Enable TouchFile for empty SHUTTLE_DONE AliEn dir too.
[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 <stdexcept>
32 #include "AliShuttle.h"
33
34 #include "AliCDBManager.h"
35 #include "AliCDBStorage.h"
36 #include "AliCDBId.h"
37 #include "AliCDBRunRange.h"
38 #include "AliCDBPath.h"
39 #include "AliCDBEntry.h"
40 #include "AliShuttleConfig.h"
41 #include "DCSClient/AliDCSClient.h"
42 #include "AliLog.h"
43 #include "AliPreprocessor.h"
44 #include "AliShuttleStatus.h"
45 #include "AliShuttleLogbookEntry.h"
46
47 #include <TSystem.h>
48 #include <TObject.h>
49 #include <TString.h>
50 #include <TTimeStamp.h>
51 #include <TObjString.h>
52 #include <TSQLServer.h>
53 #include <TSQLResult.h>
54 #include <TSQLRow.h>
55 #include <TMutex.h>
56 #include <TSystemDirectory.h>
57 #include <TSystemFile.h>
58 #include <TFile.h>
59 #include <TGrid.h>
60 #include <TGridResult.h>
61 #include <TMap.h>
62
63 #include <TMonaLisaWriter.h>
64
65 #include <fstream>
66
67 #include <sys/types.h>
68 #include <sys/wait.h>
69
70 #include <signal.h>
71 using namespace std;
72
73 ClassImp(AliShuttle)
74
75 //______________________________________________________________________________________________
76 AliShuttle::AliShuttle(const AliShuttleConfig* config,
77                 UInt_t timeout, Int_t retries):
78 fConfig(config),
79 fTimeout(timeout), fRetries(retries),
80 fPreprocessorMap(),
81 fLogbookEntry(0),
82 fCurrentDetector(),
83 fFirstProcessing(0),
84 fFXSError(-1),
85 fStatusEntry(0),
86 fMonitoringMutex(0),
87 fLastActionTime(0),
88 fLastAction(),
89 fMonaLisa(0),
90 fTestMode(kNone),
91 fReadTestMode(kFALSE),
92 fOutputRedirected(kFALSE)
93 {
94         //
95         // config: AliShuttleConfig used
96         // timeout: timeout used for AliDCSClient connection
97         // retries: the number of retries in case of connection error.
98         //
99
100         if (!fConfig->IsValid()) AliFatal("********** !!!!! Invalid configuration !!!!! **********");
101         for(int iSys=0;iSys<5;iSys++) {
102                 fServer[iSys]=0;
103                 if (iSys < 4)
104                         fFXSlist[iSys].SetOwner(kTRUE);
105         }
106         fPreprocessorMap.SetOwner(kTRUE);
107
108         for (UInt_t iDet=0; iDet<NDetectors(); iDet++)
109                 fFirstUnprocessed[iDet] = kFALSE;
110
111         fMonitoringMutex = new TMutex();
112 }
113
114 //______________________________________________________________________________________________
115 AliShuttle::~AliShuttle()
116 {
117         //
118         // destructor
119         //
120
121         fPreprocessorMap.DeleteAll();
122         for(int iSys=0;iSys<5;iSys++)
123                 if(fServer[iSys]) {
124                         fServer[iSys]->Close();
125                         delete fServer[iSys];
126                         fServer[iSys] = 0;
127                 }
128
129         if (fStatusEntry){
130                 delete fStatusEntry;
131                 fStatusEntry = 0;
132         }
133         
134         if (fMonitoringMutex) 
135         {
136                 delete fMonitoringMutex;
137                 fMonitoringMutex = 0;
138         }
139 }
140
141 //______________________________________________________________________________________________
142 void AliShuttle::RegisterPreprocessor(AliPreprocessor* preprocessor)
143 {
144         //
145         // Registers new AliPreprocessor.
146         // It uses GetName() for indentificator of the pre processor.
147         // The pre processor is registered it there isn't any other
148         // with the same identificator (GetName()).
149         //
150
151         const char* detName = preprocessor->GetName();
152         if(GetDetPos(detName) < 0)
153                 AliFatal(Form("********** !!!!! Invalid detector name: %s !!!!! **********", detName));
154
155         if (fPreprocessorMap.GetValue(detName)) {
156                 AliWarning(Form("AliPreprocessor %s is already registered!", detName));
157                 return;
158         }
159
160         fPreprocessorMap.Add(new TObjString(detName), preprocessor);
161 }
162 //______________________________________________________________________________________________
163 Bool_t AliShuttle::Store(const AliCDBPath& path, TObject* object,
164                 AliCDBMetaData* metaData, Int_t validityStart, Bool_t validityInfinite)
165 {
166         // Stores a CDB object in the storage for offline reconstruction. Objects that are not needed for
167         // offline reconstruction, but should be stored anyway (e.g. for debugging) should NOT be stored
168         // using this function. Use StoreReferenceData instead!
169         // It calls StoreLocally function which temporarily stores the data locally; when the preprocessor
170         // finishes the data are transferred to the main storage (Grid).
171
172         return StoreLocally(fgkLocalCDB, path, object, metaData, validityStart, validityInfinite);
173 }
174
175 //______________________________________________________________________________________________
176 Bool_t AliShuttle::StoreReferenceData(const AliCDBPath& path, TObject* object, AliCDBMetaData* metaData)
177 {
178         // Stores a CDB object in the storage for reference data. This objects will not be available during
179         // offline reconstrunction. Use this function for reference data only!
180         // It calls StoreLocally function which temporarily stores the data locally; when the preprocessor
181         // finishes the data are transferred to the main storage (Grid).
182
183         return StoreLocally(fgkLocalRefStorage, path, object, metaData);
184 }
185
186 //______________________________________________________________________________________________
187 Bool_t AliShuttle::StoreLocally(const TString& localUri,
188                         const AliCDBPath& path, TObject* object, AliCDBMetaData* metaData,
189                         Int_t validityStart, Bool_t validityInfinite)
190 {
191         // Store object temporarily in local storage. Parameters are passed by Store and StoreReferenceData functions.
192         // when the preprocessor finishes the data are transferred to the main storage (Grid).
193         // The parameters are:
194         //   1) Uri of the backup storage (Local)
195         //   2) the object's path.
196         //   3) the object to be stored
197         //   4) the metaData to be associated with the object
198         //   5) the validity start run number w.r.t. the current run,
199         //      if the data is valid only for this run leave the default 0
200         //   6) specifies if the calibration data is valid for infinity (this means until updated),
201         //      typical for calibration runs, the default is kFALSE
202         //
203         // returns 0 if fail, 1 otherwise
204
205
206         if (fTestMode & kErrorStorage)
207         {
208                 Log(fCurrentDetector, "StoreLocally - In TESTMODE - Simulating error while storing locally");
209                 return kFALSE;
210         }
211         
212         const char* cdbType = (localUri == fgkLocalCDB) ? "CDB" : "Reference";
213
214         Int_t firstRun = GetCurrentRun() - validityStart;
215         if(firstRun < 0) {
216                 AliWarning("First valid run happens to be less than 0! Setting it to 0.");
217                 firstRun=0;
218         }
219
220         Int_t lastRun = -1;
221         if(validityInfinite) {
222                 lastRun = AliCDBRunRange::Infinity();
223         } else {
224                 lastRun = GetCurrentRun();
225         }
226
227         // Version is set to current run, it will be used later to transfer data to Grid
228         AliCDBId id(path, firstRun, lastRun, GetCurrentRun(), -1);
229
230         if(! dynamic_cast<TObjString*> (metaData->GetProperty("RunUsed(TObjString)"))){
231                 TObjString runUsed = Form("%d", GetCurrentRun());
232                 metaData->SetProperty("RunUsed(TObjString)", runUsed.Clone());
233         }
234
235         Bool_t result = kFALSE;
236
237         if (!(AliCDBManager::Instance()->GetStorage(localUri))) {
238                 Log("SHUTTLE", Form("StoreLocally - Cannot activate local %s storage", cdbType));
239         } else {
240                 Int_t logLevel = AliLog::GetGlobalLogLevel();
241                 AliLog::SetGlobalLogLevel(AliLog::kError);
242                 result = AliCDBManager::Instance()->GetStorage(localUri)
243                                         ->Put(object, id, metaData);
244                 AliLog::SetGlobalLogLevel((AliLog::EType_t)logLevel);
245         }
246
247         if(!result) {
248
249                 Log(fCurrentDetector, Form("StoreLocally - Can't store object <%s>!", id.ToString().Data()));
250         }
251
252
253         return result;
254 }
255
256 //______________________________________________________________________________________________
257 Bool_t AliShuttle::StoreOCDB()
258 {
259         //
260         // Called when preprocessor ends successfully or when previous storage attempt failed (kStoreError status)
261         // Calls underlying StoreOCDB(const char*) function twice, for OCDB and Reference storage.
262         // Then calls StoreRefFilesToGrid to store reference files. 
263         //
264         
265         UpdateShuttleStatus(AliShuttleStatus::kStoreStarted);
266                         
267         if (fTestMode & kErrorGrid)
268         {
269                 Log("SHUTTLE", "StoreOCDB - In TESTMODE - Simulating error while storing in the Grid");
270                 Log(fCurrentDetector, "StoreOCDB - In TESTMODE - Simulating error while storing in the Grid");
271                 return kFALSE;
272         }
273         
274         Log("SHUTTLE","StoreOCDB - Storing OCDB data ...");
275         Int_t resultCDB = StoreOCDB(fgkMainCDB);
276
277         Log("SHUTTLE","StoreOCDB - Storing reference data ...");
278         Int_t resultRef = StoreOCDB(fgkMainRefStorage);
279         
280         Log("SHUTTLE","StoreOCDB - Storing reference files ...");
281         Bool_t resultRefFiles = CopyFilesToGrid("reference");
282         
283         Bool_t resultMetadata = kTRUE;
284         if(fCurrentDetector == "GRP") 
285         {
286                 Log("SHUTTLE","StoreOCDB - Storing Run Metadata file ...");
287                 resultMetadata = CopyFilesToGrid("metadata");
288         }
289         
290         Int_t storeResult = 0;
291         
292         if (resultCDB < 0 || resultRef < 0 || resultRefFiles == kFALSE || resultMetadata == kFALSE)
293                 storeResult = -1;
294         else if (resultCDB > 0 || resultRef > 0)
295                 storeResult = 1;
296                 
297         if (storeResult < 0)
298         {
299                 Log("SHUTTLE", 
300                         Form("\t\t\t****** run %d - %s: STORAGE ERROR ******",
301                                 GetCurrentRun(), fCurrentDetector.Data()));
302                 UpdateShuttleStatus(AliShuttleStatus::kStoreError);
303         } 
304         else if (storeResult > 0)
305         {
306                 Log("SHUTTLE", 
307                         Form("\t\t\t****** run %d - %s: STORAGE DELAYED ******",
308                                 GetCurrentRun(), fCurrentDetector.Data()));
309                 UpdateShuttleStatus(AliShuttleStatus::kStoreDelayed);
310         }
311         else if (storeResult == 0) 
312         {
313                 Log("SHUTTLE", 
314                         Form("\t\t\t****** run %d - %s: DONE ******",
315                                 GetCurrentRun(), fCurrentDetector.Data()));
316                 UpdateShuttleStatus(AliShuttleStatus::kDone);
317                 UpdateShuttleLogbook(fCurrentDetector, "DONE");
318         }
319
320         return (storeResult == 0);
321 }
322
323 //______________________________________________________________________________________________
324 Int_t AliShuttle::StoreOCDB(const TString& gridURI)
325 {
326         //
327         // Called by StoreOCDB(), performs actual storage to the main OCDB and reference storages (Grid)
328         //
329         // Return code:
330         //   -2 initialization error
331         //   -1 storage error
332         //   0  success
333         //   1  storage delayed (e.g. previous unprocessed runs)
334         //
335
336         TObjArray* gridIds=0;
337
338         Bool_t result = kTRUE;
339         Bool_t delayed = kFALSE;
340
341         const char* type = 0;
342         TString localURI;
343         if(gridURI == fgkMainCDB) {
344                 type = "OCDB";
345                 localURI = fgkLocalCDB;
346         } else if(gridURI == fgkMainRefStorage) {
347                 type = "reference";
348                 localURI = fgkLocalRefStorage;
349         } else {
350                 AliError(Form("Invalid storage URI: %s", gridURI.Data()));
351                 return -2;
352         }
353
354         AliCDBManager* man = AliCDBManager::Instance();
355
356         AliCDBStorage *gridSto = man->GetStorage(gridURI);
357         if(!gridSto) {
358                 Log("SHUTTLE",
359                         Form("StoreOCDB - cannot activate main %s storage", type));
360                 return -2;
361         }
362         gridSto->SetMirrorSEs(fgkMirrorSEs.Data());
363
364         gridIds = gridSto->GetQueryCDBList();
365
366         // get objects previously stored in local CDB
367         AliCDBStorage *localSto = man->GetStorage(localURI);
368         if(!localSto) {
369                 Log("SHUTTLE",
370                         Form("StoreOCDB - cannot activate local %s storage", type));
371                 return -2;
372         }
373         AliCDBPath aPath(GetOfflineDetName(fCurrentDetector.Data()),"*","*");
374         // Local objects were stored with current run as Grid version!
375         TList* localEntries = localSto->GetAll(aPath.GetPath(), GetCurrentRun(), GetCurrentRun());
376         localEntries->SetOwner(1);
377
378         // loop on local stored objects
379         TIter localIter(localEntries);
380         AliCDBEntry *aLocEntry = 0;
381         while((aLocEntry = dynamic_cast<AliCDBEntry*> (localIter.Next()))){
382                 aLocEntry->SetOwner(1);
383                 AliCDBId aLocId = aLocEntry->GetId();
384                 aLocEntry->SetVersion(-1);
385                 aLocEntry->SetSubVersion(-1);
386                                                 
387                 Log(fCurrentDetector.Data(), Form("Attempting to store %s", aLocId.ToString().Data()));
388
389                 // If local object is valid up to infinity we store it only if it is
390                 // the first unprocessed run!
391                 if (aLocId.GetLastRun() == AliCDBRunRange::Infinity() &&
392                         !fFirstUnprocessed[GetDetPos(fCurrentDetector)])
393                 {
394                         Log("SHUTTLE", Form("StoreOCDB - %s: object %s has validity infinite but "
395                                                 "there are previous unprocessed runs!",
396                                                 fCurrentDetector.Data(), aLocId.GetPath().Data()));
397                         Log(fCurrentDetector.Data(), Form("StoreOCDB - %s: object %s has validity infinite but "
398                                                 "there are previous unprocessed runs!",
399                                                 fCurrentDetector.Data(), aLocId.GetPath().Data()));
400                         delayed = kTRUE;
401                         continue;
402                 }
403
404                 // loop on Grid valid Id's
405                 Bool_t store = kTRUE;
406                 TIter gridIter(gridIds);
407                 AliCDBId* aGridId = 0;
408                 while ((aGridId = dynamic_cast<AliCDBId*> (gridIter.Next()))) {
409                         if (aGridId->GetPath() != aLocId.GetPath()) 
410                                 continue;
411                         // skip all objects valid up to infinity
412                         if (aGridId->GetLastRun() == AliCDBRunRange::Infinity()) 
413                                 continue;
414                         
415                         // if we get here, it means there's already some more recent object stored on Grid!
416                         Log(fCurrentDetector.Data(),
417                                 Form("StoreOCDB - A more recent object already exists in %s storage: <%s>",
418                                 type, aGridId->ToString().Data()));
419                         
420                         store = kFALSE;
421                         break;
422                 }
423
424                 Bool_t storeOk = kFALSE;
425                 if (store)
426                 {
427                         Log(fCurrentDetector.Data(), Form("Prechecks succeeded. Ready to store %s", aLocId.ToString().Data()));
428                         storeOk = gridSto->Put(aLocEntry);
429                         if (storeOk) {
430                                 Log("SHUTTLE",
431                                 Form("StoreOCDB - Object <%s> successfully put into %s storage",
432                                         aLocId.ToString().Data(), type));
433                                 Log(fCurrentDetector.Data(),
434                                         Form("StoreOCDB - Object <%s> successfully put into %s storage",
435                                         aLocId.ToString().Data(), type));
436                         } else  {
437                                 Log("SHUTTLE",
438                                         Form("StoreOCDB - Grid %s storage of object <%s> failed",
439                                         type, aLocId.ToString().Data()));
440                                 Log(fCurrentDetector.Data(),
441                                         Form("StoreOCDB - Grid %s storage of object <%s> failed",
442                                         type, aLocId.ToString().Data()));
443                                 result = kFALSE;
444                         }
445                 }
446                 
447                 if (!store || storeOk) {
448                         // removing local file...
449                         TString filename;
450                         localSto->IdToFilename(aLocId, filename);
451                         Log("SHUTTLE", Form("StoreOCDB - Removing local file %s", filename.Data()));
452                         RemoveFile(filename.Data());
453                 }
454         }
455         localEntries->Clear();
456
457         Int_t returnCode = 0;
458         
459         if (result == kFALSE)
460                 returnCode = -1;
461         else if (delayed != kFALSE)
462                 returnCode =  1;
463
464         Log("SHUTTLE", Form("StoreOCDB - Returning with %d (result = %d, delayed = %d)", returnCode, result, delayed));
465         Log(fCurrentDetector.Data(), Form("StoreOCDB - Returning with %d (result = %d, delayed = %d)", returnCode, result, delayed));
466         
467         return returnCode;
468 }
469
470 //______________________________________________________________________________________________
471 Bool_t AliShuttle::CleanReferenceStorage(const char* detector)
472 {
473         // clears the directory used to store reference files of a given subdetector
474   
475         AliCDBManager* man = AliCDBManager::Instance();
476         AliCDBStorage* sto = man->GetStorage(fgkLocalRefStorage);
477         TString localBaseFolder = sto->GetBaseFolder();
478
479         TString targetDir = GetRefFilePrefix(localBaseFolder.Data(), detector);
480         
481         Log("SHUTTLE", Form("CleanReferenceStorage - Cleaning %s", targetDir.Data()));
482
483         TString begin;
484         begin.Form("%d_", GetCurrentRun());
485         
486         TSystemDirectory* baseDir = new TSystemDirectory("/", targetDir);
487         if (!baseDir)
488                 return kTRUE;
489                 
490         TList* dirList = baseDir->GetListOfFiles();
491         delete baseDir;
492         
493         if (!dirList) return kTRUE;
494                         
495         if (dirList->GetEntries() < 3) // to be changed to 4?
496         {
497                 delete dirList;
498                 return kTRUE;
499         }
500                                 
501         Int_t nDirs = 0, nDel = 0;
502         TIter dirIter(dirList);
503         TSystemFile* entry = 0;
504
505         Bool_t success = kTRUE;
506         
507         while ((entry = dynamic_cast<TSystemFile*> (dirIter.Next())))
508         {                                       
509                 if (entry->IsDirectory())
510                         continue;
511                 
512                 TString fileName(entry->GetName());
513                 if (!fileName.BeginsWith(begin))
514                         continue;
515                         
516                 nDirs++;
517                                                 
518                 // delete file
519                 Int_t result = gSystem->Unlink(fileName.Data());
520                 
521                 if (result)
522                 {
523                         Log("SHUTTLE", Form("CleanReferenceStorage - Could not delete file %s!", fileName.Data()));
524                         success = kFALSE;
525                 } else {
526                         nDel++;
527                 }
528         }
529
530         if(nDirs > 0)
531                 Log("SHUTTLE", Form("CleanReferenceStorage - %d (over %d) reference files in folder %s were deleted.", 
532                         nDel, nDirs, targetDir.Data()));
533
534                 
535         delete dirList;
536         return success;
537
538
539
540
541
542
543   Int_t result = gSystem->GetPathInfo(targetDir, 0, (Long64_t*) 0, 0, 0);
544   if (result == 0)
545   {
546     // delete directory
547     result = gSystem->Exec(Form("rm -rf %s", targetDir.Data()));
548     if (result != 0)
549     {  
550       Log("SHUTTLE", Form("CleanReferenceStorage - Could not clean directory %s", targetDir.Data()));
551       return kFALSE;
552     }
553   }
554
555   result = gSystem->mkdir(targetDir, kTRUE);
556   if (result != 0)
557   {
558     Log("SHUTTLE", Form("CleanReferenceStorage - Error creating base directory %s", targetDir.Data()));
559     return kFALSE;
560   }
561         
562   return kTRUE;
563 }
564
565 //______________________________________________________________________________________________
566 Bool_t AliShuttle::StoreReferenceFile(const char* detector, const char* localFile, const char* gridFileName)
567 {
568         //
569         // Stores reference file directly (without opening it). This function stores the file locally.
570         //
571         // The file is stored under the following location: 
572         // <base folder of local reference storage>/<DET>/<RUN#>_<gridFileName>
573         // where <gridFileName> is the second parameter given to the function
574         // 
575         
576         if (fTestMode & kErrorStorage)
577         {
578                 Log(fCurrentDetector, "StoreReferenceFile - In TESTMODE - Simulating error while storing locally");
579                 return kFALSE;
580         }
581         
582         AliCDBManager* man = AliCDBManager::Instance();
583         AliCDBStorage* sto = man->GetStorage(fgkLocalRefStorage);
584         
585         TString localBaseFolder = sto->GetBaseFolder();
586         
587         TString target = GetRefFilePrefix(localBaseFolder.Data(), detector);    
588         target.Append(Form("/%d_%s", GetCurrentRun(), gridFileName));
589         
590         return CopyFileLocally(localFile, target);
591 }
592
593 //______________________________________________________________________________________________
594 Bool_t AliShuttle::StoreRunMetadataFile(const char* localFile, const char* gridFileName)
595 {
596         //
597         // Stores Run metadata file to the Grid, in the run folder
598         //
599         // Only GRP can call this function.
600         
601         if (fTestMode & kErrorStorage)
602         {
603                 Log(fCurrentDetector, "StoreRunMetaDataFile - In TESTMODE - Simulating error while storing locally");
604                 return kFALSE;
605         }
606         
607         AliCDBManager* man = AliCDBManager::Instance();
608         AliCDBStorage* sto = man->GetStorage(fgkLocalRefStorage);
609         
610         TString localBaseFolder = sto->GetBaseFolder();
611         
612         // Build Run level folder
613         // folder = /alice/data/year/lhcPeriod/runNb/raw
614         
615                 
616         TString lhcPeriod = GetLHCPeriod();     
617         if (lhcPeriod.Length() == 0) 
618         {
619                 Log("SHUTTLE","StoreRunMetaDataFile - LHCPeriod not found in logbook!");
620                 return 0;
621         }
622         
623         // TODO partitions with one detector only write data into LHCperiod_DET
624         TString partition = GetRunParameter("detector");
625         
626         if (partition.Length() > 0 && partition != "ALICE")
627         {
628                 lhcPeriod.Append(Form("_%s", partition.Data()));
629                 Log(fCurrentDetector, Form("Run data tags merged file will be written in %s", 
630                                 lhcPeriod.Data()));
631         }
632                 
633         TString target = Form("%s/GRP/RunMetadata%s%d/%s/%09d/raw/%s", 
634                                 localBaseFolder.Data(), fConfig->GetAlienPath(), GetCurrentYear(), 
635                                 lhcPeriod.Data(), GetCurrentRun(), gridFileName);
636                                         
637         return CopyFileLocally(localFile, target);
638 }
639
640 //______________________________________________________________________________________________
641 Bool_t AliShuttle::CopyFileLocally(const char* localFile, const TString& target)
642 {
643         //
644         // Stores file locally. Called by StoreReferenceFile and StoreRunMetadataFile
645         // Files are temporarily stored in the local reference storage. When the preprocessor 
646         // finishes, the Shuttle calls CopyFilesToGrid to transfer the files to AliEn 
647         // (in reference or run level folders)
648         //
649         
650         TString targetDir(target(0, target.Last('/')));
651         
652         //try to open base dir folder, if it does not exist
653         void* dir = gSystem->OpenDirectory(targetDir.Data());
654         if (dir == NULL) {
655                 if (gSystem->mkdir(targetDir.Data(), kTRUE)) {
656                         Log("SHUTTLE", Form("CopyFileLocally - Can't open directory <%s>", targetDir.Data()));
657                         return kFALSE;
658                 }
659
660         } else {
661                 gSystem->FreeDirectory(dir);
662         }
663         
664         Int_t result = 0;
665         
666         result = gSystem->GetPathInfo(localFile, 0, (Long64_t*) 0, 0, 0);
667         if (result)
668         {
669                 Log("SHUTTLE", Form("CopyFileLocally - %s does not exist", localFile));
670                 return kFALSE;
671         }
672
673         result = gSystem->GetPathInfo(target, 0, (Long64_t*) 0, 0, 0);
674         if (!result)
675         {
676                 Log("SHUTTLE", Form("CopyFileLocally - target file %s already exist, removing...", target.Data()));
677                 if (gSystem->Unlink(target.Data()))
678                 {
679                         Log("SHUTTLE", Form("CopyFileLocally - Could not remove existing target file %s!", target.Data()));
680                         return kFALSE;
681                 }
682         }       
683         
684         result = gSystem->CopyFile(localFile, target);
685
686         if (result == 0)
687         {
688                 Log("SHUTTLE", Form("CopyFileLocally - File %s stored locally to %s", localFile, target.Data()));
689                 return kTRUE;
690         }
691         else
692         {
693                 Log("SHUTTLE", Form("CopyFileLocally - Could not store file %s to %s! Error code = %d", 
694                                 localFile, target.Data(), result));
695                 return kFALSE;
696         }       
697
698
699
700 }
701
702 //______________________________________________________________________________________________
703 Bool_t AliShuttle::CopyFilesToGrid(const char* type)
704 {
705         //
706         // Transfers local files to the Grid. Local files can be reference files 
707         // or run metadata file (from GRP only).
708         //
709         // According to the type (ref, metadata) the files are stored under the following location: 
710         // ref --> <base folder of reference storage>/<DET>/<RUN#>_<gridFileName>
711         // metadata --> <run data folder>/<MetadataFileName>
712         //
713                 
714         AliCDBManager* man = AliCDBManager::Instance();
715         AliCDBStorage* sto = man->GetStorage(fgkLocalRefStorage);
716         if (!sto)
717                 return kFALSE;
718         TString localBaseFolder = sto->GetBaseFolder();
719         
720         TString dir;
721         TString alienDir;
722         TString begin;
723         
724         if (strcmp(type, "reference") == 0) 
725         {
726                 dir = GetRefFilePrefix(localBaseFolder.Data(), fCurrentDetector.Data());
727                 AliCDBStorage* gridSto = man->GetStorage(fgkMainRefStorage);
728                 if (!gridSto)
729                         return kFALSE;
730                 TString gridBaseFolder = gridSto->GetBaseFolder();
731                 alienDir = GetRefFilePrefix(gridBaseFolder.Data(), fCurrentDetector.Data());
732                 begin = Form("%d_", GetCurrentRun());
733         } 
734         else if (strcmp(type, "metadata") == 0)
735         {
736                         
737                 TString lhcPeriod = GetLHCPeriod();
738         
739                 if (lhcPeriod.Length() == 0) 
740                 {
741                         Log("SHUTTLE","CopyFilesToGrid - LHCPeriod not found in logbook!");
742                         return 0;
743                 }
744                 
745                 // TODO partitions with one detector only write data into LHCperiod_DET
746                 TString partition = GetRunParameter("detector");
747         
748                 if (partition.Length() > 0 && partition != "ALICE")
749                 {
750                         lhcPeriod.Append(Form("_%s", partition.Data()));
751                 }
752                 
753                 dir = Form("%s/GRP/RunMetadata%s%d/%s/%09d/raw", 
754                                 localBaseFolder.Data(), fConfig->GetAlienPath(), GetCurrentYear(), 
755                                 lhcPeriod.Data(), GetCurrentRun());
756                 alienDir = dir(dir.Index(fConfig->GetAlienPath()), dir.Length());
757                 
758                 begin = "";
759         }
760         else 
761         {
762                 Log("SHUTTLE", "CopyFilesToGrid - Unexpected: type label must be reference or metadata!");
763                 return kFALSE;
764         }
765                 
766         TSystemDirectory* baseDir = new TSystemDirectory("/", dir);
767         if (!baseDir)
768                 return kTRUE;
769                 
770         TList* dirList = baseDir->GetListOfFiles();
771         delete baseDir;
772         
773         if (!dirList) return kTRUE;
774                 
775         if (dirList->GetEntries() < 3) 
776         {
777                 delete dirList;
778                 return kTRUE;
779         }
780                         
781         if (!gGrid)
782         { 
783                 Log("SHUTTLE", "CopyFilesToGrid - Connection to Grid failed: Cannot continue!");
784                 delete dirList;
785                 return kFALSE;
786         }
787         
788         Int_t nDirs = 0, nTransfer = 0;
789         TIter dirIter(dirList);
790         TSystemFile* entry = 0;
791
792         Bool_t success = kTRUE;
793         Bool_t first = kTRUE;
794         
795         while ((entry = dynamic_cast<TSystemFile*> (dirIter.Next())))
796         {                       
797                 if (entry->IsDirectory())
798                         continue;
799                         
800                 TString fileName(entry->GetName());
801                 if (!fileName.BeginsWith(begin))
802                         continue;
803                         
804                 nDirs++;
805                         
806                 if (first)
807                 {
808                         first = kFALSE;
809                         // check that folder exists, otherwise create it
810                         TGridResult* result = gGrid->Ls(alienDir.Data(), "a");
811                         
812                         if (!result)
813                         {
814                                 delete dirList;
815                                 return kFALSE;
816                         }
817                         
818                         if (!result->GetFileName(1)) // TODO: It looks like element 0 is always 0!!
819                         {
820                                 // TODO It does not work currently! Bug in TAliEn::Mkdir
821                                 // TODO Manually fixed in local root v5-16-00
822                                 if (!gGrid->Mkdir(alienDir.Data(),"-p",0))
823                                 {
824                                         Log("SHUTTLE", Form("CopyFilesToGrid - Cannot create directory %s",
825                                                         alienDir.Data()));
826                                         delete dirList;
827                                         return kFALSE;
828                                 } else {
829                                         Log("SHUTTLE",Form("CopyFilesToGrid - Folder %s created", alienDir.Data()));
830                                 }
831                                 
832                         } else {
833                                         Log("SHUTTLE",Form("CopyFilesToGrid - Folder %s found", alienDir.Data()));
834                         }
835                 }
836                         
837                 TString fullLocalPath;
838                 fullLocalPath.Form("%s/%s", dir.Data(), fileName.Data());
839                 
840                 TString fullGridPath;
841                 fullGridPath.Form("alien://%s/%s", alienDir.Data(), fileName.Data());
842
843                 Bool_t result = TFile::Cp(fullLocalPath, fullGridPath);
844                 
845                 if (result)
846                 {
847                         Log("SHUTTLE", Form("CopyFilesToGrid - Copying local file %s to %s succeeded!", 
848                                                 fullLocalPath.Data(), fullGridPath.Data()));
849                         RemoveFile(fullLocalPath);
850                         nTransfer++;
851                 }
852                 else
853                 {
854                         Log("SHUTTLE", Form("CopyFilesToGrid - Copying local file %s to %s FAILED!", 
855                                                 fullLocalPath.Data(), fullGridPath.Data()));
856                         success = kFALSE;
857                 }
858         }
859
860         Log("SHUTTLE", Form("CopyFilesToGrid - %d (over %d) files in folder %s copied to Grid.", 
861                                                 nTransfer, nDirs, dir.Data()));
862
863                 
864         delete dirList;
865         return success;
866 }
867
868 //______________________________________________________________________________________________
869 const char* AliShuttle::GetRefFilePrefix(const char* base, const char* detector)
870 {
871         //
872         // Get folder name of reference files 
873         //
874
875         TString offDetStr(GetOfflineDetName(detector));
876         static TString dir;
877         if (offDetStr == "ITS" || offDetStr == "MUON" || offDetStr == "PHOS")
878         {
879                 dir.Form("%s/%s/%s", base, offDetStr.Data(), detector);
880         } else {
881                 dir.Form("%s/%s", base, offDetStr.Data());
882         }
883         
884         return dir.Data();
885 }
886
887 //______________________________________________________________________________________________
888 void AliShuttle::CleanLocalStorage(const TString& uri)
889 {
890         //
891         // Called in case the preprocessor is declared failed. Remove remaining objects from the local storages.
892         //
893
894         const char* type = 0;
895         if(uri == fgkLocalCDB) {
896                 type = "OCDB";
897         } else if(uri == fgkLocalRefStorage) {
898                 type = "Reference";
899         } else {
900                 AliError(Form("Invalid storage URI: %s", uri.Data()));
901                 return;
902         }
903
904         AliCDBManager* man = AliCDBManager::Instance();
905
906         // open local storage
907         AliCDBStorage *localSto = man->GetStorage(uri);
908         if(!localSto) {
909                 Log("SHUTTLE",
910                         Form("CleanLocalStorage - cannot activate local %s storage", type));
911                 return;
912         }
913
914         TString filename(Form("%s/%s/*/Run*_v%d_s*.root",
915                 localSto->GetBaseFolder().Data(), GetOfflineDetName(fCurrentDetector.Data()), GetCurrentRun()));
916
917         AliDebug(2, Form("filename = %s", filename.Data()));
918
919         Log("SHUTTLE", Form("Removing remaining local files for run %d and detector %s ...",
920                 GetCurrentRun(), fCurrentDetector.Data()));
921
922         RemoveFile(filename.Data());
923
924 }
925
926 //______________________________________________________________________________________________
927 void AliShuttle::RemoveFile(const char* filename)
928 {
929         //
930         // removes local file
931         //
932
933         TString command(Form("rm -f %s", filename));
934
935         Int_t result = gSystem->Exec(command.Data());
936         if(result != 0)
937         {
938                 Log("SHUTTLE", Form("RemoveFile - %s: Cannot remove file %s!",
939                         fCurrentDetector.Data(), filename));
940         }
941 }
942
943 //______________________________________________________________________________________________
944 AliShuttleStatus* AliShuttle::ReadShuttleStatus()
945 {
946         //
947         // Reads the AliShuttleStatus from the CDB
948         //
949
950         if (fStatusEntry){
951                 delete fStatusEntry;
952                 fStatusEntry = 0;
953         }
954
955         Int_t path1 = GetCurrentRun()/10000;
956         try{
957                 fStatusEntry = AliCDBManager::Instance()->GetStorage(GetLocalCDB())
958                         ->Get(Form("/SHUTTLE/%s/%d", fCurrentDetector.Data(), path1), GetCurrentRun());
959         } catch(std::exception& x) {
960                 AliInfo(TString::Format("%s",x.what()));
961         }
962
963         if (!fStatusEntry) return 0;
964         fStatusEntry->SetOwner(1);
965
966         AliShuttleStatus* status = dynamic_cast<AliShuttleStatus*> (fStatusEntry->GetObject());
967         if (!status) {
968                 AliError("Invalid object stored to CDB!");
969                 return 0;
970         }
971
972         return status;
973 }
974
975 //______________________________________________________________________________________________
976 Bool_t AliShuttle::WriteShuttleStatus(AliShuttleStatus* status)
977 {
978         //
979         // writes the status for one subdetector
980         //
981
982         if (fStatusEntry){
983                 delete fStatusEntry;
984                 fStatusEntry = 0;
985         }
986
987         Int_t run = GetCurrentRun();
988         Int_t path1 = run/10000;
989         TString path1_string = Form("%d",path1);
990
991         AliCDBId id(AliCDBPath("SHUTTLE", fCurrentDetector, path1_string), run, run);
992
993         fStatusEntry = new AliCDBEntry(status, id, new AliCDBMetaData);
994         fStatusEntry->SetOwner(1);
995
996         Int_t logLevel = AliLog::GetGlobalLogLevel();
997         AliLog::SetGlobalLogLevel(AliLog::kError);
998
999         UInt_t result = AliCDBManager::Instance()->GetStorage(fgkLocalCDB)->Put(fStatusEntry);
1000
1001         AliLog::SetGlobalLogLevel((AliLog::EType_t)logLevel);
1002
1003         if (!result) {
1004                 Log("SHUTTLE", Form("WriteShuttleStatus - Failed for %s, run %d",
1005                                                 fCurrentDetector.Data(), run));
1006                 return kFALSE;
1007         }
1008         
1009         SendMLDetInfo();
1010
1011         return kTRUE;
1012 }
1013
1014 //______________________________________________________________________________________________
1015 void AliShuttle::UpdateShuttleStatus(AliShuttleStatus::Status newStatus, Bool_t increaseCount)
1016 {
1017         //
1018         // changes the AliShuttleStatus for the given detector and run to the given status
1019         //
1020
1021         if (!fStatusEntry){
1022                 AliError("UNEXPECTED: fStatusEntry empty");
1023                 return;
1024         }
1025
1026         AliShuttleStatus* status = dynamic_cast<AliShuttleStatus*> (fStatusEntry->GetObject());
1027
1028         if (!status){
1029                 Log("SHUTTLE", "UpdateShuttleStatus - UNEXPECTED: status could not be read from current CDB entry");
1030                 return;
1031         }
1032
1033         TString actionStr = Form("UpdateShuttleStatus - %s: Changing state from %s to %s",
1034                                 fCurrentDetector.Data(),
1035                                 status->GetStatusName(),
1036                                 status->GetStatusName(newStatus));
1037         Log("SHUTTLE", actionStr);
1038         SetLastAction(actionStr);
1039
1040         status->SetStatus(newStatus);
1041         if (increaseCount) status->IncreaseCount();
1042
1043         Int_t logLevel = AliLog::GetGlobalLogLevel();
1044         AliLog::SetGlobalLogLevel(AliLog::kError);
1045
1046         AliCDBManager::Instance()->GetStorage(fgkLocalCDB)->Put(fStatusEntry);
1047
1048         AliLog::SetGlobalLogLevel((AliLog::EType_t)logLevel);
1049
1050         SendMLDetInfo();
1051 }
1052
1053 //______________________________________________________________________________________________
1054 void AliShuttle::SendMLDetInfo()
1055 {
1056         //
1057         // sends ML information about the current status of the current detector being processed
1058         //
1059         
1060         AliShuttleStatus* status = dynamic_cast<AliShuttleStatus*> (fStatusEntry->GetObject());
1061         
1062         if (!status){
1063                 Log("SHUTTLE", "SendMLDetInfo - UNEXPECTED: status could not be read from current CDB entry");
1064                 return;
1065         }
1066         
1067         TMonaLisaText  mlStatus(Form("%s_status", fCurrentDetector.Data()), status->GetStatusName());
1068         TMonaLisaValue mlRetryCount(Form("%s_count", fCurrentDetector.Data()), status->GetCount());
1069
1070         TList mlList;
1071         mlList.Add(&mlStatus);
1072         mlList.Add(&mlRetryCount);
1073
1074         TString mlID;
1075         mlID.Form("%d", GetCurrentRun());
1076         fMonaLisa->SendParameters(&mlList, mlID);
1077 }
1078
1079 //______________________________________________________________________________________________
1080 Bool_t AliShuttle::ContinueProcessing()
1081 {
1082         // this function reads the AliShuttleStatus information from CDB and
1083         // checks if the processing should be continued
1084         // if yes it returns kTRUE and updates the AliShuttleStatus with nextStatus
1085
1086         if (!fConfig->HostProcessDetector(fCurrentDetector))
1087                 return kFALSE;
1088
1089         AliPreprocessor* aPreprocessor =
1090                 dynamic_cast<AliPreprocessor*> (fPreprocessorMap.GetValue(fCurrentDetector));
1091         if (!aPreprocessor)
1092         {
1093                 Log("SHUTTLE", Form("ContinueProcessing - %s: no preprocessor registered", fCurrentDetector.Data()));
1094                 return kFALSE;
1095         }
1096
1097         AliShuttleLogbookEntry::Status entryStatus =
1098                 fLogbookEntry->GetDetectorStatus(fCurrentDetector);
1099
1100         if (entryStatus != AliShuttleLogbookEntry::kUnprocessed) {
1101                 Log("SHUTTLE", Form("ContinueProcessing - %s is %s",
1102                                 fCurrentDetector.Data(),
1103                                 fLogbookEntry->GetDetectorStatusName(entryStatus)));
1104                 return kFALSE;
1105         }
1106
1107         // if we get here, according to Shuttle logbook subdetector is in UNPROCESSED state
1108
1109         // check if current run is first unprocessed run for current detector
1110         if (fConfig->StrictRunOrder(fCurrentDetector) &&
1111                 !fFirstUnprocessed[GetDetPos(fCurrentDetector)])
1112         {
1113                 if (fTestMode == kNone)
1114                 {
1115                         Log("SHUTTLE", Form("ContinueProcessing - %s requires strict run ordering"
1116                                             " but this is not the first unprocessed run!",fCurrentDetector.Data()));
1117                         return kFALSE;
1118                 }
1119                 else
1120                 {
1121                         Log("SHUTTLE", Form("ContinueProcessing - In TESTMODE - "
1122                                         "Although %s requires strict run ordering "
1123                                         "and this is not the first unprocessed run, "
1124                                             "the SHUTTLE continues",fCurrentDetector.Data()));
1125                 }
1126         }
1127
1128         // Is the subdetector processed first time for this run?
1129         fFirstProcessing = kFALSE;
1130
1131         AliShuttleStatus* status = ReadShuttleStatus();
1132         if (!status) {
1133                 // first time
1134                 Log("SHUTTLE", Form("ContinueProcessing - %s: Processing first time",
1135                                 fCurrentDetector.Data()));
1136                 status = new AliShuttleStatus(AliShuttleStatus::kStarted);
1137                 fFirstProcessing = kTRUE;
1138                 return WriteShuttleStatus(status);
1139         }
1140
1141         // The following case shouldn't happen if Shuttle Logbook was correctly updated.
1142         // If it happens it may mean Logbook updating failed... let's do it now!
1143         if (status->GetStatus() == AliShuttleStatus::kDone ||
1144             status->GetStatus() == AliShuttleStatus::kFailed ||
1145             status->GetStatus() == AliShuttleStatus::kSkipped) {
1146                 Log("SHUTTLE", Form("ContinueProcessing - %s is already %s. Updating Shuttle Logbook",
1147                                         fCurrentDetector.Data(),
1148                                         status->GetStatusName(status->GetStatus())));
1149                 
1150                 if (status->GetStatus() == AliShuttleStatus::kSkipped)
1151                 {
1152                         UpdateShuttleLogbook(fCurrentDetector.Data(), "DONE");
1153                 }
1154                 else
1155                         UpdateShuttleLogbook(fCurrentDetector.Data(), status->GetStatusName(status->GetStatus()));
1156                         
1157                 return kFALSE;
1158         }
1159
1160         if (status->GetStatus() == AliShuttleStatus::kStoreStarted || status->GetStatus() == AliShuttleStatus::kStoreDelayed ||status->GetStatus() == AliShuttleStatus::kStoreError) {
1161                 Log("SHUTTLE",
1162                         Form("ContinueProcessing - %s: Grid storage of one or more "
1163                                 "objects failed. Trying again now",
1164                                 fCurrentDetector.Data()));
1165                 StoreOCDB();
1166                 return kFALSE;
1167         }
1168
1169         // if we get here, there is a restart
1170         Bool_t cont = kFALSE;
1171
1172         // abort conditions
1173         if (status->GetCount() >= fConfig->GetMaxRetries()) {
1174                 Log("SHUTTLE", Form("ContinueProcessing - %s failed %d times in status %s - "
1175                                 "Updating Shuttle Logbook", fCurrentDetector.Data(),
1176                                 status->GetCount(), status->GetStatusName()));
1177                 UpdateShuttleLogbook(fCurrentDetector.Data(), "FAILED");
1178                 UpdateShuttleStatus(AliShuttleStatus::kFailed);
1179
1180                 // there may still be objects in local OCDB and reference storage
1181                 // and FXS databases may be not updated: do it now!
1182                 
1183                 // TODO Currently disabled, we want to keep files in case of failure!
1184                 // CleanLocalStorage(fgkLocalCDB);
1185                 // CleanLocalStorage(fgkLocalRefStorage);
1186                 // UpdateTableFailCase();
1187                 
1188                 // Send mail to detector expert!
1189                 Log("SHUTTLE", Form("ContinueProcessing - Sending mail to %s expert...", 
1190                                     fCurrentDetector.Data()));
1191                 // det experts in to
1192                 TString to="";
1193                 TIter *iterExperts = 0;
1194                 iterExperts = new TIter(fConfig->GetResponsibles(fCurrentDetector));
1195                 TObjString *anExpert=0;
1196                 while ((anExpert = (TObjString*) iterExperts->Next()))
1197                         {
1198                                 to += Form("%s, \n", anExpert->GetName());
1199                         }
1200                 delete iterExperts;
1201                 
1202                 if (to.Length() > 0)
1203                         to.Remove(to.Length()-3);
1204                 AliDebug(2, Form("to: %s",to.Data()));
1205
1206                 if (to.IsNull()) {
1207                         Log("SHUTTLE", Form("List of %s responsibles not set!", fCurrentDetector.Data()));
1208                         return kFALSE;
1209                 }
1210
1211                 Log(fCurrentDetector.Data(), Form("ContinueProcessing - Sending mail to %s expert(s):", 
1212                                     fCurrentDetector.Data()));
1213                 Log(fCurrentDetector.Data(), Form("\n%s", to.Data()));
1214                 if (!SendMail(kPPEMail))
1215                         Log("SHUTTLE", Form("ContinueProcessing - Could not send mail to %s expert",
1216                                             fCurrentDetector.Data()));
1217
1218         } else {
1219                 Log("SHUTTLE", Form("ContinueProcessing - %s: restarting. "
1220                                 "Aborted before with %s. Retry number %d.", fCurrentDetector.Data(),
1221                                 status->GetStatusName(), status->GetCount()));
1222                 Bool_t increaseCount = kTRUE;
1223                 if (status->GetStatus() == AliShuttleStatus::kDCSError || 
1224                     status->GetStatus() == AliShuttleStatus::kDCSStarted ||
1225                     status->GetStatus() == AliShuttleStatus::kFXSError ||
1226                     status->GetStatus() == AliShuttleStatus::kOCDBError)
1227                                 increaseCount = kFALSE;
1228                                 
1229                 UpdateShuttleStatus(AliShuttleStatus::kStarted, increaseCount);
1230                 cont = kTRUE;
1231         }
1232
1233         return cont;
1234 }
1235
1236 //______________________________________________________________________________________________
1237 void AliShuttle::SendMLRunInfo(const char* status)
1238 {
1239         // 
1240         // Send information about this run to ML
1241         
1242         TMonaLisaText  mlStatus("SHUTTLE_status", status);
1243         TString runType(fLogbookEntry->GetRunType());
1244         if (strlen(fLogbookEntry->GetRunParameter("log")) > 0){
1245
1246                 runType += "(";
1247                 runType += fLogbookEntry->GetRunParameter("log");
1248                 runType += ")";
1249         }
1250         if (fLogbookEntry->GetDATestMode()){
1251                 runType += " (DATest)";
1252         }
1253         TMonaLisaText  mlRunType("SHUTTLE_runtype", runType);
1254
1255         TList mlList;
1256         mlList.Add(&mlStatus);
1257         mlList.Add(&mlRunType);
1258
1259         TString mlID;
1260         mlID.Form("%d", GetCurrentRun());
1261         fMonaLisa->SendParameters(&mlList, mlID);       
1262 }
1263
1264 //______________________________________________________________________________________________
1265 Int_t AliShuttle::GetMem(Int_t pid)
1266 {
1267         // invokes ps to get the memory consumption of the process <pid>
1268         // returns -1 in case of error
1269         
1270         TString checkStr;
1271         checkStr.Form("ps -o vsize --pid %d | tail -n 1", pid);
1272         FILE* pipe = gSystem->OpenPipe(checkStr, "r");
1273         if (!pipe)
1274         {
1275                 Log("SHUTTLE", Form("Process - Error: "
1276                         "Could not open pipe to %s", checkStr.Data()));
1277                 return -1;
1278         }
1279                 
1280         char buffer[100];
1281         if (!fgets(buffer, 100, pipe))
1282         {
1283                 Log("SHUTTLE", "Process - Error: ps did not return anything");
1284                 gSystem->ClosePipe(pipe);
1285                 return -1;
1286         }
1287         gSystem->ClosePipe(pipe);
1288         
1289         //Log("SHUTTLE", Form("ps returned %s", buffer));
1290         
1291         Int_t mem = 0;
1292         if ((sscanf(buffer, "%d\n", &mem) != 1) || !mem)
1293         {
1294                 Log("SHUTTLE", "Process - Error: Could not parse output of ps");
1295                 return -1;
1296         }
1297         
1298         return mem;
1299 }
1300
1301 //______________________________________________________________________________________________
1302 Bool_t AliShuttle::Process(AliShuttleLogbookEntry* entry)
1303 {
1304         //
1305         // Makes data retrieval for all detectors in the configuration.
1306         // entry: Shuttle logbook entry, contains run paramenters and status of detectors
1307         // (Unprocessed, Inactive, Failed or Done).
1308         // Returns kFALSE in case of error occured and kTRUE otherwise
1309         //
1310
1311         if (!entry) return kFALSE;
1312
1313         fLogbookEntry = entry;
1314
1315         Log("SHUTTLE", Form("\t\t\t^*^*^*^*^*^*^*^*^*^*^*^* run %d: START ^*^*^*^*^*^*^*^*^*^*^*^*",
1316                                         GetCurrentRun()));
1317
1318         CountOpenRuns();
1319         
1320         // Send the information to ML
1321         SendMLRunInfo("Processing");
1322
1323         if (fLogbookEntry->IsDone())
1324         {
1325                 Log("SHUTTLE","Process - Shuttle is already DONE. Updating logbook");
1326                 UpdateShuttleLogbook("shuttle_done");
1327                 fLogbookEntry = 0;
1328                 return kTRUE;
1329         }
1330
1331         // read test mode if flag is set
1332         if (fReadTestMode)
1333         {
1334                 fTestMode = kNone;
1335                 TString logEntry(entry->GetRunParameter("log"));
1336                 //printf("log entry = %s\n", logEntry.Data());
1337                 TString searchStr("Testmode: ");
1338                 Int_t pos = logEntry.Index(searchStr.Data());
1339                 //printf("%d\n", pos);
1340                 if (pos >= 0)
1341                 {
1342                         TSubString subStr = logEntry(pos + searchStr.Length(), logEntry.Length());
1343                         //printf("%s\n", subStr.String().Data());
1344                         TString newStr(subStr.Data());
1345                         TObjArray* token = newStr.Tokenize(' ');
1346                         if (token)
1347                         {
1348                                 //token->Print();
1349                                 TObjString* tmpStr = dynamic_cast<TObjString*> (token->First());
1350                                 if (tmpStr)
1351                                 {
1352                                         Int_t testMode = tmpStr->String().Atoi();
1353                                         if (testMode > 0)
1354                                         {
1355                                                 Log("SHUTTLE", Form("Process - Enabling test mode %d", testMode));
1356                                                 SetTestMode((TestMode) testMode);
1357                                         }
1358                                 }
1359                                 delete token;          
1360                         }
1361                 }
1362         }
1363                 
1364         fLogbookEntry->Print("all");
1365
1366         // Initialization
1367         Bool_t hasError = kFALSE;
1368
1369         // Set the CDB and Reference folders according to the year
1370
1371         // build cdb paths (repeat each time, run might be a DATest run)
1372         if (!fLogbookEntry->GetDATestMode()){
1373                 fgkMainCDB.Form("alien://folder=%s%d/OCDB?user=alidaq?cacheFold=/tmp/OCDBCache", 
1374                                 fConfig->GetAlienPath(), GetCurrentYear());
1375                 
1376                 fgkMainRefStorage.Form("alien://folder=%s%d/Reference?user=alidaq?cacheFold=/tmp/OCDBCache", 
1377                                        fConfig->GetAlienPath(), GetCurrentYear());
1378         }
1379         else {
1380                 fgkMainCDB.Form("alien://folder=%s%d/DATest/OCDB?user=alidaq?cacheFold=/tmp/OCDBCache",
1381                                 fConfig->GetAlienPath(), GetCurrentYear());
1382                 
1383                 fgkMainRefStorage.Form("alien://folder=%s%d/DATest/Reference?user=alidaq?cacheFold=/tmp/OCDBCache",
1384                                        fConfig->GetAlienPath(), GetCurrentYear());
1385         }
1386
1387         AliDebug(2,Form("Main CDB storage = %s",fgkMainCDB.Data()));
1388         AliDebug(2,Form("Main Reference storage = %s",fgkMainRefStorage.Data()));
1389
1390         // Loop on detectors in the configuration
1391         TIter iter(fConfig->GetDetectors());
1392         TObjString* aDetector = 0;
1393
1394         Bool_t first = kTRUE;
1395
1396         while ((aDetector = (TObjString*) iter.Next()))
1397         {
1398                 fCurrentDetector = aDetector->String();
1399
1400                 if (ContinueProcessing() == kFALSE) 
1401                         continue;
1402                 
1403                 if (first)
1404                 {
1405                   // only read QueryCDB when needed and only once
1406                   AliCDBStorage *mainCDBSto = AliCDBManager::Instance()->GetStorage(fgkMainCDB);
1407                   if(mainCDBSto) mainCDBSto->QueryCDB(GetCurrentRun());
1408                   AliCDBStorage *mainRefSto = AliCDBManager::Instance()->GetStorage(fgkMainRefStorage);
1409                   if(mainRefSto) mainRefSto->QueryCDB(GetCurrentRun());
1410                   first = kFALSE;
1411                 }
1412
1413                 Log("SHUTTLE", Form("\t\t\t****** run %d - %s: START  ******",
1414                                                 GetCurrentRun(), aDetector->GetName()));
1415
1416                 for(Int_t iSys=0;iSys<4;iSys++) fFXSCalled[iSys]=kFALSE;
1417                 
1418                 Int_t initialMem = GetMem(getpid());
1419                 Log("SHUTTLE", Form("Memory consumption before forking is %d", initialMem));
1420
1421                 Log(fCurrentDetector.Data(), "Process - Starting processing");
1422
1423                 Int_t pid = fork();
1424
1425                 if (pid < 0)
1426                 {
1427                         Log("SHUTTLE", "Process - ERROR: Forking failed");
1428                 }
1429                 else if (pid > 0)
1430                 {
1431                         // parent
1432                         Log("SHUTTLE", Form("Process - In parent process of %d - %s: Starting monitoring",
1433                                                         GetCurrentRun(), aDetector->GetName()));
1434
1435                         Long_t begin = time(0);
1436
1437                         int status; // to be used with waitpid, on purpose an int (not Int_t)!
1438                         while (waitpid(pid, &status, WNOHANG) == 0)
1439                         {
1440                                 Long_t expiredTime = time(0) - begin;
1441
1442                                 // the run-dependent timeout is the timeout from the configuration plus a twentieth of
1443                                 // the run duration, e.g. 3 additional minutes for 1h run or 1/2h for a 10h run
1444                                 Int_t runDepTimeOut = fConfig->GetPPTimeOut() + (GetCurrentEndTime() - GetCurrentStartTime()) * 0.05;
1445                                 if (expiredTime > runDepTimeOut)
1446                                 {
1447                                         TString logMsg;
1448                                         AliShuttleStatus *currentStatus = ReadShuttleStatus();
1449                                         AliShuttleStatus::Status newStatus = AliShuttleStatus::kInvalid;
1450                                         
1451                                         if (currentStatus->GetStatus() == AliShuttleStatus::kDCSStarted)
1452                                         {
1453                                                 // in case the pp goes in TimeOut while retrieving the DCS DPs
1454                                                 // set status to kDCSError
1455                                                 
1456                                                 logMsg.Form("Process - Process of %s timed out while retrieving the DCS DataPoints. Run time: %ld seconds. Killing... and setting status to DCSError.",
1457                                                                 fCurrentDetector.Data(), expiredTime);
1458                                                 newStatus = AliShuttleStatus::kDCSError;
1459                                         }
1460                                         else if (currentStatus->GetStatus() <= AliShuttleStatus::kPPDone)
1461                                         {
1462                                                 // in case pp not yet done set status to kPPTimeOut
1463                                         
1464                                                 logMsg.Form("Process - Process of %s timed out. Run time: %ld seconds. Killing...",
1465                                                                 fCurrentDetector.Data(), expiredTime);
1466                                                 newStatus = AliShuttleStatus::kPPTimeOut;
1467                                         }
1468                                         else if (currentStatus->GetStatus() == AliShuttleStatus::kStoreStarted)
1469                                         {
1470                                                 // in case the pp goes in TimeOut while storing the objects in the OCDB
1471                                                 // set status to kStoreError
1472                                                 
1473                                                 logMsg.Form("Process - Process of %s timed out while storing the OCDB object. Run time: %ld seconds. Killing... and setting status to StoreError.",
1474                                                                 fCurrentDetector.Data(), expiredTime);
1475                                                 newStatus = AliShuttleStatus::kStoreError;
1476                                         }
1477                                         else 
1478                                         {
1479                                                 // in other cases don't change the status
1480                                                 
1481                                                 logMsg.Form("Process - Process of %s timed out in status = %s. Run time: %ld seconds. Killing... without changing the status",
1482                                                                 fCurrentDetector.Data(), currentStatus->GetStatusName(), expiredTime);
1483                                         }
1484                                 
1485                                         Log("SHUTTLE", logMsg);
1486                                         Log(fCurrentDetector, logMsg);
1487
1488                                         kill(pid, 9);
1489
1490                                         if (newStatus != AliShuttleStatus::kInvalid)
1491                                                 UpdateShuttleStatus(newStatus);
1492                                         hasError = kTRUE;
1493
1494                                         gSystem->Sleep(1000);
1495                                 }
1496                                 else
1497                                 {
1498                                         gSystem->Sleep(1000);
1499                                         
1500                                         Int_t mem = GetMem(pid);
1501
1502                                         if (mem < 0)
1503                                                 continue;
1504                                                 
1505                                         mem -= initialMem;
1506                                         if (mem < 0)
1507                                                 mem = 0;
1508                                         
1509                                         if (expiredTime % 60 == 0)
1510                                         {
1511                                                 Log("SHUTTLE", Form("Process - %s: Checking process. "
1512                                                         "Run time: %ld seconds - Memory consumption: %d KB",
1513                                                         fCurrentDetector.Data(), expiredTime, mem));
1514                                                 SendAlive();
1515                                         }
1516                                         
1517                                         if (mem > fConfig->GetPPMaxMem())
1518                                         {
1519                                                 TString tmp;
1520                                                 tmp.Form("Process - Process exceeds maximum allowed memory "
1521                                                         "(%d KB > %d KB). Killing...",
1522                                                         mem, fConfig->GetPPMaxMem());
1523                                                 Log("SHUTTLE", tmp);
1524                                                 Log(fCurrentDetector, tmp);
1525         
1526                                                 kill(pid, 9);
1527         
1528                                                 UpdateShuttleStatus(AliShuttleStatus::kPPOutOfMemory);
1529                                                 hasError = kTRUE;
1530         
1531                                                 gSystem->Sleep(1000);
1532                                         }
1533                                 }
1534                         }
1535
1536                         Log("SHUTTLE", Form("Process - In parent process of %d - %s: Client has terminated.",
1537                                                                 GetCurrentRun(), aDetector->GetName()));
1538
1539                         if (WIFEXITED(status))
1540                         {
1541                                 Int_t returnCode = WEXITSTATUS(status);
1542
1543                                 Log("SHUTTLE", Form("Process - %s: the return code is %d", fCurrentDetector.Data(),
1544                                                                                 returnCode));
1545
1546                                 if (returnCode == 0) hasError = kTRUE;
1547                         }
1548                 }
1549                 else if (pid == 0)
1550                 {
1551                         // child
1552                         Log("SHUTTLE", Form("Process - In child process of %d - %s", GetCurrentRun(),
1553                                 aDetector->GetName()));
1554
1555                         Log("SHUTTLE", Form("Process - Redirecting output to %s log",fCurrentDetector.Data()));
1556
1557                         if ((freopen(GetLogFileName(fCurrentDetector), "a", stdout)) == 0)
1558                         {
1559                                 Log("SHUTTLE", "Process - Could not freopen stdout");
1560                         }
1561                         else
1562                         {
1563                                 fOutputRedirected = kTRUE;
1564                                 if ((dup2(fileno(stdout), fileno(stderr))) < 0)
1565                                         Log("SHUTTLE", "Process - Could not redirect stderr");
1566                                 
1567                         }
1568
1569                         Log("SHUTTLE", "Executing TGrid::Connect");
1570                         TGrid::Connect("alien://");
1571                         
1572                         TString wd = gSystem->WorkingDirectory();
1573                         Int_t dir_lev1 = GetCurrentRun()/10000;
1574                         TString tmpDir = Form("%s/%d/%d/%s_process", GetShuttleTempDir(), 
1575                                 dir_lev1, GetCurrentRun(), fCurrentDetector.Data());
1576                         
1577                         Int_t result = gSystem->GetPathInfo(tmpDir.Data(), 0, (Long64_t*) 0, 0, 0);
1578                         if (!result) // temp dir already exists!
1579                         {
1580                                 Log(fCurrentDetector.Data(), 
1581                                         Form("Process - %s dir already exists! Removing...", tmpDir.Data()));
1582                                 gSystem->Exec(Form("rm -rf %s",tmpDir.Data()));         
1583                         } 
1584                         
1585                         if (gSystem->mkdir(tmpDir.Data(), 1))
1586                         {
1587                                 Log(fCurrentDetector.Data(), "Process - could not make temp directory!!");
1588                                 gSystem->Exit(1);
1589                         }
1590                         
1591                         if (!gSystem->ChangeDirectory(tmpDir.Data())) 
1592                         {
1593                                 Log(fCurrentDetector.Data(), "Process - could not change directory!!");
1594                                 gSystem->Exit(1);                       
1595                         }
1596                         
1597                         Int_t success = ProcessCurrentDetector();
1598
1599                         gSystem->ChangeDirectory(wd.Data());
1600                                                 
1601                         if (success == 1) // Preprocessor finished successfully!
1602                         { 
1603                                 // remove temporary folder or DCS map
1604                                 if (!fConfig->KeepTempFolder())
1605                                 {
1606                                         gSystem->Exec(Form("rm -rf %s",tmpDir.Data()));
1607                                 } else if (!fConfig->KeepDCSMap())
1608                                 {
1609                                         gSystem->Exec(Form("rm -f %s/DCSMap.root",tmpDir.Data()));
1610                                 }
1611                                 
1612                                 // Update time_processed field in FXS DB
1613                                 if (UpdateTable() == kFALSE)
1614                                         Log("SHUTTLE", Form("Process - %s: Could not update FXS databases!", 
1615                                                         fCurrentDetector.Data()));
1616
1617                                 // Transfer the data from local storage to main storage (Grid)
1618                                 if (StoreOCDB() == kFALSE)
1619                                         success = kFALSE;
1620                         } 
1621                         else if (success == 0)
1622                         {
1623                                 Log("SHUTTLE", 
1624                                         Form("\t\t\t****** run %d - %s: ERROR ******",
1625                                                 GetCurrentRun(), aDetector->GetName()));
1626                         }
1627
1628                         for (UInt_t iSys=0; iSys<4; iSys++)
1629                         {
1630                                 if (fFXSCalled[iSys]) fFXSlist[iSys].Clear();
1631                         }
1632
1633                         Log("SHUTTLE", Form("Process - Client process of %d - %s is exiting now with %d.",
1634                                                         GetCurrentRun(), aDetector->GetName(), success));
1635
1636                         // the client exits here
1637                         gSystem->Exit(success);
1638
1639                         AliError("We should never get here!!!");
1640                 }
1641         }
1642
1643         Log("SHUTTLE", Form("\t\t\t^*^*^*^*^*^*^*^*^*^*^*^* run %d: FINISH ^*^*^*^*^*^*^*^*^*^*^*^*",
1644                                                         GetCurrentRun()));
1645
1646         //check if shuttle is done for this run, if so update logbook
1647         TObjArray checkEntryArray;
1648         checkEntryArray.SetOwner(1);
1649         TString whereClause = Form("where run=%d", GetCurrentRun());
1650         if (!QueryShuttleLogbook(whereClause.Data(), checkEntryArray) || 
1651                         checkEntryArray.GetEntries() == 0) {
1652                 Log("SHUTTLE", Form("Process - Warning: Cannot check status of run %d on Shuttle logbook!",
1653                                                 GetCurrentRun()));
1654                 return hasError == kFALSE;
1655         }
1656
1657         AliShuttleLogbookEntry* checkEntry = dynamic_cast<AliShuttleLogbookEntry*>
1658                                                 (checkEntryArray.At(0));
1659
1660         if (checkEntry)
1661         {
1662                 if (checkEntry->IsDone())
1663                 {
1664                         Log("SHUTTLE","Process - Shuttle is DONE. Updating logbook");
1665                         UpdateShuttleLogbook("shuttle_done");
1666                 }
1667                 else
1668                 {
1669                         for (UInt_t iDet=0; iDet<NDetectors(); iDet++)
1670                         {
1671                                 if (checkEntry->GetDetectorStatus(iDet) == AliShuttleLogbookEntry::kUnprocessed)
1672                                 {
1673                                         AliDebug(2, Form("Run %d: setting %s as \"not first time unprocessed\"",
1674                                                         checkEntry->GetRun(), GetDetName(iDet)));
1675                                         fFirstUnprocessed[iDet] = kFALSE;
1676                                 }
1677                         }
1678                         SendMLRunInfo("Pending");
1679                 }
1680         }
1681
1682         fLogbookEntry = 0;
1683
1684         return hasError == kFALSE;
1685 }
1686
1687 //______________________________________________________________________________________________
1688 Int_t AliShuttle::ProcessCurrentDetector()
1689 {
1690         //
1691         // Makes data retrieval just for a specific detector (fCurrentDetector).
1692         // Threre should be a configuration for this detector.
1693
1694         Log("SHUTTLE", Form("ProcessCurrentDetector - Retrieving values for %s, run %d", 
1695                                                 fCurrentDetector.Data(), GetCurrentRun()));
1696
1697         TString wd = gSystem->WorkingDirectory();
1698         
1699         if (!CleanReferenceStorage(fCurrentDetector.Data()))
1700                 return 0;
1701         
1702         gSystem->ChangeDirectory(wd.Data());
1703         
1704         // call preprocessor
1705         AliPreprocessor* aPreprocessor =
1706                 dynamic_cast<AliPreprocessor*> (fPreprocessorMap.GetValue(fCurrentDetector));
1707
1708         // check if the preprocessor wants to process this run type
1709         if (aPreprocessor->ProcessRunType() == kFALSE)
1710         {
1711                 UpdateShuttleStatus(AliShuttleStatus::kSkipped);
1712                 UpdateShuttleLogbook(fCurrentDetector, "DONE");
1713                 if (!UpdateTableSkippedCase(fCurrentDetector.Data()))
1714                 {
1715                         AliError(Form("Could not update FXS tables for run %d !", GetCurrentRun()));
1716                 }
1717                 Log(fCurrentDetector, Form("ProcessCurrentDetector - %s preprocessor is not interested in this run type", fCurrentDetector.Data()));
1718         
1719                 return 2;
1720         }
1721         
1722         // checking if OCDB is reachable
1723         AliCDBEntry* testEntry = GetFromOCDB("SHUTTLE","GRP/CTP/DummyConfig");
1724         if (!testEntry){
1725                 // OCDB is not accessible, going in OCDBError for current detector
1726                 AliError("OCDB Test entry not accessible");
1727                 UpdateShuttleStatus(AliShuttleStatus::kOCDBError);
1728                 return 0;
1729         }
1730
1731         TMap* dcsMap = new TMap();
1732         
1733         aPreprocessor->Initialize(GetCurrentRun(), GetCurrentStartTime(), GetCurrentEndTime());
1734
1735         Bool_t processDCS = aPreprocessor->ProcessDCS();
1736
1737         if (!processDCS)
1738         {
1739                 Log(fCurrentDetector, "ProcessCurrentDetector -"
1740                         " The preprocessor requested to skip the retrieval of DCS values");
1741         }
1742         else if (fTestMode & kSkipDCS)
1743         {
1744                 Log(fCurrentDetector, "ProcessCurrentDetector - In TESTMODE: Skipping DCS processing");
1745         } 
1746         else if (fTestMode & kErrorDCS)
1747         {
1748                 Log(fCurrentDetector, "ProcessCurrentDetector - In TESTMODE: Simulating DCS error");
1749                 UpdateShuttleStatus(AliShuttleStatus::kDCSStarted);
1750                 UpdateShuttleStatus(AliShuttleStatus::kDCSError);
1751                 delete dcsMap;
1752                 return 0;
1753         } else {
1754
1755                 UpdateShuttleStatus(AliShuttleStatus::kDCSStarted);
1756
1757                 // Query DCS archive
1758                 Int_t nServers = fConfig->GetNServers(fCurrentDetector);
1759                 
1760                 for (int iServ=0; iServ<nServers; iServ++)
1761                 {
1762                 
1763                         TString host(fConfig->GetDCSHost(fCurrentDetector, iServ));
1764                         Int_t port = fConfig->GetDCSPort(fCurrentDetector, iServ);
1765                         Int_t multiSplit = fConfig->GetMultiSplit(fCurrentDetector, iServ);
1766
1767                         Log(fCurrentDetector, Form("ProcessCurrentDetector -"
1768                                         " Querying DCS Amanda server %s:%d (%d of %d)", 
1769                                         host.Data(), port, iServ+1, nServers));
1770                         
1771                         TMap* aliasMap = 0;
1772                         TMap* dpMap = 0;
1773
1774                         if (fConfig->GetDCSAliases(fCurrentDetector, iServ)->GetEntries() > 0)
1775                         {
1776                                 Log(fCurrentDetector, Form("Querying %d DCS aliases", fConfig->GetDCSAliases(fCurrentDetector, iServ)->GetEntries()));
1777                                 aliasMap = GetValueSet(host, port, 
1778                                                 fConfig->GetDCSAliases(fCurrentDetector, iServ), 
1779                                                 kAlias, multiSplit);
1780                                 if (!aliasMap)
1781                                 {
1782                                         Log(fCurrentDetector, 
1783                                                 Form("ProcessCurrentDetector -"
1784                                                         " Error retrieving DCS aliases 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                                         delete dcsMap;
1793                                         return 0;
1794                                 }
1795                         }
1796                         
1797                         if (fConfig->GetDCSDataPoints(fCurrentDetector, iServ)->GetEntries() > 0)
1798                         {
1799                                 Log(fCurrentDetector, Form("Querying %d DCS data points", fConfig->GetDCSDataPoints(fCurrentDetector, iServ)->GetEntries()));
1800                                 dpMap = GetValueSet(host, port, 
1801                                                 fConfig->GetDCSDataPoints(fCurrentDetector, iServ), 
1802                                                 kDP, multiSplit);
1803                                 if (!dpMap)
1804                                 {
1805                                         Log(fCurrentDetector, 
1806                                                 Form("ProcessCurrentDetector -"
1807                                                         " Error retrieving DCS data points from server %s."
1808                                                         " Sending mail to DCS experts!", host.Data()));
1809                                         UpdateShuttleStatus(AliShuttleStatus::kDCSError);
1810                                         
1811                                         if (!SendMail(kDCSEMail))
1812                                                 Log("SHUTTLE", Form("ProcessCurrentDetector - "
1813                                                                     "Could not send mail to DCS experts!"));
1814                                         
1815                                         if (aliasMap) delete aliasMap;
1816                                         delete dcsMap;
1817                                         return 0;
1818                                 }                               
1819                         }
1820                         
1821                         // merge aliasMap and dpMap into dcsMap
1822                         if(aliasMap) {
1823                                 TIter iter(aliasMap);
1824                                 TObjString* key = 0;
1825                                 while ((key = (TObjString*) iter.Next()))
1826                                         dcsMap->Add(key, aliasMap->GetValue(key->String()));
1827                                 
1828                                 aliasMap->SetOwner(kFALSE);
1829                                 delete aliasMap;
1830                         }       
1831                         
1832                         if(dpMap) {
1833                                 TIter iter(dpMap);
1834                                 TObjString* key = 0;
1835                                 while ((key = (TObjString*) iter.Next()))
1836                                         dcsMap->Add(key, dpMap->GetValue(key->String()));
1837                                 
1838                                 dpMap->SetOwner(kFALSE);
1839                                 delete dpMap;
1840                         }
1841                 }
1842         }
1843         
1844         // save map into file, to help debugging in case of preprocessor error
1845         TFile* f = TFile::Open("DCSMap.root","recreate");
1846         f->cd();
1847         dcsMap->Write("DCSMap", TObject::kSingleKey);
1848         f->Close();
1849         delete f;
1850         
1851         // DCS Archive DB processing successful. Call Preprocessor!
1852         UpdateShuttleStatus(AliShuttleStatus::kPPStarted);
1853
1854         fFXSError = -1; // this variable is kTRUE after ::Process if an FXS error occured
1855         
1856         UInt_t returnValue = aPreprocessor->Process(dcsMap);
1857         
1858         if (fFXSError!=-1) {
1859                 UpdateShuttleStatus(AliShuttleStatus::kFXSError);
1860                 SendMail(kFXSEMail, fFXSError);
1861                 dcsMap->DeleteAll();
1862                 delete dcsMap;
1863                 return 0;
1864         }
1865
1866         if (returnValue > 0) // Preprocessor error!
1867         {
1868                 Log(fCurrentDetector, Form("ProcessCurrentDetector - "
1869                                 "Preprocessor failed. Process returned %d.", returnValue));
1870                 UpdateShuttleStatus(AliShuttleStatus::kPPError);
1871                 dcsMap->DeleteAll();
1872                 delete dcsMap;
1873                 return 0;
1874         }
1875         
1876         // preprocessor ok!
1877         UpdateShuttleStatus(AliShuttleStatus::kPPDone);
1878         Log(fCurrentDetector, Form("ProcessCurrentDetector - %s preprocessor returned success",
1879                                 fCurrentDetector.Data()));
1880
1881         dcsMap->DeleteAll();
1882         delete dcsMap;
1883
1884         return 1;
1885 }
1886
1887 //______________________________________________________________________________________________
1888 void AliShuttle::CountOpenRuns()
1889 {
1890         // Query DAQ's Shuttle logbook and sends the number of open runs to ML
1891         
1892         SendAlive();
1893         
1894         // check connection, in case connect
1895         if (!Connect(4)) 
1896                 return;
1897
1898         TString sqlQuery;
1899         sqlQuery = Form("select count(*) from %s where shuttle_done=0", fConfig->GetShuttlelbTable());
1900         
1901         TSQLResult* aResult = fServer[4]->Query(sqlQuery);
1902         if (!aResult) {
1903                 AliError(Form("Can't execute query <%s>!", sqlQuery.Data()));
1904                 return;
1905         }
1906
1907         AliDebug(2,Form("Query = %s", sqlQuery.Data()));
1908         
1909         if (aResult->GetRowCount() == 0) {
1910                 AliError(Form("No result for query %s received", sqlQuery.Data()));
1911                 return;
1912         }
1913
1914         if (aResult->GetFieldCount() != 1) {
1915                 AliError(Form("Invalid field count for query %s received", sqlQuery.Data()));
1916                 return;
1917         }
1918
1919         TSQLRow* aRow = aResult->Next();
1920         if (!aRow) {
1921                 AliError(Form("Could not receive result of query %s", sqlQuery.Data()));
1922                 return;
1923         }
1924         
1925         TString result(aRow->GetField(0), aRow->GetFieldLength(0));
1926         Int_t count = result.Atoi();
1927         
1928         Log("SHUTTLE", Form("%d unprocessed runs", count));
1929         
1930         delete aRow;
1931         delete aResult;
1932
1933         TMonaLisaValue mlStatus("SHUTTLE_openruns", count);
1934
1935         TList mlList;
1936         mlList.Add(&mlStatus);
1937
1938         fMonaLisa->SendParameters(&mlList, "__PROCESSINGINFO__");
1939 }
1940
1941 //______________________________________________________________________________________________
1942 Bool_t AliShuttle::QueryShuttleLogbook(const char* whereClause,
1943                 TObjArray& entries)
1944 {
1945         // Query DAQ's Shuttle logbook and fills detector status object.
1946         // Call QueryRunParameters to query DAQ logbook for run parameters.
1947         //
1948
1949         entries.SetOwner(1);
1950
1951         // check connection, in case connect
1952         if (!Connect(4)) return kFALSE;
1953
1954         TString sqlQuery;
1955         sqlQuery = Form("select * from %s %s order by run", fConfig->GetShuttlelbTable(), whereClause);
1956
1957         TSQLResult* aResult = fServer[4]->Query(sqlQuery);
1958         if (!aResult) {
1959                 AliError(Form("Can't execute query <%s>!", sqlQuery.Data()));
1960                 return kFALSE;
1961         }
1962
1963         AliDebug(2,Form("Query = %s", sqlQuery.Data()));
1964
1965         if(aResult->GetRowCount() == 0) {
1966                 Log("SHUTTLE", "No entries in Shuttle Logbook match request");
1967                 delete aResult;
1968                 return kTRUE;
1969         }
1970
1971         // TODO Check field count!
1972         const UInt_t nCols = 26;
1973         if (aResult->GetFieldCount() != (Int_t) nCols) {
1974                 Log("SHUTTLE", "Invalid SQL result field number!");
1975                 delete aResult;
1976                 return kFALSE;
1977         }
1978
1979         TSQLRow* aRow;
1980         while ((aRow = aResult->Next())) {
1981                 TString runString(aRow->GetField(0), aRow->GetFieldLength(0));
1982                 Int_t run = runString.Atoi();
1983
1984                 AliShuttleLogbookEntry *entry = QueryRunParameters(run);
1985                 if (!entry)
1986                         continue;
1987
1988                 // DA test mode flag
1989                 TString daTestModeString(aRow->GetField(2), aRow->GetFieldLength(2)); // field 2 = DA test mode flag 
1990                 Bool_t daTestMode = (Bool_t)daTestModeString.Atoi();
1991                 entry->SetDATestMode(daTestMode);
1992
1993                 // loop on detectors
1994                 for(UInt_t ii = 0; ii < nCols; ii++)
1995                         entry->SetDetectorStatus(aResult->GetFieldName(ii), aRow->GetField(ii));
1996
1997                 entries.AddLast(entry);
1998                 delete aRow;
1999         }
2000
2001         delete aResult;
2002         return kTRUE;
2003 }
2004
2005 //______________________________________________________________________________________________
2006 AliShuttleLogbookEntry* AliShuttle::QueryRunParameters(Int_t run)
2007 {
2008         //
2009         // Retrieve run parameters written in the DAQ logbook and sets them into AliShuttleLogbookEntry object
2010         //
2011
2012         // check connection, in case connect
2013         if (!Connect(4))
2014                 return 0;
2015
2016         TString sqlQuery;
2017         sqlQuery.Form("select * from %s where run=%d", fConfig->GetDAQlbTable(), run);
2018
2019         TSQLResult* aResult = fServer[4]->Query(sqlQuery);
2020         if (!aResult) {
2021                 Log("SHUTTLE", Form("Can't execute query <%s>!", sqlQuery.Data()));
2022                 return 0;
2023         }
2024
2025         if (aResult->GetRowCount() == 0) {
2026                 Log("SHUTTLE", Form("QueryRunParameters - No entry in DAQ Logbook for run %d. Skipping", run));
2027                 delete aResult;
2028                 return 0;
2029         }
2030
2031         if (aResult->GetRowCount() > 1) {
2032                 Log("SHUTTLE", Form("QueryRunParameters - UNEXPECTED: "
2033                                 "more than one entry in DAQ Logbook for run %d!", run));
2034                 delete aResult;
2035                 return 0;
2036         }
2037
2038         TSQLRow* aRow = aResult->Next();
2039         if (!aRow)
2040         {
2041                 Log("SHUTTLE", Form("QueryRunParameters - Could not retrieve row for run %d. Skipping", run));
2042                 delete aResult;
2043                 return 0;
2044         }
2045
2046         AliShuttleLogbookEntry* entry = new AliShuttleLogbookEntry(run);
2047
2048         for (Int_t ii = 0; ii < aResult->GetFieldCount(); ii++)
2049                 entry->SetRunParameter(aResult->GetFieldName(ii), aRow->GetField(ii));
2050
2051         delete aRow;
2052         delete aResult;
2053         
2054         UInt_t startTime = entry->GetStartTime();
2055         UInt_t endTime = entry->GetEndTime();
2056         Bool_t ecsSuccess = entry->GetECSSuccess();
2057         TString runType = entry->GetRunType();
2058         TString tmpdaqstartTime = entry->GetRunParameter("DAQ_time_start");
2059         TString recordingFlagString = entry->GetRunParameter("GDCmStreamRecording");
2060         UInt_t recordingFlag = recordingFlagString.Atoi();
2061         UInt_t daqstartTime = tmpdaqstartTime.Atoi();
2062         
2063         UInt_t now = time(0);
2064         Int_t dcsDelay = fConfig->GetDCSDelay()+fConfig->GetDCSQueryOffset();
2065
2066         Bool_t skip = kFALSE;
2067                                 
2068         // runs are processed if
2069         //   a) runType is PHYSICS and ecsSuccess is set
2070         //   b) runType is not PHYSICS and (ecsSuccess is set or DAQ_time_start is non-0)
2071         // effectively this means that all runs are processed that started properly (ecsSucess behaviour is different for PHYSICS and non-PHYSICS runs (check with ECS!)
2072         if (startTime != 0 && endTime != 0) {
2073                 if (endTime > startTime) { 
2074                         if (endTime >= now - dcsDelay) {
2075                                 Log("SHUTTLE", Form("Skipping run %d for now, because DCS buffer time is not yet expired", run));
2076                         } else {
2077                                 if ((runType == "PHYSICS" || runType == "STANDALONE") && recordingFlag == 0){
2078                                         Log("SHUTTLE", Form("QueryRunParameters - Run type for run %d is %s but the recording is OFF - Skipping!", run, runType.Data()));
2079                                         skip = kTRUE;
2080                                 } 
2081                                 else {
2082                                         if (runType == "PHYSICS") {
2083                                                 if (ecsSuccess) {
2084                                                         return entry;
2085                                                 } else {
2086                                                         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")));
2087                                                         skip = kTRUE;
2088                                                 } 
2089                                         } else {
2090                                                 if (ecsSuccess || daqstartTime > 0) {
2091                                                         if (ecsSuccess == kFALSE)
2092                                                                 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")));
2093                                                         return entry;
2094                                                 } else {
2095                                                         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")));
2096                                                         skip = kTRUE;
2097                                                 }
2098                                         }
2099                                 }
2100                         }
2101                 } else {
2102                         Log("SHUTTLE", Form("QueryRunParameters - Invalid parameters for run %d: startTime equal to endTime: %d %d - Skipping!", run, startTime, endTime));
2103                         skip = kTRUE;
2104                 }
2105         } else {
2106           Log("SHUTTLE", Form("QueryRunParameters - Invalid parameters for Run %d: "
2107                               "startTime = %d, endTime = %d. Skipping (Shuttle won't be marked as DONE)!",
2108                               run, startTime, endTime));
2109         }
2110         
2111         if (skip)
2112         {
2113                 Log("SHUTTLE", Form("Marking SHUTTLE skipped for run %d", run));
2114                 fLogbookEntry = entry;
2115                 if (!UpdateShuttleLogbook("shuttle_skipped"))
2116                 {
2117                         AliError(Form("Could not update logbook for run %d !", run));
2118                 }
2119                 if (!UpdateTableSkippedCase("ALL"))
2120                 {
2121                         AliError(Form("Could not update FXS tables for run %d !", run));
2122                 }
2123                 fLogbookEntry = 0;
2124         }
2125                         
2126         delete entry;
2127         return 0;
2128 }
2129
2130 //______________________________________________________________________________________________
2131 TMap* AliShuttle::GetValueSet(const char* host, Int_t port, const TSeqCollection* entries,
2132                               DCSType type, Int_t multiSplit)
2133 {
2134         // Retrieve all "entry" data points from the DCS server
2135         // host, port: TSocket connection parameters
2136         // entries: list of name of the alias or data point
2137         // type: kAlias or kDP
2138         // returns TMap of values, 0 when failure
2139         
2140         AliDCSClient client(host, port, fTimeout, fRetries, multiSplit);
2141
2142         TMap* result = 0;
2143         if (type == kAlias)
2144         {
2145                 //result = client.GetAliasValues(entries, GetCurrentStartTime()-offset, 
2146                 //      GetCurrentEndTime()+offset);
2147                 result = client.GetAliasValues(entries, GetStartTimeDCSQuery(), 
2148                         GetEndTimeDCSQuery());
2149         } 
2150         else if (type == kDP)
2151         {
2152                 //result = client.GetDPValues(entries, GetCurrentStartTime()-offset, 
2153                 //      GetCurrentEndTime()+offset);
2154                 result = client.GetDPValues(entries, GetStartTimeDCSQuery(), 
2155                         GetEndTimeDCSQuery());
2156         }
2157
2158         if (result == 0)
2159         {
2160                 Log(fCurrentDetector.Data(), Form("GetValueSet - Can't get entries! Reason: %s",
2161                         client.GetErrorString(client.GetResultErrorCode())));
2162                 if (client.GetResultErrorCode() == AliDCSClient::fgkServerError)        
2163                         Log(fCurrentDetector.Data(), Form("GetValueSet - Server error code: %s",
2164                                 client.GetServerError().Data()));
2165
2166                 return 0;
2167         }
2168                 
2169         return result;
2170 }
2171
2172 //______________________________________________________________________________________________
2173 const char* AliShuttle::GetFile(Int_t system, const char* detector,
2174                 const char* id, const char* source)
2175 {
2176         // Get calibration file from file exchange servers
2177         // First queris the FXS database for the file name, using the run, detector, id and source info
2178         // then calls RetrieveFile(filename) for actual copy to local disk
2179         // run: current run being processed (given by Logbook entry fLogbookEntry)
2180         // detector: the Preprocessor name
2181         // id: provided as a parameter by the Preprocessor
2182         // source: provided by the Preprocessor through GetFileSources function
2183
2184         // check if test mode should simulate a FXS error
2185         if (fTestMode & kErrorFXSFiles)
2186         {
2187                 Log(detector, Form("GetFile - In TESTMODE - Simulating error while connecting to %s FXS", GetSystemName(system)));
2188                 return 0;
2189         }
2190         
2191         // check connection, in case connect
2192         if (!Connect(system))
2193         {
2194                 Log(detector, Form("GetFile - Couldn't connect to %s FXS database", GetSystemName(system)));
2195                 fFXSError = system;
2196                 return 0;
2197         }
2198
2199         // Query preparation
2200         TString sourceName(source);
2201         Int_t nFields = 3;
2202         TString sqlQueryStart = Form("select filePath,size,fileChecksum from %s where",
2203                                                                 fConfig->GetFXSdbTable(system));
2204         TString whereClause = Form("run=%d and detector=\"%s\" and fileId=\"%s\"",
2205                                                                 GetCurrentRun(), detector, id);
2206
2207         if (system == kDAQ || system == kDQM)
2208         {
2209                 whereClause += Form(" and DAQsource=\"%s\"", source);
2210         }
2211         else if (system == kDCS)
2212         {
2213                 sourceName="none";
2214         }
2215         else if (system == kHLT)
2216         {
2217                 whereClause += Form(" and DDLnumbers=\"%s\"", source);
2218         }
2219
2220         TString sqlQuery = Form("%s %s", sqlQueryStart.Data(), whereClause.Data());
2221
2222         AliDebug(2, Form("SQL query: \n%s",sqlQuery.Data()));
2223
2224         // Query execution
2225         TSQLResult* aResult = 0;
2226         aResult = dynamic_cast<TSQLResult*> (fServer[system]->Query(sqlQuery));
2227         if (!aResult) {
2228                 Log(detector, Form("GetFile - Can't execute SQL query to %s database for: id = %s, source = %s",
2229                                 GetSystemName(system), id, sourceName.Data()));
2230                 fFXSError = system;
2231                 return 0;
2232         }
2233
2234         if (aResult->GetRowCount() == 0)
2235         {
2236                 Log(detector,
2237                         Form("GetFile - No entry in %s FXS db for: id = %s, source = %s",
2238                                 GetSystemName(system), id, sourceName.Data()));
2239                 delete aResult;
2240                 return 0;
2241         }
2242
2243         if (aResult->GetRowCount() > 1) {
2244                 Log(detector,
2245                         Form("GetFile - More than one entry in %s FXS db for: id = %s, source = %s",
2246                                 GetSystemName(system), id, sourceName.Data()));
2247                 fFXSError = system;
2248                 delete aResult;
2249                 return 0;
2250         }
2251
2252         if (aResult->GetFieldCount() != nFields) {
2253                 Log(detector,
2254                         Form("GetFileName - Wrong field count in %s FXS db for: id = %s, source = %s",
2255                                 GetSystemName(system), id, sourceName.Data()));
2256                 fFXSError = system;
2257                 delete aResult;
2258                 return 0;
2259         }
2260
2261         TSQLRow* aRow = dynamic_cast<TSQLRow*> (aResult->Next());
2262
2263         if (!aRow){
2264                 Log(detector, Form("GetFile - Empty set result in %s FXS db from query: id = %s, source = %s",
2265                                 GetSystemName(system), id, sourceName.Data()));
2266                 fFXSError = system;
2267                 delete aResult;
2268                 return 0;
2269         }
2270
2271         TString filePath(aRow->GetField(0), aRow->GetFieldLength(0));
2272         TString fileSize(aRow->GetField(1), aRow->GetFieldLength(1));
2273         TString fileChecksum(aRow->GetField(2), aRow->GetFieldLength(2));
2274
2275         delete aResult;
2276         delete aRow;
2277
2278         AliDebug(2, Form("filePath = %s; size = %s, fileChecksum = %s",
2279                                 filePath.Data(), fileSize.Data(), fileChecksum.Data()));
2280
2281         // retrieved file is renamed to make it unique
2282         Int_t dir_lev1 = GetCurrentRun()/10000;
2283         TString localFileName = Form("%s/%d/%d/%s_process/%s_%s_%d_%s_%s.shuttle",
2284                                      GetShuttleTempDir(), dir_lev1, GetCurrentRun(), detector,
2285                                         GetSystemName(system), detector, GetCurrentRun(), 
2286                                         id, sourceName.Data());
2287         Log("SHUTTLE",Form("file from FXS = %s",localFileName.Data())); 
2288
2289
2290         // file retrieval from FXS
2291         UInt_t nRetries = 0;
2292         UInt_t maxRetries = 3;
2293         Bool_t result = kFALSE;
2294
2295         // copy!! if successful TSystem::Exec returns 0
2296         while (nRetries++ < maxRetries) {
2297                 AliDebug(2, Form("Trying to copy file. Retry # %d", nRetries));
2298                 result = RetrieveFile(system, filePath.Data(), localFileName.Data());
2299                 if (!result)
2300                 {
2301                         Log(detector, Form("GetFile - Copy of file %s from %s FXS failed",
2302                                         filePath.Data(), GetSystemName(system)));
2303                         continue;
2304                 } 
2305
2306                 if (fileSize.Length()>0)
2307                 {
2308                         // compare filesize of local file with the one stored in the FXS DB
2309                         Long_t size = -1;
2310                         Int_t sizeComp = gSystem->GetPathInfo(localFileName.Data(), 0, &size, 0, 0);
2311
2312                         if (sizeComp != 0 || size != fileSize.Atoi())
2313                         {
2314                                 Log(detector, Form("GetFile - size of file %s does not match with local copy!",
2315                                                         filePath.Data()));
2316                                 result = kFALSE;
2317                                 continue;
2318                         }
2319
2320                 } else {
2321                         Log(fCurrentDetector, Form("GetFile - size of file %s not set in %s database, skipping comparison",
2322                                                 filePath.Data(), GetSystemName(system)));
2323                 }
2324
2325                 if (fileChecksum.Length()>0)
2326                 {
2327                         // compare md5sum of local file with the one stored in the FXS DB
2328                         if(fileChecksum.Contains(' ')) fileChecksum.Resize(fileChecksum.First(' '));
2329                         Int_t md5Comp = gSystem->Exec(Form("md5sum %s |grep %s > /dev/null 2> /dev/null",
2330                                                 localFileName.Data(), fileChecksum.Data()));
2331
2332                         if (md5Comp != 0)
2333                         {
2334                                 Log(detector, Form("GetFile - md5sum of file %s does not match with local copy!",
2335                                                         filePath.Data()));
2336                                 result = kFALSE;
2337                                 continue;
2338                         }
2339                 } else {
2340                         Log(fCurrentDetector, Form("GetFile - md5sum of file %s not set in %s database, skipping comparison",
2341                                                         filePath.Data(), GetSystemName(system)));
2342                 }
2343                 if (result) break;
2344         }
2345
2346         if (!result) 
2347         {
2348                 fFXSError = system;
2349                 return 0;
2350         }
2351
2352         fFXSCalled[system]=kTRUE;
2353         TObjString *fileParams = new TObjString(Form("%s#!?!#%s", id, sourceName.Data()));
2354         fFXSlist[system].Add(fileParams);
2355
2356         static TString staticLocalFileName;
2357         staticLocalFileName.Form("%s", localFileName.Data());
2358         
2359         Log(fCurrentDetector, Form("GetFile - Retrieved file with id %s and "
2360                         "source %s from %s to %s", id, source, 
2361                         GetSystemName(system), localFileName.Data()));
2362                         
2363         return staticLocalFileName.Data();
2364 }
2365
2366 //______________________________________________________________________________________________
2367 Bool_t AliShuttle::RetrieveFile(UInt_t system, const char* fxsFileName, const char* localFileName)
2368 {
2369         //
2370         // Copies file from FXS to local Shuttle machine
2371         //
2372
2373         // check temp directory: trying to cd to temp; if it does not exist, create it
2374         AliDebug(2, Form("Copy file %s from %s FXS into %s",
2375                         GetSystemName(system), fxsFileName, localFileName));
2376                         
2377         TString tmpDir(localFileName);
2378         
2379         tmpDir = tmpDir(0,tmpDir.Last('/'));
2380
2381         Int_t noDir = gSystem->GetPathInfo(tmpDir.Data(), 0, (Long64_t*) 0, 0, 0);
2382         if (noDir) // temp dir does not exists!
2383         {
2384                 if (gSystem->mkdir(tmpDir.Data(), 1))
2385                 {
2386                         Log(fCurrentDetector.Data(), "RetrieveFile - could not make temp directory!!");
2387                         return kFALSE;
2388                 }
2389         }
2390
2391         TString command = Form("scp -oPort=%d -2 %s@%s:%s/%s %s",
2392                 fConfig->GetFXSPort(system),
2393                 fConfig->GetFXSUser(system),
2394                 fConfig->GetFXSHost(system),
2395                 fConfig->GetFXSBaseFolder(system),
2396                 fxsFileName,
2397                 localFileName);
2398
2399         AliDebug(2, Form("%s",command.Data()));
2400
2401         Bool_t result = (gSystem->Exec(command.Data()) == 0);
2402
2403         return result;
2404 }
2405
2406 //______________________________________________________________________________________________
2407 TList* AliShuttle::GetFileSources(Int_t system, const char* detector, const char* id)
2408 {
2409         //
2410         // Get sources producing the condition file Id from file exchange servers
2411         // if id is NULL all sources are returned (distinct)
2412         //
2413
2414         if (id)
2415         {
2416                 Log(detector, Form("GetFileSources - Querying %s FXS for files with id %s produced by %s", GetSystemName(system), id, detector));
2417         } else {
2418                 Log(detector, Form("GetFileSources - Querying %s FXS for files produced by %s", GetSystemName(system), detector));
2419         }
2420         
2421         // check if test mode should simulate a FXS error
2422         if (fTestMode & kErrorFXSSources)
2423         {
2424                 Log(detector, Form("GetFileSources - In TESTMODE - Simulating error while connecting to %s FXS", GetSystemName(system)));
2425                 return 0;
2426         }
2427
2428         if (system == kDCS)
2429         {
2430                 Log(detector, "GetFileSources - WARNING: DCS system has only one source of data!");
2431                 TList *list = new TList();
2432                 list->SetOwner(1);
2433                 list->Add(new TObjString(" "));
2434                 return list;
2435         }
2436
2437         // check connection, in case connect
2438         if (!Connect(system))
2439         {
2440                 Log(detector, Form("GetFileSources - Couldn't connect to %s FXS database", GetSystemName(system)));
2441                 fFXSError = system;
2442                 return NULL;
2443         }
2444
2445         TString sourceName = "";
2446         if (system == kDAQ || system == kDQM)
2447         {
2448                 sourceName = "DAQsource";
2449         } else if (system == kHLT)
2450         {
2451                 sourceName = "DDLnumbers";
2452         }
2453
2454         TString sqlQueryStart = Form("select distinct %s from %s where", sourceName.Data(), fConfig->GetFXSdbTable(system));
2455         TString whereClause = Form("run=%d and detector=\"%s\"",
2456                                 GetCurrentRun(), detector);
2457         if (id)
2458                 whereClause += Form(" and fileId=\"%s\"", id);
2459         TString sqlQuery = Form("%s %s", sqlQueryStart.Data(), whereClause.Data());
2460
2461         AliDebug(2, Form("SQL query: \n%s",sqlQuery.Data()));
2462
2463         // Query execution
2464         TSQLResult* aResult;
2465         aResult = fServer[system]->Query(sqlQuery);
2466         if (!aResult) {
2467                 Log(detector, Form("GetFileSources - Can't execute SQL query to %s database for id: %s",
2468                                 GetSystemName(system), id));
2469                 fFXSError = system;
2470                 return 0;
2471         }
2472
2473         TList *list = new TList();
2474         list->SetOwner(1);
2475         
2476         if (aResult->GetRowCount() == 0)
2477         {
2478                 Log(detector,
2479                         Form("GetFileSources - No entry in %s FXS table for id: %s", GetSystemName(system), id));
2480                 delete aResult;
2481                 return list;
2482         }
2483
2484         Log(detector, Form("GetFileSources - Found %d sources", aResult->GetRowCount()));
2485
2486         TSQLRow* aRow;
2487         while ((aRow = aResult->Next()))
2488         {
2489
2490                 TString source(aRow->GetField(0), aRow->GetFieldLength(0));
2491                 AliDebug(2, Form("%s = %s", sourceName.Data(), source.Data()));
2492                 list->Add(new TObjString(source));
2493                 delete aRow;
2494         }
2495
2496         delete aResult;
2497
2498         return list;
2499 }
2500
2501 //______________________________________________________________________________________________
2502 TList* AliShuttle::GetFileIDs(Int_t system, const char* detector, const char* source)
2503 {
2504         //
2505         // Get all ids of condition files produced by a given source from file exchange servers
2506         //
2507         
2508         Log(detector, Form("GetFileIDs - Retrieving ids with source %s with %s", source, GetSystemName(system)));
2509
2510         // check if test mode should simulate a FXS error
2511         if (fTestMode & kErrorFXSSources)
2512         {
2513                 Log(detector, Form("GetFileIDs - In TESTMODE - Simulating error while connecting to %s FXS", GetSystemName(system)));
2514                 return 0;
2515         }
2516
2517         // check connection, in case connect
2518         if (!Connect(system))
2519         {
2520                 Log(detector, Form("GetFileIDs - Couldn't connect to %s FXS database", GetSystemName(system)));
2521                 return NULL;
2522         }
2523
2524         TString sourceName = "";
2525         if (system == kDAQ)
2526         {
2527                 sourceName = "DAQsource";
2528         } else if (system == kHLT)
2529         {
2530                 sourceName = "DDLnumbers";
2531         }
2532
2533         TString sqlQueryStart = Form("select fileId from %s where", fConfig->GetFXSdbTable(system));
2534         TString whereClause = Form("run=%d and detector=\"%s\"",
2535                                 GetCurrentRun(), detector);
2536         if (sourceName.Length() > 0 && source)
2537                 whereClause += Form(" and %s=\"%s\"", sourceName.Data(), source);
2538         TString sqlQuery = Form("%s %s", sqlQueryStart.Data(), whereClause.Data());
2539
2540         AliDebug(2, Form("SQL query: \n%s",sqlQuery.Data()));
2541
2542         // Query execution
2543         TSQLResult* aResult;
2544         aResult = fServer[system]->Query(sqlQuery);
2545         if (!aResult) {
2546                 Log(detector, Form("GetFileIDs - Can't execute SQL query to %s database for source: %s",
2547                                 GetSystemName(system), source));
2548                 return 0;
2549         }
2550
2551         TList *list = new TList();
2552         list->SetOwner(1);
2553         
2554         if (aResult->GetRowCount() == 0)
2555         {
2556                 Log(detector,
2557                         Form("GetFileIDs - No entry in %s FXS table for source: %s", GetSystemName(system), source));
2558                 delete aResult;
2559                 return list;
2560         }
2561
2562         Log(detector, Form("GetFileIDs - Found %d ids", aResult->GetRowCount()));
2563
2564         TSQLRow* aRow;
2565
2566         while ((aRow = aResult->Next()))
2567         {
2568
2569                 TString id(aRow->GetField(0), aRow->GetFieldLength(0));
2570                 AliDebug(2, Form("fileId = %s", id.Data()));
2571                 list->Add(new TObjString(id));
2572                 delete aRow;
2573         }
2574
2575         delete aResult;
2576
2577         return list;
2578 }
2579
2580 //______________________________________________________________________________________________
2581 Bool_t AliShuttle::Connect(Int_t system)
2582 {
2583         // Connect to MySQL Server of the system's FXS MySQL databases
2584         // DAQ Logbook, Shuttle Logbook and DAQ FXS db are on the same host
2585         //
2586
2587         // check connection: if already connected return
2588
2589         if(fServer[system] && fServer[system]->IsConnected()) {
2590                 // ping the server              
2591                 if (fServer[system]->PingVerify()==kTRUE){ // connection is still alive
2592                         return kTRUE;
2593                 }               
2594                 else{
2595                         AliWarning(Form("Connection got lost to FXS database for %s. Closing and reconnecting.",
2596                                         AliShuttleInterface::GetSystemName(system)));
2597                         fServer[system]->Close();
2598                         delete fServer[system];
2599                         fServer[system] = 0x0;
2600                 }
2601         }
2602
2603         TString dbHost, dbUser, dbPass, dbName;
2604
2605         if (system < 4) // FXS db servers
2606         {
2607                 dbHost = Form("mysql://%s:%d", fConfig->GetFXSdbHost(system), fConfig->GetFXSdbPort(system));
2608                 dbUser = fConfig->GetFXSdbUser(system);
2609                 dbPass = fConfig->GetFXSdbPass(system);
2610                 dbName =   fConfig->GetFXSdbName(system);
2611         } else { // Run & Shuttle logbook servers
2612         // TODO Will the Shuttle logbook server be the same as the Run logbook server ???
2613                 dbHost = Form("mysql://%s:%d", fConfig->GetDAQlbHost(), fConfig->GetDAQlbPort());
2614                 dbUser = fConfig->GetDAQlbUser();
2615                 dbPass = fConfig->GetDAQlbPass();
2616                 dbName =   fConfig->GetDAQlbDB();
2617         }
2618
2619         fServer[system] = TSQLServer::Connect(dbHost.Data(), dbUser.Data(), dbPass.Data());
2620                 if (!fServer[system] || !fServer[system]->IsConnected()) {
2621                 if(system < 4)
2622                 {
2623                 AliError(Form("Can't establish connection to FXS database for %s",
2624                                         AliShuttleInterface::GetSystemName(system)));
2625                 } else {
2626                 AliError("Can't establish connection to Run logbook.");
2627                 }
2628                 if(fServer[system]) delete fServer[system];
2629                 return kFALSE;
2630         }
2631
2632         // Get tables
2633         TSQLResult* aResult=0;
2634         switch(system){
2635                 case kDAQ:
2636                         aResult = fServer[kDAQ]->GetTables(dbName.Data());
2637                         break;
2638                 case kDCS:
2639                         aResult = fServer[kDCS]->GetTables(dbName.Data());
2640                         break;
2641                 case kHLT:
2642                         aResult = fServer[kHLT]->GetTables(dbName.Data());
2643                         break;
2644                 case kDQM:
2645                         aResult = fServer[kDQM]->GetTables(dbName.Data());
2646                         break;
2647                 default:
2648                         aResult = fServer[4]->GetTables(dbName.Data());
2649                         break;
2650         }
2651
2652         delete aResult;
2653         return kTRUE;
2654 }
2655
2656 //______________________________________________________________________________________________
2657 Bool_t AliShuttle::UpdateTable()
2658 {
2659         //
2660         // Update FXS table filling time_processed field in all rows corresponding to current run and detector
2661         //
2662
2663         Bool_t result = kTRUE;
2664
2665         for (UInt_t system=0; system<4; system++)
2666         {
2667                 if(!fFXSCalled[system]) continue;
2668
2669                 // check connection, in case connect
2670                 if (!Connect(system))
2671                 {
2672                         Log(fCurrentDetector, Form("UpdateTable - Couldn't connect to %s FXS database", GetSystemName(system)));
2673                         result = kFALSE;
2674                         continue;
2675                 }
2676
2677                 TTimeStamp now; // now
2678
2679                 // Loop on FXS list entries
2680                 TIter iter(&fFXSlist[system]);
2681                 TObjString *aFXSentry=0;
2682                 while ((aFXSentry = dynamic_cast<TObjString*> (iter.Next())))
2683                 {
2684                         TString aFXSentrystr = aFXSentry->String();
2685                         TObjArray *aFXSarray = aFXSentrystr.Tokenize("#!?!#");
2686                         if (!aFXSarray || aFXSarray->GetEntries() != 2 )
2687                         {
2688                                 Log(fCurrentDetector, Form("UpdateTable - error updating %s FXS entry. Check string: <%s>",
2689                                         GetSystemName(system), aFXSentrystr.Data()));
2690                                 if(aFXSarray) delete aFXSarray;
2691                                 result = kFALSE;
2692                                 continue;
2693                         }
2694                         const char* fileId = ((TObjString*) aFXSarray->At(0))->GetName();
2695                         const char* source = ((TObjString*) aFXSarray->At(1))->GetName();
2696
2697                         TString whereClause;
2698                         if (system == kDAQ || system == kDQM)
2699                         {
2700                                 whereClause = Form("where run=%d and detector=\"%s\" and fileId=\"%s\" and DAQsource=\"%s\";",
2701                                                         GetCurrentRun(), fCurrentDetector.Data(), fileId, source);
2702                         }
2703                         else if (system == kDCS)
2704                         {
2705                                 whereClause = Form("where run=%d and detector=\"%s\" and fileId=\"%s\";",
2706                                                         GetCurrentRun(), fCurrentDetector.Data(), fileId);
2707                         }
2708                         else if (system == kHLT)
2709                         {
2710                                 whereClause = Form("where run=%d and detector=\"%s\" and fileId=\"%s\" and DDLnumbers=\"%s\";",
2711                                                         GetCurrentRun(), fCurrentDetector.Data(), fileId, source);
2712                         }
2713
2714                         delete aFXSarray;
2715
2716                         TString sqlQuery = Form("update %s set time_processed=%ld %s", fConfig->GetFXSdbTable(system),
2717                                                 (ULong_t)now.GetSec(), whereClause.Data());
2718
2719                         AliDebug(2, Form("SQL query: \n%s",sqlQuery.Data()));
2720
2721                         // Query execution
2722                         TSQLResult* aResult;
2723                         aResult = dynamic_cast<TSQLResult*> (fServer[system]->Query(sqlQuery));
2724                         if (!aResult)
2725                         {
2726                                 Log(fCurrentDetector, Form("UpdateTable - %s db: can't execute SQL query <%s>",
2727                                                                 GetSystemName(system), sqlQuery.Data()));
2728                                 result = kFALSE;
2729                                 continue;
2730                         }
2731                         delete aResult;
2732                 }
2733         }
2734
2735         return result;
2736 }
2737
2738 //_______________________________________________________________________________
2739 Bool_t AliShuttle::UpdateTableSkippedCase(const char* detector)
2740 {
2741         //
2742         // Update FXS table filling time_processed field in all rows corresponding to current run and detector
2743         // if detector = "ALL" update all detectors
2744         //
2745
2746         Bool_t result = kTRUE;
2747
2748         TString detName(detector);
2749
2750         for (UInt_t system=0; system<4; system++)
2751         {
2752
2753                 // check connection, in case connect
2754                 if (!Connect(system))
2755                 {
2756                         Log(fCurrentDetector, Form("UpdateTableSkippedCase - Couldn't connect to %s FXS database", GetSystemName(system)));
2757                         result = kFALSE;
2758                         continue;
2759                 }
2760
2761                 TTimeStamp now; // now
2762
2763                 // Loop on FXS list entries
2764                 TIter iter(&fFXSlist[system]);
2765                         
2766                 TString whereClause;
2767                 if (detName == "ALL") whereClause = Form("where run=%d and time_processed IS NULL;",GetCurrentRun());
2768                 else whereClause = Form("where run=%d and detector=\"%s\" and time_processed IS NULL;",GetCurrentRun(), detector);
2769
2770                 //Log("SHUTTLE",Form(" whereClause = %s ",whereClause.Data()));
2771
2772                 TString sqlQuery = Form("update %s set time_processed=%ld %s", fConfig->GetFXSdbTable(system),
2773                                         (ULong_t)now.GetSec(), whereClause.Data());
2774
2775                 AliDebug(2, Form("SQL query: \n%s",sqlQuery.Data()));
2776
2777                 // Query execution
2778                 TSQLResult* aResult;
2779                 aResult = dynamic_cast<TSQLResult*> (fServer[system]->Query(sqlQuery));
2780                 if (!aResult)
2781                 {
2782                         Log("SHUTTLE", Form("UpdateTableSkippedCase - %s db: can't execute SQL query <%s>",
2783                                                         GetSystemName(system), sqlQuery.Data()));
2784                         result = kFALSE;
2785                         continue;
2786                 }
2787                 delete aResult;
2788                 
2789         }
2790
2791         return result;
2792 }
2793 //______________________________________________________________________________________________
2794 Bool_t AliShuttle::UpdateTableFailCase()
2795 {
2796         // Update FXS table filling time_processed field in all rows corresponding to current run and detector
2797         // this is called in case the preprocessor is declared failed for the current run, because
2798         // the fields are updated only in case of success
2799
2800         Bool_t result = kTRUE;
2801
2802         for (UInt_t system=0; system<4; system++)
2803         {
2804                 // check connection, in case connect
2805                 if (!Connect(system))
2806                 {
2807                         Log(fCurrentDetector, Form("UpdateTableFailCase - Couldn't connect to %s FXS database",
2808                                                         GetSystemName(system)));
2809                         result = kFALSE;
2810                         continue;
2811                 }
2812
2813                 TTimeStamp now; // now
2814
2815                 // Loop on FXS list entries
2816
2817                 TString whereClause = Form("where run=%d and detector=\"%s\";",
2818                                                 GetCurrentRun(), fCurrentDetector.Data());
2819
2820
2821                 TString sqlQuery = Form("update %s set time_processed=%ld %s", fConfig->GetFXSdbTable(system),
2822                                         (ULong_t)now.GetSec(), whereClause.Data());
2823
2824                 AliDebug(2, Form("SQL query: \n%s",sqlQuery.Data()));
2825
2826                 // Query execution
2827                 TSQLResult* aResult;
2828                 aResult = dynamic_cast<TSQLResult*> (fServer[system]->Query(sqlQuery));
2829                 if (!aResult)
2830                 {
2831                         Log(fCurrentDetector, Form("UpdateTableFailCase - %s db: can't execute SQL query <%s>",
2832                                                         GetSystemName(system), sqlQuery.Data()));
2833                         result = kFALSE;
2834                         continue;
2835                 }
2836                 delete aResult;
2837         }
2838
2839         return result;
2840 }
2841
2842 //______________________________________________________________________________________________
2843 Bool_t AliShuttle::UpdateShuttleLogbook(const char* detector, const char* status)
2844 {
2845         //
2846         // Update Shuttle logbook filling detector or shuttle_done column
2847         // ex. of usage: UpdateShuttleLogbook("PHOS", "DONE") or UpdateShuttleLogbook("shuttle_done")
2848         //
2849
2850         // check connection, in case connect
2851         if(!Connect(4)){
2852                 Log("SHUTTLE", "UpdateShuttleLogbook - Couldn't connect to DAQ Logbook.");
2853                 return kFALSE;
2854         }
2855
2856         TString detName(detector);
2857         TString setClause;
2858         if (detName == "shuttle_done" || detName == "shuttle_skipped")
2859         {
2860                 setClause = "set shuttle_done=1";
2861                 
2862                 if (detName == "shuttle_done")
2863                 {
2864                         if (TouchFile() != kTRUE)
2865                         {
2866                                 SendMLRunInfo("Pending");
2867                                 return kFALSE;
2868                         }
2869                         
2870                         SendMLRunInfo("Done");
2871                 }
2872                 else 
2873                         SendMLRunInfo("Skipped");
2874         } 
2875         else {
2876                 TString statusStr(status);
2877                 if(statusStr.Contains("done", TString::kIgnoreCase) ||
2878                    statusStr.Contains("failed", TString::kIgnoreCase)){
2879                         setClause = Form("set %s=\"%s\"", detector, status);
2880                 } else {
2881                         Log("SHUTTLE",
2882                                 Form("UpdateShuttleLogbook - Invalid status <%s> for detector %s",
2883                                         status, detector));
2884                         return kFALSE;
2885                 }
2886         }
2887
2888         TString whereClause = Form("where run=%d", GetCurrentRun());
2889
2890         TString sqlQuery = Form("update %s %s %s",
2891                                         fConfig->GetShuttlelbTable(), setClause.Data(), whereClause.Data());
2892
2893         AliDebug(2, Form("SQL query: \n%s",sqlQuery.Data()));
2894
2895         // Query execution
2896         TSQLResult* aResult;
2897         aResult = dynamic_cast<TSQLResult*> (fServer[4]->Query(sqlQuery));
2898         if (!aResult) {
2899                 Log("SHUTTLE", Form("UpdateShuttleLogbook - Can't execute query <%s>", sqlQuery.Data()));
2900                 return kFALSE;
2901         }
2902         delete aResult;
2903
2904         return kTRUE;
2905 }
2906
2907 //______________________________________________________________________________________________
2908 Int_t AliShuttle::GetCurrentRun() const
2909 {
2910         //
2911         // Get current run from logbook entry
2912         //
2913
2914         return fLogbookEntry ? fLogbookEntry->GetRun() : -1;
2915 }
2916
2917 //______________________________________________________________________________________________
2918 UInt_t AliShuttle::GetCurrentStartTime() const
2919 {
2920         //
2921         // get current start time
2922         //
2923
2924         return fLogbookEntry ? fLogbookEntry->GetStartTime() : 0;
2925 }
2926
2927 //______________________________________________________________________________________________
2928 UInt_t AliShuttle::GetCurrentEndTime() const
2929 {
2930         //
2931         // get current end time from logbook entry
2932         //
2933
2934         return fLogbookEntry ? fLogbookEntry->GetEndTime() : 0;
2935 }
2936 //______________________________________________________________________________________________
2937 UInt_t AliShuttle::GetCurrentYear() const
2938 {
2939         //
2940         // Get current year from logbook entry
2941         //
2942
2943         if (!fLogbookEntry) return 0;
2944         
2945         TTimeStamp startTime(GetCurrentStartTime());
2946         TString year =  Form("%d",startTime.GetDate());
2947         year = year(0,4);
2948         
2949         return year.Atoi();
2950 }
2951
2952 //______________________________________________________________________________________________
2953 const char* AliShuttle::GetLHCPeriod() const
2954 {
2955         //
2956         // Get current LHC period from logbook entry
2957         //
2958
2959         if (!fLogbookEntry) return 0;
2960                 
2961         return fLogbookEntry->GetRunParameter("LHCperiod");
2962 }
2963
2964 //______________________________________________________________________________________________
2965 void AliShuttle::Log(const char* detector, const char* message, UInt_t level)
2966 {
2967         //
2968         // Fill log string with a message
2969         //
2970         
2971         TString logRunDir = GetShuttleLogDir();
2972         if (GetCurrentRun() >=0) {
2973                 Int_t logDir_lev1 = GetCurrentRun()/10000;
2974                 logRunDir += Form("/%d/%d", logDir_lev1, GetCurrentRun());
2975         }               
2976         void* dir = gSystem->OpenDirectory(logRunDir.Data());
2977         if (dir == NULL) {
2978                 if (gSystem->mkdir(logRunDir.Data(), kTRUE)) {
2979                         AliError(Form("Can't open directory <%s>", GetShuttleLogDir()));
2980                         return;
2981                 }
2982
2983         } else {
2984                 gSystem->FreeDirectory(dir);
2985         }
2986
2987         TString toLog = Form("%s UTC (%d): %s - ", TTimeStamp(time(0)).AsString("s"), getpid(), detector);
2988         if (GetCurrentRun() >= 0) 
2989                 toLog += Form("run %d - ", GetCurrentRun());
2990         toLog += Form("%s", message);
2991
2992         AliLog::Message(level, toLog, MODULENAME(), ClassName(), FUNCTIONNAME(), __FILE__, __LINE__);
2993         
2994         // if we redirect the log output already to the file, leave here
2995         if (fOutputRedirected && strcmp(detector, "SHUTTLE") != 0)
2996                 return;
2997
2998         TString fileName = GetLogFileName(detector);
2999         
3000         gSystem->ExpandPathName(fileName);
3001
3002         ofstream logFile;
3003         logFile.open(fileName, ofstream::out | ofstream::app);
3004
3005         if (!logFile.is_open()) {
3006                 AliError(Form("Could not open file %s", fileName.Data()));
3007                 return;
3008         }
3009
3010         logFile << toLog.Data() << "\n";
3011
3012         logFile.close();
3013 }
3014
3015 //______________________________________________________________________________________________
3016 TString AliShuttle::GetLogFileName(const char* detector) const
3017 {
3018         // 
3019         // returns the name of the log file for a given sub detector
3020         //
3021         
3022         TString fileName;
3023         
3024         if (GetCurrentRun() >= 0) 
3025         {
3026                 Int_t logDir_lev1 = GetCurrentRun()/10000;
3027                 fileName.Form("%s/%d/%d/%s.log", GetShuttleLogDir(), logDir_lev1, GetCurrentRun(), 
3028                         detector);
3029         } else {
3030                 fileName.Form("%s/%s.log", GetShuttleLogDir(), detector);
3031         }
3032
3033         return fileName;
3034 }
3035
3036 //______________________________________________________________________________________________
3037 void AliShuttle::SendAlive()
3038 {
3039         // sends alive message to ML
3040         
3041         TMonaLisaText mlStatus("SHUTTLE_status", "Alive");
3042
3043         TList mlList;
3044         mlList.Add(&mlStatus);
3045
3046         fMonaLisa->SendParameters(&mlList, "__PROCESSINGINFO__");
3047 }
3048
3049 //______________________________________________________________________________________________
3050 Bool_t AliShuttle::Collect(Int_t run)
3051 {
3052         //
3053         // Collects conditions data for all UNPROCESSED run written to DAQ LogBook in case of run = -1 (default)
3054         // If a dedicated run is given this run is processed
3055         //
3056         // In operational mode, this is the Shuttle function triggered by the EOR signal.
3057         //
3058
3059         if (run == -1)
3060                 Log("SHUTTLE","Collect - Shuttle called. Collecting conditions data for unprocessed runs");
3061         else
3062                 Log("SHUTTLE", Form("Collect - Shuttle called. Collecting conditions data for run %d", run));
3063
3064         SetLastAction("Starting");
3065
3066         // create ML instance
3067         if (!fMonaLisa)
3068                 fMonaLisa = new TMonaLisaWriter(fConfig->GetMonitorHost(), fConfig->GetMonitorTable());
3069                 
3070         CountOpenRuns();
3071
3072         TString whereClause("where shuttle_done=0");
3073         if (run != -1)
3074                 whereClause += Form(" and run=%d", run);
3075
3076         TObjArray shuttleLogbookEntries;
3077         if (!QueryShuttleLogbook(whereClause, shuttleLogbookEntries))
3078         {
3079                 Log("SHUTTLE", "Collect - Can't retrieve entries from Shuttle logbook");
3080                 return kFALSE;
3081         }
3082
3083         if (shuttleLogbookEntries.GetEntries() == 0)
3084         {
3085                 if (run == -1)
3086                         Log("SHUTTLE","Collect - Found no UNPROCESSED runs in Shuttle logbook");
3087                 else
3088                         Log("SHUTTLE", Form("Collect - Run %d is already DONE "
3089                                                 "or it does not exist in Shuttle logbook", run));
3090                 return kTRUE;
3091         }
3092
3093         for (UInt_t iDet=0; iDet<NDetectors(); iDet++)
3094                 fFirstUnprocessed[iDet] = kTRUE;
3095
3096         if (run != -1)
3097         {
3098                 // query Shuttle logbook for earlier runs, check if some detectors are unprocessed,
3099                 // flag them into fFirstUnprocessed array
3100                 TString whereClauseBis(Form("where shuttle_done=0 and run < %d", run));
3101                 TObjArray tmpLogbookEntries;
3102                 if (!QueryShuttleLogbook(whereClauseBis, tmpLogbookEntries))
3103                 {
3104                         Log("SHUTTLE", "Collect - Can't retrieve entries from Shuttle logbook");
3105                         return kFALSE;
3106                 }
3107
3108                 TIter iter(&tmpLogbookEntries);
3109                 AliShuttleLogbookEntry* anEntry = 0;
3110                 while ((anEntry = dynamic_cast<AliShuttleLogbookEntry*> (iter.Next())))
3111                 {
3112                         for (UInt_t iDet=0; iDet<NDetectors(); iDet++)
3113                         {
3114                                 if (anEntry->GetDetectorStatus(iDet) == AliShuttleLogbookEntry::kUnprocessed)
3115                                 {
3116                                         AliDebug(2, Form("Run %d: setting %s as \"not first time unprocessed\"",
3117                                                         anEntry->GetRun(), GetDetName(iDet)));
3118                                         fFirstUnprocessed[iDet] = kFALSE;
3119                                 }
3120                         }
3121
3122                 }
3123
3124         }
3125
3126         if (!RetrieveConditionsData(shuttleLogbookEntries))
3127         {
3128                 Log("SHUTTLE", "Collect - Process of at least one run failed");
3129                 CountOpenRuns();
3130                 return kFALSE;
3131         }
3132
3133         Log("SHUTTLE", "Collect - Requested run(s) successfully processed");
3134         CountOpenRuns();
3135         return kTRUE;
3136 }
3137
3138 //______________________________________________________________________________________________
3139 Bool_t AliShuttle::RetrieveConditionsData(const TObjArray& dateEntries)
3140 {
3141         //
3142         // Retrieve conditions data for all runs that aren't processed yet
3143         //
3144
3145         Bool_t hasError = kFALSE;
3146
3147         TIter iter(&dateEntries);
3148         AliShuttleLogbookEntry* anEntry;
3149
3150         while ((anEntry = (AliShuttleLogbookEntry*) iter.Next())){
3151                 if (!Process(anEntry)){
3152                         hasError = kTRUE;
3153                 }
3154
3155                 // clean SHUTTLE temp directory
3156                 //TString filename = Form("%s/*.shuttle", GetShuttleTempDir());
3157                 //RemoveFile(filename.Data());
3158         }
3159
3160         return hasError == kFALSE;
3161 }
3162
3163 //______________________________________________________________________________________________
3164 ULong_t AliShuttle::GetTimeOfLastAction() const
3165 {
3166         //
3167         // Gets time of last action
3168         //
3169
3170         ULong_t tmp;
3171
3172         fMonitoringMutex->Lock();
3173
3174         tmp = fLastActionTime;
3175
3176         fMonitoringMutex->UnLock();
3177
3178         return tmp;
3179 }
3180
3181 //______________________________________________________________________________________________
3182 const TString AliShuttle::GetLastAction() const
3183 {
3184         //
3185         // returns a string description of the last action
3186         //
3187
3188         TString tmp;
3189
3190         fMonitoringMutex->Lock();
3191         
3192         tmp = fLastAction;
3193         
3194         fMonitoringMutex->UnLock();
3195
3196         return tmp;
3197 }
3198
3199 //______________________________________________________________________________________________
3200 void AliShuttle::SetLastAction(const char* action)
3201 {
3202         //
3203         // updates the monitoring variables
3204         //
3205
3206         fMonitoringMutex->Lock();
3207
3208         fLastAction = action;
3209         fLastActionTime = time(0);
3210         
3211         fMonitoringMutex->UnLock();
3212 }
3213
3214 //______________________________________________________________________________________________
3215 const char* AliShuttle::GetRunParameter(const char* param)
3216 {
3217         //
3218         // returns run parameter read from DAQ logbook
3219         //
3220
3221         if(!fLogbookEntry) {
3222                 AliError("No logbook entry!");
3223                 return 0;
3224         }
3225
3226         return fLogbookEntry->GetRunParameter(param);
3227 }
3228
3229 //______________________________________________________________________________________________
3230 AliCDBEntry* AliShuttle::GetFromOCDB(const char* detector, const AliCDBPath& path)
3231 {
3232         //
3233         // returns object from OCDB valid for current run
3234         //
3235
3236         if (fTestMode & kErrorOCDB)
3237         {
3238                 Log(detector, "GetFromOCDB - In TESTMODE - Simulating error with OCDB");
3239                 return 0;
3240         }
3241         
3242         AliCDBStorage *sto = AliCDBManager::Instance()->GetStorage(fgkMainCDB);
3243         if (!sto)
3244         {
3245                 Log(detector, "GetFromOCDB - Cannot activate main OCDB for query!");
3246                 return 0;
3247         }
3248
3249         return dynamic_cast<AliCDBEntry*> (sto->Get(path, GetCurrentRun()));
3250 }
3251
3252 //______________________________________________________________________________________________
3253 Bool_t AliShuttle::SendMail(EMailTarget target, Int_t system)
3254 {
3255         //
3256         // sends a mail to the subdetector expert in case of preprocessor error
3257         //
3258         
3259         if (fTestMode != kNone)
3260                 return kTRUE;
3261                 
3262         if (!fConfig->SendMail()) 
3263                 return kTRUE;
3264
3265         if (target == kDCSEMail || target == kFXSEMail) {
3266                 if (!fFirstProcessing)
3267                         return kTRUE;
3268         }
3269
3270         Int_t runMode = (Int_t)fConfig->GetRunMode();
3271         TString tmpStr;
3272         if (runMode == 0) tmpStr = " Nightly Test:";
3273         else tmpStr = " Data Taking:"; 
3274         void* dir = gSystem->OpenDirectory(GetShuttleLogDir());
3275         if (dir == NULL)
3276         {
3277                 if (gSystem->mkdir(GetShuttleLogDir(), kTRUE))
3278                 {
3279                         Log("SHUTTLE", Form("SendMail - Can't open directory <%s>", GetShuttleLogDir()));
3280                         return kFALSE;
3281                 }
3282
3283         } else {
3284                 gSystem->FreeDirectory(dir);
3285         }
3286
3287         // det experts in to
3288         TString to="";
3289         TIter *iterExperts = 0;
3290         if (target == kDCSEMail) {
3291                 iterExperts = new TIter(fConfig->GetAdmins(AliShuttleConfig::kAmanda));
3292         }
3293         else if (target == kFXSEMail) {
3294                 iterExperts = new TIter(fConfig->GetAdmins(system));
3295         }
3296         if (iterExperts) {
3297                 TObjString *anExpert=0;
3298                 while ((anExpert = (TObjString*) iterExperts->Next()))
3299                 {
3300                         to += Form("%s,", anExpert->GetName());
3301                 }
3302                 delete iterExperts;
3303         }
3304
3305         // add subdetector experts      
3306         iterExperts = new TIter(fConfig->GetResponsibles(fCurrentDetector));
3307         TObjString *anExpert=0;
3308         while ((anExpert = (TObjString*) iterExperts->Next()))
3309         {
3310                 to += Form("%s,", anExpert->GetName());
3311         }
3312         delete iterExperts;
3313         
3314         if (to.Length() > 0)
3315           to.Remove(to.Length()-1);
3316         AliDebug(2, Form("to: %s",to.Data()));
3317
3318         if (to.IsNull()) {
3319                 Log("SHUTTLE", Form("List of %d responsibles not set!", (Int_t) target));
3320                 return kFALSE;
3321         }
3322
3323         // SHUTTLE responsibles in cc
3324         TString cc="";
3325         TIter iterAdmins(fConfig->GetAdmins(AliShuttleConfig::kGlobal));
3326         TObjString *anAdmin=0;
3327         while ((anAdmin = (TObjString*) iterAdmins.Next()))
3328         {
3329                 cc += Form("%s,", anAdmin->GetName());
3330         }
3331         if (cc.Length() > 0)
3332           cc.Remove(cc.Length()-1);
3333         AliDebug(2, Form("cc: %s",to.Data()));
3334
3335         // mail body 
3336         TString bodyFileName;
3337         bodyFileName.Form("%s/mail.body", GetShuttleLogDir());
3338         gSystem->ExpandPathName(bodyFileName);
3339
3340         ofstream mailBody;
3341         mailBody.open(bodyFileName, ofstream::out);
3342
3343         if (!mailBody.is_open())
3344         {
3345                 Log("SHUTTLE", Form("Could not open mail body file %s", bodyFileName.Data()));
3346                 return kFALSE;
3347         }
3348
3349
3350         TString subject;
3351         TString body;
3352
3353         if (target == kDCSEMail){
3354                 subject = Form("%s CRITICAL Retrieval of data points for %s FAILED in run %d !",
3355                                 tmpStr.Data(), fCurrentDetector.Data(), GetCurrentRun());
3356                 AliDebug(2, Form("subject: %s", subject.Data()));
3357                 
3358                 body = Form("Dear DCS experts, \n\n");
3359                 body += Form("SHUTTLE couldn\'t retrieve the data points for detector %s "
3360                              "in run %d!!\n\n", fCurrentDetector.Data(), GetCurrentRun());
3361         }
3362         else if (target == kFXSEMail){
3363                 subject = Form("%s CRITICAL FXS communication for %s FAILED in run %d !",
3364                                 tmpStr.Data(), fCurrentDetector.Data(), GetCurrentRun());
3365                 AliDebug(2, Form("subject: %s", subject.Data()));
3366                 TString sys;
3367                 if (system == kDAQ) sys="DAQ";
3368                 else if (system == kDCS) sys="DCS";
3369                 else if (system == kHLT) sys="HLT";
3370                 else if (system == kDQM) sys="DQM";
3371                 else return kFALSE;
3372                 body = Form("Dear  %s FXS experts, \n\n",sys.Data());
3373                 body += Form("SHUTTLE couldn\'t retrieve data from the FXS for detector %s "
3374                              "in run %d!!\n\n", fCurrentDetector.Data(), GetCurrentRun());
3375                 body += Form("The contacted server was:\nDB: %s\nFXS:%s\n\n", fConfig->GetFXSdbHost(system), fConfig->GetFXSHost(system));
3376         }
3377         else {
3378                 subject = Form("%s %s Shuttle preprocessor FAILED in run %d (run type = %s)!",
3379                                        tmpStr.Data(), fCurrentDetector.Data(), GetCurrentRun(), GetRunType());
3380                 AliDebug(2, Form("subject: %s", subject.Data()));
3381         
3382                 body = Form("Dear %s expert(s), \n\n", fCurrentDetector.Data());
3383                 body += Form("SHUTTLE just detected that your preprocessor "
3384                              "failed processing run %d (run type = %s)!!\n\n", 
3385                              GetCurrentRun(), GetRunType());
3386         }
3387
3388         body += Form("Please check %s status on the SHUTTLE monitoring page: \n\n", 
3389                                 fCurrentDetector.Data());
3390         if (fConfig->GetRunMode() == AliShuttleConfig::kTest)
3391         {
3392                 body += Form("\thttp://pcalimonitor.cern.ch/shuttle.jsp?time=24 \n\n");
3393         } else {
3394                 body += Form("\thttp://pcalimonitor.cern.ch/shuttle.jsp?instance=PROD&time=24 \n\n");
3395         }
3396         
3397         
3398         TString logFolder = "logs";
3399         if (fConfig->GetRunMode() == AliShuttleConfig::kProd) 
3400                 logFolder += "_PROD";
3401         
3402         
3403         body += Form("Find the %s log for the current run on \n\n"
3404                 "\thttp://pcalishuttle02.cern.ch/%s/%d/%d/%s.log \n\n", 
3405                      fCurrentDetector.Data(), logFolder.Data(), GetCurrentRun()/10000,  
3406                                 GetCurrentRun(), fCurrentDetector.Data());
3407         body += Form("The last 15 lines of %s log file are following:\n\n", fCurrentDetector.Data());
3408
3409         AliDebug(2, Form("Body begin: %s", body.Data()));
3410
3411         mailBody << body.Data();
3412         mailBody.close();
3413         mailBody.open(bodyFileName, ofstream::out | ofstream::app);
3414
3415         TString logFileName = Form("%s/%d/%d/%s.log", GetShuttleLogDir(), 
3416                 GetCurrentRun()/10000, GetCurrentRun(), fCurrentDetector.Data());
3417         TString tailCommand = Form("tail -n 15 %s >> %s", logFileName.Data(), bodyFileName.Data());
3418         if (gSystem->Exec(tailCommand.Data()))
3419         {
3420                 mailBody << Form("%s log file not found ...\n\n", fCurrentDetector.Data());
3421         }
3422
3423         TString endBody = Form("------------------------------------------------------\n\n");
3424         endBody += Form("In case of problems please contact the SHUTTLE core team.\n\n");
3425         endBody += "Please do not answer this message directly, it is automatically generated.\n\n";
3426         endBody += "Greetings,\n\n \t\t\tthe SHUTTLE\n";
3427
3428         AliDebug(2, Form("Body end: %s", endBody.Data()));
3429
3430         mailBody << endBody.Data();
3431
3432         mailBody.close();
3433
3434         // send mail!
3435         TString mailCommand = Form("mail -s \"%s\" -c %s %s < %s",
3436                                                 subject.Data(),
3437                                                 cc.Data(),
3438                                                 to.Data(),
3439                                                 bodyFileName.Data());
3440         AliDebug(2, Form("mail command: %s", mailCommand.Data()));
3441
3442         Bool_t result = gSystem->Exec(mailCommand.Data());
3443
3444         return result == 0;
3445 }
3446 //______________________________________________________________________________________________
3447 const char* AliShuttle::GetRunType()
3448 {
3449         //
3450         // returns run type read from "run type" logbook
3451         //
3452
3453         if(!fLogbookEntry) {
3454                 AliError("No logbook entry!");
3455                 return 0;
3456         }
3457
3458         return fLogbookEntry->GetRunType();
3459 }
3460
3461 //______________________________________________________________________________________________
3462 Bool_t AliShuttle::GetHLTStatus()
3463 {
3464         // Return HLT status (ON=1 OFF=0)
3465         // Converts the HLT status from the mode string read in the run logbook (not just a bool)
3466
3467         if(!fLogbookEntry) {
3468                 AliError("No logbook entry!");
3469                 return 0;
3470         }
3471