TObjString *anExpert=0;
1176                 while ((anExpert = (TObjString*) iterExperts->Next()))
1177                         {
1178                                 to += Form("%s, \n", anExpert->GetName());
1179                         }
1180                 delete iterExperts;
1181                 
1182                 if (to.Length() > 0)
1183                         to.Remove(to.Length()-3);
1184                 AliDebug(2, Form("to: %s",to.Data()));
1185
1186                 if (to.IsNull()) {
1187                         Log("SHUTTLE", Form("List of %s responsibles not set!", fCurrentDetector.Data()));
1188                         return kFALSE;
1189                 }
1190
1191                 Log(fCurrentDetector.Data(), Form("ContinueProcessing - Sending mail to %s expert(s):", 
1192                                     fCurrentDetector.Data()));
1193                 Log(fCurrentDetector.Data(), Form("\n%s", to.Data()));
1194                 if (!SendMail(kPPEMail))
1195                         Log("SHUTTLE", Form("ContinueProcessing - Could not send mail to %s expert",
1196                                             fCurrentDetector.Data()));
1197
1198         } else {
1199                 Log("SHUTTLE", Form("ContinueProcessing - %s: restarting. "
1200                                 "Aborted before with %s. Retry number %d.", fCurrentDetector.Data(),
1201                                 status->GetStatusName(), status->GetCount()));
1202                 Bool_t increaseCount = kTRUE;
1203                 if (status->GetStatus() == AliShuttleStatus::kDCSError || 
1204                     status->GetStatus() == AliShuttleStatus::kDCSStarted ||
1205                     status->GetStatus() == AliShuttleStatus::kFXSError)
1206                                 increaseCount = kFALSE;
1207                                 
1208                 UpdateShuttleStatus(AliShuttleStatus::kStarted, increaseCount);
1209                 cont = kTRUE;
1210         }
1211
1212         return cont;
1213 }
1214
1215 //______________________________________________________________________________________________
1216 void AliShuttle::SendMLRunInfo(const char* status)
1217 {
1218         // 
1219         // Send information about this run to ML
1220         
1221         TMonaLisaText  mlStatus("SHUTTLE_status", status);
1222         TString runType(fLogbookEntry->GetRunType());
1223         if (strlen(fLogbookEntry->GetRunParameter("log")) > 0){
1224
1225                 runType += "(";
1226                 runType += fLogbookEntry->GetRunParameter("log");
1227                 runType += ")";
1228         }
1229         if (fLogbookEntry->GetDATestMode()){
1230                 runType += " (DATest)";
1231         }
1232         TMonaLisaText  mlRunType("SHUTTLE_runtype", runType);
1233
1234         TList mlList;
1235         mlList.Add(&mlStatus);
1236         mlList.Add(&mlRunType);
1237
1238         TString mlID;
1239         mlID.Form("%d", GetCurrentRun());
1240         fMonaLisa->SendParameters(&mlList, mlID);       
1241 }
1242
1243 //______________________________________________________________________________________________
1244 Int_t AliShuttle::GetMem(Int_t pid)
1245 {
1246         // invokes ps to get the memory consumption of the process <pid>
1247         // returns -1 in case of error
1248         
1249         TString checkStr;
1250         checkStr.Form("ps -o vsize --pid %d | tail -n 1", pid);
1251         FILE* pipe = gSystem->OpenPipe(checkStr, "r");
1252         if (!pipe)
1253         {
1254                 Log("SHUTTLE", Form("Process - Error: "
1255                         "Could not open pipe to %s", checkStr.Data()));
1256                 return -1;
1257         }
1258                 
1259         char buffer[100];
1260         if (!fgets(buffer, 100, pipe))
1261         {
1262                 Log("SHUTTLE", "Process - Error: ps did not return anything");
1263                 gSystem->ClosePipe(pipe);
1264                 return -1;
1265         }
1266         gSystem->ClosePipe(pipe);
1267         
1268         //Log("SHUTTLE", Form("ps returned %s", buffer));
1269         
1270         Int_t mem = 0;
1271         if ((sscanf(buffer, "%d\n", &mem) != 1) || !mem)
1272         {
1273                 Log("SHUTTLE", "Process - Error: Could not parse output of ps");
1274                 return -1;
1275         }
1276         
1277         return mem;
1278 }
1279
1280 //______________________________________________________________________________________________
1281 Bool_t AliShuttle::Process(AliShuttleLogbookEntry* entry)
1282 {
1283         //
1284         // Makes data retrieval for all detectors in the configuration.
1285         // entry: Shuttle logbook entry, contains run paramenters and status of detectors
1286         // (Unprocessed, Inactive, Failed or Done).
1287         // Returns kFALSE in case of error occured and kTRUE otherwise
1288         //
1289
1290         if (!entry) return kFALSE;
1291
1292         fLogbookEntry = entry;
1293
1294         Log("SHUTTLE", Form("\t\t\t^*^*^*^*^*^*^*^*^*^*^*^* run %d: START ^*^*^*^*^*^*^*^*^*^*^*^*",
1295                                         GetCurrentRun()));
1296
1297         CountOpenRuns();
1298         
1299         // Send the information to ML
1300         SendMLRunInfo("Processing");
1301
1302         if (fLogbookEntry->IsDone())
1303         {
1304                 Log("SHUTTLE","Process - Shuttle is already DONE. Updating logbook");
1305                 UpdateShuttleLogbook("shuttle_done");
1306                 fLogbookEntry = 0;
1307                 return kTRUE;
1308         }
1309
1310         // read test mode if flag is set
1311         if (fReadTestMode)
1312         {
1313                 fTestMode = kNone;
1314                 TString logEntry(entry->GetRunParameter("log"));
1315                 //printf("log entry = %s\n", logEntry.Data());
1316                 TString searchStr("Testmode: ");
1317                 Int_t pos = logEntry.Index(searchStr.Data());
1318                 //printf("%d\n", pos);
1319                 if (pos >= 0)
1320                 {
1321                         TSubString subStr = logEntry(pos + searchStr.Length(), logEntry.Length());
1322                         //printf("%s\n", subStr.String().Data());
1323                         TString newStr(subStr.Data());
1324                         TObjArray* token = newStr.Tokenize(' ');
1325                         if (token)
1326                         {
1327                                 //token->Print();
1328                                 TObjString* tmpStr = dynamic_cast<TObjString*> (token->First());
1329                                 if (tmpStr)
1330                                 {
1331                                         Int_t testMode = tmpStr->String().Atoi();
1332                                         if (testMode > 0)
1333                                         {
1334                                                 Log("SHUTTLE", Form("Process - Enabling test mode %d", testMode));
1335                                                 SetTestMode((TestMode) testMode);
1336                                         }
1337                                 }
1338                                 delete token;          
1339                         }
1340                 }
1341         }
1342                 
1343         fLogbookEntry->Print("all");
1344
1345         // Initialization
1346         Bool_t hasError = kFALSE;
1347
1348         // Set the CDB and Reference folders according to the year
1349
1350         // build cdb paths (repeat each time, run might be a DATest run)
1351         if (!fLogbookEntry->GetDATestMode()){
1352                 fgkMainCDB.Form("alien://folder=%s%d/OCDB?user=alidaq?cacheFold=/tmp/OCDBCache", 
1353                                 fConfig->GetAlienPath(), GetCurrentYear());
1354                 
1355                 fgkMainRefStorage.Form("alien://folder=%s%d/Reference?user=alidaq?cacheFold=/tmp/OCDBCache", 
1356                                        fConfig->GetAlienPath(), GetCurrentYear());
1357         }
1358         else {
1359                 fgkMainCDB.Form("alien://folder=%s%d/DATest/OCDB?user=alidaq?cacheFold=/tmp/OCDBCache",
1360                                 fConfig->GetAlienPath(), GetCurrentYear());
1361                 
1362                 fgkMainRefStorage.Form("alien://folder=%s%d/DATest/Reference?user=alidaq?cacheFold=/tmp/OCDBCache",
1363                                        fConfig->GetAlienPath(), GetCurrentYear());
1364         }
1365
1366         AliDebug(2,Form("Main CDB storage = %s",fgkMainCDB.Data()));
1367         AliDebug(2,Form("Main Reference storage = %s",fgkMainRefStorage.Data()));
1368
1369         // Loop on detectors in the configuration
1370         TIter iter(fConfig->GetDetectors());
1371         TObjString* aDetector = 0;
1372
1373         Bool_t first = kTRUE;
1374
1375         while ((aDetector = (TObjString*) iter.Next()))
1376         {
1377                 fCurrentDetector = aDetector->String();
1378
1379                 if (ContinueProcessing() == kFALSE) 
1380                         continue;
1381                 
1382                 if (first)
1383                 {
1384                   // only read QueryCDB when needed and only once
1385                   AliCDBStorage *mainCDBSto = AliCDBManager::Instance()->GetStorage(fgkMainCDB);
1386                   if(mainCDBSto) mainCDBSto->QueryCDB(GetCurrentRun());
1387                   AliCDBStorage *mainRefSto = AliCDBManager::Instance()->GetStorage(fgkMainRefStorage);
1388                   if(mainRefSto) mainRefSto->QueryCDB(GetCurrentRun());
1389                   first = kFALSE;
1390                 }
1391
1392                 Log("SHUTTLE", Form("\t\t\t****** run %d - %s: START  ******",
1393                                                 GetCurrentRun(), aDetector->GetName()));
1394
1395                 for(Int_t iSys=0;iSys<3;iSys++) fFXSCalled[iSys]=kFALSE;
1396                 
1397                 Int_t initialMem = GetMem(getpid());
1398                 Log("SHUTTLE", Form("Memory consumption before forking is %d", initialMem));
1399
1400                 Log(fCurrentDetector.Data(), "Process - Starting processing");
1401
1402                 Int_t pid = fork();
1403
1404                 if (pid < 0)
1405                 {
1406                         Log("SHUTTLE", "Process - ERROR: Forking failed");
1407                 }
1408                 else if (pid > 0)
1409                 {
1410                         // parent
1411                         Log("SHUTTLE", Form("Process - In parent process of %d - %s: Starting monitoring",
1412                                                         GetCurrentRun(), aDetector->GetName()));
1413
1414                         Long_t begin = time(0);
1415
1416                         int status; // to be used with waitpid, on purpose an int (not Int_t)!
1417                         while (waitpid(pid, &status, WNOHANG) == 0)
1418                         {
1419                                 Long_t expiredTime = time(0) - begin;
1420
1421                                 if (expiredTime > fConfig->GetPPTimeOut())
1422                                 {
1423                                         TString logMsg;
1424                                         AliShuttleStatus *currentStatus = ReadShuttleStatus();
1425                                         AliShuttleStatus::Status newStatus = AliShuttleStatus::kInvalid;
1426                                         
1427                                         if (currentStatus->GetStatus() <= AliShuttleStatus::kPPDone)
1428                                         {
1429                                                 // in case pp not yet done set status to kPPTimeOut
1430                                         
1431                                                 logMsg.Form("Process - Process of %s timed out. Run time: %d seconds. Killing...",
1432                                                                 fCurrentDetector.Data(), expiredTime);
1433                                                 newStatus = AliShuttleStatus::kPPTimeOut;
1434                                         }
1435                                         else if (currentStatus->GetStatus() == AliShuttleStatus::kStoreStarted)
1436                                         {
1437                                                 // in case the pp goes in TimeOut while storing the objects in the OCDB
1438                                                 // set status to kStoreError
1439                                                 
1440                                                 logMsg.Form("Process - Process of %s timed out while storing the OCDB object. Run time: %d seconds. Killing... and setting status to StoreError.",
1441                                                                 fCurrentDetector.Data(), expiredTime);
1442                                                 newStatus = AliShuttleStatus::kStoreError;
1443                                         }
1444                                         else 
1445                                         {
1446                                                 // in other cases don't change the status
1447                                                 
1448                                                 logMsg.Form("Process - Process of %s timed out in status = %s. Run time: %d seconds. Killing... without changing the status",
1449                                                                 fCurrentDetector.Data(), currentStatus->GetStatusName(), expiredTime);
1450                                         }
1451                                 
1452                                         Log("SHUTTLE", logMsg);
1453                                         Log(fCurrentDetector, logMsg);
1454
1455                                         kill(pid, 9);
1456
1457                                         if (newStatus != AliShuttleStatus::kInvalid)
1458                                                 UpdateShuttleStatus(newStatus);
1459                                         hasError = kTRUE;
1460
1461                                         gSystem->Sleep(1000);
1462                                 }
1463                                 else
1464                                 {
1465                                         gSystem->Sleep(1000);
1466                                         
1467                                         Int_t mem = GetMem(pid);
1468
1469                                         if (mem < 0)
1470                                                 continue;
1471                                                 
1472                                         mem -= initialMem;
1473                                         if (mem < 0)
1474                                                 mem = 0;
1475                                         
1476                                         if (expiredTime % 60 == 0)
1477                                         {
1478                                                 Log("SHUTTLE", Form("Process - %s: Checking process. "
1479                                                         "Run time: %d seconds - Memory consumption: %d KB",
1480                                                         fCurrentDetector.Data(), expiredTime, mem));
1481                                                 SendAlive();
1482                                         }
1483                                         
1484                                         if (mem > fConfig->GetPPMaxMem())
1485                                         {
1486                                                 TString tmp;
1487                                                 tmp.Form("Process - Process exceeds maximum allowed memory "
1488                                                         "(%d KB > %d KB). Killing...",
1489                                                         mem, fConfig->GetPPMaxMem());
1490                                                 Log("SHUTTLE", tmp);
1491                                                 Log(fCurrentDetector, tmp);
1492         
1493                                                 kill(pid, 9);
1494         
1495                                                 UpdateShuttleStatus(AliShuttleStatus::kPPOutOfMemory);
1496                                                 hasError = kTRUE;
1497         
1498                                                 gSystem->Sleep(1000);
1499                                         }
1500                                 }
1501                         }
1502
1503                         Log("SHUTTLE", Form("Process - In parent process of %d - %s: Client has terminated.",
1504                                                                 GetCurrentRun(), aDetector->GetName()));
1505
1506                         if (WIFEXITED(status))
1507                         {
1508                                 Int_t returnCode = WEXITSTATUS(status);
1509
1510                                 Log("SHUTTLE", Form("Process - %s: the return code is %d", fCurrentDetector.Data(),
1511                                                                                 returnCode));
1512
1513                                 if (returnCode == 0) hasError = kTRUE;
1514                         }
1515                 }
1516                 else if (pid == 0)
1517                 {
1518                         // child
1519                         Log("SHUTTLE", Form("Process - In child process of %d - %s", GetCurrentRun(),
1520                                 aDetector->GetName()));
1521
1522                         Log("SHUTTLE", Form("Process - Redirecting output to %s log",fCurrentDetector.Data()));
1523
1524                         if ((freopen(GetLogFileName(fCurrentDetector), "a", stdout)) == 0)
1525                         {
1526                                 Log("SHUTTLE", "Process - Could not freopen stdout");
1527                         }
1528                         else
1529                         {
1530                                 fOutputRedirected = kTRUE;
1531                                 if ((dup2(fileno(stdout), fileno(stderr))) < 0)
1532                                         Log("SHUTTLE", "Process - Could not redirect stderr");
1533                                 
1534                         }
1535                         
1536                         TString wd = gSystem->WorkingDirectory();
1537                         Int_t dir_lev1 = GetCurrentRun()/10000;
1538                         TString tmpDir = Form("%s/%d/%d/%s_process", GetShuttleTempDir(), 
1539                                 dir_lev1, GetCurrentRun(), fCurrentDetector.Data());
1540                         
1541                         Int_t result = gSystem->GetPathInfo(tmpDir.Data(), 0, (Long64_t*) 0, 0, 0);
1542                         if (!result) // temp dir already exists!
1543                         {
1544                                 Log(fCurrentDetector.Data(), 
1545                                         Form("Process - %s dir already exists! Removing...", tmpDir.Data()));
1546                                 gSystem->Exec(Form("rm -rf %s",tmpDir.Data()));         
1547                         } 
1548                         
1549                         if (gSystem->mkdir(tmpDir.Data(), 1))
1550                         {
1551                                 Log(fCurrentDetector.Data(), "Process - could not make temp directory!!");
1552                                 gSystem->Exit(1);
1553                         }
1554                         
1555                         if (!gSystem->ChangeDirectory(tmpDir.Data())) 
1556                         {
1557                                 Log(fCurrentDetector.Data(), "Process - could not change directory!!");
1558                                 gSystem->Exit(1);                       
1559                         }
1560                         
1561                         Int_t success = ProcessCurrentDetector();
1562
1563                         gSystem->ChangeDirectory(wd.Data());
1564                                                 
1565                         if (success == 1) // Preprocessor finished successfully!
1566                         { 
1567                                 // remove temporary folder or DCS map
1568                                 if (!fConfig->KeepTempFolder())
1569                                 {
1570                                         gSystem->Exec(Form("rm -rf %s",tmpDir.Data()));
1571                                 } else if (!fConfig->KeepDCSMap())
1572                                 {
1573                                         gSystem->Exec(Form("rm -f %s/DCSMap.root",tmpDir.Data()));
1574                                 }
1575                                 
1576                                 // Update time_processed field in FXS DB
1577                                 if (UpdateTable() == kFALSE)
1578                                         Log("SHUTTLE", Form("Process - %s: Could not update FXS databases!", 
1579                                                         fCurrentDetector.Data()));
1580
1581                                 // Transfer the data from local storage to main storage (Grid)
1582                                 if (StoreOCDB() == kFALSE)
1583                                         success = kFALSE;
1584                         } 
1585                         else if (success == 0)
1586                         {
1587                                 Log("SHUTTLE", 
1588                                         Form("\t\t\t****** run %d - %s: PP ERROR ******",
1589                                                 GetCurrentRun(), aDetector->GetName()));
1590                         }
1591
1592                         for (UInt_t iSys=0; iSys<3; iSys++)
1593                         {
1594                                 if (fFXSCalled[iSys]) fFXSlist[iSys].Clear();
1595                         }
1596
1597                         Log("SHUTTLE", Form("Process - Client process of %d - %s is exiting now with %d.",
1598                                                         GetCurrentRun(), aDetector->GetName(), success));
1599
1600                         // the client exits here
1601                         gSystem->Exit(success);
1602
1603                         AliError("We should never get here!!!");
1604                 }
1605         }
1606
1607         Log("SHUTTLE", Form("\t\t\t^*^*^*^*^*^*^*^*^*^*^*^* run %d: FINISH ^*^*^*^*^*^*^*^*^*^*^*^*",
1608                                                         GetCurrentRun()));
1609
1610         //check if shuttle is done for this run, if so update logbook
1611         TObjArray checkEntryArray;
1612         checkEntryArray.SetOwner(1);
1613         TString whereClause = Form("where run=%d", GetCurrentRun());
1614         if (!QueryShuttleLogbook(whereClause.Data(), checkEntryArray) || 
1615                         checkEntryArray.GetEntries() == 0) {
1616                 Log("SHUTTLE", Form("Process - Warning: Cannot check status of run %d on Shuttle logbook!",
1617                                                 GetCurrentRun()));
1618                 return hasError == kFALSE;
1619         }
1620
1621         AliShuttleLogbookEntry* checkEntry = dynamic_cast<AliShuttleLogbookEntry*>
1622                                                 (checkEntryArray.At(0));
1623
1624         if (checkEntry)
1625         {
1626                 if (checkEntry->IsDone())
1627                 {
1628                         Log("SHUTTLE","Process - Shuttle is DONE. Updating logbook");
1629                         UpdateShuttleLogbook("shuttle_done");
1630                 }
1631                 else
1632                 {
1633                         for (UInt_t iDet=0; iDet<NDetectors(); iDet++)
1634                         {
1635                                 if (checkEntry->GetDetectorStatus(iDet) == AliShuttleLogbookEntry::kUnprocessed)
1636                                 {
1637                                         AliDebug(2, Form("Run %d: setting %s as \"not first time unprocessed\"",
1638                                                         checkEntry->GetRun(), GetDetName(iDet)));
1639                                         fFirstUnprocessed[iDet] = kFALSE;
1640                                 }
1641                         }
1642                         SendMLRunInfo("Pending");
1643                 }
1644         }
1645
1646         fLogbookEntry = 0;
1647
1648         return hasError == kFALSE;
1649 }
1650
1651 //______________________________________________________________________________________________
1652 Int_t AliShuttle::ProcessCurrentDetector()
1653 {
1654         //
1655         // Makes data retrieval just for a specific detector (fCurrentDetector).
1656         // Threre should be a configuration for this detector.
1657
1658         Log("SHUTTLE", Form("ProcessCurrentDetector - Retrieving values for %s, run %d", 
1659                                                 fCurrentDetector.Data(), GetCurrentRun()));
1660
1661         TString wd = gSystem->WorkingDirectory();
1662         
1663         if (!CleanReferenceStorage(fCurrentDetector.Data()))
1664                 return 0;
1665         
1666         gSystem->ChangeDirectory(wd.Data());
1667         
1668         // call preprocessor
1669         AliPreprocessor* aPreprocessor =
1670                 dynamic_cast<AliPreprocessor*> (fPreprocessorMap.GetValue(fCurrentDetector));
1671
1672         // check if the preprocessor wants to process this run type
1673         if (aPreprocessor->ProcessRunType() == kFALSE)
1674         {
1675                 UpdateShuttleStatus(AliShuttleStatus::kSkipped);
1676                 UpdateShuttleLogbook(fCurrentDetector, "DONE");
1677                 if (!UpdateTableSkippedCase(fCurrentDetector.Data()))
1678                 {
1679                         AliError(Form("Could not update FXS tables for run %d !", GetCurrentRun()));
1680                 }
1681                 Log(fCurrentDetector, Form("ProcessCurrentDetector - %s preprocessor is not interested in this run type", fCurrentDetector.Data()));
1682         
1683                 return 2;
1684         }
1685         
1686         TMap* dcsMap = new TMap();
1687         
1688         aPreprocessor->Initialize(GetCurrentRun(), GetCurrentStartTime(), GetCurrentEndTime());
1689
1690         Bool_t processDCS = aPreprocessor->ProcessDCS();
1691
1692         if (!processDCS)
1693         {
1694                 Log(fCurrentDetector, "ProcessCurrentDetector -"
1695                         " The preprocessor requested to skip the retrieval of DCS values");
1696         }
1697         else if (fTestMode & kSkipDCS)
1698         {
1699                 Log(fCurrentDetector, "ProcessCurrentDetector - In TESTMODE: Skipping DCS processing");
1700         } 
1701         else if (fTestMode & kErrorDCS)
1702         {
1703                 Log(fCurrentDetector, "ProcessCurrentDetector - In TESTMODE: Simulating DCS error");
1704                 UpdateShuttleStatus(AliShuttleStatus::kDCSStarted);
1705                 UpdateShuttleStatus(AliShuttleStatus::kDCSError);
1706                 delete dcsMap;
1707                 return 0;
1708         } else {
1709
1710                 UpdateShuttleStatus(AliShuttleStatus::kDCSStarted);
1711
1712                 // Query DCS archive
1713                 Int_t nServers = fConfig->GetNServers(fCurrentDetector);
1714                 
1715                 for (int iServ=0; iServ<nServers; iServ++)
1716                 {
1717                 
1718                         TString host(fConfig->GetDCSHost(fCurrentDetector, iServ));
1719                         Int_t port = fConfig->GetDCSPort(fCurrentDetector, iServ);
1720                         Int_t multiSplit = fConfig->GetMultiSplit(fCurrentDetector, iServ);
1721
1722                         Log(fCurrentDetector, Form("ProcessCurrentDetector -"
1723                                         " Querying DCS Amanda server %s:%d (%d of %d)", 
1724                                         host.Data(), port, iServ+1, nServers));
1725                         
1726                         TMap* aliasMap = 0;
1727                         TMap* dpMap = 0;
1728
1729                         if (fConfig->GetDCSAliases(fCurrentDetector, iServ)->GetEntries() > 0)
1730                         {
1731                                 Log(fCurrentDetector, Form("Querying %d DCS aliases", fConfig->GetDCSAliases(fCurrentDetector, iServ)->GetEntries()));
1732                                 aliasMap = GetValueSet(host, port, 
1733                                                 fConfig->GetDCSAliases(fCurrentDetector, iServ), 
1734                                                 kAlias, multiSplit);
1735                                 if (!aliasMap)
1736                                 {
1737                                         Log(fCurrentDetector, 
1738                                                 Form("ProcessCurrentDetector -"
1739                                                         " Error retrieving DCS aliases from server %s."
1740                                                         " Sending mail to DCS experts!", host.Data()));
1741                                         UpdateShuttleStatus(AliShuttleStatus::kDCSError);
1742                                         
1743                                         if (!SendMail(kDCSEMail))
1744                                                 Log("SHUTTLE", Form("ProcessCurrentDetector - "
1745                                                                     "Could not send mail to DCS experts!"));
1746
1747                                         delete dcsMap;
1748                                         return 0;
1749                                 }
1750                         }
1751                         
1752                         if (fConfig->GetDCSDataPoints(fCurrentDetector, iServ)->GetEntries() > 0)
1753                         {
1754                                 Log(fCurrentDetector, Form("Querying %d DCS data points", fConfig->GetDCSDataPoints(fCurrentDetector, iServ)->GetEntries()));
1755                                 dpMap = GetValueSet(host, port, 
1756                                                 fConfig->GetDCSDataPoints(fCurrentDetector, iServ), 
1757                                                 kDP, multiSplit);
1758                                 if (!dpMap)
1759                                 {
1760                                         Log(fCurrentDetector, 
1761                                                 Form("ProcessCurrentDetector -"
1762                                                         " Error retrieving DCS data points from server %s."
1763                                                         " Sending mail to DCS experts!", host.Data()));
1764                                         UpdateShuttleStatus(AliShuttleStatus::kDCSError);
1765                                         
1766                                         if (!SendMail(kDCSEMail))
1767                                                 Log("SHUTTLE", Form("ProcessCurrentDetector - "
1768                                                                     "Could not send mail to DCS experts!"));
1769                                         
1770                                         if (aliasMap) delete aliasMap;
1771                                         delete dcsMap;
1772                                         return 0;
1773                                 }                               
1774                         }
1775                         
1776                         // merge aliasMap and dpMap into dcsMap
1777                         if(aliasMap) {
1778                                 TIter iter(aliasMap);
1779                                 TObjString* key = 0;
1780                                 while ((key = (TObjString*) iter.Next()))
1781                                         dcsMap->Add(key, aliasMap->GetValue(key->String()));
1782                                 
1783                                 aliasMap->SetOwner(kFALSE);
1784                                 delete aliasMap;
1785                         }       
1786                         
1787                         if(dpMap) {
1788                                 TIter iter(dpMap);
1789                                 TObjString* key = 0;
1790                                 while ((key = (TObjString*) iter.Next()))
1791                                         dcsMap->Add(key, dpMap->GetValue(key->String()));
1792                                 
1793                                 dpMap->SetOwner(kFALSE);
1794                                 delete dpMap;
1795                         }
1796                 }
1797         }
1798         
1799         // save map into file, to help debugging in case of preprocessor error
1800         TFile* f = TFile::Open("DCSMap.root","recreate");
1801         f->cd();
1802         dcsMap->Write("DCSMap", TObject::kSingleKey);
1803         f->Close();
1804         delete f;
1805         
1806         // DCS Archive DB processing successful. Call Preprocessor!
1807         UpdateShuttleStatus(AliShuttleStatus::kPPStarted);
1808
1809         fFXSError = -1; // this variable is kTRUE after ::Process if an FXS error occured
1810         
1811         UInt_t returnValue = aPreprocessor->Process(dcsMap);
1812         
1813         if (fFXSError!=-1) {
1814                 UpdateShuttleStatus(AliShuttleStatus::kFXSError);
1815                 SendMail(kFXSEMail, fFXSError);
1816                 dcsMap->DeleteAll();
1817                 delete dcsMap;
1818                 return 0;
1819         }
1820
1821         if (returnValue > 0) // Preprocessor error!
1822         {
1823                 Log(fCurrentDetector, Form("ProcessCurrentDetector - "
1824                                 "Preprocessor failed. Process returned %d.", returnValue));
1825                 UpdateShuttleStatus(AliShuttleStatus::kPPError);
1826                 dcsMap->DeleteAll();
1827                 delete dcsMap;
1828                 return 0;
1829         }
1830         
1831         // preprocessor ok!
1832         UpdateShuttleStatus(AliShuttleStatus::kPPDone);
1833         Log(fCurrentDetector, Form("ProcessCurrentDetector - %s preprocessor returned success",
1834                                 fCurrentDetector.Data()));
1835
1836         dcsMap->DeleteAll();
1837         delete dcsMap;
1838
1839         return 1;
1840 }
1841
1842 //______________________________________________________________________________________________
1843 void AliShuttle::CountOpenRuns()
1844 {
1845         // Query DAQ's Shuttle logbook and sends the number of open runs to ML
1846         
1847         SendAlive();
1848         
1849         // check connection, in case connect
1850         if (!Connect(3)) 
1851                 return;
1852
1853         TString sqlQuery;
1854         sqlQuery = Form("select count(*) from %s where shuttle_done=0", fConfig->GetShuttlelbTable());
1855         
1856         TSQLResult* aResult = fServer[3]->Query(sqlQuery);
1857         if (!aResult) {
1858                 AliError(Form("Can't execute query <%s>!", sqlQuery.Data()));
1859                 return;
1860         }
1861
1862         AliDebug(2,Form("Query = %s", sqlQuery.Data()));
1863         
1864         if (aResult->GetRowCount() == 0) {
1865                 AliError(Form("No result for query %s received", sqlQuery.Data()));
1866                 return;
1867         }
1868
1869         if (aResult->GetFieldCount() != 1) {
1870                 AliError(Form("Invalid field count for query %s received", sqlQuery.Data()));
1871                 return;
1872         }
1873
1874         TSQLRow* aRow = aResult->Next();
1875         if (!aRow) {
1876                 AliError(Form("Could not receive result of query %s", sqlQuery.Data()));
1877                 return;
1878         }
1879         
1880         TString result(aRow->GetField(0), aRow->GetFieldLength(0));
1881         Int_t count = result.Atoi();
1882         
1883         Log("SHUTTLE", Form("%d unprocessed runs", count));
1884         
1885         delete aRow;
1886         delete aResult;
1887
1888         TMonaLisaValue mlStatus("SHUTTLE_openruns", count);
1889
1890         TList mlList;
1891         mlList.Add(&mlStatus);
1892
1893         fMonaLisa->SendParameters(&mlList, "__PROCESSINGINFO__");
1894 }
1895
1896 //______________________________________________________________________________________________
1897 Bool_t AliShuttle::QueryShuttleLogbook(const char* whereClause,
1898                 TObjArray& entries)
1899 {
1900         // Query DAQ's Shuttle logbook and fills detector status object.
1901         // Call QueryRunParameters to query DAQ logbook for run parameters.
1902         //
1903
1904         entries.SetOwner(1);
1905
1906         // check connection, in case connect
1907         if (!Connect(3)) return kFALSE;
1908
1909         TString sqlQuery;
1910         sqlQuery = Form("select * from %s %s order by run", fConfig->GetShuttlelbTable(), whereClause);
1911
1912         TSQLResult* aResult = fServer[3]->Query(sqlQuery);
1913         if (!aResult) {
1914                 AliError(Form("Can't execute query <%s>!", sqlQuery.Data()));
1915                 return kFALSE;
1916         }
1917
1918         AliDebug(2,Form("Query = %s", sqlQuery.Data()));
1919
1920         if(aResult->GetRowCount() == 0) {
1921                 Log("SHUTTLE", "No entries in Shuttle Logbook match request");
1922                 delete aResult;
1923                 return kTRUE;
1924         }
1925
1926         // TODO Check field count!
1927         const UInt_t nCols = 25;
1928         if (aResult->GetFieldCount() != (Int_t) nCols) {
1929                 Log("SHUTTLE", "Invalid SQL result field number!");
1930                 delete aResult;
1931                 return kFALSE;
1932         }
1933
1934         TSQLRow* aRow;
1935         while ((aRow = aResult->Next())) {
1936                 TString runString(aRow->GetField(0), aRow->GetFieldLength(0));
1937                 Int_t run = runString.Atoi();
1938
1939                 AliShuttleLogbookEntry *entry = QueryRunParameters(run);
1940                 if (!entry)
1941                         continue;
1942
1943                 // DA test mode flag
1944                 TString daTestModeString(aRow->GetField(2), aRow->GetFieldLength(2)); // field 2 = DA test mode flag 
1945                 Bool_t daTestMode = (Bool_t)daTestModeString.Atoi();
1946                 entry->SetDATestMode(daTestMode);
1947
1948                 // loop on detectors
1949                 for(UInt_t ii = 0; ii < nCols; ii++)
1950                         entry->SetDetectorStatus(aResult->GetFieldName(ii), aRow->GetField(ii));
1951
1952                 entries.AddLast(entry);
1953                 delete aRow;
1954         }
1955
1956         delete aResult;
1957         return kTRUE;
1958 }
1959
1960 //______________________________________________________________________________________________
1961 AliShuttleLogbookEntry* AliShuttle::QueryRunParameters(Int_t run)
1962 {
1963         //
1964         // Retrieve run parameters written in the DAQ logbook and sets them into AliShuttleLogbookEntry object
1965         //
1966
1967         // check connection, in case connect
1968         if (!Connect(3))
1969                 return 0;
1970
1971         TString sqlQuery;
1972         sqlQuery.Form("select * from %s where run=%d", fConfig->GetDAQlbTable(), run);
1973
1974         TSQLResult* aResult = fServer[3]->Query(sqlQuery);
1975         if (!aResult) {
1976                 Log("SHUTTLE", Form("Can't execute query <%s>!", sqlQuery.Data()));
1977                 return 0;
1978         }
1979
1980         if (aResult->GetRowCount() == 0) {
1981                 Log("SHUTTLE", Form("QueryRunParameters - No entry in DAQ Logbook for run %d. Skipping", run));
1982                 delete aResult;
1983                 return 0;
1984         }
1985
1986         if (aResult->GetRowCount() > 1) {
1987                 Log("SHUTTLE", Form("QueryRunParameters - UNEXPECTED: "
1988                                 "more than one entry in DAQ Logbook for run %d!", run));
1989                 delete aResult;
1990                 return 0;
1991         }
1992
1993         TSQLRow* aRow = aResult->Next();
1994         if (!aRow)
1995         {
1996                 Log("SHUTTLE", Form("QueryRunParameters - Could not retrieve row for run %d. Skipping", run));
1997                 delete aResult;
1998                 return 0;
1999         }
2000
2001         AliShuttleLogbookEntry* entry = new AliShuttleLogbookEntry(run);
2002
2003         for (Int_t ii = 0; ii < aResult->GetFieldCount(); ii++)
2004                 entry->SetRunParameter(aResult->GetFieldName(ii), aRow->GetField(ii));
2005
2006         delete aRow;
2007         delete aResult;
2008         
2009         UInt_t startTime = entry->GetStartTime();
2010         UInt_t endTime = entry->GetEndTime();
2011         Bool_t ecsSuccess = entry->GetECSSuccess();
2012         
2013         TString totEventsStr = entry->GetRunParameter("totalEvents");  
2014         Int_t totEvents = totEventsStr.Atoi();
2015         
2016         UInt_t now = time(0);
2017         // TODO make this a configuration parameter
2018         Int_t dcsDelay = fConfig->GetDCSDelay()+fConfig->GetDCSQueryOffset();
2019         
2020         // runs are accepted if they have ecsSuccess set or more than 1 event
2021         if (startTime != 0 && endTime != 0 && endTime > startTime && (totEvents > 1 || ecsSuccess) && (endTime < now - dcsDelay))
2022         {
2023                 if (ecsSuccess == kFALSE)
2024                         Log("SHUTTLE", Form("Processing run %d although in status ECS failure, Reason: %s", run, entry->GetRunParameter("eor_reason")));
2025                 return entry;
2026         }
2027
2028         Bool_t skip = kFALSE;
2029                                 
2030         if (endTime != 0 && endTime >= now - dcsDelay)
2031         {
2032                 Log("SHUTTLE", Form("Skipping run %d for now, because DCS buffer time is not yet expired", run));
2033         }
2034         else if (totEvents <= 1) 
2035         {
2036                 Log("SHUTTLE", Form("QueryRunParameters - Run %d has 1 event or less - Skipping!", run));
2037                 skip = kTRUE;
2038         }
2039         else
2040         {
2041                 Log("SHUTTLE", Form("QueryRunParameters - Invalid parameters for Run %d: "
2042                         "startTime = %d, endTime = %d. Skipping (Shuttle won't be marked as DONE)!",
2043                         run, startTime, endTime));
2044         }
2045         
2046         if (skip)
2047         {
2048                 Log("SHUTTLE", Form("Marking SHUTTLE skipped for run %d", run));
2049                 fLogbookEntry = entry;
2050                 if (!UpdateShuttleLogbook("shuttle_skipped"))
2051                 {
2052                         AliError(Form("Could not update logbook for run %d !", run));
2053                 }
2054                 if (!UpdateTableSkippedCase("ALL"))
2055                 {
2056                         AliError(Form("Could not update FXS tables for run %d !", run));
2057                 }
2058                 fLogbookEntry = 0;
2059         }
2060                         
2061         delete entry;
2062         return 0;
2063 }
2064
2065 //______________________________________________________________________________________________
2066 TMap* AliShuttle::GetValueSet(const char* host, Int_t port, const TSeqCollection* entries,
2067                               DCSType type, Int_t multiSplit)
2068 {
2069         // Retrieve all "entry" data points from the DCS server
2070         // host, port: TSocket connection parameters
2071         // entries: list of name of the alias or data point
2072         // type: kAlias or kDP
2073         // returns TMap of values, 0 when failure
2074         
2075         AliDCSClient client(host, port, fTimeout, fRetries, multiSplit);
2076
2077         TMap* result = 0;
2078         if (type == kAlias)
2079         {
2080                 //result = client.GetAliasValues(entries, GetCurrentStartTime()-offset, 
2081                 //      GetCurrentEndTime()+offset);
2082                 result = client.GetAliasValues(entries, GetStartTimeDCSQuery(), 
2083                         GetEndTimeDCSQuery());
2084         } 
2085         else if (type == kDP)
2086         {
2087                 //result = client.GetDPValues(entries, GetCurrentStartTime()-offset, 
2088                 //      GetCurrentEndTime()+offset);
2089                 result = client.GetDPValues(entries, GetStartTimeDCSQuery(), 
2090                         GetEndTimeDCSQuery());
2091         }
2092
2093         if (result == 0)
2094         {
2095                 Log(fCurrentDetector.Data(), Form("GetValueSet - Can't get entries! Reason: %s",
2096                         client.GetErrorString(client.GetResultErrorCode())));
2097                 if (client.GetResultErrorCode() == AliDCSClient::fgkServerError)        
2098                         Log(fCurrentDetector.Data(), Form("GetValueSet - Server error code: %s",
2099                                 client.GetServerError().Data()));
2100
2101                 return 0;
2102         }
2103                 
2104         return result;
2105 }
2106
2107 //______________________________________________________________________________________________
2108 const char* AliShuttle::GetFile(Int_t system, const char* detector,
2109                 const char* id, const char* source)
2110 {
2111         // Get calibration file from file exchange servers
2112         // First queris the FXS database for the file name, using the run, detector, id and source info
2113         // then calls RetrieveFile(filename) for actual copy to local disk
2114         // run: current run being processed (given by Logbook entry fLogbookEntry)
2115         // detector: the Preprocessor name
2116         // id: provided as a parameter by the Preprocessor
2117         // source: provided by the Preprocessor through GetFileSources function
2118
2119         // check if test mode should simulate a FXS error
2120         if (fTestMode & kErrorFXSFiles)
2121         {
2122                 Log(detector, Form("GetFile - In TESTMODE - Simulating error while connecting to %s FXS", GetSystemName(system)));
2123                 return 0;
2124         }
2125         
2126         // check connection, in case connect
2127         if (!Connect(system))
2128         {
2129                 Log(detector, Form("GetFile - Couldn't connect to %s FXS database", GetSystemName(system)));
2130                 fFXSError = system;
2131                 return 0;
2132         }
2133
2134         // Query preparation
2135         TString sourceName(source);
2136         Int_t nFields = 3;
2137         TString sqlQueryStart = Form("select filePath,size,fileChecksum from %s where",
2138                                                                 fConfig->GetFXSdbTable(system));
2139         TString whereClause = Form("run=%d and detector=\"%s\" and fileId=\"%s\"",
2140                                                                 GetCurrentRun(), detector, id);
2141
2142         if (system == kDAQ)
2143         {
2144                 whereClause += Form(" and DAQsource=\"%s\"", source);
2145         }
2146         else if (system == kDCS)
2147         {
2148                 sourceName="none";
2149         }
2150         else if (system == kHLT)
2151         {
2152                 whereClause += Form(" and DDLnumbers=\"%s\"", source);
2153         }
2154
2155         TString sqlQuery = Form("%s %s", sqlQueryStart.Data(), whereClause.Data());
2156
2157         AliDebug(2, Form("SQL query: \n%s",sqlQuery.Data()));
2158
2159         // Query execution
2160         TSQLResult* aResult = 0;
2161         aResult = dynamic_cast<TSQLResult*> (fServer[system]->Query(sqlQuery));
2162         if (!aResult) {
2163                 Log(detector, Form("GetFile - Can't execute SQL query to %s database for: id = %s, source = %s",
2164                                 GetSystemName(system), id, sourceName.Data()));
2165                 fFXSError = system;
2166                 return 0;
2167         }
2168
2169         if (aResult->GetRowCount() == 0)
2170         {
2171                 Log(detector,
2172                         Form("GetFile - No entry in %s FXS db for: id = %s, source = %s",
2173                                 GetSystemName(system), id, sourceName.Data()));
2174                 delete aResult;
2175                 return 0;
2176         }
2177
2178         if (aResult->GetRowCount() > 1) {
2179                 Log(detector,
2180                         Form("GetFile - More than one entry in %s FXS db for: id = %s, source = %s",
2181                                 GetSystemName(system), id, sourceName.Data()));
2182                 fFXSError = system;
2183                 delete aResult;
2184                 return 0;
2185         }
2186
2187         if (aResult->GetFieldCount() != nFields) {
2188                 Log(detector,
2189                         Form("GetFileName - Wrong field count in %s FXS db for: id = %s, source = %s",
2190                                 GetSystemName(system), id, sourceName.Data()));
2191                 fFXSError = system;
2192                 delete aResult;
2193                 return 0;
2194         }
2195
2196         TSQLRow* aRow = dynamic_cast<TSQLRow*> (aResult->Next());
2197
2198         if (!aRow){
2199                 Log(detector, Form("GetFile - Empty set result in %s FXS db from query: id = %s, source = %s",
2200                                 GetSystemName(system), id, sourceName.Data()));
2201                 fFXSError = system;
2202                 delete aResult;
2203                 return 0;
2204         }
2205
2206         TString filePath(aRow->GetField(0), aRow->GetFieldLength(0));
2207         TString fileSize(aRow->GetField(1), aRow->GetFieldLength(1));
2208         TString fileChecksum(aRow->GetField(2), aRow->GetFieldLength(2));
2209
2210         delete aResult;
2211         delete aRow;
2212
2213         AliDebug(2, Form("filePath = %s; size = %s, fileChecksum = %s",
2214                                 filePath.Data(), fileSize.Data(), fileChecksum.Data()));
2215
2216         // retrieved file is renamed to make it unique
2217         Int_t dir_lev1 = GetCurrentRun()/10000;
2218         TString localFileName = Form("%s/%d/%d/%s_process/%s_%s_%d_%s_%s.shuttle",
2219                                      GetShuttleTempDir(), dir_lev1, GetCurrentRun(), detector,
2220                                         GetSystemName(system), detector, GetCurrentRun(), 
2221                                         id, sourceName.Data());
2222         Log("SHUTTLE",Form("file from FXS = %s",localFileName.Data())); 
2223
2224
2225         // file retrieval from FXS
2226         UInt_t nRetries = 0;
2227         UInt_t maxRetries = 3;
2228         Bool_t result = kFALSE;
2229
2230         // copy!! if successful TSystem::Exec returns 0
2231         while (nRetries++ < maxRetries) {
2232                 AliDebug(2, Form("Trying to copy file. Retry # %d", nRetries));
2233                 result = RetrieveFile(system, filePath.Data(), localFileName.Data());
2234                 if (!result)
2235                 {
2236                         Log(detector, Form("GetFile - Copy of file %s from %s FXS failed",
2237                                         filePath.Data(), GetSystemName(system)));
2238                         continue;
2239                 } 
2240
2241                 if (fileSize.Length()>0)
2242                 {
2243                         // compare filesize of local file with the one stored in the FXS DB
2244                         Long_t size = -1;
2245                         Int_t sizeComp = gSystem->GetPathInfo(localFileName.Data(), 0, &size, 0, 0);
2246
2247                         if (sizeComp != 0 || size != fileSize.Atoi())
2248                         {
2249                                 Log(detector, Form("GetFile - size of file %s does not match with local copy!",
2250                                                         filePath.Data()));
2251                                 result = kFALSE;
2252                                 continue;
2253                         }
2254
2255                 } else {
2256                         Log(fCurrentDetector, Form("GetFile - size of file %s not set in %s database, skipping comparison",
2257                                                 filePath.Data(), GetSystemName(system)));
2258                 }
2259
2260                 if (fileChecksum.Length()>0)
2261                 {
2262                         // compare md5sum of local file with the one stored in the FXS DB
2263                         if(fileChecksum.Contains(' ')) fileChecksum.Resize(fileChecksum.First(' '));
2264                         Int_t md5Comp = gSystem->Exec(Form("md5sum %s |grep %s > /dev/null 2> /dev/null",
2265                                                 localFileName.Data(), fileChecksum.Data()));
2266
2267                         if (md5Comp != 0)
2268                         {
2269                                 Log(detector, Form("GetFile - md5sum of file %s does not match with local copy!",
2270                                                         filePath.Data()));
2271                                 result = kFALSE;
2272                                 continue;
2273                         }
2274                 } else {
2275                         Log(fCurrentDetector, Form("GetFile - md5sum of file %s not set in %s database, skipping comparison",
2276                                                         filePath.Data(), GetSystemName(system)));
2277                 }
2278                 if (result) break;
2279         }
2280
2281         if (!result) 
2282         {
2283                 fFXSError = system;
2284                 return 0;
2285         }
2286
2287         fFXSCalled[system]=kTRUE;
2288         TObjString *fileParams = new TObjString(Form("%s#!?!#%s", id, sourceName.Data()));
2289         fFXSlist[system].Add(fileParams);
2290
2291         static TString staticLocalFileName;
2292         staticLocalFileName.Form("%s", localFileName.Data());
2293         
2294         Log(fCurrentDetector, Form("GetFile - Retrieved file with id %s and "
2295                         "source %s from %s to %s", id, source, 
2296                         GetSystemName(system), localFileName.Data()));
2297                         
2298         return staticLocalFileName.Data();
2299 }
2300
2301 //______________________________________________________________________________________________
2302 Bool_t AliShuttle::RetrieveFile(UInt_t system, const char* fxsFileName, const char* localFileName)
2303 {
2304         //
2305         // Copies file from FXS to local Shuttle machine
2306         //
2307
2308         // check temp directory: trying to cd to temp; if it does not exist, create it
2309         AliDebug(2, Form("Copy file %s from %s FXS into %s",
2310                         GetSystemName(system), fxsFileName, localFileName));
2311                         
2312         TString tmpDir(localFileName);
2313         
2314         tmpDir = tmpDir(0,tmpDir.Last('/'));
2315
2316         Int_t noDir = gSystem->GetPathInfo(tmpDir.Data(), 0, (Long64_t*) 0, 0, 0);
2317         if (noDir) // temp dir does not exists!
2318         {
2319                 if (gSystem->mkdir(tmpDir.Data(), 1))
2320                 {
2321                         Log(fCurrentDetector.Data(), "RetrieveFile - could not make temp directory!!");
2322                         return kFALSE;
2323                 }
2324         }
2325
2326         TString command = Form("scp -oPort=%d -2 %s@%s:%s/%s %s",
2327                 fConfig->GetFXSPort(system),
2328                 fConfig->GetFXSUser(system),
2329                 fConfig->GetFXSHost(system),
2330                 fConfig->GetFXSBaseFolder(system),
2331                 fxsFileName,
2332                 localFileName);
2333
2334         AliDebug(2, Form("%s",command.Data()));
2335
2336         Bool_t result = (gSystem->Exec(command.Data()) == 0);
2337
2338         return result;
2339 }
2340
2341 //______________________________________________________________________________________________
2342 TList* AliShuttle::GetFileSources(Int_t system, const char* detector, const char* id)
2343 {
2344         //
2345         // Get sources producing the condition file Id from file exchange servers
2346         // if id is NULL all sources are returned (distinct)
2347         //
2348
2349         if (id)
2350         {
2351                 Log(detector, Form("GetFileSources - Querying %s FXS for files with id %s produced by %s", GetSystemName(system), id, detector));
2352         } else {
2353                 Log(detector, Form("GetFileSources - Querying %s FXS for files produced by %s", GetSystemName(system), detector));
2354         }
2355         
2356         // check if test mode should simulate a FXS error
2357         if (fTestMode & kErrorFXSSources)
2358         {
2359                 Log(detector, Form("GetFileSources - In TESTMODE - Simulating error while connecting to %s FXS", GetSystemName(system)));
2360                 return 0;
2361         }
2362
2363         if (system == kDCS)
2364         {
2365                 Log(detector, "GetFileSources - WARNING: DCS system has only one source of data!");
2366                 TList *list = new TList();
2367                 list->SetOwner(1);
2368                 list->Add(new TObjString(" "));
2369                 return list;
2370         }
2371
2372         // check connection, in case connect
2373         if (!Connect(system))
2374         {
2375                 Log(detector, Form("GetFileSources - Couldn't connect to %s FXS database", GetSystemName(system)));
2376                 fFXSError = system;
2377                 return NULL;
2378         }
2379
2380         TString sourceName = "";
2381         if (system == kDAQ)
2382         {
2383                 sourceName = "DAQsource";
2384         } else if (system == kHLT)
2385         {
2386                 sourceName = "DDLnumbers";
2387         }
2388
2389         TString sqlQueryStart = Form("select distinct %s from %s where", sourceName.Data(), fConfig->GetFXSdbTable(system));
2390         TString whereClause = Form("run=%d and detector=\"%s\"",
2391                                 GetCurrentRun(), detector);
2392         if (id)
2393                 whereClause += Form(" and fileId=\"%s\"", id);
2394         TString sqlQuery = Form("%s %s", sqlQueryStart.Data(), whereClause.Data());
2395
2396         AliDebug(2, Form("SQL query: \n%s",sqlQuery.Data()));
2397
2398         // Query execution
2399         TSQLResult* aResult;
2400         aResult = fServer[system]->Query(sqlQuery);
2401         if (!aResult) {
2402                 Log(detector, Form("GetFileSources - Can't execute SQL query to %s database for id: %s",
2403                                 GetSystemName(system), id));
2404                 fFXSError = system;
2405                 return 0;
2406         }
2407
2408         TList *list = new TList();
2409         list->SetOwner(1);
2410         
2411         if (aResult->GetRowCount() == 0)
2412         {
2413                 Log(detector,
2414                         Form("GetFileSources - No entry in %s FXS table for id: %s", GetSystemName(system), id));
2415                 delete aResult;
2416                 return list;
2417         }
2418
2419         Log(detector, Form("GetFileSources - Found %d sources", aResult->GetRowCount()));
2420
2421         TSQLRow* aRow;
2422         while ((aRow = aResult->Next()))
2423         {
2424
2425                 TString source(aRow->GetField(0), aRow->GetFieldLength(0));
2426                 AliDebug(2, Form("%s = %s", sourceName.Data(), source.Data()));
2427                 list->Add(new TObjString(source));
2428                 delete aRow;
2429         }
2430
2431         delete aResult;
2432
2433         return list;
2434 }
2435
2436 //______________________________________________________________________________________________
2437 TList* AliShuttle::GetFileIDs(Int_t system, const char* detector, const char* source)
2438 {
2439         //
2440         // Get all ids of condition files produced by a given source from file exchange servers
2441         //
2442         
2443         Log(detector, Form("GetFileIDs - Retrieving ids with source %s with %s", source, GetSystemName(system)));
2444
2445         // check if test mode should simulate a FXS error
2446         if (fTestMode & kErrorFXSSources)
2447         {
2448                 Log(detector, Form("GetFileIDs - In TESTMODE - Simulating error while connecting to %s FXS", GetSystemName(system)));
2449                 return 0;
2450         }
2451
2452         // check connection, in case connect
2453         if (!Connect(system))
2454         {
2455                 Log(detector, Form("GetFileIDs - Couldn't connect to %s FXS database", GetSystemName(system)));
2456                 return NULL;
2457         }
2458
2459         TString sourceName = "";
2460         if (system == kDAQ)
2461         {
2462                 sourceName = "DAQsource";
2463         } else if (system == kHLT)
2464         {
2465                 sourceName = "DDLnumbers";
2466         }
2467
2468         TString sqlQueryStart = Form("select fileId from %s where", fConfig->GetFXSdbTable(system));
2469         TString whereClause = Form("run=%d and detector=\"%s\"",
2470                                 GetCurrentRun(), detector);
2471         if (sourceName.Length() > 0 && source)
2472                 whereClause += Form(" and %s=\"%s\"", sourceName.Data(), source);
2473         TString sqlQuery = Form("%s %s", sqlQueryStart.Data(), whereClause.Data());
2474
2475         AliDebug(2, Form("SQL query: \n%s",sqlQuery.Data()));
2476
2477         // Query execution
2478         TSQLResult* aResult;
2479         aResult = fServer[system]->Query(sqlQuery);
2480         if (!aResult) {
2481                 Log(detector, Form("GetFileIDs - Can't execute SQL query to %s database for source: %s",
2482                                 GetSystemName(system), source));
2483                 return 0;
2484         }
2485
2486         TList *list = new TList();
2487         list->SetOwner(1);
2488         
2489         if (aResult->GetRowCount() == 0)
2490         {
2491                 Log(detector,
2492                         Form("GetFileIDs - No entry in %s FXS table for source: %s", GetSystemName(system), source));
2493                 delete aResult;
2494                 return list;
2495         }
2496
2497         Log(detector, Form("GetFileIDs - Found %d ids", aResult->GetRowCount()));
2498
2499         TSQLRow* aRow;
2500
2501         while ((aRow = aResult->Next()))
2502         {
2503
2504                 TString id(aRow->GetField(0), aRow->GetFieldLength(0));
2505                 AliDebug(2, Form("fileId = %s", id.Data()));
2506                 list->Add(new TObjString(id));
2507                 delete aRow;
2508         }
2509
2510         delete aResult;
2511
2512         return list;
2513 }
2514
2515 //______________________________________________________________________________________________
2516 Bool_t AliShuttle::Connect(Int_t system)
2517 {
2518         // Connect to MySQL Server of the system's FXS MySQL databases
2519         // DAQ Logbook, Shuttle Logbook and DAQ FXS db are on the same host
2520         //
2521
2522         // check connection: if already connected return
2523         if(fServer[system] && fServer[system]->IsConnected()) {
2524                 // ping the server --> automatic reconnection should occur if it was broken but the
2525                 // server is still alive
2526                 fServer[system]->Ping();
2527                 return kTRUE;
2528         }
2529
2530         TString dbHost, dbUser, dbPass, dbName;
2531
2532         if (system < 3) // FXS db servers
2533         {
2534                 dbHost = Form("mysql://%s:%d", fConfig->GetFXSdbHost(system), fConfig->GetFXSdbPort(system));
2535                 dbUser = fConfig->GetFXSdbUser(system);
2536                 dbPass = fConfig->GetFXSdbPass(system);
2537                 dbName =   fConfig->GetFXSdbName(system);
2538         } else { // Run & Shuttle logbook servers
2539         // TODO Will the Shuttle logbook server be the same as the Run logbook server ???
2540                 dbHost = Form("mysql://%s:%d", fConfig->GetDAQlbHost(), fConfig->GetDAQlbPort());
2541                 dbUser = fConfig->GetDAQlbUser();
2542                 dbPass = fConfig->GetDAQlbPass();
2543                 dbName =   fConfig->GetDAQlbDB();
2544         }
2545
2546         fServer[system] = TSQLServer::Connect(dbHost.Data(), dbUser.Data(), dbPass.Data());
2547         if (!fServer[system] || !fServer[system]->IsConnected()) {
2548                 if(system < 3)
2549                 {
2550                 AliError(Form("Can't establish connection to FXS database for %s",
2551                                         AliShuttleInterface::GetSystemName(system)));
2552                 } else {
2553                 AliError("Can't establish connection to Run logbook.");
2554                 }
2555                 if(fServer[system]) delete fServer[system];
2556                 return kFALSE;
2557         }
2558
2559         // Get tables
2560         TSQLResult* aResult=0;
2561         switch(system){
2562                 case kDAQ:
2563                         aResult = fServer[kDAQ]->GetTables(dbName.Data());
2564                         break;
2565                 case kDCS:
2566                         aResult = fServer[kDCS]->GetTables(dbName.Data());
2567                         break;
2568                 case kHLT:
2569                         aResult = fServer[kHLT]->GetTables(dbName.Data());
2570                         break;
2571                 default:
2572                         aResult = fServer[3]->GetTables(dbName.Data());
2573                         break;
2574         }
2575
2576         delete aResult;
2577         return kTRUE;
2578 }
2579
2580 //______________________________________________________________________________________________
2581 Bool_t AliShuttle::UpdateTable()
2582 {
2583         //
2584         // Update FXS table filling time_processed field in all rows corresponding to current run and detector
2585         //
2586
2587         Bool_t result = kTRUE;
2588
2589         for (UInt_t system=0; system<3; system++)
2590         {
2591                 if(!fFXSCalled[system]) continue;
2592
2593                 // check connection, in case connect
2594                 if (!Connect(system))
2595                 {
2596                         Log(fCurrentDetector, Form("UpdateTable - Couldn't connect to %s FXS database", GetSystemName(system)));
2597                         result = kFALSE;
2598                         continue;
2599                 }
2600
2601                 TTimeStamp now; // now
2602
2603                 // Loop on FXS list entries
2604                 TIter iter(&fFXSlist[system]);
2605                 TObjString *aFXSentry=0;
2606                 while ((aFXSentry = dynamic_cast<TObjString*> (iter.Next())))
2607                 {
2608                         TString aFXSentrystr = aFXSentry->String();
2609                         TObjArray *aFXSarray = aFXSentrystr.Tokenize("#!?!#");
2610                         if (!aFXSarray || aFXSarray->GetEntries() != 2 )
2611                         {
2612                                 Log(fCurrentDetector, Form("UpdateTable - error updating %s FXS entry. Check string: <%s>",
2613                                         GetSystemName(system), aFXSentrystr.Data()));
2614                                 if(aFXSarray) delete aFXSarray;
2615                                 result = kFALSE;
2616                                 continue;
2617                         }
2618                         const char* fileId = ((TObjString*) aFXSarray->At(0))->GetName();
2619                         const char* source = ((TObjString*) aFXSarray->At(1))->GetName();
2620
2621                         TString whereClause;
2622                         if (system == kDAQ)
2623                         {
2624                                 whereClause = Form("where run=%d and detector=\"%s\" and fileId=\"%s\" and DAQsource=\"%s\";",
2625                                                         GetCurrentRun(), fCurrentDetector.Data(), fileId, source);
2626                         }
2627                         else if (system == kDCS)
2628                         {
2629                                 whereClause = Form("where run=%d and detector=\"%s\" and fileId=\"%s\";",
2630                                                         GetCurrentRun(), fCurrentDetector.Data(), fileId);
2631                         }
2632                         else if (system == kHLT)
2633                         {
2634                                 whereClause = Form("where run=%d and detector=\"%s\" and fileId=\"%s\" and DDLnumbers=\"%s\";",
2635                                                         GetCurrentRun(), fCurrentDetector.Data(), fileId, source);
2636                         }
2637
2638                         delete aFXSarray;
2639
2640                         TString sqlQuery = Form("update %s set time_processed=%d %s", fConfig->GetFXSdbTable(system),
2641                                                                 now.GetSec(), whereClause.Data());
2642
2643                         AliDebug(2, Form("SQL query: \n%s",sqlQuery.Data()));
2644
2645                         // Query execution
2646                         TSQLResult* aResult;
2647                         aResult = dynamic_cast<TSQLResult*> (fServer[system]->Query(sqlQuery));
2648                         if (!aResult)
2649                         {
2650                                 Log(fCurrentDetector, Form("UpdateTable - %s db: can't execute SQL query <%s>",
2651                                                                 GetSystemName(system), sqlQuery.Data()));
2652                                 result = kFALSE;
2653                                 continue;
2654                         }
2655                         delete aResult;
2656                 }
2657         }
2658
2659         return result;
2660 }
2661
2662 //_______________________________________________________________________________
2663 Bool_t AliShuttle::UpdateTableSkippedCase(const char* detector)
2664 {
2665         //
2666         // Update FXS table filling time_processed field in all rows corresponding to current run and detector
2667         // if detector = "ALL" update all detectors
2668         //
2669
2670         Bool_t result = kTRUE;
2671
2672         TString detName(detector);
2673
2674         for (UInt_t system=0; system<3; system++)
2675         {
2676
2677                 // check connection, in case connect
2678                 if (!Connect(system))
2679                 {
2680                         Log(fCurrentDetector, Form("UpdateTableSkippedCase - Couldn't connect to %s FXS database", GetSystemName(system)));
2681                         result = kFALSE;
2682                         continue;
2683                 }
2684
2685                 TTimeStamp now; // now
2686
2687                 // Loop on FXS list entries
2688                 TIter iter(&fFXSlist[system]);
2689                         
2690                 TString whereClause;
2691                 if (detName == "ALL") whereClause = Form("where run=%d and time_processed IS NULL;",GetCurrentRun());
2692                 else whereClause = Form("where run=%d and detector=\"%s\" and time_processed IS NULL;",GetCurrentRun(), detector);
2693
2694                 Log("SHUTTLE",Form(" whereClause = %s ",whereClause.Data()));
2695
2696                 TString sqlQuery = Form("update %s set time_processed=%d %s", fConfig->GetFXSdbTable(system),
2697                                         now.GetSec(), whereClause.Data());
2698
2699                 AliDebug(2, Form("SQL query: \n%s",sqlQuery.Data()));
2700
2701                 // Query execution
2702                 TSQLResult* aResult;
2703                 aResult = dynamic_cast<TSQLResult*> (fServer[system]->Query(sqlQuery));
2704                 if (!aResult)
2705                 {
2706                         Log("SHUTTLE", Form("UpdateTableSkippedCase - %s db: can't execute SQL query <%s>",
2707                                                         GetSystemName(system), sqlQuery.Data()));
2708                         result = kFALSE;
2709                         continue;
2710                 }
2711                 delete aResult;
2712                 
2713         }
2714
2715         return result;
2716 }
2717 //______________________________________________________________________________________________
2718 Bool_t AliShuttle::UpdateTableFailCase()
2719 {
2720         // Update FXS table filling time_processed field in all rows corresponding to current run and detector
2721         // this is called in case the preprocessor is declared failed for the current run, because
2722         // the fields are updated only in case of success
2723
2724         Bool_t result = kTRUE;
2725
2726         for (UInt_t system=0; system<3; system++)
2727         {
2728                 // check connection, in case connect
2729                 if (!Connect(system))
2730                 {
2731                         Log(fCurrentDetector, Form("UpdateTableFailCase - Couldn't connect to %s FXS database",
2732                                                         GetSystemName(system)));
2733                         result = kFALSE;
2734                         continue;
2735                 }
2736
2737                 TTimeStamp now; // now
2738
2739                 // Loop on FXS list entries
2740
2741                 TString whereClause = Form("where run=%d and detector=\"%s\";",
2742                                                 GetCurrentRun(), fCurrentDetector.Data());
2743
2744
2745                 TString sqlQuery = Form("update %s set time_processed=%d %s", fConfig->GetFXSdbTable(system),
2746                                                         now.GetSec(), whereClause.Data());
2747
2748                 AliDebug(2, Form("SQL query: \n%s",sqlQuery.Data()));
2749
2750                 // Query execution
2751                 TSQLResult* aResult;
2752                 aResult = dynamic_cast<TSQLResult*> (fServer[system]->Query(sqlQuery));
2753                 if (!aResult)
2754                 {
2755                         Log(fCurrentDetector, Form("UpdateTableFailCase - %s db: can't execute SQL query <%s>",
2756                                                         GetSystemName(system), sqlQuery.Data()));
2757                         result = kFALSE;
2758                         continue;
2759                 }
2760                 delete aResult;
2761         }
2762
2763         return result;
2764 }
2765
2766 //______________________________________________________________________________________________
2767 Bool_t AliShuttle::UpdateShuttleLogbook(const char* detector, const char* status)
2768 {
2769         //
2770         // Update Shuttle logbook filling detector or shuttle_done column
2771         // ex. of usage: UpdateShuttleLogbook("PHOS", "DONE") or UpdateShuttleLogbook("shuttle_done")
2772         //
2773
2774         // check connection, in case connect
2775         if(!Connect(3)){
2776                 Log("SHUTTLE", "UpdateShuttleLogbook - Couldn't connect to DAQ Logbook.");
2777                 return kFALSE;
2778         }
2779
2780         TString detName(detector);
2781         TString setClause;
2782         if (detName == "shuttle_done" || detName == "shuttle_skipped")
2783         {
2784                 setClause = "set shuttle_done=1";
2785                 
2786                 if (detName == "shuttle_done")
2787                 {
2788                         if (TouchFile() != kTRUE)
2789                         {
2790                                 SendMLRunInfo("Pending");
2791                                 return kFALSE;
2792                         }
2793                         
2794                         SendMLRunInfo("Done");
2795                 }
2796                 else 
2797                         SendMLRunInfo("Skipped");
2798         } 
2799         else {
2800                 TString statusStr(status);
2801                 if(statusStr.Contains("done", TString::kIgnoreCase) ||
2802                    statusStr.Contains("failed", TString::kIgnoreCase)){
2803                         setClause = Form("set %s=\"%s\"", detector, status);
2804                 } else {
2805                         Log("SHUTTLE",
2806                                 Form("UpdateShuttleLogbook - Invalid status <%s> for detector %s",
2807                                         status, detector));
2808                         return kFALSE;
2809                 }
2810         }
2811
2812         TString whereClause = Form("where run=%d", GetCurrentRun());
2813
2814         TString sqlQuery = Form("update %s %s %s",
2815                                         fConfig->GetShuttlelbTable(), setClause.Data(), whereClause.Data());
2816
2817         AliDebug(2, Form("SQL query: \n%s",sqlQuery.Data()));
2818
2819         // Query execution
2820         TSQLResult* aResult;
2821         aResult = dynamic_cast<TSQLResult*> (fServer[3]->Query(sqlQuery));
2822         if (!aResult) {
2823                 Log("SHUTTLE", Form("UpdateShuttleLogbook - Can't execute query <%s>", sqlQuery.Data()));
2824                 return kFALSE;
2825         }
2826         delete aResult;
2827
2828         return kTRUE;
2829 }
2830
2831 //______________________________________________________________________________________________
2832 Int_t AliShuttle::GetCurrentRun() const
2833 {
2834         //
2835         // Get current run from logbook entry
2836         //
2837
2838         return fLogbookEntry ? fLogbookEntry->GetRun() : -1;
2839 }
2840
2841 //______________________________________________________________________________________________
2842 UInt_t AliShuttle::GetCurrentStartTime() const
2843 {
2844         //
2845         // get current start time
2846         //
2847
2848         return fLogbookEntry ? fLogbookEntry->GetStartTime() : 0;
2849 }
2850
2851 //______________________________________________________________________________________________
2852 UInt_t AliShuttle::GetCurrentEndTime() const
2853 {
2854         //
2855         // get current end time from logbook entry
2856         //
2857
2858         return fLogbookEntry ? fLogbookEntry->GetEndTime() : 0;
2859 }
2860 //______________________________________________________________________________________________
2861 UInt_t AliShuttle::GetCurrentYear() const
2862 {
2863         //
2864         // Get current year from logbook entry
2865         //
2866
2867         if (!fLogbookEntry) return 0;
2868         
2869         TTimeStamp startTime(GetCurrentStartTime());
2870         TString year =  Form("%d",startTime.GetDate());
2871         year = year(0,4);
2872         
2873         return year.Atoi();
2874 }
2875
2876 //______________________________________________________________________________________________
2877 const char* AliShuttle::GetLHCPeriod() const
2878 {
2879         //
2880         // Get current LHC period from logbook entry
2881         //
2882
2883         if (!fLogbookEntry) return 0;
2884                 
2885         return fLogbookEntry->GetRunParameter("LHCperiod");
2886 }
2887
2888 //______________________________________________________________________________________________
2889 void AliShuttle::Log(const char* detector, const char* message)
2890 {
2891         //
2892         // Fill log string with a message
2893         //
2894         
2895         TString logRunDir = GetShuttleLogDir();
2896         if (GetCurrentRun() >=0) {
2897                 Int_t logDir_lev1 = GetCurrentRun()/10000;
2898                 logRunDir += Form("/%d/%d", logDir_lev1, GetCurrentRun());
2899         }               
2900         void* dir = gSystem->OpenDirectory(logRunDir.Data());
2901         if (dir == NULL) {
2902                 if (gSystem->mkdir(logRunDir.Data(), kTRUE)) {
2903                         AliError(Form("Can't open directory <%s>", GetShuttleLogDir()));
2904                         return;
2905                 }
2906
2907         } else {
2908                 gSystem->FreeDirectory(dir);
2909         }
2910
2911         TString toLog = Form("%s UTC (%d): %s - ", TTimeStamp(time(0)).AsString("s"), getpid(), detector);
2912         if (GetCurrentRun() >= 0) 
2913                 toLog += Form("run %d - ", GetCurrentRun());
2914         toLog += Form("%s", message);
2915
2916         AliInfo(toLog.Data());
2917         
2918         // if we redirect the log output already to the file, leave here
2919         if (fOutputRedirected && strcmp(detector, "SHUTTLE") != 0)
2920                 return;
2921
2922         TString fileName = GetLogFileName(detector);
2923         
2924         gSystem->ExpandPathName(fileName);
2925
2926         ofstream logFile;
2927         logFile.open(fileName, ofstream::out | ofstream::app);
2928
2929         if (!logFile.is_open()) {
2930                 AliError(Form("Could not open file %s", fileName.Data()));
2931                 return;
2932         }
2933
2934         logFile << toLog.Data() << "\n";
2935
2936         logFile.close();
2937 }
2938
2939 //______________________________________________________________________________________________
2940 TString AliShuttle::GetLogFileName(const char* detector) const
2941 {
2942         // 
2943         // returns the name of the log file for a given sub detector
2944         //
2945         
2946         TString fileName;
2947         
2948         if (GetCurrentRun() >= 0) 
2949         {
2950                 Int_t logDir_lev1 = GetCurrentRun()/10000;
2951                 fileName.Form("%s/%d/%d/%s.log", GetShuttleLogDir(), logDir_lev1, GetCurrentRun(), 
2952                         detector);
2953         } else {
2954                 fileName.Form("%s/%s.log", GetShuttleLogDir(), detector);
2955         }
2956
2957         return fileName;
2958 }
2959
2960 //______________________________________________________________________________________________
2961 void AliShuttle::SendAlive()
2962 {
2963         // sends alive message to ML
2964         
2965         TMonaLisaText mlStatus("SHUTTLE_status", "Alive");
2966
2967         TList mlList;
2968         mlList.Add(&mlStatus);
2969
2970         fMonaLisa->SendParameters(&mlList, "__PROCESSINGINFO__");
2971 }
2972
2973 //______________________________________________________________________________________________
2974 Bool_t AliShuttle::Collect(Int_t run)
2975 {
2976         //
2977         // Collects conditions data for all UNPROCESSED run written to DAQ LogBook in case of run = -1 (default)
2978         // If a dedicated run is given this run is processed
2979         //
2980         // In operational mode, this is the Shuttle function triggered by the EOR signal.
2981         //
2982
2983         if (run == -1)
2984                 Log("SHUTTLE","Collect - Shuttle called. Collecting conditions data for unprocessed runs");
2985         else
2986                 Log("SHUTTLE", Form("Collect - Shuttle called. Collecting conditions data for run %d", run));
2987
2988         SetLastAction("Starting");
2989
2990         // create ML instance
2991         if (!fMonaLisa)
2992                 fMonaLisa = new TMonaLisaWriter(fConfig->GetMonitorHost(), fConfig->GetMonitorTable());
2993                 
2994         CountOpenRuns();
2995
2996         TString whereClause("where shuttle_done=0");
2997         if (run != -1)
2998                 whereClause += Form(" and run=%d", run);
2999
3000         TObjArray shuttleLogbookEntries;
3001         if (!QueryShuttleLogbook(whereClause, shuttleLogbookEntries))
3002         {
3003                 Log("SHUTTLE", "Collect - Can't retrieve entries from Shuttle logbook");
3004                 return kFALSE;
3005         }
3006
3007         if (shuttleLogbookEntries.GetEntries() == 0)
3008         {
3009                 if (run == -1)
3010                         Log("SHUTTLE","Collect - Found no UNPROCESSED runs in Shuttle logbook");
3011                 else
3012                         Log("SHUTTLE", Form("Collect - Run %d is already DONE "
3013                                                 "or it does not exist in Shuttle logbook", run));
3014                 return kTRUE;
3015         }
3016
3017         for (UInt_t iDet=0; iDet<NDetectors(); iDet++)
3018                 fFirstUnprocessed[iDet] = kTRUE;
3019
3020         if (run != -1)
3021         {
3022                 // query Shuttle logbook for earlier runs, check if some detectors are unprocessed,
3023                 // flag them into fFirstUnprocessed array
3024                 TString whereClauseBis(Form("where shuttle_done=0 and run < %d", run));
3025                 TObjArray tmpLogbookEntries;
3026                 if (!QueryShuttleLogbook(whereClauseBis, tmpLogbookEntries))
3027                 {
3028                         Log("SHUTTLE", "Collect - Can't retrieve entries from Shuttle logbook");
3029                         return kFALSE;
3030                 }
3031
3032                 TIter iter(&tmpLogbookEntries);
3033                 AliShuttleLogbookEntry* anEntry = 0;
3034                 while ((anEntry = dynamic_cast<AliShuttleLogbookEntry*> (iter.Next())))
3035                 {
3036                         for (UInt_t iDet=0; iDet<NDetectors(); iDet++)
3037                         {
3038                                 if (anEntry->GetDetectorStatus(iDet) == AliShuttleLogbookEntry::kUnprocessed)
3039                                 {
3040                                         AliDebug(2, Form("Run %d: setting %s as \"not first time unprocessed\"",
3041                                                         anEntry->GetRun(), GetDetName(iDet)));
3042                                         fFirstUnprocessed[iDet] = kFALSE;
3043                                 }
3044                         }
3045
3046                 }
3047
3048         }
3049
3050         if (!RetrieveConditionsData(shuttleLogbookEntries))
3051         {
3052                 Log("SHUTTLE", "Collect - Process of at least one run failed");
3053                 CountOpenRuns();
3054                 return kFALSE;
3055         }
3056
3057         Log("SHUTTLE", "Collect - Requested run(s) successfully processed");
3058         CountOpenRuns();
3059         return kTRUE;
3060 }
3061
3062 //______________________________________________________________________________________________
3063 Bool_t AliShuttle::RetrieveConditionsData(const TObjArray& dateEntries)
3064 {
3065         //
3066         // Retrieve conditions data for all runs that aren't processed yet
3067         //
3068
3069         Bool_t hasError = kFALSE;
3070
3071         TIter iter(&dateEntries);
3072         AliShuttleLogbookEntry* anEntry;
3073
3074         while ((anEntry = (AliShuttleLogbookEntry*) iter.Next())){
3075                 if (!Process(anEntry)){
3076                         hasError = kTRUE;
3077                 }
3078
3079                 // clean SHUTTLE temp directory
3080                 //TString filename = Form("%s/*.shuttle", GetShuttleTempDir());
3081                 //RemoveFile(filename.Data());
3082         }
3083
3084         return hasError == kFALSE;
3085 }
3086
3087 //______________________________________________________________________________________________
3088 ULong_t AliShuttle::GetTimeOfLastAction() const
3089 {
3090         //
3091         // Gets time of last action
3092         //
3093
3094         ULong_t tmp;
3095
3096         fMonitoringMutex->Lock();
3097
3098         tmp = fLastActionTime;
3099
3100         fMonitoringMutex->UnLock();
3101
3102         return tmp;
3103 }
3104
3105 //______________________________________________________________________________________________
3106 const TString AliShuttle::GetLastAction() const
3107 {
3108         //
3109         // returns a string description of the last action
3110         //
3111
3112         TString tmp;
3113
3114         fMonitoringMutex->Lock();
3115         
3116         tmp = fLastAction;
3117         
3118         fMonitoringMutex->UnLock();
3119
3120         return tmp;
3121 }
3122
3123 //______________________________________________________________________________________________
3124 void AliShuttle::SetLastAction(const char* action)
3125 {
3126         //
3127         // updates the monitoring variables
3128         //
3129
3130         fMonitoringMutex->Lock();
3131
3132         fLastAction = action;
3133         fLastActionTime = time(0);
3134         
3135         fMonitoringMutex->UnLock();
3136 }
3137
3138 //______________________________________________________________________________________________
3139 const char* AliShuttle::GetRunParameter(const char* param)
3140 {
3141         //
3142         // returns run parameter read from DAQ logbook
3143         //
3144
3145         if(!fLogbookEntry) {
3146                 AliError("No logbook entry!");
3147                 return 0;
3148         }
3149
3150         return fLogbookEntry->GetRunParameter(param);
3151 }
3152
3153 //______________________________________________________________________________________________
3154 AliCDBEntry* AliShuttle::GetFromOCDB(const char* detector, const AliCDBPath& path)
3155 {
3156         //
3157         // returns object from OCDB valid for current run
3158         //
3159
3160         if (fTestMode & kErrorOCDB)
3161         {
3162                 Log(detector, "GetFromOCDB - In TESTMODE - Simulating error with OCDB");
3163                 return 0;
3164         }
3165         
3166         AliCDBStorage *sto = AliCDBManager::Instance()->GetStorage(fgkMainCDB);
3167         if (!sto)
3168         {
3169                 Log(detector, "GetFromOCDB - Cannot activate main OCDB for query!");
3170                 return 0;
3171         }
3172
3173         return dynamic_cast<AliCDBEntry*> (sto->Get(path, GetCurrentRun()));
3174 }
3175
3176 //______________________________________________________________________________________________
3177 Bool_t AliShuttle::SendMail(EMailTarget target, Int_t system)
3178 {
3179         //
3180         // sends a mail to the subdetector expert in case of preprocessor error
3181         //
3182         
3183         if (fTestMode != kNone)
3184                 return kTRUE;
3185                 
3186         if (!fConfig->SendMail()) 
3187                 return kTRUE;
3188
3189         if (target == kDCSEMail || target == kFXSEMail) {
3190                 if (!fFirstProcessing)
3191                         return kTRUE;
3192         }
3193
3194         Int_t runMode = (Int_t)fConfig->GetRunMode();
3195         TString tmpStr;
3196         if (runMode == 0) tmpStr = " Nightly Test:";
3197         else tmpStr = " Data Taking:"; 
3198         void* dir = gSystem->OpenDirectory(GetShuttleLogDir());
3199         if (dir == NULL)
3200         {
3201                 if (gSystem->mkdir(GetShuttleLogDir(), kTRUE))
3202                 {
3203                         Log("SHUTTLE", Form("SendMail - Can't open directory <%s>", GetShuttleLogDir()));
3204                         return kFALSE;
3205                 }
3206
3207         } else {
3208                 gSystem->FreeDirectory(dir);
3209         }
3210
3211         // det experts in to
3212         TString to="";
3213         TIter *iterExperts = 0;
3214         if (target == kDCSEMail) {
3215                 iterExperts = new TIter(fConfig->GetAdmins(AliShuttleConfig::kAmanda));
3216         }
3217         else if (target == kFXSEMail) {
3218                 iterExperts = new TIter(fConfig->GetAdmins(system));
3219         }
3220         if (iterExperts) {
3221                 TObjString *anExpert=0;
3222                 while ((anExpert = (TObjString*) iterExperts->Next()))
3223                 {
3224                         to += Form("%s,", anExpert->GetName());
3225                 }
3226                 delete iterExperts;
3227         }
3228
3229         // add subdetector experts      
3230         iterExperts = new TIter(fConfig->GetResponsibles(fCurrentDetector));
3231         TObjString *anExpert=0;
3232         while ((anExpert = (TObjString*) iterExperts->Next()))
3233         {
3234                 to += Form("%s,", anExpert->GetName());
3235         }
3236         delete iterExperts;
3237         
3238         if (to.Length() > 0)
3239           to.Remove(to.Length()-1);
3240         AliDebug(2, Form("to: %s",to.Data()));
3241
3242         if (to.IsNull()) {
3243                 Log("SHUTTLE", Form("List of %d responsibles not set!", (Int_t) target));
3244                 return kFALSE;
3245         }
3246
3247         // SHUTTLE responsibles in cc
3248         TString cc="";
3249         TIter iterAdmins(fConfig->GetAdmins(AliShuttleConfig::kGlobal));
3250         TObjString *anAdmin=0;
3251         while ((anAdmin = (TObjString*) iterAdmins.Next()))
3252         {
3253                 cc += Form("%s,", anAdmin->GetName());
3254         }
3255         if (cc.Length() > 0)
3256           cc.Remove(cc.Length()-1);
3257         AliDebug(2, Form("cc: %s",to.Data()));
3258
3259         // mail body 
3260         TString bodyFileName;
3261         bodyFileName.Form("%s/mail.body", GetShuttleLogDir());
3262         gSystem->ExpandPathName(bodyFileName);
3263
3264         ofstream mailBody;
3265         mailBody.open(bodyFileName, ofstream::out);
3266
3267         if (!mailBody.is_open())
3268         {
3269                 Log("SHUTTLE", Form("Could not open mail body file %s", bodyFileName.Data()));
3270                 return kFALSE;
3271         }
3272
3273
3274         TString subject;
3275         TString body;
3276
3277         if (target == kDCSEMail){
3278                 subject = Form("%s CRITICAL Retrieval of data points for %s FAILED in run %d !",
3279                                 tmpStr.Data(), fCurrentDetector.Data(), GetCurrentRun());
3280                 AliDebug(2, Form("subject: %s", subject.Data()));
3281                 
3282                 body = Form("Dear DCS experts, \n\n");
3283                 body += Form("SHUTTLE couldn\'t retrieve the data points for detector %s "
3284                              "in run %d!!\n\n", fCurrentDetector.Data(), GetCurrentRun());
3285         }
3286         else if (target == kFXSEMail){
3287                 subject = Form("%s CRITICAL FXS communication for %s FAILED in run %d !",
3288                                 tmpStr.Data(), fCurrentDetector.Data(), GetCurrentRun());
3289                 AliDebug(2, Form("subject: %s", subject.Data()));
3290                 TString sys;
3291                 if (system == kDAQ) sys="DAQ";
3292                 else if (system == kDCS) sys="DCS";
3293                 else if (system == kHLT) sys="HLT";
3294                 else return kFALSE;
3295                 body = Form("Dear  %s FXS experts, \n\n",sys.Data());
3296                 body += Form("SHUTTLE couldn\'t retrieve data from the FXS for detector %s "
3297                              "in run %d!!\n\n", fCurrentDetector.Data(), GetCurrentRun());
3298         }
3299         else {
3300                 subject = Form("%s %s Shuttle preprocessor FAILED in run %d (run type = %s)!",
3301                                        tmpStr.Data(), fCurrentDetector.Data(), GetCurrentRun(), GetRunType());
3302                 AliDebug(2, Form("subject: %s", subject.Data()));
3303         
3304                 body = Form("Dear %s expert(s), \n\n", fCurrentDetector.Data());
3305                 body += Form("SHUTTLE just detected that your preprocessor "
3306                              "failed processing run %d (run type = %s)!!\n\n", 
3307                              GetCurrentRun(), GetRunType());
3308         }
3309
3310         body += Form("Please check %s status on the SHUTTLE monitoring page: \n\n", 
3311                                 fCurrentDetector.Data());
3312         if (fConfig->GetRunMode() == AliShuttleConfig::kTest)
3313         {
3314                 body += Form("\thttp://pcalimonitor.cern.ch/shuttle.jsp?time=24 \n\n");
3315         } else {
3316                 body += Form("\thttp://pcalimonitor.cern.ch/shuttle.jsp?instance=PROD&time=24 \n\n");
3317         }
3318         
3319         
3320         TString logFolder = "logs";
3321         if (fConfig->GetRunMode() == AliShuttleConfig::kProd) 
3322                 logFolder += "_PROD";
3323         
3324         
3325         body += Form("Find the %s log for the current run on \n\n"
3326                 "\thttp://pcalishuttle01.cern.ch:8880/%s/%d/%d/%s.log \n\n", 
3327                      fCurrentDetector.Data(), logFolder.Data(), GetCurrentRun()/10000,  
3328                                 GetCurrentRun(), fCurrentDetector.Data());
3329         body += Form("The last 15 lines of %s log file are following:\n\n", fCurrentDetector.Data());
3330
3331         AliDebug(2, Form("Body begin: %s", body.Data()));
3332
3333         mailBody << body.Data();
3334         mailBody.close();
3335         mailBody.open(bodyFileName, ofstream::out | ofstream::app);
3336
3337         TString logFileName = Form("%s/%d/%d/%s.log", GetShuttleLogDir(), 
3338                 GetCurrentRun()/10000, GetCurrentRun(), fCurrentDetector.Data());
3339         TString tailCommand = Form("tail -n 15 %s >> %s", logFileName.Data(), bodyFileName.Data());
3340         if (gSystem->Exec(tailCommand.Data()))
3341         {
3342                 mailBody << Form("%s log file not found ...\n\n", fCurrentDetector.Data());
3343         }
3344
3345         TString endBody = Form("------------------------------------------------------\n\n");
3346         endBody += Form("In case of problems please contact the SHUTTLE core team.\n\n");
3347         endBody += "Please do not answer this message directly, it is automatically generated.\n\n";
3348         endBody += "Greetings,\n\n \t\t\tthe SHUTTLE\n";
3349
3350         AliDebug(2, Form("Body end: %s", endBody.Data()));
3351
3352         mailBody << endBody.Data();
3353
3354         mailBody.close();
3355
3356         // send mail!
3357         TString mailCommand = Form("mail -s \"%s\" -c %s %s < %s",
3358                                                 subject.Data(),
3359                                                 cc.Data(),
3360                                                 to.Data(),
3361                                                 bodyFileName.Data());
3362         AliDebug(2, Form("mail command: %s", mailCommand.Data()));
3363
3364         Bool_t result = gSystem->Exec(mailCommand.Data());
3365
3366         return result == 0;
3367 }
3368 //______________________________________________________________________________________________
3369 const char* AliShuttle::GetRunType()
3370 {
3371         //
3372         // returns run type read from "run type" logbook
3373         //
3374
3375         if(!fLogbookEntry) {
3376                 AliError("No logbook entry!");
3377                 return 0;
3378         }
3379
3380         return fLogbookEntry->GetRunType();
3381 }
3382
3383 //______________________________________________________________________________________________
3384 Bool_t AliShuttle::GetHLTStatus()
3385 {
3386         // Return HLT status (ON=1 OFF=0)
3387         // Converts the HLT status from the mode string read in the run logbook (not just a bool)
3388
3389         if(!fLogbookEntry) {
3390                 AliError("No logbook entry!");
3391                 return 0;
3392         }
3393
3394         // TODO implement when HLTMode is inserted in run logbook
3395         TString hltMode = fLogbookEntry->GetRunParameter("HLTmode");
3396         TSubString firstChar = hltMode(0,1);
3397         AliDebug(2,Form("First char = %s ",firstChar.Data())); 
3398         if (firstChar == "A") {
3399                 return kFALSE;
3400         }
3401         else if ((firstChar == "B") || (firstChar == "C") || (firstChar == "D") || (firstChar == "E")) {
3402                 return kTRUE;
3403         }
3404         else {
3405                 Log("SHUTTLE","Unexpected HLT mode! Returning 0....");
3406                 return kFALSE;
3407         }
3408 }
3409
3410 //______________________________________________________________________________________________
3411 const char* AliShuttle::GetTriggerConfiguration()
3412 {
3413         // Receives the trigger configuration from the DAQ logbook for the current run
3414         
3415         // check connection, if needed reconnect
3416         if (!Connect(3)) 
3417                 return 0;
3418
3419         TString sqlQuery;
3420         sqlQuery.Form("SELECT configFile FROM logbook_trigger_config WHERE run = %d", GetCurrentRun());
3421         TSQLResult* result = fServer[3]->Query(sqlQuery);
3422         if (!result)
3423         {
3424                 Log("SHUTTLE", Form("ERROR: Can't execute query <%s>!", sqlQuery.Data()));
3425                 return 0;
3426         }
3427         
3428         if (result->GetRowCount() == 0)
3429         {
3430                 Log("SHUTTLE", "ERROR: Trigger configuration not found in logbook_trigger_config");
3431                 delete result;
3432                 return 0;
3433         }
3434         
3435         TSQLRow* row = result->Next();
3436         if (!row)
3437         {
3438                 Log("SHUTTLE", "ERROR: Could not receive logbook_trigger_config data");
3439                 delete result;
3440                 return 0;
3441         }
3442
3443         // static, so that pointer remains valid when it is returned to the calling class       
3444         static TString triggerConfig(row->GetField(0));
3445         
3446         delete row;
3447         row = 0;
3448         
3449         delete result;
3450         result = 0;
3451         
3452         Log("SHUTTLE", Form("Found trigger configuration: %s", triggerConfig.Data()));
3453         
3454         return triggerConfig;
3455 }
3456
3457 //______________________________________________________________________________________________
3458 void AliShuttle::SetShuttleTempDir(const char* tmpDir)
3459 {
3460         //
3461         // sets Shuttle temp directory
3462         //
3463
3464         fgkShuttleTempDir = gSystem->ExpandPathName(tmpDir);
3465 }
3466
3467 //______________________________________________________________________________________________
3468 void AliShuttle::SetShuttleLogDir(const char* logDir)
3469 {
3470         //
3471         // sets Shuttle log directory
3472         //
3473
3474         fgkShuttleLogDir = gSystem->ExpandPathName(logDir);
3475 }
3476 //______________________________________________________________________________________________
3477 Bool_t AliShuttle::TouchFile()
3478 {
3479         //
3480         // touching a file on the grid if run has been DONE
3481         //
3482         
3483         if (!gGrid)
3484         {
3485                 Log("SHUTTLE",Form("No TGrid connection estabilished!"));
3486                 Log("SHUTTLE",Form("Could not touch file for run %i",GetCurrentRun()));
3487                 return kFALSE;
3488         }
3489
3490         TString dir;
3491         dir.Form("%s%d/SHUTTLE_DONE", fConfig->GetAlienPath(), GetCurrentYear());
3492         // checking whether directory for touch command exists
3493         TString commandLs;
3494         commandLs.Form("ls %s",dir.Data());
3495         TGridResult *resultLs = dynamic_cast<TGridResult*>(gGrid->Command(commandLs));
3496         if (!resultLs){
3497                 Log("SHUTTLE",Form("No result for %s command, returning without touching",commandLs.Data()));
3498                 return kFALSE;
3499         }
3500         TMap *mapLs = dynamic_cast<TMap*>(resultLs->At(0));
3501         if (!mapLs){
3502                 Log("SHUTTLE",Form("No map for %s command, returning without touching",commandLs.Data()));
3503                 delete resultLs;
3504                 resultLs = 0x0;
3505                 return kFALSE;
3506         }
3507         TObjString *valueLsPath = dynamic_cast<TObjString*>(mapLs->GetValue("path"));
3508         if (!valueLsPath || (valueLsPath->GetString()).CompareTo(dir)!=1){ 
3509                 Log("SHUTTLE",Form("No directory %s found, creating it",dir.Data()));
3510
3511                 // creating the directory
3512
3513                 Bool_t boolMkdir = gGrid->Mkdir(dir.Data());
3514                 if (!boolMkdir) {
3515                         Log("SHUTTLE",Form("Impossible to create dir %s in