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