memory consumption measures only memory used by preprocessor
[u/mrichter/AliRoot.git] / SHUTTLE / AliShuttle.cxx
1 /**************************************************************************
2  * Copyright(c) 1998-1999, ALICE Experiment at CERN, All rights reserved. *
3  *                                                                        *
4  * Author: The ALICE Off-line Project.                                    *
5  * Contributors are mentioned in the code where appropriate.              *
6  *                                                                        *
7  * Permission to use, copy, modify and distribute this software and its   *
8  * documentation strictly for non-commercial purposes is hereby granted   *
9  * without fee, provided that the above copyright notice appears in all   *
10  * copies and that both the copyright notice and this permission notice   *
11  * appear in the supporting documentation. The authors make no claims     *
12  * about the suitability of this software for any purpose. It is          *
13  * provided "as is" without express or implied warranty.                  *
14  **************************************************************************/
15
16 /* $Id$ */
17
18 //
19 // This class is the main manager for AliShuttle. 
20 // It organizes the data retrieval from DCS and call the 
21 // interface methods of AliPreprocessor.
22 // For every detector in AliShuttleConfgi (see AliShuttleConfig),
23 // data for its set of aliases is retrieved. If there is registered
24 // AliPreprocessor for this detector then it will be used
25 // accroding to the schema (see AliPreprocessor).
26 // If there isn't registered AliPreprocessor than the retrieved
27 // data is stored automatically to the undelying AliCDBStorage.
28 // For detSpec is used the alias name.
29 //
30
31 #include "AliShuttle.h"
32
33 #include "AliCDBManager.h"
34 #include "AliCDBStorage.h"
35 #include "AliCDBId.h"
36 #include "AliCDBRunRange.h"
37 #include "AliCDBPath.h"
38 #include "AliCDBEntry.h"
39 #include "AliShuttleConfig.h"
40 #include "DCSClient/AliDCSClient.h"
41 #include "AliLog.h"
42 #include "AliPreprocessor.h"
43 #include "AliShuttleStatus.h"
44 #include "AliShuttleLogbookEntry.h"
45
46 #include <TSystem.h>
47 #include <TObject.h>
48 #include <TString.h>
49 #include <TTimeStamp.h>
50 #include <TObjString.h>
51 #include <TSQLServer.h>
52 #include <TSQLResult.h>
53 #include <TSQLRow.h>
54 #include <TMutex.h>
55 #include <TSystemDirectory.h>
56 #include <TSystemFile.h>
57 #include <TFile.h>
58 #include <TGrid.h>
59 #include <TGridResult.h>
60 #include <TMap.h>
61
62 #include <TMonaLisaWriter.h>
63
64 #include <fstream>
65
66 #include <sys/types.h>
67 #include <sys/wait.h>
68
69 #include <signal.h>
70
71 ClassImp(AliShuttle)
72
73 //______________________________________________________________________________________________
74 AliShuttle::AliShuttle(const AliShuttleConfig* config,
75                 UInt_t timeout, Int_t retries):
76 fConfig(config),
77 fTimeout(timeout), fRetries(retries),
78 fPreprocessorMap(),
79 fLogbookEntry(0),
80 fCurrentDetector(),
81 fFirstProcessing(0),
82 fFXSError(-1),
83 fStatusEntry(0),
84 fMonitoringMutex(0),
85 fLastActionTime(0),
86 fLastAction(),
87 fMonaLisa(0),
88 fTestMode(kNone),
89 fReadTestMode(kFALSE),
90 fOutputRedirected(kFALSE)
91 {
92         //
93         // config: AliShuttleConfig used
94         // timeout: timeout used for AliDCSClient connection
95         // retries: the number of retries in case of connection error.
96         //
97
98         if (!fConfig->IsValid()) AliFatal("********** !!!!! Invalid configuration !!!!! **********");
99         for(int iSys=0;iSys<4;iSys++) {
100                 fServer[iSys]=0;
101                 if (iSys < 3)
102                         fFXSlist[iSys].SetOwner(kTRUE);
103         }
104         fPreprocessorMap.SetOwner(kTRUE);
105
106         for (UInt_t iDet=0; iDet<NDetectors(); iDet++)
107                 fFirstUnprocessed[iDet] = kFALSE;
108
109         fMonitoringMutex = new TMutex();
110 }
111
112 //______________________________________________________________________________________________
113 AliShuttle::~AliShuttle()
114 {
115         //
116         // destructor
117         //
118
119         fPreprocessorMap.DeleteAll();
120         for(int iSys=0;iSys<4;iSys++)
121                 if(fServer[iSys]) {
122                         fServer[iSys]->Close();
123                         delete fServer[iSys];
124                         fServer[iSys] = 0;
125                 }
126
127         if (fStatusEntry){
128                 delete fStatusEntry;
129                 fStatusEntry = 0;
130         }
131         
132         if (fMonitoringMutex) 
133         {
134                 delete fMonitoringMutex;
135                 fMonitoringMutex = 0;
136         }
137 }
138
139 //______________________________________________________________________________________________
140 void AliShuttle::RegisterPreprocessor(AliPreprocessor* preprocessor)
141 {
142         //
143         // Registers new AliPreprocessor.
144         // It uses GetName() for indentificator of the pre processor.
145         // The pre processor is registered it there isn't any other
146         // with the same identificator (GetName()).
147         //
148
149         const char* detName = preprocessor->GetName();
150         if(GetDetPos(detName) < 0)
151                 AliFatal(Form("********** !!!!! Invalid detector name: %s !!!!! **********", detName));
152
153         if (fPreprocessorMap.GetValue(detName)) {
154                 AliWarning(Form("AliPreprocessor %s is already registered!", detName));
155                 return;
156         }
157
158         fPreprocessorMap.Add(new TObjString(detName), preprocessor);
159 }
160 //______________________________________________________________________________________________
161 Bool_t AliShuttle::Store(const AliCDBPath& path, TObject* object,
162                 AliCDBMetaData* metaData, Int_t validityStart, Bool_t validityInfinite)
163 {
164         // Stores a CDB object in the storage for offline reconstruction. Objects that are not needed for
165         // offline reconstruction, but should be stored anyway (e.g. for debugging) should NOT be stored
166         // using this function. Use StoreReferenceData instead!
167         // It calls StoreLocally function which temporarily stores the data locally; when the preprocessor
168         // finishes the data are transferred to the main storage (Grid).
169
170         return StoreLocally(fgkLocalCDB, path, object, metaData, validityStart, validityInfinite);
171 }
172
173 //______________________________________________________________________________________________
174 Bool_t AliShuttle::StoreReferenceData(const AliCDBPath& path, TObject* object, AliCDBMetaData* metaData)
175 {
176         // Stores a CDB object in the storage for reference data. This objects will not be available during
177         // offline reconstrunction. Use this function for reference data only!
178         // It calls StoreLocally function which temporarily stores the data locally; when the preprocessor
179         // finishes the data are transferred to the main storage (Grid).
180
181         return StoreLocally(fgkLocalRefStorage, path, object, metaData);
182 }
183
184 //______________________________________________________________________________________________
185 Bool_t AliShuttle::StoreLocally(const TString& localUri,
186                         const AliCDBPath& path, TObject* object, AliCDBMetaData* metaData,
187                         Int_t validityStart, Bool_t validityInfinite)
188 {
189         // Store object temporarily in local storage. Parameters are passed by Store and StoreReferenceData functions.
190         // when the preprocessor finishes the data are transferred to the main storage (Grid).
191         // The parameters are:
192         //   1) Uri of the backup storage (Local)
193         //   2) the object's path.
194         //   3) the object to be stored
195         //   4) the metaData to be associated with the object
196         //   5) the validity start run number w.r.t. the current run,
197         //      if the data is valid only for this run leave the default 0
198         //   6) specifies if the calibration data is valid for infinity (this means until updated),
199         //      typical for calibration runs, the default is kFALSE
200         //
201         // returns 0 if fail, 1 otherwise
202
203         if (fTestMode & kErrorStorage)
204         {
205                 Log(fCurrentDetector, "StoreLocally - In TESTMODE - Simulating error while storing locally");
206                 return kFALSE;
207         }
208         
209         const char* cdbType = (localUri == fgkLocalCDB) ? "CDB" : "Reference";
210
211         Int_t firstRun = GetCurrentRun() - validityStart;
212         if(firstRun < 0) {
213                 AliWarning("First valid run happens to be less than 0! Setting it to 0.");
214                 firstRun=0;
215         }
216
217         Int_t lastRun = -1;
218         if(validityInfinite) {
219                 lastRun = AliCDBRunRange::Infinity();
220         } else {
221                 lastRun = GetCurrentRun();
222         }
223
224         // Version is set to current run, it will be used later to transfer data to Grid
225         AliCDBId id(path, firstRun, lastRun, GetCurrentRun(), -1);
226
227         if(! dynamic_cast<TObjString*> (metaData->GetProperty("RunUsed(TObjString)"))){
228                 TObjString runUsed = Form("%d", GetCurrentRun());
229                 metaData->SetProperty("RunUsed(TObjString)", runUsed.Clone());
230         }
231
232         Bool_t result = kFALSE;
233
234         if (!(AliCDBManager::Instance()->GetStorage(localUri))) {
235                 Log("SHUTTLE", Form("StoreLocally - Cannot activate local %s storage", cdbType));
236         } else {
237                 result = AliCDBManager::Instance()->GetStorage(localUri)
238                                         ->Put(object, id, metaData);
239         }
240
241         if(!result) {
242
243                 Log(fCurrentDetector, Form("StoreLocally - Can't store object <%s>!", id.ToString().Data()));
244         }
245
246         return result;
247 }
248
249 //______________________________________________________________________________________________
250 Bool_t AliShuttle::StoreOCDB()
251 {
252         //
253         // Called when preprocessor ends successfully or when previous storage attempt failed (kStoreError status)
254         // Calls underlying StoreOCDB(const char*) function twice, for OCDB and Reference storage.
255         // Then calls StoreRefFilesToGrid to store reference files. 
256         //
257         
258         UpdateShuttleStatus(AliShuttleStatus::kStoreStarted);
259                                 
260         if (fTestMode & kErrorGrid)
261         {
262                 Log("SHUTTLE", "StoreOCDB - In TESTMODE - Simulating error while storing in the Grid");
263                 Log(fCurrentDetector, "StoreOCDB - In TESTMODE - Simulating error while storing in the Grid");
264                 return kFALSE;
265         }
266         
267         Log("SHUTTLE","StoreOCDB - Storing OCDB data ...");
268         Int_t resultCDB = StoreOCDB(fgkMainCDB);
269
270         Log("SHUTTLE","StoreOCDB - Storing reference data ...");
271         Int_t resultRef = StoreOCDB(fgkMainRefStorage);
272         
273         Log("SHUTTLE","StoreOCDB - Storing reference files ...");
274         Bool_t resultRefFiles = CopyFilesToGrid("reference");
275         
276         Bool_t resultMetadata = kTRUE;
277         if(fCurrentDetector == "GRP") 
278         {
279                 Log("StoreOCDB - SHUTTLE","Storing Run Metadata file ...");
280                 resultMetadata = CopyFilesToGrid("metadata");
281         }
282         
283         Int_t storeResult = 0;
284         
285         if (resultCDB < 0 || resultRef < 0 || resultRefFiles == kFALSE || resultMetadata == kFALSE)
286                 storeResult = -1;
287         else if (resultCDB > 0 || resultRef > 0)
288                 storeResult = 1;
289                 
290         if (storeResult < 0)
291         {
292                 Log("SHUTTLE", 
293                         Form("\t\t\t****** run %d - %s: STORAGE ERROR ******",
294                                 GetCurrentRun(), fCurrentDetector.Data()));
295                 UpdateShuttleStatus(AliShuttleStatus::kStoreError);
296         } 
297         else if (storeResult > 0)
298         {
299                 Log("SHUTTLE", 
300                         Form("\t\t\t****** run %d - %s: STORAGE DELAYED ******",
301                                 GetCurrentRun(), fCurrentDetector.Data()));
302                 UpdateShuttleStatus(AliShuttleStatus::kStoreDelayed);
303         }
304         else if (storeResult == 0) 
305         {
306                 Log("SHUTTLE", 
307                         Form("\t\t\t****** run %d - %s: DONE ******",
308                                 GetCurrentRun(), fCurrentDetector.Data()));
309                 UpdateShuttleStatus(AliShuttleStatus::kDone);
310                 UpdateShuttleLogbook(fCurrentDetector, "DONE");
311         }
312
313         return (storeResult == 0);
314 }
315
316 //______________________________________________________________________________________________
317 Int_t AliShuttle::StoreOCDB(const TString& gridURI)
318 {
319         //
320         // Called by StoreOCDB(), performs actual storage to the main OCDB and reference storages (Grid)
321         //
322         // Return code:
323         //   -2 initialization error
324         //   -1 storage error
325         //   0  success
326         //   1  storage delayed (e.g. previous unprocessed runs)
327         //
328
329         TObjArray* gridIds=0;
330
331         Bool_t result = kTRUE;
332         Bool_t delayed = kFALSE;
333
334         const char* type = 0;
335         TString localURI;
336         if(gridURI == fgkMainCDB) {
337                 type = "OCDB";
338                 localURI = fgkLocalCDB;
339         } else if(gridURI == fgkMainRefStorage) {
340                 type = "reference";
341                 localURI = fgkLocalRefStorage;
342         } else {
343                 AliError(Form("Invalid storage URI: %s", gridURI.Data()));
344                 return -2;
345         }
346
347         AliCDBManager* man = AliCDBManager::Instance();
348
349         AliCDBStorage *gridSto = man->GetStorage(gridURI);
350         if(!gridSto) {
351                 Log("SHUTTLE",
352                         Form("StoreOCDB - cannot activate main %s storage", type));
353                 return -2;
354         }
355
356         gridIds = gridSto->GetQueryCDBList();
357
358         // get objects previously stored in local CDB
359         AliCDBStorage *localSto = man->GetStorage(localURI);
360         if(!localSto) {
361                 Log("SHUTTLE",
362                         Form("StoreOCDB - cannot activate local %s storage", type));
363                 return -2;
364         }
365         AliCDBPath aPath(GetOfflineDetName(fCurrentDetector.Data()),"*","*");
366         // Local objects were stored with current run as Grid version!
367         TList* localEntries = localSto->GetAll(aPath.GetPath(), GetCurrentRun(), GetCurrentRun());
368         localEntries->SetOwner(1);
369
370         // loop on local stored objects
371         TIter localIter(localEntries);
372         AliCDBEntry *aLocEntry = 0;
373         while((aLocEntry = dynamic_cast<AliCDBEntry*> (localIter.Next()))){
374                 aLocEntry->SetOwner(1);
375                 AliCDBId aLocId = aLocEntry->GetId();
376                 aLocEntry->SetVersion(-1);
377                 aLocEntry->SetSubVersion(-1);
378
379                 // If local object is valid up to infinity we store it only if it is
380                 // the first unprocessed run!
381                 if (aLocId.GetLastRun() == AliCDBRunRange::Infinity() &&
382                         !fFirstUnprocessed[GetDetPos(fCurrentDetector)])
383                 {
384                         Log("SHUTTLE", Form("StoreOCDB - %s: object %s has validity infinite but "
385                                                 "there are previous unprocessed runs!",
386                                                 fCurrentDetector.Data(), aLocId.GetPath().Data()));
387                         Log(fCurrentDetector.Data(), Form("StoreOCDB - %s: object %s has validity infinite but "
388                                                 "there are previous unprocessed runs!",
389                                                 fCurrentDetector.Data(), aLocId.GetPath().Data()));
390                         delayed = kTRUE;
391                         continue;
392                 }
393
394                 // loop on Grid valid Id's
395                 Bool_t store = kTRUE;
396                 TIter gridIter(gridIds);
397                 AliCDBId* aGridId = 0;
398                 while((aGridId = dynamic_cast<AliCDBId*> (gridIter.Next()))){
399                         if(aGridId->GetPath() != aLocId.GetPath()) continue;
400                         // skip all objects valid up to infinity
401                         if(aGridId->GetLastRun() == AliCDBRunRange::Infinity()) continue;
402                         // if we get here, it means there's already some more recent object stored on Grid!
403                         store = kFALSE;
404                         break;
405                 }
406
407                 // If we get here, the file can be stored!
408                 Bool_t storeOk = gridSto->Put(aLocEntry);
409                 if(!store || storeOk){
410
411                         if (!store)
412                         {
413                                 Log(fCurrentDetector.Data(),
414                                         Form("StoreOCDB - A more recent object already exists in %s storage: <%s>",
415                                                 type, aGridId->ToString().Data()));
416                         } else {
417                                 Log("SHUTTLE",
418                                         Form("StoreOCDB - Object <%s> successfully put into %s storage",
419                                                 aLocId.ToString().Data(), type));
420                                 Log(fCurrentDetector.Data(),
421                                         Form("StoreOCDB - Object <%s> successfully put into %s storage",
422                                                 aLocId.ToString().Data(), type));
423                         }
424
425                         // removing local filename...
426                         TString filename;
427                         localSto->IdToFilename(aLocId, filename);
428                         Log("SHUTTLE", Form("StoreOCDB - Removing local file %s", filename.Data()));
429                         RemoveFile(filename.Data());
430                         continue;
431                 } else  {
432                         Log("SHUTTLE",
433                                 Form("StoreOCDB - Grid %s storage of object <%s> failed",
434                                         type, aLocId.ToString().Data()));
435                         Log(fCurrentDetector.Data(),
436                                 Form("StoreOCDB - Grid %s storage of object <%s> failed",
437                                         type, aLocId.ToString().Data()));
438                         result = kFALSE;
439                 }
440         }
441         localEntries->Clear();
442
443         Int_t returnCode = 0;
444         
445         if (result == kFALSE)
446                 returnCode = -1;
447         else if (delayed != kFALSE)
448                 returnCode =  1;
449
450         Log("SHUTTLE", Form("StoreOCDB - Returning with %d (result = %d, delayed = %d)", returnCode, result, delayed));
451         Log(fCurrentDetector.Data(), Form("StoreOCDB - Returning with %d (result = %d, delayed = %d)", returnCode, result, delayed));
452         
453         return returnCode;
454 }
455
456 //______________________________________________________________________________________________
457 Bool_t AliShuttle::CleanReferenceStorage(const char* detector)
458 {
459         // clears the directory used to store reference files of a given subdetector
460   
461         AliCDBManager* man = AliCDBManager::Instance();
462         AliCDBStorage* sto = man->GetStorage(fgkLocalRefStorage);
463         TString localBaseFolder = sto->GetBaseFolder();
464
465         TString targetDir = GetRefFilePrefix(localBaseFolder.Data(), detector);
466         
467         Log("SHUTTLE", Form("CleanReferenceStorage - Cleaning %s", targetDir.Data()));
468
469         TString begin;
470         begin.Form("%d_", GetCurrentRun());
471         
472         TSystemDirectory* baseDir = new TSystemDirectory("/", targetDir);
473         if (!baseDir)
474                 return kTRUE;
475                 
476         TList* dirList = baseDir->GetListOfFiles();
477         delete baseDir;
478         
479         if (!dirList) return kTRUE;
480                         
481         if (dirList->GetEntries() < 3) 
482         {
483                 delete dirList;
484                 return kTRUE;
485         }
486                                 
487         Int_t nDirs = 0, nDel = 0;
488         TIter dirIter(dirList);
489         TSystemFile* entry = 0;
490
491         Bool_t success = kTRUE;
492         
493         while ((entry = dynamic_cast<TSystemFile*> (dirIter.Next())))
494         {                                       
495                 if (entry->IsDirectory())
496                         continue;
497                 
498                 TString fileName(entry->GetName());
499                 if (!fileName.BeginsWith(begin))
500                         continue;
501                         
502                 nDirs++;
503                                                 
504                 // delete file
505                 Int_t result = gSystem->Unlink(fileName.Data());
506                 
507                 if (result)
508                 {
509                         Log("SHUTTLE", Form("CleanReferenceStorage - Could not delete file %s!", fileName.Data()));
510                         success = kFALSE;
511                 } else {
512                         nDel++;
513                 }
514         }
515
516         if(nDirs > 0)
517                 Log("SHUTTLE", Form("CleanReferenceStorage - %d (over %d) reference files in folder %s were deleted.", 
518                         nDel, nDirs, targetDir.Data()));
519
520                 
521         delete dirList;
522         return success;
523
524
525
526
527
528
529   Int_t result = gSystem->GetPathInfo(targetDir, 0, (Long64_t*) 0, 0, 0);
530   if (result == 0)
531   {
532     // delete directory
533     result = gSystem->Exec(Form("rm -rf %s", targetDir.Data()));
534     if (result != 0)
535     {  
536       Log("SHUTTLE", Form("CleanReferenceStorage - Could not clean directory %s", targetDir.Data()));
537       return kFALSE;
538     }
539   }
540
541   result = gSystem->mkdir(targetDir, kTRUE);
542   if (result != 0)
543   {
544     Log("SHUTTLE", Form("CleanReferenceStorage - Error creating base directory %s", targetDir.Data()));
545     return kFALSE;
546   }
547         
548   return kTRUE;
549 }
550
551 //______________________________________________________________________________________________
552 Bool_t AliShuttle::StoreReferenceFile(const char* detector, const char* localFile, const char* gridFileName)
553 {
554         //
555         // Stores reference file directly (without opening it). This function stores the file locally.
556         //
557         // The file is stored under the following location: 
558         // <base folder of local reference storage>/<DET>/<RUN#>_<gridFileName>
559         // where <gridFileName> is the second parameter given to the function
560         // 
561         
562         if (fTestMode & kErrorStorage)
563         {
564                 Log(fCurrentDetector, "StoreReferenceFile - In TESTMODE - Simulating error while storing locally");
565                 return kFALSE;
566         }
567         
568         AliCDBManager* man = AliCDBManager::Instance();
569         AliCDBStorage* sto = man->GetStorage(fgkLocalRefStorage);
570         
571         TString localBaseFolder = sto->GetBaseFolder();
572         
573         TString target = GetRefFilePrefix(localBaseFolder.Data(), detector);    
574         target.Append(Form("/%d_%s", GetCurrentRun(), gridFileName));
575         
576         return CopyFileLocally(localFile, target);
577 }
578
579 //______________________________________________________________________________________________
580 Bool_t AliShuttle::StoreRunMetadataFile(const char* localFile, const char* gridFileName)
581 {
582         //
583         // Stores Run metadata file to the Grid, in the run folder
584         //
585         // Only GRP can call this function.
586         
587         if (fTestMode & kErrorStorage)
588         {
589                 Log(fCurrentDetector, "StoreRunMetaDataFile - In TESTMODE - Simulating error while storing locally");
590                 return kFALSE;
591         }
592         
593         AliCDBManager* man = AliCDBManager::Instance();
594         AliCDBStorage* sto = man->GetStorage(fgkLocalRefStorage);
595         
596         TString localBaseFolder = sto->GetBaseFolder();
597         
598         // Build Run level folder
599         // folder = /alice/data/year/lhcPeriod/runNb/raw
600         
601                 
602         TString lhcPeriod = GetLHCPeriod();     
603         if (lhcPeriod.Length() == 0) 
604         {
605                 Log("SHUTTLE","StoreRunMetaDataFile - LHCPeriod not found in logbook!");
606                 return 0;
607         }
608         
609         // TODO partitions with one detector only write data into LHCperiod_DET
610         TString partition = GetRunParameter("detector");
611         
612         if (partition.Length() > 0 && partition != "ALICE")
613         {
614                 lhcPeriod.Append(Form("_%s", partition.Data()));
615                 Log(fCurrentDetector, Form("Run data tags merged file will be written in %s", 
616                                 lhcPeriod.Data()));
617         }
618                 
619         TString target = Form("%s/GRP/RunMetadata%s%d/%s/%09d/raw/%s", 
620                                 localBaseFolder.Data(), fConfig->GetAlienPath(), GetCurrentYear(), 
621                                 lhcPeriod.Data(), GetCurrentRun(), gridFileName);
622                                         
623         return CopyFileLocally(localFile, target);
624 }
625
626 //______________________________________________________________________________________________
627 Bool_t AliShuttle::CopyFileLocally(const char* localFile, const TString& target)
628 {
629         //
630         // Stores file locally. Called by StoreReferenceFile and StoreRunMetadataFile
631         // Files are temporarily stored in the local reference storage. When the preprocessor 
632         // finishes, the Shuttle calls CopyFilesToGrid to transfer the files to AliEn 
633         // (in reference or run level folders)
634         //
635         
636         TString targetDir(target(0, target.Last('/')));
637         
638         //try to open base dir folder, if it does not exist
639         void* dir = gSystem->OpenDirectory(targetDir.Data());
640         if (dir == NULL) {
641                 if (gSystem->mkdir(targetDir.Data(), kTRUE)) {
642                         Log("SHUTTLE", Form("CopyFileLocally - Can't open directory <%s>", targetDir.Data()));
643                         return kFALSE;
644                 }
645
646         } else {
647                 gSystem->FreeDirectory(dir);
648         }
649         
650         Int_t result = 0;
651         
652         result = gSystem->GetPathInfo(localFile, 0, (Long64_t*) 0, 0, 0);
653         if (result)
654         {
655                 Log("SHUTTLE", Form("CopyFileLocally - %s does not exist", localFile));
656                 return kFALSE;
657         }
658
659         result = gSystem->GetPathInfo(target, 0, (Long64_t*) 0, 0, 0);
660         if (!result)
661         {
662                 Log("SHUTTLE", Form("CopyFileLocally - target file %s already exist, removing...", target.Data()));
663                 if (gSystem->Unlink(target.Data()))
664                 {
665                         Log("SHUTTLE", Form("CopyFileLocally - Could not remove existing target file %s!", target.Data()));
666                         return kFALSE;
667                 }
668         }       
669         
670         result = gSystem->CopyFile(localFile, target);
671
672         if (result == 0)
673         {
674                 Log("SHUTTLE", Form("CopyFileLocally - File %s stored locally to %s", localFile, target.Data()));
675                 return kTRUE;
676         }
677         else
678         {
679                 Log("SHUTTLE", Form("CopyFileLocally - Could not store file %s to %s! Error code = %d", 
680                                 localFile, target.Data(), result));
681                 return kFALSE;
682         }       
683
684
685
686 }
687
688 //______________________________________________________________________________________________
689 Bool_t AliShuttle::CopyFilesToGrid(const char* type)
690 {
691         //
692         // Transfers local files to the Grid. Local files can be reference files 
693         // or run metadata file (from GRP only).
694         //
695         // According to the type (ref, metadata) the files are stored under the following location: 
696         // ref --> <base folder of reference storage>/<DET>/<RUN#>_<gridFileName>
697         // metadata --> <run data folder>/<MetadataFileName>
698         //
699                 
700         AliCDBManager* man = AliCDBManager::Instance();
701         AliCDBStorage* sto = man->GetStorage(fgkLocalRefStorage);
702         if (!sto)
703                 return kFALSE;
704         TString localBaseFolder = sto->GetBaseFolder();
705         
706         TString dir;
707         TString alienDir;
708         TString begin;
709         
710         if (strcmp(type, "reference") == 0) 
711         {
712                 dir = GetRefFilePrefix(localBaseFolder.Data(), fCurrentDetector.Data());
713                 AliCDBStorage* gridSto = man->GetStorage(fgkMainRefStorage);
714                 if (!gridSto)
715                         return kFALSE;
716                 TString gridBaseFolder = gridSto->GetBaseFolder();
717                 alienDir = GetRefFilePrefix(gridBaseFolder.Data(), fCurrentDetector.Data());
718                 begin = Form("%d_", GetCurrentRun());
719         } 
720         else if (strcmp(type, "metadata") == 0)
721         {
722                         
723                 TString lhcPeriod = GetLHCPeriod();
724         
725                 if (lhcPeriod.Length() == 0) 
726                 {
727                         Log("SHUTTLE","CopyFilesToGrid - LHCPeriod not found in logbook!");
728                         return 0;
729                 }
730                 
731                 // TODO partitions with one detector only write data into LHCperiod_DET
732                 TString partition = GetRunParameter("detector");
733         
734                 if (partition.Length() > 0 && partition != "ALICE")
735                 {
736                         lhcPeriod.Append(Form("_%s", partition.Data()));
737                 }
738                 
739                 dir = Form("%s/GRP/RunMetadata%s%d/%s/%09d/raw", 
740                                 localBaseFolder.Data(), fConfig->GetAlienPath(), GetCurrentYear(), 
741                                 lhcPeriod.Data(), GetCurrentRun());
742                 alienDir = dir(dir.Index(fConfig->GetAlienPath()), dir.Length());
743                 
744                 begin = "";
745         }
746         else 
747         {
748                 Log("SHUTTLE", "CopyFilesToGrid - Unexpected: type label must be reference or metadata!");
749                 return kFALSE;
750         }
751                 
752         TSystemDirectory* baseDir = new TSystemDirectory("/", dir);
753         if (!baseDir)
754                 return kTRUE;
755                 
756         TList* dirList = baseDir->GetListOfFiles();
757         delete baseDir;
758         
759         if (!dirList) return kTRUE;
760                 
761         if (dirList->GetEntries() < 3) 
762         {
763                 delete dirList;
764                 return kTRUE;
765         }
766                         
767         if (!gGrid)
768         { 
769                 Log("SHUTTLE", "CopyFilesToGrid - Connection to Grid failed: Cannot continue!");
770                 delete dirList;
771                 return kFALSE;
772         }
773         
774         Int_t nDirs = 0, nTransfer = 0;
775         TIter dirIter(dirList);
776         TSystemFile* entry = 0;
777
778         Bool_t success = kTRUE;
779         Bool_t first = kTRUE;
780         
781         while ((entry = dynamic_cast<TSystemFile*> (dirIter.Next())))
782         {                       
783                 if (entry->IsDirectory())
784                         continue;
785                         
786                 TString fileName(entry->GetName());
787                 if (!fileName.BeginsWith(begin))
788                         continue;
789                         
790                 nDirs++;
791                         
792                 if (first)
793                 {
794                         first = kFALSE;
795                         // check that folder exists, otherwise create it
796                         TGridResult* result = gGrid->Ls(alienDir.Data(), "a");
797                         
798                         if (!result)
799                         {
800                                 delete dirList;
801                                 return kFALSE;
802                         }
803                         
804                         if (!result->GetFileName(1)) // TODO: It looks like element 0 is always 0!!
805                         {
806                                 // TODO It does not work currently! Bug in TAliEn::Mkdir
807                                 // TODO Manually fixed in local root v5-16-00
808                                 if (!gGrid->Mkdir(alienDir.Data(),"-p",0))
809                                 {
810                                         Log("SHUTTLE", Form("CopyFilesToGrid - Cannot create directory %s",
811                                                         alienDir.Data()));
812                                         delete dirList;
813                                         return kFALSE;
814                                 } else {
815                                         Log("SHUTTLE",Form("CopyFilesToGrid - Folder %s created", alienDir.Data()));
816                                 }
817                                 
818                         } else {
819                                         Log("SHUTTLE",Form("CopyFilesToGrid - Folder %s found", alienDir.Data()));
820                         }
821                 }
822                         
823                 TString fullLocalPath;
824                 fullLocalPath.Form("%s/%s", dir.Data(), fileName.Data());
825                 
826                 TString fullGridPath;
827                 fullGridPath.Form("alien://%s/%s", alienDir.Data(), fileName.Data());
828
829                 Bool_t result = TFile::Cp(fullLocalPath, fullGridPath);
830                 
831                 if (result)
832                 {
833                         Log("SHUTTLE", Form("CopyFilesToGrid - Copying local file %s to %s succeeded!", 
834                                                 fullLocalPath.Data(), fullGridPath.Data()));
835                         RemoveFile(fullLocalPath);
836                         nTransfer++;
837                 }
838                 else
839                 {
840                         Log("SHUTTLE", Form("CopyFilesToGrid - Copying local file %s to %s FAILED!", 
841                                                 fullLocalPath.Data(), fullGridPath.Data()));
842                         success = kFALSE;
843                 }
844         }
845
846         Log("SHUTTLE", Form("CopyFilesToGrid - %d (over %d) files in folder %s copied to Grid.", 
847                                                 nTransfer, nDirs, dir.Data()));
848
849                 
850         delete dirList;
851         return success;
852 }
853
854 //______________________________________________________________________________________________
855 const char* AliShuttle::GetRefFilePrefix(const char* base, const char* detector)
856 {
857         //
858         // Get folder name of reference files 
859         //
860
861         TString offDetStr(GetOfflineDetName(detector));
862         TString dir;
863         if (offDetStr == "ITS" || offDetStr == "MUON" || offDetStr == "PHOS")
864         {
865                 dir.Form("%s/%s/%s", base, offDetStr.Data(), detector);
866         } else {
867                 dir.Form("%s/%s", base, offDetStr.Data());
868         }
869         
870         return dir.Data();
871         
872
873 }
874
875 //______________________________________________________________________________________________
876 void AliShuttle::CleanLocalStorage(const TString& uri)
877 {
878         //
879         // Called in case the preprocessor is declared failed. Remove remaining objects from the local storages.
880         //
881
882         const char* type = 0;
883         if(uri == fgkLocalCDB) {
884                 type = "OCDB";
885         } else if(uri == fgkLocalRefStorage) {
886                 type = "Reference";
887         } else {
888                 AliError(Form("Invalid storage URI: %s", uri.Data()));
889                 return;
890         }
891
892         AliCDBManager* man = AliCDBManager::Instance();
893
894         // open local storage
895         AliCDBStorage *localSto = man->GetStorage(uri);
896         if(!localSto) {
897                 Log("SHUTTLE",
898                         Form("CleanLocalStorage - cannot activate local %s storage", type));
899                 return;
900         }
901
902         TString filename(Form("%s/%s/*/Run*_v%d_s*.root",
903                 localSto->GetBaseFolder().Data(), GetOfflineDetName(fCurrentDetector.Data()), GetCurrentRun()));
904
905         AliDebug(2, Form("filename = %s", filename.Data()));
906
907         Log("SHUTTLE", Form("Removing remaining local files for run %d and detector %s ...",
908                 GetCurrentRun(), fCurrentDetector.Data()));
909
910         RemoveFile(filename.Data());
911
912 }
913
914 //______________________________________________________________________________________________
915 void AliShuttle::RemoveFile(const char* filename)
916 {
917         //
918         // removes local file
919         //
920
921         TString command(Form("rm -f %s", filename));
922
923         Int_t result = gSystem->Exec(command.Data());
924         if(result != 0)
925         {
926                 Log("SHUTTLE", Form("RemoveFile - %s: Cannot remove file %s!",
927                         fCurrentDetector.Data(), filename));
928         }
929 }
930
931 //______________________________________________________________________________________________
932 AliShuttleStatus* AliShuttle::ReadShuttleStatus()
933 {
934         //
935         // Reads the AliShuttleStatus from the CDB
936         //
937
938         if (fStatusEntry){
939                 delete fStatusEntry;
940                 fStatusEntry = 0;
941         }
942
943         fStatusEntry = AliCDBManager::Instance()->GetStorage(GetLocalCDB())
944                 ->Get(Form("/SHUTTLE/STATUS/%s", fCurrentDetector.Data()), GetCurrentRun());
945
946         if (!fStatusEntry) return 0;
947         fStatusEntry->SetOwner(1);
948
949         AliShuttleStatus* status = dynamic_cast<AliShuttleStatus*> (fStatusEntry->GetObject());
950         if (!status) {
951                 AliError("Invalid object stored to CDB!");
952                 return 0;
953         }
954
955         return status;
956 }
957
958 //______________________________________________________________________________________________
959 Bool_t AliShuttle::WriteShuttleStatus(AliShuttleStatus* status)
960 {
961         //
962         // writes the status for one subdetector
963         //
964
965         if (fStatusEntry){
966                 delete fStatusEntry;
967                 fStatusEntry = 0;
968         }
969
970         Int_t run = GetCurrentRun();
971
972         AliCDBId id(AliCDBPath("SHUTTLE", "STATUS", fCurrentDetector), run, run);
973
974         fStatusEntry = new AliCDBEntry(status, id, new AliCDBMetaData);
975         fStatusEntry->SetOwner(1);
976
977         UInt_t result = AliCDBManager::Instance()->GetStorage(fgkLocalCDB)->Put(fStatusEntry);
978
979         if (!result) {
980                 Log("SHUTTLE", Form("WriteShuttleStatus - Failed for %s, run %d",
981                                                 fCurrentDetector.Data(), run));
982                 return kFALSE;
983         }
984         
985         SendMLDetInfo();
986
987         return kTRUE;
988 }
989
990 //______________________________________________________________________________________________
991 void AliShuttle::UpdateShuttleStatus(AliShuttleStatus::Status newStatus, Bool_t increaseCount)
992 {
993         //
994         // changes the AliShuttleStatus for the given detector and run to the given status
995         //
996
997         if (!fStatusEntry){
998                 AliError("UNEXPECTED: fStatusEntry empty");
999                 return;
1000         }
1001
1002         AliShuttleStatus* status = dynamic_cast<AliShuttleStatus*> (fStatusEntry->GetObject());
1003
1004         if (!status){
1005                 Log("SHUTTLE", "UpdateShuttleStatus - UNEXPECTED: status could not be read from current CDB entry");
1006                 return;
1007         }
1008
1009         TString actionStr = Form("UpdateShuttleStatus - %s: Changing state from %s to %s",
1010                                 fCurrentDetector.Data(),
1011                                 status->GetStatusName(),
1012                                 status->GetStatusName(newStatus));
1013         Log("SHUTTLE", actionStr);
1014         SetLastAction(actionStr);
1015
1016         status->SetStatus(newStatus);
1017         if (increaseCount) status->IncreaseCount();
1018
1019         AliCDBManager::Instance()->GetStorage(fgkLocalCDB)->Put(fStatusEntry);
1020
1021         SendMLDetInfo();
1022 }
1023
1024 //______________________________________________________________________________________________
1025 void AliShuttle::SendMLDetInfo()
1026 {
1027         //
1028         // sends ML information about the current status of the current detector being processed
1029         //
1030         
1031         AliShuttleStatus* status = dynamic_cast<AliShuttleStatus*> (fStatusEntry->GetObject());
1032         
1033         if (!status){
1034                 Log("SHUTTLE", "SendMLDetInfo - UNEXPECTED: status could not be read from current CDB entry");
1035                 return;
1036         }
1037         
1038         TMonaLisaText  mlStatus(Form("%s_status", fCurrentDetector.Data()), status->GetStatusName());
1039         TMonaLisaValue mlRetryCount(Form("%s_count", fCurrentDetector.Data()), status->GetCount());
1040
1041         TList mlList;
1042         mlList.Add(&mlStatus);
1043         mlList.Add(&mlRetryCount);
1044
1045         TString mlID;
1046         mlID.Form("%d", GetCurrentRun());
1047         fMonaLisa->SendParameters(&mlList, mlID);
1048 }
1049
1050 //______________________________________________________________________________________________
1051 Bool_t AliShuttle::ContinueProcessing()
1052 {
1053         // this function reads the AliShuttleStatus information from CDB and
1054         // checks if the processing should be continued
1055         // if yes it returns kTRUE and updates the AliShuttleStatus with nextStatus
1056
1057         if (!fConfig->HostProcessDetector(fCurrentDetector))
1058                 return kFALSE;
1059
1060         AliPreprocessor* aPreprocessor =
1061                 dynamic_cast<AliPreprocessor*> (fPreprocessorMap.GetValue(fCurrentDetector));
1062         if (!aPreprocessor)
1063         {
1064                 Log("SHUTTLE", Form("ContinueProcessing - %s: no preprocessor registered", fCurrentDetector.Data()));
1065                 return kFALSE;
1066         }
1067
1068         AliShuttleLogbookEntry::Status entryStatus =
1069                 fLogbookEntry->GetDetectorStatus(fCurrentDetector);
1070
1071         if (entryStatus != AliShuttleLogbookEntry::kUnprocessed) {
1072                 Log("SHUTTLE", Form("ContinueProcessing - %s is %s",
1073                                 fCurrentDetector.Data(),
1074                                 fLogbookEntry->GetDetectorStatusName(entryStatus)));
1075                 return kFALSE;
1076         }
1077
1078         // if we get here, according to Shuttle logbook subdetector is in UNPROCESSED state
1079
1080         // check if current run is first unprocessed run for current detector
1081         if (fConfig->StrictRunOrder(fCurrentDetector) &&
1082                 !fFirstUnprocessed[GetDetPos(fCurrentDetector)])
1083         {
1084                 if (fTestMode == kNone)
1085                 {
1086                         Log("SHUTTLE", Form("ContinueProcessing - %s requires strict run ordering"
1087                                         " but this is not the first unprocessed run!"));
1088                         return kFALSE;
1089                 }
1090                 else
1091                 {
1092                         Log("SHUTTLE", Form("ContinueProcessing - In TESTMODE - "
1093                                         "Although %s requires strict run ordering "
1094                                         "and this is not the first unprocessed run, "
1095                                         "the SHUTTLE continues"));
1096                 }
1097         }
1098
1099         // Is the subdetector processed first time for this run?
1100         fFirstProcessing = kFALSE;
1101
1102         AliShuttleStatus* status = ReadShuttleStatus();
1103         if (!status) {
1104                 // first time
1105                 Log("SHUTTLE", Form("ContinueProcessing - %s: Processing first time",
1106                                 fCurrentDetector.Data()));
1107                 status = new AliShuttleStatus(AliShuttleStatus::kStarted);
1108                 fFirstProcessing = kTRUE;
1109                 return WriteShuttleStatus(status);
1110         }
1111
1112         // The following case shouldn't happen if Shuttle Logbook was correctly updated.
1113         // If it happens it may mean Logbook updating failed... let's do it now!
1114         if (status->GetStatus() == AliShuttleStatus::kDone ||
1115             status->GetStatus() == AliShuttleStatus::kFailed ||
1116             status->GetStatus() == AliShuttleStatus::kSkipped) {
1117                 Log("SHUTTLE", Form("ContinueProcessing - %s is already %s. Updating Shuttle Logbook",
1118                                         fCurrentDetector.Data(),
1119                                         status->GetStatusName(status->GetStatus())));
1120                 
1121                 if (status->GetStatus() == AliShuttleStatus::kSkipped)
1122                 {
1123                         UpdateShuttleLogbook(fCurrentDetector.Data(), "DONE");
1124                 }
1125                 else
1126                         UpdateShuttleLogbook(fCurrentDetector.Data(), status->GetStatusName(status->GetStatus()));
1127                         
1128                 return kFALSE;
1129         }
1130
1131         if (status->GetStatus() == AliShuttleStatus::kStoreStarted || status->GetStatus() == AliShuttleStatus::kStoreDelayed ||status->GetStatus() == AliShuttleStatus::kStoreError) {
1132                 Log("SHUTTLE",
1133                         Form("ContinueProcessing - %s: Grid storage of one or more "
1134                                 "objects failed. Trying again now",
1135                                 fCurrentDetector.Data()));
1136                 StoreOCDB();
1137                 return kFALSE;
1138         }
1139
1140         // if we get here, there is a restart
1141         Bool_t cont = kFALSE;
1142
1143         // abort conditions
1144         if (status->GetCount() >= fConfig->GetMaxRetries()) {
1145                 Log("SHUTTLE", Form("ContinueProcessing - %s failed %d times in status %s - "
1146                                 "Updating Shuttle Logbook", fCurrentDetector.Data(),
1147                                 status->GetCount(), status->GetStatusName()));
1148                 UpdateShuttleLogbook(fCurrentDetector.Data(), "FAILED");
1149                 UpdateShuttleStatus(AliShuttleStatus::kFailed);
1150
1151                 // there may still be objects in local OCDB and reference storage
1152                 // and FXS databases may be not updated: do it now!
1153                 
1154                 // TODO Currently disabled, we want to keep files in case of failure!
1155                 // CleanLocalStorage(fgkLocalCDB);
1156                 // CleanLocalStorage(fgkLocalRefStorage);
1157                 // UpdateTableFailCase();
1158                 
1159                 // Send mail to detector expert!
1160                 Log("SHUTTLE", Form("ContinueProcessing - Sending mail to %s expert...", 
1161                                     fCurrentDetector.Data()));
1162                 if (!SendMail(kPPEMail))
1163                         Log("SHUTTLE", Form("ContinueProcessing - Could not send mail to %s expert",
1164                                             fCurrentDetector.Data()));
1165
1166         } else {
1167                 Log("SHUTTLE", Form("ContinueProcessing - %s: restarting. "
1168                                 "Aborted before with %s. Retry number %d.", fCurrentDetector.Data(),
1169                                 status->GetStatusName(), status->GetCount()));
1170                 Bool_t increaseCount = kTRUE;
1171                 if (status->GetStatus() == AliShuttleStatus::kDCSError || 
1172                     status->GetStatus() == AliShuttleStatus::kDCSStarted ||
1173                     status->GetStatus() == AliShuttleStatus::kFXSError)
1174                                 increaseCount = kFALSE;
1175                                 
1176                 UpdateShuttleStatus(AliShuttleStatus::kStarted, increaseCount);
1177                 cont = kTRUE;
1178         }
1179
1180         return cont;
1181 }
1182
1183 //______________________________________________________________________________________________
1184 void AliShuttle::SendMLRunInfo(const char* status)
1185 {
1186         // 
1187         // Send information about this run to ML
1188         
1189         TMonaLisaText  mlStatus("SHUTTLE_status", status);
1190         TString runType(fLogbookEntry->GetRunType());
1191         if (strlen(fLogbookEntry->GetRunParameter("log")) > 0){
1192
1193                 runType += "(";
1194                 runType += fLogbookEntry->GetRunParameter("log");
1195                 runType += ")";
1196         }
1197         TMonaLisaText  mlRunType("SHUTTLE_runtype", runType);
1198
1199         TList mlList;
1200         mlList.Add(&mlStatus);
1201         mlList.Add(&mlRunType);
1202
1203         TString mlID;
1204         mlID.Form("%d", GetCurrentRun());
1205         fMonaLisa->SendParameters(&mlList, mlID);       
1206 }
1207
1208 //______________________________________________________________________________________________
1209 Int_t AliShuttle::GetMem(Int_t pid)
1210 {
1211         // invokes ps to get the memory consumption of the process <pid>
1212         // returns -1 in case of error
1213         
1214         TString checkStr;
1215         checkStr.Form("ps -o vsize --pid %d | tail -n 1", pid);
1216         FILE* pipe = gSystem->OpenPipe(checkStr, "r");
1217         if (!pipe)
1218         {
1219                 Log("SHUTTLE", Form("Process - Error: "
1220                         "Could not open pipe to %s", checkStr.Data()));
1221                 return -1;
1222         }
1223                 
1224         char buffer[100];
1225         if (!fgets(buffer, 100, pipe))
1226         {
1227                 Log("SHUTTLE", "Process - Error: ps did not return anything");
1228                 gSystem->ClosePipe(pipe);
1229                 return -1;
1230         }
1231         gSystem->ClosePipe(pipe);
1232         
1233         //Log("SHUTTLE", Form("ps returned %s", buffer));
1234         
1235         Int_t mem = 0;
1236         if ((sscanf(buffer, "%d\n", &mem) != 1) || !mem)
1237         {
1238                 Log("SHUTTLE", "Process - Error: Could not parse output of ps");
1239                 return -1;
1240         }
1241         
1242         return mem;
1243 }
1244
1245 //______________________________________________________________________________________________
1246 Bool_t AliShuttle::Process(AliShuttleLogbookEntry* entry)
1247 {
1248         //
1249         // Makes data retrieval for all detectors in the configuration.
1250         // entry: Shuttle logbook entry, contains run paramenters and status of detectors
1251         // (Unprocessed, Inactive, Failed or Done).
1252         // Returns kFALSE in case of error occured and kTRUE otherwise
1253         //
1254
1255         if (!entry) return kFALSE;
1256
1257         fLogbookEntry = entry;
1258
1259         Log("SHUTTLE", Form("\t\t\t^*^*^*^*^*^*^*^*^*^*^*^* run %d: START ^*^*^*^*^*^*^*^*^*^*^*^*",
1260                                         GetCurrentRun()));
1261
1262         CountOpenRuns();
1263         
1264         // Send the information to ML
1265         SendMLRunInfo("Processing");
1266
1267         if (fLogbookEntry->IsDone())
1268         {
1269                 Log("SHUTTLE","Process - Shuttle is already DONE. Updating logbook");
1270                 UpdateShuttleLogbook("shuttle_done");
1271                 fLogbookEntry = 0;
1272                 return kTRUE;
1273         }
1274
1275         // read test mode if flag is set
1276         if (fReadTestMode)
1277         {
1278                 fTestMode = kNone;
1279                 TString logEntry(entry->GetRunParameter("log"));
1280                 //printf("log entry = %s\n", logEntry.Data());
1281                 TString searchStr("Testmode: ");
1282                 Int_t pos = logEntry.Index(searchStr.Data());
1283                 //printf("%d\n", pos);
1284                 if (pos >= 0)
1285                 {
1286                         TSubString subStr = logEntry(pos + searchStr.Length(), logEntry.Length());
1287                         //printf("%s\n", subStr.String().Data());
1288                         TString newStr(subStr.Data());
1289                         TObjArray* token = newStr.Tokenize(' ');
1290                         if (token)
1291                         {
1292                                 //token->Print();
1293                                 TObjString* tmpStr = dynamic_cast<TObjString*> (token->First());
1294                                 if (tmpStr)
1295                                 {
1296                                         Int_t testMode = tmpStr->String().Atoi();
1297                                         if (testMode > 0)
1298                                         {
1299                                                 Log("SHUTTLE", Form("Process - Enabling test mode %d", testMode));
1300                                                 SetTestMode((TestMode) testMode);
1301                                         }
1302                                 }
1303                                 delete token;          
1304                         }
1305                 }
1306         }
1307                 
1308         fLogbookEntry->Print("all");
1309
1310         // Initialization
1311         Bool_t hasError = kFALSE;
1312
1313         // Set the CDB and Reference folders according to the year and LHC period
1314         TString lhcPeriod(GetLHCPeriod());
1315         if (lhcPeriod.Length() == 0) 
1316         {
1317                 Log("SHUTTLE","Process - LHCPeriod not found in logbook!");
1318                 return 0; 
1319         }       
1320         
1321         // build cdb paths (repeat each time, LHCperiod might have changed)
1322         fgkMainCDB.Form("alien://folder=%s%d/%s/OCDB?user=alidaq?cacheFold=/tmp/OCDBCache", 
1323                                         fConfig->GetAlienPath(), GetCurrentYear(), lhcPeriod.Data());
1324         
1325         fgkMainRefStorage.Form("alien://folder=%s%d/%s/Reference?user=alidaq?cacheFold=/tmp/OCDBCache", 
1326                                         fConfig->GetAlienPath(), GetCurrentYear(), lhcPeriod.Data());
1327         
1328         // Loop on detectors in the configuration
1329         TIter iter(fConfig->GetDetectors());
1330         TObjString* aDetector = 0;
1331
1332         Bool_t first = kTRUE;
1333
1334         while ((aDetector = (TObjString*) iter.Next()))
1335         {
1336                 fCurrentDetector = aDetector->String();
1337
1338                 if (ContinueProcessing() == kFALSE) 
1339                         continue;
1340                 
1341                 if (first)
1342                 {
1343                   // only read QueryCDB when needed and only once
1344                   AliCDBStorage *mainCDBSto = AliCDBManager::Instance()->GetStorage(fgkMainCDB);
1345                   if(mainCDBSto) mainCDBSto->QueryCDB(GetCurrentRun());
1346                   AliCDBStorage *mainRefSto = AliCDBManager::Instance()->GetStorage(fgkMainRefStorage);
1347                   if(mainRefSto) mainRefSto->QueryCDB(GetCurrentRun());
1348                   first = kFALSE;
1349                 }
1350
1351                 Log("SHUTTLE", Form("\t\t\t****** run %d - %s: START  ******",
1352                                                 GetCurrentRun(), aDetector->GetName()));
1353
1354                 for(Int_t iSys=0;iSys<3;iSys++) fFXSCalled[iSys]=kFALSE;
1355                 
1356                 Int_t initialMem = GetMem(getpid());
1357                 Log("SHUTTLE", Form("Memory consumption before forking is %d", initialMem));
1358
1359                 Log(fCurrentDetector.Data(), "Process - Starting processing");
1360
1361                 Int_t pid = fork();
1362
1363                 if (pid < 0)
1364                 {
1365                         Log("SHUTTLE", "Process - ERROR: Forking failed");
1366                 }
1367                 else if (pid > 0)
1368                 {
1369                         // parent
1370                         Log("SHUTTLE", Form("Process - In parent process of %d - %s: Starting monitoring",
1371                                                         GetCurrentRun(), aDetector->GetName()));
1372
1373                         Long_t begin = time(0);
1374
1375                         int status; // to be used with waitpid, on purpose an int (not Int_t)!
1376                         while (waitpid(pid, &status, WNOHANG) == 0)
1377                         {
1378                                 Long_t expiredTime = time(0) - begin;
1379
1380                                 if (expiredTime > fConfig->GetPPTimeOut())
1381                                 {
1382                                         TString tmp;
1383                                         tmp.Form("Process - Process of %s time out. "
1384                                                         "Run time: %d seconds. Killing...",
1385                                                         fCurrentDetector.Data(), expiredTime);
1386                                         Log("SHUTTLE", tmp);
1387                                         Log(fCurrentDetector, tmp);
1388
1389                                         kill(pid, 9);
1390
1391                                         UpdateShuttleStatus(AliShuttleStatus::kPPTimeOut);
1392                                         hasError = kTRUE;
1393
1394                                         gSystem->Sleep(1000);
1395                                 }
1396                                 else
1397                                 {
1398                                         gSystem->Sleep(1000);
1399                                         
1400                                         Int_t mem = GetMem(pid);
1401
1402                                         if (mem < 0)
1403                                                 continue;
1404                                                 
1405                                         mem -= initialMem;
1406                                         if (mem < 0)
1407                                                 mem = 0;
1408                                         
1409                                         if (expiredTime % 60 == 0)
1410                                         {
1411                                                 Log("SHUTTLE", Form("Process - %s: Checking process. "
1412                                                         "Run time: %d seconds - Memory consumption: %d KB",
1413                                                         fCurrentDetector.Data(), expiredTime, mem));
1414                                                 SendAlive();
1415                                         }
1416                                         
1417                                         if (mem > fConfig->GetPPMaxMem())
1418                                         {
1419                                                 TString tmp;
1420                                                 tmp.Form("Process - Process exceeds maximum allowed memory "
1421                                                         "(%d KB > %d KB). Killing...",
1422                                                         mem, fConfig->GetPPMaxMem());
1423                                                 Log("SHUTTLE", tmp);
1424                                                 Log(fCurrentDetector, tmp);
1425         
1426                                                 kill(pid, 9);
1427         
1428                                                 UpdateShuttleStatus(AliShuttleStatus::kPPOutOfMemory);
1429                                                 hasError = kTRUE;
1430         
1431                                                 gSystem->Sleep(1000);
1432                                         }
1433                                 }
1434                         }
1435
1436                         Log("SHUTTLE", Form("Process - In parent process of %d - %s: Client has terminated.",
1437                                                                 GetCurrentRun(), aDetector->GetName()));
1438
1439                         if (WIFEXITED(status))
1440                         {
1441                                 Int_t returnCode = WEXITSTATUS(status);
1442
1443                                 Log("SHUTTLE", Form("Process - %s: the return code is %d", fCurrentDetector.Data(),
1444                                                                                 returnCode));
1445
1446                                 if (returnCode == 0) hasError = kTRUE;
1447                         }
1448                 }
1449                 else if (pid == 0)
1450                 {
1451                         // child
1452                         Log("SHUTTLE", Form("Process - In child process of %d - %s", GetCurrentRun(),
1453                                 aDetector->GetName()));
1454
1455                         Log("SHUTTLE", Form("Process - Redirecting output to %s log",fCurrentDetector.Data()));
1456
1457                         if ((freopen(GetLogFileName(fCurrentDetector), "a", stdout)) == 0)
1458                         {
1459                                 Log("SHUTTLE", "Process - Could not freopen stdout");
1460                         }
1461                         else
1462                         {
1463                                 fOutputRedirected = kTRUE;
1464                                 if ((dup2(fileno(stdout), fileno(stderr))) < 0)
1465                                         Log("SHUTTLE", "Process - Could not redirect stderr");
1466                                 
1467                         }
1468                         
1469                         TString wd = gSystem->WorkingDirectory();
1470                         TString tmpDir = Form("%s/%s_%d_process", GetShuttleTempDir(), 
1471                                 fCurrentDetector.Data(), GetCurrentRun());
1472                         
1473                         Int_t result = gSystem->GetPathInfo(tmpDir.Data(), 0, (Long64_t*) 0, 0, 0);
1474                         if (!result) // temp dir already exists!
1475                         {
1476                                 Log(fCurrentDetector.Data(), 
1477                                         Form("Process - %s dir already exists! Removing...", tmpDir.Data()));
1478                                 gSystem->Exec(Form("rm -rf %s",tmpDir.Data()));         
1479                         } 
1480                         
1481                         if (gSystem->mkdir(tmpDir.Data(), 1))
1482                         {
1483                                 Log(fCurrentDetector.Data(), "Process - could not make temp directory!!");
1484                                 gSystem->Exit(1);
1485                         }
1486                         
1487                         if (!gSystem->ChangeDirectory(tmpDir.Data())) 
1488                         {
1489                                 Log(fCurrentDetector.Data(), "Process - could not change directory!!");
1490                                 gSystem->Exit(1);                       
1491                         }
1492                         
1493                         Int_t success = ProcessCurrentDetector();
1494
1495                         gSystem->ChangeDirectory(wd.Data());
1496                                                 
1497                         if (success == 1) // Preprocessor finished successfully!
1498                         { 
1499                                 // remove temporary folder or DCS map
1500                                 if (!fConfig->KeepTempFolder())
1501                                 {
1502                                         gSystem->Exec(Form("rm -rf %s",tmpDir.Data()));
1503                                 } else if (!fConfig->KeepDCSMap())
1504                                 {
1505                                         gSystem->Exec(Form("rm -f %s/DCSMap.root",tmpDir.Data()));
1506                                 }
1507                                 
1508                                 // Update time_processed field in FXS DB
1509                                 if (UpdateTable() == kFALSE)
1510                                         Log("SHUTTLE", Form("Process - %s: Could not update FXS databases!", 
1511                                                         fCurrentDetector.Data()));
1512
1513                                 // Transfer the data from local storage to main storage (Grid)
1514                                 if (StoreOCDB() == kFALSE)
1515                                         success = kFALSE;
1516                         } 
1517                         else if (success == 0)
1518                         {
1519                                 Log("SHUTTLE", 
1520                                         Form("\t\t\t****** run %d - %s: PP ERROR ******",
1521                                                 GetCurrentRun(), aDetector->GetName()));
1522                         }
1523
1524                         for (UInt_t iSys=0; iSys<3; iSys++)
1525                         {
1526                                 if (fFXSCalled[iSys]) fFXSlist[iSys].Clear();
1527                         }
1528
1529                         Log("SHUTTLE", Form("Process - Client process of %d - %s is exiting now with %d.",
1530                                                         GetCurrentRun(), aDetector->GetName(), success));
1531
1532                         // the client exits here
1533                         gSystem->Exit(success);
1534
1535                         AliError("We should never get here!!!");
1536                 }
1537         }
1538
1539         Log("SHUTTLE", Form("\t\t\t^*^*^*^*^*^*^*^*^*^*^*^* run %d: FINISH ^*^*^*^*^*^*^*^*^*^*^*^*",
1540                                                         GetCurrentRun()));
1541
1542         //check if shuttle is done for this run, if so update logbook
1543         TObjArray checkEntryArray;
1544         checkEntryArray.SetOwner(1);
1545         TString whereClause = Form("where run=%d", GetCurrentRun());
1546         if (!QueryShuttleLogbook(whereClause.Data(), checkEntryArray) || 
1547                         checkEntryArray.GetEntries() == 0) {
1548                 Log("SHUTTLE", Form("Process - Warning: Cannot check status of run %d on Shuttle logbook!",
1549                                                 GetCurrentRun()));
1550                 return hasError == kFALSE;
1551         }
1552
1553         AliShuttleLogbookEntry* checkEntry = dynamic_cast<AliShuttleLogbookEntry*>
1554                                                 (checkEntryArray.At(0));
1555
1556         if (checkEntry)
1557         {
1558                 if (checkEntry->IsDone())
1559                 {
1560                         Log("SHUTTLE","Process - Shuttle is DONE. Updating logbook");
1561                         UpdateShuttleLogbook("shuttle_done");
1562                 }
1563                 else
1564                 {
1565                         for (UInt_t iDet=0; iDet<NDetectors(); iDet++)
1566                         {
1567                                 if (checkEntry->GetDetectorStatus(iDet) == AliShuttleLogbookEntry::kUnprocessed)
1568                                 {
1569                                         AliDebug(2, Form("Run %d: setting %s as \"not first time unprocessed\"",
1570                                                         checkEntry->GetRun(), GetDetName(iDet)));
1571                                         fFirstUnprocessed[iDet] = kFALSE;
1572                                 }
1573                         }
1574                         SendMLRunInfo("Pending");
1575                 }
1576         }
1577
1578         fLogbookEntry = 0;
1579
1580         return hasError == kFALSE;
1581 }
1582
1583 //______________________________________________________________________________________________
1584 Int_t AliShuttle::ProcessCurrentDetector()
1585 {
1586         //
1587         // Makes data retrieval just for a specific detector (fCurrentDetector).
1588         // Threre should be a configuration for this detector.
1589
1590         Log("SHUTTLE", Form("ProcessCurrentDetector - Retrieving values for %s, run %d", 
1591                                                 fCurrentDetector.Data(), GetCurrentRun()));
1592
1593         TString wd = gSystem->WorkingDirectory();
1594         
1595         if (!CleanReferenceStorage(fCurrentDetector.Data()))
1596                 return 0;
1597         
1598         gSystem->ChangeDirectory(wd.Data());
1599         
1600         // call preprocessor
1601         AliPreprocessor* aPreprocessor =
1602                 dynamic_cast<AliPreprocessor*> (fPreprocessorMap.GetValue(fCurrentDetector));
1603
1604         // check if the preprocessor wants to process this run type
1605         if (aPreprocessor->ProcessRunType() == kFALSE)
1606         {
1607                 UpdateShuttleStatus(AliShuttleStatus::kSkipped);
1608                 UpdateShuttleLogbook(fCurrentDetector, "DONE");
1609                 Log(fCurrentDetector, Form("ProcessCurrentDetector - %s preprocessor is not interested in this run type", fCurrentDetector.Data()));
1610         
1611                 return 2;
1612         }
1613         
1614         TMap* dcsMap = new TMap();
1615         
1616         aPreprocessor->Initialize(GetCurrentRun(), GetCurrentStartTime(), GetCurrentEndTime());
1617
1618         Bool_t processDCS = aPreprocessor->ProcessDCS();
1619
1620         if (!processDCS)
1621         {
1622                 Log(fCurrentDetector, "ProcessCurrentDetector -"
1623                         " The preprocessor requested to skip the retrieval of DCS values");
1624         }
1625         else if (fTestMode & kSkipDCS)
1626         {
1627                 Log(fCurrentDetector, "ProcessCurrentDetector - In TESTMODE: Skipping DCS processing");
1628         } 
1629         else if (fTestMode & kErrorDCS)
1630         {
1631                 Log(fCurrentDetector, "ProcessCurrentDetector - In TESTMODE: Simulating DCS error");
1632                 UpdateShuttleStatus(AliShuttleStatus::kDCSStarted);
1633                 UpdateShuttleStatus(AliShuttleStatus::kDCSError);
1634                 delete dcsMap;
1635                 return 0;
1636         } else {
1637
1638                 UpdateShuttleStatus(AliShuttleStatus::kDCSStarted);
1639
1640                 // Query DCS archive
1641                 Int_t nServers = fConfig->GetNServers(fCurrentDetector);
1642                 
1643                 for (int iServ=0; iServ<nServers; iServ++)
1644                 {
1645                 
1646                         TString host(fConfig->GetDCSHost(fCurrentDetector, iServ));
1647                         Int_t port = fConfig->GetDCSPort(fCurrentDetector, iServ);
1648                         Int_t multiSplit = fConfig->GetMultiSplit(fCurrentDetector, iServ);
1649
1650                         Log(fCurrentDetector, Form("ProcessCurrentDetector -"
1651                                         " Querying DCS Amanda server %s:%d (%d of %d)", 
1652                                         host.Data(), port, iServ+1, nServers));
1653                         
1654                         TMap* aliasMap = 0;
1655                         TMap* dpMap = 0;
1656
1657                         if (fConfig->GetDCSAliases(fCurrentDetector, iServ)->GetEntries() > 0)
1658                         {
1659                                 aliasMap = GetValueSet(host, port, 
1660                                                 fConfig->GetDCSAliases(fCurrentDetector, iServ), 
1661                                                 kAlias, multiSplit);
1662                                 if (!aliasMap)
1663                                 {
1664                                         Log(fCurrentDetector, 
1665                                                 Form("ProcessCurrentDetector -"
1666                                                         " Error retrieving DCS aliases from server %s."
1667                                                         " Sending mail to DCS experts!", host.Data()));
1668                                         UpdateShuttleStatus(AliShuttleStatus::kDCSError);
1669                                         
1670                                         if (!SendMail(kDCSEMail))
1671                                                 Log("SHUTTLE", Form("ProcessCurrentDetector - "
1672                                                                     "Could not send mail to DCS experts!"));
1673
1674                                         delete dcsMap;
1675                                         return 0;
1676                                 }
1677                         }
1678                         
1679                         if (fConfig->GetDCSDataPoints(fCurrentDetector, iServ)->GetEntries() > 0)
1680                         {
1681                                 dpMap = GetValueSet(host, port, 
1682                                                 fConfig->GetDCSDataPoints(fCurrentDetector, iServ), 
1683                                                 kDP, multiSplit);
1684                                 if (!dpMap)
1685                                 {
1686                                         Log(fCurrentDetector, 
1687                                                 Form("ProcessCurrentDetector -"
1688                                                         " Error retrieving DCS data points from server %s."
1689                                                         " Sending mail to DCS experts!", host.Data()));
1690                                         UpdateShuttleStatus(AliShuttleStatus::kDCSError);
1691                                         
1692                                         if (!SendMail(kDCSEMail))
1693                                                 Log("SHUTTLE", Form("ProcessCurrentDetector - "
1694                                                                     "Could not send mail to DCS experts!"));
1695                                         
1696                                         if (aliasMap) delete aliasMap;
1697                                         delete dcsMap;
1698                                         return 0;
1699                                 }                               
1700                         }
1701                         
1702                         // merge aliasMap and dpMap into dcsMap
1703                         if(aliasMap) {
1704                                 TIter iter(aliasMap);
1705                                 TObjString* key = 0;
1706                                 while ((key = (TObjString*) iter.Next()))
1707                                         dcsMap->Add(key, aliasMap->GetValue(key->String()));
1708                                 
1709                                 aliasMap->SetOwner(kFALSE);
1710                                 delete aliasMap;
1711                         }       
1712                         
1713                         if(dpMap) {
1714                                 TIter iter(dpMap);
1715                                 TObjString* key = 0;
1716                                 while ((key = (TObjString*) iter.Next()))
1717                                         dcsMap->Add(key, dpMap->GetValue(key->String()));
1718                                 
1719                                 dpMap->SetOwner(kFALSE);
1720                                 delete dpMap;
1721                         }
1722                 }
1723         }
1724         
1725         // save map into file, to help debugging in case of preprocessor error
1726         TFile* f = TFile::Open("DCSMap.root","recreate");
1727         f->cd();
1728         dcsMap->Write("DCSMap", TObject::kSingleKey);
1729         f->Close();
1730         delete f;
1731         
1732         // DCS Archive DB processing successful. Call Preprocessor!
1733         UpdateShuttleStatus(AliShuttleStatus::kPPStarted);
1734
1735         fFXSError = -1; // this variable is kTRUE after ::Process if an FXS error occured
1736         
1737         UInt_t returnValue = aPreprocessor->Process(dcsMap);
1738         
1739         if (fFXSError!=-1) {
1740                 UpdateShuttleStatus(AliShuttleStatus::kFXSError);
1741                 SendMail(kFXSEMail, fFXSError);
1742                 dcsMap->DeleteAll();
1743                 delete dcsMap;
1744                 return 0;
1745         }
1746
1747         if (returnValue > 0) // Preprocessor error!
1748         {
1749                 Log(fCurrentDetector, Form("ProcessCurrentDetector - "
1750                                 "Preprocessor failed. Process returned %d.", returnValue));
1751                 UpdateShuttleStatus(AliShuttleStatus::kPPError);
1752                 dcsMap->DeleteAll();
1753                 delete dcsMap;
1754                 return 0;
1755         }
1756         
1757         // preprocessor ok!
1758         UpdateShuttleStatus(AliShuttleStatus::kPPDone);
1759         Log(fCurrentDetector, Form("ProcessCurrentDetector - %s preprocessor returned success",
1760                                 fCurrentDetector.Data()));
1761
1762         dcsMap->DeleteAll();
1763         delete dcsMap;
1764
1765         return 1;
1766 }
1767
1768 //______________________________________________________________________________________________
1769 void AliShuttle::CountOpenRuns()
1770 {
1771         // Query DAQ's Shuttle logbook and sends the number of open runs to ML
1772         
1773         // check connection, in case connect
1774         if (!Connect(3)) 
1775                 return;
1776
1777         TString sqlQuery;
1778         sqlQuery = Form("select count(*) from %s where shuttle_done=0", fConfig->GetShuttlelbTable());
1779         
1780         TSQLResult* aResult = fServer[3]->Query(sqlQuery);
1781         if (!aResult) {
1782                 AliError(Form("Can't execute query <%s>!", sqlQuery.Data()));
1783                 return;
1784         }
1785
1786         AliDebug(2,Form("Query = %s", sqlQuery.Data()));
1787         
1788         if (aResult->GetRowCount() == 0) {
1789                 AliError(Form("No result for query %s received", sqlQuery.Data()));
1790                 return;
1791         }
1792
1793         if (aResult->GetFieldCount() != 1) {
1794                 AliError(Form("Invalid field count for query %s received", sqlQuery.Data()));
1795                 return;
1796         }
1797
1798         TSQLRow* aRow = aResult->Next();
1799         if (!aRow) {
1800                 AliError(Form("Could not receive result of query %s", sqlQuery.Data()));
1801                 return;
1802         }
1803         
1804         TString result(aRow->GetField(0), aRow->GetFieldLength(0));
1805         Int_t count = result.Atoi();
1806         
1807         Log("SHUTTLE", Form("%d unprocessed runs", count));
1808         
1809         delete aRow;
1810         delete aResult;
1811
1812         TMonaLisaValue mlStatus("SHUTTLE_openruns", count);
1813
1814         TList mlList;
1815         mlList.Add(&mlStatus);
1816
1817         fMonaLisa->SendParameters(&mlList, "__PROCESSINGINFO__");
1818 }
1819
1820 //______________________________________________________________________________________________
1821 Bool_t AliShuttle::QueryShuttleLogbook(const char* whereClause,
1822                 TObjArray& entries)
1823 {
1824         // Query DAQ's Shuttle logbook and fills detector status object.
1825         // Call QueryRunParameters to query DAQ logbook for run parameters.
1826         //
1827
1828         entries.SetOwner(1);
1829
1830         // check connection, in case connect
1831         if (!Connect(3)) return kFALSE;
1832
1833         TString sqlQuery;
1834         sqlQuery = Form("select * from %s %s order by run", fConfig->GetShuttlelbTable(), whereClause);
1835
1836         TSQLResult* aResult = fServer[3]->Query(sqlQuery);
1837         if (!aResult) {
1838                 AliError(Form("Can't execute query <%s>!", sqlQuery.Data()));
1839                 return kFALSE;
1840         }
1841
1842         AliDebug(2,Form("Query = %s", sqlQuery.Data()));
1843
1844         if(aResult->GetRowCount() == 0) {
1845                 Log("SHUTTLE", "No entries in Shuttle Logbook match request");
1846                 delete aResult;
1847                 return kTRUE;
1848         }
1849
1850         // TODO Check field count!
1851         const UInt_t nCols = 23;
1852         if (aResult->GetFieldCount() != (Int_t) nCols) {
1853                 Log("SHUTTLE", "Invalid SQL result field number!");
1854                 delete aResult;
1855                 return kFALSE;
1856         }
1857
1858         TSQLRow* aRow;
1859         while ((aRow = aResult->Next())) {
1860                 TString runString(aRow->GetField(0), aRow->GetFieldLength(0));
1861                 Int_t run = runString.Atoi();
1862
1863                 AliShuttleLogbookEntry *entry = QueryRunParameters(run);
1864                 if (!entry)
1865                         continue;
1866
1867                 // loop on detectors
1868                 for(UInt_t ii = 0; ii < nCols; ii++)
1869                         entry->SetDetectorStatus(aResult->GetFieldName(ii), aRow->GetField(ii));
1870
1871                 entries.AddLast(entry);
1872                 delete aRow;
1873         }
1874
1875         delete aResult;
1876         return kTRUE;
1877 }
1878
1879 //______________________________________________________________________________________________
1880 AliShuttleLogbookEntry* AliShuttle::QueryRunParameters(Int_t run)
1881 {
1882         //
1883         // Retrieve run parameters written in the DAQ logbook and sets them into AliShuttleLogbookEntry object
1884         //
1885
1886         // check connection, in case connect
1887         if (!Connect(3))
1888                 return 0;
1889
1890         TString sqlQuery;
1891         sqlQuery.Form("select * from %s where run=%d", fConfig->GetDAQlbTable(), run);
1892
1893         TSQLResult* aResult = fServer[3]->Query(sqlQuery);
1894         if (!aResult) {
1895                 Log("SHUTTLE", Form("Can't execute query <%s>!", sqlQuery.Data()));
1896                 return 0;
1897         }
1898
1899         if (aResult->GetRowCount() == 0) {
1900                 Log("SHUTTLE", Form("QueryRunParameters - No entry in DAQ Logbook for run %d. Skipping", run));
1901                 delete aResult;
1902                 return 0;
1903         }
1904
1905         if (aResult->GetRowCount() > 1) {
1906                 Log("SHUTTLE", Form("QueryRunParameters - UNEXPECTED: "
1907                                 "more than one entry in DAQ Logbook for run %d!", run));
1908                 delete aResult;
1909                 return 0;
1910         }
1911
1912         TSQLRow* aRow = aResult->Next();
1913         if (!aRow)
1914         {
1915                 Log("SHUTTLE", Form("QueryRunParameters - Could not retrieve row for run %d. Skipping", run));
1916                 delete aResult;
1917                 return 0;
1918         }
1919
1920         AliShuttleLogbookEntry* entry = new AliShuttleLogbookEntry(run);
1921
1922         for (Int_t ii = 0; ii < aResult->GetFieldCount(); ii++)
1923                 entry->SetRunParameter(aResult->GetFieldName(ii), aRow->GetField(ii));
1924
1925         delete aRow;
1926         delete aResult;
1927         
1928         UInt_t startTime = entry->GetStartTime();
1929         UInt_t endTime = entry->GetEndTime();
1930         Bool_t ecsSuccess = entry->GetECSSuccess();
1931         
1932         TString totEventsStr = entry->GetRunParameter("totalEvents");  
1933         Int_t totEvents = totEventsStr.Atoi();
1934         
1935         // runs are accepted if they have ecsSuccess set or more than 1 event
1936         if (startTime != 0 && endTime != 0 && endTime > startTime && (totEvents > 1 || ecsSuccess))
1937         {
1938                 if (ecsSuccess == kFALSE)
1939                         Log("SHUTTLE", Form("Processing run %d although in status ECS failure, Reason: %s", run, entry->GetRunParameter("eor_reason")));
1940                 return entry;
1941         }
1942
1943         Bool_t skip = kFALSE;
1944                                 
1945         if (totEvents <= 1) 
1946         {
1947                 Log("SHUTTLE", Form("QueryRunParameters - Run %d has 1 event or less - Skipping!", run));
1948                 skip = kTRUE;
1949         } 
1950         else
1951         {
1952                 Log("SHUTTLE", Form("QueryRunParameters - Invalid parameters for Run %d: "
1953                         "startTime = %d, endTime = %d. Skipping (Shuttle won't be marked as DONE)!",
1954                         run, startTime, endTime));
1955         }
1956         
1957         if (skip)
1958         {
1959                 Log("SHUTTLE", Form("Marking SHUTTLE skipped for run %d", run));
1960                 fLogbookEntry = entry;
1961                 if (!UpdateShuttleLogbook("shuttle_skipped"))
1962                 {
1963                         AliError(Form("Could not update logbook for run %d !", run));
1964                 }
1965                 fLogbookEntry = 0;
1966         }
1967                         
1968         delete entry;
1969         return 0;
1970 }
1971
1972 //______________________________________________________________________________________________
1973 TMap* AliShuttle::GetValueSet(const char* host, Int_t port, const TSeqCollection* entries,
1974                               DCSType type, Int_t multiSplit)
1975 {
1976         // Retrieve all "entry" data points from the DCS server
1977         // host, port: TSocket connection parameters
1978         // entries: list of name of the alias or data point
1979         // type: kAlias or kDP
1980         // returns TMap of values, 0 when failure
1981         
1982         AliDCSClient client(host, port, fTimeout, fRetries, multiSplit);
1983
1984         TMap* result = 0;
1985         if (type == kAlias)
1986         {
1987                 result = client.GetAliasValues(entries, GetCurrentStartTime(), 
1988                         GetCurrentEndTime());
1989         } 
1990         else if (type == kDP)
1991         {
1992                 result = client.GetDPValues(entries, GetCurrentStartTime(), 
1993                         GetCurrentEndTime());
1994         }
1995
1996         if (result == 0)
1997         {
1998                 Log(fCurrentDetector.Data(), Form("GetValueSet - Can't get entries! Reason: %s",
1999                         client.GetErrorString(client.GetResultErrorCode())));
2000                 if (client.GetResultErrorCode() == AliDCSClient::fgkServerError)        
2001                         Log(fCurrentDetector.Data(), Form("GetValueSet - Server error code: %s",
2002                                 client.GetServerError().Data()));
2003
2004                 return 0;
2005         }
2006                 
2007         return result;
2008 }
2009
2010 //______________________________________________________________________________________________
2011 const char* AliShuttle::GetFile(Int_t system, const char* detector,
2012                 const char* id, const char* source)
2013 {
2014         // Get calibration file from file exchange servers
2015         // First queris the FXS database for the file name, using the run, detector, id and source info
2016         // then calls RetrieveFile(filename) for actual copy to local disk
2017         // run: current run being processed (given by Logbook entry fLogbookEntry)
2018         // detector: the Preprocessor name
2019         // id: provided as a parameter by the Preprocessor
2020         // source: provided by the Preprocessor through GetFileSources function
2021
2022         // check if test mode should simulate a FXS error
2023         if (fTestMode & kErrorFXSFiles)
2024         {
2025                 Log(detector, Form("GetFile - In TESTMODE - Simulating error while connecting to %s FXS", GetSystemName(system)));
2026                 return 0;
2027         }
2028         
2029         // check connection, in case connect
2030         if (!Connect(system))
2031         {
2032                 Log(detector, Form("GetFile - Couldn't connect to %s FXS database", GetSystemName(system)));
2033                 fFXSError = system;
2034                 return 0;
2035         }
2036
2037         // Query preparation
2038         TString sourceName(source);
2039         Int_t nFields = 3;
2040         TString sqlQueryStart = Form("select filePath,size,fileChecksum from %s where",
2041                                                                 fConfig->GetFXSdbTable(system));
2042         TString whereClause = Form("run=%d and detector=\"%s\" and fileId=\"%s\"",
2043                                                                 GetCurrentRun(), detector, id);
2044
2045         if (system == kDAQ)
2046         {
2047                 whereClause += Form(" and DAQsource=\"%s\"", source);
2048         }
2049         else if (system == kDCS)
2050         {
2051                 sourceName="none";
2052         }
2053         else if (system == kHLT)
2054         {
2055                 whereClause += Form(" and DDLnumbers=\"%s\"", source);
2056         }
2057
2058         TString sqlQuery = Form("%s %s", sqlQueryStart.Data(), whereClause.Data());
2059
2060         AliDebug(2, Form("SQL query: \n%s",sqlQuery.Data()));
2061
2062         // Query execution
2063         TSQLResult* aResult = 0;
2064         aResult = dynamic_cast<TSQLResult*> (fServer[system]->Query(sqlQuery));
2065         if (!aResult) {
2066                 Log(detector, Form("GetFile - Can't execute SQL query to %s database for: id = %s, source = %s",
2067                                 GetSystemName(system), id, sourceName.Data()));
2068                 fFXSError = system;
2069                 return 0;
2070         }
2071
2072         if (aResult->GetRowCount() == 0)
2073         {
2074                 Log(detector,
2075                         Form("GetFile - No entry in %s FXS db for: id = %s, source = %s",
2076                                 GetSystemName(system), id, sourceName.Data()));
2077                 delete aResult;
2078                 return 0;
2079         }
2080
2081         if (aResult->GetRowCount() > 1) {
2082                 Log(detector,
2083                         Form("GetFile - More than one entry in %s FXS db for: id = %s, source = %s",
2084                                 GetSystemName(system), id, sourceName.Data()));
2085                 fFXSError = system;
2086                 delete aResult;
2087                 return 0;
2088         }
2089
2090         if (aResult->GetFieldCount() != nFields) {
2091                 Log(detector,
2092                         Form("GetFileName - Wrong field count in %s FXS db for: id = %s, source = %s",
2093                                 GetSystemName(system), id, sourceName.Data()));
2094                 fFXSError = system;
2095                 delete aResult;
2096                 return 0;
2097         }
2098
2099         TSQLRow* aRow = dynamic_cast<TSQLRow*> (aResult->Next());
2100
2101         if (!aRow){
2102                 Log(detector, Form("GetFile - Empty set result in %s FXS db from query: id = %s, source = %s",
2103                                 GetSystemName(system), id, sourceName.Data()));
2104                 fFXSError = system;
2105                 delete aResult;
2106                 return 0;
2107         }
2108
2109         TString filePath(aRow->GetField(0), aRow->GetFieldLength(0));
2110         TString fileSize(aRow->GetField(1), aRow->GetFieldLength(1));
2111         TString fileChecksum(aRow->GetField(2), aRow->GetFieldLength(2));
2112
2113         delete aResult;
2114         delete aRow;
2115
2116         AliDebug(2, Form("filePath = %s; size = %s, fileChecksum = %s",
2117                                 filePath.Data(), fileSize.Data(), fileChecksum.Data()));
2118
2119         // retrieved file is renamed to make it unique
2120         TString localFileName = Form("%s/%s_%d_process/%s_%s_%d_%s_%s.shuttle",
2121                                         GetShuttleTempDir(), detector, GetCurrentRun(),
2122                                         GetSystemName(system), detector, GetCurrentRun(), 
2123                                         id, sourceName.Data());
2124
2125
2126         // file retrieval from FXS
2127         UInt_t nRetries = 0;
2128         UInt_t maxRetries = 3;
2129         Bool_t result = kFALSE;
2130
2131         // copy!! if successful TSystem::Exec returns 0
2132         while (nRetries++ < maxRetries) {
2133                 AliDebug(2, Form("Trying to copy file. Retry # %d", nRetries));
2134                 result = RetrieveFile(system, filePath.Data(), localFileName.Data());
2135                 if (!result)
2136                 {
2137                         Log(detector, Form("GetFile - Copy of file %s from %s FXS failed",
2138                                         filePath.Data(), GetSystemName(system)));
2139                         continue;
2140                 } 
2141
2142                 if (fileSize.Length()>0)
2143                 {
2144                         // compare filesize of local file with the one stored in the FXS DB
2145                         Long_t size = -1;
2146                         Int_t sizeComp = gSystem->GetPathInfo(localFileName.Data(), 0, &size, 0, 0);
2147
2148                         if (sizeComp != 0 || size != fileSize.Atoi())
2149                         {
2150                                 Log(detector, Form("GetFile - size of file %s does not match with local copy!",
2151                                                         filePath.Data()));
2152                                 result = kFALSE;
2153                                 continue;
2154                         }
2155
2156                 } else {
2157                         Log(fCurrentDetector, Form("GetFile - size of file %s not set in %s database, skipping comparison",
2158                                                 filePath.Data(), GetSystemName(system)));
2159                 }
2160
2161                 if (fileChecksum.Length()>0)
2162                 {
2163                         // compare md5sum of local file with the one stored in the FXS DB
2164                         if(fileChecksum.Contains(' ')) fileChecksum.Resize(fileChecksum.First(' '));
2165                         Int_t md5Comp = gSystem->Exec(Form("md5sum %s |grep %s > /dev/null 2> /dev/null",
2166                                                 localFileName.Data(), fileChecksum.Data()));
2167
2168                         if (md5Comp != 0)
2169                         {
2170                                 Log(detector, Form("GetFile - md5sum of file %s does not match with local copy!",
2171                                                         filePath.Data()));
2172                                 result = kFALSE;
2173                                 continue;
2174                         }
2175                 } else {
2176                         Log(fCurrentDetector, Form("GetFile - md5sum of file %s not set in %s database, skipping comparison",
2177                                                         filePath.Data(), GetSystemName(system)));
2178                 }
2179                 if (result) break;
2180         }
2181
2182         if (!result) 
2183         {
2184                 fFXSError = system;
2185                 return 0;
2186         }
2187
2188         fFXSCalled[system]=kTRUE;
2189         TObjString *fileParams = new TObjString(Form("%s#!?!#%s", id, sourceName.Data()));
2190         fFXSlist[system].Add(fileParams);
2191
2192         static TString staticLocalFileName;
2193         staticLocalFileName.Form("%s", localFileName.Data());
2194         
2195         Log(fCurrentDetector, Form("GetFile - Retrieved file with id %s and "
2196                         "source %s from %s to %s", id, source, 
2197                         GetSystemName(system), localFileName.Data()));
2198                         
2199         return staticLocalFileName.Data();
2200 }
2201
2202 //______________________________________________________________________________________________
2203 Bool_t AliShuttle::RetrieveFile(UInt_t system, const char* fxsFileName, const char* localFileName)
2204 {
2205         //
2206         // Copies file from FXS to local Shuttle machine
2207         //
2208
2209         // check temp directory: trying to cd to temp; if it does not exist, create it
2210         AliDebug(2, Form("Copy file %s from %s FXS into %s",
2211                         GetSystemName(system), fxsFileName, localFileName));
2212                         
2213         TString tmpDir(localFileName);
2214         
2215         tmpDir = tmpDir(0,tmpDir.Last('/'));
2216
2217         Int_t noDir = gSystem->GetPathInfo(tmpDir.Data(), 0, (Long64_t*) 0, 0, 0);
2218         if (noDir) // temp dir does not exists!
2219         {
2220                 if (gSystem->mkdir(tmpDir.Data(), 1))
2221                 {
2222                         Log(fCurrentDetector.Data(), "RetrieveFile - could not make temp directory!!");
2223                         return kFALSE;
2224                 }
2225         }
2226
2227         TString command = Form("scp -oPort=%d -2 %s@%s:%s/%s %s",
2228                 fConfig->GetFXSPort(system),
2229                 fConfig->GetFXSUser(system),
2230                 fConfig->GetFXSHost(system),
2231                 fConfig->GetFXSBaseFolder(system),
2232                 fxsFileName,
2233                 localFileName);
2234
2235         AliDebug(2, Form("%s",command.Data()));
2236
2237         Bool_t result = (gSystem->Exec(command.Data()) == 0);
2238
2239         return result;
2240 }
2241
2242 //______________________________________________________________________________________________
2243 TList* AliShuttle::GetFileSources(Int_t system, const char* detector, const char* id)
2244 {
2245         //
2246         // Get sources producing the condition file Id from file exchange servers
2247         // if id is NULL all sources are returned (distinct)
2248         //
2249
2250         if (id)
2251         {
2252                 Log(detector, Form("GetFileSources - Querying %s FXS for files with id %s produced by %s", GetSystemName(system), id, detector));
2253         } else {
2254                 Log(detector, Form("GetFileSources - Querying %s FXS for files produced by %s", GetSystemName(system), detector));
2255         }
2256         
2257         // check if test mode should simulate a FXS error
2258         if (fTestMode & kErrorFXSSources)
2259         {
2260                 Log(detector, Form("GetFileSources - In TESTMODE - Simulating error while connecting to %s FXS", GetSystemName(system)));
2261                 return 0;
2262         }
2263
2264         if (system == kDCS)
2265         {
2266                 Log(detector, "GetFileSources - WARNING: DCS system has only one source of data!");
2267                 TList *list = new TList();
2268                 list->SetOwner(1);
2269                 list->Add(new TObjString(" "));
2270                 return list;
2271         }
2272
2273         // check connection, in case connect
2274         if (!Connect(system))
2275         {
2276                 Log(detector, Form("GetFileSources - Couldn't connect to %s FXS database", GetSystemName(system)));
2277                 fFXSError = system;
2278                 return NULL;
2279         }
2280
2281         TString sourceName = 0;
2282         if (system == kDAQ)
2283         {
2284                 sourceName = "DAQsource";
2285         } else if (system == kHLT)
2286         {
2287                 sourceName = "DDLnumbers";
2288         }
2289
2290         TString sqlQueryStart = Form("select distinct %s from %s where", sourceName.Data(), fConfig->GetFXSdbTable(system));
2291         TString whereClause = Form("run=%d and detector=\"%s\"",
2292                                 GetCurrentRun(), detector);
2293         if (id)
2294                 whereClause += Form(" and fileId=\"%s\"", id);
2295         TString sqlQuery = Form("%s %s", sqlQueryStart.Data(), whereClause.Data());
2296
2297         AliDebug(2, Form("SQL query: \n%s",sqlQuery.Data()));
2298
2299         // Query execution
2300         TSQLResult* aResult;
2301         aResult = fServer[system]->Query(sqlQuery);
2302         if (!aResult) {
2303                 Log(detector, Form("GetFileSources - Can't execute SQL query to %s database for id: %s",
2304                                 GetSystemName(system), id));
2305                 fFXSError = system;
2306                 return 0;
2307         }
2308
2309         TList *list = new TList();
2310         list->SetOwner(1);
2311         
2312         if (aResult->GetRowCount() == 0)
2313         {
2314                 Log(detector,
2315                         Form("GetFileSources - No entry in %s FXS table for id: %s", GetSystemName(system), id));
2316                 delete aResult;
2317                 return list;
2318         }
2319
2320         Log(detector, Form("GetFileSources - Found %d sources", aResult->GetRowCount()));
2321
2322         TSQLRow* aRow;
2323         while ((aRow = aResult->Next()))
2324         {
2325
2326                 TString source(aRow->GetField(0), aRow->GetFieldLength(0));
2327                 AliDebug(2, Form("%s = %s", sourceName.Data(), source.Data()));
2328                 list->Add(new TObjString(source));
2329                 delete aRow;
2330         }
2331
2332         delete aResult;
2333
2334         return list;
2335 }
2336
2337 //______________________________________________________________________________________________
2338 TList* AliShuttle::GetFileIDs(Int_t system, const char* detector, const char* source)
2339 {
2340         //
2341         // Get all ids of condition files produced by a given source from file exchange servers
2342         //
2343         
2344         Log(detector, Form("GetFileIDs - Retrieving ids with source %s with %s", source, GetSystemName(system)));
2345
2346         // check if test mode should simulate a FXS error
2347         if (fTestMode & kErrorFXSSources)
2348         {
2349                 Log(detector, Form("GetFileIDs - In TESTMODE - Simulating error while connecting to %s FXS", GetSystemName(system)));
2350                 return 0;
2351         }
2352
2353         // check connection, in case connect
2354         if (!Connect(system))
2355         {
2356                 Log(detector, Form("GetFileIDs - Couldn't connect to %s FXS database", GetSystemName(system)));
2357                 return NULL;
2358         }
2359
2360         TString sourceName = 0;
2361         if (system == kDAQ)
2362         {
2363                 sourceName = "DAQsource";
2364         } else if (system == kHLT)
2365         {
2366                 sourceName = "DDLnumbers";
2367         }
2368
2369         TString sqlQueryStart = Form("select fileId from %s where", fConfig->GetFXSdbTable(system));
2370         TString whereClause = Form("run=%d and detector=\"%s\"",
2371                                 GetCurrentRun(), detector);
2372         if (sourceName.Length() > 0 && source)
2373                 whereClause += Form(" and %s=\"%s\"", sourceName.Data(), source);
2374         TString sqlQuery = Form("%s %s", sqlQueryStart.Data(), whereClause.Data());
2375
2376         AliDebug(2, Form("SQL query: \n%s",sqlQuery.Data()));
2377
2378         // Query execution
2379         TSQLResult* aResult;
2380         aResult = fServer[system]->Query(sqlQuery);
2381         if (!aResult) {
2382                 Log(detector, Form("GetFileIDs - Can't execute SQL query to %s database for source: %s",
2383                                 GetSystemName(system), source));
2384                 return 0;
2385         }
2386
2387         TList *list = new TList();
2388         list->SetOwner(1);
2389         
2390         if (aResult->GetRowCount() == 0)
2391         {
2392                 Log(detector,
2393                         Form("GetFileIDs - No entry in %s FXS table for source: %s", GetSystemName(system), source));
2394                 delete aResult;
2395                 return list;
2396         }
2397
2398         Log(detector, Form("GetFileIDs - Found %d ids", aResult->GetRowCount()));
2399
2400         TSQLRow* aRow;
2401
2402         while ((aRow = aResult->Next()))
2403         {
2404
2405                 TString id(aRow->GetField(0), aRow->GetFieldLength(0));
2406                 AliDebug(2, Form("fileId = %s", id.Data()));
2407                 list->Add(new TObjString(id));
2408                 delete aRow;
2409         }
2410
2411         delete aResult;
2412
2413         return list;
2414 }
2415
2416 //______________________________________________________________________________________________
2417 Bool_t AliShuttle::Connect(Int_t system)
2418 {
2419         // Connect to MySQL Server of the system's FXS MySQL databases
2420         // DAQ Logbook, Shuttle Logbook and DAQ FXS db are on the same host
2421         //
2422
2423         // check connection: if already connected return
2424         if(fServer[system] && fServer[system]->IsConnected()) return kTRUE;
2425
2426         TString dbHost, dbUser, dbPass, dbName;
2427
2428         if (system < 3) // FXS db servers
2429         {
2430                 dbHost = Form("mysql://%s:%d", fConfig->GetFXSdbHost(system), fConfig->GetFXSdbPort(system));
2431                 dbUser = fConfig->GetFXSdbUser(system);
2432                 dbPass = fConfig->GetFXSdbPass(system);
2433                 dbName =   fConfig->GetFXSdbName(system);
2434         } else { // Run & Shuttle logbook servers
2435         // TODO Will the Shuttle logbook server be the same as the Run logbook server ???
2436                 dbHost = Form("mysql://%s:%d", fConfig->GetDAQlbHost(), fConfig->GetDAQlbPort());
2437                 dbUser = fConfig->GetDAQlbUser();
2438                 dbPass = fConfig->GetDAQlbPass();
2439                 dbName =   fConfig->GetDAQlbDB();
2440         }
2441
2442         fServer[system] = TSQLServer::Connect(dbHost.Data(), dbUser.Data(), dbPass.Data());
2443         if (!fServer[system] || !fServer[system]->IsConnected()) {
2444                 if(system < 3)
2445                 {
2446                 AliError(Form("Can't establish connection to FXS database for %s",
2447                                         AliShuttleInterface::GetSystemName(system)));
2448                 } else {
2449                 AliError("Can't establish connection to Run logbook.");
2450                 }
2451                 if(fServer[system]) delete fServer[system];
2452                 return kFALSE;
2453         }
2454
2455         // Get tables
2456         TSQLResult* aResult=0;
2457         switch(system){
2458                 case kDAQ:
2459                         aResult = fServer[kDAQ]->GetTables(dbName.Data());
2460                         break;
2461                 case kDCS:
2462                         aResult = fServer[kDCS]->GetTables(dbName.Data());
2463                         break;
2464                 case kHLT:
2465                         aResult = fServer[kHLT]->GetTables(dbName.Data());
2466                         break;
2467                 default:
2468                         aResult = fServer[3]->GetTables(dbName.Data());
2469                         break;
2470         }
2471
2472         delete aResult;
2473         return kTRUE;
2474 }
2475
2476 //______________________________________________________________________________________________
2477 Bool_t AliShuttle::UpdateTable()
2478 {
2479         //
2480         // Update FXS table filling time_processed field in all rows corresponding to current run and detector
2481         //
2482
2483         Bool_t result = kTRUE;
2484
2485         for (UInt_t system=0; system<3; system++)
2486         {
2487                 if(!fFXSCalled[system]) continue;
2488
2489                 // check connection, in case connect
2490                 if (!Connect(system))
2491                 {
2492                         Log(fCurrentDetector, Form("UpdateTable - Couldn't connect to %s FXS database", GetSystemName(system)));
2493                         result = kFALSE;
2494                         continue;
2495                 }
2496
2497                 TTimeStamp now; // now
2498
2499                 // Loop on FXS list entries
2500                 TIter iter(&fFXSlist[system]);
2501                 TObjString *aFXSentry=0;
2502                 while ((aFXSentry = dynamic_cast<TObjString*> (iter.Next())))
2503                 {
2504                         TString aFXSentrystr = aFXSentry->String();
2505                         TObjArray *aFXSarray = aFXSentrystr.Tokenize("#!?!#");
2506                         if (!aFXSarray || aFXSarray->GetEntries() != 2 )
2507                         {
2508                                 Log(fCurrentDetector, Form("UpdateTable - error updating %s FXS entry. Check string: <%s>",
2509                                         GetSystemName(system), aFXSentrystr.Data()));
2510                                 if(aFXSarray) delete aFXSarray;
2511                                 result = kFALSE;
2512                                 continue;
2513                         }
2514                         const char* fileId = ((TObjString*) aFXSarray->At(0))->GetName();
2515                         const char* source = ((TObjString*) aFXSarray->At(1))->GetName();
2516
2517                         TString whereClause;
2518                         if (system == kDAQ)
2519                         {
2520                                 whereClause = Form("where run=%d and detector=\"%s\" and fileId=\"%s\" and DAQsource=\"%s\";",
2521                                                         GetCurrentRun(), fCurrentDetector.Data(), fileId, source);
2522                         }
2523                         else if (system == kDCS)
2524                         {
2525                                 whereClause = Form("where run=%d and detector=\"%s\" and fileId=\"%s\";",
2526                                                         GetCurrentRun(), fCurrentDetector.Data(), fileId);
2527                         }
2528                         else if (system == kHLT)
2529                         {
2530                                 whereClause = Form("where run=%d and detector=\"%s\" and fileId=\"%s\" and DDLnumbers=\"%s\";",
2531                                                         GetCurrentRun(), fCurrentDetector.Data(), fileId, source);
2532                         }
2533
2534                         delete aFXSarray;
2535
2536                         TString sqlQuery = Form("update %s set time_processed=%d %s", fConfig->GetFXSdbTable(system),
2537                                                                 now.GetSec(), whereClause.Data());
2538
2539                         AliDebug(2, Form("SQL query: \n%s",sqlQuery.Data()));
2540
2541                         // Query execution
2542                         TSQLResult* aResult;
2543                         aResult = dynamic_cast<TSQLResult*> (fServer[system]->Query(sqlQuery));
2544                         if (!aResult)
2545                         {
2546                                 Log(fCurrentDetector, Form("UpdateTable - %s db: can't execute SQL query <%s>",
2547                                                                 GetSystemName(system), sqlQuery.Data()));
2548                                 result = kFALSE;
2549                                 continue;
2550                         }
2551                         delete aResult;
2552                 }
2553         }
2554
2555         return result;
2556 }
2557
2558 //______________________________________________________________________________________________
2559 Bool_t AliShuttle::UpdateTableFailCase()
2560 {
2561         // Update FXS table filling time_processed field in all rows corresponding to current run and detector
2562         // this is called in case the preprocessor is declared failed for the current run, because
2563         // the fields are updated only in case of success
2564
2565         Bool_t result = kTRUE;
2566
2567         for (UInt_t system=0; system<3; system++)
2568         {
2569                 // check connection, in case connect
2570                 if (!Connect(system))
2571                 {
2572                         Log(fCurrentDetector, Form("UpdateTableFailCase - Couldn't connect to %s FXS database",
2573                                                         GetSystemName(system)));
2574                         result = kFALSE;
2575                         continue;
2576                 }
2577
2578                 TTimeStamp now; // now
2579
2580                 // Loop on FXS list entries
2581
2582                 TString whereClause = Form("where run=%d and detector=\"%s\";",
2583                                                 GetCurrentRun(), fCurrentDetector.Data());
2584
2585
2586                 TString sqlQuery = Form("update %s set time_processed=%d %s", fConfig->GetFXSdbTable(system),
2587                                                         now.GetSec(), whereClause.Data());
2588
2589                 AliDebug(2, Form("SQL query: \n%s",sqlQuery.Data()));
2590
2591                 // Query execution
2592                 TSQLResult* aResult;
2593                 aResult = dynamic_cast<TSQLResult*> (fServer[system]->Query(sqlQuery));
2594                 if (!aResult)
2595                 {
2596                         Log(fCurrentDetector, Form("UpdateTableFailCase - %s db: can't execute SQL query <%s>",
2597                                                         GetSystemName(system), sqlQuery.Data()));
2598                         result = kFALSE;
2599                         continue;
2600                 }
2601                 delete aResult;
2602         }
2603
2604         return result;
2605 }
2606
2607 //______________________________________________________________________________________________
2608 Bool_t AliShuttle::UpdateShuttleLogbook(const char* detector, const char* status)
2609 {
2610         //
2611         // Update Shuttle logbook filling detector or shuttle_done column
2612         // ex. of usage: UpdateShuttleLogbook("PHOS", "DONE") or UpdateShuttleLogbook("shuttle_done")
2613         //
2614
2615         // check connection, in case connect
2616         if(!Connect(3)){
2617                 Log("SHUTTLE", "UpdateShuttleLogbook - Couldn't connect to DAQ Logbook.");
2618                 return kFALSE;
2619         }
2620
2621         TString detName(detector);
2622         TString setClause;
2623         if (detName == "shuttle_done" || detName == "shuttle_skipped")
2624         {
2625                 setClause = "set shuttle_done=1";
2626                 
2627                 if (detName == "shuttle_done")
2628                 {
2629                         if (TouchFile() != kTRUE)
2630                                 return kFALSE;
2631                         
2632                         SendMLRunInfo("Done");
2633                 }
2634                 else 
2635                         SendMLRunInfo("Skipped");
2636         } 
2637         else {
2638                 TString statusStr(status);
2639                 if(statusStr.Contains("done", TString::kIgnoreCase) ||
2640                    statusStr.Contains("failed", TString::kIgnoreCase)){
2641                         setClause = Form("set %s=\"%s\"", detector, status);
2642                 } else {
2643                         Log("SHUTTLE",
2644                                 Form("UpdateShuttleLogbook - Invalid status <%s> for detector %s",
2645                                         status, detector));
2646                         return kFALSE;
2647                 }
2648         }
2649
2650         TString whereClause = Form("where run=%d", GetCurrentRun());
2651
2652         TString sqlQuery = Form("update %s %s %s",
2653                                         fConfig->GetShuttlelbTable(), setClause.Data(), whereClause.Data());
2654
2655         AliDebug(2, Form("SQL query: \n%s",sqlQuery.Data()));
2656
2657         // Query execution
2658         TSQLResult* aResult;
2659         aResult = dynamic_cast<TSQLResult*> (fServer[3]->Query(sqlQuery));
2660         if (!aResult) {
2661                 Log("SHUTTLE", Form("UpdateShuttleLogbook - Can't execute query <%s>", sqlQuery.Data()));
2662                 return kFALSE;
2663         }
2664         delete aResult;
2665
2666         return kTRUE;
2667 }
2668
2669 //______________________________________________________________________________________________
2670 Int_t AliShuttle::GetCurrentRun() const
2671 {
2672         //
2673         // Get current run from logbook entry
2674         //
2675
2676         return fLogbookEntry ? fLogbookEntry->GetRun() : -1;
2677 }
2678
2679 //______________________________________________________________________________________________
2680 UInt_t AliShuttle::GetCurrentStartTime() const
2681 {
2682         //
2683         // get current start time
2684         //
2685
2686         return fLogbookEntry ? fLogbookEntry->GetStartTime() : 0;
2687 }
2688
2689 //______________________________________________________________________________________________
2690 UInt_t AliShuttle::GetCurrentEndTime() const
2691 {
2692         //
2693         // get current end time from logbook entry
2694         //
2695
2696         return fLogbookEntry ? fLogbookEntry->GetEndTime() : 0;
2697 }
2698
2699 //______________________________________________________________________________________________
2700 UInt_t AliShuttle::GetCurrentYear() const
2701 {
2702         //
2703         // Get current year from logbook entry
2704         //
2705
2706         if (!fLogbookEntry) return 0;
2707         
2708         TTimeStamp startTime(GetCurrentStartTime());
2709         TString year =  Form("%d",startTime.GetDate());
2710         year = year(0,4);
2711         
2712         return year.Atoi();
2713 }
2714
2715 //______________________________________________________________________________________________
2716 const char* AliShuttle::GetLHCPeriod() const
2717 {
2718         //
2719         // Get current LHC period from logbook entry
2720         //
2721
2722         if (!fLogbookEntry) return 0;
2723                 
2724         return fLogbookEntry->GetRunParameter("LHCperiod");
2725 }
2726
2727 //______________________________________________________________________________________________
2728 void AliShuttle::Log(const char* detector, const char* message)
2729 {
2730         //
2731         // Fill log string with a message
2732         //
2733
2734         TString logRunDir = GetShuttleLogDir();
2735         if (GetCurrentRun() >=0)
2736                 logRunDir += Form("/%d", GetCurrentRun());
2737         
2738         void* dir = gSystem->OpenDirectory(logRunDir.Data());
2739         if (dir == NULL) {
2740                 if (gSystem->mkdir(logRunDir.Data(), kTRUE)) {
2741                         AliError(Form("Can't open directory <%s>", GetShuttleLogDir()));
2742                         return;
2743                 }
2744
2745         } else {
2746                 gSystem->FreeDirectory(dir);
2747         }
2748
2749         TString toLog = Form("%s UTC (%d): %s - ", TTimeStamp(time(0)).AsString("s"), getpid(), detector);
2750         if (GetCurrentRun() >= 0) 
2751                 toLog += Form("run %d - ", GetCurrentRun());
2752         toLog += Form("%s", message);
2753
2754         AliInfo(toLog.Data());
2755         
2756         // if we redirect the log output already to the file, leave here
2757         if (fOutputRedirected && strcmp(detector, "SHUTTLE") != 0)
2758                 return;
2759
2760         TString fileName = GetLogFileName(detector);
2761         
2762         gSystem->ExpandPathName(fileName);
2763
2764         ofstream logFile;
2765         logFile.open(fileName, ofstream::out | ofstream::app);
2766
2767         if (!logFile.is_open()) {
2768                 AliError(Form("Could not open file %s", fileName.Data()));
2769                 return;
2770         }
2771
2772         logFile << toLog.Data() << "\n";
2773
2774         logFile.close();
2775 }
2776
2777 //______________________________________________________________________________________________
2778 TString AliShuttle::GetLogFileName(const char* detector) const
2779 {
2780         // 
2781         // returns the name of the log file for a given sub detector
2782         //
2783         
2784         TString fileName;
2785         
2786         if (GetCurrentRun() >= 0) 
2787         {
2788                 fileName.Form("%s/%d/%s_%d.log", GetShuttleLogDir(), GetCurrentRun(), 
2789                         detector, GetCurrentRun());
2790         } else {
2791                 fileName.Form("%s/%s.log", GetShuttleLogDir(), detector);
2792         }
2793
2794         return fileName;
2795 }
2796
2797 //______________________________________________________________________________________________
2798 void AliShuttle::SendAlive()
2799 {
2800         // sends alive message to ML
2801         
2802         TMonaLisaText mlStatus("SHUTTLE_status", "Alive");
2803
2804         TList mlList;
2805         mlList.Add(&mlStatus);
2806
2807         fMonaLisa->SendParameters(&mlList, "__PROCESSINGINFO__");
2808 }
2809
2810 //______________________________________________________________________________________________
2811 Bool_t AliShuttle::Collect(Int_t run)
2812 {
2813         //
2814         // Collects conditions data for all UNPROCESSED run written to DAQ LogBook in case of run = -1 (default)
2815         // If a dedicated run is given this run is processed
2816         //
2817         // In operational mode, this is the Shuttle function triggered by the EOR signal.
2818         //
2819
2820         if (run == -1)
2821                 Log("SHUTTLE","Collect - Shuttle called. Collecting conditions data for unprocessed runs");
2822         else
2823                 Log("SHUTTLE", Form("Collect - Shuttle called. Collecting conditions data for run %d", run));
2824
2825         SetLastAction("Starting");
2826
2827         // create ML instance
2828         if (!fMonaLisa)
2829                 fMonaLisa = new TMonaLisaWriter(fConfig->GetMonitorHost(), fConfig->GetMonitorTable());
2830                 
2831         SendAlive();
2832         CountOpenRuns();
2833
2834         TString whereClause("where shuttle_done=0");
2835         if (run != -1)
2836                 whereClause += Form(" and run=%d", run);
2837
2838         TObjArray shuttleLogbookEntries;
2839         if (!QueryShuttleLogbook(whereClause, shuttleLogbookEntries))
2840         {
2841                 Log("SHUTTLE", "Collect - Can't retrieve entries from Shuttle logbook");
2842                 return kFALSE;
2843         }
2844
2845         if (shuttleLogbookEntries.GetEntries() == 0)
2846         {
2847                 if (run == -1)
2848                         Log("SHUTTLE","Collect - Found no UNPROCESSED runs in Shuttle logbook");
2849                 else
2850                         Log("SHUTTLE", Form("Collect - Run %d is already DONE "
2851                                                 "or it does not exist in Shuttle logbook", run));
2852                 return kTRUE;
2853         }
2854
2855         for (UInt_t iDet=0; iDet<NDetectors(); iDet++)
2856                 fFirstUnprocessed[iDet] = kTRUE;
2857
2858         if (run != -1)
2859         {
2860                 // query Shuttle logbook for earlier runs, check if some detectors are unprocessed,
2861                 // flag them into fFirstUnprocessed array
2862                 TString whereClauseBis(Form("where shuttle_done=0 and run < %d", run));
2863                 TObjArray tmpLogbookEntries;
2864                 if (!QueryShuttleLogbook(whereClauseBis, tmpLogbookEntries))
2865                 {
2866                         Log("SHUTTLE", "Collect - Can't retrieve entries from Shuttle logbook");
2867                         return kFALSE;
2868                 }
2869
2870                 TIter iter(&tmpLogbookEntries);
2871                 AliShuttleLogbookEntry* anEntry = 0;
2872                 while ((anEntry = dynamic_cast<AliShuttleLogbookEntry*> (iter.Next())))
2873                 {
2874                         for (UInt_t iDet=0; iDet<NDetectors(); iDet++)
2875                         {
2876                                 if (anEntry->GetDetectorStatus(iDet) == AliShuttleLogbookEntry::kUnprocessed)
2877                                 {
2878                                         AliDebug(2, Form("Run %d: setting %s as \"not first time unprocessed\"",
2879                                                         anEntry->GetRun(), GetDetName(iDet)));
2880                                         fFirstUnprocessed[iDet] = kFALSE;
2881                                 }
2882                         }
2883
2884                 }
2885
2886         }
2887
2888         if (!RetrieveConditionsData(shuttleLogbookEntries))
2889         {
2890                 Log("SHUTTLE", "Collect - Process of at least one run failed");
2891                 CountOpenRuns();
2892                 return kFALSE;
2893         }
2894
2895         Log("SHUTTLE", "Collect - Requested run(s) successfully processed");
2896         CountOpenRuns();
2897         return kTRUE;
2898 }
2899
2900 //______________________________________________________________________________________________
2901 Bool_t AliShuttle::RetrieveConditionsData(const TObjArray& dateEntries)
2902 {
2903         //
2904         // Retrieve conditions data for all runs that aren't processed yet
2905         //
2906
2907         Bool_t hasError = kFALSE;
2908
2909         TIter iter(&dateEntries);
2910         AliShuttleLogbookEntry* anEntry;
2911
2912         while ((anEntry = (AliShuttleLogbookEntry*) iter.Next())){
2913                 if (!Process(anEntry)){
2914                         hasError = kTRUE;
2915                 }
2916
2917                 // clean SHUTTLE temp directory
2918                 //TString filename = Form("%s/*.shuttle", GetShuttleTempDir());
2919                 //RemoveFile(filename.Data());
2920         }
2921
2922         return hasError == kFALSE;
2923 }
2924
2925 //______________________________________________________________________________________________
2926 ULong_t AliShuttle::GetTimeOfLastAction() const
2927 {
2928         //
2929         // Gets time of last action
2930         //
2931
2932         ULong_t tmp;
2933
2934         fMonitoringMutex->Lock();
2935
2936         tmp = fLastActionTime;
2937
2938         fMonitoringMutex->UnLock();
2939
2940         return tmp;
2941 }
2942
2943 //______________________________________________________________________________________________
2944 const TString AliShuttle::GetLastAction() const
2945 {
2946         //
2947         // returns a string description of the last action
2948         //
2949
2950         TString tmp;
2951
2952         fMonitoringMutex->Lock();
2953         
2954         tmp = fLastAction;
2955         
2956         fMonitoringMutex->UnLock();
2957
2958         return tmp;
2959 }
2960
2961 //______________________________________________________________________________________________
2962 void AliShuttle::SetLastAction(const char* action)
2963 {
2964         //
2965         // updates the monitoring variables
2966         //
2967
2968         fMonitoringMutex->Lock();
2969
2970         fLastAction = action;
2971         fLastActionTime = time(0);
2972         
2973         fMonitoringMutex->UnLock();
2974 }
2975
2976 //______________________________________________________________________________________________
2977 const char* AliShuttle::GetRunParameter(const char* param)
2978 {
2979         //
2980         // returns run parameter read from DAQ logbook
2981         //
2982
2983         if(!fLogbookEntry) {
2984                 AliError("No logbook entry!");
2985                 return 0;
2986         }
2987
2988         return fLogbookEntry->GetRunParameter(param);
2989 }
2990
2991 //______________________________________________________________________________________________
2992 AliCDBEntry* AliShuttle::GetFromOCDB(const char* detector, const AliCDBPath& path)
2993 {
2994         //
2995         // returns object from OCDB valid for current run
2996         //
2997
2998         if (fTestMode & kErrorOCDB)
2999         {
3000                 Log(detector, "GetFromOCDB - In TESTMODE - Simulating error with OCDB");
3001                 return 0;
3002         }
3003         
3004         AliCDBStorage *sto = AliCDBManager::Instance()->GetStorage(fgkMainCDB);
3005         if (!sto)
3006         {
3007                 Log(detector, "GetFromOCDB - Cannot activate main OCDB for query!");
3008                 return 0;
3009         }
3010
3011         return dynamic_cast<AliCDBEntry*> (sto->Get(path, GetCurrentRun()));
3012 }
3013
3014 //______________________________________________________________________________________________
3015 Bool_t AliShuttle::SendMail(EMailTarget target, Int_t system)
3016 {
3017         //
3018         // sends a mail to the subdetector expert in case of preprocessor error
3019         //
3020         
3021         if (fTestMode != kNone)
3022                 return kTRUE;
3023                 
3024         if (!fConfig->SendMail()) 
3025                 return kTRUE;
3026
3027         if (target == kDCSEMail || target == kFXSEMail) {
3028                 if (!fFirstProcessing)
3029                 return kTRUE;
3030         }
3031
3032         Int_t runMode = (Int_t)fConfig->GetRunMode();
3033         TString tmpStr;
3034         if (runMode == 0) tmpStr = " Nightly Test:";
3035         else tmpStr = " Data Taking:"; 
3036         void* dir = gSystem->OpenDirectory(GetShuttleLogDir());
3037         if (dir == NULL)
3038         {
3039                 if (gSystem->mkdir(GetShuttleLogDir(), kTRUE))
3040                 {
3041                         Log("SHUTTLE", Form("SendMail - Can't open directory <%s>", GetShuttleLogDir()));
3042                         return kFALSE;
3043                 }
3044
3045         } else {
3046                 gSystem->FreeDirectory(dir);
3047         }
3048
3049         // det experts in to
3050         TString to="";
3051         TIter *iterExperts = 0;
3052         if (target == kDCSEMail) {
3053                 iterExperts = new TIter(fConfig->GetAdmins(AliShuttleConfig::kAmanda));
3054         }
3055         else if (target == kFXSEMail) {
3056                 iterExperts = new TIter(fConfig->GetAdmins(system));
3057         }
3058         if (iterExperts) {
3059                 TObjString *anExpert=0;
3060                 while ((anExpert = (TObjString*) iterExperts->Next()))
3061                 {
3062                         to += Form("%s,", anExpert->GetName());
3063                 }
3064                 delete iterExperts;
3065         }
3066
3067         // add subdetector experts      
3068         iterExperts = new TIter(fConfig->GetResponsibles(fCurrentDetector));
3069         TObjString *anExpert=0;
3070         while ((anExpert = (TObjString*) iterExperts->Next()))
3071         {
3072                 to += Form("%s,", anExpert->GetName());
3073         }
3074         delete iterExperts;
3075         
3076         if (to.Length() > 0)
3077           to.Remove(to.Length()-1);
3078         AliDebug(2, Form("to: %s",to.Data()));
3079
3080         if (to.IsNull()) {
3081                 Log("SHUTTLE", Form("List of %d responsibles not set!", (Int_t) target));
3082                 return kFALSE;
3083         }
3084