New method Invert() for changing alpha by pi (forbiden operation via Rotate())
[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<5;iSys++) {
100                 fServer[iSys]=0;
101                 if (iSys < 4)
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<5;iSys++)
121                 if(fServer[iSys]) {
122                         fServer[iSys]->Close();
123                         delete fServer[iSys];
124                         fServer[iSys] = 0;
125                 }
126
127         if (fStatusEntry){
128                 delete fStatusEntry;
129                 fStatusEntry = 0;
130         }
131         
132         if (fMonitoringMutex) 
133         {
134                 delete fMonitoringMutex;
135                 fMonitoringMutex = 0;
136         }
137 }
138
139 //______________________________________________________________________________________________
140 void AliShuttle::RegisterPreprocessor(AliPreprocessor* preprocessor)
141 {
142         //
143         // Registers new AliPreprocessor.
144         // It uses GetName() for indentificator of the pre processor.
145         // The pre processor is registered it there isn't any other
146         // with the same identificator (GetName()).
147         //
148
149         const char* detName = preprocessor->GetName();
150         if(GetDetPos(detName) < 0)
151                 AliFatal(Form("********** !!!!! Invalid detector name: %s !!!!! **********", detName));
152
153         if (fPreprocessorMap.GetValue(detName)) {
154                 AliWarning(Form("AliPreprocessor %s is already registered!", detName));
155                 return;
156         }
157
158         fPreprocessorMap.Add(new TObjString(detName), preprocessor);
159 }
160 //______________________________________________________________________________________________
161 Bool_t AliShuttle::Store(const AliCDBPath& path, TObject* object,
162                 AliCDBMetaData* metaData, Int_t validityStart, Bool_t validityInfinite)
163 {
164         // Stores a CDB object in the storage for offline reconstruction. Objects that are not needed for
165         // offline reconstruction, but should be stored anyway (e.g. for debugging) should NOT be stored
166         // using this function. Use StoreReferenceData instead!
167         // It calls StoreLocally function which temporarily stores the data locally; when the preprocessor
168         // finishes the data are transferred to the main storage (Grid).
169
170         return StoreLocally(fgkLocalCDB, path, object, metaData, validityStart, validityInfinite);
171 }
172
173 //______________________________________________________________________________________________
174 Bool_t AliShuttle::StoreReferenceData(const AliCDBPath& path, TObject* object, AliCDBMetaData* metaData)
175 {
176         // Stores a CDB object in the storage for reference data. This objects will not be available during
177         // offline reconstrunction. Use this function for reference data only!
178         // It calls StoreLocally function which temporarily stores the data locally; when the preprocessor
179         // finishes the data are transferred to the main storage (Grid).
180
181         return StoreLocally(fgkLocalRefStorage, path, object, metaData);
182 }
183
184 //______________________________________________________________________________________________
185 Bool_t AliShuttle::StoreLocally(const TString& localUri,
186                         const AliCDBPath& path, TObject* object, AliCDBMetaData* metaData,
187                         Int_t validityStart, Bool_t validityInfinite)
188 {
189         // Store object temporarily in local storage. Parameters are passed by Store and StoreReferenceData functions.
190         // when the preprocessor finishes the data are transferred to the main storage (Grid).
191         // The parameters are:
192         //   1) Uri of the backup storage (Local)
193         //   2) the object's path.
194         //   3) the object to be stored
195         //   4) the metaData to be associated with the object
196         //   5) the validity start run number w.r.t. the current run,
197         //      if the data is valid only for this run leave the default 0
198         //   6) specifies if the calibration data is valid for infinity (this means until updated),
199         //      typical for calibration runs, the default is kFALSE
200         //
201         // returns 0 if fail, 1 otherwise
202
203
204         if (fTestMode & kErrorStorage)
205         {
206                 Log(fCurrentDetector, "StoreLocally - In TESTMODE - Simulating error while storing locally");
207                 return kFALSE;
208         }
209         
210         const char* cdbType = (localUri == fgkLocalCDB) ? "CDB" : "Reference";
211
212         Int_t firstRun = GetCurrentRun() - validityStart;
213         if(firstRun < 0) {
214                 AliWarning("First valid run happens to be less than 0! Setting it to 0.");
215                 firstRun=0;
216         }
217
218         Int_t lastRun = -1;
219         if(validityInfinite) {
220                 lastRun = AliCDBRunRange::Infinity();
221         } else {
222                 lastRun = GetCurrentRun();
223         }
224
225         // Version is set to current run, it will be used later to transfer data to Grid
226         AliCDBId id(path, firstRun, lastRun, GetCurrentRun(), -1);
227
228         if(! dynamic_cast<TObjString*> (metaData->GetProperty("RunUsed(TObjString)"))){
229                 TObjString runUsed = Form("%d", GetCurrentRun());
230                 metaData->SetProperty("RunUsed(TObjString)", runUsed.Clone());
231         }
232
233         Bool_t result = kFALSE;
234
235         if (!(AliCDBManager::Instance()->GetStorage(localUri))) {
236                 Log("SHUTTLE", Form("StoreLocally - Cannot activate local %s storage", cdbType));
237         } else {
238                 Int_t logLevel = AliLog::GetGlobalLogLevel();
239                 AliLog::SetGlobalLogLevel(AliLog::kError);
240                 result = AliCDBManager::Instance()->GetStorage(localUri)
241                                         ->Put(object, id, metaData);
242                 AliLog::SetGlobalLogLevel((AliLog::EType_t)logLevel);
243         }
244
245         if(!result) {
246
247                 Log(fCurrentDetector, Form("StoreLocally - Can't store object <%s>!", id.ToString().Data()));
248         }
249
250
251         return result;
252 }
253
254 //______________________________________________________________________________________________
255 Bool_t AliShuttle::StoreOCDB()
256 {
257         //
258         // Called when preprocessor ends successfully or when previous storage attempt failed (kStoreError status)
259         // Calls underlying StoreOCDB(const char*) function twice, for OCDB and Reference storage.
260         // Then calls StoreRefFilesToGrid to store reference files. 
261         //
262         
263         UpdateShuttleStatus(AliShuttleStatus::kStoreStarted);
264                         
265         if (fTestMode & kErrorGrid)
266         {
267                 Log("SHUTTLE", "StoreOCDB - In TESTMODE - Simulating error while storing in the Grid");
268                 Log(fCurrentDetector, "StoreOCDB - In TESTMODE - Simulating error while storing in the Grid");
269                 return kFALSE;
270         }
271         
272         Log("SHUTTLE","StoreOCDB - Storing OCDB data ...");
273         Int_t resultCDB = StoreOCDB(fgkMainCDB);
274
275         Log("SHUTTLE","StoreOCDB - Storing reference data ...");
276         Int_t resultRef = StoreOCDB(fgkMainRefStorage);
277         
278         Log("SHUTTLE","StoreOCDB - Storing reference files ...");
279         Bool_t resultRefFiles = CopyFilesToGrid("reference");
280         
281         Bool_t resultMetadata = kTRUE;
282         if(fCurrentDetector == "GRP") 
283         {
284                 Log("SHUTTLE","StoreOCDB - Storing Run Metadata file ...");
285                 resultMetadata = CopyFilesToGrid("metadata");
286         }
287         
288         Int_t storeResult = 0;
289         
290         if (resultCDB < 0 || resultRef < 0 || resultRefFiles == kFALSE || resultMetadata == kFALSE)
291                 storeResult = -1;
292         else if (resultCDB > 0 || resultRef > 0)
293                 storeResult = 1;
294                 
295         if (storeResult < 0)
296         {
297                 Log("SHUTTLE", 
298                         Form("\t\t\t****** run %d - %s: STORAGE ERROR ******",
299                                 GetCurrentRun(), fCurrentDetector.Data()));
300                 UpdateShuttleStatus(AliShuttleStatus::kStoreError);
301         } 
302         else if (storeResult > 0)
303         {
304                 Log("SHUTTLE", 
305                         Form("\t\t\t****** run %d - %s: STORAGE DELAYED ******",
306                                 GetCurrentRun(), fCurrentDetector.Data()));
307                 UpdateShuttleStatus(AliShuttleStatus::kStoreDelayed);
308         }
309         else if (storeResult == 0) 
310         {
311                 Log("SHUTTLE", 
312                         Form("\t\t\t****** run %d - %s: DONE ******",
313                                 GetCurrentRun(), fCurrentDetector.Data()));
314                 UpdateShuttleStatus(AliShuttleStatus::kDone);
315                 UpdateShuttleLogbook(fCurrentDetector, "DONE");
316         }
317
318         return (storeResult == 0);
319 }
320
321 //______________________________________________________________________________________________
322 Int_t AliShuttle::StoreOCDB(const TString& gridURI)
323 {
324         //
325         // Called by StoreOCDB(), performs actual storage to the main OCDB and reference storages (Grid)
326         //
327         // Return code:
328         //   -2 initialization error
329         //   -1 storage error
330         //   0  success
331         //   1  storage delayed (e.g. previous unprocessed runs)
332         //
333
334         TObjArray* gridIds=0;
335
336         Bool_t result = kTRUE;
337         Bool_t delayed = kFALSE;
338
339         const char* type = 0;
340         TString localURI;
341         if(gridURI == fgkMainCDB) {
342                 type = "OCDB";
343                 localURI = fgkLocalCDB;
344         } else if(gridURI == fgkMainRefStorage) {
345                 type = "reference";
346                 localURI = fgkLocalRefStorage;
347         } else {
348                 AliError(Form("Invalid storage URI: %s", gridURI.Data()));
349                 return -2;
350         }
351
352         AliCDBManager* man = AliCDBManager::Instance();
353
354         AliCDBStorage *gridSto = man->GetStorage(gridURI);
355         if(!gridSto) {
356                 Log("SHUTTLE",
357                         Form("StoreOCDB - cannot activate main %s storage", type));
358                 return -2;
359         }
360
361         gridIds = gridSto->GetQueryCDBList();
362
363         // get objects previously stored in local CDB
364         AliCDBStorage *localSto = man->GetStorage(localURI);
365         if(!localSto) {
366                 Log("SHUTTLE",
367                         Form("StoreOCDB - cannot activate local %s storage", type));
368                 return -2;
369         }
370         AliCDBPath aPath(GetOfflineDetName(fCurrentDetector.Data()),"*","*");
371         // Local objects were stored with current run as Grid version!
372         TList* localEntries = localSto->GetAll(aPath.GetPath(), GetCurrentRun(), GetCurrentRun());
373         localEntries->SetOwner(1);
374
375         // loop on local stored objects
376         TIter localIter(localEntries);
377         AliCDBEntry *aLocEntry = 0;
378         while((aLocEntry = dynamic_cast<AliCDBEntry*> (localIter.Next()))){
379                 aLocEntry->SetOwner(1);
380                 AliCDBId aLocId = aLocEntry->GetId();
381                 aLocEntry->SetVersion(-1);
382                 aLocEntry->SetSubVersion(-1);
383                                                 
384                 Log(fCurrentDetector.Data(), Form("Attempting to store %s", aLocId.ToString().Data()));
385
386                 // If local object is valid up to infinity we store it only if it is
387                 // the first unprocessed run!
388                 if (aLocId.GetLastRun() == AliCDBRunRange::Infinity() &&
389                         !fFirstUnprocessed[GetDetPos(fCurrentDetector)])
390                 {
391                         Log("SHUTTLE", Form("StoreOCDB - %s: object %s has validity infinite but "
392                                                 "there are previous unprocessed runs!",
393                                                 fCurrentDetector.Data(), aLocId.GetPath().Data()));
394                         Log(fCurrentDetector.Data(), Form("StoreOCDB - %s: object %s has validity infinite but "
395                                                 "there are previous unprocessed runs!",
396                                                 fCurrentDetector.Data(), aLocId.GetPath().Data()));
397                         delayed = kTRUE;
398                         continue;
399                 }
400
401                 // loop on Grid valid Id's
402                 Bool_t store = kTRUE;
403                 TIter gridIter(gridIds);
404                 AliCDBId* aGridId = 0;
405                 while ((aGridId = dynamic_cast<AliCDBId*> (gridIter.Next()))) {
406                         if (aGridId->GetPath() != aLocId.GetPath()) 
407                                 continue;
408                         // skip all objects valid up to infinity
409                         if (aGridId->GetLastRun() == AliCDBRunRange::Infinity()) 
410                                 continue;
411                         
412                         // if we get here, it means there's already some more recent object stored on Grid!
413                         Log(fCurrentDetector.Data(),
414                                 Form("StoreOCDB - A more recent object already exists in %s storage: <%s>",
415                                 type, aGridId->ToString().Data()));
416                         
417                         store = kFALSE;
418                         break;
419                 }
420
421                 Bool_t storeOk = kFALSE;
422                 if (store)
423                 {
424                         Log(fCurrentDetector.Data(), Form("Prechecks succeeded. Ready to store %s", aLocId.ToString().Data()));
425                         storeOk = gridSto->Put(aLocEntry);
426                         if (storeOk) {
427                                 Log("SHUTTLE",
428                                 Form("StoreOCDB - Object <%s> successfully put into %s storage",
429                                         aLocId.ToString().Data(), type));
430                                 Log(fCurrentDetector.Data(),
431                                         Form("StoreOCDB - Object <%s> successfully put into %s storage",
432                                         aLocId.ToString().Data(), type));
433                         } else  {
434                                 Log("SHUTTLE",
435                                         Form("StoreOCDB - Grid %s storage of object <%s> failed",
436                                         type, aLocId.ToString().Data()));
437                                 Log(fCurrentDetector.Data(),
438                                         Form("StoreOCDB - Grid %s storage of object <%s> failed",
439                                         type, aLocId.ToString().Data()));
440                                 result = kFALSE;
441                         }
442                 }
443                 
444                 if (!store || storeOk) {
445                         // removing local file...
446                         TString filename;
447                         localSto->IdToFilename(aLocId, filename);
448                         Log("SHUTTLE", Form("StoreOCDB - Removing local file %s", filename.Data()));
449                         RemoveFile(filename.Data());
450                 }
451         }
452         localEntries->Clear();
453
454         Int_t returnCode = 0;
455         
456         if (result == kFALSE)
457                 returnCode = -1;
458         else if (delayed != kFALSE)
459                 returnCode =  1;
460
461         Log("SHUTTLE", Form("StoreOCDB - Returning with %d (result = %d, delayed = %d)", returnCode, result, delayed));
462         Log(fCurrentDetector.Data(), Form("StoreOCDB - Returning with %d (result = %d, delayed = %d)", returnCode, result, delayed));
463         
464         return returnCode;
465 }
466
467 //______________________________________________________________________________________________
468 Bool_t AliShuttle::CleanReferenceStorage(const char* detector)
469 {
470         // clears the directory used to store reference files of a given subdetector
471   
472         AliCDBManager* man = AliCDBManager::Instance();
473         AliCDBStorage* sto = man->GetStorage(fgkLocalRefStorage);
474         TString localBaseFolder = sto->GetBaseFolder();
475
476         TString targetDir = GetRefFilePrefix(localBaseFolder.Data(), detector);
477         
478         Log("SHUTTLE", Form("CleanReferenceStorage - Cleaning %s", targetDir.Data()));
479
480         TString begin;
481         begin.Form("%d_", GetCurrentRun());
482         
483         TSystemDirectory* baseDir = new TSystemDirectory("/", targetDir);
484         if (!baseDir)
485                 return kTRUE;
486                 
487         TList* dirList = baseDir->GetListOfFiles();
488         delete baseDir;
489         
490         if (!dirList) return kTRUE;
491                         
492         if (dirList->GetEntries() < 3) // to be changed to 4?
493         {
494                 delete dirList;
495                 return kTRUE;
496         }
497                                 
498         Int_t nDirs = 0, nDel = 0;
499         TIter dirIter(dirList);
500         TSystemFile* entry = 0;
501
502         Bool_t success = kTRUE;
503         
504         while ((entry = dynamic_cast<TSystemFile*> (dirIter.Next())))
505         {                                       
506                 if (entry->IsDirectory())
507                         continue;
508                 
509                 TString fileName(entry->GetName());
510                 if (!fileName.BeginsWith(begin))
511                         continue;
512                         
513                 nDirs++;
514                                                 
515                 // delete file
516                 Int_t result = gSystem->Unlink(fileName.Data());
517                 
518                 if (result)
519                 {
520                         Log("SHUTTLE", Form("CleanReferenceStorage - Could not delete file %s!", fileName.Data()));
521                         success = kFALSE;
522                 } else {
523                         nDel++;
524                 }
525         }
526
527         if(nDirs > 0)
528                 Log("SHUTTLE", Form("CleanReferenceStorage - %d (over %d) reference files in folder %s were deleted.", 
529                         nDel, nDirs, targetDir.Data()));
530
531                 
532         delete dirList;
533         return success;
534
535
536
537
538
539
540   Int_t result = gSystem->GetPathInfo(targetDir, 0, (Long64_t*) 0, 0, 0);
541   if (result == 0)
542   {
543     // delete directory
544     result = gSystem->Exec(Form("rm -rf %s", targetDir.Data()));
545     if (result != 0)
546     {  
547       Log("SHUTTLE", Form("CleanReferenceStorage - Could not clean directory %s", targetDir.Data()));
548       return kFALSE;
549     }
550   }
551
552   result = gSystem->mkdir(targetDir, kTRUE);
553   if (result != 0)
554   {
555     Log("SHUTTLE", Form("CleanReferenceStorage - Error creating base directory %s", targetDir.Data()));
556     return kFALSE;
557   }
558         
559   return kTRUE;
560 }
561
562 //______________________________________________________________________________________________
563 Bool_t AliShuttle::StoreReferenceFile(const char* detector, const char* localFile, const char* gridFileName)
564 {
565         //
566         // Stores reference file directly (without opening it). This function stores the file locally.
567         //
568         // The file is stored under the following location: 
569         // <base folder of local reference storage>/<DET>/<RUN#>_<gridFileName>
570         // where <gridFileName> is the second parameter given to the function
571         // 
572         
573         if (fTestMode & kErrorStorage)
574         {
575                 Log(fCurrentDetector, "StoreReferenceFile - In TESTMODE - Simulating error while storing locally");
576                 return kFALSE;
577         }
578         
579         AliCDBManager* man = AliCDBManager::Instance();
580         AliCDBStorage* sto = man->GetStorage(fgkLocalRefStorage);
581         
582         TString localBaseFolder = sto->GetBaseFolder();
583         
584         TString target = GetRefFilePrefix(localBaseFolder.Data(), detector);    
585         target.Append(Form("/%d_%s", GetCurrentRun(), gridFileName));
586         
587         return CopyFileLocally(localFile, target);
588 }
589
590 //______________________________________________________________________________________________
591 Bool_t AliShuttle::StoreRunMetadataFile(const char* localFile, const char* gridFileName)
592 {
593         //
594         // Stores Run metadata file to the Grid, in the run folder
595         //
596         // Only GRP can call this function.
597         
598         if (fTestMode & kErrorStorage)
599         {
600                 Log(fCurrentDetector, "StoreRunMetaDataFile - In TESTMODE - Simulating error while storing locally");
601                 return kFALSE;
602         }
603         
604         AliCDBManager* man = AliCDBManager::Instance();
605         AliCDBStorage* sto = man->GetStorage(fgkLocalRefStorage);
606         
607         TString localBaseFolder = sto->GetBaseFolder();
608         
609         // Build Run level folder
610         // folder = /alice/data/year/lhcPeriod/runNb/raw
611         
612                 
613         TString lhcPeriod = GetLHCPeriod();     
614         if (lhcPeriod.Length() == 0) 
615         {
616                 Log("SHUTTLE","StoreRunMetaDataFile - LHCPeriod not found in logbook!");
617                 return 0;
618         }
619         
620         // TODO partitions with one detector only write data into LHCperiod_DET
621         TString partition = GetRunParameter("detector");
622         
623         if (partition.Length() > 0 && partition != "ALICE")
624         {
625                 lhcPeriod.Append(Form("_%s", partition.Data()));
626                 Log(fCurrentDetector, Form("Run data tags merged file will be written in %s", 
627                                 lhcPeriod.Data()));
628         }
629                 
630         TString target = Form("%s/GRP/RunMetadata%s%d/%s/%09d/raw/%s", 
631                                 localBaseFolder.Data(), fConfig->GetAlienPath(), GetCurrentYear(), 
632                                 lhcPeriod.Data(), GetCurrentRun(), gridFileName);
633                                         
634         return CopyFileLocally(localFile, target);
635 }
636
637 //______________________________________________________________________________________________
638 Bool_t AliShuttle::CopyFileLocally(const char* localFile, const TString& target)
639 {
640         //
641         // Stores file locally. Called by StoreReferenceFile and StoreRunMetadataFile
642         // Files are temporarily stored in the local reference storage. When the preprocessor 
643         // finishes, the Shuttle calls CopyFilesToGrid to transfer the files to AliEn 
644         // (in reference or run level folders)
645         //
646         
647         TString targetDir(target(0, target.Last('/')));
648         
649         //try to open base dir folder, if it does not exist
650         void* dir = gSystem->OpenDirectory(targetDir.Data());
651         if (dir == NULL) {
652                 if (gSystem->mkdir(targetDir.Data(), kTRUE)) {
653                         Log("SHUTTLE", Form("CopyFileLocally - Can't open directory <%s>", targetDir.Data()));
654                         return kFALSE;
655                 }
656
657         } else {
658                 gSystem->FreeDirectory(dir);
659         }
660         
661         Int_t result = 0;
662         
663         result = gSystem->GetPathInfo(localFile, 0, (Long64_t*) 0, 0, 0);
664         if (result)
665         {
666                 Log("SHUTTLE", Form("CopyFileLocally - %s does not exist", localFile));
667                 return kFALSE;
668         }
669
670         result = gSystem->GetPathInfo(target, 0, (Long64_t*) 0, 0, 0);
671         if (!result)
672         {
673                 Log("SHUTTLE", Form("CopyFileLocally - target file %s already exist, removing...", target.Data()));
674                 if (gSystem->Unlink(target.Data()))
675                 {
676                         Log("SHUTTLE", Form("CopyFileLocally - Could not remove existing target file %s!", target.Data()));
677                         return kFALSE;
678                 }
679         }       
680         
681         result = gSystem->CopyFile(localFile, target);
682
683         if (result == 0)
684         {
685                 Log("SHUTTLE", Form("CopyFileLocally - File %s stored locally to %s", localFile, target.Data()));
686                 return kTRUE;
687         }
688         else
689         {
690                 Log("SHUTTLE", Form("CopyFileLocally - Could not store file %s to %s! Error code = %d", 
691                                 localFile, target.Data(), result));
692                 return kFALSE;
693         }       
694
695
696
697 }
698
699 //______________________________________________________________________________________________
700 Bool_t AliShuttle::CopyFilesToGrid(const char* type)
701 {
702         //
703         // Transfers local files to the Grid. Local files can be reference files 
704         // or run metadata file (from GRP only).
705         //
706         // According to the type (ref, metadata) the files are stored under the following location: 
707         // ref --> <base folder of reference storage>/<DET>/<RUN#>_<gridFileName>
708         // metadata --> <run data folder>/<MetadataFileName>
709         //
710                 
711         AliCDBManager* man = AliCDBManager::Instance();
712         AliCDBStorage* sto = man->GetStorage(fgkLocalRefStorage);
713         if (!sto)
714                 return kFALSE;
715         TString localBaseFolder = sto->GetBaseFolder();
716         
717         TString dir;
718         TString alienDir;
719         TString begin;
720         
721         if (strcmp(type, "reference") == 0) 
722         {
723                 dir = GetRefFilePrefix(localBaseFolder.Data(), fCurrentDetector.Data());
724                 AliCDBStorage* gridSto = man->GetStorage(fgkMainRefStorage);
725                 if (!gridSto)
726                         return kFALSE;
727                 TString gridBaseFolder = gridSto->GetBaseFolder();
728                 alienDir = GetRefFilePrefix(gridBaseFolder.Data(), fCurrentDetector.Data());
729                 begin = Form("%d_", GetCurrentRun());
730         } 
731         else if (strcmp(type, "metadata") == 0)
732         {
733                         
734                 TString lhcPeriod = GetLHCPeriod();
735         
736                 if (lhcPeriod.Length() == 0) 
737                 {
738                         Log("SHUTTLE","CopyFilesToGrid - LHCPeriod not found in logbook!");
739                         return 0;
740                 }
741                 
742                 // TODO partitions with one detector only write data into LHCperiod_DET
743                 TString partition = GetRunParameter("detector");
744         
745                 if (partition.Length() > 0 && partition != "ALICE")
746                 {
747                         lhcPeriod.Append(Form("_%s", partition.Data()));
748                 }
749                 
750                 dir = Form("%s/GRP/RunMetadata%s%d/%s/%09d/raw", 
751                                 localBaseFolder.Data(), fConfig->GetAlienPath(), GetCurrentYear(), 
752                                 lhcPeriod.Data(), GetCurrentRun());
753                 alienDir = dir(dir.Index(fConfig->GetAlienPath()), dir.Length());
754                 
755                 begin = "";
756         }
757         else 
758         {
759                 Log("SHUTTLE", "CopyFilesToGrid - Unexpected: type label must be reference or metadata!");
760                 return kFALSE;
761         }
762                 
763         TSystemDirectory* baseDir = new TSystemDirectory("/", dir);
764         if (!baseDir)
765                 return kTRUE;
766                 
767         TList* dirList = baseDir->GetListOfFiles();
768         delete baseDir;
769         
770         if (!dirList) return kTRUE;
771                 
772         if (dirList->GetEntries() < 3) 
773         {
774                 delete dirList;
775                 return kTRUE;
776         }
777                         
778         if (!gGrid)
779         { 
780                 Log("SHUTTLE", "CopyFilesToGrid - Connection to Grid failed: Cannot continue!");
781                 delete dirList;
782                 return kFALSE;
783         }
784         
785         Int_t nDirs = 0, nTransfer = 0;
786         TIter dirIter(dirList);
787         TSystemFile* entry = 0;
788
789         Bool_t success = kTRUE;
790         Bool_t first = kTRUE;
791         
792         while ((entry = dynamic_cast<TSystemFile*> (dirIter.Next())))
793         {                       
794                 if (entry->IsDirectory())
795                         continue;
796                         
797                 TString fileName(entry->GetName());
798                 if (!fileName.BeginsWith(begin))
799                         continue;
800                         
801                 nDirs++;
802                         
803                 if (first)
804                 {
805                         first = kFALSE;
806                         // check that folder exists, otherwise create it
807                         TGridResult* result = gGrid->Ls(alienDir.Data(), "a");
808                         
809                         if (!result)
810                         {
811                                 delete dirList;
812                                 return kFALSE;
813                         }
814                         
815                         if (!result->GetFileName(1)) // TODO: It looks like element 0 is always 0!!
816                         {
817                                 // TODO It does not work currently! Bug in TAliEn::Mkdir
818                                 // TODO Manually fixed in local root v5-16-00
819                                 if (!gGrid->Mkdir(alienDir.Data(),"-p",0))
820                                 {
821                                         Log("SHUTTLE", Form("CopyFilesToGrid - Cannot create directory %s",
822                                                         alienDir.Data()));
823                                         delete dirList;
824                                         return kFALSE;
825                                 } else {
826                                         Log("SHUTTLE",Form("CopyFilesToGrid - Folder %s created", alienDir.Data()));
827                                 }
828                                 
829                         } else {
830                                         Log("SHUTTLE",Form("CopyFilesToGrid - Folder %s found", alienDir.Data()));
831                         }
832                 }
833                         
834                 TString fullLocalPath;
835                 fullLocalPath.Form("%s/%s", dir.Data(), fileName.Data());
836                 
837                 TString fullGridPath;
838                 fullGridPath.Form("alien://%s/%s", alienDir.Data(), fileName.Data());
839
840                 Bool_t result = TFile::Cp(fullLocalPath, fullGridPath);
841                 
842                 if (result)
843                 {
844                         Log("SHUTTLE", Form("CopyFilesToGrid - Copying local file %s to %s succeeded!", 
845                                                 fullLocalPath.Data(), fullGridPath.Data()));
846                         RemoveFile(fullLocalPath);
847                         nTransfer++;
848                 }
849                 else
850                 {
851                         Log("SHUTTLE", Form("CopyFilesToGrid - Copying local file %s to %s FAILED!", 
852                                                 fullLocalPath.Data(), fullGridPath.Data()));
853                         success = kFALSE;
854                 }
855         }
856
857         Log("SHUTTLE", Form("CopyFilesToGrid - %d (over %d) files in folder %s copied to Grid.", 
858                                                 nTransfer, nDirs, dir.Data()));
859
860                 
861         delete dirList;
862         return success;
863 }
864
865 //______________________________________________________________________________________________
866 const char* AliShuttle::GetRefFilePrefix(const char* base, const char* detector)
867 {
868         //
869         // Get folder name of reference files 
870         //
871
872         TString offDetStr(GetOfflineDetName(detector));
873         static TString dir;
874         if (offDetStr == "ITS" || offDetStr == "MUON" || offDetStr == "PHOS")
875         {
876                 dir.Form("%s/%s/%s", base, offDetStr.Data(), detector);
877         } else {
878                 dir.Form("%s/%s", base, offDetStr.Data());
879         }
880         
881         return dir.Data();
882 }
883
884 //______________________________________________________________________________________________
885 void AliShuttle::CleanLocalStorage(const TString& uri)
886 {
887         //
888         // Called in case the preprocessor is declared failed. Remove remaining objects from the local storages.
889         //
890
891         const char* type = 0;
892         if(uri == fgkLocalCDB) {
893                 type = "OCDB";
894         } else if(uri == fgkLocalRefStorage) {
895                 type = "Reference";
896         } else {
897                 AliError(Form("Invalid storage URI: %s", uri.Data()));
898                 return;
899         }
900
901         AliCDBManager* man = AliCDBManager::Instance();
902
903         // open local storage
904         AliCDBStorage *localSto = man->GetStorage(uri);
905         if(!localSto) {
906                 Log("SHUTTLE",
907                         Form("CleanLocalStorage - cannot activate local %s storage", type));
908                 return;
909         }
910
911         TString filename(Form("%s/%s/*/Run*_v%d_s*.root",
912                 localSto->GetBaseFolder().Data(), GetOfflineDetName(fCurrentDetector.Data()), GetCurrentRun()));
913
914         AliDebug(2, Form("filename = %s", filename.Data()));
915
916         Log("SHUTTLE", Form("Removing remaining local files for run %d and detector %s ...",
917                 GetCurrentRun(), fCurrentDetector.Data()));
918
919         RemoveFile(filename.Data());
920
921 }
922
923 //______________________________________________________________________________________________
924 void AliShuttle::RemoveFile(const char* filename)
925 {
926         //
927         // removes local file
928         //
929
930         TString command(Form("rm -f %s", filename));
931
932         Int_t result = gSystem->Exec(command.Data());
933         if(result != 0)
934         {
935                 Log("SHUTTLE", Form("RemoveFile - %s: Cannot remove file %s!",
936                         fCurrentDetector.Data(), filename));
937         }
938 }
939
940 //______________________________________________________________________________________________
941 AliShuttleStatus* AliShuttle::ReadShuttleStatus()
942 {
943         //
944         // Reads the AliShuttleStatus from the CDB
945         //
946
947         if (fStatusEntry){
948                 delete fStatusEntry;
949                 fStatusEntry = 0;
950         }
951
952         Int_t path1 = GetCurrentRun()/10000;
953         fStatusEntry = AliCDBManager::Instance()->GetStorage(GetLocalCDB())
954                 ->Get(Form("/SHUTTLE/%s/%d", fCurrentDetector.Data(), path1), GetCurrentRun());
955
956         if (!fStatusEntry) return 0;
957         fStatusEntry->SetOwner(1);
958
959         AliShuttleStatus* status = dynamic_cast<AliShuttleStatus*> (fStatusEntry->GetObject());
960         if (!status) {
961                 AliError("Invalid object stored to CDB!");
962                 return 0;
963         }
964
965         return status;
966 }
967
968 //______________________________________________________________________________________________
969 Bool_t AliShuttle::WriteShuttleStatus(AliShuttleStatus* status)
970 {
971         //
972         // writes the status for one subdetector
973         //
974
975         if (fStatusEntry){
976                 delete fStatusEntry;
977                 fStatusEntry = 0;
978         }
979
980         Int_t run = GetCurrentRun();
981         Int_t path1 = run/10000;
982         TString path1_string = Form("%d",path1);
983
984         AliCDBId id(AliCDBPath("SHUTTLE", fCurrentDetector, path1_string), run, run);
985
986         fStatusEntry = new AliCDBEntry(status, id, new AliCDBMetaData);
987         fStatusEntry->SetOwner(1);
988
989         Int_t logLevel = AliLog::GetGlobalLogLevel();
990         AliLog::SetGlobalLogLevel(AliLog::kError);
991
992         UInt_t result = AliCDBManager::Instance()->GetStorage(fgkLocalCDB)->Put(fStatusEntry);
993
994         AliLog::SetGlobalLogLevel((AliLog::EType_t)logLevel);
995
996         if (!result) {
997                 Log("SHUTTLE", Form("WriteShuttleStatus - Failed for %s, run %d",
998                                                 fCurrentDetector.Data(), run));
999                 return kFALSE;
1000         }
1001         
1002         SendMLDetInfo();
1003
1004         return kTRUE;
1005 }
1006
1007 //______________________________________________________________________________________________
1008 void AliShuttle::UpdateShuttleStatus(AliShuttleStatus::Status newStatus, Bool_t increaseCount)
1009 {
1010         //
1011         // changes the AliShuttleStatus for the given detector and run to the given status
1012         //
1013
1014         if (!fStatusEntry){
1015                 AliError("UNEXPECTED: fStatusEntry empty");
1016                 return;
1017         }
1018
1019         AliShuttleStatus* status = dynamic_cast<AliShuttleStatus*> (fStatusEntry->GetObject());
1020
1021         if (!status){
1022                 Log("SHUTTLE", "UpdateShuttleStatus - UNEXPECTED: status could not be read from current CDB entry");
1023                 return;
1024         }
1025
1026         TString actionStr = Form("UpdateShuttleStatus - %s: Changing state from %s to %s",
1027                                 fCurrentDetector.Data(),
1028                                 status->GetStatusName(),
1029                                 status->GetStatusName(newStatus));
1030         Log("SHUTTLE", actionStr);
1031         SetLastAction(actionStr);
1032
1033         status->SetStatus(newStatus);
1034         if (increaseCount) status->IncreaseCount();
1035
1036         Int_t logLevel = AliLog::GetGlobalLogLevel();
1037         AliLog::SetGlobalLogLevel(AliLog::kError);
1038
1039         AliCDBManager::Instance()->GetStorage(fgkLocalCDB)->Put(fStatusEntry);
1040
1041         AliLog::SetGlobalLogLevel((AliLog::EType_t)logLevel);
1042
1043         SendMLDetInfo();
1044 }
1045
1046 //______________________________________________________________________________________________
1047 void AliShuttle::SendMLDetInfo()
1048 {
1049         //
1050         // sends ML information about the current status of the current detector being processed
1051         //
1052         
1053         AliShuttleStatus* status = dynamic_cast<AliShuttleStatus*> (fStatusEntry->GetObject());
1054         
1055         if (!status){
1056                 Log("SHUTTLE", "SendMLDetInfo - UNEXPECTED: status could not be read from current CDB entry");
1057                 return;
1058         }
1059         
1060         TMonaLisaText  mlStatus(Form("%s_status", fCurrentDetector.Data()), status->GetStatusName());
1061         TMonaLisaValue mlRetryCount(Form("%s_count", fCurrentDetector.Data()), status->GetCount());
1062
1063         TList mlList;
1064         mlList.Add(&mlStatus);
1065         mlList.Add(&mlRetryCount);
1066
1067         TString mlID;
1068         mlID.Form("%d", GetCurrentRun());
1069         fMonaLisa->SendParameters(&mlList, mlID);
1070 }
1071
1072 //______________________________________________________________________________________________
1073 Bool_t AliShuttle::ContinueProcessing()
1074 {
1075         // this function reads the AliShuttleStatus information from CDB and
1076         // checks if the processing should be continued
1077         // if yes it returns kTRUE and updates the AliShuttleStatus with nextStatus
1078
1079         if (!fConfig->HostProcessDetector(fCurrentDetector))
1080                 return kFALSE;
1081
1082         AliPreprocessor* aPreprocessor =
1083                 dynamic_cast<AliPreprocessor*> (fPreprocessorMap.GetValue(fCurrentDetector));
1084         if (!aPreprocessor)
1085         {
1086             Log("SHUTTLE", Form("ContinueProcessing - %s: no preprocessor registered", fCurrentDetector.Data()));
1087             if(fCurrentDetector==TString("HLT")){
1088                 Log("SHUTTLE", "ContinueProcessing: Fake processing HLT, to avoid pending runs! => temporary, till an HLT preprocessor is made available.");
1089                 UpdateShuttleLogbook("HLT","DONE");
1090                 Log(fCurrentDetector, "ContinueProcessing - skipping HLT preprocessor.");
1091
1092                 AliShuttleStatus* status = ReadShuttleStatus();
1093                 if (!status) {
1094                     status = new AliShuttleStatus(AliShuttleStatus::kDone);
1095                     WriteShuttleStatus(status);
1096                 }
1097             }
1098             return kFALSE;
1099         }
1100
1101         AliShuttleLogbookEntry::Status entryStatus =
1102                 fLogbookEntry->GetDetectorStatus(fCurrentDetector);
1103
1104         if (entryStatus != AliShuttleLogbookEntry::kUnprocessed) {
1105                 Log("SHUTTLE", Form("ContinueProcessing - %s is %s",
1106                                 fCurrentDetector.Data(),
1107                                 fLogbookEntry->GetDetectorStatusName(entryStatus)));
1108                 return kFALSE;
1109         }
1110
1111         // if we get here, according to Shuttle logbook subdetector is in UNPROCESSED state
1112
1113         // check if current run is first unprocessed run for current detector
1114         if (fConfig->StrictRunOrder(fCurrentDetector) &&
1115                 !fFirstUnprocessed[GetDetPos(fCurrentDetector)])
1116         {
1117                 if (fTestMode == kNone)
1118                 {
1119                         Log("SHUTTLE", Form("ContinueProcessing - %s requires strict run ordering"
1120                                             " but this is not the first unprocessed run!",fCurrentDetector.Data()));
1121                         return kFALSE;
1122                 }
1123                 else
1124                 {
1125                         Log("SHUTTLE", Form("ContinueProcessing - In TESTMODE - "
1126                                         "Although %s requires strict run ordering "
1127                                         "and this is not the first unprocessed run, "
1128                                             "the SHUTTLE continues",fCurrentDetector.Data()));
1129                 }
1130         }
1131
1132         // Is the subdetector processed first time for this run?
1133         fFirstProcessing = kFALSE;
1134
1135         AliShuttleStatus* status = ReadShuttleStatus();
1136         if (!status) {
1137                 // first time
1138                 Log("SHUTTLE", Form("ContinueProcessing - %s: Processing first time",
1139                                 fCurrentDetector.Data()));
1140                 status = new AliShuttleStatus(AliShuttleStatus::kStarted);
1141                 fFirstProcessing = kTRUE;
1142                 return WriteShuttleStatus(status);
1143         }
1144
1145         // The following case shouldn't happen if Shuttle Logbook was correctly updated.
1146         // If it happens it may mean Logbook updating failed... let's do it now!
1147         if (status->GetStatus() == AliShuttleStatus::kDone ||
1148             status->GetStatus() == AliShuttleStatus::kFailed ||
1149             status->GetStatus() == AliShuttleStatus::kSkipped) {
1150                 Log("SHUTTLE", Form("ContinueProcessing - %s is already %s. Updating Shuttle Logbook",
1151                                         fCurrentDetector.Data(),
1152                                         status->GetStatusName(status->GetStatus())));
1153                 
1154                 if (status->GetStatus() == AliShuttleStatus::kSkipped)
1155                 {
1156                         UpdateShuttleLogbook(fCurrentDetector.Data(), "DONE");
1157                 }
1158                 else
1159                         UpdateShuttleLogbook(fCurrentDetector.Data(), status->GetStatusName(status->GetStatus()));
1160                         
1161                 return kFALSE;
1162         }
1163
1164         if (status->GetStatus() == AliShuttleStatus::kStoreStarted || status->GetStatus() == AliShuttleStatus::kStoreDelayed ||status->GetStatus() == AliShuttleStatus::kStoreError) {
1165                 Log("SHUTTLE",
1166                         Form("ContinueProcessing - %s: Grid storage of one or more "
1167                                 "objects failed. Trying again now",
1168                                 fCurrentDetector.Data()));
1169                 StoreOCDB();
1170                 return kFALSE;
1171         }
1172
1173         // if we get here, there is a restart
1174         Bool_t cont = kFALSE;
1175
1176         // abort conditions
1177         if (status->GetCount() >= fConfig->GetMaxRetries()) {
1178                 Log("SHUTTLE", Form("ContinueProcessing - %s failed %d times in status %s - "
1179                                 "Updating Shuttle Logbook", fCurrentDetector.Data(),
1180                                 status->GetCount(), status->GetStatusName()));
1181                 UpdateShuttleLogbook(fCurrentDetector.Data(), "FAILED");
1182                 UpdateShuttleStatus(AliShuttleStatus::kFailed);
1183
1184                 // there may still be objects in local OCDB and reference storage
1185                 // and FXS databases may be not updated: do it now!
1186                 
1187                 // TODO Currently disabled, we want to keep files in case of failure!
1188                 // CleanLocalStorage(fgkLocalCDB);
1189                 // CleanLocalStorage(fgkLocalRefStorage);
1190                 // UpdateTableFailCase();
1191                 
1192                 // Send mail to detector expert!
1193                 Log("SHUTTLE", Form("ContinueProcessing - Sending mail to %s expert...", 
1194                                     fCurrentDetector.Data()));
1195                 // det experts in to
1196                 TString to="";
1197                 TIter *iterExperts = 0;
1198                 iterExperts = new TIter(fConfig->GetResponsibles(fCurrentDetector));
1199                 TObjString *anExpert=0;
1200                 while ((anExpert = (TObjString*) iterExperts->Next()))
1201                         {
1202                                 to += Form("%s, \n", anExpert->GetName());
1203                         }
1204                 delete iterExperts;
1205                 
1206                 if (to.Length() > 0)
1207                         to.Remove(to.Length()-3);
1208                 AliDebug(2, Form("to: %s",to.Data()));
1209
1210                 if (to.IsNull()) {
1211                         Log("SHUTTLE", Form("List of %s responsibles not set!", fCurrentDetector.Data()));
1212                         return kFALSE;
1213                 }
1214
1215                 Log(fCurrentDetector.Data(), Form("ContinueProcessing - Sending mail to %s expert(s):", 
1216                                     fCurrentDetector.Data()));
1217                 Log(fCurrentDetector.Data(), Form("\n%s", to.Data()));
1218                 if (!SendMail(kPPEMail))
1219                         Log("SHUTTLE", Form("ContinueProcessing - Could not send mail to %s expert",
1220                                             fCurrentDetector.Data()));
1221
1222         } else {
1223                 Log("SHUTTLE", Form("ContinueProcessing - %s: restarting. "
1224                                 "Aborted before with %s. Retry number %d.", fCurrentDetector.Data(),
1225                                 status->GetStatusName(), status->GetCount()));
1226                 Bool_t increaseCount = kTRUE;
1227                 if (status->GetStatus() == AliShuttleStatus::kDCSError || 
1228                     status->GetStatus() == AliShuttleStatus::kDCSStarted ||
1229                     status->GetStatus() == AliShuttleStatus::kFXSError ||
1230                     status->GetStatus() == AliShuttleStatus::kOCDBError)
1231                                 increaseCount = kFALSE;
1232                                 
1233                 UpdateShuttleStatus(AliShuttleStatus::kStarted, increaseCount);
1234                 cont = kTRUE;
1235         }
1236
1237         return cont;
1238 }
1239
1240 //______________________________________________________________________________________________
1241 void AliShuttle::SendMLRunInfo(const char* status)
1242 {
1243         // 
1244         // Send information about this run to ML
1245         
1246         TMonaLisaText  mlStatus("SHUTTLE_status", status);
1247         TString runType(fLogbookEntry->GetRunType());
1248         if (strlen(fLogbookEntry->GetRunParameter("log")) > 0){
1249
1250                 runType += "(";
1251                 runType += fLogbookEntry->GetRunParameter("log");
1252                 runType += ")";
1253         }
1254         if (fLogbookEntry->GetDATestMode()){
1255                 runType += " (DATest)";
1256         }
1257         TMonaLisaText  mlRunType("SHUTTLE_runtype", runType);
1258
1259         TList mlList;
1260         mlList.Add(&mlStatus);
1261         mlList.Add(&mlRunType);
1262
1263         TString mlID;
1264         mlID.Form("%d", GetCurrentRun());
1265         fMonaLisa->SendParameters(&mlList, mlID);       
1266 }
1267
1268 //______________________________________________________________________________________________
1269 Int_t AliShuttle::GetMem(Int_t pid)
1270 {
1271         // invokes ps to get the memory consumption of the process <pid>
1272         // returns -1 in case of error
1273         
1274         TString checkStr;
1275         checkStr.Form("ps -o vsize --pid %d | tail -n 1", pid);
1276         FILE* pipe = gSystem->OpenPipe(checkStr, "r");
1277         if (!pipe)
1278         {
1279                 Log("SHUTTLE", Form("Process - Error: "
1280                         "Could not open pipe to %s", checkStr.Data()));
1281                 return -1;
1282         }
1283                 
1284         char buffer[100];
1285         if (!fgets(buffer, 100, pipe))
1286         {
1287                 Log("SHUTTLE", "Process - Error: ps did not return anything");
1288                 gSystem->ClosePipe(pipe);
1289                 return -1;
1290         }
1291         gSystem->ClosePipe(pipe);
1292         
1293         //Log("SHUTTLE", Form("ps returned %s", buffer));
1294         
1295         Int_t mem = 0;
1296         if ((sscanf(buffer, "%d\n", &mem) != 1) || !mem)
1297         {
1298                 Log("SHUTTLE", "Process - Error: Could not parse output of ps");
1299                 return -1;
1300         }
1301         
1302         return mem;
1303 }
1304
1305 //______________________________________________________________________________________________
1306 Bool_t AliShuttle::Process(AliShuttleLogbookEntry* entry)
1307 {
1308         //
1309         // Makes data retrieval for all detectors in the configuration.
1310         // entry: Shuttle logbook entry, contains run paramenters and status of detectors
1311         // (Unprocessed, Inactive, Failed or Done).
1312         // Returns kFALSE in case of error occured and kTRUE otherwise
1313         //
1314
1315         if (!entry) return kFALSE;
1316
1317         fLogbookEntry = entry;
1318
1319         Log("SHUTTLE", Form("\t\t\t^*^*^*^*^*^*^*^*^*^*^*^* run %d: START ^*^*^*^*^*^*^*^*^*^*^*^*",
1320                                         GetCurrentRun()));
1321
1322         CountOpenRuns();
1323         
1324         // Send the information to ML
1325         SendMLRunInfo("Processing");
1326
1327         if (fLogbookEntry->IsDone())
1328         {
1329                 Log("SHUTTLE","Process - Shuttle is already DONE. Updating logbook");
1330                 UpdateShuttleLogbook("shuttle_done");
1331                 fLogbookEntry = 0;
1332                 return kTRUE;
1333         }
1334
1335         // read test mode if flag is set
1336         if (fReadTestMode)
1337         {
1338                 fTestMode = kNone;
1339                 TString logEntry(entry->GetRunParameter("log"));
1340                 //printf("log entry = %s\n", logEntry.Data());
1341                 TString searchStr("Testmode: ");
1342                 Int_t pos = logEntry.Index(searchStr.Data());
1343                 //printf("%d\n", pos);
1344                 if (pos >= 0)
1345                 {
1346                         TSubString subStr = logEntry(pos + searchStr.Length(), logEntry.Length());
1347                         //printf("%s\n", subStr.String().Data());
1348                         TString newStr(subStr.Data());
1349                         TObjArray* token = newStr.Tokenize(' ');
1350                         if (token)
1351                         {
1352                                 //token->Print();
1353                                 TObjString* tmpStr = dynamic_cast<TObjString*> (token->First());
1354                                 if (tmpStr)
1355                                 {
1356                                         Int_t testMode = tmpStr->String().Atoi();
1357                                         if (testMode > 0)
1358                                         {
1359                                                 Log("SHUTTLE", Form("Process - Enabling test mode %d", testMode));
1360                                                 SetTestMode((TestMode) testMode);
1361                                         }
1362                                 }
1363                                 delete token;          
1364                         }
1365                 }
1366         }
1367                 
1368         fLogbookEntry->Print("all");
1369
1370         // Initialization
1371         Bool_t hasError = kFALSE;
1372
1373         // Set the CDB and Reference folders according to the year
1374
1375         // build cdb paths (repeat each time, run might be a DATest run)
1376         if (!fLogbookEntry->GetDATestMode()){
1377                 fgkMainCDB.Form("alien://folder=%s%d/OCDB?user=alidaq?cacheFold=/tmp/OCDBCache", 
1378                                 fConfig->GetAlienPath(), GetCurrentYear());
1379                 
1380                 fgkMainRefStorage.Form("alien://folder=%s%d/Reference?user=alidaq?cacheFold=/tmp/OCDBCache", 
1381                                        fConfig->GetAlienPath(), GetCurrentYear());
1382         }
1383         else {
1384                 fgkMainCDB.Form("alien://folder=%s%d/DATest/OCDB?user=alidaq?cacheFold=/tmp/OCDBCache",
1385                                 fConfig->GetAlienPath(), GetCurrentYear());
1386                 
1387                 fgkMainRefStorage.Form("alien://folder=%s%d/DATest/Reference?user=alidaq?cacheFold=/tmp/OCDBCache",
1388                                        fConfig->GetAlienPath(), GetCurrentYear());
1389         }
1390
1391         AliDebug(2,Form("Main CDB storage = %s",fgkMainCDB.Data()));
1392         AliDebug(2,Form("Main Reference storage = %s",fgkMainRefStorage.Data()));
1393
1394         // Loop on detectors in the configuration
1395         TIter iter(fConfig->GetDetectors());
1396         TObjString* aDetector = 0;
1397
1398         Bool_t first = kTRUE;
1399
1400         while ((aDetector = (TObjString*) iter.Next()))
1401         {
1402                 fCurrentDetector = aDetector->String();
1403
1404                 if (ContinueProcessing() == kFALSE) 
1405                         continue;
1406                 
1407                 if (first)
1408                 {
1409                   // only read QueryCDB when needed and only once
1410                   AliCDBStorage *mainCDBSto = AliCDBManager::Instance()->GetStorage(fgkMainCDB);
1411                   if(mainCDBSto) mainCDBSto->QueryCDB(GetCurrentRun());
1412                   AliCDBStorage *mainRefSto = AliCDBManager::Instance()->GetStorage(fgkMainRefStorage);
1413                   if(mainRefSto) mainRefSto->QueryCDB(GetCurrentRun());
1414                   first = kFALSE;
1415                 }
1416
1417                 Log("SHUTTLE", Form("\t\t\t****** run %d - %s: START  ******",
1418                                                 GetCurrentRun(), aDetector->GetName()));
1419
1420                 for(Int_t iSys=0;iSys<3;iSys++) fFXSCalled[iSys]=kFALSE;
1421                 
1422                 Int_t initialMem = GetMem(getpid());
1423                 Log("SHUTTLE", Form("Memory consumption before forking is %d", initialMem));
1424
1425                 Log(fCurrentDetector.Data(), "Process - Starting processing");
1426
1427                 Int_t pid = fork();
1428
1429                 if (pid < 0)
1430                 {
1431                         Log("SHUTTLE", "Process - ERROR: Forking failed");
1432                 }
1433                 else if (pid > 0)
1434                 {
1435                         // parent
1436                         Log("SHUTTLE", Form("Process - In parent process of %d - %s: Starting monitoring",
1437                                                         GetCurrentRun(), aDetector->GetName()));
1438
1439                         Long_t begin = time(0);
1440
1441                         int status; // to be used with waitpid, on purpose an int (not Int_t)!
1442                         while (waitpid(pid, &status, WNOHANG) == 0)
1443                         {
1444                                 Long_t expiredTime = time(0) - begin;
1445
1446                                 if (expiredTime > fConfig->GetPPTimeOut())
1447                                 {
1448                                         TString logMsg;
1449                                         AliShuttleStatus *currentStatus = ReadShuttleStatus();
1450                                         AliShuttleStatus::Status newStatus = AliShuttleStatus::kInvalid;
1451                                         
1452                                         if (currentStatus->GetStatus() == AliShuttleStatus::kDCSStarted)
1453                                         {
1454                                                 // in case the pp goes in TimeOut while retrieving the DCS DPs
1455                                                 // set status to kDCSError
1456                                                 
1457                                                 logMsg.Form("Process - Process of %s timed out while retrieving the DCS DataPoints. Run time: %ld seconds. Killing... and setting status to DCSError.",
1458                                                                 fCurrentDetector.Data(), expiredTime);
1459                                                 newStatus = AliShuttleStatus::kDCSError;
1460                                         }
1461                                         else if (currentStatus->GetStatus() <= AliShuttleStatus::kPPDone)
1462                                         {
1463                                                 // in case pp not yet done set status to kPPTimeOut
1464                                         
1465                                                 logMsg.Form("Process - Process of %s timed out. Run time: %ld seconds. Killing...",
1466                                                                 fCurrentDetector.Data(), expiredTime);
1467                                                 newStatus = AliShuttleStatus::kPPTimeOut;
1468                                         }
1469                                         else if (currentStatus->GetStatus() == AliShuttleStatus::kStoreStarted)
1470                                         {
1471                                                 // in case the pp goes in TimeOut while storing the objects in the OCDB
1472                                                 // set status to kStoreError
1473                                                 
1474                                                 logMsg.Form("Process - Process of %s timed out while storing the OCDB object. Run time: %ld seconds. Killing... and setting status to StoreError.",
1475                                                                 fCurrentDetector.Data(), expiredTime);
1476                                                 newStatus = AliShuttleStatus::kStoreError;
1477                                         }
1478                                         else 
1479                                         {
1480                                                 // in other cases don't change the status
1481                                                 
1482                                                 logMsg.Form("Process - Process of %s timed out in status = %s. Run time: %ld seconds. Killing... without changing the status",
1483                                                                 fCurrentDetector.Data(), currentStatus->GetStatusName(), expiredTime);
1484                                         }
1485                                 
1486                                         Log("SHUTTLE", logMsg);
1487                                         Log(fCurrentDetector, logMsg);
1488
1489                                         kill(pid, 9);
1490
1491                                         if (newStatus != AliShuttleStatus::kInvalid)
1492                                                 UpdateShuttleStatus(newStatus);
1493                                         hasError = kTRUE;
1494
1495                                         gSystem->Sleep(1000);
1496                                 }
1497                                 else
1498                                 {
1499                                         gSystem->Sleep(1000);
1500                                         
1501                                         Int_t mem = GetMem(pid);
1502
1503                                         if (mem < 0)
1504                                                 continue;
1505                                                 
1506                                         mem -= initialMem;
1507                                         if (mem < 0)
1508                                                 mem = 0;
1509                                         
1510                                         if (expiredTime % 60 == 0)
1511                                         {
1512                                                 Log("SHUTTLE", Form("Process - %s: Checking process. "
1513                                                         "Run time: %ld seconds - Memory consumption: %d KB",
1514                                                         fCurrentDetector.Data(), expiredTime, mem));
1515                                                 SendAlive();
1516                                         }
1517                                         
1518                                         if (mem > fConfig->GetPPMaxMem())
1519                                         {
1520                                                 TString tmp;
1521                                                 tmp.Form("Process - Process exceeds maximum allowed memory "
1522                                                         "(%d KB > %d KB). Killing...",
1523                                                         mem, fConfig->GetPPMaxMem());
1524                                                 Log("SHUTTLE", tmp);
1525                                                 Log(fCurrentDetector, tmp);
1526         
1527                                                 kill(pid, 9);
1528         
1529                                                 UpdateShuttleStatus(AliShuttleStatus::kPPOutOfMemory);
1530                                                 hasError = kTRUE;
1531         
1532                                                 gSystem->Sleep(1000);
1533                                         }
1534                                 }
1535                         }
1536
1537                         Log("SHUTTLE", Form("Process - In parent process of %d - %s: Client has terminated.",
1538                                                                 GetCurrentRun(), aDetector->GetName()));
1539
1540                         if (WIFEXITED(status))
1541                         {
1542                                 Int_t returnCode = WEXITSTATUS(status);
1543
1544                                 Log("SHUTTLE", Form("Process - %s: the return code is %d", fCurrentDetector.Data(),
1545                                                                                 returnCode));
1546
1547                                 if (returnCode == 0) hasError = kTRUE;
1548                         }
1549                 }
1550                 else if (pid == 0)
1551                 {
1552                         // child
1553                         Log("SHUTTLE", Form("Process - In child process of %d - %s", GetCurrentRun(),
1554                                 aDetector->GetName()));
1555
1556                         Log("SHUTTLE", Form("Process - Redirecting output to %s log",fCurrentDetector.Data()));
1557
1558                         if ((freopen(GetLogFileName(fCurrentDetector), "a", stdout)) == 0)
1559                         {
1560                                 Log("SHUTTLE", "Process - Could not freopen stdout");
1561                         }
1562                         else
1563                         {
1564                                 fOutputRedirected = kTRUE;
1565                                 if ((dup2(fileno(stdout), fileno(stderr))) < 0)
1566                                         Log("SHUTTLE", "Process - Could not redirect stderr");
1567                                 
1568                         }
1569
1570                         Log("SHUTTLE", "Executing TGrid::Connect");
1571                         TGrid::Connect("alien://");
1572                         
1573                         TString wd = gSystem->WorkingDirectory();
1574                         Int_t dir_lev1 = GetCurrentRun()/10000;
1575                         TString tmpDir = Form("%s/%d/%d/%s_process", GetShuttleTempDir(), 
1576                                 dir_lev1, GetCurrentRun(), fCurrentDetector.Data());
1577                         
1578                         Int_t result = gSystem->GetPathInfo(tmpDir.Data(), 0, (Long64_t*) 0, 0, 0);
1579                         if (!result) // temp dir already exists!
1580                         {
1581                                 Log(fCurrentDetector.Data(), 
1582                                         Form("Process - %s dir already exists! Removing...", tmpDir.Data()));
1583                                 gSystem->Exec(Form("rm -rf %s",tmpDir.Data()));         
1584                         } 
1585                         
1586                         if (gSystem->mkdir(tmpDir.Data(), 1))
1587                         {
1588                                 Log(fCurrentDetector.Data(), "Process - could not make temp directory!!");
1589                                 gSystem->Exit(1);
1590                         }
1591                         
1592                         if (!gSystem->ChangeDirectory(tmpDir.Data())) 
1593                         {
1594                                 Log(fCurrentDetector.Data(), "Process - could not change directory!!");
1595                                 gSystem->Exit(1);                       
1596                         }
1597                         
1598                         Int_t success = ProcessCurrentDetector();
1599
1600                         gSystem->ChangeDirectory(wd.Data());
1601                                                 
1602                         if (success == 1) // Preprocessor finished successfully!
1603                         { 
1604                                 // remove temporary folder or DCS map
1605                                 if (!fConfig->KeepTempFolder())
1606                                 {
1607                                         gSystem->Exec(Form("rm -rf %s",tmpDir.Data()));
1608                                 } else if (!fConfig->KeepDCSMap())
1609                                 {
1610                                         gSystem->Exec(Form("rm -f %s/DCSMap.root",tmpDir.Data()));
1611                                 }
1612                                 
1613                                 // Update time_processed field in FXS DB
1614                                 if (UpdateTable() == kFALSE)
1615                                         Log("SHUTTLE", Form("Process - %s: Could not update FXS databases!", 
1616                                                         fCurrentDetector.Data()));
1617
1618                                 // Transfer the data from local storage to main storage (Grid)
1619                                 if (StoreOCDB() == kFALSE)
1620                                         success = kFALSE;
1621                         } 
1622                         else if (success == 0)
1623                         {
1624                                 Log("SHUTTLE", 
1625                                         Form("\t\t\t****** run %d - %s: ERROR ******",
1626                                                 GetCurrentRun(), aDetector->GetName()));
1627                         }
1628
1629                         for (UInt_t iSys=0; iSys<3; iSys++)
1630                         {
1631                                 if (fFXSCalled[iSys]) fFXSlist[iSys].Clear();
1632                         }
1633
1634                         Log("SHUTTLE", Form("Process - Client process of %d - %s is exiting now with %d.",
1635                                                         GetCurrentRun(), aDetector->GetName(), success));
1636
1637                         // the client exits here
1638                         gSystem->Exit(success);
1639
1640                         AliError("We should never get here!!!");
1641                 }
1642         }
1643
1644         Log("SHUTTLE", Form("\t\t\t^*^*^*^*^*^*^*^*^*^*^*^* run %d: FINISH ^*^*^*^*^*^*^*^*^*^*^*^*",
1645                                                         GetCurrentRun()));
1646
1647         //check if shuttle is done for this run, if so update logbook
1648         TObjArray checkEntryArray;
1649         checkEntryArray.SetOwner(1);
1650         TString whereClause = Form("where run=%d", GetCurrentRun());
1651         if (!QueryShuttleLogbook(whereClause.Data(), checkEntryArray) || 
1652                         checkEntryArray.GetEntries() == 0) {
1653                 Log("SHUTTLE", Form("Process - Warning: Cannot check status of run %d on Shuttle logbook!",
1654                                                 GetCurrentRun()));
1655                 return hasError == kFALSE;
1656         }
1657
1658         AliShuttleLogbookEntry* checkEntry = dynamic_cast<AliShuttleLogbookEntry*>
1659                                                 (checkEntryArray.At(0));
1660
1661         if (checkEntry)
1662         {
1663                 if (checkEntry->IsDone())
1664                 {
1665                         Log("SHUTTLE","Process - Shuttle is DONE. Updating logbook");
1666                         UpdateShuttleLogbook("shuttle_done");
1667                 }
1668                 else
1669                 {
1670                         for (UInt_t iDet=0; iDet<NDetectors(); iDet++)
1671                         {
1672                                 if (checkEntry->GetDetectorStatus(iDet) == AliShuttleLogbookEntry::kUnprocessed)
1673                                 {
1674                                         AliDebug(2, Form("Run %d: setting %s as \"not first time unprocessed\"",
1675                                                         checkEntry->GetRun(), GetDetName(iDet)));
1676                                         fFirstUnprocessed[iDet] = kFALSE;
1677                                 }
1678                         }
1679                         SendMLRunInfo("Pending");
1680                 }
1681         }
1682
1683         fLogbookEntry = 0;
1684
1685         return hasError == kFALSE;
1686 }
1687
1688 //______________________________________________________________________________________________
1689 Int_t AliShuttle::ProcessCurrentDetector()
1690 {
1691         //
1692         // Makes data retrieval just for a specific detector (fCurrentDetector).
1693         // Threre should be a configuration for this detector.
1694
1695         Log("SHUTTLE", Form("ProcessCurrentDetector - Retrieving values for %s, run %d", 
1696                                                 fCurrentDetector.Data(), GetCurrentRun()));
1697
1698         TString wd = gSystem->WorkingDirectory();
1699         
1700         if (!CleanReferenceStorage(fCurrentDetector.Data()))
1701                 return 0;
1702         
1703         gSystem->ChangeDirectory(wd.Data());
1704         
1705         // call preprocessor
1706         AliPreprocessor* aPreprocessor =
1707                 dynamic_cast<AliPreprocessor*> (fPreprocessorMap.GetValue(fCurrentDetector));
1708
1709         // check if the preprocessor wants to process this run type
1710         if (aPreprocessor->ProcessRunType() == kFALSE)
1711         {
1712                 UpdateShuttleStatus(AliShuttleStatus::kSkipped);
1713                 UpdateShuttleLogbook(fCurrentDetector, "DONE");
1714                 if (!UpdateTableSkippedCase(fCurrentDetector.Data()))
1715                 {
1716                         AliError(Form("Could not update FXS tables for run %d !", GetCurrentRun()));
1717                 }
1718                 Log(fCurrentDetector, Form("ProcessCurrentDetector - %s preprocessor is not interested in this run type", fCurrentDetector.Data()));
1719         
1720                 return 2;
1721         }
1722         
1723         // checking if OCDB is reachable
1724         AliCDBEntry* testEntry = GetFromOCDB("SHUTTLE","GRP/CTP/DummyConfig");
1725         if (!testEntry){
1726                 // OCDB is not accessible, going in OCDBError for current detector
1727                 AliError("OCDB Test entry not accessible");
1728                 UpdateShuttleStatus(AliShuttleStatus::kOCDBError);
1729                 return 0;
1730         }
1731
1732         TMap* dcsMap = new TMap();
1733         
1734         aPreprocessor->Initialize(GetCurrentRun(), GetCurrentStartTime(), GetCurrentEndTime());
1735
1736         Bool_t processDCS = aPreprocessor->ProcessDCS();
1737
1738         if (!processDCS)
1739         {
1740                 Log(fCurrentDetector, "ProcessCurrentDetector -"
1741                         " The preprocessor requested to skip the retrieval of DCS values");
1742         }
1743         else if (fTestMode & kSkipDCS)
1744         {
1745                 Log(fCurrentDetector, "ProcessCurrentDetector - In TESTMODE: Skipping DCS processing");
1746         } 
1747         else if (fTestMode & kErrorDCS)
1748         {
1749                 Log(fCurrentDetector, "ProcessCurrentDetector - In TESTMODE: Simulating DCS error");
1750                 UpdateShuttleStatus(AliShuttleStatus::kDCSStarted);
1751                 UpdateShuttleStatus(AliShuttleStatus::kDCSError);
1752                 delete dcsMap;
1753                 return 0;
1754         } else {
1755
1756                 UpdateShuttleStatus(AliShuttleStatus::kDCSStarted);
1757
1758                 // Query DCS archive
1759                 Int_t nServers = fConfig->GetNServers(fCurrentDetector);
1760                 
1761                 for (int iServ=0; iServ<nServers; iServ++)
1762                 {
1763                 
1764                         TString host(fConfig->GetDCSHost(fCurrentDetector, iServ));
1765                         Int_t port = fConfig->GetDCSPort(fCurrentDetector, iServ);
1766                         Int_t multiSplit = fConfig->GetMultiSplit(fCurrentDetector, iServ);
1767
1768                         Log(fCurrentDetector, Form("ProcessCurrentDetector -"
1769                                         " Querying DCS Amanda server %s:%d (%d of %d)", 
1770                                         host.Data(), port, iServ+1, nServers));
1771                         
1772                         TMap* aliasMap = 0;
1773                         TMap* dpMap = 0;
1774
1775                         if (fConfig->GetDCSAliases(fCurrentDetector, iServ)->GetEntries() > 0)
1776                         {
1777                                 Log(fCurrentDetector, Form("Querying %d DCS aliases", fConfig->GetDCSAliases(fCurrentDetector, iServ)->GetEntries()));
1778                                 aliasMap = GetValueSet(host, port, 
1779                                                 fConfig->GetDCSAliases(fCurrentDetector, iServ), 
1780                                                 kAlias, multiSplit);
1781                                 if (!aliasMap)
1782                                 {
1783                                         Log(fCurrentDetector, 
1784                                                 Form("ProcessCurrentDetector -"
1785                                                         " Error retrieving DCS aliases from server %s."
1786                                                         " Sending mail to DCS experts!", host.Data()));
1787                                         UpdateShuttleStatus(AliShuttleStatus::kDCSError);
1788                                         
1789                                         if (!SendMail(kDCSEMail))
1790                                                 Log("SHUTTLE", Form("ProcessCurrentDetector - "
1791                                                                     "Could not send mail to DCS experts!"));
1792
1793                                         delete dcsMap;
1794                                         return 0;
1795                                 }
1796                         }
1797                         
1798                         if (fConfig->GetDCSDataPoints(fCurrentDetector, iServ)->GetEntries() > 0)
1799                         {
1800                                 Log(fCurrentDetector, Form("Querying %d DCS data points", fConfig->GetDCSDataPoints(fCurrentDetector, iServ)->GetEntries()));
1801                                 dpMap = GetValueSet(host, port, 
1802                                                 fConfig->GetDCSDataPoints(fCurrentDetector, iServ), 
1803                                                 kDP, multiSplit);
1804                                 if (!dpMap)
1805                                 {
1806                                         Log(fCurrentDetector, 
1807                                                 Form("ProcessCurrentDetector -"
1808                                                         " Error retrieving DCS data points from server %s."
1809                                                         " Sending mail to DCS experts!", host.Data()));
1810                                         UpdateShuttleStatus(AliShuttleStatus::kDCSError);
1811                                         
1812                                         if (!SendMail(kDCSEMail))
1813                                                 Log("SHUTTLE", Form("ProcessCurrentDetector - "
1814                                                                     "Could not send mail to DCS experts!"));
1815                                         
1816                                         if (aliasMap) delete aliasMap;
1817                                         delete dcsMap;
1818                                         return 0;
1819                                 }                               
1820                         }
1821                         
1822                         // merge aliasMap and dpMap into dcsMap
1823                         if(aliasMap) {
1824                                 TIter iter(aliasMap);
1825                                 TObjString* key = 0;
1826                                 while ((key = (TObjString*) iter.Next()))
1827                                         dcsMap->Add(key, aliasMap->GetValue(key->String()));
1828                                 
1829                                 aliasMap->SetOwner(kFALSE);
1830                                 delete aliasMap;
1831                         }       
1832                         
1833                         if(dpMap) {
1834                                 TIter iter(dpMap);
1835                                 TObjString* key = 0;
1836                                 while ((key = (TObjString*) iter.Next()))
1837                                         dcsMap->Add(key, dpMap->GetValue(key->String()));
1838                                 
1839                                 dpMap->SetOwner(kFALSE);
1840                                 delete dpMap;
1841                         }
1842                 }
1843         }
1844         
1845         // save map into file, to help debugging in case of preprocessor error
1846         TFile* f = TFile::Open("DCSMap.root","recreate");
1847         f->cd();
1848         dcsMap->Write("DCSMap", TObject::kSingleKey);
1849         f->Close();
1850         delete f;
1851         
1852         // DCS Archive DB processing successful. Call Preprocessor!
1853         UpdateShuttleStatus(AliShuttleStatus::kPPStarted);
1854
1855         fFXSError = -1; // this variable is kTRUE after ::Process if an FXS error occured
1856         
1857         UInt_t returnValue = aPreprocessor->Process(dcsMap);
1858         
1859         if (fFXSError!=-1) {
1860                 UpdateShuttleStatus(AliShuttleStatus::kFXSError);
1861                 SendMail(kFXSEMail, fFXSError);
1862                 dcsMap->DeleteAll();
1863                 delete dcsMap;
1864                 return 0;
1865         }
1866
1867         if (returnValue > 0) // Preprocessor error!
1868         {
1869                 Log(fCurrentDetector, Form("ProcessCurrentDetector - "
1870                                 "Preprocessor failed. Process returned %d.", returnValue));
1871                 UpdateShuttleStatus(AliShuttleStatus::kPPError);
1872                 dcsMap->DeleteAll();
1873                 delete dcsMap;
1874                 return 0;
1875         }
1876         
1877         // preprocessor ok!
1878         UpdateShuttleStatus(AliShuttleStatus::kPPDone);
1879         Log(fCurrentDetector, Form("ProcessCurrentDetector - %s preprocessor returned success",
1880                                 fCurrentDetector.Data()));
1881
1882         dcsMap->DeleteAll();
1883         delete dcsMap;
1884
1885         return 1;
1886 }
1887
1888 //______________________________________________________________________________________________
1889 void AliShuttle::CountOpenRuns()
1890 {
1891         // Query DAQ's Shuttle logbook and sends the number of open runs to ML
1892         
1893         SendAlive();
1894         
1895         // check connection, in case connect
1896         if (!Connect(4)) 
1897                 return;
1898
1899         TString sqlQuery;
1900         sqlQuery = Form("select count(*) from %s where shuttle_done=0", fConfig->GetShuttlelbTable());
1901         
1902         TSQLResult* aResult = fServer[4]->Query(sqlQuery);
1903         if (!aResult) {
1904                 AliError(Form("Can't execute query <%s>!", sqlQuery.Data()));
1905                 return;
1906         }
1907
1908         AliDebug(2,Form("Query = %s", sqlQuery.Data()));
1909         
1910         if (aResult->GetRowCount() == 0) {
1911                 AliError(Form("No result for query %s received", sqlQuery.Data()));
1912                 return;
1913         }
1914
1915         if (aResult->GetFieldCount() != 1) {
1916                 AliError(Form("Invalid field count for query %s received", sqlQuery.Data()));
1917                 return;
1918         }
1919
1920         TSQLRow* aRow = aResult->Next();
1921         if (!aRow) {
1922                 AliError(Form("Could not receive result of query %s", sqlQuery.Data()));
1923                 return;
1924         }
1925         
1926         TString result(aRow->GetField(0), aRow->GetFieldLength(0));
1927         Int_t count = result.Atoi();
1928         
1929         Log("SHUTTLE", Form("%d unprocessed runs", count));
1930         
1931         delete aRow;
1932         delete aResult;
1933
1934         TMonaLisaValue mlStatus("SHUTTLE_openruns", count);
1935
1936         TList mlList;
1937         mlList.Add(&mlStatus);
1938
1939         fMonaLisa->SendParameters(&mlList, "__PROCESSINGINFO__");
1940 }
1941
1942 //______________________________________________________________________________________________
1943 Bool_t AliShuttle::QueryShuttleLogbook(const char* whereClause,
1944                 TObjArray& entries)
1945 {
1946         // Query DAQ's Shuttle logbook and fills detector status object.
1947         // Call QueryRunParameters to query DAQ logbook for run parameters.
1948         //
1949
1950         entries.SetOwner(1);
1951
1952         // check connection, in case connect
1953         if (!Connect(4)) return kFALSE;
1954
1955         TString sqlQuery;
1956         sqlQuery = Form("select * from %s %s order by run", fConfig->GetShuttlelbTable(), whereClause);
1957
1958         TSQLResult* aResult = fServer[4]->Query(sqlQuery);
1959         if (!aResult) {
1960                 AliError(Form("Can't execute query <%s>!", sqlQuery.Data()));
1961                 return kFALSE;
1962         }
1963
1964         AliDebug(2,Form("Query = %s", sqlQuery.Data()));
1965
1966         if(aResult->GetRowCount() == 0) {
1967                 Log("SHUTTLE", "No entries in Shuttle Logbook match request");
1968                 delete aResult;
1969                 return kTRUE;
1970         }
1971
1972         // TODO Check field count!
1973         const UInt_t nCols = 26;
1974         if (aResult->GetFieldCount() != (Int_t) nCols) {
1975                 Log("SHUTTLE", "Invalid SQL result field number!");
1976                 delete aResult;
1977                 return kFALSE;
1978         }
1979
1980         TSQLRow* aRow;
1981         while ((aRow = aResult->Next())) {
1982                 TString runString(aRow->GetField(0), aRow->GetFieldLength(0));
1983                 Int_t run = runString.Atoi();
1984
1985                 AliShuttleLogbookEntry *entry = QueryRunParameters(run);
1986                 if (!entry)
1987                         continue;
1988
1989                 // DA test mode flag
1990                 TString daTestModeString(aRow->GetField(2), aRow->GetFieldLength(2)); // field 2 = DA test mode flag 
1991                 Bool_t daTestMode = (Bool_t)daTestModeString.Atoi();
1992                 entry->SetDATestMode(daTestMode);
1993
1994                 // loop on detectors
1995                 for(UInt_t ii = 0; ii < nCols; ii++)
1996                         entry->SetDetectorStatus(aResult->GetFieldName(ii), aRow->GetField(ii));
1997
1998                 entries.AddLast(entry);
1999                 delete aRow;
2000         }
2001
2002         delete aResult;
2003         return kTRUE;
2004 }
2005
2006 //______________________________________________________________________________________________
2007 AliShuttleLogbookEntry* AliShuttle::QueryRunParameters(Int_t run)
2008 {
2009         //
2010         // Retrieve run parameters written in the DAQ logbook and sets them into AliShuttleLogbookEntry object
2011         //
2012
2013         // check connection, in case connect
2014         if (!Connect(4))
2015                 return 0;
2016
2017         TString sqlQuery;
2018         sqlQuery.Form("select * from %s where run=%d", fConfig->GetDAQlbTable(), run);
2019
2020         TSQLResult* aResult = fServer[4]->Query(sqlQuery);
2021         if (!aResult) {
2022                 Log("SHUTTLE", Form("Can't execute query <%s>!", sqlQuery.Data()));
2023                 return 0;
2024         }
2025
2026         if (aResult->GetRowCount() == 0) {
2027                 Log("SHUTTLE", Form("QueryRunParameters - No entry in DAQ Logbook for run %d. Skipping", run));
2028                 delete aResult;
2029                 return 0;
2030         }
2031
2032         if (aResult->GetRowCount() > 1) {
2033                 Log("SHUTTLE", Form("QueryRunParameters - UNEXPECTED: "
2034                                 "more than one entry in DAQ Logbook for run %d!", run));
2035                 delete aResult;
2036                 return 0;
2037         }
2038
2039         TSQLRow* aRow = aResult->Next();
2040         if (!aRow)
2041         {
2042                 Log("SHUTTLE", Form("QueryRunParameters - Could not retrieve row for run %d. Skipping", run));
2043                 delete aResult;
2044                 return 0;
2045         }
2046
2047         AliShuttleLogbookEntry* entry = new AliShuttleLogbookEntry(run);
2048
2049         for (Int_t ii = 0; ii < aResult->GetFieldCount(); ii++)
2050                 entry->SetRunParameter(aResult->GetFieldName(ii), aRow->GetField(ii));
2051
2052         delete aRow;
2053         delete aResult;
2054         
2055         UInt_t startTime = entry->GetStartTime();
2056         UInt_t endTime = entry->GetEndTime();
2057         Bool_t ecsSuccess = entry->GetECSSuccess();
2058         TString runType = entry->GetRunType();
2059         TString tmpdaqstartTime = entry->GetRunParameter("DAQ_time_start");
2060         TString recordingFlagString = entry->GetRunParameter("GDCmStreamRecording");
2061         UInt_t recordingFlag = recordingFlagString.Atoi();
2062         UInt_t daqstartTime = tmpdaqstartTime.Atoi();
2063         
2064         UInt_t now = time(0);
2065         Int_t dcsDelay = fConfig->GetDCSDelay()+fConfig->GetDCSQueryOffset();
2066
2067         Bool_t skip = kFALSE;
2068                                 
2069         // runs are processed if
2070         //   a) runType is PHYSICS and ecsSuccess is set
2071         //   b) runType is not PHYSICS and (ecsSuccess is set or DAQ_time_start is non-0)
2072         // effectively this means that all runs are processed that started properly (ecsSucess behaviour is different for PHYSICS and non-PHYSICS runs (check with ECS!)
2073         if (startTime != 0 && endTime != 0) {
2074                 if (endTime > startTime) { 
2075                         if (endTime >= now - dcsDelay) {
2076                                 Log("SHUTTLE", Form("Skipping run %d for now, because DCS buffer time is not yet expired", run));
2077                         } else {
2078                                 if ((runType == "PHYSICS" || runType == "STANDALONE") && recordingFlag == 0){
2079                                         Log("SHUTTLE", Form("QueryRunParameters - Run type for run %d is %s but the recording is OFF - Skipping!", run, runType.Data()));
2080                                         skip = kTRUE;
2081                                 } 
2082                                 else {
2083                                         if (runType == "PHYSICS") {
2084                                                 if (ecsSuccess) {
2085                                                         return entry;
2086                                                 } else {
2087                                                         Log("SHUTTLE", Form("QueryRunParameters - Run type for run %d is PHYSICS but ECS success flag not set (Reason = %s) - Skipping!", run, entry->GetRunParameter("eor_reason")));
2088                                                         skip = kTRUE;
2089                                                 } 
2090                                         } else {
2091                                                 if (ecsSuccess || daqstartTime > 0) {
2092                                                         if (ecsSuccess == kFALSE)
2093                                                                 Log("SHUTTLE", Form("Processing run %d although in status ECS failure (Reason: %s), since run type != PHYSICS and DAQ_time_start != 0", run, entry->GetRunParameter("eor_reason")));
2094                                                         return entry;
2095                                                 } else {
2096                                                         Log("SHUTTLE", Form("QueryRunParameters - Run type for run %d is %s, ECS success flag was not set (Reason = %s) and DAQ_time_start was NULL - Skipping!", run, runType.Data(), entry->GetRunParameter("eor_reason")));
2097                                                         skip = kTRUE;
2098                                                 }
2099                                         }
2100                                 }
2101                         }
2102                 } else {
2103                         Log("SHUTTLE", Form("QueryRunParameters - Invalid parameters for run %d: startTime equal to endTime: %d %d - Skipping!", run, startTime, endTime));
2104                         skip = kTRUE;
2105                 }
2106         } else {
2107           Log("SHUTTLE", Form("QueryRunParameters - Invalid parameters for Run %d: "
2108                               "startTime = %d, endTime = %d. Skipping (Shuttle won't be marked as DONE)!",
2109                               run, startTime, endTime));
2110         }
2111         
2112         if (skip)
2113         {
2114                 Log("SHUTTLE", Form("Marking SHUTTLE skipped for run %d", run));
2115                 fLogbookEntry = entry;
2116                 if (!UpdateShuttleLogbook("shuttle_skipped"))
2117                 {
2118                         AliError(Form("Could not update logbook for run %d !", run));
2119                 }
2120                 if (!UpdateTableSkippedCase("ALL"))
2121                 {
2122                         AliError(Form("Could not update FXS tables for run %d !", run));
2123                 }
2124                 fLogbookEntry = 0;
2125         }
2126                         
2127         delete entry;
2128         return 0;
2129 }
2130
2131 //______________________________________________________________________________________________
2132 TMap* AliShuttle::GetValueSet(const char* host, Int_t port, const TSeqCollection* entries,
2133                               DCSType type, Int_t multiSplit)
2134 {
2135         // Retrieve all "entry" data points from the DCS server
2136         // host, port: TSocket connection parameters
2137         // entries: list of name of the alias or data point
2138         // type: kAlias or kDP
2139         // returns TMap of values, 0 when failure
2140         
2141         AliDCSClient client(host, port, fTimeout, fRetries, multiSplit);
2142
2143         TMap* result = 0;
2144         if (type == kAlias)
2145         {
2146                 //result = client.GetAliasValues(entries, GetCurrentStartTime()-offset, 
2147                 //      GetCurrentEndTime()+offset);
2148                 result = client.GetAliasValues(entries, GetStartTimeDCSQuery(), 
2149                         GetEndTimeDCSQuery());
2150         } 
2151         else if (type == kDP)
2152         {
2153                 //result = client.GetDPValues(entries, GetCurrentStartTime()-offset, 
2154                 //      GetCurrentEndTime()+offset);
2155                 result = client.GetDPValues(entries, GetStartTimeDCSQuery(), 
2156                         GetEndTimeDCSQuery());
2157         }
2158
2159         if (result == 0)
2160         {
2161                 Log(fCurrentDetector.Data(), Form("GetValueSet - Can't get entries! Reason: %s",
2162                         client.GetErrorString(client.GetResultErrorCode())));
2163                 if (client.GetResultErrorCode() == AliDCSClient::fgkServerError)        
2164                         Log(fCurrentDetector.Data(), Form("GetValueSet - Server error code: %s",
2165                                 client.GetServerError().Data()));
2166
2167                 return 0;
2168         }
2169                 
2170         return result;
2171 }
2172
2173 //______________________________________________________________________________________________
2174 const char* AliShuttle::GetFile(Int_t system, const char* detector,
2175                 const char* id, const char* source)
2176 {
2177         // Get calibration file from file exchange servers
2178         // First queris the FXS database for the file name, using the run, detector, id and source info
2179         // then calls RetrieveFile(filename) for actual copy to local disk
2180         // run: current run being processed (given by Logbook entry fLogbookEntry)
2181         // detector: the Preprocessor name
2182         // id: provided as a parameter by the Preprocessor
2183         // source: provided by the Preprocessor through GetFileSources function
2184
2185         // check if test mode should simulate a FXS error
2186         if (fTestMode & kErrorFXSFiles)
2187         {
2188                 Log(detector, Form("GetFile - In TESTMODE - Simulating error while connecting to %s FXS", GetSystemName(system)));
2189                 return 0;
2190         }
2191         
2192         // check connection, in case connect
2193         if (!Connect(system))
2194         {
2195                 Log(detector, Form("GetFile - Couldn't connect to %s FXS database", GetSystemName(system)));
2196                 fFXSError = system;
2197                 return 0;
2198         }
2199
2200         // Query preparation
2201         TString sourceName(source);
2202         Int_t nFields = 3;
2203         TString sqlQueryStart = Form("select filePath,size,fileChecksum from %s where",
2204                                                                 fConfig->GetFXSdbTable(system));
2205         TString whereClause = Form("run=%d and detector=\"%s\" and fileId=\"%s\"",
2206                                                                 GetCurrentRun(), detector, id);
2207
2208         if (system == kDAQ || system == kDQM)
2209         {
2210                 whereClause += Form(" and DAQsource=\"%s\"", source);
2211         }
2212         else if (system == kDCS)
2213         {
2214                 sourceName="none";
2215         }
2216         else if (system == kHLT)
2217         {
2218                 whereClause += Form(" and DDLnumbers=\"%s\"", source);
2219         }
2220
2221         TString sqlQuery = Form("%s %s", sqlQueryStart.Data(), whereClause.Data());
2222
2223         AliDebug(2, Form("SQL query: \n%s",sqlQuery.Data()));
2224
2225         // Query execution
2226         TSQLResult* aResult = 0;
2227         aResult = dynamic_cast<TSQLResult*> (fServer[system]->Query(sqlQuery));
2228         if (!aResult) {
2229                 Log(detector, Form("GetFile - Can't execute SQL query to %s database for: id = %s, source = %s",
2230                                 GetSystemName(system), id, sourceName.Data()));
2231                 fFXSError = system;
2232                 return 0;
2233         }
2234
2235         if (aResult->GetRowCount() == 0)
2236         {
2237                 Log(detector,
2238                         Form("GetFile - No entry in %s FXS db for: id = %s, source = %s",
2239                                 GetSystemName(system), id, sourceName.Data()));
2240                 delete aResult;
2241                 return 0;
2242         }
2243
2244         if (aResult->GetRowCount() > 1) {
2245                 Log(detector,
2246                         Form("GetFile - More than one entry in %s FXS db for: id = %s, source = %s",
2247                                 GetSystemName(system), id, sourceName.Data()));
2248                 fFXSError = system;
2249                 delete aResult;
2250                 return 0;
2251         }
2252
2253         if (aResult->GetFieldCount() != nFields) {
2254                 Log(detector,
2255                         Form("GetFileName - Wrong field count in %s FXS db for: id = %s, source = %s",
2256                                 GetSystemName(system), id, sourceName.Data()));
2257                 fFXSError = system;
2258                 delete aResult;
2259                 return 0;
2260         }
2261
2262         TSQLRow* aRow = dynamic_cast<TSQLRow*> (aResult->Next());
2263
2264         if (!aRow){
2265                 Log(detector, Form("GetFile - Empty set result in %s FXS db from query: id = %s, source = %s",
2266                                 GetSystemName(system), id, sourceName.Data()));
2267                 fFXSError = system;
2268                 delete aResult;
2269                 return 0;
2270         }
2271
2272         TString filePath(aRow->GetField(0), aRow->GetFieldLength(0));
2273         TString fileSize(aRow->GetField(1), aRow->GetFieldLength(1));
2274         TString fileChecksum(aRow->GetField(2), aRow->GetFieldLength(2));
2275
2276         delete aResult;
2277         delete aRow;
2278
2279         AliDebug(2, Form("filePath = %s; size = %s, fileChecksum = %s",
2280                                 filePath.Data(), fileSize.Data(), fileChecksum.Data()));
2281
2282         // retrieved file is renamed to make it unique
2283         Int_t dir_lev1 = GetCurrentRun()/10000;
2284         TString localFileName = Form("%s/%d/%d/%s_process/%s_%s_%d_%s_%s.shuttle",
2285                                      GetShuttleTempDir(), dir_lev1, GetCurrentRun(), detector,
2286                                         GetSystemName(system), detector, GetCurrentRun(), 
2287                                         id, sourceName.Data());
2288         Log("SHUTTLE",Form("file from FXS = %s",localFileName.Data())); 
2289
2290
2291         // file retrieval from FXS
2292         UInt_t nRetries = 0;
2293         UInt_t maxRetries = 3;
2294         Bool_t result = kFALSE;
2295
2296         // copy!! if successful TSystem::Exec returns 0
2297         while (nRetries++ < maxRetries) {
2298                 AliDebug(2, Form("Trying to copy file. Retry # %d", nRetries));
2299                 result = RetrieveFile(system, filePath.Data(), localFileName.Data());
2300                 if (!result)
2301                 {
2302                         Log(detector, Form("GetFile - Copy of file %s from %s FXS failed",
2303                                         filePath.Data(), GetSystemName(system)));
2304                         continue;
2305                 } 
2306
2307                 if (fileSize.Length()>0)
2308                 {
2309                         // compare filesize of local file with the one stored in the FXS DB
2310                         Long_t size = -1;
2311                         Int_t sizeComp = gSystem->GetPathInfo(localFileName.Data(), 0, &size, 0, 0);
2312
2313                         if (sizeComp != 0 || size != fileSize.Atoi())
2314                         {
2315                                 Log(detector, Form("GetFile - size of file %s does not match with local copy!",
2316                                                         filePath.Data()));
2317                                 result = kFALSE;
2318                                 continue;
2319                         }
2320
2321                 } else {
2322                         Log(fCurrentDetector, Form("GetFile - size of file %s not set in %s database, skipping comparison",
2323                                                 filePath.Data(), GetSystemName(system)));
2324                 }
2325
2326                 if (fileChecksum.Length()>0)
2327                 {
2328                         // compare md5sum of local file with the one stored in the FXS DB
2329                         if(fileChecksum.Contains(' ')) fileChecksum.Resize(fileChecksum.First(' '));
2330                         Int_t md5Comp = gSystem->Exec(Form("md5sum %s |grep %s > /dev/null 2> /dev/null",
2331                                                 localFileName.Data(), fileChecksum.Data()));
2332
2333                         if (md5Comp != 0)
2334                         {
2335                                 Log(detector, Form("GetFile - md5sum of file %s does not match with local copy!",
2336                                                         filePath.Data()));
2337                                 result = kFALSE;
2338                                 continue;
2339                         }
2340                 } else {
2341                         Log(fCurrentDetector, Form("GetFile - md5sum of file %s not set in %s database, skipping comparison",
2342                                                         filePath.Data(), GetSystemName(system)));
2343                 }
2344                 if (result) break;
2345         }
2346
2347         if (!result) 
2348         {
2349                 fFXSError = system;
2350                 return 0;
2351         }
2352
2353         fFXSCalled[system]=kTRUE;
2354         TObjString *fileParams = new TObjString(Form("%s#!?!#%s", id, sourceName.Data()));
2355         fFXSlist[system].Add(fileParams);
2356
2357         static TString staticLocalFileName;
2358         staticLocalFileName.Form("%s", localFileName.Data());
2359         
2360         Log(fCurrentDetector, Form("GetFile - Retrieved file with id %s and "
2361                         "source %s from %s to %s", id, source, 
2362                         GetSystemName(system), localFileName.Data()));
2363                         
2364         return staticLocalFileName.Data();
2365 }
2366
2367 //______________________________________________________________________________________________
2368 Bool_t AliShuttle::RetrieveFile(UInt_t system, const char* fxsFileName, const char* localFileName)
2369 {
2370         //
2371         // Copies file from FXS to local Shuttle machine
2372         //
2373
2374         // check temp directory: trying to cd to temp; if it does not exist, create it
2375         AliDebug(2, Form("Copy file %s from %s FXS into %s",
2376                         GetSystemName(system), fxsFileName, localFileName));
2377                         
2378         TString tmpDir(localFileName);
2379         
2380         tmpDir = tmpDir(0,tmpDir.Last('/'));
2381
2382         Int_t noDir = gSystem->GetPathInfo(tmpDir.Data(), 0, (Long64_t*) 0, 0, 0);
2383         if (noDir) // temp dir does not exists!
2384         {
2385                 if (gSystem->mkdir(tmpDir.Data(), 1))
2386                 {
2387                         Log(fCurrentDetector.Data(), "RetrieveFile - could not make temp directory!!");
2388                         return kFALSE;
2389                 }
2390         }
2391
2392         TString command = Form("scp -oPort=%d -2 %s@%s:%s/%s %s",
2393                 fConfig->GetFXSPort(system),
2394                 fConfig->GetFXSUser(system),
2395                 fConfig->GetFXSHost(system),
2396                 fConfig->GetFXSBaseFolder(system),
2397                 fxsFileName,
2398                 localFileName);
2399
2400         AliDebug(2, Form("%s",command.Data()));
2401
2402         Bool_t result = (gSystem->Exec(command.Data()) == 0);
2403
2404         return result;
2405 }
2406
2407 //______________________________________________________________________________________________
2408 TList* AliShuttle::GetFileSources(Int_t system, const char* detector, const char* id)
2409 {
2410         //
2411         // Get sources producing the condition file Id from file exchange servers
2412         // if id is NULL all sources are returned (distinct)
2413         //
2414
2415         if (id)
2416         {
2417                 Log(detector, Form("GetFileSources - Querying %s FXS for files with id %s produced by %s", GetSystemName(system), id, detector));
2418         } else {
2419                 Log(detector, Form("GetFileSources - Querying %s FXS for files produced by %s", GetSystemName(system), detector));
2420         }
2421         
2422         // check if test mode should simulate a FXS error
2423         if (fTestMode & kErrorFXSSources)
2424         {
2425                 Log(detector, Form("GetFileSources - In TESTMODE - Simulating error while connecting to %s FXS", GetSystemName(system)));
2426                 return 0;
2427         }
2428
2429         if (system == kDCS)
2430         {
2431                 Log(detector, "GetFileSources - WARNING: DCS system has only one source of data!");
2432                 TList *list = new TList();
2433                 list->SetOwner(1);
2434                 list->Add(new TObjString(" "));
2435                 return list;
2436         }
2437
2438         // check connection, in case connect
2439         if (!Connect(system))
2440         {
2441                 Log(detector, Form("GetFileSources - Couldn't connect to %s FXS database", GetSystemName(system)));
2442                 fFXSError = system;
2443                 return NULL;
2444         }
2445
2446         TString sourceName = "";
2447         if (system == kDAQ || system == kDQM)
2448         {
2449                 sourceName = "DAQsource";
2450         } else if (system == kHLT)
2451         {
2452                 sourceName = "DDLnumbers";
2453         }
2454
2455         TString sqlQueryStart = Form("select distinct %s from %s where", sourceName.Data(), fConfig->GetFXSdbTable(system));
2456         TString whereClause = Form("run=%d and detector=\"%s\"",
2457                                 GetCurrentRun(), detector);
2458         if (id)
2459                 whereClause += Form(" and fileId=\"%s\"", id);
2460         TString sqlQuery = Form("%s %s", sqlQueryStart.Data(), whereClause.Data());
2461
2462         AliDebug(2, Form("SQL query: \n%s",sqlQuery.Data()));
2463
2464         // Query execution
2465         TSQLResult* aResult;
2466         aResult = fServer[system]->Query(sqlQuery);
2467         if (!aResult) {
2468                 Log(detector, Form("GetFileSources - Can't execute SQL query to %s database for id: %s",
2469                                 GetSystemName(system), id));
2470                 fFXSError = system;
2471                 return 0;
2472         }
2473
2474         TList *list = new TList();
2475         list->SetOwner(1);
2476         
2477         if (aResult->GetRowCount() == 0)
2478         {
2479                 Log(detector,
2480                         Form("GetFileSources - No entry in %s FXS table for id: %s", GetSystemName(system), id));
2481                 delete aResult;
2482                 return list;
2483         }
2484
2485         Log(detector, Form("GetFileSources - Found %d sources", aResult->GetRowCount()));
2486
2487         TSQLRow* aRow;
2488         while ((aRow = aResult->Next()))
2489         {
2490
2491                 TString source(aRow->GetField(0), aRow->GetFieldLength(0));
2492                 AliDebug(2, Form("%s = %s", sourceName.Data(), source.Data()));
2493                 list->Add(new TObjString(source));
2494                 delete aRow;
2495         }
2496
2497         delete aResult;
2498
2499         return list;
2500 }
2501
2502 //______________________________________________________________________________________________
2503 TList* AliShuttle::GetFileIDs(Int_t system, const char* detector, const char* source)
2504 {
2505         //
2506         // Get all ids of condition files produced by a given source from file exchange servers
2507         //
2508         
2509         Log(detector, Form("GetFileIDs - Retrieving ids with source %s with %s", source, GetSystemName(system)));
2510
2511         // check if test mode should simulate a FXS error
2512         if (fTestMode & kErrorFXSSources)
2513         {
2514                 Log(detector, Form("GetFileIDs - In TESTMODE - Simulating error while connecting to %s FXS", GetSystemName(system)));
2515                 return 0;
2516         }
2517
2518         // check connection, in case connect
2519         if (!Connect(system))
2520         {
2521                 Log(detector, Form("GetFileIDs - Couldn't connect to %s FXS database", GetSystemName(system)));
2522                 return NULL;
2523         }
2524
2525         TString sourceName = "";
2526         if (system == kDAQ)
2527         {
2528                 sourceName = "DAQsource";
2529         } else if (system == kHLT)
2530         {
2531                 sourceName = "DDLnumbers";
2532         }
2533
2534         TString sqlQueryStart = Form("select fileId from %s where", fConfig->GetFXSdbTable(system));
2535         TString whereClause = Form("run=%d and detector=\"%s\"",
2536                                 GetCurrentRun(), detector);
2537         if (sourceName.Length() > 0 && source)
2538                 whereClause += Form(" and %s=\"%s\"", sourceName.Data(), source);
2539         TString sqlQuery = Form("%s %s", sqlQueryStart.Data(), whereClause.Data());
2540
2541         AliDebug(2, Form("SQL query: \n%s",sqlQuery.Data()));
2542
2543         // Query execution
2544         TSQLResult* aResult;
2545         aResult = fServer[system]->Query(sqlQuery);
2546         if (!aResult) {
2547                 Log(detector, Form("GetFileIDs - Can't execute SQL query to %s database for source: %s",
2548                                 GetSystemName(system), source));
2549                 return 0;
2550         }
2551
2552         TList *list = new TList();
2553         list->SetOwner(1);
2554         
2555         if (aResult->GetRowCount() == 0)
2556         {
2557                 Log(detector,
2558                         Form("GetFileIDs - No entry in %s FXS table for source: %s", GetSystemName(system), source));
2559                 delete aResult;
2560                 return list;
2561         }
2562
2563         Log(detector, Form("GetFileIDs - Found %d ids", aResult->GetRowCount()));
2564
2565         TSQLRow* aRow;
2566
2567         while ((aRow = aResult->Next()))
2568         {
2569
2570                 TString id(aRow->GetField(0), aRow->GetFieldLength(0));
2571                 AliDebug(2, Form("fileId = %s", id.Data()));
2572                 list->Add(new TObjString(id));
2573                 delete aRow;
2574         }
2575
2576         delete aResult;
2577
2578         return list;
2579 }
2580
2581 //______________________________________________________________________________________________
2582 Bool_t AliShuttle::Connect(Int_t system)
2583 {
2584         // Connect to MySQL Server of the system's FXS MySQL databases
2585         // DAQ Logbook, Shuttle Logbook and DAQ FXS db are on the same host
2586         //
2587
2588         // check connection: if already connected return
2589
2590         if(fServer[system] && fServer[system]->IsConnected()) {
2591                 // ping the server              
2592                 if (fServer[system]->PingVerify()==kTRUE){ // connection is still alive
2593                         return kTRUE;
2594                 }               
2595                 else{
2596                         AliWarning(Form("Connection got lost to FXS database for %s. Closing and reconnecting.",
2597                                         AliShuttleInterface::GetSystemName(system)));
2598                         fServer[system]->Close();
2599                         delete fServer[system];
2600                         fServer[system] = 0x0;
2601                 }
2602         }
2603
2604         TString dbHost, dbUser, dbPass, dbName;
2605
2606         if (system < 4) // FXS db servers
2607         {
2608                 dbHost = Form("mysql://%s:%d", fConfig->GetFXSdbHost(system), fConfig->GetFXSdbPort(system));
2609                 dbUser = fConfig->GetFXSdbUser(system);
2610                 dbPass = fConfig->GetFXSdbPass(system);
2611                 dbName =   fConfig->GetFXSdbName(system);
2612         } else { // Run & Shuttle logbook servers
2613         // TODO Will the Shuttle logbook server be the same as the Run logbook server ???
2614                 dbHost = Form("mysql://%s:%d", fConfig->GetDAQlbHost(), fConfig->GetDAQlbPort());
2615                 dbUser = fConfig->GetDAQlbUser();
2616                 dbPass = fConfig->GetDAQlbPass();
2617                 dbName =   fConfig->GetDAQlbDB();
2618         }
2619
2620         fServer[system] = TSQLServer::Connect(dbHost.Data(), dbUser.Data(), dbPass.Data());
2621                 if (!fServer[system] || !fServer[system]->IsConnected()) {
2622                 if(system < 4)
2623                 {
2624                 AliError(Form("Can't establish connection to FXS database for %s",
2625                                         AliShuttleInterface::GetSystemName(system)));
2626                 } else {
2627                 AliError("Can't establish connection to Run logbook.");
2628                 }
2629                 if(fServer[system]) delete fServer[system];
2630                 return kFALSE;
2631         }
2632
2633         // Get tables
2634         TSQLResult* aResult=0;
2635         switch(system){
2636                 case kDAQ:
2637                         aResult = fServer[kDAQ]->GetTables(dbName.Data());
2638                         break;
2639                 case kDCS:
2640                         aResult = fServer[kDCS]->GetTables(dbName.Data());
2641                         break;
2642                 case kHLT:
2643                         aResult = fServer[kHLT]->GetTables(dbName.Data());
2644                         break;
2645                 case kDQM:
2646                         aResult = fServer[kDQM]->GetTables(dbName.Data());
2647                         break;
2648                 default:
2649                         aResult = fServer[4]->GetTables(dbName.Data());
2650                         break;
2651         }
2652
2653         delete aResult;
2654         return kTRUE;
2655 }
2656
2657 //______________________________________________________________________________________________
2658 Bool_t AliShuttle::UpdateTable()
2659 {
2660         //
2661         // Update FXS table filling time_processed field in all rows corresponding to current run and detector
2662         //
2663
2664         Bool_t result = kTRUE;
2665
2666         for (UInt_t system=0; system<3; system++)
2667         {
2668                 if(!fFXSCalled[system]) continue;
2669
2670                 // check connection, in case connect
2671                 if (!Connect(system))
2672                 {
2673                         Log(fCurrentDetector, Form("UpdateTable - Couldn't connect to %s FXS database", GetSystemName(system)));
2674                         result = kFALSE;
2675                         continue;
2676                 }
2677
2678                 TTimeStamp now; // now
2679
2680                 // Loop on FXS list entries
2681                 TIter iter(&fFXSlist[system]);
2682                 TObjString *aFXSentry=0;
2683                 while ((aFXSentry = dynamic_cast<TObjString*> (iter.Next())))
2684                 {
2685                         TString aFXSentrystr = aFXSentry->String();
2686                         TObjArray *aFXSarray = aFXSentrystr.Tokenize("#!?!#");
2687                         if (!aFXSarray || aFXSarray->GetEntries() != 2 )
2688                         {
2689                                 Log(fCurrentDetector, Form("UpdateTable - error updating %s FXS entry. Check string: <%s>",
2690                                         GetSystemName(system), aFXSentrystr.Data()));
2691                                 if(aFXSarray) delete aFXSarray;
2692                                 result = kFALSE;
2693                                 continue;
2694                         }
2695                         const char* fileId = ((TObjString*) aFXSarray->At(0))->GetName();
2696                         const char* source = ((TObjString*) aFXSarray->At(1))->GetName();
2697
2698                         TString whereClause;
2699                         if (system == kDAQ || system == kDQM)
2700                         {
2701                                 whereClause = Form("where run=%d and detector=\"%s\" and fileId=\"%s\" and DAQsource=\"%s\";",
2702                                                         GetCurrentRun(), fCurrentDetector.Data(), fileId, source);
2703                         }
2704                         else if (system == kDCS)
2705                         {
2706                                 whereClause = Form("where run=%d and detector=\"%s\" and fileId=\"%s\";",
2707                                                         GetCurrentRun(), fCurrentDetector.Data(), fileId);
2708                         }
2709                         else if (system == kHLT)
2710                         {
2711                                 whereClause = Form("where run=%d and detector=\"%s\" and fileId=\"%s\" and DDLnumbers=\"%s\";",
2712                                                         GetCurrentRun(), fCurrentDetector.Data(), fileId, source);
2713                         }
2714
2715                         delete aFXSarray;
2716
2717                         TString sqlQuery = Form("update %s set time_processed=%ld %s", fConfig->GetFXSdbTable(system),
2718                                                 (ULong_t)now.GetSec(), whereClause.Data());
2719
2720                         AliDebug(2, Form("SQL query: \n%s",sqlQuery.Data()));
2721
2722                         // Query execution
2723                         TSQLResult* aResult;
2724                         aResult = dynamic_cast<TSQLResult*> (fServer[system]->Query(sqlQuery));
2725                         if (!aResult)
2726                         {
2727                                 Log(fCurrentDetector, Form("UpdateTable - %s db: can't execute SQL query <%s>",
2728                                                                 GetSystemName(system), sqlQuery.Data()));
2729                                 result = kFALSE;
2730                                 continue;
2731                         }
2732                         delete aResult;
2733                 }
2734         }
2735
2736         return result;
2737 }
2738
2739 //_______________________________________________________________________________
2740 Bool_t AliShuttle::UpdateTableSkippedCase(const char* detector)
2741 {
2742         //
2743         // Update FXS table filling time_processed field in all rows corresponding to current run and detector
2744         // if detector = "ALL" update all detectors
2745         //
2746
2747         Bool_t result = kTRUE;
2748
2749         TString detName(detector);
2750
2751         for (UInt_t system=0; system<3; system++)
2752         {
2753
2754                 // check connection, in case connect
2755                 if (!Connect(system))
2756                 {
2757                         Log(fCurrentDetector, Form("UpdateTableSkippedCase - Couldn't connect to %s FXS database", GetSystemName(system)));
2758                         result = kFALSE;
2759                         continue;
2760                 }
2761
2762                 TTimeStamp now; // now
2763
2764                 // Loop on FXS list entries
2765                 TIter iter(&fFXSlist[system]);
2766                         
2767                 TString whereClause;
2768                 if (detName == "ALL") whereClause = Form("where run=%d and time_processed IS NULL;",GetCurrentRun());
2769                 else whereClause = Form("where run=%d and detector=\"%s\" and time_processed IS NULL;",GetCurrentRun(), detector);
2770
2771                 //Log("SHUTTLE",Form(" whereClause = %s ",whereClause.Data()));
2772
2773                 TString sqlQuery = Form("update %s set time_processed=%ld %s", fConfig->GetFXSdbTable(system),
2774                                         (ULong_t)now.GetSec(), whereClause.Data());
2775
2776                 AliDebug(2, Form("SQL query: \n%s",sqlQuery.Data()));
2777
2778                 // Query execution
2779                 TSQLResult* aResult;
2780                 aResult = dynamic_cast<TSQLResult*> (fServer[system]->Query(sqlQuery));
2781                 if (!aResult)
2782                 {
2783                         Log("SHUTTLE", Form("UpdateTableSkippedCase - %s db: can't execute SQL query <%s>",
2784                                                         GetSystemName(system), sqlQuery.Data()));
2785                         result = kFALSE;
2786                         continue;
2787                 }
2788                 delete aResult;
2789                 
2790         }
2791
2792         return result;
2793 }
2794 //______________________________________________________________________________________________
2795 Bool_t AliShuttle::UpdateTableFailCase()
2796 {
2797         // Update FXS table filling time_processed field in all rows corresponding to current run and detector
2798         // this is called in case the preprocessor is declared failed for the current run, because
2799         // the fields are updated only in case of success
2800
2801         Bool_t result = kTRUE;
2802
2803         for (UInt_t system=0; system<3; system++)
2804         {
2805                 // check connection, in case connect
2806                 if (!Connect(system))
2807                 {
2808                         Log(fCurrentDetector, Form("UpdateTableFailCase - Couldn't connect to %s FXS database",
2809                                                         GetSystemName(system)));
2810                         result = kFALSE;
2811                         continue;
2812                 }
2813
2814                 TTimeStamp now; // now
2815
2816                 // Loop on FXS list entries
2817
2818                 TString whereClause = Form("where run=%d and detector=\"%s\";",
2819                                                 GetCurrentRun(), fCurrentDetector.Data());
2820
2821
2822                 TString sqlQuery = Form("update %s set time_processed=%ld %s", fConfig->GetFXSdbTable(system),
2823                                         (ULong_t)now.GetSec(), whereClause.Data());
2824
2825                 AliDebug(2, Form("SQL query: \n%s",sqlQuery.Data()));
2826
2827                 // Query execution
2828                 TSQLResult* aResult;
2829                 aResult = dynamic_cast<TSQLResult*> (fServer[system]->Query(sqlQuery));
2830                 if (!aResult)
2831                 {
2832                         Log(fCurrentDetector, Form("UpdateTableFailCase - %s db: can't execute SQL query <%s>",
2833                                                         GetSystemName(system), sqlQuery.Data()));
2834                         result = kFALSE;
2835                         continue;
2836                 }
2837                 delete aResult;
2838         }
2839
2840         return result;
2841 }
2842
2843 //______________________________________________________________________________________________
2844 Bool_t AliShuttle::UpdateShuttleLogbook(const char* detector, const char* status)
2845 {
2846         //
2847         // Update Shuttle logbook filling detector or shuttle_done column
2848         // ex. of usage: UpdateShuttleLogbook("PHOS", "DONE") or UpdateShuttleLogbook("shuttle_done")
2849         //
2850
2851         // check connection, in case connect
2852         if(!Connect(4)){
2853                 Log("SHUTTLE", "UpdateShuttleLogbook - Couldn't connect to DAQ Logbook.");
2854                 return kFALSE;
2855         }
2856
2857         TString detName(detector);
2858         TString setClause;
2859         if (detName == "shuttle_done" || detName == "shuttle_skipped")
2860         {
2861                 setClause = "set shuttle_done=1";
2862                 
2863                 if (detName == "shuttle_done")
2864                 {
2865                         if (TouchFile() != kTRUE)
2866                         {
2867                                 SendMLRunInfo("Pending");
2868                                 return kFALSE;
2869                         }
2870                         
2871                         SendMLRunInfo("Done");
2872                 }
2873                 else 
2874                         SendMLRunInfo("Skipped");
2875         } 
2876         else {
2877                 TString statusStr(status);
2878                 if(statusStr.Contains("done", TString::kIgnoreCase) ||
2879                    statusStr.Contains("failed", TString::kIgnoreCase)){
2880                         setClause = Form("set %s=\"%s\"", detector, status);
2881                 } else {
2882                         Log("SHUTTLE",
2883                                 Form("UpdateShuttleLogbook - Invalid status <%s> for detector %s",
2884                                         status, detector));
2885                         return kFALSE;
2886                 }
2887         }
2888
2889         TString whereClause = Form("where run=%d", GetCurrentRun());
2890
2891         TString sqlQuery = Form("update %s %s %s",
2892                                         fConfig->GetShuttlelbTable(), setClause.Data(), whereClause.Data());
2893
2894         AliDebug(2, Form("SQL query: \n%s",sqlQuery.Data()));
2895
2896         // Query execution
2897         TSQLResult* aResult;
2898         aResult = dynamic_cast<TSQLResult*> (fServer[4]->Query(sqlQuery));
2899         if (!aResult) {
2900                 Log("SHUTTLE", Form("UpdateShuttleLogbook - Can't execute query <%s>", sqlQuery.Data()));
2901                 return kFALSE;
2902         }
2903         delete aResult;
2904
2905         return kTRUE;
2906 }
2907
2908 //______________________________________________________________________________________________
2909 Int_t AliShuttle::GetCurrentRun() const
2910 {
2911         //
2912         // Get current run from logbook entry
2913         //
2914
2915         return fLogbookEntry ? fLogbookEntry->GetRun() : -1;
2916 }
2917
2918 //______________________________________________________________________________________________
2919 UInt_t AliShuttle::GetCurrentStartTime() const
2920 {
2921         //
2922         // get current start time
2923         //
2924
2925         return fLogbookEntry ? fLogbookEntry->GetStartTime() : 0;
2926 }
2927
2928 //______________________________________________________________________________________________
2929 UInt_t AliShuttle::GetCurrentEndTime() const
2930 {
2931         //
2932         // get current end time from logbook entry
2933         //
2934
2935         return fLogbookEntry ? fLogbookEntry->GetEndTime() : 0;
2936 }
2937 //______________________________________________________________________________________________
2938 UInt_t AliShuttle::GetCurrentYear() const
2939 {
2940         //
2941         // Get current year from logbook entry
2942         //
2943
2944         if (!fLogbookEntry) return 0;
2945         
2946         TTimeStamp startTime(GetCurrentStartTime());
2947         TString year =  Form("%d",startTime.GetDate());
2948         year = year(0,4);
2949         
2950         return year.Atoi();
2951 }
2952
2953 //______________________________________________________________________________________________
2954 const char* AliShuttle::GetLHCPeriod() const
2955 {
2956         //
2957         // Get current LHC period from logbook entry
2958         //
2959
2960         if (!fLogbookEntry) return 0;
2961                 
2962         return fLogbookEntry->GetRunParameter("LHCperiod");
2963 }
2964
2965 //______________________________________________________________________________________________
2966 void AliShuttle::Log(const char* detector, const char* message, UInt_t level)
2967 {
2968         //
2969         // Fill log string with a message
2970         //
2971         
2972         TString logRunDir = GetShuttleLogDir();
2973         if (GetCurrentRun() >=0) {
2974                 Int_t logDir_lev1 = GetCurrentRun()/10000;
2975                 logRunDir += Form("/%d/%d", logDir_lev1, GetCurrentRun());
2976         }               
2977         void* dir = gSystem->OpenDirectory(logRunDir.Data());
2978         if (dir == NULL) {
2979                 if (gSystem->mkdir(logRunDir.Data(), kTRUE)) {
2980                         AliError(Form("Can't open directory <%s>", GetShuttleLogDir()));
2981                         return;
2982                 }
2983
2984         } else {
2985                 gSystem->FreeDirectory(dir);
2986         }
2987
2988         TString toLog = Form("%s UTC (%d): %s - ", TTimeStamp(time(0)).AsString("s"), getpid(), detector);
2989         if (GetCurrentRun() >= 0) 
2990                 toLog += Form("run %d - ", GetCurrentRun());
2991         toLog += Form("%s", message);
2992
2993         AliLog::Message(level, toLog, MODULENAME(), ClassName(), FUNCTIONNAME(), __FILE__, __LINE__);
2994         
2995         // if we redirect the log output already to the file, leave here
2996         if (fOutputRedirected && strcmp(detector, "SHUTTLE") != 0)
2997                 return;
2998
2999         TString fileName = GetLogFileName(detector);
3000         
3001         gSystem->ExpandPathName(fileName);
3002