Changing pinging to FXSs. Manual reconnection in case connection gets lost.
[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         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
886 //______________________________________________________________________________________________
887 void AliShuttle::CleanLocalStorage(const TString& uri)
888 {
889         //
890         // Called in case the preprocessor is declared failed. Remove remaining objects from the local storages.
891         //
892
893         const char* type = 0;
894         if(uri == fgkLocalCDB) {
895                 type = "OCDB";
896         } else if(uri == fgkLocalRefStorage) {
897                 type = "Reference";
898         } else {
899                 AliError(Form("Invalid storage URI: %s", uri.Data()));
900                 return;
901         }
902
903         AliCDBManager* man = AliCDBManager::Instance();
904
905         // open local storage
906         AliCDBStorage *localSto = man->GetStorage(uri);
907         if(!localSto) {
908                 Log("SHUTTLE",
909                         Form("CleanLocalStorage - cannot activate local %s storage", type));
910                 return;
911         }
912
913         TString filename(Form("%s/%s/*/Run*_v%d_s*.root",
914                 localSto->GetBaseFolder().Data(), GetOfflineDetName(fCurrentDetector.Data()), GetCurrentRun()));
915
916         AliDebug(2, Form("filename = %s", filename.Data()));
917
918         Log("SHUTTLE", Form("Removing remaining local files for run %d and detector %s ...",
919                 GetCurrentRun(), fCurrentDetector.Data()));
920
921         RemoveFile(filename.Data());
922
923 }
924
925 //______________________________________________________________________________________________
926 void AliShuttle::RemoveFile(const char* filename)
927 {
928         //
929         // removes local file
930         //
931
932         TString command(Form("rm -f %s", filename));
933
934         Int_t result = gSystem->Exec(command.Data());
935         if(result != 0)
936         {
937                 Log("SHUTTLE", Form("RemoveFile - %s: Cannot remove file %s!",
938                         fCurrentDetector.Data(), filename));
939         }
940 }
941
942 //______________________________________________________________________________________________
943 AliShuttleStatus* AliShuttle::ReadShuttleStatus()
944 {
945         //
946         // Reads the AliShuttleStatus from the CDB
947         //
948
949         if (fStatusEntry){
950                 delete fStatusEntry;
951                 fStatusEntry = 0;
952         }
953
954         Int_t path1 = GetCurrentRun()/10000;
955         fStatusEntry = AliCDBManager::Instance()->GetStorage(GetLocalCDB())
956                 ->Get(Form("/SHUTTLE/%s/%d", fCurrentDetector.Data(), path1), GetCurrentRun());
957
958         if (!fStatusEntry) return 0;
959         fStatusEntry->SetOwner(1);
960
961         AliShuttleStatus* status = dynamic_cast<AliShuttleStatus*> (fStatusEntry->GetObject());
962         if (!status) {
963                 AliError("Invalid object stored to CDB!");
964                 return 0;
965         }
966
967         return status;
968 }
969
970 //______________________________________________________________________________________________
971 Bool_t AliShuttle::WriteShuttleStatus(AliShuttleStatus* status)
972 {
973         //
974         // writes the status for one subdetector
975         //
976
977         if (fStatusEntry){
978                 delete fStatusEntry;
979                 fStatusEntry = 0;
980         }
981
982         Int_t run = GetCurrentRun();
983         Int_t path1 = run/10000;
984         TString path1_string = Form("%d",path1);
985
986         AliCDBId id(AliCDBPath("SHUTTLE", fCurrentDetector, path1_string), run, run);
987
988         fStatusEntry = new AliCDBEntry(status, id, new AliCDBMetaData);
989         fStatusEntry->SetOwner(1);
990
991         Int_t logLevel = AliLog::GetGlobalLogLevel();
992         AliLog::SetGlobalLogLevel(AliLog::kError);
993
994         UInt_t result = AliCDBManager::Instance()->GetStorage(fgkLocalCDB)->Put(fStatusEntry);
995
996         AliLog::SetGlobalLogLevel((AliLog::EType_t)logLevel);
997
998         if (!result) {
999                 Log("SHUTTLE", Form("WriteShuttleStatus - Failed for %s, run %d",
1000                                                 fCurrentDetector.Data(), run));
1001                 return kFALSE;
1002         }
1003         
1004         SendMLDetInfo();
1005
1006         return kTRUE;
1007 }
1008
1009 //______________________________________________________________________________________________
1010 void AliShuttle::UpdateShuttleStatus(AliShuttleStatus::Status newStatus, Bool_t increaseCount)
1011 {
1012         //
1013         // changes the AliShuttleStatus for the given detector and run to the given status
1014         //
1015
1016         if (!fStatusEntry){
1017                 AliError("UNEXPECTED: fStatusEntry empty");
1018                 return;
1019         }
1020
1021         AliShuttleStatus* status = dynamic_cast<AliShuttleStatus*> (fStatusEntry->GetObject());
1022
1023         if (!status){
1024                 Log("SHUTTLE", "UpdateShuttleStatus - UNEXPECTED: status could not be read from current CDB entry");
1025                 return;
1026         }
1027
1028         TString actionStr = Form("UpdateShuttleStatus - %s: Changing state from %s to %s",
1029                                 fCurrentDetector.Data(),
1030                                 status->GetStatusName(),
1031                                 status->GetStatusName(newStatus));
1032         Log("SHUTTLE", actionStr);
1033         SetLastAction(actionStr);
1034
1035         status->SetStatus(newStatus);
1036         if (increaseCount) status->IncreaseCount();
1037
1038         Int_t logLevel = AliLog::GetGlobalLogLevel();
1039         AliLog::SetGlobalLogLevel(AliLog::kError);
1040
1041         AliCDBManager::Instance()->GetStorage(fgkLocalCDB)->Put(fStatusEntry);
1042
1043         AliLog::SetGlobalLogLevel((AliLog::EType_t)logLevel);
1044
1045         SendMLDetInfo();
1046 }
1047
1048 //______________________________________________________________________________________________
1049 void AliShuttle::SendMLDetInfo()
1050 {
1051         //
1052         // sends ML information about the current status of the current detector being processed
1053         //
1054         
1055         AliShuttleStatus* status = dynamic_cast<AliShuttleStatus*> (fStatusEntry->GetObject());
1056         
1057         if (!status){
1058                 Log("SHUTTLE", "SendMLDetInfo - UNEXPECTED: status could not be read from current CDB entry");
1059                 return;
1060         }
1061         
1062         TMonaLisaText  mlStatus(Form("%s_status", fCurrentDetector.Data()), status->GetStatusName());
1063         TMonaLisaValue mlRetryCount(Form("%s_count", fCurrentDetector.Data()), status->GetCount());
1064
1065         TList mlList;
1066         mlList.Add(&mlStatus);
1067         mlList.Add(&mlRetryCount);
1068
1069         TString mlID;
1070         mlID.Form("%d", GetCurrentRun());
1071         fMonaLisa->SendParameters(&mlList, mlID);
1072 }
1073
1074 //______________________________________________________________________________________________
1075 Bool_t AliShuttle::ContinueProcessing()
1076 {
1077         // this function reads the AliShuttleStatus information from CDB and
1078         // checks if the processing should be continued
1079         // if yes it returns kTRUE and updates the AliShuttleStatus with nextStatus
1080
1081         if (!fConfig->HostProcessDetector(fCurrentDetector))
1082                 return kFALSE;
1083
1084         AliPreprocessor* aPreprocessor =
1085                 dynamic_cast<AliPreprocessor*> (fPreprocessorMap.GetValue(fCurrentDetector));
1086         if (!aPreprocessor)
1087         {
1088                 Log("SHUTTLE", Form("ContinueProcessing - %s: no preprocessor registered", fCurrentDetector.Data()));
1089                 return kFALSE;
1090         }
1091
1092         AliShuttleLogbookEntry::Status entryStatus =
1093                 fLogbookEntry->GetDetectorStatus(fCurrentDetector);
1094
1095         if (entryStatus != AliShuttleLogbookEntry::kUnprocessed) {
1096                 Log("SHUTTLE", Form("ContinueProcessing - %s is %s",
1097                                 fCurrentDetector.Data(),
1098                                 fLogbookEntry->GetDetectorStatusName(entryStatus)));
1099                 return kFALSE;
1100         }
1101
1102         // if we get here, according to Shuttle logbook subdetector is in UNPROCESSED state
1103
1104         // check if current run is first unprocessed run for current detector
1105         if (fConfig->StrictRunOrder(fCurrentDetector) &&
1106                 !fFirstUnprocessed[GetDetPos(fCurrentDetector)])
1107         {
1108                 if (fTestMode == kNone)
1109                 {
1110                         Log("SHUTTLE", Form("ContinueProcessing - %s requires strict run ordering"
1111                                         " but this is not the first unprocessed run!"));
1112                         return kFALSE;
1113                 }
1114                 else
1115                 {
1116                         Log("SHUTTLE", Form("ContinueProcessing - In TESTMODE - "
1117                                         "Although %s requires strict run ordering "
1118                                         "and this is not the first unprocessed run, "
1119                                         "the SHUTTLE continues"));
1120                 }
1121         }
1122
1123         // Is the subdetector processed first time for this run?
1124         fFirstProcessing = kFALSE;
1125
1126         AliShuttleStatus* status = ReadShuttleStatus();
1127         if (!status) {
1128                 // first time
1129                 Log("SHUTTLE", Form("ContinueProcessing - %s: Processing first time",
1130                                 fCurrentDetector.Data()));
1131                 status = new AliShuttleStatus(AliShuttleStatus::kStarted);
1132                 fFirstProcessing = kTRUE;
1133                 return WriteShuttleStatus(status);
1134         }
1135
1136         // The following case shouldn't happen if Shuttle Logbook was correctly updated.
1137         // If it happens it may mean Logbook updating failed... let's do it now!
1138         if (status->GetStatus() == AliShuttleStatus::kDone ||
1139             status->GetStatus() == AliShuttleStatus::kFailed ||
1140             status->GetStatus() == AliShuttleStatus::kSkipped) {
1141                 Log("SHUTTLE", Form("ContinueProcessing - %s is already %s. Updating Shuttle Logbook",
1142                                         fCurrentDetector.Data(),
1143                                         status->GetStatusName(status->GetStatus())));
1144                 
1145                 if (status->GetStatus() == AliShuttleStatus::kSkipped)
1146                 {
1147                         UpdateShuttleLogbook(fCurrentDetector.Data(), "DONE");
1148                 }
1149                 else
1150                         UpdateShuttleLogbook(fCurrentDetector.Data(), status->GetStatusName(status->GetStatus()));
1151                         
1152                 return kFALSE;
1153         }
1154
1155         if (status->GetStatus() == AliShuttleStatus::kStoreStarted || status->GetStatus() == AliShuttleStatus::kStoreDelayed ||status->GetStatus() == AliShuttleStatus::kStoreError) {
1156                 Log("SHUTTLE",
1157                         Form("ContinueProcessing - %s: Grid storage of one or more "
1158                                 "objects failed. Trying again now",
1159                                 fCurrentDetector.Data()));
1160                 StoreOCDB();
1161                 return kFALSE;
1162         }
1163
1164         // if we get here, there is a restart
1165         Bool_t cont = kFALSE;
1166
1167         // abort conditions
1168         if (status->GetCount() >= fConfig->GetMaxRetries()) {
1169                 Log("SHUTTLE", Form("ContinueProcessing - %s failed %d times in status %s - "
1170                                 "Updating Shuttle Logbook", fCurrentDetector.Data(),
1171                                 status->GetCount(), status->GetStatusName()));
1172                 UpdateShuttleLogbook(fCurrentDetector.Data(), "FAILED");
1173                 UpdateShuttleStatus(AliShuttleStatus::kFailed);
1174
1175                 // there may still be objects in local OCDB and reference storage
1176                 // and FXS databases may be not updated: do it now!
1177                 
1178                 // TODO Currently disabled, we want to keep files in case of failure!
1179                 // CleanLocalStorage(fgkLocalCDB);
1180                 // CleanLocalStorage(fgkLocalRefStorage);
1181                 // UpdateTableFailCase();
1182                 
1183                 // Send mail to detector expert!
1184                 Log("SHUTTLE", Form("ContinueProcessing - Sending mail to %s expert...", 
1185                                     fCurrentDetector.Data()));
1186                 // det experts in to
1187                 TString to="";
1188                 TIter *iterExperts = 0;
1189                 iterExperts = new TIter(fConfig->GetResponsibles(fCurrentDetector));
1190                 TObjString *anExpert=0;
1191                 while ((anExpert = (TObjString*) iterExperts->Next()))
1192                         {
1193                                 to += Form("%s, \n", anExpert->GetName());
1194                         }
1195                 delete iterExperts;
1196                 
1197                 if (to.Length() > 0)
1198                         to.Remove(to.Length()-3);
1199                 AliDebug(2, Form("to: %s",to.Data()));
1200
1201                 if (to.IsNull()) {
1202                         Log("SHUTTLE", Form("List of %s responsibles not set!", fCurrentDetector.Data()));
1203                         return kFALSE;
1204                 }
1205
1206                 Log(fCurrentDetector.Data(), Form("ContinueProcessing - Sending mail to %s expert(s):", 
1207                                     fCurrentDetector.Data()));
1208                 Log(fCurrentDetector.Data(), Form("\n%s", to.Data()));
1209                 if (!SendMail(kPPEMail))
1210                         Log("SHUTTLE", Form("ContinueProcessing - Could not send mail to %s expert",
1211                                             fCurrentDetector.Data()));
1212
1213         } else {
1214                 Log("SHUTTLE", Form("ContinueProcessing - %s: restarting. "
1215                                 "Aborted before with %s. Retry number %d.", fCurrentDetector.Data(),
1216                                 status->GetStatusName(), status->GetCount()));
1217                 Bool_t increaseCount = kTRUE;
1218                 if (status->GetStatus() == AliShuttleStatus::kDCSError || 
1219                     status->GetStatus() == AliShuttleStatus::kDCSStarted ||
1220                     status->GetStatus() == AliShuttleStatus::kFXSError)
1221                                 increaseCount = kFALSE;
1222                                 
1223                 UpdateShuttleStatus(AliShuttleStatus::kStarted, increaseCount);
1224                 cont = kTRUE;
1225         }
1226
1227         return cont;
1228 }
1229
1230 //______________________________________________________________________________________________
1231 void AliShuttle::SendMLRunInfo(const char* status)
1232 {
1233         // 
1234         // Send information about this run to ML
1235         
1236         TMonaLisaText  mlStatus("SHUTTLE_status", status);
1237         TString runType(fLogbookEntry->GetRunType());
1238         if (strlen(fLogbookEntry->GetRunParameter("log")) > 0){
1239
1240                 runType += "(";
1241                 runType += fLogbookEntry->GetRunParameter("log");
1242                 runType += ")";
1243         }
1244         if (fLogbookEntry->GetDATestMode()){
1245                 runType += " (DATest)";
1246         }
1247         TMonaLisaText  mlRunType("SHUTTLE_runtype", runType);
1248
1249         TList mlList;
1250         mlList.Add(&mlStatus);
1251         mlList.Add(&mlRunType);
1252
1253         TString mlID;
1254         mlID.Form("%d", GetCurrentRun());
1255         fMonaLisa->SendParameters(&mlList, mlID);       
1256 }
1257
1258 //______________________________________________________________________________________________
1259 Int_t AliShuttle::GetMem(Int_t pid)
1260 {
1261         // invokes ps to get the memory consumption of the process <pid>
1262         // returns -1 in case of error
1263         
1264         TString checkStr;
1265         checkStr.Form("ps -o vsize --pid %d | tail -n 1", pid);
1266         FILE* pipe = gSystem->OpenPipe(checkStr, "r");
1267         if (!pipe)
1268         {
1269                 Log("SHUTTLE", Form("Process - Error: "
1270                         "Could not open pipe to %s", checkStr.Data()));
1271                 return -1;
1272         }
1273                 
1274         char buffer[100];
1275         if (!fgets(buffer, 100, pipe))
1276         {
1277                 Log("SHUTTLE", "Process - Error: ps did not return anything");
1278                 gSystem->ClosePipe(pipe);
1279                 return -1;
1280         }
1281         gSystem->ClosePipe(pipe);
1282         
1283         //Log("SHUTTLE", Form("ps returned %s", buffer));
1284         
1285         Int_t mem = 0;
1286         if ((sscanf(buffer, "%d\n", &mem) != 1) || !mem)
1287         {
1288                 Log("SHUTTLE", "Process - Error: Could not parse output of ps");
1289                 return -1;
1290         }
1291         
1292         return mem;
1293 }
1294
1295 //______________________________________________________________________________________________
1296 Bool_t AliShuttle::Process(AliShuttleLogbookEntry* entry)
1297 {
1298         //
1299         // Makes data retrieval for all detectors in the configuration.
1300         // entry: Shuttle logbook entry, contains run paramenters and status of detectors
1301         // (Unprocessed, Inactive, Failed or Done).
1302         // Returns kFALSE in case of error occured and kTRUE otherwise
1303         //
1304
1305         if (!entry) return kFALSE;
1306
1307         fLogbookEntry = entry;
1308
1309         Log("SHUTTLE", Form("\t\t\t^*^*^*^*^*^*^*^*^*^*^*^* run %d: START ^*^*^*^*^*^*^*^*^*^*^*^*",
1310                                         GetCurrentRun()));
1311
1312         CountOpenRuns();
1313         
1314         // Send the information to ML
1315         SendMLRunInfo("Processing");
1316
1317         if (fLogbookEntry->IsDone())
1318         {
1319                 Log("SHUTTLE","Process - Shuttle is already DONE. Updating logbook");
1320                 UpdateShuttleLogbook("shuttle_done");
1321                 fLogbookEntry = 0;
1322                 return kTRUE;
1323         }
1324
1325         // read test mode if flag is set
1326         if (fReadTestMode)
1327         {
1328                 fTestMode = kNone;
1329                 TString logEntry(entry->GetRunParameter("log"));
1330                 //printf("log entry = %s\n", logEntry.Data());
1331                 TString searchStr("Testmode: ");
1332                 Int_t pos = logEntry.Index(searchStr.Data());
1333                 //printf("%d\n", pos);
1334                 if (pos >= 0)
1335                 {
1336                         TSubString subStr = logEntry(pos + searchStr.Length(), logEntry.Length());
1337                         //printf("%s\n", subStr.String().Data());
1338                         TString newStr(subStr.Data());
1339                         TObjArray* token = newStr.Tokenize(' ');
1340                         if (token)
1341                         {
1342                                 //token->Print();
1343                                 TObjString* tmpStr = dynamic_cast<TObjString*> (token->First());
1344                                 if (tmpStr)
1345                                 {
1346                                         Int_t testMode = tmpStr->String().Atoi();
1347                                         if (testMode > 0)
1348                                         {
1349                                                 Log("SHUTTLE", Form("Process - Enabling test mode %d", testMode));
1350                                                 SetTestMode((TestMode) testMode);
1351                                         }
1352                                 }
1353                                 delete token;          
1354                         }
1355                 }
1356         }
1357                 
1358         fLogbookEntry->Print("all");
1359
1360         // Initialization
1361         Bool_t hasError = kFALSE;
1362
1363         // Set the CDB and Reference folders according to the year
1364
1365         // build cdb paths (repeat each time, run might be a DATest run)
1366         if (!fLogbookEntry->GetDATestMode()){
1367                 fgkMainCDB.Form("alien://folder=%s%d/OCDB?user=alidaq?cacheFold=/tmp/OCDBCache", 
1368                                 fConfig->GetAlienPath(), GetCurrentYear());
1369                 
1370                 fgkMainRefStorage.Form("alien://folder=%s%d/Reference?user=alidaq?cacheFold=/tmp/OCDBCache", 
1371                                        fConfig->GetAlienPath(), GetCurrentYear());
1372         }
1373         else {
1374                 fgkMainCDB.Form("alien://folder=%s%d/DATest/OCDB?user=alidaq?cacheFold=/tmp/OCDBCache",
1375                                 fConfig->GetAlienPath(), GetCurrentYear());
1376                 
1377                 fgkMainRefStorage.Form("alien://folder=%s%d/DATest/Reference?user=alidaq?cacheFold=/tmp/OCDBCache",
1378                                        fConfig->GetAlienPath(), GetCurrentYear());
1379         }
1380
1381         AliDebug(2,Form("Main CDB storage = %s",fgkMainCDB.Data()));
1382         AliDebug(2,Form("Main Reference storage = %s",fgkMainRefStorage.Data()));
1383
1384         // Loop on detectors in the configuration
1385         TIter iter(fConfig->GetDetectors());
1386         TObjString* aDetector = 0;
1387
1388         Bool_t first = kTRUE;
1389
1390         while ((aDetector = (TObjString*) iter.Next()))
1391         {
1392                 fCurrentDetector = aDetector->String();
1393
1394                 if (ContinueProcessing() == kFALSE) 
1395                         continue;
1396                 
1397                 if (first)
1398                 {
1399                   // only read QueryCDB when needed and only once
1400                   AliCDBStorage *mainCDBSto = AliCDBManager::Instance()->GetStorage(fgkMainCDB);
1401                   if(mainCDBSto) mainCDBSto->QueryCDB(GetCurrentRun());
1402                   AliCDBStorage *mainRefSto = AliCDBManager::Instance()->GetStorage(fgkMainRefStorage);
1403                   if(mainRefSto) mainRefSto->QueryCDB(GetCurrentRun());
1404                   first = kFALSE;
1405                 }
1406
1407                 Log("SHUTTLE", Form("\t\t\t****** run %d - %s: START  ******",
1408                                                 GetCurrentRun(), aDetector->GetName()));
1409
1410                 for(Int_t iSys=0;iSys<3;iSys++) fFXSCalled[iSys]=kFALSE;
1411                 
1412                 Int_t initialMem = GetMem(getpid());
1413                 Log("SHUTTLE", Form("Memory consumption before forking is %d", initialMem));
1414
1415                 Log(fCurrentDetector.Data(), "Process - Starting processing");
1416
1417                 Int_t pid = fork();
1418
1419                 if (pid < 0)
1420                 {
1421                         Log("SHUTTLE", "Process - ERROR: Forking failed");
1422                 }
1423                 else if (pid > 0)
1424                 {
1425                         // parent
1426                         Log("SHUTTLE", Form("Process - In parent process of %d - %s: Starting monitoring",
1427                                                         GetCurrentRun(), aDetector->GetName()));
1428
1429                         Long_t begin = time(0);
1430
1431                         int status; // to be used with waitpid, on purpose an int (not Int_t)!
1432                         while (waitpid(pid, &status, WNOHANG) == 0)
1433                         {
1434                                 Long_t expiredTime = time(0) - begin;
1435
1436                                 if (expiredTime > fConfig->GetPPTimeOut())
1437                                 {
1438                                         TString logMsg;
1439                                         AliShuttleStatus *currentStatus = ReadShuttleStatus();
1440                                         AliShuttleStatus::Status newStatus = AliShuttleStatus::kInvalid;
1441                                         
1442                                         if (currentStatus->GetStatus() <= AliShuttleStatus::kPPDone)
1443                                         {
1444                                                 // in case pp not yet done set status to kPPTimeOut
1445                                         
1446                                                 logMsg.Form("Process - Process of %s timed out. Run time: %d seconds. Killing...",
1447                                                                 fCurrentDetector.Data(), expiredTime);
1448                                                 newStatus = AliShuttleStatus::kPPTimeOut;
1449                                         }
1450                                         else if (currentStatus->GetStatus() == AliShuttleStatus::kStoreStarted)
1451                                         {
1452                                                 // in case the pp goes in TimeOut while storing the objects in the OCDB
1453                                                 // set status to kStoreError
1454                                                 
1455                                                 logMsg.Form("Process - Process of %s timed out while storing the OCDB object. Run time: %d seconds. Killing... and setting status to StoreError.",
1456                                                                 fCurrentDetector.Data(), expiredTime);
1457                                                 newStatus = AliShuttleStatus::kStoreError;
1458                                         }
1459                                         else 
1460                                         {
1461                                                 // in other cases don't change the status
1462                                                 
1463                                                 logMsg.Form("Process - Process of %s timed out in status = %s. Run time: %d seconds. Killing... without changing the status",
1464                                                                 fCurrentDetector.Data(), currentStatus->GetStatusName(), expiredTime);
1465                                         }
1466                                 
1467                                         Log("SHUTTLE", logMsg);
1468                                         Log(fCurrentDetector, logMsg);
1469
1470                                         kill(pid, 9);
1471
1472                                         if (newStatus != AliShuttleStatus::kInvalid)
1473                                                 UpdateShuttleStatus(newStatus);
1474                                         hasError = kTRUE;
1475
1476                                         gSystem->Sleep(1000);
1477                                 }
1478                                 else
1479                                 {
1480                                         gSystem->Sleep(1000);
1481                                         
1482                                         Int_t mem = GetMem(pid);
1483
1484                                         if (mem < 0)
1485                                                 continue;
1486                                                 
1487                                         mem -= initialMem;
1488                                         if (mem < 0)
1489                                                 mem = 0;
1490                                         
1491                                         if (expiredTime % 60 == 0)
1492                                         {
1493                                                 Log("SHUTTLE", Form("Process - %s: Checking process. "
1494                                                         "Run time: %d seconds - Memory consumption: %d KB",
1495                                                         fCurrentDetector.Data(), expiredTime, mem));
1496                                                 SendAlive();
1497                                         }
1498                                         
1499                                         if (mem > fConfig->GetPPMaxMem())
1500                                         {
1501                                                 TString tmp;
1502                                                 tmp.Form("Process - Process exceeds maximum allowed memory "
1503                                                         "(%d KB > %d KB). Killing...",
1504                                                         mem, fConfig->GetPPMaxMem());
1505                                                 Log("SHUTTLE", tmp);
1506                                                 Log(fCurrentDetector, tmp);
1507         
1508                                                 kill(pid, 9);
1509         
1510                                                 UpdateShuttleStatus(AliShuttleStatus::kPPOutOfMemory);
1511                                                 hasError = kTRUE;
1512         
1513                                                 gSystem->Sleep(1000);
1514                                         }
1515                                 }
1516                         }
1517
1518                         Log("SHUTTLE", Form("Process - In parent process of %d - %s: Client has terminated.",
1519                                                                 GetCurrentRun(), aDetector->GetName()));
1520
1521                         if (WIFEXITED(status))
1522                         {
1523                                 Int_t returnCode = WEXITSTATUS(status);
1524
1525                                 Log("SHUTTLE", Form("Process - %s: the return code is %d", fCurrentDetector.Data(),
1526                                                                                 returnCode));
1527
1528                                 if (returnCode == 0) hasError = kTRUE;
1529                         }
1530                 }
1531                 else if (pid == 0)
1532                 {
1533                         // child
1534                         Log("SHUTTLE", Form("Process - In child process of %d - %s", GetCurrentRun(),
1535                                 aDetector->GetName()));
1536
1537                         Log("SHUTTLE", Form("Process - Redirecting output to %s log",fCurrentDetector.Data()));
1538
1539                         if ((freopen(GetLogFileName(fCurrentDetector), "a", stdout)) == 0)
1540                         {
1541                                 Log("SHUTTLE", "Process - Could not freopen stdout");
1542                         }
1543                         else
1544                         {
1545                                 fOutputRedirected = kTRUE;
1546                                 if ((dup2(fileno(stdout), fileno(stderr))) < 0)
1547                                         Log("SHUTTLE", "Process - Could not redirect stderr");
1548                                 
1549                         }
1550                         
1551                         TString wd = gSystem->WorkingDirectory();
1552                         Int_t dir_lev1 = GetCurrentRun()/10000;
1553                         TString tmpDir = Form("%s/%d/%d/%s_process", GetShuttleTempDir(), 
1554                                 dir_lev1, GetCurrentRun(), fCurrentDetector.Data());
1555                         
1556                         Int_t result = gSystem->GetPathInfo(tmpDir.Data(), 0, (Long64_t*) 0, 0, 0);
1557                         if (!result) // temp dir already exists!
1558                         {
1559                                 Log(fCurrentDetector.Data(), 
1560                                         Form("Process - %s dir already exists! Removing...", tmpDir.Data()));
1561                                 gSystem->Exec(Form("rm -rf %s",tmpDir.Data()));         
1562                         } 
1563                         
1564                         if (gSystem->mkdir(tmpDir.Data(), 1))
1565                         {
1566                                 Log(fCurrentDetector.Data(), "Process - could not make temp directory!!");
1567                                 gSystem->Exit(1);
1568                         }
1569                         
1570                         if (!gSystem->ChangeDirectory(tmpDir.Data())) 
1571                         {
1572                                 Log(fCurrentDetector.Data(), "Process - could not change directory!!");
1573                                 gSystem->Exit(1);                       
1574                         }
1575                         
1576                         Int_t success = ProcessCurrentDetector();
1577
1578                         gSystem->ChangeDirectory(wd.Data());
1579                                                 
1580                         if (success == 1) // Preprocessor finished successfully!
1581                         { 
1582                                 // remove temporary folder or DCS map
1583                                 if (!fConfig->KeepTempFolder())
1584                                 {
1585                                         gSystem->Exec(Form("rm -rf %s",tmpDir.Data()));
1586                                 } else if (!fConfig->KeepDCSMap())
1587                                 {
1588                                         gSystem->Exec(Form("rm -f %s/DCSMap.root",tmpDir.Data()));
1589                                 }
1590                                 
1591                                 // Update time_processed field in FXS DB
1592                                 if (UpdateTable() == kFALSE)
1593                                         Log("SHUTTLE", Form("Process - %s: Could not update FXS databases!", 
1594                                                         fCurrentDetector.Data()));
1595
1596                                 // Transfer the data from local storage to main storage (Grid)
1597                                 if (StoreOCDB() == kFALSE)
1598                                         success = kFALSE;
1599                         } 
1600                         else if (success == 0)
1601                         {
1602                                 Log("SHUTTLE", 
1603                                         Form("\t\t\t****** run %d - %s: PP ERROR ******",
1604                                                 GetCurrentRun(), aDetector->GetName()));
1605                         }
1606
1607                         for (UInt_t iSys=0; iSys<3; iSys++)
1608                         {
1609                                 if (fFXSCalled[iSys]) fFXSlist[iSys].Clear();
1610                         }
1611
1612                         Log("SHUTTLE", Form("Process - Client process of %d - %s is exiting now with %d.",
1613                                                         GetCurrentRun(), aDetector->GetName(), success));
1614
1615                         // the client exits here
1616                         gSystem->Exit(success);
1617
1618                         AliError("We should never get here!!!");
1619                 }
1620         }
1621
1622         Log("SHUTTLE", Form("\t\t\t^*^*^*^*^*^*^*^*^*^*^*^* run %d: FINISH ^*^*^*^*^*^*^*^*^*^*^*^*",
1623                                                         GetCurrentRun()));
1624
1625         //check if shuttle is done for this run, if so update logbook
1626         TObjArray checkEntryArray;
1627         checkEntryArray.SetOwner(1);
1628         TString whereClause = Form("where run=%d", GetCurrentRun());
1629         if (!QueryShuttleLogbook(whereClause.Data(), checkEntryArray) || 
1630                         checkEntryArray.GetEntries() == 0) {
1631                 Log("SHUTTLE", Form("Process - Warning: Cannot check status of run %d on Shuttle logbook!",
1632                                                 GetCurrentRun()));
1633                 return hasError == kFALSE;
1634         }
1635
1636         AliShuttleLogbookEntry* checkEntry = dynamic_cast<AliShuttleLogbookEntry*>
1637                                                 (checkEntryArray.At(0));
1638
1639         if (checkEntry)
1640         {
1641                 if (checkEntry->IsDone())
1642                 {
1643                         Log("SHUTTLE","Process - Shuttle is DONE. Updating logbook");
1644                         UpdateShuttleLogbook("shuttle_done");
1645                 }
1646                 else
1647                 {
1648                         for (UInt_t iDet=0; iDet<NDetectors(); iDet++)
1649                         {
1650                                 if (checkEntry->GetDetectorStatus(iDet) == AliShuttleLogbookEntry::kUnprocessed)
1651                                 {
1652                                         AliDebug(2, Form("Run %d: setting %s as \"not first time unprocessed\"",
1653                                                         checkEntry->GetRun(), GetDetName(iDet)));
1654                                         fFirstUnprocessed[iDet] = kFALSE;
1655                                 }
1656                         }
1657                         SendMLRunInfo("Pending");
1658                 }
1659         }
1660
1661         fLogbookEntry = 0;
1662
1663         return hasError == kFALSE;
1664 }
1665
1666 //______________________________________________________________________________________________
1667 Int_t AliShuttle::ProcessCurrentDetector()
1668 {
1669         //
1670         // Makes data retrieval just for a specific detector (fCurrentDetector).
1671         // Threre should be a configuration for this detector.
1672
1673         Log("SHUTTLE", Form("ProcessCurrentDetector - Retrieving values for %s, run %d", 
1674                                                 fCurrentDetector.Data(), GetCurrentRun()));
1675
1676         TString wd = gSystem->WorkingDirectory();
1677         
1678         if (!CleanReferenceStorage(fCurrentDetector.Data()))
1679                 return 0;
1680         
1681         gSystem->ChangeDirectory(wd.Data());
1682         
1683         // call preprocessor
1684         AliPreprocessor* aPreprocessor =
1685                 dynamic_cast<AliPreprocessor*> (fPreprocessorMap.GetValue(fCurrentDetector));
1686
1687         // check if the preprocessor wants to process this run type
1688         if (aPreprocessor->ProcessRunType() == kFALSE)
1689         {
1690                 UpdateShuttleStatus(AliShuttleStatus::kSkipped);
1691                 UpdateShuttleLogbook(fCurrentDetector, "DONE");
1692                 if (!UpdateTableSkippedCase(fCurrentDetector.Data()))
1693                 {
1694                         AliError(Form("Could not update FXS tables for run %d !", GetCurrentRun()));
1695                 }
1696                 Log(fCurrentDetector, Form("ProcessCurrentDetector - %s preprocessor is not interested in this run type", fCurrentDetector.Data()));
1697         
1698                 return 2;
1699         }
1700         
1701         TMap* dcsMap = new TMap();
1702         
1703         aPreprocessor->Initialize(GetCurrentRun(), GetCurrentStartTime(), GetCurrentEndTime());
1704
1705         Bool_t processDCS = aPreprocessor->ProcessDCS();
1706
1707         if (!processDCS)
1708         {
1709                 Log(fCurrentDetector, "ProcessCurrentDetector -"
1710                         " The preprocessor requested to skip the retrieval of DCS values");
1711         }
1712         else if (fTestMode & kSkipDCS)
1713         {
1714                 Log(fCurrentDetector, "ProcessCurrentDetector - In TESTMODE: Skipping DCS processing");
1715         } 
1716         else if (fTestMode & kErrorDCS)
1717         {
1718                 Log(fCurrentDetector, "ProcessCurrentDetector - In TESTMODE: Simulating DCS error");
1719                 UpdateShuttleStatus(AliShuttleStatus::kDCSStarted);
1720                 UpdateShuttleStatus(AliShuttleStatus::kDCSError);
1721                 delete dcsMap;
1722                 return 0;
1723         } else {
1724
1725                 UpdateShuttleStatus(AliShuttleStatus::kDCSStarted);
1726
1727                 // Query DCS archive
1728                 Int_t nServers = fConfig->GetNServers(fCurrentDetector);
1729                 
1730                 for (int iServ=0; iServ<nServers; iServ++)
1731                 {
1732                 
1733                         TString host(fConfig->GetDCSHost(fCurrentDetector, iServ));
1734                         Int_t port = fConfig->GetDCSPort(fCurrentDetector, iServ);
1735                         Int_t multiSplit = fConfig->GetMultiSplit(fCurrentDetector, iServ);
1736
1737                         Log(fCurrentDetector, Form("ProcessCurrentDetector -"
1738                                         " Querying DCS Amanda server %s:%d (%d of %d)", 
1739                                         host.Data(), port, iServ+1, nServers));
1740                         
1741                         TMap* aliasMap = 0;
1742                         TMap* dpMap = 0;
1743
1744                         if (fConfig->GetDCSAliases(fCurrentDetector, iServ)->GetEntries() > 0)
1745                         {
1746                                 Log(fCurrentDetector, Form("Querying %d DCS aliases", fConfig->GetDCSAliases(fCurrentDetector, iServ)->GetEntries()));
1747                                 aliasMap = GetValueSet(host, port, 
1748                                                 fConfig->GetDCSAliases(fCurrentDetector, iServ), 
1749                                                 kAlias, multiSplit);
1750                                 if (!aliasMap)
1751                                 {
1752                                         Log(fCurrentDetector, 
1753                                                 Form("ProcessCurrentDetector -"
1754                                                         " Error retrieving DCS aliases from server %s."
1755                                                         " Sending mail to DCS experts!", host.Data()));
1756                                         UpdateShuttleStatus(AliShuttleStatus::kDCSError);
1757                                         
1758                                         if (!SendMail(kDCSEMail))
1759                                                 Log("SHUTTLE", Form("ProcessCurrentDetector - "
1760                                                                     "Could not send mail to DCS experts!"));
1761
1762                                         delete dcsMap;
1763                                         return 0;
1764                                 }
1765                         }
1766                         
1767                         if (fConfig->GetDCSDataPoints(fCurrentDetector, iServ)->GetEntries() > 0)
1768                         {
1769                                 Log(fCurrentDetector, Form("Querying %d DCS data points", fConfig->GetDCSDataPoints(fCurrentDetector, iServ)->GetEntries()));
1770                                 dpMap = GetValueSet(host, port, 
1771                                                 fConfig->GetDCSDataPoints(fCurrentDetector, iServ), 
1772                                                 kDP, multiSplit);
1773                                 if (!dpMap)
1774                                 {
1775                                         Log(fCurrentDetector, 
1776                                                 Form("ProcessCurrentDetector -"
1777                                                         " Error retrieving DCS data points from server %s."
1778                                                         " Sending mail to DCS experts!", host.Data()));
1779                                         UpdateShuttleStatus(AliShuttleStatus::kDCSError);
1780                                         
1781                                         if (!SendMail(kDCSEMail))
1782                                                 Log("SHUTTLE", Form("ProcessCurrentDetector - "
1783                                                                     "Could not send mail to DCS experts!"));
1784                                         
1785                                         if (aliasMap) delete aliasMap;
1786                                         delete dcsMap;
1787                                         return 0;
1788                                 }                               
1789                         }
1790                         
1791                         // merge aliasMap and dpMap into dcsMap
1792                         if(aliasMap) {
1793                                 TIter iter(aliasMap);
1794                                 TObjString* key = 0;
1795                                 while ((key = (TObjString*) iter.Next()))
1796                                         dcsMap->Add(key, aliasMap->GetValue(key->String()));
1797                                 
1798                                 aliasMap->SetOwner(kFALSE);
1799                                 delete aliasMap;
1800                         }       
1801                         
1802                         if(dpMap) {
1803                                 TIter iter(dpMap);
1804                                 TObjString* key = 0;
1805                                 while ((key = (TObjString*) iter.Next()))
1806                                         dcsMap->Add(key, dpMap->GetValue(key->String()));
1807                                 
1808                                 dpMap->SetOwner(kFALSE);
1809                                 delete dpMap;
1810                         }
1811                 }
1812         }
1813         
1814         // save map into file, to help debugging in case of preprocessor error
1815         TFile* f = TFile::Open("DCSMap.root","recreate");
1816         f->cd();
1817         dcsMap->Write("DCSMap", TObject::kSingleKey);
1818         f->Close();
1819         delete f;
1820         
1821         // DCS Archive DB processing successful. Call Preprocessor!
1822         UpdateShuttleStatus(AliShuttleStatus::kPPStarted);
1823
1824         fFXSError = -1; // this variable is kTRUE after ::Process if an FXS error occured
1825         
1826         UInt_t returnValue = aPreprocessor->Process(dcsMap);
1827         
1828         if (fFXSError!=-1) {
1829                 UpdateShuttleStatus(AliShuttleStatus::kFXSError);
1830                 SendMail(kFXSEMail, fFXSError);
1831                 dcsMap->DeleteAll();
1832                 delete dcsMap;
1833                 return 0;
1834         }
1835
1836         if (returnValue > 0) // Preprocessor error!
1837         {
1838                 Log(fCurrentDetector, Form("ProcessCurrentDetector - "
1839                                 "Preprocessor failed. Process returned %d.", returnValue));
1840                 UpdateShuttleStatus(AliShuttleStatus::kPPError);
1841                 dcsMap->DeleteAll();
1842                 delete dcsMap;
1843                 return 0;
1844         }
1845         
1846         // preprocessor ok!
1847         UpdateShuttleStatus(AliShuttleStatus::kPPDone);
1848         Log(fCurrentDetector, Form("ProcessCurrentDetector - %s preprocessor returned success",
1849                                 fCurrentDetector.Data()));
1850
1851         dcsMap->DeleteAll();
1852         delete dcsMap;
1853
1854         return 1;
1855 }
1856
1857 //______________________________________________________________________________________________
1858 void AliShuttle::CountOpenRuns()
1859 {
1860         // Query DAQ's Shuttle logbook and sends the number of open runs to ML
1861         
1862         SendAlive();
1863         
1864         // check connection, in case connect
1865         if (!Connect(3)) 
1866                 return;
1867
1868         TString sqlQuery;
1869         sqlQuery = Form("select count(*) from %s where shuttle_done=0", fConfig->GetShuttlelbTable());
1870         
1871         TSQLResult* aResult = fServer[3]->Query(sqlQuery);
1872         if (!aResult) {
1873                 AliError(Form("Can't execute query <%s>!", sqlQuery.Data()));
1874                 return;
1875         }
1876
1877         AliDebug(2,Form("Query = %s", sqlQuery.Data()));
1878         
1879         if (aResult->GetRowCount() == 0) {
1880                 AliError(Form("No result for query %s received", sqlQuery.Data()));
1881                 return;
1882         }
1883
1884         if (aResult->GetFieldCount() != 1) {
1885                 AliError(Form("Invalid field count for query %s received", sqlQuery.Data()));
1886                 return;
1887         }
1888
1889         TSQLRow* aRow = aResult->Next();
1890         if (!aRow) {
1891                 AliError(Form("Could not receive result of query %s", sqlQuery.Data()));
1892                 return;
1893         }
1894         
1895         TString result(aRow->GetField(0), aRow->GetFieldLength(0));
1896         Int_t count = result.Atoi();
1897         
1898         Log("SHUTTLE", Form("%d unprocessed runs", count));
1899         
1900         delete aRow;
1901         delete aResult;
1902
1903         TMonaLisaValue mlStatus("SHUTTLE_openruns", count);
1904
1905         TList mlList;
1906         mlList.Add(&mlStatus);
1907
1908         fMonaLisa->SendParameters(&mlList, "__PROCESSINGINFO__");
1909 }
1910
1911 //______________________________________________________________________________________________
1912 Bool_t AliShuttle::QueryShuttleLogbook(const char* whereClause,
1913                 TObjArray& entries)
1914 {
1915         // Query DAQ's Shuttle logbook and fills detector status object.
1916         // Call QueryRunParameters to query DAQ logbook for run parameters.
1917         //
1918
1919         entries.SetOwner(1);
1920
1921         // check connection, in case connect
1922         if (!Connect(3)) return kFALSE;
1923
1924         TString sqlQuery;
1925         sqlQuery = Form("select * from %s %s order by run", fConfig->GetShuttlelbTable(), whereClause);
1926
1927         TSQLResult* aResult = fServer[3]->Query(sqlQuery);
1928         if (!aResult) {
1929                 AliError(Form("Can't execute query <%s>!", sqlQuery.Data()));
1930                 return kFALSE;
1931         }
1932
1933         AliDebug(2,Form("Query = %s", sqlQuery.Data()));
1934
1935         if(aResult->GetRowCount() == 0) {
1936                 Log("SHUTTLE", "No entries in Shuttle Logbook match request");
1937                 delete aResult;
1938                 return kTRUE;
1939         }
1940
1941         // TODO Check field count!
1942         const UInt_t nCols = 25;
1943         if (aResult->GetFieldCount() != (Int_t) nCols) {
1944                 Log("SHUTTLE", "Invalid SQL result field number!");
1945                 delete aResult;
1946                 return kFALSE;
1947         }
1948
1949         TSQLRow* aRow;
1950         while ((aRow = aResult->Next())) {
1951                 TString runString(aRow->GetField(0), aRow->GetFieldLength(0));
1952                 Int_t run = runString.Atoi();
1953
1954                 AliShuttleLogbookEntry *entry = QueryRunParameters(run);
1955                 if (!entry)
1956                         continue;
1957
1958                 // DA test mode flag
1959                 TString daTestModeString(aRow->GetField(2), aRow->GetFieldLength(2)); // field 2 = DA test mode flag 
1960                 Bool_t daTestMode = (Bool_t)daTestModeString.Atoi();
1961                 entry->SetDATestMode(daTestMode);
1962
1963                 // loop on detectors
1964                 for(UInt_t ii = 0; ii < nCols; ii++)
1965                         entry->SetDetectorStatus(aResult->GetFieldName(ii), aRow->GetField(ii));
1966
1967                 entries.AddLast(entry);
1968                 delete aRow;
1969         }
1970
1971         delete aResult;
1972         return kTRUE;
1973 }
1974
1975 //______________________________________________________________________________________________
1976 AliShuttleLogbookEntry* AliShuttle::QueryRunParameters(Int_t run)
1977 {
1978         //
1979         // Retrieve run parameters written in the DAQ logbook and sets them into AliShuttleLogbookEntry object
1980         //
1981
1982         // check connection, in case connect
1983         if (!Connect(3))
1984                 return 0;
1985
1986         TString sqlQuery;
1987         sqlQuery.Form("select * from %s where run=%d", fConfig->GetDAQlbTable(), run);
1988
1989         TSQLResult* aResult = fServer[3]->Query(sqlQuery);
1990         if (!aResult) {
1991                 Log("SHUTTLE", Form("Can't execute query <%s>!", sqlQuery.Data()));
1992                 return 0;
1993         }
1994
1995         if (aResult->GetRowCount() == 0) {
1996                 Log("SHUTTLE", Form("QueryRunParameters - No entry in DAQ Logbook for run %d. Skipping", run));
1997                 delete aResult;
1998                 return 0;
1999         }
2000
2001         if (aResult->GetRowCount() > 1) {
2002                 Log("SHUTTLE", Form("QueryRunParameters - UNEXPECTED: "
2003                                 "more than one entry in DAQ Logbook for run %d!", run));
2004                 delete aResult;
2005                 return 0;
2006         }
2007
2008         TSQLRow* aRow = aResult->Next();
2009         if (!aRow)
2010         {
2011                 Log("SHUTTLE", Form("QueryRunParameters - Could not retrieve row for run %d. Skipping", run));
2012                 delete aResult;
2013                 return 0;
2014         }
2015
2016         AliShuttleLogbookEntry* entry = new AliShuttleLogbookEntry(run);
2017
2018         for (Int_t ii = 0; ii < aResult->GetFieldCount(); ii++)
2019                 entry->SetRunParameter(aResult->GetFieldName(ii), aRow->GetField(ii));
2020
2021         delete aRow;
2022         delete aResult;
2023         
2024         UInt_t startTime = entry->GetStartTime();
2025         UInt_t endTime = entry->GetEndTime();
2026         Bool_t ecsSuccess = entry->GetECSSuccess();
2027         
2028         TString totEventsStr = entry->GetRunParameter("totalEvents");  
2029         Int_t totEvents = totEventsStr.Atoi();
2030         
2031         UInt_t now = time(0);
2032         // TODO make this a configuration parameter
2033         Int_t dcsDelay = fConfig->GetDCSDelay()+fConfig->GetDCSQueryOffset();
2034         
2035         // runs are accepted if they have ecsSuccess set or more than 1 event
2036         if (startTime != 0 && endTime != 0 && endTime > startTime && (totEvents > 1 || ecsSuccess) && (endTime < now - dcsDelay))
2037         {
2038                 if (ecsSuccess == kFALSE)
2039                         Log("SHUTTLE", Form("Processing run %d although in status ECS failure, Reason: %s", run, entry->GetRunParameter("eor_reason")));
2040                 return entry;
2041         }
2042
2043         Bool_t skip = kFALSE;
2044                                 
2045         if (endTime != 0 && endTime >= now - dcsDelay)
2046         {
2047                 Log("SHUTTLE", Form("Skipping run %d for now, because DCS buffer time is not yet expired", run));
2048         }
2049         else if (totEvents <= 1) 
2050         {
2051                 Log("SHUTTLE", Form("QueryRunParameters - Run %d has 1 event or less - Skipping!", run));
2052                 skip = kTRUE;
2053         }
2054         else
2055         {
2056                 Log("SHUTTLE", Form("QueryRunParameters - Invalid parameters for Run %d: "
2057                         "startTime = %d, endTime = %d. Skipping (Shuttle won't be marked as DONE)!",
2058                         run, startTime, endTime));
2059         }
2060         
2061         if (skip)
2062         {
2063                 Log("SHUTTLE", Form("Marking SHUTTLE skipped for run %d", run));
2064                 fLogbookEntry = entry;
2065                 if (!UpdateShuttleLogbook("shuttle_skipped"))
2066                 {
2067                         AliError(Form("Could not update logbook for run %d !", run));
2068                 }
2069                 if (!UpdateTableSkippedCase("ALL"))
2070                 {
2071                         AliError(Form("Could not update FXS tables for run %d !", run));
2072                 }
2073                 fLogbookEntry = 0;
2074         }
2075                         
2076         delete entry;
2077         return 0;
2078 }
2079
2080 //______________________________________________________________________________________________
2081 TMap* AliShuttle::GetValueSet(const char* host, Int_t port, const TSeqCollection* entries,
2082                               DCSType type, Int_t multiSplit)
2083 {
2084         // Retrieve all "entry" data points from the DCS server
2085         // host, port: TSocket connection parameters
2086         // entries: list of name of the alias or data point
2087         // type: kAlias or kDP
2088         // returns TMap of values, 0 when failure
2089         
2090         AliDCSClient client(host, port, fTimeout, fRetries, multiSplit);
2091
2092         TMap* result = 0;
2093         if (type == kAlias)
2094         {
2095                 //result = client.GetAliasValues(entries, GetCurrentStartTime()-offset, 
2096                 //      GetCurrentEndTime()+offset);
2097                 result = client.GetAliasValues(entries, GetStartTimeDCSQuery(), 
2098                         GetEndTimeDCSQuery());
2099         } 
2100         else if (type == kDP)
2101         {
2102                 //result = client.GetDPValues(entries, GetCurrentStartTime()-offset, 
2103                 //      GetCurrentEndTime()+offset);
2104                 result = client.GetDPValues(entries, GetStartTimeDCSQuery(), 
2105                         GetEndTimeDCSQuery());
2106         }
2107
2108         if (result == 0)
2109         {
2110                 Log(fCurrentDetector.Data(), Form("GetValueSet - Can't get entries! Reason: %s",
2111                         client.GetErrorString(client.GetResultErrorCode())));
2112                 if (client.GetResultErrorCode() == AliDCSClient::fgkServerError)        
2113                         Log(fCurrentDetector.Data(), Form("GetValueSet - Server error code: %s",
2114                                 client.GetServerError().Data()));
2115
2116                 return 0;
2117         }
2118                 
2119         return result;
2120 }
2121
2122 //______________________________________________________________________________________________
2123 const char* AliShuttle::GetFile(Int_t system, const char* detector,
2124                 const char* id, const char* source)
2125 {
2126         // Get calibration file from file exchange servers
2127         // First queris the FXS database for the file name, using the run, detector, id and source info
2128         // then calls RetrieveFile(filename) for actual copy to local disk
2129         // run: current run being processed (given by Logbook entry fLogbookEntry)
2130         // detector: the Preprocessor name
2131         // id: provided as a parameter by the Preprocessor
2132         // source: provided by the Preprocessor through GetFileSources function
2133
2134         // check if test mode should simulate a FXS error
2135         if (fTestMode & kErrorFXSFiles)
2136         {
2137                 Log(detector, Form("GetFile - In TESTMODE - Simulating error while connecting to %s FXS", GetSystemName(system)));
2138                 return 0;
2139         }
2140         
2141         // check connection, in case connect
2142         if (!Connect(system))
2143         {
2144                 Log(detector, Form("GetFile - Couldn't connect to %s FXS database", GetSystemName(system)));
2145                 fFXSError = system;
2146                 return 0;
2147         }
2148
2149         // Query preparation
2150         TString sourceName(source);
2151         Int_t nFields = 3;
2152         TString sqlQueryStart = Form("select filePath,size,fileChecksum from %s where",
2153                                                                 fConfig->GetFXSdbTable(system));
2154         TString whereClause = Form("run=%d and detector=\"%s\" and fileId=\"%s\"",
2155                                                                 GetCurrentRun(), detector, id);
2156
2157         if (system == kDAQ)
2158         {
2159                 whereClause += Form(" and DAQsource=\"%s\"", source);
2160         }
2161         else if (system == kDCS)
2162         {
2163                 sourceName="none";
2164         }
2165         else if (system == kHLT)
2166         {
2167                 whereClause += Form(" and DDLnumbers=\"%s\"", source);
2168         }
2169
2170         TString sqlQuery = Form("%s %s", sqlQueryStart.Data(), whereClause.Data());
2171
2172         AliDebug(2, Form("SQL query: \n%s",sqlQuery.Data()));
2173
2174         // Query execution
2175         TSQLResult* aResult = 0;
2176         aResult = dynamic_cast<TSQLResult*> (fServer[system]->Query(sqlQuery));
2177         if (!aResult) {
2178                 Log(detector, Form("GetFile - Can't execute SQL query to %s database for: id = %s, source = %s",
2179                                 GetSystemName(system), id, sourceName.Data()));
2180                 fFXSError = system;
2181                 return 0;
2182         }
2183
2184         if (aResult->GetRowCount() == 0)
2185         {
2186                 Log(detector,
2187                         Form("GetFile - No entry in %s FXS db for: id = %s, source = %s",
2188                                 GetSystemName(system), id, sourceName.Data()));
2189                 delete aResult;
2190                 return 0;
2191         }
2192
2193         if (aResult->GetRowCount() > 1) {
2194                 Log(detector,
2195                         Form("GetFile - More than one entry in %s FXS db for: id = %s, source = %s",
2196                                 GetSystemName(system), id, sourceName.Data()));
2197                 fFXSError = system;
2198                 delete aResult;
2199                 return 0;
2200         }
2201
2202         if (aResult->GetFieldCount() != nFields) {
2203                 Log(detector,
2204                         Form("GetFileName - Wrong field count in %s FXS db for: id = %s, source = %s",
2205                                 GetSystemName(system), id, sourceName.Data()));
2206                 fFXSError = system;
2207                 delete aResult;
2208                 return 0;
2209         }
2210
2211         TSQLRow* aRow = dynamic_cast<TSQLRow*> (aResult->Next());
2212
2213         if (!aRow){
2214                 Log(detector, Form("GetFile - Empty set result in %s FXS db from query: id = %s, source = %s",
2215                                 GetSystemName(system), id, sourceName.Data()));
2216                 fFXSError = system;
2217                 delete aResult;
2218                 return 0;
2219         }
2220
2221         TString filePath(aRow->GetField(0), aRow->GetFieldLength(0));
2222         TString fileSize(aRow->GetField(1), aRow->GetFieldLength(1));
2223         TString fileChecksum(aRow->GetField(2), aRow->GetFieldLength(2));
2224
2225         delete aResult;
2226         delete aRow;
2227
2228         AliDebug(2, Form("filePath = %s; size = %s, fileChecksum = %s",
2229                                 filePath.Data(), fileSize.Data(), fileChecksum.Data()));
2230
2231         // retrieved file is renamed to make it unique
2232         Int_t dir_lev1 = GetCurrentRun()/10000;
2233         TString localFileName = Form("%s/%d/%d/%s_process/%s_%s_%d_%s_%s.shuttle",
2234                                      GetShuttleTempDir(), dir_lev1, GetCurrentRun(), detector,
2235                                         GetSystemName(system), detector, GetCurrentRun(), 
2236                                         id, sourceName.Data());
2237         Log("SHUTTLE",Form("file from FXS = %s",localFileName.Data())); 
2238
2239
2240         // file retrieval from FXS
2241         UInt_t nRetries = 0;
2242         UInt_t maxRetries = 3;
2243         Bool_t result = kFALSE;
2244
2245         // copy!! if successful TSystem::Exec returns 0
2246         while (nRetries++ < maxRetries) {
2247                 AliDebug(2, Form("Trying to copy file. Retry # %d", nRetries));
2248                 result = RetrieveFile(system, filePath.Data(), localFileName.Data());
2249                 if (!result)
2250                 {
2251                         Log(detector, Form("GetFile - Copy of file %s from %s FXS failed",
2252                                         filePath.Data(), GetSystemName(system)));
2253                         continue;
2254                 } 
2255
2256                 if (fileSize.Length()>0)
2257                 {
2258                         // compare filesize of local file with the one stored in the FXS DB
2259                         Long_t size = -1;
2260                         Int_t sizeComp = gSystem->GetPathInfo(localFileName.Data(), 0, &size, 0, 0);
2261
2262                         if (sizeComp != 0 || size != fileSize.Atoi())
2263                         {
2264                                 Log(detector, Form("GetFile - size of file %s does not match with local copy!",
2265                                                         filePath.Data()));
2266                                 result = kFALSE;
2267                                 continue;
2268                         }
2269
2270                 } else {
2271                         Log(fCurrentDetector, Form("GetFile - size of file %s not set in %s database, skipping comparison",
2272                                                 filePath.Data(), GetSystemName(system)));
2273                 }
2274
2275                 if (fileChecksum.Length()>0)
2276                 {
2277                         // compare md5sum of local file with the one stored in the FXS DB
2278                         if(fileChecksum.Contains(' ')) fileChecksum.Resize(fileChecksum.First(' '));
2279                         Int_t md5Comp = gSystem->Exec(Form("md5sum %s |grep %s > /dev/null 2> /dev/null",
2280                                                 localFileName.Data(), fileChecksum.Data()));
2281
2282                         if (md5Comp != 0)
2283                         {
2284                                 Log(detector, Form("GetFile - md5sum of file %s does not match with local copy!",
2285                                                         filePath.Data()));
2286                                 result = kFALSE;
2287                                 continue;
2288                         }
2289                 } else {
2290                         Log(fCurrentDetector, Form("GetFile - md5sum of file %s not set in %s database, skipping comparison",
2291                                                         filePath.Data(), GetSystemName(system)));
2292                 }
2293                 if (result) break;
2294         }
2295
2296         if (!result) 
2297         {
2298                 fFXSError = system;
2299                 return 0;
2300         }
2301
2302         fFXSCalled[system]=kTRUE;
2303         TObjString *fileParams = new TObjString(Form("%s#!?!#%s", id, sourceName.Data()));
2304         fFXSlist[system].Add(fileParams);
2305
2306         static TString staticLocalFileName;
2307         staticLocalFileName.Form("%s", localFileName.Data());
2308         
2309         Log(fCurrentDetector, Form("GetFile - Retrieved file with id %s and "
2310                         "source %s from %s to %s", id, source, 
2311                         GetSystemName(system), localFileName.Data()));
2312                         
2313         return staticLocalFileName.Data();
2314 }
2315
2316 //______________________________________________________________________________________________
2317 Bool_t AliShuttle::RetrieveFile(UInt_t system, const char* fxsFileName, const char* localFileName)
2318 {
2319         //
2320         // Copies file from FXS to local Shuttle machine
2321         //
2322
2323         // check temp directory: trying to cd to temp; if it does not exist, create it
2324         AliDebug(2, Form("Copy file %s from %s FXS into %s",
2325                         GetSystemName(system), fxsFileName, localFileName));
2326                         
2327         TString tmpDir(localFileName);
2328         
2329         tmpDir = tmpDir(0,tmpDir.Last('/'));
2330
2331         Int_t noDir = gSystem->GetPathInfo(tmpDir.Data(), 0, (Long64_t*) 0, 0, 0);
2332         if (noDir) // temp dir does not exists!
2333         {
2334                 if (gSystem->mkdir(tmpDir.Data(), 1))
2335                 {
2336                         Log(fCurrentDetector.Data(), "RetrieveFile - could not make temp directory!!");
2337                         return kFALSE;
2338                 }
2339         }
2340
2341         TString command = Form("scp -oPort=%d -2 %s@%s:%s/%s %s",
2342                 fConfig->GetFXSPort(system),
2343                 fConfig->GetFXSUser(system),
2344                 fConfig->GetFXSHost(system),
2345                 fConfig->GetFXSBaseFolder(system),
2346                 fxsFileName,
2347                 localFileName);
2348
2349         AliDebug(2, Form("%s",command.Data()));
2350
2351         Bool_t result = (gSystem->Exec(command.Data()) == 0);
2352
2353         return result;
2354 }
2355
2356 //______________________________________________________________________________________________
2357 TList* AliShuttle::GetFileSources(Int_t system, const char* detector, const char* id)
2358 {
2359         //
2360         // Get sources producing the condition file Id from file exchange servers
2361         // if id is NULL all sources are returned (distinct)
2362         //
2363
2364         if (id)
2365         {
2366                 Log(detector, Form("GetFileSources - Querying %s FXS for files with id %s produced by %s", GetSystemName(system), id, detector));
2367         } else {
2368                 Log(detector, Form("GetFileSources - Querying %s FXS for files produced by %s", GetSystemName(system), detector));
2369         }
2370         
2371         // check if test mode should simulate a FXS error
2372         if (fTestMode & kErrorFXSSources)
2373         {
2374                 Log(detector, Form("GetFileSources - In TESTMODE - Simulating error while connecting to %s FXS", GetSystemName(system)));
2375                 return 0;
2376         }
2377
2378         if (system == kDCS)
2379         {
2380                 Log(detector, "GetFileSources - WARNING: DCS system has only one source of data!");
2381                 TList *list = new TList();
2382                 list->SetOwner(1);
2383                 list->Add(new TObjString(" "));
2384                 return list;
2385         }
2386
2387         // check connection, in case connect
2388         if (!Connect(system))
2389         {
2390                 Log(detector, Form("GetFileSources - Couldn't connect to %s FXS database", GetSystemName(system)));
2391                 fFXSError = system;
2392                 return NULL;
2393         }
2394
2395         TString sourceName = "";
2396         if (system == kDAQ)
2397         {
2398                 sourceName = "DAQsource";
2399         } else if (system == kHLT)
2400         {
2401                 sourceName = "DDLnumbers";
2402         }
2403
2404         TString sqlQueryStart = Form("select distinct %s from %s where", sourceName.Data(), fConfig->GetFXSdbTable(system));
2405         TString whereClause = Form("run=%d and detector=\"%s\"",
2406                                 GetCurrentRun(), detector);
2407         if (id)
2408                 whereClause += Form(" and fileId=\"%s\"", id);
2409         TString sqlQuery = Form("%s %s", sqlQueryStart.Data(), whereClause.Data());
2410
2411         AliDebug(2, Form("SQL query: \n%s",sqlQuery.Data()));
2412
2413         // Query execution
2414         TSQLResult* aResult;
2415         aResult = fServer[system]->Query(sqlQuery);
2416         if (!aResult) {
2417                 Log(detector, Form("GetFileSources - Can't execute SQL query to %s database for id: %s",
2418                                 GetSystemName(system), id));
2419                 fFXSError = system;
2420                 return 0;
2421         }
2422
2423         TList *list = new TList();
2424         list->SetOwner(1);
2425         
2426         if (aResult->GetRowCount() == 0)
2427         {
2428                 Log(detector,
2429                         Form("GetFileSources - No entry in %s FXS table for id: %s", GetSystemName(system), id));
2430                 delete aResult;
2431                 return list;
2432         }
2433
2434         Log(detector, Form("GetFileSources - Found %d sources", aResult->GetRowCount()));
2435
2436         TSQLRow* aRow;
2437         while ((aRow = aResult->Next()))
2438         {
2439
2440                 TString source(aRow->GetField(0), aRow->GetFieldLength(0));
2441                 AliDebug(2, Form("%s = %s", sourceName.Data(), source.Data()));
2442                 list->Add(new TObjString(source));
2443                 delete aRow;
2444         }
2445
2446         delete aResult;
2447
2448         return list;
2449 }
2450
2451 //______________________________________________________________________________________________
2452 TList* AliShuttle::GetFileIDs(Int_t system, const char* detector, const char* source)
2453 {
2454         //
2455         // Get all ids of condition files produced by a given source from file exchange servers
2456         //
2457         
2458         Log(detector, Form("GetFileIDs - Retrieving ids with source %s with %s", source, GetSystemName(system)));
2459
2460         // check if test mode should simulate a FXS error
2461         if (fTestMode & kErrorFXSSources)
2462         {
2463                 Log(detector, Form("GetFileIDs - In TESTMODE - Simulating error while connecting to %s FXS", GetSystemName(system)));
2464                 return 0;
2465         }
2466
2467         // check connection, in case connect
2468         if (!Connect(system))
2469         {
2470                 Log(detector, Form("GetFileIDs - Couldn't connect to %s FXS database", GetSystemName(system)));
2471                 return NULL;
2472         }
2473
2474         TString sourceName = "";
2475         if (system == kDAQ)
2476         {
2477                 sourceName = "DAQsource";
2478         } else if (system == kHLT)
2479         {
2480                 sourceName = "DDLnumbers";
2481         }
2482
2483         TString sqlQueryStart = Form("select fileId from %s where", fConfig->GetFXSdbTable(system));
2484         TString whereClause = Form("run=%d and detector=\"%s\"",
2485                                 GetCurrentRun(), detector);
2486         if (sourceName.Length() > 0 && source)
2487                 whereClause += Form(" and %s=\"%s\"", sourceName.Data(), source);
2488         TString sqlQuery = Form("%s %s", sqlQueryStart.Data(), whereClause.Data());
2489
2490         AliDebug(2, Form("SQL query: \n%s",sqlQuery.Data()));
2491
2492         // Query execution
2493         TSQLResult* aResult;
2494         aResult = fServer[system]->Query(sqlQuery);
2495         if (!aResult) {
2496                 Log(detector, Form("GetFileIDs - Can't execute SQL query to %s database for source: %s",
2497                                 GetSystemName(system), source));
2498                 return 0;
2499         }
2500
2501         TList *list = new TList();
2502         list->SetOwner(1);
2503         
2504         if (aResult->GetRowCount() == 0)
2505         {
2506                 Log(detector,
2507                         Form("GetFileIDs - No entry in %s FXS table for source: %s", GetSystemName(system), source));
2508                 delete aResult;
2509                 return list;
2510         }
2511
2512         Log(detector, Form("GetFileIDs - Found %d ids", aResult->GetRowCount()));
2513
2514         TSQLRow* aRow;
2515
2516         while ((aRow = aResult->Next()))
2517         {
2518
2519                 TString id(aRow->GetField(0), aRow->GetFieldLength(0));
2520                 AliDebug(2, Form("fileId = %s", id.Data()));
2521                 list->Add(new TObjString(id));
2522                 delete aRow;
2523         }
2524
2525         delete aResult;
2526
2527         return list;
2528 }
2529
2530 //______________________________________________________________________________________________
2531 Bool_t AliShuttle::Connect(Int_t system)
2532 {
2533         // Connect to MySQL Server of the system's FXS MySQL databases
2534         // DAQ Logbook, Shuttle Logbook and DAQ FXS db are on the same host
2535         //
2536
2537         // check connection: if already connected return
2538
2539         if(fServer[system] && fServer[system]->IsConnected()) {
2540                 // ping the server              
2541                 if (fServer[system]->PingVerify()==kTRUE){ // connection is still alive
2542                         return kTRUE;
2543                 }               
2544                 else{
2545                         AliWarning(Form("Connection got lost to FXS database for %s. Closing and reconnecting.",
2546                                         AliShuttleInterface::GetSystemName(system)));
2547                         fServer[system]->Close();
2548                         delete fServer[system];
2549                         fServer[system] = 0x0;
2550                 }
2551         }
2552
2553         TString dbHost, dbUser, dbPass, dbName;
2554
2555         if (system < 3) // FXS db servers
2556         {
2557                 dbHost = Form("mysql://%s:%d", fConfig->GetFXSdbHost(system), fConfig->GetFXSdbPort(system));
2558                 dbUser = fConfig->GetFXSdbUser(system);
2559                 dbPass = fConfig->GetFXSdbPass(system);
2560                 dbName =   fConfig->GetFXSdbName(system);
2561         } else { // Run & Shuttle logbook servers
2562         // TODO Will the Shuttle logbook server be the same as the Run logbook server ???
2563                 dbHost = Form("mysql://%s:%d", fConfig->GetDAQlbHost(), fConfig->GetDAQlbPort());
2564                 dbUser = fConfig->GetDAQlbUser();
2565                 dbPass = fConfig->GetDAQlbPass();
2566                 dbName =   fConfig->GetDAQlbDB();
2567         }
2568
2569         fServer[system] = TSQLServer::Connect(dbHost.Data(), dbUser.Data(), dbPass.Data());
2570         if (!fServer[system] || !fServer[system]->IsConnected()) {
2571                 if(system < 3)
2572                 {
2573                 AliError(Form("Can't establish connection to FXS database for %s",
2574                                         AliShuttleInterface::GetSystemName(system)));
2575                 } else {
2576                 AliError("Can't establish connection to Run logbook.");
2577                 }
2578                 if(fServer[system]) delete fServer[system];
2579                 return kFALSE;
2580         }
2581
2582         // Get tables
2583         TSQLResult* aResult=0;
2584         switch(system){
2585                 case kDAQ:
2586                         aResult = fServer[kDAQ]->GetTables(dbName.Data());
2587                         break;
2588                 case kDCS:
2589                         aResult = fServer[kDCS]->GetTables(dbName.Data());
2590                         break;
2591                 case kHLT:
2592                         aResult = fServer[kHLT]->GetTables(dbName.Data());
2593                         break;
2594                 default:
2595                         aResult = fServer[3]->GetTables(dbName.Data());
2596                         break;
2597         }
2598
2599         delete aResult;
2600         return kTRUE;
2601 }
2602
2603 //______________________________________________________________________________________________
2604 Bool_t AliShuttle::UpdateTable()
2605 {
2606         //
2607         // Update FXS table filling time_processed field in all rows corresponding to current run and detector
2608         //
2609
2610         Bool_t result = kTRUE;
2611
2612         for (UInt_t system=0; system<3; system++)
2613         {
2614                 if(!fFXSCalled[system]) continue;
2615
2616                 // check connection, in case connect
2617                 if (!Connect(system))
2618                 {
2619                         Log(fCurrentDetector, Form("UpdateTable - Couldn't connect to %s FXS database", GetSystemName(system)));
2620                         result = kFALSE;
2621                         continue;
2622                 }
2623
2624                 TTimeStamp now; // now
2625
2626                 // Loop on FXS list entries
2627                 TIter iter(&fFXSlist[system]);
2628                 TObjString *aFXSentry=0;
2629                 while ((aFXSentry = dynamic_cast<TObjString*> (iter.Next())))
2630                 {
2631                         TString aFXSentrystr = aFXSentry->String();
2632                         TObjArray *aFXSarray = aFXSentrystr.Tokenize("#!?!#");
2633                         if (!aFXSarray || aFXSarray->GetEntries() != 2 )
2634                         {
2635                                 Log(fCurrentDetector, Form("UpdateTable - error updating %s FXS entry. Check string: <%s>",
2636                                         GetSystemName(system), aFXSentrystr.Data()));
2637                                 if(aFXSarray) delete aFXSarray;
2638                                 result = kFALSE;
2639                                 continue;
2640                         }
2641                         const char* fileId = ((TObjString*) aFXSarray->At(0))->GetName();
2642                         const char* source = ((TObjString*) aFXSarray->At(1))->GetName();
2643
2644                         TString whereClause;
2645                         if (system == kDAQ)
2646                         {
2647                                 whereClause = Form("where run=%d and detector=\"%s\" and fileId=\"%s\" and DAQsource=\"%s\";",
2648                                                         GetCurrentRun(), fCurrentDetector.Data(), fileId, source);
2649                         }
2650                         else if (system == kDCS)
2651                         {
2652                                 whereClause = Form("where run=%d and detector=\"%s\" and fileId=\"%s\";",
2653                                                         GetCurrentRun(), fCurrentDetector.Data(), fileId);
2654                         }
2655                         else if (system == kHLT)
2656                         {
2657                                 whereClause = Form("where run=%d and detector=\"%s\" and fileId=\"%s\" and DDLnumbers=\"%s\";",
2658                                                         GetCurrentRun(), fCurrentDetector.Data(), fileId, source);
2659                         }
2660
2661                         delete aFXSarray;
2662
2663                         TString sqlQuery = Form("update %s set time_processed=%d %s", fConfig->GetFXSdbTable(system),
2664                                                                 now.GetSec(), whereClause.Data());
2665
2666                         AliDebug(2, Form("SQL query: \n%s",sqlQuery.Data()));
2667
2668                         // Query execution
2669                         TSQLResult* aResult;
2670                         aResult = dynamic_cast<TSQLResult*> (fServer[system]->Query(sqlQuery));
2671                         if (!aResult)
2672                         {
2673                                 Log(fCurrentDetector, Form("UpdateTable - %s db: can't execute SQL query <%s>",
2674                                                                 GetSystemName(system), sqlQuery.Data()));
2675                                 result = kFALSE;
2676                                 continue;
2677                         }
2678                         delete aResult;
2679                 }
2680         }
2681
2682         return result;
2683 }
2684
2685 //_______________________________________________________________________________
2686 Bool_t AliShuttle::UpdateTableSkippedCase(const char* detector)
2687 {
2688         //
2689         // Update FXS table filling time_processed field in all rows corresponding to current run and detector
2690         // if detector = "ALL" update all detectors
2691         //
2692
2693         Bool_t result = kTRUE;
2694
2695         TString detName(detector);
2696
2697         for (UInt_t system=0; system<3; system++)
2698         {
2699
2700                 // check connection, in case connect
2701                 if (!Connect(system))
2702                 {
2703                         Log(fCurrentDetector, Form("UpdateTableSkippedCase - Couldn't connect to %s FXS database", GetSystemName(system)));
2704                         result = kFALSE;
2705                         continue;
2706                 }
2707
2708                 TTimeStamp now; // now
2709
2710                 // Loop on FXS list entries
2711                 TIter iter(&fFXSlist[system]);
2712                         
2713                 TString whereClause;
2714                 if (detName == "ALL") whereClause = Form("where run=%d and time_processed IS NULL;",GetCurrentRun());
2715                 else whereClause = Form("where run=%d and detector=\"%s\" and time_processed IS NULL;",GetCurrentRun(), detector);
2716
2717                 Log("SHUTTLE",Form(" whereClause = %s ",whereClause.Data()));
2718
2719                 TString sqlQuery = Form("update %s set time_processed=%d %s", fConfig->GetFXSdbTable(system),
2720                                         now.GetSec(), whereClause.Data());
2721
2722                 AliDebug(2, Form("SQL query: \n%s",sqlQuery.Data()));
2723
2724                 // Query execution
2725                 TSQLResult* aResult;
2726                 aResult = dynamic_cast<TSQLResult*> (fServer[system]->Query(sqlQuery));
2727                 if (!aResult)
2728                 {
2729                         Log("SHUTTLE", Form("UpdateTableSkippedCase - %s db: can't execute SQL query <%s>",
2730                                                         GetSystemName(system), sqlQuery.Data()));
2731                         result = kFALSE;
2732                         continue;
2733                 }
2734                 delete aResult;
2735                 
2736         }
2737
2738         return result;
2739 }
2740 //______________________________________________________________________________________________
2741 Bool_t AliShuttle::UpdateTableFailCase()
2742 {
2743         // Update FXS table filling time_processed field in all rows corresponding to current run and detector
2744         // this is called in case the preprocessor is declared failed for the current run, because
2745         // the fields are updated only in case of success
2746
2747         Bool_t result = kTRUE;
2748
2749         for (UInt_t system=0; system<3; system++)
2750         {
2751                 // check connection, in case connect
2752                 if (!Connect(system))
2753                 {
2754                         Log(fCurrentDetector, Form("UpdateTableFailCase - Couldn't connect to %s FXS database",
2755                                                         GetSystemName(system)));
2756                         result = kFALSE;
2757                         continue;
2758                 }
2759
2760                 TTimeStamp now; // now
2761
2762                 // Loop on FXS list entries
2763
2764                 TString whereClause = Form("where run=%d and detector=\"%s\";",
2765                                                 GetCurrentRun(), fCurrentDetector.Data());
2766
2767
2768                 TString sqlQuery = Form("update %s set time_processed=%d %s", fConfig->GetFXSdbTable(system),
2769                                                         now.GetSec(), whereClause.Data());
2770
2771                 AliDebug(2, Form("SQL query: \n%s",sqlQuery.Data()));
2772
2773                 // Query execution
2774                 TSQLResult* aResult;
2775                 aResult = dynamic_cast<TSQLResult*> (fServer[system]->Query(sqlQuery));
2776                 if (!aResult)
2777                 {
2778                         Log(fCurrentDetector, Form("UpdateTableFailCase - %s db: can't execute SQL query <%s>",
2779                                                         GetSystemName(system), sqlQuery.Data()));
2780                         result = kFALSE;
2781                         continue;
2782                 }
2783                 delete aResult;
2784         }
2785
2786         return result;
2787 }
2788
2789 //______________________________________________________________________________________________
2790 Bool_t AliShuttle::UpdateShuttleLogbook(const char* detector, const char* status)
2791 {
2792         //
2793         // Update Shuttle logbook filling detector or shuttle_done column
2794         // ex. of usage: UpdateShuttleLogbook("PHOS", "DONE") or UpdateShuttleLogbook("shuttle_done")
2795         //
2796
2797         // check connection, in case connect
2798         if(!Connect(3)){
2799                 Log("SHUTTLE", "UpdateShuttleLogbook - Couldn't connect to DAQ Logbook.");
2800                 return kFALSE;
2801         }
2802
2803         TString detName(detector);
2804         TString setClause;
2805         if (detName == "shuttle_done" || detName == "shuttle_skipped")
2806         {
2807                 setClause = "set shuttle_done=1";
2808                 
2809                 if (detName == "shuttle_done")
2810                 {
2811                         if (TouchFile() != kTRUE)
2812                         {
2813                                 SendMLRunInfo("Pending");
2814                                 return kFALSE;
2815                         }
2816                         
2817                         SendMLRunInfo("Done");
2818                 }
2819                 else 
2820                         SendMLRunInfo("Skipped");
2821         } 
2822         else {
2823                 TString statusStr(status);
2824                 if(statusStr.Contains("done", TString::kIgnoreCase) ||
2825                    statusStr.Contains("failed", TString::kIgnoreCase)){
2826                         setClause = Form("set %s=\"%s\"", detector, status);
2827                 } else {
2828                         Log("SHUTTLE",
2829                                 Form("UpdateShuttleLogbook - Invalid status <%s> for detector %s",
2830                                         status, detector));
2831                         return kFALSE;
2832                 }
2833         }
2834
2835         TString whereClause = Form("where run=%d", GetCurrentRun());
2836
2837         TString sqlQuery = Form("update %s %s %s",
2838                                         fConfig->GetShuttlelbTable(), setClause.Data(), whereClause.Data());
2839
2840         AliDebug(2, Form("SQL query: \n%s",sqlQuery.Data()));
2841
2842         // Query execution
2843         TSQLResult* aResult;
2844         aResult = dynamic_cast<TSQLResult*> (fServer[3]->Query(sqlQuery));
2845         if (!aResult) {
2846                 Log("SHUTTLE", Form("UpdateShuttleLogbook - Can't execute query <%s>", sqlQuery.Data()));
2847                 return kFALSE;
2848         }
2849         delete aResult;
2850
2851         return kTRUE;
2852 }
2853
2854 //______________________________________________________________________________________________
2855 Int_t AliShuttle::GetCurrentRun() const
2856 {
2857         //
2858         // Get current run from logbook entry
2859         //
2860
2861         return fLogbookEntry ? fLogbookEntry->GetRun() : -1;
2862 }
2863
2864 //______________________________________________________________________________________________
2865 UInt_t AliShuttle::GetCurrentStartTime() const
2866 {
2867         //
2868         // get current start time
2869         //
2870
2871         return fLogbookEntry ? fLogbookEntry->GetStartTime() : 0;
2872 }
2873
2874 //______________________________________________________________________________________________
2875 UInt_t AliShuttle::GetCurrentEndTime() const
2876 {
2877         //
2878         // get current end time from logbook entry
2879         //
2880
2881         return fLogbookEntry ? fLogbookEntry->GetEndTime() : 0;
2882 }
2883 //______________________________________________________________________________________________
2884 UInt_t AliShuttle::GetCurrentYear() const
2885 {
2886         //
2887         // Get current year from logbook entry
2888         //
2889
2890         if (!fLogbookEntry) return 0;
2891         
2892         TTimeStamp startTime(GetCurrentStartTime());
2893         TString year =  Form("%d",startTime.GetDate());
2894         year = year(0,4);
2895         
2896         return year.Atoi();
2897 }
2898
2899 //______________________________________________________________________________________________
2900 const char* AliShuttle::GetLHCPeriod() const
2901 {
2902         //
2903         // Get current LHC period from logbook entry
2904         //
2905
2906         if (!fLogbookEntry) return 0;
2907                 
2908         return fLogbookEntry->GetRunParameter("LHCperiod");
2909 }
2910
2911 //______________________________________________________________________________________________
2912 void AliShuttle::Log(const char* detector, const char* message)
2913 {
2914         //
2915         // Fill log string with a message
2916         //
2917         
2918         TString logRunDir = GetShuttleLogDir();
2919         if (GetCurrentRun() >=0) {
2920                 Int_t logDir_lev1 = GetCurrentRun()/10000;
2921                 logRunDir += Form("/%d/%d", logDir_lev1, GetCurrentRun());
2922         }               
2923         void* dir = gSystem->OpenDirectory(logRunDir.Data());
2924         if (dir == NULL) {
2925                 if (gSystem->mkdir(logRunDir.Data(), kTRUE)) {
2926                         AliError(Form("Can't open directory <%s>", GetShuttleLogDir()));
2927                         return;
2928                 }
2929
2930         } else {
2931                 gSystem->FreeDirectory(dir);
2932         }
2933
2934         TString toLog = Form("%s UTC (%d): %s - ", TTimeStamp(time(0)).AsString("s"), getpid(), detector);
2935         if (GetCurrentRun() >= 0) 
2936                 toLog += Form("run %d - ", GetCurrentRun());
2937         toLog += Form("%s", message);
2938
2939         AliInfo(toLog.Data());
2940         
2941         // if we redirect the log output already to the file, leave here
2942         if (fOutputRedirected && strcmp(detector, "SHUTTLE") != 0)
2943                 return;
2944
2945         TString fileName = GetLogFileName(detector);
2946         
2947         gSystem->ExpandPathName(fileName);
2948
2949         ofstream logFile;
2950         logFile.open(fileName, ofstream::out | ofstream::app);
2951
2952         if (!logFile.is_open()) {
2953                 AliError(Form("Could not open file %s", fileName.Data()));
2954                 return;
2955         }
2956
2957         logFile << toLog.Data() << "\n";
2958
2959         logFile.close();
2960 }
2961
2962 //______________________________________________________________________________________________
2963 TString AliShuttle::GetLogFileName(const char* detector) const
2964 {
2965         // 
2966         // returns the name of the log file for a given sub detector
2967         //
2968         
2969         TString fileName;
2970         
2971         if (GetCurrentRun() >= 0) 
2972         {
2973                 Int_t logDir_lev1 = GetCurrentRun()/10000;
2974                 fileName.Form("%s/%d/%d/%s.log", GetShuttleLogDir(), logDir_lev1, GetCurrentRun(), 
2975                         detector);
2976         } else {
2977                 fileName.Form("%s/%s.log", GetShuttleLogDir(), detector);
2978         }
2979
2980         return fileName;
2981 }
2982
2983 //______________________________________________________________________________________________
2984 void AliShuttle::SendAlive()
2985 {
2986         // sends alive message to ML
2987         
2988         TMonaLisaText mlStatus("SHUTTLE_status", "Alive");
2989
2990         TList mlList;
2991         mlList.Add(&mlStatus);
2992
2993         fMonaLisa->SendParameters(&mlList, "__PROCESSINGINFO__");
2994 }
2995
2996 //______________________________________________________________________________________________
2997 Bool_t AliShuttle::Collect(Int_t run)
2998 {
2999         //
3000         // Collects conditions data for all UNPROCESSED run written to DAQ LogBook in case of run = -1 (default)
3001         // If a dedicated run is given this run is processed
3002         //
3003         // In operational mode, this is the Shuttle function triggered by the EOR signal.
3004         //
3005
3006         if (run == -1)
3007                 Log("SHUTTLE","Collect - Shuttle called. Collecting conditions data for unprocessed runs");
3008         else
3009                 Log("SHUTTLE", Form("Collect - Shuttle called. Collecting conditions data for run %d", run));
3010
3011         SetLastAction("Starting");
3012
3013         // create ML instance
3014         if (!fMonaLisa)
3015                 fMonaLisa = new TMonaLisaWriter(fConfig->GetMonitorHost(), fConfig->GetMonitorTable());
3016                 
3017         CountOpenRuns();
3018
3019         TString whereClause("where shuttle_done=0");
3020         if (run != -1)
3021                 whereClause += Form(" and run=%d", run);
3022
3023         TObjArray shuttleLogbookEntries;
3024         if (!QueryShuttleLogbook(whereClause, shuttleLogbookEntries))
3025         {
3026                 Log("SHUTTLE", "Collect - Can't retrieve entries from Shuttle logbook");
3027                 return kFALSE;
3028         }
3029
3030         if (shuttleLogbookEntries.GetEntries() == 0)
3031         {
3032                 if (run == -1)
3033                         Log("SHUTTLE","Collect - Found no UNPROCESSED runs in Shuttle logbook");
3034                 else
3035                         Log("SHUTTLE", Form("Collect - Run %d is already DONE "
3036                                                 "or it does not exist in Shuttle logbook", run));
3037                 return kTRUE;
3038         }
3039
3040         for (UInt_t iDet=0; iDet<NDetectors(); iDet++)
3041                 fFirstUnprocessed[iDet] = kTRUE;
3042
3043         if (run != -1)
3044         {
3045                 // query Shuttle logbook for earlier runs, check if some detectors are unprocessed,
3046                 // flag them into fFirstUnprocessed array
3047                 TString whereClauseBis(Form("where shuttle_done=0 and run < %d", run));
3048                 TObjArray tmpLogbookEntries;
3049                 if (!QueryShuttleLogbook(whereClauseBis, tmpLogbookEntries))
3050                 {
3051                         Log("SHUTTLE", "Collect - Can't retrieve entries from Shuttle logbook");
3052                         return kFALSE;
3053                 }
3054
3055                 TIter iter(&tmpLogbookEntries);
3056                 AliShuttleLogbookEntry* anEntry = 0;
3057                 while ((anEntry = dynamic_cast<AliShuttleLogbookEntry*> (iter.Next())))
3058                 {
3059                         for (UInt_t iDet=0; iDet<NDetectors(); iDet++)
3060                         {
3061                                 if (anEntry->GetDetectorStatus(iDet) == AliShuttleLogbookEntry::kUnprocessed)
3062                                 {
3063                                         AliDebug(2, Form("Run %d: setting %s as \"not first time unprocessed\"",
3064                                                         anEntry->GetRun(), GetDetName(iDet)));
3065                                         fFirstUnprocessed[iDet] = kFALSE;
3066                                 }
3067                         }
3068
3069                 }
3070
3071         }
3072
3073         if (!RetrieveConditionsData(shuttleLogbookEntries))
3074         {
3075                 Log("SHUTTLE", "Collect - Process of at least one run failed");
3076                 CountOpenRuns();
3077                 return kFALSE;
3078         }
3079
3080         Log("SHUTTLE", "Collect - Requested run(s) successfully processed");
3081         CountOpenRuns();
3082         return kTRUE;
3083 }
3084
3085 //______________________________________________________________________________________________
3086 Bool_t AliShuttle::RetrieveConditionsData(const TObjArray& dateEntries)
3087 {
3088         //
3089         // Retrieve conditions data for all runs that aren't processed yet
3090         //
3091
3092         Bool_t hasError = kFALSE;
3093
3094         TIter iter(&dateEntries);
3095         AliShuttleLogbookEntry* anEntry;
3096
3097         while ((anEntry = (AliShuttleLogbookEntry*) iter.Next())){
3098                 if (!Process(anEntry)){
3099                         hasError = kTRUE;
3100                 }
3101
3102                 // clean SHUTTLE temp directory
3103                 //TString filename = Form("%s/*.shuttle", GetShuttleTempDir());
3104                 //RemoveFile(filename.Data());
3105         }
3106
3107         return hasError == kFALSE;
3108 }
3109
3110 //______________________________________________________________________________________________
3111 ULong_t AliShuttle::GetTimeOfLastAction() const
3112 {
3113         //
3114         // Gets time of last action
3115         //
3116
3117         ULong_t tmp;
3118
3119         fMonitoringMutex->Lock();
3120
3121         tmp = fLastActionTime;
3122
3123         fMonitoringMutex->UnLock();
3124
3125         return tmp;
3126 }
3127
3128 //______________________________________________________________________________________________
3129 const TString AliShuttle::GetLastAction() const
3130 {
3131         //
3132         // returns a string description of the last action
3133         //
3134
3135         TString tmp;
3136
3137         fMonitoringMutex->Lock();
3138         
3139         tmp = fLastAction;
3140         
3141         fMonitoringMutex->UnLock();
3142
3143         return tmp;
3144 }
3145
3146 //______________________________________________________________________________________________
3147 void AliShuttle::SetLastAction(const char* action)
3148 {
3149         //
3150         // updates the monitoring variables
3151         //
3152
3153         fMonitoringMutex->Lock();
3154
3155         fLastAction = action;
3156         fLastActionTime = time(0);
3157         
3158         fMonitoringMutex->UnLock();
3159 }
3160
3161 //______________________________________________________________________________________________
3162 const char* AliShuttle::GetRunParameter(const char* param)
3163 {
3164         //
3165         // returns run parameter read from DAQ logbook
3166         //
3167
3168         if(!fLogbookEntry) {
3169                 AliError("No logbook entry!");
3170                 return 0;
3171         }
3172
3173         return fLogbookEntry->GetRunParameter(param);
3174 }
3175
3176 //______________________________________________________________________________________________
3177 AliCDBEntry* AliShuttle::GetFromOCDB(const char* detector, const AliCDBPath& path)
3178 {
3179         //
3180         // returns object from OCDB valid for current run
3181         //
3182
3183         if (fTestMode & kErrorOCDB)
3184         {
3185                 Log(detector, "GetFromOCDB - In TESTMODE - Simulating error with OCDB");
3186                 return 0;
3187         }
3188         
3189         AliCDBStorage *sto = AliCDBManager::Instance()->GetStorage(fgkMainCDB);
3190         if (!sto)
3191         {
3192                 Log(detector, "GetFromOCDB - Cannot activate main OCDB for query!");
3193                 return 0;
3194         }
3195
3196         return dynamic_cast<AliCDBEntry*> (sto->Get(path, GetCurrentRun()));
3197 }
3198
3199 //______________________________________________________________________________________________
3200 Bool_t AliShuttle::SendMail(EMailTarget target, Int_t system)
3201 {
3202         //
3203         // sends a mail to the subdetector expert in case of preprocessor error
3204         //
3205         
3206         if (fTestMode != kNone)
3207                 return kTRUE;
3208                 
3209         if (!fConfig->SendMail()) 
3210                 return kTRUE;
3211
3212         if (target == kDCSEMail || target == kFXSEMail) {
3213                 if (!fFirstProcessing)
3214                         return kTRUE;
3215         }
3216
3217         Int_t runMode = (Int_t)fConfig->GetRunMode();
3218         TString tmpStr;
3219         if (runMode == 0) tmpStr = " Nightly Test:";
3220         else tmpStr = " Data Taking:"; 
3221         void* dir = gSystem->OpenDirectory(GetShuttleLogDir());
3222         if (dir == NULL)
3223         {
3224                 if (gSystem->mkdir(GetShuttleLogDir(), kTRUE))
3225                 {
3226                         Log("SHUTTLE", Form("SendMail - Can't open directory <%s>", GetShuttleLogDir()));
3227                         return kFALSE;
3228                 }
3229
3230         } else {
3231                 gSystem->FreeDirectory(dir);
3232         }
3233
3234         // det experts in to
3235         TString to="";
3236         TIter *iterExperts = 0;
3237         if (target == kDCSEMail) {
3238                 iterExperts = new TIter(fConfig->GetAdmins(AliShuttleConfig::kAmanda));
3239         }
3240         else if (target == kFXSEMail) {
3241                 iterExperts = new TIter(fConfig->GetAdmins(system));
3242         }
3243         if (iterExperts) {
3244                 TObjString *anExpert=0;
3245                 while ((anExpert = (TObjString*) iterExperts->Next()))
3246                 {
3247                         to += Form("%s,", anExpert->GetName());
3248                 }
3249                 delete iterExperts;
3250         }
3251
3252         // add subdetector experts      
3253         iterExperts = new TIter(fConfig->GetResponsibles(fCurrentDetector));
3254         TObjString *anExpert=0;
3255         while ((anExpert = (TObjString*) iterExperts->Next()))
3256         {
3257                 to += Form("%s,", anExpert->GetName());
3258         }
3259         delete iterExperts;
3260         
3261         if (to.Length() > 0)
3262           to.Remove(to.Length()-1);
3263         AliDebug(2, Form("to: %s",to.Data()));
3264
3265         if (to.IsNull()) {
3266                 Log("SHUTTLE", Form("List of %d responsibles not set!", (Int_t) target));
3267                 return kFALSE;
3268         }
3269
3270         // SHUTTLE responsibles in cc
3271         TString cc="";
3272         TIter iterAdmins(fConfig->GetAdmins(AliShuttleConfig::kGlobal));
3273         TObjString *anAdmin=0;
3274         while ((anAdmin = (TObjString*) iterAdmins.Next()))
3275         {
3276                 cc += Form("%s,", anAdmin->GetName());
3277         }
3278         if (cc.Length() > 0)
3279           cc.Remove(cc.Length()-1);
3280         AliDebug(2, Form("cc: %s",to.Data()));
3281
3282         // mail body 
3283         TString bodyFileName;
3284         bodyFileName.Form("%s/mail.body", GetShuttleLogDir());
3285         gSystem->ExpandPathName(bodyFileName);
3286
3287         ofstream mailBody;
3288         mailBody.open(bodyFileName, ofstream::out);
3289
3290         if (!mailBody.is_open())
3291         {
3292                 Log("SHUTTLE", Form("Could not open mail body file %s", bodyFileName.Data()));
3293                 return kFALSE;
3294         }
3295
3296
3297         TString subject;
3298         TString body;
3299
3300         if (target == kDCSEMail){
3301                 subject = Form("%s CRITICAL Retrieval of data points for %s FAILED in run %d !",
3302                                 tmpStr.Data(), fCurrentDetector.Data(), GetCurrentRun());
3303                 AliDebug(2, Form("subject: %s", subject.Data()));
3304                 
3305                 body = Form("Dear DCS experts, \n\n");
3306                 body += Form("SHUTTLE couldn\'t retrieve the data points for detector %s "
3307                              "in run %d!!\n\n", fCurrentDetector.Data(), GetCurrentRun());
3308         }
3309         else if (target == kFXSEMail){
3310                 subject = Form("%s CRITICAL FXS communication for %s FAILED in run %d !",
3311                                 tmpStr.Data(), fCurrentDetector.Data(), GetCurrentRun());
3312                 AliDebug(2, Form("subject: %s", subject.Data()));
3313                 TString sys;
3314                 if (system == kDAQ) sys="DAQ";
3315                 else if (system == kDCS) sys="DCS";
3316                 else if (system == kHLT) sys="HLT";
3317                 else return kFALSE;
3318                 body = Form("Dear  %s FXS experts, \n\n",sys.Data());
3319                 body += Form("SHUTTLE couldn\'t retrieve data from the FXS for detector %s "
3320                              "in run %d!!\n\n", fCurrentDetector.Data(), GetCurrentRun());
3321         }
3322         else {
3323                 subject = Form("%s %s Shuttle preprocessor FAILED in run %d (run type = %s)!",
3324                                        tmpStr.Data(), fCurrentDetector.Data(), GetCurrentRun(), GetRunType());
3325                 AliDebug(2, Form("subject: %s", subject.Data()));
3326         
3327                 body = Form("Dear %s expert(s), \n\n", fCurrentDetector.Data());
3328                 body += Form("SHUTTLE just detected that your preprocessor "
3329                              "failed processing run %d (run type = %s)!!\n\n", 
3330                              GetCurrentRun(), GetRunType());
3331         }
3332
3333         body += Form("Please check %s status on the SHUTTLE monitoring page: \n\n", 
3334                                 fCurrentDetector.Data());
3335         if (fConfig->GetRunMode() == AliShuttleConfig::kTest)
3336         {
3337                 body += Form("\thttp://pcalimonitor.cern.ch/shuttle.jsp?time=24 \n\n");
3338         } else {
3339                 body += Form("\thttp://pcalimonitor.cern.ch/shuttle.jsp?instance=PROD&time=24 \n\n");
3340         }
3341         
3342         
3343         TString logFolder = "logs";
3344         if (fConfig->GetRunMode() == AliShuttleConfig::kProd) 
3345                 logFolder += "_PROD";
3346         
3347         
3348         body += Form("Find the %s log for the current run on \n\n"
3349                 "\thttp://pcalishuttle02.cern.ch/%s/%d/%d/%s.log \n\n", 
3350                      fCurrentDetector.Data(), logFolder.Data(), GetCurrentRun()/10000,  
3351                                 GetCurrentRun(), fCurrentDetector.Data());
3352         body += Form("The last 15 lines of %s log file are following:\n\n", fCurrentDetector.Data());
3353
3354         AliDebug(2, Form("Body begin: %s", body.Data()));
3355
3356         mailBody << body.Data();
3357         mailBody.close();
3358         mailBody.open(bodyFileName, ofstream::out | ofstream::app);
3359
3360         TString logFileName = Form("%s/%d/%d/%s.log", GetShuttleLogDir(), 
3361                 GetCurrentRun()/10000, GetCurrentRun(), fCurrentDetector.Data());
3362         TString tailCommand = Form("tail -n 15 %s >> %s", logFileName.Data(), bodyFileName.Data());
3363         if (gSystem->Exec(tailCommand.Data()))
3364         {
3365                 mailBody << Form("%s log file not found ...\n\n", fCurrentDetector.Data());
3366         }
3367
3368         TString endBody = Form("------------------------------------------------------\n\n");
3369         endBody += Form("In case of problems please contact the SHUTTLE core team.\n\n");
3370         endBody += "Please do not answer this message directly, it is automatically generated.\n\n";
3371         endBody += "Greetings,\n\n \t\t\tthe SHUTTLE\n";
3372
3373         AliDebug(2, Form("Body end: %s", endBody.Data()));
3374
3375         mailBody << endBody.Data();
3376
3377         mailBody.close();
3378
3379         // send mail!
3380         TString mailCommand = Form("mail -s \"%s\" -c %s %s < %s",
3381                                                 subject.Data(),
3382                                                 cc.Data(),
3383                                                 to.Data(),
3384                                                 bodyFileName.Data());
3385         AliDebug(2, Form("mail command: %s", mailCommand.Data()));
3386
3387         Bool_t result = gSystem->Exec(mailCommand.Data());
3388
3389         return result == 0;
3390 }
3391 //______________________________________________________________________________________________
3392 const char* AliShuttle::GetRunType()
3393 {
3394         //
3395         // returns run type read from "run type" logbook
3396         //
3397
3398         if(!fLogbookEntry) {
3399                 AliError("No logbook entry!");
3400                 return 0;
3401         }
3402
3403         return fLogbookEntry->GetRunType();
3404 }
3405
3406 //______________________________________________________________________________________________
3407 Bool_t AliShuttle::GetHLTStatus()
3408 {
3409         // Return HLT status (ON=1 OFF=0)
3410         // Converts the HLT status from the mode string read in the run logbook (not just a bool)
3411
3412         if(!fLogbookEntry) {
3413                 AliError("No logbook entry!");
3414                 return 0;
3415         }
3416
3417         // TODO implement when HLTMode is inserted in run logbook
3418         TString hltMode = fLogbookEntry->GetRunParameter("HLTmode");
3419         TSubString firstChar = hltMode(0,1);
3420         AliDebug(2,Form("First char = %s ",firstChar.Data())); 
3421         if (firstChar == "A") {
3422                 return kFALSE;
3423         }
3424         else if ((firstChar == "B") || (firstChar == "C") || (firstChar == "D") || (firstChar == "E")) {
3425                 return kTRUE;
3426         }
3427         else {
3428                 Log("SHUTTLE","Unexpected HLT mode! Returning 0....");
3429                 return kFALSE;
3430         }
3431 }
3432
3433 //______________________________________________________________________________________________
3434 const char* AliShuttle::GetTriggerConfiguration()
3435 {
3436         // Receives the trigger configuration from the DAQ logbook for the current run
3437         
3438         // check connection, if needed reconnect
3439         if (!Connect(3)) 
3440                 return 0;
3441
3442         TString sqlQuery;
3443         sqlQuery.Form("SELECT configFile FROM logbook_trigger_config WHERE run = %d", GetCurrentRun());
3444         TSQLResult* result = fServer[3]->Query(sqlQuery);
3445         if (!result)
3446         {
3447                 Log("SHUTTLE", Form("ERROR: Can't execute query <%s>!", sqlQuery.Data()));
3448                 return 0;
3449         }
3450         
3451         if (result->GetRowCount() == 0)
3452         {
3453                 Log("SHUTTLE", "ERROR: Trigger configuration not found in logbook_trigger_config");
3454                 delete result;
3455                 return 0;
3456         }
3457         
3458         TSQLRow* row = result->Next();
3459         if (!row)
3460         {
3461                 Log("SHUTTLE", "ERROR: Could not receive logbook_trigger_config data");
3462                 delete result;
3463                 return 0;
3464         }
3465
3466         // static, so that pointer remains valid when it is returned to the calling class       
3467         static TString triggerConfig(row->GetField(0));
3468         
3469         delete row;
3470         row = 0;
3471         
3472         delete result;
3473         result = 0;
3474         
3475         Log("SHUTTLE", Form("Found trigger configuration: %s", triggerConfig.Data()));
3476         
3477         return triggerConfig;
3478 }
3479
3480 //______________________________________________________________________________________________
3481 void AliShuttle::SetShuttleTempDir(const char* tmpDir)
3482 {
3483         //
3484         // sets Shuttle temp directory
3485         //
3486
3487         fgkShuttleTempDir = gSystem->ExpandPathName(tmpDir);
3488 }
3489
3490 //______________________________________________________________________________________________
3491 void AliShuttle::SetShuttleLogDir(const char* logDir)
3492 {
3493         //
3494         // sets Shuttle log directory
3495         //
3496
3497         fgkShuttleLogDir = gSystem->ExpandPathName(logDir);
3498 }
3499 //______________________________________________________________________________________________
3500 Bool_t AliShuttle::TouchFile()
3501 {
3502         //
3503         // touching a file on the grid if run has been DONE
3504         //
3505         
3506         if (!gGrid)
3507         {
3508                 Log("SHUTTLE",Form("No TGrid connection estabilished!"));
3509                 Log("SHUTTLE",Form("Could not touch file for run %i",GetCurrentRun()));
3510                 return kFALSE;
3511         }
3512
3513         TString dir;
3514         dir.Form("%s%d/SHUTTLE_DONE", fConfig->GetAlienPath(), GetCurrentYear());
3515         // checking whether directory for touch command exists
3516         TString commandLs;