Increasing number of expected fields in logbook_shuttle table.
[u/mrichter/AliRoot.git] / SHUTTLE / AliShuttle.cxx
1 /**************************************************************************
2  * Copyright(c) 1998-1999, ALICE Experiment at CERN, All rights reserved. *
3  *                                                                        *
4  * Author: The ALICE Off-line Project.                                    *
5  * Contributors are mentioned in the code where appropriate.              *
6  *                                                                        *
7  * Permission to use, copy, modify and distribute this software and its   *
8  * documentation strictly for non-commercial purposes is hereby granted   *
9  * without fee, provided that the above copyright notice appears in all   *
10  * copies and that both the copyright notice and this permission notice   *
11  * appear in the supporting documentation. The authors make no claims     *
12  * about the suitability of this software for any purpose. It is          *
13  * provided "as is" without express or implied warranty.                  *
14  **************************************************************************/
15
16 /* $Id$ */
17
18 //
19 // This class is the main manager for AliShuttle. 
20 // It organizes the data retrieval from DCS and call the 
21 // interface methods of AliPreprocessor.
22 // For every detector in AliShuttleConfgi (see AliShuttleConfig),
23 // data for its set of aliases is retrieved. If there is registered
24 // AliPreprocessor for this detector then it will be used
25 // accroding to the schema (see AliPreprocessor).
26 // If there isn't registered AliPreprocessor than the retrieved
27 // data is stored automatically to the undelying AliCDBStorage.
28 // For detSpec is used the alias name.
29 //
30
31 #include "AliShuttle.h"
32
33 #include "AliCDBManager.h"
34 #include "AliCDBStorage.h"
35 #include "AliCDBId.h"
36 #include "AliCDBRunRange.h"
37 #include "AliCDBPath.h"
38 #include "AliCDBEntry.h"
39 #include "AliShuttleConfig.h"
40 #include "DCSClient/AliDCSClient.h"
41 #include "AliLog.h"
42 #include "AliPreprocessor.h"
43 #include "AliShuttleStatus.h"
44 #include "AliShuttleLogbookEntry.h"
45
46 #include <TSystem.h>
47 #include <TObject.h>
48 #include <TString.h>
49 #include <TTimeStamp.h>
50 #include <TObjString.h>
51 #include <TSQLServer.h>
52 #include <TSQLResult.h>
53 #include <TSQLRow.h>
54 #include <TMutex.h>
55 #include <TSystemDirectory.h>
56 #include <TSystemFile.h>
57 #include <TFile.h>
58 #include <TGrid.h>
59 #include <TGridResult.h>
60 #include <TMap.h>
61
62 #include <TMonaLisaWriter.h>
63
64 #include <fstream>
65
66 #include <sys/types.h>
67 #include <sys/wait.h>
68
69 #include <signal.h>
70
71 ClassImp(AliShuttle)
72
73 //______________________________________________________________________________________________
74 AliShuttle::AliShuttle(const AliShuttleConfig* config,
75                 UInt_t timeout, Int_t retries):
76 fConfig(config),
77 fTimeout(timeout), fRetries(retries),
78 fPreprocessorMap(),
79 fLogbookEntry(0),
80 fCurrentDetector(),
81 fFirstProcessing(0),
82 fFXSError(-1),
83 fStatusEntry(0),
84 fMonitoringMutex(0),
85 fLastActionTime(0),
86 fLastAction(),
87 fMonaLisa(0),
88 fTestMode(kNone),
89 fReadTestMode(kFALSE),
90 fOutputRedirected(kFALSE)
91 {
92         //
93         // config: AliShuttleConfig used
94         // timeout: timeout used for AliDCSClient connection
95         // retries: the number of retries in case of connection error.
96         //
97
98         if (!fConfig->IsValid()) AliFatal("********** !!!!! Invalid configuration !!!!! **********");
99         for(int iSys=0;iSys<4;iSys++) {
100                 fServer[iSys]=0;
101                 if (iSys < 3)
102                         fFXSlist[iSys].SetOwner(kTRUE);
103         }
104         fPreprocessorMap.SetOwner(kTRUE);
105
106         for (UInt_t iDet=0; iDet<NDetectors(); iDet++)
107                 fFirstUnprocessed[iDet] = kFALSE;
108
109         fMonitoringMutex = new TMutex();
110 }
111
112 //______________________________________________________________________________________________
113 AliShuttle::~AliShuttle()
114 {
115         //
116         // destructor
117         //
118
119         fPreprocessorMap.DeleteAll();
120         for(int iSys=0;iSys<4;iSys++)
121                 if(fServer[iSys]) {
122                         fServer[iSys]->Close();
123                         delete fServer[iSys];
124                         fServer[iSys] = 0;
125                 }
126
127         if (fStatusEntry){
128                 delete fStatusEntry;
129                 fStatusEntry = 0;
130         }
131         
132         if (fMonitoringMutex) 
133         {
134                 delete fMonitoringMutex;
135                 fMonitoringMutex = 0;
136         }
137 }
138
139 //______________________________________________________________________________________________
140 void AliShuttle::RegisterPreprocessor(AliPreprocessor* preprocessor)
141 {
142         //
143         // Registers new AliPreprocessor.
144         // It uses GetName() for indentificator of the pre processor.
145         // The pre processor is registered it there isn't any other
146         // with the same identificator (GetName()).
147         //
148
149         const char* detName = preprocessor->GetName();
150         if(GetDetPos(detName) < 0)
151                 AliFatal(Form("********** !!!!! Invalid detector name: %s !!!!! **********", detName));
152
153         if (fPreprocessorMap.GetValue(detName)) {
154                 AliWarning(Form("AliPreprocessor %s is already registered!", detName));
155                 return;
156         }
157
158         fPreprocessorMap.Add(new TObjString(detName), preprocessor);
159 }
160 //______________________________________________________________________________________________
161 Bool_t AliShuttle::Store(const AliCDBPath& path, TObject* object,
162                 AliCDBMetaData* metaData, Int_t validityStart, Bool_t validityInfinite)
163 {
164         // Stores a CDB object in the storage for offline reconstruction. Objects that are not needed for
165         // offline reconstruction, but should be stored anyway (e.g. for debugging) should NOT be stored
166         // using this function. Use StoreReferenceData instead!
167         // It calls StoreLocally function which temporarily stores the data locally; when the preprocessor
168         // finishes the data are transferred to the main storage (Grid).
169
170         return StoreLocally(fgkLocalCDB, path, object, metaData, validityStart, validityInfinite);
171 }
172
173 //______________________________________________________________________________________________
174 Bool_t AliShuttle::StoreReferenceData(const AliCDBPath& path, TObject* object, AliCDBMetaData* metaData)
175 {
176         // Stores a CDB object in the storage for reference data. This objects will not be available during
177         // offline reconstrunction. Use this function for reference data only!
178         // It calls StoreLocally function which temporarily stores the data locally; when the preprocessor
179         // finishes the data are transferred to the main storage (Grid).
180
181         return StoreLocally(fgkLocalRefStorage, path, object, metaData);
182 }
183
184 //______________________________________________________________________________________________
185 Bool_t AliShuttle::StoreLocally(const TString& localUri,
186                         const AliCDBPath& path, TObject* object, AliCDBMetaData* metaData,
187                         Int_t validityStart, Bool_t validityInfinite)
188 {
189         // Store object temporarily in local storage. Parameters are passed by Store and StoreReferenceData functions.
190         // when the preprocessor finishes the data are transferred to the main storage (Grid).
191         // The parameters are:
192         //   1) Uri of the backup storage (Local)
193         //   2) the object's path.
194         //   3) the object to be stored
195         //   4) the metaData to be associated with the object
196         //   5) the validity start run number w.r.t. the current run,
197         //      if the data is valid only for this run leave the default 0
198         //   6) specifies if the calibration data is valid for infinity (this means until updated),
199         //      typical for calibration runs, the default is kFALSE
200         //
201         // returns 0 if fail, 1 otherwise
202
203
204         if (fTestMode & kErrorStorage)
205         {
206                 Log(fCurrentDetector, "StoreLocally - In TESTMODE - Simulating error while storing locally");
207                 return kFALSE;
208         }
209         
210         const char* cdbType = (localUri == fgkLocalCDB) ? "CDB" : "Reference";
211
212         Int_t firstRun = GetCurrentRun() - validityStart;
213         if(firstRun < 0) {
214                 AliWarning("First valid run happens to be less than 0! Setting it to 0.");
215                 firstRun=0;
216         }
217
218         Int_t lastRun = -1;
219         if(validityInfinite) {
220                 lastRun = AliCDBRunRange::Infinity();
221         } else {
222                 lastRun = GetCurrentRun();
223         }
224
225         // Version is set to current run, it will be used later to transfer data to Grid
226         AliCDBId id(path, firstRun, lastRun, GetCurrentRun(), -1);
227
228         if(! dynamic_cast<TObjString*> (metaData->GetProperty("RunUsed(TObjString)"))){
229                 TObjString runUsed = Form("%d", GetCurrentRun());
230                 metaData->SetProperty("RunUsed(TObjString)", runUsed.Clone());
231         }
232
233         Bool_t result = kFALSE;
234
235         if (!(AliCDBManager::Instance()->GetStorage(localUri))) {
236                 Log("SHUTTLE", Form("StoreLocally - Cannot activate local %s storage", cdbType));
237         } else {
238                 Int_t logLevel = AliLog::GetGlobalLogLevel();
239                 AliLog::SetGlobalLogLevel(AliLog::kError);
240                 result = AliCDBManager::Instance()->GetStorage(localUri)
241                                         ->Put(object, id, metaData);
242                 AliLog::SetGlobalLogLevel((AliLog::EType_t)logLevel);
243         }
244
245         if(!result) {
246
247                 Log(fCurrentDetector, Form("StoreLocally - Can't store object <%s>!", id.ToString().Data()));
248         }
249
250
251         return result;
252 }
253
254 //______________________________________________________________________________________________
255 Bool_t AliShuttle::StoreOCDB()
256 {
257         //
258         // Called when preprocessor ends successfully or when previous storage attempt failed (kStoreError status)
259         // Calls underlying StoreOCDB(const char*) function twice, for OCDB and Reference storage.
260         // Then calls StoreRefFilesToGrid to store reference files. 
261         //
262         
263         UpdateShuttleStatus(AliShuttleStatus::kStoreStarted);
264                         
265         if (fTestMode & kErrorGrid)
266         {
267                 Log("SHUTTLE", "StoreOCDB - In TESTMODE - Simulating error while storing in the Grid");
268                 Log(fCurrentDetector, "StoreOCDB - In TESTMODE - Simulating error while storing in the Grid");
269                 return kFALSE;
270         }
271         
272         Log("SHUTTLE","StoreOCDB - Storing OCDB data ...");
273         Int_t resultCDB = StoreOCDB(fgkMainCDB);
274
275         Log("SHUTTLE","StoreOCDB - Storing reference data ...");
276         Int_t resultRef = StoreOCDB(fgkMainRefStorage);
277         
278         Log("SHUTTLE","StoreOCDB - Storing reference files ...");
279         Bool_t resultRefFiles = CopyFilesToGrid("reference");
280         
281         Bool_t resultMetadata = kTRUE;
282         if(fCurrentDetector == "GRP") 
283         {
284                 Log("SHUTTLE","StoreOCDB - Storing Run Metadata file ...");
285                 resultMetadata = CopyFilesToGrid("metadata");
286         }
287         
288         Int_t storeResult = 0;
289         
290         if (resultCDB < 0 || resultRef < 0 || resultRefFiles == kFALSE || resultMetadata == kFALSE)
291                 storeResult = -1;
292         else if (resultCDB > 0 || resultRef > 0)
293                 storeResult = 1;
294                 
295         if (storeResult < 0)
296         {
297                 Log("SHUTTLE", 
298                         Form("\t\t\t****** run %d - %s: STORAGE ERROR ******",
299                                 GetCurrentRun(), fCurrentDetector.Data()));
300                 UpdateShuttleStatus(AliShuttleStatus::kStoreError);
301         } 
302         else if (storeResult > 0)
303         {
304                 Log("SHUTTLE", 
305                         Form("\t\t\t****** run %d - %s: STORAGE DELAYED ******",
306                                 GetCurrentRun(), fCurrentDetector.Data()));
307                 UpdateShuttleStatus(AliShuttleStatus::kStoreDelayed);
308         }
309         else if (storeResult == 0) 
310         {
311                 Log("SHUTTLE", 
312                         Form("\t\t\t****** run %d - %s: DONE ******",
313                                 GetCurrentRun(), fCurrentDetector.Data()));
314                 UpdateShuttleStatus(AliShuttleStatus::kDone);
315                 UpdateShuttleLogbook(fCurrentDetector, "DONE");
316         }
317
318         return (storeResult == 0);
319 }
320
321 //______________________________________________________________________________________________
322 Int_t AliShuttle::StoreOCDB(const TString& gridURI)
323 {
324         //
325         // Called by StoreOCDB(), performs actual storage to the main OCDB and reference storages (Grid)
326         //
327         // Return code:
328         //   -2 initialization error
329         //   -1 storage error
330         //   0  success
331         //   1  storage delayed (e.g. previous unprocessed runs)
332         //
333
334         TObjArray* gridIds=0;
335
336         Bool_t result = kTRUE;
337         Bool_t delayed = kFALSE;
338
339         const char* type = 0;
340         TString localURI;
341         if(gridURI == fgkMainCDB) {
342                 type = "OCDB";
343                 localURI = fgkLocalCDB;
344         } else if(gridURI == fgkMainRefStorage) {
345                 type = "reference";
346                 localURI = fgkLocalRefStorage;
347         } else {
348                 AliError(Form("Invalid storage URI: %s", gridURI.Data()));
349                 return -2;
350         }
351
352         AliCDBManager* man = AliCDBManager::Instance();
353
354         AliCDBStorage *gridSto = man->GetStorage(gridURI);
355         if(!gridSto) {
356                 Log("SHUTTLE",
357                         Form("StoreOCDB - cannot activate main %s storage", type));
358                 return -2;
359         }
360
361         gridIds = gridSto->GetQueryCDBList();
362
363         // get objects previously stored in local CDB
364         AliCDBStorage *localSto = man->GetStorage(localURI);
365         if(!localSto) {
366                 Log("SHUTTLE",
367                         Form("StoreOCDB - cannot activate local %s storage", type));
368                 return -2;
369         }
370         AliCDBPath aPath(GetOfflineDetName(fCurrentDetector.Data()),"*","*");
371         // Local objects were stored with current run as Grid version!
372         TList* localEntries = localSto->GetAll(aPath.GetPath(), GetCurrentRun(), GetCurrentRun());
373         localEntries->SetOwner(1);
374
375         // loop on local stored objects
376         TIter localIter(localEntries);
377         AliCDBEntry *aLocEntry = 0;
378         while((aLocEntry = dynamic_cast<AliCDBEntry*> (localIter.Next()))){
379                 aLocEntry->SetOwner(1);
380                 AliCDBId aLocId = aLocEntry->GetId();
381                 aLocEntry->SetVersion(-1);
382                 aLocEntry->SetSubVersion(-1);
383                                                 
384                 Log(fCurrentDetector.Data(), Form("Attempting to store %s", aLocId.ToString().Data()));
385
386                 // If local object is valid up to infinity we store it only if it is
387                 // the first unprocessed run!
388                 if (aLocId.GetLastRun() == AliCDBRunRange::Infinity() &&
389                         !fFirstUnprocessed[GetDetPos(fCurrentDetector)])
390                 {
391                         Log("SHUTTLE", Form("StoreOCDB - %s: object %s has validity infinite but "
392                                                 "there are previous unprocessed runs!",
393                                                 fCurrentDetector.Data(), aLocId.GetPath().Data()));
394                         Log(fCurrentDetector.Data(), Form("StoreOCDB - %s: object %s has validity infinite but "
395                                                 "there are previous unprocessed runs!",
396                                                 fCurrentDetector.Data(), aLocId.GetPath().Data()));
397                         delayed = kTRUE;
398                         continue;
399                 }
400
401                 // loop on Grid valid Id's
402                 Bool_t store = kTRUE;
403                 TIter gridIter(gridIds);
404                 AliCDBId* aGridId = 0;
405                 while ((aGridId = dynamic_cast<AliCDBId*> (gridIter.Next()))) {
406                         if (aGridId->GetPath() != aLocId.GetPath()) 
407                                 continue;
408                         // skip all objects valid up to infinity
409                         if (aGridId->GetLastRun() == AliCDBRunRange::Infinity()) 
410                                 continue;
411                         
412                         // if we get here, it means there's already some more recent object stored on Grid!
413                         Log(fCurrentDetector.Data(),
414                                 Form("StoreOCDB - A more recent object already exists in %s storage: <%s>",
415                                 type, aGridId->ToString().Data()));
416                         
417                         store = kFALSE;
418                         break;
419                 }
420
421                 Bool_t storeOk = kFALSE;
422                 if (store)
423                 {
424                         Log(fCurrentDetector.Data(), Form("Prechecks succeeded. Ready to store %s", aLocId.ToString().Data()));
425                         storeOk = gridSto->Put(aLocEntry);
426                         if (storeOk) {
427                                 Log("SHUTTLE",
428                                 Form("StoreOCDB - Object <%s> successfully put into %s storage",
429                                         aLocId.ToString().Data(), type));
430                                 Log(fCurrentDetector.Data(),
431                                         Form("StoreOCDB - Object <%s> successfully put into %s storage",
432                                         aLocId.ToString().Data(), type));
433                         } else  {
434                                 Log("SHUTTLE",
435                                         Form("StoreOCDB - Grid %s storage of object <%s> failed",
436                                         type, aLocId.ToString().Data()));
437                                 Log(fCurrentDetector.Data(),
438                                         Form("StoreOCDB - Grid %s storage of object <%s> failed",
439                                         type, aLocId.ToString().Data()));
440                                 result = kFALSE;
441                         }
442                 }
443                 
444                 if (!store || storeOk) {
445                         // removing local file...
446                         TString filename;
447                         localSto->IdToFilename(aLocId, filename);
448                         Log("SHUTTLE", Form("StoreOCDB - Removing local file %s", filename.Data()));
449                         RemoveFile(filename.Data());
450                 }
451         }
452         localEntries->Clear();
453
454         Int_t returnCode = 0;
455         
456         if (result == kFALSE)
457                 returnCode = -1;
458         else if (delayed != kFALSE)
459                 returnCode =  1;
460
461         Log("SHUTTLE", Form("StoreOCDB - Returning with %d (result = %d, delayed = %d)", returnCode, result, delayed));
462         Log(fCurrentDetector.Data(), Form("StoreOCDB - Returning with %d (result = %d, delayed = %d)", returnCode, result, delayed));
463         
464         return returnCode;
465 }
466
467 //______________________________________________________________________________________________
468 Bool_t AliShuttle::CleanReferenceStorage(const char* detector)
469 {
470         // clears the directory used to store reference files of a given subdetector
471   
472         AliCDBManager* man = AliCDBManager::Instance();
473         AliCDBStorage* sto = man->GetStorage(fgkLocalRefStorage);
474         TString localBaseFolder = sto->GetBaseFolder();
475
476         TString targetDir = GetRefFilePrefix(localBaseFolder.Data(), detector);
477         
478         Log("SHUTTLE", Form("CleanReferenceStorage - Cleaning %s", targetDir.Data()));
479
480         TString begin;
481         begin.Form("%d_", GetCurrentRun());
482         
483         TSystemDirectory* baseDir = new TSystemDirectory("/", targetDir);
484         if (!baseDir)
485                 return kTRUE;
486                 
487         TList* dirList = baseDir->GetListOfFiles();
488         delete baseDir;
489         
490         if (!dirList) return kTRUE;
491                         
492         if (dirList->GetEntries() < 3) 
493         {
494                 delete dirList;
495                 return kTRUE;
496         }
497                                 
498         Int_t nDirs = 0, nDel = 0;
499         TIter dirIter(dirList);
500         TSystemFile* entry = 0;
501
502         Bool_t success = kTRUE;
503         
504         while ((entry = dynamic_cast<TSystemFile*> (dirIter.Next())))
505         {                                       
506                 if (entry->IsDirectory())
507                         continue;
508                 
509                 TString fileName(entry->GetName());
510                 if (!fileName.BeginsWith(begin))
511                         continue;
512                         
513                 nDirs++;
514                                                 
515                 // delete file
516                 Int_t result = gSystem->Unlink(fileName.Data());
517                 
518                 if (result)
519                 {
520                         Log("SHUTTLE", Form("CleanReferenceStorage - Could not delete file %s!", fileName.Data()));
521                         success = kFALSE;
522                 } else {
523                         nDel++;
524                 }
525         }
526
527         if(nDirs > 0)
528                 Log("SHUTTLE", Form("CleanReferenceStorage - %d (over %d) reference files in folder %s were deleted.", 
529                         nDel, nDirs, targetDir.Data()));
530
531                 
532         delete dirList;
533         return success;
534
535
536
537
538
539
540   Int_t result = gSystem->GetPathInfo(targetDir, 0, (Long64_t*) 0, 0, 0);
541   if (result == 0)
542   {
543     // delete directory
544     result = gSystem->Exec(Form("rm -rf %s", targetDir.Data()));
545     if (result != 0)
546     {  
547       Log("SHUTTLE", Form("CleanReferenceStorage - Could not clean directory %s", targetDir.Data()));
548       return kFALSE;
549     }
550   }
551
552   result = gSystem->mkdir(targetDir, kTRUE);
553   if (result != 0)
554   {
555     Log("SHUTTLE", Form("CleanReferenceStorage - Error creating base directory %s", targetDir.Data()));
556     return kFALSE;
557   }
558         
559   return kTRUE;
560 }
561
562 //______________________________________________________________________________________________
563 Bool_t AliShuttle::StoreReferenceFile(const char* detector, const char* localFile, const char* gridFileName)
564 {
565         //
566         // Stores reference file directly (without opening it). This function stores the file locally.
567         //
568         // The file is stored under the following location: 
569         // <base folder of local reference storage>/<DET>/<RUN#>_<gridFileName>
570         // where <gridFileName> is the second parameter given to the function
571         // 
572         
573         if (fTestMode & kErrorStorage)
574         {
575                 Log(fCurrentDetector, "StoreReferenceFile - In TESTMODE - Simulating error while storing locally");
576                 return kFALSE;
577         }
578         
579         AliCDBManager* man = AliCDBManager::Instance();
580         AliCDBStorage* sto = man->GetStorage(fgkLocalRefStorage);
581         
582         TString localBaseFolder = sto->GetBaseFolder();
583         
584         TString target = GetRefFilePrefix(localBaseFolder.Data(), detector);    
585         target.Append(Form("/%d_%s", GetCurrentRun(), gridFileName));
586         
587         return CopyFileLocally(localFile, target);
588 }
589
590 //______________________________________________________________________________________________
591 Bool_t AliShuttle::StoreRunMetadataFile(const char* localFile, const char* gridFileName)
592 {
593         //
594         // Stores Run metadata file to the Grid, in the run folder
595         //
596         // Only GRP can call this function.
597         
598         if (fTestMode & kErrorStorage)
599         {
600                 Log(fCurrentDetector, "StoreRunMetaDataFile - In TESTMODE - Simulating error while storing locally");
601                 return kFALSE;
602         }
603         
604         AliCDBManager* man = AliCDBManager::Instance();
605         AliCDBStorage* sto = man->GetStorage(fgkLocalRefStorage);
606         
607         TString localBaseFolder = sto->GetBaseFolder();
608         
609         // Build Run level folder
610         // folder = /alice/data/year/lhcPeriod/runNb/raw
611         
612                 
613         TString lhcPeriod = GetLHCPeriod();     
614         if (lhcPeriod.Length() == 0) 
615         {
616                 Log("SHUTTLE","StoreRunMetaDataFile - LHCPeriod not found in logbook!");
617                 return 0;
618         }
619         
620         // TODO partitions with one detector only write data into LHCperiod_DET
621         TString partition = GetRunParameter("detector");
622         
623         if (partition.Length() > 0 && partition != "ALICE")
624         {
625                 lhcPeriod.Append(Form("_%s", partition.Data()));
626                 Log(fCurrentDetector, Form("Run data tags merged file will be written in %s", 
627                                 lhcPeriod.Data()));
628         }
629                 
630         TString target = Form("%s/GRP/RunMetadata%s%d/%s/%09d/raw/%s", 
631                                 localBaseFolder.Data(), fConfig->GetAlienPath(), GetCurrentYear(), 
632                                 lhcPeriod.Data(), GetCurrentRun(), gridFileName);
633                                         
634         return CopyFileLocally(localFile, target);
635 }
636
637 //______________________________________________________________________________________________
638 Bool_t AliShuttle::CopyFileLocally(const char* localFile, const TString& target)
639 {
640         //
641         // Stores file locally. Called by StoreReferenceFile and StoreRunMetadataFile
642         // Files are temporarily stored in the local reference storage. When the preprocessor 
643         // finishes, the Shuttle calls CopyFilesToGrid to transfer the files to AliEn 
644         // (in reference or run level folders)
645         //
646         
647         TString targetDir(target(0, target.Last('/')));
648         
649         //try to open base dir folder, if it does not exist
650         void* dir = gSystem->OpenDirectory(targetDir.Data());
651         if (dir == NULL) {
652                 if (gSystem->mkdir(targetDir.Data(), kTRUE)) {
653                         Log("SHUTTLE", Form("CopyFileLocally - Can't open directory <%s>", targetDir.Data()));
654                         return kFALSE;
655                 }
656
657         } else {
658                 gSystem->FreeDirectory(dir);
659         }
660         
661         Int_t result = 0;
662         
663         result = gSystem->GetPathInfo(localFile, 0, (Long64_t*) 0, 0, 0);
664         if (result)
665         {
666                 Log("SHUTTLE", Form("CopyFileLocally - %s does not exist", localFile));
667                 return kFALSE;
668         }
669
670         result = gSystem->GetPathInfo(target, 0, (Long64_t*) 0, 0, 0);
671         if (!result)
672         {
673                 Log("SHUTTLE", Form("CopyFileLocally - target file %s already exist, removing...", target.Data()));
674                 if (gSystem->Unlink(target.Data()))
675                 {
676                         Log("SHUTTLE", Form("CopyFileLocally - Could not remove existing target file %s!", target.Data()));
677                         return kFALSE;
678                 }
679         }       
680         
681         result = gSystem->CopyFile(localFile, target);
682
683         if (result == 0)
684         {
685                 Log("SHUTTLE", Form("CopyFileLocally - File %s stored locally to %s", localFile, target.Data()));
686                 return kTRUE;
687         }
688         else
689         {
690                 Log("SHUTTLE", Form("CopyFileLocally - Could not store file %s to %s! Error code = %d", 
691                                 localFile, target.Data(), result));
692                 return kFALSE;
693         }       
694
695
696
697 }
698
699 //______________________________________________________________________________________________
700 Bool_t AliShuttle::CopyFilesToGrid(const char* type)
701 {
702         //
703         // Transfers local files to the Grid. Local files can be reference files 
704         // or run metadata file (from GRP only).
705         //
706         // According to the type (ref, metadata) the files are stored under the following location: 
707         // ref --> <base folder of reference storage>/<DET>/<RUN#>_<gridFileName>
708         // metadata --> <run data folder>/<MetadataFileName>
709         //
710                 
711         AliCDBManager* man = AliCDBManager::Instance();
712         AliCDBStorage* sto = man->GetStorage(fgkLocalRefStorage);
713         if (!sto)
714                 return kFALSE;
715         TString localBaseFolder = sto->GetBaseFolder();
716         
717         TString dir;
718         TString alienDir;
719         TString begin;
720         
721         if (strcmp(type, "reference") == 0) 
722         {
723                 dir = GetRefFilePrefix(localBaseFolder.Data(), fCurrentDetector.Data());
724                 AliCDBStorage* gridSto = man->GetStorage(fgkMainRefStorage);
725                 if (!gridSto)
726                         return kFALSE;
727                 TString gridBaseFolder = gridSto->GetBaseFolder();
728                 alienDir = GetRefFilePrefix(gridBaseFolder.Data(), fCurrentDetector.Data());
729                 begin = Form("%d_", GetCurrentRun());
730         } 
731         else if (strcmp(type, "metadata") == 0)
732         {
733                         
734                 TString lhcPeriod = GetLHCPeriod();
735         
736                 if (lhcPeriod.Length() == 0) 
737                 {
738                         Log("SHUTTLE","CopyFilesToGrid - LHCPeriod not found in logbook!");
739                         return 0;
740                 }
741                 
742                 // TODO partitions with one detector only write data into LHCperiod_DET
743                 TString partition = GetRunParameter("detector");
744         
745                 if (partition.Length() > 0 && partition != "ALICE")
746                 {
747                         lhcPeriod.Append(Form("_%s", partition.Data()));
748                 }
749                 
750                 dir = Form("%s/GRP/RunMetadata%s%d/%s/%09d/raw", 
751                                 localBaseFolder.Data(), fConfig->GetAlienPath(), GetCurrentYear(), 
752                                 lhcPeriod.Data(), GetCurrentRun());
753                 alienDir = dir(dir.Index(fConfig->GetAlienPath()), dir.Length());
754                 
755                 begin = "";
756         }
757         else 
758         {
759                 Log("SHUTTLE", "CopyFilesToGrid - Unexpected: type label must be reference or metadata!");
760                 return kFALSE;
761         }
762                 
763         TSystemDirectory* baseDir = new TSystemDirectory("/", dir);
764         if (!baseDir)
765                 return kTRUE;
766                 
767         TList* dirList = baseDir->GetListOfFiles();
768         delete baseDir;
769         
770         if (!dirList) return kTRUE;
771                 
772         if (dirList->GetEntries() < 3) 
773         {
774                 delete dirList;
775                 return kTRUE;
776         }
777                         
778         if (!gGrid)
779         { 
780                 Log("SHUTTLE", "CopyFilesToGrid - Connection to Grid failed: Cannot continue!");
781                 delete dirList;
782                 return kFALSE;
783         }
784         
785         Int_t nDirs = 0, nTransfer = 0;
786         TIter dirIter(dirList);
787         TSystemFile* entry = 0;
788
789         Bool_t success = kTRUE;
790         Bool_t first = kTRUE;
791         
792         while ((entry = dynamic_cast<TSystemFile*> (dirIter.Next())))
793         {                       
794                 if (entry->IsDirectory())
795                         continue;
796                         
797                 TString fileName(entry->GetName());
798                 if (!fileName.BeginsWith(begin))
799                         continue;
800                         
801                 nDirs++;
802                         
803                 if (first)
804                 {
805                         first = kFALSE;
806                         // check that folder exists, otherwise create it
807                         TGridResult* result = gGrid->Ls(alienDir.Data(), "a");
808                         
809                         if (!result)
810                         {
811                                 delete dirList;
812                                 return kFALSE;
813                         }
814                         
815                         if (!result->GetFileName(1)) // TODO: It looks like element 0 is always 0!!
816                         {
817                                 // TODO It does not work currently! Bug in TAliEn::Mkdir
818                                 // TODO Manually fixed in local root v5-16-00
819                                 if (!gGrid->Mkdir(alienDir.Data(),"-p",0))
820                                 {
821                                         Log("SHUTTLE", Form("CopyFilesToGrid - Cannot create directory %s",
822                                                         alienDir.Data()));
823                                         delete dirList;
824                                         return kFALSE;
825                                 } else {
826                                         Log("SHUTTLE",Form("CopyFilesToGrid - Folder %s created", alienDir.Data()));
827                                 }
828                                 
829                         } else {
830                                         Log("SHUTTLE",Form("CopyFilesToGrid - Folder %s found", alienDir.Data()));
831                         }
832                 }
833                         
834                 TString fullLocalPath;
835                 fullLocalPath.Form("%s/%s", dir.Data(), fileName.Data());
836                 
837                 TString fullGridPath;
838                 fullGridPath.Form("alien://%s/%s", alienDir.Data(), fileName.Data());
839
840                 Bool_t result = TFile::Cp(fullLocalPath, fullGridPath);
841                 
842                 if (result)
843                 {
844                         Log("SHUTTLE", Form("CopyFilesToGrid - Copying local file %s to %s succeeded!", 
845                                                 fullLocalPath.Data(), fullGridPath.Data()));
846                         RemoveFile(fullLocalPath);
847                         nTransfer++;
848                 }
849                 else
850                 {
851                         Log("SHUTTLE", Form("CopyFilesToGrid - Copying local file %s to %s FAILED!", 
852                                                 fullLocalPath.Data(), fullGridPath.Data()));
853                         success = kFALSE;
854                 }
855         }
856
857         Log("SHUTTLE", Form("CopyFilesToGrid - %d (over %d) files in folder %s copied to Grid.", 
858                                                 nTransfer, nDirs, dir.Data()));
859
860                 
861         delete dirList;
862         return success;
863 }
864
865 //______________________________________________________________________________________________
866 const char* AliShuttle::GetRefFilePrefix(const char* base, const char* detector)
867 {
868         //
869         // Get folder name of reference files 
870         //
871
872         TString offDetStr(GetOfflineDetName(detector));
873         static TString dir;
874         if (offDetStr == "ITS" || offDetStr == "MUON" || offDetStr == "PHOS")
875         {
876                 dir.Form("%s/%s/%s", base, offDetStr.Data(), detector);
877         } else {
878                 dir.Form("%s/%s", base, offDetStr.Data());
879         }
880         
881         return dir.Data();
882 }
883
884 //______________________________________________________________________________________________
885 void AliShuttle::CleanLocalStorage(const TString& uri)
886 {
887         //
888         // Called in case the preprocessor is declared failed. Remove remaining objects from the local storages.
889         //
890
891         const char* type = 0;
892         if(uri == fgkLocalCDB) {
893                 type = "OCDB";
894         } else if(uri == fgkLocalRefStorage) {
895                 type = "Reference";
896         } else {
897                 AliError(Form("Invalid storage URI: %s", uri.Data()));
898                 return;
899         }
900
901         AliCDBManager* man = AliCDBManager::Instance();
902
903         // open local storage
904         AliCDBStorage *localSto = man->GetStorage(uri);
905         if(!localSto) {
906                 Log("SHUTTLE",
907                         Form("CleanLocalStorage - cannot activate local %s storage", type));
908                 return;
909         }
910
911         TString filename(Form("%s/%s/*/Run*_v%d_s*.root",
912                 localSto->GetBaseFolder().Data(), GetOfflineDetName(fCurrentDetector.Data()), GetCurrentRun()));
913
914         AliDebug(2, Form("filename = %s", filename.Data()));
915
916         Log("SHUTTLE", Form("Removing remaining local files for run %d and detector %s ...",
917                 GetCurrentRun(), fCurrentDetector.Data()));
918
919         RemoveFile(filename.Data());
920
921 }
922
923 //______________________________________________________________________________________________
924 void AliShuttle::RemoveFile(const char* filename)
925 {
926         //
927         // removes local file
928         //
929
930         TString command(Form("rm -f %s", filename));
931
932         Int_t result = gSystem->Exec(command.Data());
933         if(result != 0)
934         {
935                 Log("SHUTTLE", Form("RemoveFile - %s: Cannot remove file %s!",
936                         fCurrentDetector.Data(), filename));
937         }
938 }
939
940 //______________________________________________________________________________________________
941 AliShuttleStatus* AliShuttle::ReadShuttleStatus()
942 {
943         //
944         // Reads the AliShuttleStatus from the CDB
945         //
946
947         if (fStatusEntry){
948                 delete fStatusEntry;
949                 fStatusEntry = 0;
950         }
951
952         Int_t path1 = GetCurrentRun()/10000;
953         fStatusEntry = AliCDBManager::Instance()->GetStorage(GetLocalCDB())
954                 ->Get(Form("/SHUTTLE/%s/%d", fCurrentDetector.Data(), path1), GetCurrentRun());
955
956         if (!fStatusEntry) return 0;
957         fStatusEntry->SetOwner(1);
958
959         AliShuttleStatus* status = dynamic_cast<AliShuttleStatus*> (fStatusEntry->GetObject());
960         if (!status) {
961                 AliError("Invalid object stored to CDB!");
962                 return 0;
963         }
964
965         return status;
966 }
967
968 //______________________________________________________________________________________________
969 Bool_t AliShuttle::WriteShuttleStatus(AliShuttleStatus* status)
970 {
971         //
972         // writes the status for one subdetector
973         //
974
975         if (fStatusEntry){
976                 delete fStatusEntry;
977                 fStatusEntry = 0;
978         }
979
980         Int_t run = GetCurrentRun();
981         Int_t path1 = run/10000;
982         TString path1_string = Form("%d",path1);
983
984         AliCDBId id(AliCDBPath("SHUTTLE", fCurrentDetector, path1_string), run, run);
985
986         fStatusEntry = new AliCDBEntry(status, id, new AliCDBMetaData);
987         fStatusEntry->SetOwner(1);
988
989         Int_t logLevel = AliLog::GetGlobalLogLevel();
990         AliLog::SetGlobalLogLevel(AliLog::kError);
991
992         UInt_t result = AliCDBManager::Instance()->GetStorage(fgkLocalCDB)->Put(fStatusEntry);
993
994         AliLog::SetGlobalLogLevel((AliLog::EType_t)logLevel);
995
996         if (!result) {
997                 Log("SHUTTLE", Form("WriteShuttleStatus - Failed for %s, run %d",
998                                                 fCurrentDetector.Data(), run));
999                 return kFALSE;
1000         }
1001         
1002         SendMLDetInfo();
1003
1004         return kTRUE;
1005 }
1006
1007 //______________________________________________________________________________________________
1008 void AliShuttle::UpdateShuttleStatus(AliShuttleStatus::Status newStatus, Bool_t increaseCount)
1009 {
1010         //
1011         // changes the AliShuttleStatus for the given detector and run to the given status
1012         //
1013
1014         if (!fStatusEntry){
1015                 AliError("UNEXPECTED: fStatusEntry empty");
1016                 return;
1017         }
1018
1019         AliShuttleStatus* status = dynamic_cast<AliShuttleStatus*> (fStatusEntry->GetObject());
1020
1021         if (!status){
1022                 Log("SHUTTLE", "UpdateShuttleStatus - UNEXPECTED: status could not be read from current CDB entry");
1023                 return;
1024         }
1025
1026         TString actionStr = Form("UpdateShuttleStatus - %s: Changing state from %s to %s",
1027                                 fCurrentDetector.Data(),
1028                                 status->GetStatusName(),
1029                                 status->GetStatusName(newStatus));
1030         Log("SHUTTLE", actionStr);
1031         SetLastAction(actionStr);
1032
1033         status->SetStatus(newStatus);
1034         if (increaseCount) status->IncreaseCount();
1035
1036         Int_t logLevel = AliLog::GetGlobalLogLevel();
1037         AliLog::SetGlobalLogLevel(AliLog::kError);
1038
1039         AliCDBManager::Instance()->GetStorage(fgkLocalCDB)->Put(fStatusEntry);
1040
1041         AliLog::SetGlobalLogLevel((AliLog::EType_t)logLevel);
1042
1043         SendMLDetInfo();
1044 }
1045
1046 //______________________________________________________________________________________________
1047 void AliShuttle::SendMLDetInfo()
1048 {
1049         //
1050         // sends ML information about the current status of the current detector being processed
1051         //
1052         
1053         AliShuttleStatus* status = dynamic_cast<AliShuttleStatus*> (fStatusEntry->GetObject());
1054         
1055         if (!status){
1056                 Log("SHUTTLE", "SendMLDetInfo - UNEXPECTED: status could not be read from current CDB entry");
1057                 return;
1058         }
1059         
1060         TMonaLisaText  mlStatus(Form("%s_status", fCurrentDetector.Data()), status->GetStatusName());
1061         TMonaLisaValue mlRetryCount(Form("%s_count", fCurrentDetector.Data()), status->GetCount());
1062
1063         TList mlList;
1064         mlList.Add(&mlStatus);
1065         mlList.Add(&mlRetryCount);
1066
1067         TString mlID;
1068         mlID.Form("%d", GetCurrentRun());
1069         fMonaLisa->SendParameters(&mlList, mlID);
1070 }
1071
1072 //______________________________________________________________________________________________
1073 Bool_t AliShuttle::ContinueProcessing()
1074 {
1075         // this function reads the AliShuttleStatus information from CDB and
1076         // checks if the processing should be continued
1077         // if yes it returns kTRUE and updates the AliShuttleStatus with nextStatus
1078
1079         if (!fConfig->HostProcessDetector(fCurrentDetector))
1080                 return kFALSE;
1081
1082         AliPreprocessor* aPreprocessor =
1083                 dynamic_cast<AliPreprocessor*> (fPreprocessorMap.GetValue(fCurrentDetector));
1084         if (!aPreprocessor)
1085         {
1086                 Log("SHUTTLE", Form("ContinueProcessing - %s: no preprocessor registered", fCurrentDetector.Data()));
1087                 return kFALSE;
1088         }
1089
1090         AliShuttleLogbookEntry::Status entryStatus =
1091                 fLogbookEntry->GetDetectorStatus(fCurrentDetector);
1092
1093         if (entryStatus != AliShuttleLogbookEntry::kUnprocessed) {
1094                 Log("SHUTTLE", Form("ContinueProcessing - %s is %s",
1095                                 fCurrentDetector.Data(),
1096                                 fLogbookEntry->GetDetectorStatusName(entryStatus)));
1097                 return kFALSE;
1098         }
1099
1100         // if we get here, according to Shuttle logbook subdetector is in UNPROCESSED state
1101
1102         // check if current run is first unprocessed run for current detector
1103         if (fConfig->StrictRunOrder(fCurrentDetector) &&
1104                 !fFirstUnprocessed[GetDetPos(fCurrentDetector)])
1105         {
1106                 if (fTestMode == kNone)
1107                 {
1108                         Log("SHUTTLE", Form("ContinueProcessing - %s requires strict run ordering"
1109                                         " but this is not the first unprocessed run!"));
1110                         return kFALSE;
1111                 }
1112                 else
1113                 {
1114                         Log("SHUTTLE", Form("ContinueProcessing - In TESTMODE - "
1115                                         "Although %s requires strict run ordering "
1116                                         "and this is not the first unprocessed run, "
1117                                         "the SHUTTLE continues"));
1118                 }
1119         }
1120
1121         // Is the subdetector processed first time for this run?
1122         fFirstProcessing = kFALSE;
1123
1124         AliShuttleStatus* status = ReadShuttleStatus();
1125         if (!status) {
1126                 // first time
1127                 Log("SHUTTLE", Form("ContinueProcessing - %s: Processing first time",
1128                                 fCurrentDetector.Data()));
1129                 status = new AliShuttleStatus(AliShuttleStatus::kStarted);
1130                 fFirstProcessing = kTRUE;
1131                 return WriteShuttleStatus(status);
1132         }
1133
1134         // The following case shouldn't happen if Shuttle Logbook was correctly updated.
1135         // If it happens it may mean Logbook updating failed... let's do it now!
1136         if (status->GetStatus() == AliShuttleStatus::kDone ||
1137             status->GetStatus() == AliShuttleStatus::kFailed ||
1138             status->GetStatus() == AliShuttleStatus::kSkipped) {
1139                 Log("SHUTTLE", Form("ContinueProcessing - %s is already %s. Updating Shuttle Logbook",
1140                                         fCurrentDetector.Data(),
1141                                         status->GetStatusName(status->GetStatus())));
1142                 
1143                 if (status->GetStatus() == AliShuttleStatus::kSkipped)
1144                 {
1145                         UpdateShuttleLogbook(fCurrentDetector.Data(), "DONE");
1146                 }
1147                 else
1148                         UpdateShuttleLogbook(fCurrentDetector.Data(), status->GetStatusName(status->GetStatus()));
1149                         
1150                 return kFALSE;
1151         }
1152
1153         if (status->GetStatus() == AliShuttleStatus::kStoreStarted || status->GetStatus() == AliShuttleStatus::kStoreDelayed ||status->GetStatus() == AliShuttleStatus::kStoreError) {
1154                 Log("SHUTTLE",
1155                         Form("ContinueProcessing - %s: Grid storage of one or more "
1156                                 "objects failed. Trying again now",
1157                                 fCurrentDetector.Data()));
1158                 StoreOCDB();
1159                 return kFALSE;
1160         }
1161
1162         // if we get here, there is a restart
1163         Bool_t cont = kFALSE;
1164
1165         // abort conditions
1166         if (status->GetCount() >= fConfig->GetMaxRetries()) {
1167                 Log("SHUTTLE", Form("ContinueProcessing - %s failed %d times in status %s - "
1168                                 "Updating Shuttle Logbook", fCurrentDetector.Data(),
1169                                 status->GetCount(), status->GetStatusName()));
1170                 UpdateShuttleLogbook(fCurrentDetector.Data(), "FAILED");
1171                 UpdateShuttleStatus(AliShuttleStatus::kFailed);
1172
1173                 // there may still be objects in local OCDB and reference storage
1174                 // and FXS databases may be not updated: do it now!
1175                 
1176                 // TODO Currently disabled, we want to keep files in case of failure!
1177                 // CleanLocalStorage(fgkLocalCDB);
1178                 // CleanLocalStorage(fgkLocalRefStorage);
1179                 // UpdateTableFailCase();
1180                 
1181                 // Send mail to detector expert!
1182                 Log("SHUTTLE", Form("ContinueProcessing - Sending mail to %s expert...", 
1183                                     fCurrentDetector.Data()));
1184                 // det experts in to
1185                 TString to="";
1186                 TIter *iterExperts = 0;
1187                 iterExperts = new TIter(fConfig->GetResponsibles(fCurrentDetector));
1188                 TObjString *anExpert=0;
1189                 while ((anExpert = (TObjString*) iterExperts->Next()))
1190                         {
1191                                 to += Form("%s, \n", anExpert->GetName());
1192                         }
1193                 delete iterExperts;
1194                 
1195                 if (to.Length() > 0)
1196                         to.Remove(to.Length()-3);
1197                 AliDebug(2, Form("to: %s",to.Data()));
1198
1199                 if (to.IsNull()) {
1200                         Log("SHUTTLE", Form("List of %s responsibles not set!", fCurrentDetector.Data()));
1201                         return kFALSE;
1202                 }
1203
1204                 Log(fCurrentDetector.Data(), Form("ContinueProcessing - Sending mail to %s expert(s):", 
1205                                     fCurrentDetector.Data()));
1206                 Log(fCurrentDetector.Data(), Form("\n%s", to.Data()));
1207                 if (!SendMail(kPPEMail))
1208                         Log("SHUTTLE", Form("ContinueProcessing - Could not send mail to %s expert",
1209                                             fCurrentDetector.Data()));
1210
1211         } else {
1212                 Log("SHUTTLE", Form("ContinueProcessing - %s: restarting. "
1213                                 "Aborted before with %s. Retry number %d.", fCurrentDetector.Data(),
1214                                 status->GetStatusName(), status->GetCount()));
1215                 Bool_t increaseCount = kTRUE;
1216                 if (status->GetStatus() == AliShuttleStatus::kDCSError || 
1217                     status->GetStatus() == AliShuttleStatus::kDCSStarted ||
1218                     status->GetStatus() == AliShuttleStatus::kFXSError)
1219                                 increaseCount = kFALSE;
1220                                 
1221                 UpdateShuttleStatus(AliShuttleStatus::kStarted, increaseCount);
1222                 cont = kTRUE;
1223         }
1224
1225         return cont;
1226 }
1227
1228 //______________________________________________________________________________________________
1229 void AliShuttle::SendMLRunInfo(const char* status)
1230 {
1231         // 
1232         // Send information about this run to ML
1233         
1234         TMonaLisaText  mlStatus("SHUTTLE_status", status);
1235         TString runType(fLogbookEntry->GetRunType());
1236         if (strlen(fLogbookEntry->GetRunParameter("log")) > 0){
1237
1238                 runType += "(";
1239                 runType += fLogbookEntry->GetRunParameter("log");
1240                 runType += ")";
1241         }
1242         if (fLogbookEntry->GetDATestMode()){
1243                 runType += " (DATest)";
1244         }
1245         TMonaLisaText  mlRunType("SHUTTLE_runtype", runType);
1246
1247         TList mlList;
1248         mlList.Add(&mlStatus);
1249         mlList.Add(&mlRunType);
1250
1251         TString mlID;
1252         mlID.Form("%d", GetCurrentRun());
1253         fMonaLisa->SendParameters(&mlList, mlID);       
1254 }
1255
1256 //______________________________________________________________________________________________
1257 Int_t AliShuttle::GetMem(Int_t pid)
1258 {
1259         // invokes ps to get the memory consumption of the process <pid>
1260         // returns -1 in case of error
1261         
1262         TString checkStr;
1263         checkStr.Form("ps -o vsize --pid %d | tail -n 1", pid);
1264         FILE* pipe = gSystem->OpenPipe(checkStr, "r");
1265         if (!pipe)
1266         {
1267                 Log("SHUTTLE", Form("Process - Error: "
1268                         "Could not open pipe to %s", checkStr.Data()));
1269                 return -1;
1270         }
1271                 
1272         char buffer[100];
1273         if (!fgets(buffer, 100, pipe))
1274         {
1275                 Log("SHUTTLE", "Process - Error: ps did not return anything");
1276                 gSystem->ClosePipe(pipe);
1277                 return -1;
1278         }
1279         gSystem->ClosePipe(pipe);
1280         
1281         //Log("SHUTTLE", Form("ps returned %s", buffer));
1282         
1283         Int_t mem = 0;
1284         if ((sscanf(buffer, "%d\n", &mem) != 1) || !mem)
1285         {
1286                 Log("SHUTTLE", "Process - Error: Could not parse output of ps");
1287                 return -1;
1288         }
1289         
1290         return mem;
1291 }
1292
1293 //______________________________________________________________________________________________
1294 Bool_t AliShuttle::Process(AliShuttleLogbookEntry* entry)
1295 {
1296         //
1297         // Makes data retrieval for all detectors in the configuration.
1298         // entry: Shuttle logbook entry, contains run paramenters and status of detectors
1299         // (Unprocessed, Inactive, Failed or Done).
1300         // Returns kFALSE in case of error occured and kTRUE otherwise
1301         //
1302
1303         if (!entry) return kFALSE;
1304
1305         fLogbookEntry = entry;
1306
1307         Log("SHUTTLE", Form("\t\t\t^*^*^*^*^*^*^*^*^*^*^*^* run %d: START ^*^*^*^*^*^*^*^*^*^*^*^*",
1308                                         GetCurrentRun()));
1309
1310         CountOpenRuns();
1311         
1312         // Send the information to ML
1313         SendMLRunInfo("Processing");
1314
1315         if (fLogbookEntry->IsDone())
1316         {
1317                 Log("SHUTTLE","Process - Shuttle is already DONE. Updating logbook");
1318                 UpdateShuttleLogbook("shuttle_done");
1319                 fLogbookEntry = 0;
1320                 return kTRUE;
1321         }
1322
1323         // read test mode if flag is set
1324         if (fReadTestMode)
1325         {
1326                 fTestMode = kNone;
1327                 TString logEntry(entry->GetRunParameter("log"));
1328                 //printf("log entry = %s\n", logEntry.Data());
1329                 TString searchStr("Testmode: ");
1330                 Int_t pos = logEntry.Index(searchStr.Data());
1331                 //printf("%d\n", pos);
1332                 if (pos >= 0)
1333                 {
1334                         TSubString subStr = logEntry(pos + searchStr.Length(), logEntry.Length());
1335                         //printf("%s\n", subStr.String().Data());
1336                         TString newStr(subStr.Data());
1337                         TObjArray* token = newStr.Tokenize(' ');
1338                         if (token)
1339                         {
1340                                 //token->Print();
1341                                 TObjString* tmpStr = dynamic_cast<TObjString*> (token->First());
1342                                 if (tmpStr)
1343                                 {
1344                                         Int_t testMode = tmpStr->String().Atoi();
1345                                         if (testMode > 0)
1346                                         {
1347                                                 Log("SHUTTLE", Form("Process - Enabling test mode %d", testMode));
1348                                                 SetTestMode((TestMode) testMode);
1349                                         }
1350                                 }
1351                                 delete token;          
1352                         }
1353                 }
1354         }
1355                 
1356         fLogbookEntry->Print("all");
1357
1358         // Initialization
1359         Bool_t hasError = kFALSE;
1360
1361         // Set the CDB and Reference folders according to the year
1362
1363         // build cdb paths (repeat each time, run might be a DATest run)
1364         if (!fLogbookEntry->GetDATestMode()){
1365                 fgkMainCDB.Form("alien://folder=%s%d/OCDB?user=alidaq?cacheFold=/tmp/OCDBCache", 
1366                                 fConfig->GetAlienPath(), GetCurrentYear());
1367                 
1368                 fgkMainRefStorage.Form("alien://folder=%s%d/Reference?user=alidaq?cacheFold=/tmp/OCDBCache", 
1369                                        fConfig->GetAlienPath(), GetCurrentYear());
1370         }
1371         else {
1372                 fgkMainCDB.Form("alien://folder=%s%d/DATest/OCDB?user=alidaq?cacheFold=/tmp/OCDBCache",
1373                                 fConfig->GetAlienPath(), GetCurrentYear());
1374                 
1375                 fgkMainRefStorage.Form("alien://folder=%s%d/DATest/Reference?user=alidaq?cacheFold=/tmp/OCDBCache",
1376                                        fConfig->GetAlienPath(), GetCurrentYear());
1377         }
1378
1379         AliDebug(2,Form("Main CDB storage = %s",fgkMainCDB.Data()));
1380         AliDebug(2,Form("Main Reference storage = %s",fgkMainRefStorage.Data()));
1381
1382         // Loop on detectors in the configuration
1383         TIter iter(fConfig->GetDetectors());
1384         TObjString* aDetector = 0;
1385
1386         Bool_t first = kTRUE;
1387
1388         while ((aDetector = (TObjString*) iter.Next()))
1389         {
1390                 fCurrentDetector = aDetector->String();
1391
1392                 if (ContinueProcessing() == kFALSE) 
1393                         continue;
1394                 
1395                 if (first)
1396                 {
1397                   // only read QueryCDB when needed and only once
1398                   AliCDBStorage *mainCDBSto = AliCDBManager::Instance()->GetStorage(fgkMainCDB);
1399                   if(mainCDBSto) mainCDBSto->QueryCDB(GetCurrentRun());
1400                   AliCDBStorage *mainRefSto = AliCDBManager::Instance()->GetStorage(fgkMainRefStorage);
1401                   if(mainRefSto) mainRefSto->QueryCDB(GetCurrentRun());
1402                   first = kFALSE;
1403                 }
1404
1405                 Log("SHUTTLE", Form("\t\t\t****** run %d - %s: START  ******",
1406                                                 GetCurrentRun(), aDetector->GetName()));
1407
1408                 for(Int_t iSys=0;iSys<3;iSys++) fFXSCalled[iSys]=kFALSE;
1409                 
1410                 Int_t initialMem = GetMem(getpid());
1411                 Log("SHUTTLE", Form("Memory consumption before forking is %d", initialMem));
1412
1413                 Log(fCurrentDetector.Data(), "Process - Starting processing");
1414
1415                 Int_t pid = fork();
1416
1417                 if (pid < 0)
1418                 {
1419                         Log("SHUTTLE", "Process - ERROR: Forking failed");
1420                 }
1421                 else if (pid > 0)
1422                 {
1423                         // parent
1424                         Log("SHUTTLE", Form("Process - In parent process of %d - %s: Starting monitoring",
1425                                                         GetCurrentRun(), aDetector->GetName()));
1426
1427                         Long_t begin = time(0);
1428
1429                         int status; // to be used with waitpid, on purpose an int (not Int_t)!
1430                         while (waitpid(pid, &status, WNOHANG) == 0)
1431                         {
1432                                 Long_t expiredTime = time(0) - begin;
1433
1434                                 if (expiredTime > fConfig->GetPPTimeOut())
1435                                 {
1436                                         TString logMsg;
1437                                         AliShuttleStatus *currentStatus = ReadShuttleStatus();
1438                                         AliShuttleStatus::Status newStatus = AliShuttleStatus::kInvalid;
1439                                         
1440                                         if (currentStatus->GetStatus() <= AliShuttleStatus::kPPDone)
1441                                         {
1442                                                 // in case pp not yet done set status to kPPTimeOut
1443                                         
1444                                                 logMsg.Form("Process - Process of %s timed out. Run time: %d seconds. Killing...",
1445                                                                 fCurrentDetector.Data(), expiredTime);
1446                                                 newStatus = AliShuttleStatus::kPPTimeOut;
1447                                         }
1448                                         else if (currentStatus->GetStatus() == AliShuttleStatus::kStoreStarted)
1449                                         {
1450                                                 // in case the pp goes in TimeOut while storing the objects in the OCDB
1451                                                 // set status to kStoreError
1452                                                 
1453                                                 logMsg.Form("Process - Process of %s timed out while storing the OCDB object. Run time: %d seconds. Killing... and setting status to StoreError.",
1454                                                                 fCurrentDetector.Data(), expiredTime);
1455                                                 newStatus = AliShuttleStatus::kStoreError;
1456                                         }
1457                                         else 
1458                                         {
1459                                                 // in other cases don't change the status
1460                                                 
1461                                                 logMsg.Form("Process - Process of %s timed out in status = %s. Run time: %d seconds. Killing... without changing the status",
1462                                                                 fCurrentDetector.Data(), currentStatus->GetStatusName(), expiredTime);
1463                                         }
1464                                 
1465                                         Log("SHUTTLE", logMsg);
1466                                         Log(fCurrentDetector, logMsg);
1467
1468                                         kill(pid, 9);
1469
1470                                         if (newStatus != AliShuttleStatus::kInvalid)
1471                                                 UpdateShuttleStatus(newStatus);
1472                                         hasError = kTRUE;
1473
1474                                         gSystem->Sleep(1000);
1475                                 }
1476                                 else
1477                                 {
1478                                         gSystem->Sleep(1000);
1479                                         
1480                                         Int_t mem = GetMem(pid);
1481
1482                                         if (mem < 0)
1483                                                 continue;
1484                                                 
1485                                         mem -= initialMem;
1486                                         if (mem < 0)
1487                                                 mem = 0;
1488                                         
1489                                         if (expiredTime % 60 == 0)
1490                                         {
1491                                                 Log("SHUTTLE", Form("Process - %s: Checking process. "
1492                                                         "Run time: %d seconds - Memory consumption: %d KB",
1493                                                         fCurrentDetector.Data(), expiredTime, mem));
1494                                                 SendAlive();
1495                                         }
1496                                         
1497                                         if (mem > fConfig->GetPPMaxMem())
1498                                         {
1499                                                 TString tmp;
1500                                                 tmp.Form("Process - Process exceeds maximum allowed memory "
1501                                                         "(%d KB > %d KB). Killing...",
1502                                                         mem, fConfig->GetPPMaxMem());
1503                                                 Log("SHUTTLE", tmp);
1504                                                 Log(fCurrentDetector, tmp);
1505         
1506                                                 kill(pid, 9);
1507         
1508                                                 UpdateShuttleStatus(AliShuttleStatus::kPPOutOfMemory);
1509                                                 hasError = kTRUE;
1510         
1511                                                 gSystem->Sleep(1000);
1512                                         }
1513                                 }
1514                         }
1515
1516                         Log("SHUTTLE", Form("Process - In parent process of %d - %s: Client has terminated.",
1517                                                                 GetCurrentRun(), aDetector->GetName()));
1518
1519                         if (WIFEXITED(status))
1520                         {
1521                                 Int_t returnCode = WEXITSTATUS(status);
1522
1523                                 Log("SHUTTLE", Form("Process - %s: the return code is %d", fCurrentDetector.Data(),
1524                                                                                 returnCode));
1525
1526                                 if (returnCode == 0) hasError = kTRUE;
1527                         }
1528                 }
1529                 else if (pid == 0)
1530                 {
1531                         // child
1532                         Log("SHUTTLE", Form("Process - In child process of %d - %s", GetCurrentRun(),
1533                                 aDetector->GetName()));
1534
1535                         Log("SHUTTLE", Form("Process - Redirecting output to %s log",fCurrentDetector.Data()));
1536
1537                         if ((freopen(GetLogFileName(fCurrentDetector), "a", stdout)) == 0)
1538                         {
1539                                 Log("SHUTTLE", "Process - Could not freopen stdout");
1540                         }
1541                         else
1542                         {
1543                                 fOutputRedirected = kTRUE;
1544                                 if ((dup2(fileno(stdout), fileno(stderr))) < 0)
1545                                         Log("SHUTTLE", "Process - Could not redirect stderr");
1546                                 
1547                         }
1548                         
1549                         TString wd = gSystem->WorkingDirectory();
1550                         Int_t dir_lev1 = GetCurrentRun()/10000;
1551                         TString tmpDir = Form("%s/%d/%d/%s_process", GetShuttleTempDir(), 
1552                                 dir_lev1, GetCurrentRun(), fCurrentDetector.Data());
1553                         
1554                         Int_t result = gSystem->GetPathInfo(tmpDir.Data(), 0, (Long64_t*) 0, 0, 0);
1555                         if (!result) // temp dir already exists!
1556                         {
1557                                 Log(fCurrentDetector.Data(), 
1558                                         Form("Process - %s dir already exists! Removing...", tmpDir.Data()));
1559                                 gSystem->Exec(Form("rm -rf %s",tmpDir.Data()));         
1560                         } 
1561                         
1562                         if (gSystem->mkdir(tmpDir.Data(), 1))
1563                         {
1564                                 Log(fCurrentDetector.Data(), "Process - could not make temp directory!!");
1565                                 gSystem->Exit(1);
1566                         }
1567                         
1568                         if (!gSystem->ChangeDirectory(tmpDir.Data())) 
1569                         {
1570                                 Log(fCurrentDetector.Data(), "Process - could not change directory!!");
1571                                 gSystem->Exit(1);                       
1572                         }
1573                         
1574                         Int_t success = ProcessCurrentDetector();
1575
1576                         gSystem->ChangeDirectory(wd.Data());
1577                                                 
1578                         if (success == 1) // Preprocessor finished successfully!
1579                         { 
1580                                 // remove temporary folder or DCS map
1581                                 if (!fConfig->KeepTempFolder())
1582                                 {
1583                                         gSystem->Exec(Form("rm -rf %s",tmpDir.Data()));
1584                                 } else if (!fConfig->KeepDCSMap())
1585                                 {
1586                                         gSystem->Exec(Form("rm -f %s/DCSMap.root",tmpDir.Data()));
1587                                 }
1588                                 
1589                                 // Update time_processed field in FXS DB
1590                                 if (UpdateTable() == kFALSE)
1591                                         Log("SHUTTLE", Form("Process - %s: Could not update FXS databases!", 
1592                                                         fCurrentDetector.Data()));
1593
1594                                 // Transfer the data from local storage to main storage (Grid)
1595                                 if (StoreOCDB() == kFALSE)
1596                                         success = kFALSE;
1597                         } 
1598                         else if (success == 0)
1599                         {
1600                                 Log("SHUTTLE", 
1601                                         Form("\t\t\t****** run %d - %s: PP ERROR ******",
1602                                                 GetCurrentRun(), aDetector->GetName()));
1603                         }
1604
1605                         for (UInt_t iSys=0; iSys<3; iSys++)
1606                         {
1607                                 if (fFXSCalled[iSys]) fFXSlist[iSys].Clear();
1608                         }
1609
1610                         Log("SHUTTLE", Form("Process - Client process of %d - %s is exiting now with %d.",
1611                                                         GetCurrentRun(), aDetector->GetName(), success));
1612
1613                         // the client exits here
1614                         gSystem->Exit(success);
1615
1616                         AliError("We should never get here!!!");
1617                 }
1618         }
1619
1620         Log("SHUTTLE", Form("\t\t\t^*^*^*^*^*^*^*^*^*^*^*^* run %d: FINISH ^*^*^*^*^*^*^*^*^*^*^*^*",
1621                                                         GetCurrentRun()));
1622
1623         //check if shuttle is done for this run, if so update logbook
1624         TObjArray checkEntryArray;
1625         checkEntryArray.SetOwner(1);
1626         TString whereClause = Form("where run=%d", GetCurrentRun());
1627         if (!QueryShuttleLogbook(whereClause.Data(), checkEntryArray) || 
1628                         checkEntryArray.GetEntries() == 0) {
1629                 Log("SHUTTLE", Form("Process - Warning: Cannot check status of run %d on Shuttle logbook!",
1630                                                 GetCurrentRun()));
1631                 return hasError == kFALSE;
1632         }
1633
1634         AliShuttleLogbookEntry* checkEntry = dynamic_cast<AliShuttleLogbookEntry*>
1635                                                 (checkEntryArray.At(0));
1636
1637         if (checkEntry)
1638         {
1639                 if (checkEntry->IsDone())
1640                 {
1641                         Log("SHUTTLE","Process - Shuttle is DONE. Updating logbook");
1642                         UpdateShuttleLogbook("shuttle_done");
1643                 }
1644                 else
1645                 {
1646                         for (UInt_t iDet=0; iDet<NDetectors(); iDet++)
1647                         {
1648                                 if (checkEntry->GetDetectorStatus(iDet) == AliShuttleLogbookEntry::kUnprocessed)
1649                                 {
1650                                         AliDebug(2, Form("Run %d: setting %s as \"not first time unprocessed\"",
1651                                                         checkEntry->GetRun(), GetDetName(iDet)));
1652                                         fFirstUnprocessed[iDet] = kFALSE;
1653                                 }
1654                         }
1655                         SendMLRunInfo("Pending");
1656                 }
1657         }
1658
1659         fLogbookEntry = 0;
1660
1661         return hasError == kFALSE;
1662 }
1663
1664 //______________________________________________________________________________________________
1665 Int_t AliShuttle::ProcessCurrentDetector()
1666 {
1667         //
1668         // Makes data retrieval just for a specific detector (fCurrentDetector).
1669         // Threre should be a configuration for this detector.
1670
1671         Log("SHUTTLE", Form("ProcessCurrentDetector - Retrieving values for %s, run %d", 
1672                                                 fCurrentDetector.Data(), GetCurrentRun()));
1673
1674         TString wd = gSystem->WorkingDirectory();
1675         
1676         if (!CleanReferenceStorage(fCurrentDetector.Data()))
1677                 return 0;
1678         
1679         gSystem->ChangeDirectory(wd.Data());
1680         
1681         // call preprocessor
1682         AliPreprocessor* aPreprocessor =
1683                 dynamic_cast<AliPreprocessor*> (fPreprocessorMap.GetValue(fCurrentDetector));
1684
1685         // check if the preprocessor wants to process this run type
1686         if (aPreprocessor->ProcessRunType() == kFALSE)
1687         {
1688                 UpdateShuttleStatus(AliShuttleStatus::kSkipped);
1689                 UpdateShuttleLogbook(fCurrentDetector, "DONE");
1690                 if (!UpdateTableSkippedCase(fCurrentDetector.Data()))
1691                 {
1692                         AliError(Form("Could not update FXS tables for run %d !", GetCurrentRun()));
1693                 }
1694                 Log(fCurrentDetector, Form("ProcessCurrentDetector - %s preprocessor is not interested in this run type", fCurrentDetector.Data()));
1695         
1696                 return 2;
1697         }
1698         
1699         TMap* dcsMap = new TMap();
1700         
1701         aPreprocessor->Initialize(GetCurrentRun(), GetCurrentStartTime(), GetCurrentEndTime());
1702
1703         Bool_t processDCS = aPreprocessor->ProcessDCS();
1704
1705         if (!processDCS)
1706         {
1707                 Log(fCurrentDetector, "ProcessCurrentDetector -"
1708                         " The preprocessor requested to skip the retrieval of DCS values");
1709         }
1710         else if (fTestMode & kSkipDCS)
1711         {
1712                 Log(fCurrentDetector, "ProcessCurrentDetector - In TESTMODE: Skipping DCS processing");
1713         } 
1714         else if (fTestMode & kErrorDCS)
1715         {
1716                 Log(fCurrentDetector, "ProcessCurrentDetector - In TESTMODE: Simulating DCS error");
1717                 UpdateShuttleStatus(AliShuttleStatus::kDCSStarted);
1718                 UpdateShuttleStatus(AliShuttleStatus::kDCSError);
1719                 delete dcsMap;
1720                 return 0;
1721         } else {
1722
1723                 UpdateShuttleStatus(AliShuttleStatus::kDCSStarted);
1724
1725                 // Query DCS archive
1726                 Int_t nServers = fConfig->GetNServers(fCurrentDetector);
1727                 
1728                 for (int iServ=0; iServ<nServers; iServ++)
1729                 {
1730                 
1731                         TString host(fConfig->GetDCSHost(fCurrentDetector, iServ));
1732                         Int_t port = fConfig->GetDCSPort(fCurrentDetector, iServ);
1733                         Int_t multiSplit = fConfig->GetMultiSplit(fCurrentDetector, iServ);
1734
1735                         Log(fCurrentDetector, Form("ProcessCurrentDetector -"
1736                                         " Querying DCS Amanda server %s:%d (%d of %d)", 
1737                                         host.Data(), port, iServ+1, nServers));
1738                         
1739                         TMap* aliasMap = 0;
1740                         TMap* dpMap = 0;
1741
1742                         if (fConfig->GetDCSAliases(fCurrentDetector, iServ)->GetEntries() > 0)
1743                         {
1744                                 Log(fCurrentDetector, Form("Querying %d DCS aliases", fConfig->GetDCSAliases(fCurrentDetector, iServ)->GetEntries()));
1745                                 aliasMap = GetValueSet(host, port, 
1746                                                 fConfig->GetDCSAliases(fCurrentDetector, iServ), 
1747                                                 kAlias, multiSplit);
1748                                 if (!aliasMap)
1749                                 {
1750                                         Log(fCurrentDetector, 
1751                                                 Form("ProcessCurrentDetector -"
1752                                                         " Error retrieving DCS aliases from server %s."
1753                                                         " Sending mail to DCS experts!", host.Data()));
1754                                         UpdateShuttleStatus(AliShuttleStatus::kDCSError);
1755                                         
1756                                         if (!SendMail(kDCSEMail))
1757                                                 Log("SHUTTLE", Form("ProcessCurrentDetector - "
1758                                                                     "Could not send mail to DCS experts!"));
1759
1760                                         delete dcsMap;
1761                                         return 0;
1762                                 }
1763                         }
1764                         
1765                         if (fConfig->GetDCSDataPoints(fCurrentDetector, iServ)->GetEntries() > 0)
1766                         {
1767                                 Log(fCurrentDetector, Form("Querying %d DCS data points", fConfig->GetDCSDataPoints(fCurrentDetector, iServ)->GetEntries()));
1768                                 dpMap = GetValueSet(host, port, 
1769                                                 fConfig->GetDCSDataPoints(fCurrentDetector, iServ), 
1770                                                 kDP, multiSplit);
1771                                 if (!dpMap)
1772                                 {
1773                                         Log(fCurrentDetector, 
1774                                                 Form("ProcessCurrentDetector -"
1775                                                         " Error retrieving DCS data points from server %s."
1776                                                         " Sending mail to DCS experts!", host.Data()));
1777                                         UpdateShuttleStatus(AliShuttleStatus::kDCSError);
1778                                         
1779                                         if (!SendMail(kDCSEMail))
1780                                                 Log("SHUTTLE", Form("ProcessCurrentDetector - "
1781                                                                     "Could not send mail to DCS experts!"));
1782                                         
1783                                         if (aliasMap) delete aliasMap;
1784                                         delete dcsMap;
1785                                         return 0;
1786                                 }                               
1787                         }
1788                         
1789                         // merge aliasMap and dpMap into dcsMap
1790                         if(aliasMap) {
1791                                 TIter iter(aliasMap);
1792                                 TObjString* key = 0;
1793                                 while ((key = (TObjString*) iter.Next()))
1794                                         dcsMap->Add(key, aliasMap->GetValue(key->String()));
1795                                 
1796                                 aliasMap->SetOwner(kFALSE);
1797                                 delete aliasMap;
1798                         }       
1799                         
1800                         if(dpMap) {
1801                                 TIter iter(dpMap);
1802                                 TObjString* key = 0;
1803                                 while ((key = (TObjString*) iter.Next()))
1804                                         dcsMap->Add(key, dpMap->GetValue(key->String()));
1805                                 
1806                                 dpMap->SetOwner(kFALSE);
1807                                 delete dpMap;
1808                         }
1809                 }
1810         }
1811         
1812         // save map into file, to help debugging in case of preprocessor error
1813         TFile* f = TFile::Open("DCSMap.root","recreate");
1814         f->cd();
1815         dcsMap->Write("DCSMap", TObject::kSingleKey);
1816         f->Close();
1817         delete f;
1818         
1819         // DCS Archive DB processing successful. Call Preprocessor!
1820         UpdateShuttleStatus(AliShuttleStatus::kPPStarted);
1821
1822         fFXSError = -1; // this variable is kTRUE after ::Process if an FXS error occured
1823         
1824         UInt_t returnValue = aPreprocessor->Process(dcsMap);
1825         
1826         if (fFXSError!=-1) {
1827                 UpdateShuttleStatus(AliShuttleStatus::kFXSError);
1828                 SendMail(kFXSEMail, fFXSError);
1829                 dcsMap->DeleteAll();
1830                 delete dcsMap;
1831                 return 0;
1832         }
1833
1834         if (returnValue > 0) // Preprocessor error!
1835         {
1836                 Log(fCurrentDetector, Form("ProcessCurrentDetector - "
1837                                 "Preprocessor failed. Process returned %d.", returnValue));
1838                 UpdateShuttleStatus(AliShuttleStatus::kPPError);
1839                 dcsMap->DeleteAll();
1840                 delete dcsMap;
1841                 return 0;
1842         }
1843         
1844         // preprocessor ok!
1845         UpdateShuttleStatus(AliShuttleStatus::kPPDone);
1846         Log(fCurrentDetector, Form("ProcessCurrentDetector - %s preprocessor returned success",
1847                                 fCurrentDetector.Data()));
1848
1849         dcsMap->DeleteAll();
1850         delete dcsMap;
1851
1852         return 1;
1853 }
1854
1855 //______________________________________________________________________________________________
1856 void AliShuttle::CountOpenRuns()
1857 {
1858         // Query DAQ's Shuttle logbook and sends the number of open runs to ML
1859         
1860         SendAlive();
1861         
1862         // check connection, in case connect
1863         if (!Connect(3)) 
1864                 return;
1865
1866         TString sqlQuery;
1867         sqlQuery = Form("select count(*) from %s where shuttle_done=0", fConfig->GetShuttlelbTable());
1868         
1869         TSQLResult* aResult = fServer[3]->Query(sqlQuery);
1870         if (!aResult) {
1871                 AliError(Form("Can't execute query <%s>!", sqlQuery.Data()));
1872                 return;
1873         }
1874
1875         AliDebug(2,Form("Query = %s", sqlQuery.Data()));
1876         
1877         if (aResult->GetRowCount() == 0) {
1878                 AliError(Form("No result for query %s received", sqlQuery.Data()));
1879                 return;
1880         }
1881
1882         if (aResult->GetFieldCount() != 1) {
1883                 AliError(Form("Invalid field count for query %s received", sqlQuery.Data()));
1884                 return;
1885         }
1886
1887         TSQLRow* aRow = aResult->Next();
1888         if (!aRow) {
1889                 AliError(Form("Could not receive result of query %s", sqlQuery.Data()));
1890                 return;
1891         }
1892         
1893         TString result(aRow->GetField(0), aRow->GetFieldLength(0));
1894         Int_t count = result.Atoi();
1895         
1896         Log("SHUTTLE", Form("%d unprocessed runs", count));
1897         
1898         delete aRow;
1899         delete aResult;
1900
1901         TMonaLisaValue mlStatus("SHUTTLE_openruns", count);
1902
1903         TList mlList;
1904         mlList.Add(&mlStatus);
1905
1906         fMonaLisa->SendParameters(&mlList, "__PROCESSINGINFO__");
1907 }
1908
1909 //______________________________________________________________________________________________
1910 Bool_t AliShuttle::QueryShuttleLogbook(const char* whereClause,
1911                 TObjArray& entries)
1912 {
1913         // Query DAQ's Shuttle logbook and fills detector status object.
1914         // Call QueryRunParameters to query DAQ logbook for run parameters.
1915         //
1916
1917         entries.SetOwner(1);
1918
1919         // check connection, in case connect
1920         if (!Connect(3)) return kFALSE;
1921
1922         TString sqlQuery;
1923         sqlQuery = Form("select * from %s %s order by run", fConfig->GetShuttlelbTable(), whereClause);
1924
1925         TSQLResult* aResult = fServer[3]->Query(sqlQuery);
1926         if (!aResult) {
1927                 AliError(Form("Can't execute query <%s>!", sqlQuery.Data()));
1928                 return kFALSE;
1929         }
1930
1931         AliDebug(2,Form("Query = %s", sqlQuery.Data()));
1932
1933         if(aResult->GetRowCount() == 0) {
1934                 Log("SHUTTLE", "No entries in Shuttle Logbook match request");
1935                 delete aResult;
1936                 return kTRUE;
1937         }
1938
1939         // TODO Check field count!
1940         const UInt_t nCols = 26;
1941         if (aResult->GetFieldCount() != (Int_t) nCols) {
1942                 Log("SHUTTLE", "Invalid SQL result field number!");
1943                 delete aResult;
1944                 return kFALSE;
1945         }
1946
1947         TSQLRow* aRow;
1948         while ((aRow = aResult->Next())) {
1949                 TString runString(aRow->GetField(0), aRow->GetFieldLength(0));
1950                 Int_t run = runString.Atoi();
1951
1952                 AliShuttleLogbookEntry *entry = QueryRunParameters(run);
1953                 if (!entry)
1954                         continue;
1955
1956                 // DA test mode flag
1957                 TString daTestModeString(aRow->GetField(2), aRow->GetFieldLength(2)); // field 2 = DA test mode flag 
1958                 Bool_t daTestMode = (Bool_t)daTestModeString.Atoi();
1959                 entry->SetDATestMode(daTestMode);
1960
1961                 // loop on detectors
1962                 for(UInt_t ii = 0; ii < nCols; ii++)
1963                         entry->SetDetectorStatus(aResult->GetFieldName(ii), aRow->GetField(ii));
1964
1965                 entries.AddLast(entry);
1966                 delete aRow;
1967         }
1968
1969         delete aResult;
1970         return kTRUE;
1971 }
1972
1973 //______________________________________________________________________________________________
1974 AliShuttleLogbookEntry* AliShuttle::QueryRunParameters(Int_t run)
1975 {
1976         //
1977         // Retrieve run parameters written in the DAQ logbook and sets them into AliShuttleLogbookEntry object
1978         //
1979
1980         // check connection, in case connect
1981         if (!Connect(3))
1982                 return 0;
1983
1984         TString sqlQuery;
1985         sqlQuery.Form("select * from %s where run=%d", fConfig->GetDAQlbTable(), run);
1986
1987         TSQLResult* aResult = fServer[3]->Query(sqlQuery);
1988         if (!aResult) {
1989                 Log("SHUTTLE", Form("Can't execute query <%s>!", sqlQuery.Data()));
1990                 return 0;
1991         }
1992
1993         if (aResult->GetRowCount() == 0) {
1994                 Log("SHUTTLE", Form("QueryRunParameters - No entry in DAQ Logbook for run %d. Skipping", run));
1995                 delete aResult;
1996                 return 0;
1997         }
1998
1999         if (aResult->GetRowCount() > 1) {
2000                 Log("SHUTTLE", Form("QueryRunParameters - UNEXPECTED: "
2001                                 "more than one entry in DAQ Logbook for run %d!", run));
2002                 delete aResult;
2003                 return 0;
2004         }
2005
2006         TSQLRow* aRow = aResult->Next();
2007         if (!aRow)
2008         {
2009                 Log("SHUTTLE", Form("QueryRunParameters - Could not retrieve row for run %d. Skipping", run));
2010                 delete aResult;
2011                 return 0;
2012         }
2013
2014         AliShuttleLogbookEntry* entry = new AliShuttleLogbookEntry(run);
2015
2016         for (Int_t ii = 0; ii < aResult->GetFieldCount(); ii++)
2017                 entry->SetRunParameter(aResult->GetFieldName(ii), aRow->GetField(ii));
2018
2019         delete aRow;
2020         delete aResult;
2021         
2022         UInt_t startTime = entry->GetStartTime();
2023         UInt_t endTime = entry->GetEndTime();
2024         Bool_t ecsSuccess = entry->GetECSSuccess();
2025         
2026         TString totEventsStr = entry->GetRunParameter("totalEvents");  
2027         Int_t totEvents = totEventsStr.Atoi();
2028         
2029         UInt_t now = time(0);
2030         // TODO make this a configuration parameter
2031         Int_t dcsDelay = fConfig->GetDCSDelay()+fConfig->GetDCSQueryOffset();
2032         
2033         // runs are accepted if they have ecsSuccess set or more than 1 event
2034         if (startTime != 0 && endTime != 0 && endTime > startTime && (totEvents > 1 || ecsSuccess) && (endTime < now - dcsDelay))
2035         {
2036                 if (ecsSuccess == kFALSE)
2037                         Log("SHUTTLE", Form("Processing run %d although in status ECS failure, Reason: %s", run, entry->GetRunParameter("eor_reason")));
2038                 return entry;
2039         }
2040
2041         Bool_t skip = kFALSE;
2042                                 
2043         if (endTime != 0 && endTime >= now - dcsDelay)
2044         {
2045                 Log("SHUTTLE", Form("Skipping run %d for now, because DCS buffer time is not yet expired", run));
2046         }
2047         else if (totEvents <= 1) 
2048         {
2049                 Log("SHUTTLE", Form("QueryRunParameters - Run %d has 1 event or less - Skipping!", run));
2050                 skip = kTRUE;
2051         }
2052         else
2053         {
2054                 Log("SHUTTLE", Form("QueryRunParameters - Invalid parameters for Run %d: "
2055                         "startTime = %d, endTime = %d. Skipping (Shuttle won't be marked as DONE)!",
2056                         run, startTime, endTime));
2057         }
2058         
2059         if (skip)
2060         {
2061                 Log("SHUTTLE", Form("Marking SHUTTLE skipped for run %d", run));
2062                 fLogbookEntry = entry;
2063                 if (!UpdateShuttleLogbook("shuttle_skipped"))
2064                 {
2065                         AliError(Form("Could not update logbook for run %d !", run));
2066                 }
2067                 if (!UpdateTableSkippedCase("ALL"))
2068                 {
2069                         AliError(Form("Could not update FXS tables for run %d !", run));
2070                 }
2071                 fLogbookEntry = 0;
2072         }
2073                         
2074         delete entry;
2075         return 0;
2076 }
2077
2078 //______________________________________________________________________________________________
2079 TMap* AliShuttle::GetValueSet(const char* host, Int_t port, const TSeqCollection* entries,
2080                               DCSType type, Int_t multiSplit)
2081 {
2082         // Retrieve all "entry" data points from the DCS server
2083         // host, port: TSocket connection parameters
2084         // entries: list of name of the alias or data point
2085         // type: kAlias or kDP
2086         // returns TMap of values, 0 when failure
2087         
2088         AliDCSClient client(host, port, fTimeout, fRetries, multiSplit);
2089
2090         TMap* result = 0;
2091         if (type == kAlias)
2092         {
2093                 //result = client.GetAliasValues(entries, GetCurrentStartTime()-offset, 
2094                 //      GetCurrentEndTime()+offset);
2095                 result = client.GetAliasValues(entries, GetStartTimeDCSQuery(), 
2096                         GetEndTimeDCSQuery());
2097         } 
2098         else if (type == kDP)
2099         {
2100                 //result = client.GetDPValues(entries, GetCurrentStartTime()-offset, 
2101                 //      GetCurrentEndTime()+offset);
2102                 result = client.GetDPValues(entries, GetStartTimeDCSQuery(), 
2103                         GetEndTimeDCSQuery());
2104         }
2105
2106         if (result == 0)
2107         {
2108                 Log(fCurrentDetector.Data(), Form("GetValueSet - Can't get entries! Reason: %s",
2109                         client.GetErrorString(client.GetResultErrorCode())));
2110                 if (client.GetResultErrorCode() == AliDCSClient::fgkServerError)        
2111                         Log(fCurrentDetector.Data(), Form("GetValueSet - Server error code: %s",
2112                                 client.GetServerError().Data()));
2113
2114                 return 0;
2115         }
2116                 
2117         return result;
2118 }
2119
2120 //______________________________________________________________________________________________
2121 const char* AliShuttle::GetFile(Int_t system, const char* detector,
2122                 const char* id, const char* source)
2123 {
2124         // Get calibration file from file exchange servers
2125         // First queris the FXS database for the file name, using the run, detector, id and source info
2126         // then calls RetrieveFile(filename) for actual copy to local disk
2127         // run: current run being processed (given by Logbook entry fLogbookEntry)
2128         // detector: the Preprocessor name
2129         // id: provided as a parameter by the Preprocessor
2130         // source: provided by the Preprocessor through GetFileSources function
2131
2132         // check if test mode should simulate a FXS error
2133         if (fTestMode & kErrorFXSFiles)
2134         {
2135                 Log(detector, Form("GetFile - In TESTMODE - Simulating error while connecting to %s FXS", GetSystemName(system)));
2136                 return 0;
2137         }
2138         
2139         // check connection, in case connect
2140         if (!Connect(system))
2141         {
2142                 Log(detector, Form("GetFile - Couldn't connect to %s FXS database", GetSystemName(system)));
2143                 fFXSError = system;
2144                 return 0;
2145         }
2146
2147         // Query preparation
2148         TString sourceName(source);
2149         Int_t nFields = 3;
2150         TString sqlQueryStart = Form("select filePath,size,fileChecksum from %s where",
2151                                                                 fConfig->GetFXSdbTable(system));
2152         TString whereClause = Form("run=%d and detector=\"%s\" and fileId=\"%s\"",
2153                                                                 GetCurrentRun(), detector, id);
2154
2155         if (system == kDAQ)
2156         {
2157                 whereClause += Form(" and DAQsource=\"%s\"", source);
2158         }
2159         else if (system == kDCS)
2160         {
2161                 sourceName="none";
2162         }
2163         else if (system == kHLT)
2164         {
2165                 whereClause += Form(" and DDLnumbers=\"%s\"", source);
2166         }
2167
2168         TString sqlQuery = Form("%s %s", sqlQueryStart.Data(), whereClause.Data());
2169
2170         AliDebug(2, Form("SQL query: \n%s",sqlQuery.Data()));
2171
2172         // Query execution
2173         TSQLResult* aResult = 0;
2174         aResult = dynamic_cast<TSQLResult*> (fServer[system]->Query(sqlQuery));
2175         if (!aResult) {
2176                 Log(detector, Form("GetFile - Can't execute SQL query to %s database for: id = %s, source = %s",
2177                                 GetSystemName(system), id, sourceName.Data()));
2178                 fFXSError = system;
2179                 return 0;
2180         }
2181
2182         if (aResult->GetRowCount() == 0)
2183         {
2184                 Log(detector,
2185                         Form("GetFile - No entry in %s FXS db for: id = %s, source = %s",
2186                                 GetSystemName(system), id, sourceName.Data()));
2187                 delete aResult;
2188                 return 0;
2189         }
2190
2191         if (aResult->GetRowCount() > 1) {
2192                 Log(detector,
2193                         Form("GetFile - More than one entry in %s FXS db for: id = %s, source = %s",
2194                                 GetSystemName(system), id, sourceName.Data()));
2195                 fFXSError = system;
2196                 delete aResult;
2197                 return 0;
2198         }
2199
2200         if (aResult->GetFieldCount() != nFields) {
2201                 Log(detector,
2202                         Form("GetFileName - Wrong field count in %s FXS db for: id = %s, source = %s",
2203                                 GetSystemName(system), id, sourceName.Data()));
2204                 fFXSError = system;
2205                 delete aResult;
2206                 return 0;
2207         }
2208
2209         TSQLRow* aRow = dynamic_cast<TSQLRow*> (aResult->Next());
2210
2211         if (!aRow){
2212                 Log(detector, Form("GetFile - Empty set result in %s FXS db from query: id = %s, source = %s",
2213                                 GetSystemName(system), id, sourceName.Data()));
2214                 fFXSError = system;
2215                 delete aResult;
2216                 return 0;
2217         }
2218
2219         TString filePath(aRow->GetField(0), aRow->GetFieldLength(0));
2220         TString fileSize(aRow->GetField(1), aRow->GetFieldLength(1));
2221         TString fileChecksum(aRow->GetField(2), aRow->GetFieldLength(2));
2222
2223         delete aResult;
2224         delete aRow;
2225
2226         AliDebug(2, Form("filePath = %s; size = %s, fileChecksum = %s",
2227                                 filePath.Data(), fileSize.Data(), fileChecksum.Data()));
2228
2229         // retrieved file is renamed to make it unique
2230         Int_t dir_lev1 = GetCurrentRun()/10000;
2231         TString localFileName = Form("%s/%d/%d/%s_process/%s_%s_%d_%s_%s.shuttle",
2232                                      GetShuttleTempDir(), dir_lev1, GetCurrentRun(), detector,
2233                                         GetSystemName(system), detector, GetCurrentRun(), 
2234                                         id, sourceName.Data());
2235         Log("SHUTTLE",Form("file from FXS = %s",localFileName.Data())); 
2236
2237
2238         // file retrieval from FXS
2239         UInt_t nRetries = 0;
2240         UInt_t maxRetries = 3;
2241         Bool_t result = kFALSE;
2242
2243         // copy!! if successful TSystem::Exec returns 0
2244         while (nRetries++ < maxRetries) {
2245                 AliDebug(2, Form("Trying to copy file. Retry # %d", nRetries));
2246                 result = RetrieveFile(system, filePath.Data(), localFileName.Data());
2247                 if (!result)
2248                 {
2249                         Log(detector, Form("GetFile - Copy of file %s from %s FXS failed",
2250                                         filePath.Data(), GetSystemName(system)));
2251                         continue;
2252                 } 
2253
2254                 if (fileSize.Length()>0)
2255                 {
2256                         // compare filesize of local file with the one stored in the FXS DB
2257                         Long_t size = -1;
2258                         Int_t sizeComp = gSystem->GetPathInfo(localFileName.Data(), 0, &size, 0, 0);
2259
2260                         if (sizeComp != 0 || size != fileSize.Atoi())
2261                         {
2262                                 Log(detector, Form("GetFile - size of file %s does not match with local copy!",
2263                                                         filePath.Data()));
2264                                 result = kFALSE;
2265                                 continue;
2266                         }
2267
2268                 } else {
2269                         Log(fCurrentDetector, Form("GetFile - size of file %s not set in %s database, skipping comparison",
2270                                                 filePath.Data(), GetSystemName(system)));
2271                 }
2272
2273                 if (fileChecksum.Length()>0)
2274                 {
2275                         // compare md5sum of local file with the one stored in the FXS DB
2276                         if(fileChecksum.Contains(' ')) fileChecksum.Resize(fileChecksum.First(' '));
2277                         Int_t md5Comp = gSystem->Exec(Form("md5sum %s |grep %s > /dev/null 2> /dev/null",
2278                                                 localFileName.Data(), fileChecksum.Data()));
2279
2280                         if (md5Comp != 0)
2281                         {
2282                                 Log(detector, Form("GetFile - md5sum of file %s does not match with local copy!",
2283                                                         filePath.Data()));
2284                                 result = kFALSE;
2285                                 continue;
2286                         }
2287                 } else {
2288                         Log(fCurrentDetector, Form("GetFile - md5sum of file %s not set in %s database, skipping comparison",
2289                                                         filePath.Data(), GetSystemName(system)));
2290                 }
2291                 if (result) break;
2292         }
2293
2294         if (!result) 
2295         {
2296                 fFXSError = system;
2297                 return 0;
2298         }
2299
2300         fFXSCalled[system]=kTRUE;
2301         TObjString *fileParams = new TObjString(Form("%s#!?!#%s", id, sourceName.Data()));
2302         fFXSlist[system].Add(fileParams);
2303
2304         static TString staticLocalFileName;
2305         staticLocalFileName.Form("%s", localFileName.Data());
2306         
2307         Log(fCurrentDetector, Form("GetFile - Retrieved file with id %s and "
2308                         "source %s from %s to %s", id, source, 
2309                         GetSystemName(system), localFileName.Data()));
2310                         
2311         return staticLocalFileName.Data();
2312 }
2313
2314 //______________________________________________________________________________________________
2315 Bool_t AliShuttle::RetrieveFile(UInt_t system, const char* fxsFileName, const char* localFileName)
2316 {
2317         //
2318         // Copies file from FXS to local Shuttle machine
2319         //
2320
2321         // check temp directory: trying to cd to temp; if it does not exist, create it
2322         AliDebug(2, Form("Copy file %s from %s FXS into %s",
2323                         GetSystemName(system), fxsFileName, localFileName));
2324                         
2325         TString tmpDir(localFileName);
2326         
2327         tmpDir = tmpDir(0,tmpDir.Last('/'));
2328
2329         Int_t noDir = gSystem->GetPathInfo(tmpDir.Data(), 0, (Long64_t*) 0, 0, 0);
2330         if (noDir) // temp dir does not exists!
2331         {
2332                 if (gSystem->mkdir(tmpDir.Data(), 1))
2333                 {
2334                         Log(fCurrentDetector.Data(), "RetrieveFile - could not make temp directory!!");
2335                         return kFALSE;
2336                 }
2337         }
2338
2339         TString command = Form("scp -oPort=%d -2 %s@%s:%s/%s %s",
2340                 fConfig->GetFXSPort(system),
2341                 fConfig->GetFXSUser(system),
2342                 fConfig->GetFXSHost(system),
2343                 fConfig->GetFXSBaseFolder(system),
2344                 fxsFileName,
2345                 localFileName);
2346
2347         AliDebug(2, Form("%s",command.Data()));
2348
2349         Bool_t result = (gSystem->Exec(command.Data()) == 0);
2350
2351         return result;
2352 }
2353
2354 //______________________________________________________________________________________________
2355 TList* AliShuttle::GetFileSources(Int_t system, const char* detector, const char* id)
2356 {
2357         //
2358         // Get sources producing the condition file Id from file exchange servers
2359         // if id is NULL all sources are returned (distinct)
2360         //
2361
2362         if (id)
2363         {
2364                 Log(detector, Form("GetFileSources - Querying %s FXS for files with id %s produced by %s", GetSystemName(system), id, detector));
2365         } else {
2366                 Log(detector, Form("GetFileSources - Querying %s FXS for files produced by %s", GetSystemName(system), detector));
2367         }
2368         
2369         // check if test mode should simulate a FXS error
2370         if (fTestMode & kErrorFXSSources)
2371         {
2372                 Log(detector, Form("GetFileSources - In TESTMODE - Simulating error while connecting to %s FXS", GetSystemName(system)));
2373                 return 0;
2374         }
2375
2376         if (system == kDCS)
2377         {
2378                 Log(detector, "GetFileSources - WARNING: DCS system has only one source of data!");
2379                 TList *list = new TList();
2380                 list->SetOwner(1);
2381                 list->Add(new TObjString(" "));
2382                 return list;
2383         }
2384
2385         // check connection, in case connect
2386         if (!Connect(system))
2387         {
2388                 Log(detector, Form("GetFileSources - Couldn't connect to %s FXS database", GetSystemName(system)));
2389                 fFXSError = system;
2390                 return NULL;
2391         }
2392
2393         TString sourceName = "";
2394         if (system == kDAQ)
2395         {
2396                 sourceName = "DAQsource";
2397         } else if (system == kHLT)
2398         {
2399                 sourceName = "DDLnumbers";
2400         }
2401
2402         TString sqlQueryStart = Form("select distinct %s from %s where", sourceName.Data(), fConfig->GetFXSdbTable(system));
2403         TString whereClause = Form("run=%d and detector=\"%s\"",
2404                                 GetCurrentRun(), detector);
2405         if (id)
2406                 whereClause += Form(" and fileId=\"%s\"", id);
2407         TString sqlQuery = Form("%s %s", sqlQueryStart.Data(), whereClause.Data());
2408
2409         AliDebug(2, Form("SQL query: \n%s",sqlQuery.Data()));
2410
2411         // Query execution
2412         TSQLResult* aResult;
2413         aResult = fServer[system]->Query(sqlQuery);
2414         if (!aResult) {
2415                 Log(detector, Form("GetFileSources - Can't execute SQL query to %s database for id: %s",
2416                                 GetSystemName(system), id));
2417                 fFXSError = system;
2418                 return 0;
2419         }
2420
2421         TList *list = new TList();
2422         list->SetOwner(1);
2423         
2424         if (aResult->GetRowCount() == 0)
2425         {
2426                 Log(detector,
2427                         Form("GetFileSources - No entry in %s FXS table for id: %s", GetSystemName(system), id));
2428                 delete aResult;
2429                 return list;
2430         }
2431
2432         Log(detector, Form("GetFileSources - Found %d sources", aResult->GetRowCount()));
2433
2434         TSQLRow* aRow;
2435         while ((aRow = aResult->Next()))
2436         {
2437
2438                 TString source(aRow->GetField(0), aRow->GetFieldLength(0));
2439                 AliDebug(2, Form("%s = %s", sourceName.Data(), source.Data()));
2440                 list->Add(new TObjString(source));
2441                 delete aRow;
2442         }
2443
2444         delete aResult;
2445
2446         return list;
2447 }
2448
2449 //______________________________________________________________________________________________
2450 TList* AliShuttle::GetFileIDs(Int_t system, const char* detector, const char* source)
2451 {
2452         //
2453         // Get all ids of condition files produced by a given source from file exchange servers
2454         //
2455         
2456         Log(detector, Form("GetFileIDs - Retrieving ids with source %s with %s", source, GetSystemName(system)));
2457
2458         // check if test mode should simulate a FXS error
2459         if (fTestMode & kErrorFXSSources)
2460         {
2461                 Log(detector, Form("GetFileIDs - In TESTMODE - Simulating error while connecting to %s FXS", GetSystemName(system)));
2462                 return 0;
2463         }
2464
2465         // check connection, in case connect
2466         if (!Connect(system))
2467         {
2468                 Log(detector, Form("GetFileIDs - Couldn't connect to %s FXS database", GetSystemName(system)));
2469                 return NULL;
2470         }
2471
2472         TString sourceName = "";
2473         if (system == kDAQ)
2474         {
2475                 sourceName = "DAQsource";
2476         } else if (system == kHLT)
2477         {
2478                 sourceName = "DDLnumbers";
2479         }
2480
2481         TString sqlQueryStart = Form("select fileId from %s where", fConfig->GetFXSdbTable(system));
2482         TString whereClause = Form("run=%d and detector=\"%s\"",
2483                                 GetCurrentRun(), detector);
2484         if (sourceName.Length() > 0 && source)
2485                 whereClause += Form(" and %s=\"%s\"", sourceName.Data(), source);
2486         TString sqlQuery = Form("%s %s", sqlQueryStart.Data(), whereClause.Data());
2487
2488         AliDebug(2, Form("SQL query: \n%s",sqlQuery.Data()));
2489
2490         // Query execution
2491         TSQLResult* aResult;
2492         aResult = fServer[system]->Query(sqlQuery);
2493         if (!aResult) {
2494                 Log(detector, Form("GetFileIDs - Can't execute SQL query to %s database for source: %s",
2495                                 GetSystemName(system), source));
2496                 return 0;
2497         }
2498
2499         TList *list = new TList();
2500         list->SetOwner(1);
2501         
2502         if (aResult->GetRowCount() == 0)
2503         {
2504                 Log(detector,
2505                         Form("GetFileIDs - No entry in %s FXS table for source: %s", GetSystemName(system), source));
2506                 delete aResult;
2507                 return list;
2508         }
2509
2510         Log(detector, Form("GetFileIDs - Found %d ids", aResult->GetRowCount()));
2511
2512         TSQLRow* aRow;
2513
2514         while ((aRow = aResult->Next()))
2515         {
2516
2517                 TString id(aRow->GetField(0), aRow->GetFieldLength(0));
2518                 AliDebug(2, Form("fileId = %s", id.Data()));
2519                 list->Add(new TObjString(id));
2520                 delete aRow;
2521         }
2522
2523         delete aResult;
2524
2525         return list;
2526 }
2527
2528 //______________________________________________________________________________________________
2529 Bool_t AliShuttle::Connect(Int_t system)
2530 {
2531         // Connect to MySQL Server of the system's FXS MySQL databases
2532         // DAQ Logbook, Shuttle Logbook and DAQ FXS db are on the same host
2533         //
2534
2535         // check connection: if already connected return
2536
2537         if(fServer[system] && fServer[system]->IsConnected()) {
2538                 // ping the server              
2539                 if (fServer[system]->PingVerify()==kTRUE){ // connection is still alive
2540                         return kTRUE;
2541                 }               
2542                 else{
2543                         AliWarning(Form("Connection got lost to FXS database for %s. Closing and reconnecting.",
2544                                         AliShuttleInterface::GetSystemName(system)));
2545                         fServer[system]->Close();
2546                         delete fServer[system];
2547                         fServer[system] = 0x0;
2548                 }
2549         }
2550
2551         TString dbHost, dbUser, dbPass, dbName;
2552
2553         if (system < 3) // FXS db servers
2554         {
2555                 dbHost = Form("mysql://%s:%d", fConfig->GetFXSdbHost(system), fConfig->GetFXSdbPort(system));
2556                 dbUser = fConfig->GetFXSdbUser(system);
2557                 dbPass = fConfig->GetFXSdbPass(system);
2558                 dbName =   fConfig->GetFXSdbName(system);
2559         } else { // Run & Shuttle logbook servers
2560         // TODO Will the Shuttle logbook server be the same as the Run logbook server ???
2561                 dbHost = Form("mysql://%s:%d", fConfig->GetDAQlbHost(), fConfig->GetDAQlbPort());
2562                 dbUser = fConfig->GetDAQlbUser();
2563                 dbPass = fConfig->GetDAQlbPass();
2564                 dbName =   fConfig->GetDAQlbDB();
2565         }
2566
2567         fServer[system] = TSQLServer::Connect(dbHost.Data(), dbUser.Data(), dbPass.Data());
2568         if (!fServer[system] || !fServer[system]->IsConnected()) {
2569                 if(system < 3)
2570                 {
2571                 AliError(Form("Can't establish connection to FXS database for %s",
2572                                         AliShuttleInterface::GetSystemName(system)));
2573                 } else {
2574                 AliError("Can't establish connection to Run logbook.");
2575                 }
2576                 if(fServer[system]) delete fServer[system];
2577                 return kFALSE;
2578         }
2579
2580         // Get tables
2581         TSQLResult* aResult=0;
2582         switch(system){
2583                 case kDAQ:
2584                         aResult = fServer[kDAQ]->GetTables(dbName.Data());
2585                         break;
2586                 case kDCS:
2587                         aResult = fServer[kDCS]->GetTables(dbName.Data());
2588                         break;
2589                 case kHLT:
2590                         aResult = fServer[kHLT]->GetTables(dbName.Data());
2591                         break;
2592                 default:
2593                         aResult = fServer[3]->GetTables(dbName.Data());
2594                         break;
2595         }
2596
2597         delete aResult;
2598         return kTRUE;
2599 }
2600
2601 //______________________________________________________________________________________________
2602 Bool_t AliShuttle::UpdateTable()
2603 {
2604         //
2605         // Update FXS table filling time_processed field in all rows corresponding to current run and detector
2606         //
2607
2608         Bool_t result = kTRUE;
2609
2610         for (UInt_t system=0; system<3; system++)
2611         {
2612                 if(!fFXSCalled[system]) continue;
2613
2614                 // check connection, in case connect
2615                 if (!Connect(system))
2616                 {
2617                         Log(fCurrentDetector, Form("UpdateTable - Couldn't connect to %s FXS database", GetSystemName(system)));
2618                         result = kFALSE;
2619                         continue;
2620                 }
2621
2622                 TTimeStamp now; // now
2623
2624                 // Loop on FXS list entries
2625                 TIter iter(&fFXSlist[system]);
2626                 TObjString *aFXSentry=0;
2627                 while ((aFXSentry = dynamic_cast<TObjString*> (iter.Next())))
2628                 {
2629                         TString aFXSentrystr = aFXSentry->String();
2630                         TObjArray *aFXSarray = aFXSentrystr.Tokenize("#!?!#");
2631                         if (!aFXSarray || aFXSarray->GetEntries() != 2 )
2632                         {
2633                                 Log(fCurrentDetector, Form("UpdateTable - error updating %s FXS entry. Check string: <%s>",
2634                                         GetSystemName(system), aFXSentrystr.Data()));
2635                                 if(aFXSarray) delete aFXSarray;
2636                                 result = kFALSE;
2637                                 continue;
2638                         }
2639                         const char* fileId = ((TObjString*) aFXSarray->At(0))->GetName();
2640                         const char* source = ((TObjString*) aFXSarray->At(1))->GetName();
2641
2642                         TString whereClause;
2643                         if (system == kDAQ)
2644                         {
2645                                 whereClause = Form("where run=%d and detector=\"%s\" and fileId=\"%s\" and DAQsource=\"%s\";",
2646                                                         GetCurrentRun(), fCurrentDetector.Data(), fileId, source);
2647                         }
2648                         else if (system == kDCS)
2649                         {
2650                                 whereClause = Form("where run=%d and detector=\"%s\" and fileId=\"%s\";",
2651                                                         GetCurrentRun(), fCurrentDetector.Data(), fileId);
2652                         }
2653                         else if (system == kHLT)
2654                         {
2655                                 whereClause = Form("where run=%d and detector=\"%s\" and fileId=\"%s\" and DDLnumbers=\"%s\";",
2656                                                         GetCurrentRun(), fCurrentDetector.Data(), fileId, source);
2657                         }
2658
2659                         delete aFXSarray;
2660
2661                         TString sqlQuery = Form("update %s set time_processed=%d %s", fConfig->GetFXSdbTable(system),
2662                                                                 now.GetSec(), whereClause.Data());
2663
2664                         AliDebug(2, Form("SQL query: \n%s",sqlQuery.Data()));
2665
2666                         // Query execution
2667                         TSQLResult* aResult;
2668                         aResult = dynamic_cast<TSQLResult*> (fServer[system]->Query(sqlQuery));
2669                         if (!aResult)
2670                         {
2671                                 Log(fCurrentDetector, Form("UpdateTable - %s db: can't execute SQL query <%s>",
2672                                                                 GetSystemName(system), sqlQuery.Data()));
2673                                 result = kFALSE;
2674                                 continue;
2675                         }
2676                         delete aResult;
2677                 }
2678         }
2679
2680         return result;
2681 }
2682
2683 //_______________________________________________________________________________
2684 Bool_t AliShuttle::UpdateTableSkippedCase(const char* detector)
2685 {
2686         //
2687         // Update FXS table filling time_processed field in all rows corresponding to current run and detector
2688         // if detector = "ALL" update all detectors
2689         //
2690
2691         Bool_t result = kTRUE;
2692
2693         TString detName(detector);
2694
2695         for (UInt_t system=0; system<3; system++)
2696         {
2697
2698                 // check connection, in case connect
2699                 if (!Connect(system))
2700                 {
2701                         Log(fCurrentDetector, Form("UpdateTableSkippedCase - Couldn't connect to %s FXS database", GetSystemName(system)));
2702                         result = kFALSE;
2703                         continue;
2704                 }
2705
2706                 TTimeStamp now; // now
2707
2708                 // Loop on FXS list entries
2709                 TIter iter(&fFXSlist[system]);
2710                         
2711                 TString whereClause;
2712                 if (detName == "ALL") whereClause = Form("where run=%d and time_processed IS NULL;",GetCurrentRun());
2713                 else whereClause = Form("where run=%d and detector=\"%s\" and time_processed IS NULL;",GetCurrentRun(), detector);
2714
2715                 //Log("SHUTTLE",Form(" whereClause = %s ",whereClause.Data()));
2716
2717                 TString sqlQuery = Form("update %s set time_processed=%d %s", fConfig->GetFXSdbTable(system),
2718                                         now.GetSec(), whereClause.Data());
2719
2720                 AliDebug(2, Form("SQL query: \n%s",sqlQuery.Data()));
2721
2722                 // Query execution
2723                 TSQLResult* aResult;
2724                 aResult = dynamic_cast<TSQLResult*> (fServer[system]->Query(sqlQuery));
2725                 if (!aResult)
2726                 {
2727                         Log("SHUTTLE", Form("UpdateTableSkippedCase - %s db: can't execute SQL query <%s>",
2728                                                         GetSystemName(system), sqlQuery.Data()));
2729                         result = kFALSE;
2730                         continue;
2731                 }
2732                 delete aResult;
2733                 
2734         }
2735
2736         return result;
2737 }
2738 //______________________________________________________________________________________________
2739 Bool_t AliShuttle::UpdateTableFailCase()
2740 {
2741         // Update FXS table filling time_processed field in all rows corresponding to current run and detector
2742         // this is called in case the preprocessor is declared failed for the current run, because
2743         // the fields are updated only in case of success
2744
2745         Bool_t result = kTRUE;
2746
2747         for (UInt_t system=0; system<3; system++)
2748         {
2749                 // check connection, in case connect
2750                 if (!Connect(system))
2751                 {
2752                         Log(fCurrentDetector, Form("UpdateTableFailCase - Couldn't connect to %s FXS database",
2753                                                         GetSystemName(system)));
2754                         result = kFALSE;
2755                         continue;
2756                 }
2757
2758                 TTimeStamp now; // now
2759
2760                 // Loop on FXS list entries
2761
2762                 TString whereClause = Form("where run=%d and detector=\"%s\";",
2763                                                 GetCurrentRun(), fCurrentDetector.Data());
2764
2765
2766                 TString sqlQuery = Form("update %s set time_processed=%d %s", fConfig->GetFXSdbTable(system),
2767                                                         now.GetSec(), whereClause.Data());
2768
2769                 AliDebug(2, Form("SQL query: \n%s",sqlQuery.Data()));
2770
2771                 // Query execution
2772                 TSQLResult* aResult;
2773                 aResult = dynamic_cast<TSQLResult*> (fServer[system]->Query(sqlQuery));
2774                 if (!aResult)
2775                 {
2776                         Log(fCurrentDetector, Form("UpdateTableFailCase - %s db: can't execute SQL query <%s>",
2777                                                         GetSystemName(system), sqlQuery.Data()));
2778                         result = kFALSE;
2779                         continue;
2780                 }
2781                 delete aResult;
2782         }
2783
2784         return result;
2785 }
2786
2787 //______________________________________________________________________________________________
2788 Bool_t AliShuttle::UpdateShuttleLogbook(const char* detector, const char* status)
2789 {
2790         //
2791         // Update Shuttle logbook filling detector or shuttle_done column
2792         // ex. of usage: UpdateShuttleLogbook("PHOS", "DONE") or UpdateShuttleLogbook("shuttle_done")
2793         //
2794
2795         // check connection, in case connect
2796         if(!Connect(3)){
2797                 Log("SHUTTLE", "UpdateShuttleLogbook - Couldn't connect to DAQ Logbook.");
2798                 return kFALSE;
2799         }
2800
2801         TString detName(detector);
2802         TString setClause;
2803         if (detName == "shuttle_done" || detName == "shuttle_skipped")
2804         {
2805                 setClause = "set shuttle_done=1";
2806                 
2807                 if (detName == "shuttle_done")
2808                 {
2809                         if (TouchFile() != kTRUE)
2810                         {
2811                                 SendMLRunInfo("Pending");
2812                                 return kFALSE;
2813                         }
2814                         
2815                         SendMLRunInfo("Done");
2816                 }
2817                 else 
2818                         SendMLRunInfo("Skipped");
2819         } 
2820         else {
2821                 TString statusStr(status);
2822                 if(statusStr.Contains("done", TString::kIgnoreCase) ||
2823                    statusStr.Contains("failed", TString::kIgnoreCase)){
2824                         setClause = Form("set %s=\"%s\"", detector, status);
2825                 } else {
2826                         Log("SHUTTLE",
2827                                 Form("UpdateShuttleLogbook - Invalid status <%s> for detector %s",
2828                                         status, detector));
2829                         return kFALSE;
2830                 }
2831         }
2832
2833         TString whereClause = Form("where run=%d", GetCurrentRun());
2834
2835         TString sqlQuery = Form("update %s %s %s",
2836                                         fConfig->GetShuttlelbTable(), setClause.Data(), whereClause.Data());
2837
2838         AliDebug(2, Form("SQL query: \n%s",sqlQuery.Data()));
2839
2840         // Query execution
2841         TSQLResult* aResult;
2842         aResult = dynamic_cast<TSQLResult*> (fServer[3]->Query(sqlQuery));
2843         if (!aResult) {
2844                 Log("SHUTTLE", Form("UpdateShuttleLogbook - Can't execute query <%s>", sqlQuery.Data()));
2845                 return kFALSE;
2846         }
2847         delete aResult;
2848
2849         return kTRUE;
2850 }
2851
2852 //______________________________________________________________________________________________
2853 Int_t AliShuttle::GetCurrentRun() const
2854 {
2855         //
2856         // Get current run from logbook entry
2857         //
2858
2859         return fLogbookEntry ? fLogbookEntry->GetRun() : -1;
2860 }
2861
2862 //______________________________________________________________________________________________
2863 UInt_t AliShuttle::GetCurrentStartTime() const
2864 {
2865         //
2866         // get current start time
2867         //
2868
2869         return fLogbookEntry ? fLogbookEntry->GetStartTime() : 0;
2870 }
2871
2872 //______________________________________________________________________________________________
2873 UInt_t AliShuttle::GetCurrentEndTime() const
2874 {
2875         //
2876         // get current end time from logbook entry
2877         //
2878
2879         return fLogbookEntry ? fLogbookEntry->GetEndTime() : 0;
2880 }
2881 //______________________________________________________________________________________________
2882 UInt_t AliShuttle::GetCurrentYear() const
2883 {
2884         //
2885         // Get current year from logbook entry
2886         //
2887
2888         if (!fLogbookEntry) return 0;
2889         
2890         TTimeStamp startTime(GetCurrentStartTime());
2891         TString year =  Form("%d",startTime.GetDate());
2892         year = year(0,4);
2893         
2894         return year.Atoi();
2895 }
2896
2897 //______________________________________________________________________________________________
2898 const char* AliShuttle::GetLHCPeriod() const
2899 {
2900         //
2901         // Get current LHC period from logbook entry
2902         //
2903
2904         if (!fLogbookEntry) return 0;
2905                 
2906         return fLogbookEntry->GetRunParameter("LHCperiod");
2907 }
2908
2909 //______________________________________________________________________________________________
2910 void AliShuttle::Log(const char* detector, const char* message)
2911 {
2912         //
2913         // Fill log string with a message
2914         //
2915         
2916         TString logRunDir = GetShuttleLogDir();
2917         if (GetCurrentRun() >=0) {
2918                 Int_t logDir_lev1 = GetCurrentRun()/10000;
2919                 logRunDir += Form("/%d/%d", logDir_lev1, GetCurrentRun());
2920         }               
2921         void* dir = gSystem->OpenDirectory(logRunDir.Data());
2922         if (dir == NULL) {
2923                 if (gSystem->mkdir(logRunDir.Data(), kTRUE)) {
2924                         AliError(Form("Can't open directory <%s>", GetShuttleLogDir()));
2925                         return;
2926                 }
2927
2928         } else {
2929                 gSystem->FreeDirectory(dir);
2930         }
2931
2932         TString toLog = Form("%s UTC (%d): %s - ", TTimeStamp(time(0)).AsString("s"), getpid(), detector);
2933         if (GetCurrentRun() >= 0) 
2934                 toLog += Form("run %d - ", GetCurrentRun());
2935         toLog += Form("%s", message);
2936
2937         AliInfo(toLog.Data());
2938         
2939         // if we redirect the log output already to the file, leave here
2940         if (fOutputRedirected && strcmp(detector, "SHUTTLE") != 0)
2941                 return;
2942
2943         TString fileName = GetLogFileName(detector);
2944         
2945         gSystem->ExpandPathName(fileName);
2946
2947         ofstream logFile;
2948         logFile.open(fileName, ofstream::out | ofstream::app);
2949
2950         if (!logFile.is_open()) {
2951                 AliError(Form("Could not open file %s", fileName.Data()));
2952                 return;
2953         }
2954
2955         logFile << toLog.Data() << "\n";
2956
2957         logFile.close();
2958 }
2959
2960 //______________________________________________________________________________________________
2961 TString AliShuttle::GetLogFileName(const char* detector) const
2962 {
2963         // 
2964         // returns the name of the log file for a given sub detector
2965         //
2966         
2967         TString fileName;
2968         
2969         if (GetCurrentRun() >= 0) 
2970         {
2971                 Int_t logDir_lev1 = GetCurrentRun()/10000;
2972                 fileName.Form("%s/%d/%d/%s.log", GetShuttleLogDir(), logDir_lev1, GetCurrentRun(), 
2973                         detector);
2974         } else {
2975                 fileName.Form("%s/%s.log", GetShuttleLogDir(), detector);
2976         }
2977
2978         return fileName;
2979 }
2980
2981 //______________________________________________________________________________________________
2982 void AliShuttle::SendAlive()
2983 {
2984         // sends alive message to ML
2985         
2986         TMonaLisaText mlStatus("SHUTTLE_status", "Alive");
2987
2988         TList mlList;
2989         mlList.Add(&mlStatus);
2990
2991         fMonaLisa->SendParameters(&mlList, "__PROCESSINGINFO__");
2992 }
2993
2994 //______________________________________________________________________________________________
2995 Bool_t AliShuttle::Collect(Int_t run)
2996 {
2997         //
2998         // Collects conditions data for all UNPROCESSED run written to DAQ LogBook in case of run = -1 (default)
2999         // If a dedicated run is given this run is processed
3000         //
3001         // In operational mode, this is the Shuttle function triggered by the EOR signal.
3002         //
3003
3004         if (run == -1)
3005                 Log("SHUTTLE","Collect - Shuttle called. Collecting conditions data for unprocessed runs");
3006         else
3007                 Log("SHUTTLE", Form("Collect - Shuttle called. Collecting conditions data for run %d", run));
3008
3009         SetLastAction("Starting");
3010
3011         // create ML instance
3012         if (!fMonaLisa)
3013                 fMonaLisa = new TMonaLisaWriter(fConfig->GetMonitorHost(), fConfig->GetMonitorTable());
3014                 
3015         CountOpenRuns();
3016
3017         TString whereClause("where shuttle_done=0");
3018         if (run != -1)
3019                 whereClause += Form(" and run=%d", run);
3020
3021         TObjArray shuttleLogbookEntries;
3022         if (!QueryShuttleLogbook(whereClause, shuttleLogbookEntries))
3023         {
3024                 Log("SHUTTLE", "Collect - Can't retrieve entries from Shuttle logbook");
3025                 return kFALSE;
3026         }
3027
3028         if (shuttleLogbookEntries.GetEntries() == 0)
3029         {
3030                 if (run == -1)
3031                         Log("SHUTTLE","Collect - Found no UNPROCESSED runs in Shuttle logbook");
3032                 else
3033                         Log("SHUTTLE", Form("Collect - Run %d is already DONE "
3034                                                 "or it does not exist in Shuttle logbook", run));
3035                 return kTRUE;
3036         }
3037
3038         for (UInt_t iDet=0; iDet<NDetectors(); iDet++)
3039                 fFirstUnprocessed[iDet] = kTRUE;
3040
3041         if (run != -1)
3042         {
3043                 // query Shuttle logbook for earlier runs, check if some detectors are unprocessed,
3044                 // flag them into fFirstUnprocessed array
3045                 TString whereClauseBis(Form("where shuttle_done=0 and run < %d", run));
3046                 TObjArray tmpLogbookEntries;
3047                 if (!QueryShuttleLogbook(whereClauseBis, tmpLogbookEntries))
3048                 {
3049                         Log("SHUTTLE", "Collect - Can't retrieve entries from Shuttle logbook");
3050                         return kFALSE;
3051                 }
3052
3053                 TIter iter(&tmpLogbookEntries);
3054                 AliShuttleLogbookEntry* anEntry = 0;
3055                 while ((anEntry = dynamic_cast<AliShuttleLogbookEntry*> (iter.Next())))
3056                 {
3057                         for (UInt_t iDet=0; iDet<NDetectors(); iDet++)
3058                         {
3059                                 if (anEntry->GetDetectorStatus(iDet) == AliShuttleLogbookEntry::kUnprocessed)
3060                                 {
3061                                         AliDebug(2, Form("Run %d: setting %s as \"not first time unprocessed\"",
3062                                                         anEntry->GetRun(), GetDetName(iDet)));
3063                                         fFirstUnprocessed[iDet] = kFALSE;
3064                                 }
3065                         }
3066
3067                 }
3068
3069         }
3070
3071         if (!RetrieveConditionsData(shuttleLogbookEntries))
3072         {
3073                 Log("SHUTTLE", "Collect - Process of at least one run failed");
3074                 CountOpenRuns();
3075                 return kFALSE;
3076         }
3077
3078         Log("SHUTTLE", "Collect - Requested run(s) successfully processed");
3079         CountOpenRuns();
3080         return kTRUE;
3081 }
3082
3083 //______________________________________________________________________________________________
3084 Bool_t AliShuttle::RetrieveConditionsData(const TObjArray& dateEntries)
3085 {
3086         //
3087         // Retrieve conditions data for all runs that aren't processed yet
3088         //
3089
3090         Bool_t hasError = kFALSE;
3091
3092         TIter iter(&dateEntries);
3093         AliShuttleLogbookEntry* anEntry;
3094
3095         while ((anEntry = (AliShuttleLogbookEntry*) iter.Next())){
3096                 if (!Process(anEntry)){
3097                         hasError = kTRUE;
3098                 }
3099
3100                 // clean SHUTTLE temp directory
3101                 //TString filename = Form("%s/*.shuttle", GetShuttleTempDir());
3102                 //RemoveFile(filename.Data());
3103         }
3104
3105         return hasError == kFALSE;
3106 }
3107
3108 //______________________________________________________________________________________________
3109 ULong_t AliShuttle::GetTimeOfLastAction() const
3110 {
3111         //
3112         // Gets time of last action
3113         //
3114
3115         ULong_t tmp;
3116
3117         fMonitoringMutex->Lock();
3118
3119         tmp = fLastActionTime;
3120
3121         fMonitoringMutex->UnLock();
3122
3123         return tmp;
3124 }
3125
3126 //______________________________________________________________________________________________
3127 const TString AliShuttle::GetLastAction() const
3128 {
3129         //
3130         // returns a string description of the last action
3131         //
3132
3133         TString tmp;
3134
3135         fMonitoringMutex->Lock();
3136         
3137         tmp = fLastAction;
3138         
3139         fMonitoringMutex->UnLock();
3140
3141         return tmp;
3142 }
3143
3144 //______________________________________________________________________________________________
3145 void AliShuttle::SetLastAction(const char* action)
3146 {
3147         //
3148         // updates the monitoring variables
3149         //
3150
3151         fMonitoringMutex->Lock();
3152
3153         fLastAction = action;
3154         fLastActionTime = time(0);
3155         
3156         fMonitoringMutex->UnLock();
3157 }
3158
3159 //______________________________________________________________________________________________
3160 const char* AliShuttle::GetRunParameter(const char* param)
3161 {
3162         //
3163         // returns run parameter read from DAQ logbook
3164         //
3165
3166         if(!fLogbookEntry) {
3167                 AliError("No logbook entry!");
3168                 return 0;
3169         }
3170
3171         return fLogbookEntry->GetRunParameter(param);
3172 }
3173
3174 //______________________________________________________________________________________________
3175 AliCDBEntry* AliShuttle::GetFromOCDB(const char* detector, const AliCDBPath& path)
3176 {
3177         //
3178         // returns object from OCDB valid for current run
3179         //
3180
3181         if (fTestMode & kErrorOCDB)
3182         {
3183                 Log(detector, "GetFromOCDB - In TESTMODE - Simulating error with OCDB");
3184                 return 0;
3185         }
3186         
3187         AliCDBStorage *sto = AliCDBManager::Instance()->GetStorage(fgkMainCDB);
3188         if (!sto)
3189         {
3190                 Log(detector, "GetFromOCDB - Cannot activate main OCDB for query!");
3191                 return 0;
3192         }
3193
3194         return dynamic_cast<AliCDBEntry*> (sto->Get(path, GetCurrentRun()));
3195 }
3196
3197 //______________________________________________________________________________________________
3198 Bool_t AliShuttle::SendMail(EMailTarget target, Int_t system)
3199 {
3200         //
3201         // sends a mail to the subdetector expert in case of preprocessor error
3202         //
3203         
3204         if (fTestMode != kNone)
3205                 return kTRUE;
3206                 
3207         if (!fConfig->SendMail()) 
3208                 return kTRUE;
3209
3210         if (target == kDCSEMail || target == kFXSEMail) {
3211                 if (!fFirstProcessing)
3212                         return kTRUE;
3213         }
3214
3215         Int_t runMode = (Int_t)fConfig->GetRunMode();
3216         TString tmpStr;
3217         if (runMode == 0) tmpStr = " Nightly Test:";
3218         else tmpStr = " Data Taking:"; 
3219         void* dir = gSystem->OpenDirectory(GetShuttleLogDir());
3220         if (dir == NULL)
3221         {
3222                 if (gSystem->mkdir(GetShuttleLogDir(), kTRUE))
3223                 {
3224                         Log("SHUTTLE", Form("SendMail - Can't open directory <%s>", GetShuttleLogDir()));
3225                         return kFALSE;
3226                 }
3227
3228         } else {
3229                 gSystem->FreeDirectory(dir);
3230         }
3231
3232         // det experts in to
3233         TString to="";
3234         TIter *iterExperts = 0;
3235         if (target == kDCSEMail) {
3236                 iterExperts = new TIter(fConfig->GetAdmins(AliShuttleConfig::kAmanda));
3237         }
3238         else if (target == kFXSEMail) {
3239                 iterExperts = new TIter(fConfig->GetAdmins(system));
3240         }
3241         if (iterExperts) {
3242                 TObjString *anExpert=0;
3243                 while ((anExpert = (TObjString*) iterExperts->Next()))
3244                 {
3245                         to += Form("%s,", anExpert->GetName());
3246                 }
3247                 delete iterExperts;
3248         }
3249
3250         // add subdetector experts      
3251         iterExperts = new TIter(fConfig->GetResponsibles(fCurrentDetector));
3252         TObjString *anExpert=0;
3253         while ((anExpert = (TObjString*) iterExperts->Next()))
3254         {
3255                 to += Form("%s,", anExpert->GetName());
3256         }
3257         delete iterExperts;
3258         
3259         if (to.Length() > 0)
3260           to.Remove(to.Length()-1);
3261         AliDebug(2, Form("to: %s",to.Data()));
3262
3263         if (to.IsNull()) {
3264                 Log("SHUTTLE", Form("List of %d responsibles not set!", (Int_t) target));
3265                 return kFALSE;
3266         }
3267
3268         // SHUTTLE responsibles in cc
3269         TString cc="";
3270         TIter iterAdmins(fConfig->GetAdmins(AliShuttleConfig::kGlobal));
3271         TObjString *anAdmin=0;
3272         while ((anAdmin = (TObjString*) iterAdmins.Next()))
3273         {
3274                 cc += Form("%s,", anAdmin->GetName());
3275         }
3276         if (cc.Length() > 0)
3277           cc.Remove(cc.Length()-1);
3278         AliDebug(2, Form("cc: %s",to.Data()));
3279
3280         // mail body 
3281         TString bodyFileName;
3282         bodyFileName.Form("%s/mail.body", GetShuttleLogDir());
3283         gSystem->ExpandPathName(bodyFileName);
3284
3285         ofstream mailBody;
3286         mailBody.open(bodyFileName, ofstream::out);
3287
3288         if (!mailBody.is_open())
3289         {
3290                 Log("SHUTTLE", Form("Could not open mail body file %s", bodyFileName.Data()));
3291                 return kFALSE;
3292         }
3293
3294
3295         TString subject;
3296         TString body;
3297
3298         if (target == kDCSEMail){
3299                 subject = Form("%s CRITICAL Retrieval of data points for %s FAILED in run %d !",
3300                                 tmpStr.Data(), fCurrentDetector.Data(), GetCurrentRun());
3301                 AliDebug(2, Form("subject: %s", subject.Data()));
3302                 
3303                 body = Form("Dear DCS experts, \n\n");
3304                 body += Form("SHUTTLE couldn\'t retrieve the data points for detector %s "
3305                              "in run %d!!\n\n", fCurrentDetector.Data(), GetCurrentRun());
3306         }
3307         else if (target == kFXSEMail){
3308                 subject = Form("%s CRITICAL FXS communication for %s FAILED in run %d !",
3309                                 tmpStr.Data(), fCurrentDetector.Data(), GetCurrentRun());
3310                 AliDebug(2, Form("subject: %s", subject.Data()));
3311                 TString sys;
3312                 if (system == kDAQ) sys="DAQ";
3313                 else if (system == kDCS) sys="DCS";
3314                 else if (system == kHLT) sys="HLT";
3315                 else return kFALSE;
3316                 body = Form("Dear  %s FXS experts, \n\n",sys.Data());
3317                 body += Form("SHUTTLE couldn\'t retrieve data from the FXS for detector %s "
3318                              "in run %d!!\n\n", fCurrentDetector.Data(), GetCurrentRun());
3319                 body += Form("The contacted server was:\nDB: %s\nFXS:%s\n\n", fConfig->GetFXSdbHost(system), fConfig->GetFXSHost(system));
3320         }
3321         else {
3322                 subject = Form("%s %s Shuttle preprocessor FAILED in run %d (run type = %s)!",
3323                                        tmpStr.Data(), fCurrentDetector.Data(), GetCurrentRun(), GetRunType());
3324                 AliDebug(2, Form("subject: %s", subject.Data()));
3325         
3326                 body = Form("Dear %s expert(s), \n\n", fCurrentDetector.Data());
3327                 body += Form("SHUTTLE just detected that your preprocessor "
3328                              "failed processing run %d (run type = %s)!!\n\n", 
3329                              GetCurrentRun(), GetRunType());
3330         }
3331
3332         body += Form("Please check %s status on the SHUTTLE monitoring page: \n\n", 
3333                                 fCurrentDetector.Data());
3334         if (fConfig->GetRunMode() == AliShuttleConfig::kTest)
3335         {
3336                 body += Form("\thttp://pcalimonitor.cern.ch/shuttle.jsp?time=24 \n\n");
3337         } else {
3338                 body += Form("\thttp://pcalimonitor.cern.ch/shuttle.jsp?instance=PROD&time=24 \n\n");
3339         }
3340         
3341         
3342         TString logFolder = "logs";
3343         if (fConfig->GetRunMode() == AliShuttleConfig::kProd) 
3344                 logFolder += "_PROD";
3345         
3346         
3347         body += Form("Find the %s log for the current run on \n\n"
3348                 "\thttp://pcalishuttle02.cern.ch/%s/%d/%d/%s.log \n\n", 
3349                      fCurrentDetector.Data(), logFolder.Data(), GetCurrentRun()/10000,  
3350                                 GetCurrentRun(), fCurrentDetector.Data());
3351         body += Form("The last 15 lines of %s log file are following:\n\n", fCurrentDetector.Data());
3352
3353         AliDebug(2, Form("Body begin: %s", body.Data()));
3354
3355         mailBody << body.Data();
3356         mailBody.close();
3357         mailBody.open(bodyFileName, ofstream::out | ofstream::app);
3358
3359         TString logFileName = Form("%s/%d/%d/%s.log", GetShuttleLogDir(), 
3360                 GetCurrentRun()/10000, GetCurrentRun(), fCurrentDetector.Data());
3361         TString tailCommand = Form("tail -n 15 %s >> %s", logFileName.Data(), bodyFileName.Data());
3362         if (gSystem->Exec(tailCommand.Data()))
3363         {
3364                 mailBody << Form("%s log file not found ...\n\n", fCurrentDetector.Data());
3365         }
3366
3367         TString endBody = Form("------------------------------------------------------\n\n");
3368         endBody += Form("In case of problems please contact the SHUTTLE core team.\n\n");
3369         endBody += "Please do not answer this message directly, it is automatically generated.\n\n";
3370         endBody += "Greetings,\n\n \t\t\tthe SHUTTLE\n";
3371
3372         AliDebug(2, Form("Body end: %s", endBody.Data()));
3373
3374         mailBody << endBody.Data();
3375
3376         mailBody.close();
3377
3378         // send mail!
3379         TString mailCommand = Form("mail -s \"%s\" -c %s %s < %s",
3380                                                 subject.Data(),
3381                                                 cc.Data(),
3382                                                 to.Data(),
3383                                                 bodyFileName.Data());
3384         AliDebug(2, Form("mail command: %s", mailCommand.Data()));
3385
3386         Bool_t result = gSystem->Exec(mailCommand.Data());
3387
3388         return result == 0;
3389 }
3390 //______________________________________________________________________________________________
3391 const char* AliShuttle::GetRunType()
3392 {
3393         //
3394         // returns run type read from "run type" logbook
3395         //
3396
3397         if(!fLogbookEntry) {
3398                 AliError("No logbook entry!");
3399                 return 0;
3400         }
3401
3402         return fLogbookEntry->GetRunType();
3403 }
3404
3405 //______________________________________________________________________________________________
3406 Bool_t AliShuttle::GetHLTStatus()
3407 {
3408         // Return HLT status (ON=1 OFF=0)
3409         // Converts the HLT status from the mode string read in the run logbook (not just a bool)
3410
3411         if(!fLogbookEntry) {
3412                 AliError("No logbook entry!");
3413                 return 0;
3414         }
3415
3416         // TODO implement when HLTMode is inserted in run logbook
3417         TString hltMode = fLogbookEntry->GetRunParameter("HLTmode");
3418         TSubString firstChar = hltMode(0,1);
3419         AliDebug(2,Form("First char = %s ",firstChar.Data())); 
3420         if (firstChar == "A") {
3421                 return kFALSE;
3422         }
3423         else if ((firstChar == "B") || (firstChar == "C") || (firstChar == "D") || (firstChar == "E")) {
3424                 return kTRUE;
3425         }
3426         else {
3427                 Log("SHUTTLE","Unexpected HLT mode! Returning 0....");
3428                 return kFALSE;
3429         }
3430 }
3431
3432 //______________________________________________________________________________________________
3433 const char* AliShuttle::GetTriggerConfiguration()
3434 {
3435         // Receives the trigger configuration from the DAQ logbook for the current run
3436         
3437         // check connection, if needed reconnect
3438         if (!Connect(3)) 
3439                 return 0;
3440
3441         TString sqlQuery;
3442         sqlQuery.Form("SELECT configFile FROM logbook_trigger_config WHERE run = %d", GetCurrentRun());
3443         TSQLResult* result = fServer[3]->Query(sqlQuery);
3444         if (!result)
3445         {
3446                 Log("SHUTTLE", Form("ERROR: Can't execute query <%s>!", sqlQuery.Data()));
3447                 return 0;
3448         }
3449         
3450         if (result->GetRowCount() == 0)
3451         {
3452                 Log("SHUTTLE", "ERROR: Trigger configuration not found in logbook_trigger_config");
3453                 delete result;
3454                 return 0;
3455         }
3456         
3457         TSQLRow* row = result->Next();
3458         if (!row)
3459         {
3460                 Log("SHUTTLE", "ERROR: Could not receive logbook_trigger_config data");
3461                 delete result;
3462                 return 0;
3463         }
3464
3465         // static, so that pointer remains valid when it is returned to the calling class       
3466         static TString triggerConfig(row->GetField(0));
3467         
3468         delete row;
3469         row = 0;
3470         
3471         delete result;
3472         result = 0;
3473         
3474         Log("SHUTTLE", Form("Found trigger configuration: %s", triggerConfig.Data()));
3475         
3476         return triggerConfig;
3477 }
3478
3479 //______________________________________________________________________________________________
3480 void AliShuttle::SetShuttleTempDir(const char* tmpDir)
3481 {
3482         //
3483         // sets Shuttle temp directory
3484         //
3485
3486         fgkShuttleTempDir = gSystem->ExpandPathName(tmpDir);
3487 }
3488
3489 //______________________________________________________________________________________________
3490 void AliShuttle::SetShuttleLogDir(const char* logDir)
3491 {
3492         //
3493         // sets Shuttle log directory
3494         //
3495
3496         fgkShuttleLogDir = gSystem->ExpandPathName(logDir);
3497 }
3498 //______________________________________________________________________________________________
3499 Bool_t AliShuttle::TouchFile()
3500 {
3501         //
3502         // touching a file on the grid if run has been DONE
3503         //
3504         
3505         if (!gGrid)
3506         {
3507                 Log("SHUTTLE",Form("No TGrid connection estabilished!"));
3508                 Log("SHUTTLE",Form("Could not touch file for run %i",GetCurrentRun()));
3509                 return kFALSE;
3510         }
3511
3512         TString dir;
3513         dir.Form("%s%d/SHUTTLE_DONE", fConfig->GetAlienPath(), GetCurrentYear());
3514         // checking whether directory for touch command exists
3515         TString commandLs;