1 /**************************************************************************
2 * Copyright(c) 1998-1999, ALICE Experiment at CERN, All rights reserved. *
4 * Author: The ALICE Off-line Project. *
5 * Contributors are mentioned in the code where appropriate. *
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 **************************************************************************/
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.
32 #include "AliShuttle.h"
34 #include "AliCDBManager.h"
35 #include "AliCDBStorage.h"
37 #include "AliCDBRunRange.h"
38 #include "AliCDBPath.h"
39 #include "AliCDBEntry.h"
40 #include "AliShuttleConfig.h"
41 #include "DCSClient/AliDCSClient.h"
43 #include "AliPreprocessor.h"
44 #include "AliShuttleStatus.h"
45 #include "AliShuttleLogbookEntry.h"
50 #include <TTimeStamp.h>
51 #include <TObjString.h>
52 #include <TSQLServer.h>
53 #include <TSQLResult.h>
56 #include <TSystemDirectory.h>
57 #include <TSystemFile.h>
60 #include <TGridResult.h>
63 #include <TMonaLisaWriter.h>
67 #include <sys/types.h>
75 //______________________________________________________________________________________________
76 AliShuttle::AliShuttle(const AliShuttleConfig* config,
77 UInt_t timeout, Int_t retries):
79 fTimeout(timeout), fRetries(retries),
91 fReadTestMode(kFALSE),
92 fOutputRedirected(kFALSE)
95 // config: AliShuttleConfig used
96 // timeout: timeout used for AliDCSClient connection
97 // retries: the number of retries in case of connection error.
100 if (!fConfig->IsValid()) AliFatal("********** !!!!! Invalid configuration !!!!! **********");
101 for(int iSys=0;iSys<5;iSys++) {
104 fFXSlist[iSys].SetOwner(kTRUE);
106 fPreprocessorMap.SetOwner(kTRUE);
108 for (UInt_t iDet=0; iDet<NDetectors(); iDet++)
109 fFirstUnprocessed[iDet] = kFALSE;
111 fMonitoringMutex = new TMutex();
114 //______________________________________________________________________________________________
115 AliShuttle::~AliShuttle()
121 fPreprocessorMap.DeleteAll();
122 for(int iSys=0;iSys<5;iSys++)
124 fServer[iSys]->Close();
125 delete fServer[iSys];
134 if (fMonitoringMutex)
136 delete fMonitoringMutex;
137 fMonitoringMutex = 0;
141 //______________________________________________________________________________________________
142 void AliShuttle::RegisterPreprocessor(AliPreprocessor* preprocessor)
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()).
151 const char* detName = preprocessor->GetName();
152 if(GetDetPos(detName) < 0)
153 AliFatal(Form("********** !!!!! Invalid detector name: %s !!!!! **********", detName));
155 if (fPreprocessorMap.GetValue(detName)) {
156 AliWarning(Form("AliPreprocessor %s is already registered!", detName));
160 fPreprocessorMap.Add(new TObjString(detName), preprocessor);
162 //______________________________________________________________________________________________
163 Bool_t AliShuttle::Store(const AliCDBPath& path, TObject* object,
164 AliCDBMetaData* metaData, Int_t validityStart, Bool_t validityInfinite)
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).
172 return StoreLocally(fgkLocalCDB, path, object, metaData, validityStart, validityInfinite);
175 //______________________________________________________________________________________________
176 Bool_t AliShuttle::StoreReferenceData(const AliCDBPath& path, TObject* object, AliCDBMetaData* metaData)
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).
183 return StoreLocally(fgkLocalRefStorage, path, object, metaData);
186 //______________________________________________________________________________________________
187 Bool_t AliShuttle::StoreLocally(const TString& localUri,
188 const AliCDBPath& path, TObject* object, AliCDBMetaData* metaData,
189 Int_t validityStart, Bool_t validityInfinite)
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
203 // returns 0 if fail, 1 otherwise
206 if (fTestMode & kErrorStorage)
208 Log(fCurrentDetector, "StoreLocally - In TESTMODE - Simulating error while storing locally");
212 const char* cdbType = (localUri == fgkLocalCDB) ? "CDB" : "Reference";
214 Int_t firstRun = GetCurrentRun() - validityStart;
216 AliWarning("First valid run happens to be less than 0! Setting it to 0.");
221 if(validityInfinite) {
222 lastRun = AliCDBRunRange::Infinity();
224 lastRun = GetCurrentRun();
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);
230 if(! dynamic_cast<TObjString*> (metaData->GetProperty("RunUsed(TObjString)"))){
231 TObjString runUsed = Form("%d", GetCurrentRun());
232 metaData->SetProperty("RunUsed(TObjString)", runUsed.Clone());
235 Bool_t result = kFALSE;
237 if (!(AliCDBManager::Instance()->GetStorage(localUri))) {
238 Log("SHUTTLE", Form("StoreLocally - Cannot activate local %s storage", cdbType));
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);
249 Log(fCurrentDetector, Form("StoreLocally - Can't store object <%s>!", id.ToString().Data()));
256 //______________________________________________________________________________________________
257 Bool_t AliShuttle::StoreOCDB()
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.
265 UpdateShuttleStatus(AliShuttleStatus::kStoreStarted);
267 if (fTestMode & kErrorGrid)
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");
274 Log("SHUTTLE","StoreOCDB - Storing OCDB data ...");
275 Int_t resultCDB = StoreOCDB(fgkMainCDB);
277 Log("SHUTTLE","StoreOCDB - Storing reference data ...");
278 Int_t resultRef = StoreOCDB(fgkMainRefStorage);
280 Log("SHUTTLE","StoreOCDB - Storing reference files ...");
281 Bool_t resultRefFiles = CopyFilesToGrid("reference");
283 Bool_t resultMetadata = kTRUE;
284 if(fCurrentDetector == "GRP")
286 Log("SHUTTLE","StoreOCDB - Storing Run Metadata file ...");
287 resultMetadata = CopyFilesToGrid("metadata");
290 Int_t storeResult = 0;
292 if (resultCDB < 0 || resultRef < 0 || resultRefFiles == kFALSE || resultMetadata == kFALSE)
294 else if (resultCDB > 0 || resultRef > 0)
300 Form("\t\t\t****** run %d - %s: STORAGE ERROR ******",
301 GetCurrentRun(), fCurrentDetector.Data()));
302 UpdateShuttleStatus(AliShuttleStatus::kStoreError);
304 else if (storeResult > 0)
307 Form("\t\t\t****** run %d - %s: STORAGE DELAYED ******",
308 GetCurrentRun(), fCurrentDetector.Data()));
309 UpdateShuttleStatus(AliShuttleStatus::kStoreDelayed);
311 else if (storeResult == 0)
314 Form("\t\t\t****** run %d - %s: DONE ******",
315 GetCurrentRun(), fCurrentDetector.Data()));
316 UpdateShuttleStatus(AliShuttleStatus::kDone);
317 UpdateShuttleLogbook(fCurrentDetector, "DONE");
320 return (storeResult == 0);
323 //______________________________________________________________________________________________
324 Int_t AliShuttle::StoreOCDB(const TString& gridURI)
327 // Called by StoreOCDB(), performs actual storage to the main OCDB and reference storages (Grid)
330 // -2 initialization error
333 // 1 storage delayed (e.g. previous unprocessed runs)
336 TObjArray* gridIds=0;
338 Bool_t result = kTRUE;
339 Bool_t delayed = kFALSE;
341 const char* type = 0;
343 if(gridURI == fgkMainCDB) {
345 localURI = fgkLocalCDB;
346 } else if(gridURI == fgkMainRefStorage) {
348 localURI = fgkLocalRefStorage;
350 AliError(Form("Invalid storage URI: %s", gridURI.Data()));
354 AliCDBManager* man = AliCDBManager::Instance();
356 AliCDBStorage *gridSto = man->GetStorage(gridURI);
359 Form("StoreOCDB - cannot activate main %s storage", type));
362 gridSto->SetMirrorSEs(fgkMirrorSEs.Data());
364 gridIds = gridSto->GetQueryCDBList();
366 // get objects previously stored in local CDB
367 AliCDBStorage *localSto = man->GetStorage(localURI);
370 Form("StoreOCDB - cannot activate local %s storage", type));
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);
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);
387 Log(fCurrentDetector.Data(), Form("Attempting to store %s", aLocId.ToString().Data()));
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)])
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()));
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())
411 // skip all objects valid up to infinity
412 if (aGridId->GetLastRun() == AliCDBRunRange::Infinity())
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()));
424 Bool_t storeOk = kFALSE;
427 Log(fCurrentDetector.Data(), Form("Prechecks succeeded. Ready to store %s", aLocId.ToString().Data()));
428 storeOk = gridSto->Put(aLocEntry);
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));
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()));
447 if (!store || storeOk) {
448 // removing local file...
450 localSto->IdToFilename(aLocId, filename);
451 Log("SHUTTLE", Form("StoreOCDB - Removing local file %s", filename.Data()));
452 RemoveFile(filename.Data());
455 localEntries->Clear();
457 Int_t returnCode = 0;
459 if (result == kFALSE)
461 else if (delayed != kFALSE)
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));
470 //______________________________________________________________________________________________
471 Bool_t AliShuttle::CleanReferenceStorage(const char* detector)
473 // clears the directory used to store reference files of a given subdetector
475 AliCDBManager* man = AliCDBManager::Instance();
476 AliCDBStorage* sto = man->GetStorage(fgkLocalRefStorage);
477 TString localBaseFolder = sto->GetBaseFolder();
479 TString targetDir = GetRefFilePrefix(localBaseFolder.Data(), detector);
481 Log("SHUTTLE", Form("CleanReferenceStorage - Cleaning %s", targetDir.Data()));
484 begin.Form("%d_", GetCurrentRun());
486 TSystemDirectory* baseDir = new TSystemDirectory("/", targetDir);
490 TList* dirList = baseDir->GetListOfFiles();
493 if (!dirList) return kTRUE;
495 if (dirList->GetEntries() < 3) // to be changed to 4?
501 Int_t nDirs = 0, nDel = 0;
502 TIter dirIter(dirList);
503 TSystemFile* entry = 0;
505 Bool_t success = kTRUE;
507 while ((entry = dynamic_cast<TSystemFile*> (dirIter.Next())))
509 if (entry->IsDirectory())
512 TString fileName(entry->GetName());
513 if (!fileName.BeginsWith(begin))
519 Int_t result = gSystem->Unlink(fileName.Data());
523 Log("SHUTTLE", Form("CleanReferenceStorage - Could not delete file %s!", fileName.Data()));
531 Log("SHUTTLE", Form("CleanReferenceStorage - %d (over %d) reference files in folder %s were deleted.",
532 nDel, nDirs, targetDir.Data()));
543 Int_t result = gSystem->GetPathInfo(targetDir, 0, (Long64_t*) 0, 0, 0);
547 result = gSystem->Exec(Form("rm -rf %s", targetDir.Data()));
550 Log("SHUTTLE", Form("CleanReferenceStorage - Could not clean directory %s", targetDir.Data()));
555 result = gSystem->mkdir(targetDir, kTRUE);
558 Log("SHUTTLE", Form("CleanReferenceStorage - Error creating base directory %s", targetDir.Data()));
565 //______________________________________________________________________________________________
566 Bool_t AliShuttle::StoreReferenceFile(const char* detector, const char* localFile, const char* gridFileName)
569 // Stores reference file directly (without opening it). This function stores the file locally.
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
576 if (fTestMode & kErrorStorage)
578 Log(fCurrentDetector, "StoreReferenceFile - In TESTMODE - Simulating error while storing locally");
582 AliCDBManager* man = AliCDBManager::Instance();
583 AliCDBStorage* sto = man->GetStorage(fgkLocalRefStorage);
585 TString localBaseFolder = sto->GetBaseFolder();
587 TString target = GetRefFilePrefix(localBaseFolder.Data(), detector);
588 target.Append(Form("/%d_%s", GetCurrentRun(), gridFileName));
590 return CopyFileLocally(localFile, target);
593 //______________________________________________________________________________________________
594 Bool_t AliShuttle::StoreRunMetadataFile(const char* localFile, const char* gridFileName)
597 // Stores Run metadata file to the Grid, in the run folder
599 // Only GRP can call this function.
601 if (fTestMode & kErrorStorage)
603 Log(fCurrentDetector, "StoreRunMetaDataFile - In TESTMODE - Simulating error while storing locally");
607 AliCDBManager* man = AliCDBManager::Instance();
608 AliCDBStorage* sto = man->GetStorage(fgkLocalRefStorage);
610 TString localBaseFolder = sto->GetBaseFolder();
612 // Build Run level folder
613 // folder = /alice/data/year/lhcPeriod/runNb/raw
616 TString lhcPeriod = GetLHCPeriod();
617 if (lhcPeriod.Length() == 0)
619 Log("SHUTTLE","StoreRunMetaDataFile - LHCPeriod not found in logbook!");
623 // TODO partitions with one detector only write data into LHCperiod_DET
624 TString partition = GetRunParameter("detector");
626 if (partition.Length() > 0 && partition != "ALICE")
628 lhcPeriod.Append(Form("_%s", partition.Data()));
629 Log(fCurrentDetector, Form("Run data tags merged file will be written in %s",
633 TString target = Form("%s/GRP/RunMetadata%s%d/%s/%09d/raw/%s",
634 localBaseFolder.Data(), fConfig->GetAlienPath(), GetCurrentYear(),
635 lhcPeriod.Data(), GetCurrentRun(), gridFileName);
637 return CopyFileLocally(localFile, target);
640 //______________________________________________________________________________________________
641 Bool_t AliShuttle::CopyFileLocally(const char* localFile, const TString& target)
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)
650 TString targetDir(target(0, target.Last('/')));
652 //try to open base dir folder, if it does not exist
653 void* dir = gSystem->OpenDirectory(targetDir.Data());
655 if (gSystem->mkdir(targetDir.Data(), kTRUE)) {
656 Log("SHUTTLE", Form("CopyFileLocally - Can't open directory <%s>", targetDir.Data()));
661 gSystem->FreeDirectory(dir);
666 result = gSystem->GetPathInfo(localFile, 0, (Long64_t*) 0, 0, 0);
669 Log("SHUTTLE", Form("CopyFileLocally - %s does not exist", localFile));
673 result = gSystem->GetPathInfo(target, 0, (Long64_t*) 0, 0, 0);
676 Log("SHUTTLE", Form("CopyFileLocally - target file %s already exist, removing...", target.Data()));
677 if (gSystem->Unlink(target.Data()))
679 Log("SHUTTLE", Form("CopyFileLocally - Could not remove existing target file %s!", target.Data()));
684 result = gSystem->CopyFile(localFile, target);
688 Log("SHUTTLE", Form("CopyFileLocally - File %s stored locally to %s", localFile, target.Data()));
693 Log("SHUTTLE", Form("CopyFileLocally - Could not store file %s to %s! Error code = %d",
694 localFile, target.Data(), result));
702 //______________________________________________________________________________________________
703 Bool_t AliShuttle::CopyFilesToGrid(const char* type)
706 // Transfers local files to the Grid. Local files can be reference files
707 // or run metadata file (from GRP only).
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>
714 AliCDBManager* man = AliCDBManager::Instance();
715 AliCDBStorage* sto = man->GetStorage(fgkLocalRefStorage);
718 TString localBaseFolder = sto->GetBaseFolder();
724 if (strcmp(type, "reference") == 0)
726 dir = GetRefFilePrefix(localBaseFolder.Data(), fCurrentDetector.Data());
727 AliCDBStorage* gridSto = man->GetStorage(fgkMainRefStorage);
730 TString gridBaseFolder = gridSto->GetBaseFolder();
731 alienDir = GetRefFilePrefix(gridBaseFolder.Data(), fCurrentDetector.Data());
732 begin = Form("%d_", GetCurrentRun());
734 else if (strcmp(type, "metadata") == 0)
737 TString lhcPeriod = GetLHCPeriod();
739 if (lhcPeriod.Length() == 0)
741 Log("SHUTTLE","CopyFilesToGrid - LHCPeriod not found in logbook!");
745 // TODO partitions with one detector only write data into LHCperiod_DET
746 TString partition = GetRunParameter("detector");
748 if (partition.Length() > 0 && partition != "ALICE")
750 lhcPeriod.Append(Form("_%s", partition.Data()));
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());
762 Log("SHUTTLE", "CopyFilesToGrid - Unexpected: type label must be reference or metadata!");
766 TSystemDirectory* baseDir = new TSystemDirectory("/", dir);
770 TList* dirList = baseDir->GetListOfFiles();
773 if (!dirList) return kTRUE;
775 if (dirList->GetEntries() < 3)
783 Log("SHUTTLE", "CopyFilesToGrid - Connection to Grid failed: Cannot continue!");
788 Int_t nDirs = 0, nTransfer = 0;
789 TIter dirIter(dirList);
790 TSystemFile* entry = 0;
792 Bool_t success = kTRUE;
793 Bool_t first = kTRUE;
795 while ((entry = dynamic_cast<TSystemFile*> (dirIter.Next())))
797 if (entry->IsDirectory())
800 TString fileName(entry->GetName());
801 if (!fileName.BeginsWith(begin))
809 // check that folder exists, otherwise create it
810 TGridResult* result = gGrid->Ls(alienDir.Data(), "a");
818 if (!result->GetFileName(1)) // TODO: It looks like element 0 is always 0!!
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))
824 Log("SHUTTLE", Form("CopyFilesToGrid - Cannot create directory %s",
829 Log("SHUTTLE",Form("CopyFilesToGrid - Folder %s created", alienDir.Data()));
833 Log("SHUTTLE",Form("CopyFilesToGrid - Folder %s found", alienDir.Data()));
837 TString fullLocalPath;
838 fullLocalPath.Form("%s/%s", dir.Data(), fileName.Data());
840 TString fullGridPath;
841 fullGridPath.Form("alien://%s/%s", alienDir.Data(), fileName.Data());
843 Bool_t result = TFile::Cp(fullLocalPath, fullGridPath);
847 Log("SHUTTLE", Form("CopyFilesToGrid - Copying local file %s to %s succeeded!",
848 fullLocalPath.Data(), fullGridPath.Data()));
849 RemoveFile(fullLocalPath);
854 Log("SHUTTLE", Form("CopyFilesToGrid - Copying local file %s to %s FAILED!",
855 fullLocalPath.Data(), fullGridPath.Data()));
860 Log("SHUTTLE", Form("CopyFilesToGrid - %d (over %d) files in folder %s copied to Grid.",
861 nTransfer, nDirs, dir.Data()));
868 //______________________________________________________________________________________________
869 const char* AliShuttle::GetRefFilePrefix(const char* base, const char* detector)
872 // Get folder name of reference files
875 TString offDetStr(GetOfflineDetName(detector));
877 if (offDetStr == "ITS" || offDetStr == "MUON" || offDetStr == "PHOS")
879 dir.Form("%s/%s/%s", base, offDetStr.Data(), detector);
881 dir.Form("%s/%s", base, offDetStr.Data());
887 //______________________________________________________________________________________________
888 void AliShuttle::CleanLocalStorage(const TString& uri)
891 // Called in case the preprocessor is declared failed. Remove remaining objects from the local storages.
894 const char* type = 0;
895 if(uri == fgkLocalCDB) {
897 } else if(uri == fgkLocalRefStorage) {
900 AliError(Form("Invalid storage URI: %s", uri.Data()));
904 AliCDBManager* man = AliCDBManager::Instance();
906 // open local storage
907 AliCDBStorage *localSto = man->GetStorage(uri);
910 Form("CleanLocalStorage - cannot activate local %s storage", type));
914 TString filename(Form("%s/%s/*/Run*_v%d_s*.root",
915 localSto->GetBaseFolder().Data(), GetOfflineDetName(fCurrentDetector.Data()), GetCurrentRun()));
917 AliDebug(2, Form("filename = %s", filename.Data()));
919 Log("SHUTTLE", Form("Removing remaining local files for run %d and detector %s ...",
920 GetCurrentRun(), fCurrentDetector.Data()));
922 RemoveFile(filename.Data());
926 //______________________________________________________________________________________________
927 void AliShuttle::RemoveFile(const char* filename)
930 // removes local file
933 TString command(Form("rm -f %s", filename));
935 Int_t result = gSystem->Exec(command.Data());
938 Log("SHUTTLE", Form("RemoveFile - %s: Cannot remove file %s!",
939 fCurrentDetector.Data(), filename));
943 //______________________________________________________________________________________________
944 AliShuttleStatus* AliShuttle::ReadShuttleStatus()
947 // Reads the AliShuttleStatus from the CDB
955 Int_t path1 = GetCurrentRun()/10000;
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()));
963 if (!fStatusEntry) return 0;
964 fStatusEntry->SetOwner(1);
966 AliShuttleStatus* status = dynamic_cast<AliShuttleStatus*> (fStatusEntry->GetObject());
968 AliError("Invalid object stored to CDB!");
975 //______________________________________________________________________________________________
976 Bool_t AliShuttle::WriteShuttleStatus(AliShuttleStatus* status)
979 // writes the status for one subdetector
987 Int_t run = GetCurrentRun();
988 Int_t path1 = run/10000;
989 TString path1_string = Form("%d",path1);
991 AliCDBId id(AliCDBPath("SHUTTLE", fCurrentDetector, path1_string), run, run);
993 fStatusEntry = new AliCDBEntry(status, id, new AliCDBMetaData);
994 fStatusEntry->SetOwner(1);
996 Int_t logLevel = AliLog::GetGlobalLogLevel();
997 AliLog::SetGlobalLogLevel(AliLog::kError);
999 UInt_t result = AliCDBManager::Instance()->GetStorage(fgkLocalCDB)->Put(fStatusEntry);
1001 AliLog::SetGlobalLogLevel((AliLog::EType_t)logLevel);
1004 Log("SHUTTLE", Form("WriteShuttleStatus - Failed for %s, run %d",
1005 fCurrentDetector.Data(), run));
1014 //______________________________________________________________________________________________
1015 void AliShuttle::UpdateShuttleStatus(AliShuttleStatus::Status newStatus, Bool_t increaseCount)
1018 // changes the AliShuttleStatus for the given detector and run to the given status
1022 AliError("UNEXPECTED: fStatusEntry empty");
1026 AliShuttleStatus* status = dynamic_cast<AliShuttleStatus*> (fStatusEntry->GetObject());
1029 Log("SHUTTLE", "UpdateShuttleStatus - UNEXPECTED: status could not be read from current CDB entry");
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);
1040 status->SetStatus(newStatus);
1041 if (increaseCount) status->IncreaseCount();
1043 Int_t logLevel = AliLog::GetGlobalLogLevel();
1044 AliLog::SetGlobalLogLevel(AliLog::kError);
1046 AliCDBManager::Instance()->GetStorage(fgkLocalCDB)->Put(fStatusEntry);
1048 AliLog::SetGlobalLogLevel((AliLog::EType_t)logLevel);
1053 //______________________________________________________________________________________________
1054 void AliShuttle::SendMLDetInfo()
1057 // sends ML information about the current status of the current detector being processed
1060 AliShuttleStatus* status = dynamic_cast<AliShuttleStatus*> (fStatusEntry->GetObject());
1063 Log("SHUTTLE", "SendMLDetInfo - UNEXPECTED: status could not be read from current CDB entry");
1067 TMonaLisaText mlStatus(Form("%s_status", fCurrentDetector.Data()), status->GetStatusName());
1068 TMonaLisaValue mlRetryCount(Form("%s_count", fCurrentDetector.Data()), status->GetCount());
1071 mlList.Add(&mlStatus);
1072 mlList.Add(&mlRetryCount);
1075 mlID.Form("%d", GetCurrentRun());
1076 fMonaLisa->SendParameters(&mlList, mlID);
1079 //______________________________________________________________________________________________
1080 Bool_t AliShuttle::ContinueProcessing()
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
1086 if (!fConfig->HostProcessDetector(fCurrentDetector))
1089 AliPreprocessor* aPreprocessor =
1090 dynamic_cast<AliPreprocessor*> (fPreprocessorMap.GetValue(fCurrentDetector));
1093 Log("SHUTTLE", Form("ContinueProcessing - %s: no preprocessor registered", fCurrentDetector.Data()));
1097 AliShuttleLogbookEntry::Status entryStatus =
1098 fLogbookEntry->GetDetectorStatus(fCurrentDetector);
1100 if (entryStatus != AliShuttleLogbookEntry::kUnprocessed) {
1101 Log("SHUTTLE", Form("ContinueProcessing - %s is %s",
1102 fCurrentDetector.Data(),
1103 fLogbookEntry->GetDetectorStatusName(entryStatus)));
1107 // if we get here, according to Shuttle logbook subdetector is in UNPROCESSED state
1109 // check if current run is first unprocessed run for current detector
1110 if (fConfig->StrictRunOrder(fCurrentDetector) &&
1111 !fFirstUnprocessed[GetDetPos(fCurrentDetector)])
1113 if (fTestMode == kNone)
1115 Log("SHUTTLE", Form("ContinueProcessing - %s requires strict run ordering"
1116 " but this is not the first unprocessed run!",fCurrentDetector.Data()));
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()));
1128 // Is the subdetector processed first time for this run?
1129 fFirstProcessing = kFALSE;
1131 AliShuttleStatus* status = ReadShuttleStatus();
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);
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())));
1150 if (status->GetStatus() == AliShuttleStatus::kSkipped)
1152 UpdateShuttleLogbook(fCurrentDetector.Data(), "DONE");
1155 UpdateShuttleLogbook(fCurrentDetector.Data(), status->GetStatusName(status->GetStatus()));
1160 if (status->GetStatus() == AliShuttleStatus::kStoreStarted || status->GetStatus() == AliShuttleStatus::kStoreDelayed ||status->GetStatus() == AliShuttleStatus::kStoreError) {
1162 Form("ContinueProcessing - %s: Grid storage of one or more "
1163 "objects failed. Trying again now",
1164 fCurrentDetector.Data()));
1169 // if we get here, there is a restart
1170 Bool_t cont = kFALSE;
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);
1180 // there may still be objects in local OCDB and reference storage
1181 // and FXS databases may be not updated: do it now!
1183 // TODO Currently disabled, we want to keep files in case of failure!
1184 // CleanLocalStorage(fgkLocalCDB);
1185 // CleanLocalStorage(fgkLocalRefStorage);
1186 // UpdateTableFailCase();
1188 // Send mail to detector expert!
1189 Log("SHUTTLE", Form("ContinueProcessing - Sending mail to %s expert...",
1190 fCurrentDetector.Data()));
1191 // det experts in to
1193 TIter *iterExperts = 0;
1194 iterExperts = new TIter(fConfig->GetResponsibles(fCurrentDetector));
1195 TObjString *anExpert=0;
1196 while ((anExpert = (TObjString*) iterExperts->Next()))
1198 to += Form("%s, \n", anExpert->GetName());
1202 if (to.Length() > 0)
1203 to.Remove(to.Length()-3);
1204 AliDebug(2, Form("to: %s",to.Data()));
1207 Log("SHUTTLE", Form("List of %s responsibles not set!", fCurrentDetector.Data()));
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()));
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;
1229 UpdateShuttleStatus(AliShuttleStatus::kStarted, increaseCount);
1236 //______________________________________________________________________________________________
1237 void AliShuttle::SendMLRunInfo(const char* status)
1240 // Send information about this run to ML
1242 TMonaLisaText mlStatus("SHUTTLE_status", status);
1243 TString runType(fLogbookEntry->GetRunType());
1244 if (strlen(fLogbookEntry->GetRunParameter("log")) > 0){
1247 runType += fLogbookEntry->GetRunParameter("log");
1250 if (fLogbookEntry->GetDATestMode()){
1251 runType += " (DATest)";
1253 TMonaLisaText mlRunType("SHUTTLE_runtype", runType);
1256 mlList.Add(&mlStatus);
1257 mlList.Add(&mlRunType);
1260 mlID.Form("%d", GetCurrentRun());
1261 fMonaLisa->SendParameters(&mlList, mlID);
1264 //______________________________________________________________________________________________
1265 Int_t AliShuttle::GetMem(Int_t pid)
1267 // invokes ps to get the memory consumption of the process <pid>
1268 // returns -1 in case of error
1271 checkStr.Form("ps -o vsize --pid %d | tail -n 1", pid);
1272 FILE* pipe = gSystem->OpenPipe(checkStr, "r");
1275 Log("SHUTTLE", Form("Process - Error: "
1276 "Could not open pipe to %s", checkStr.Data()));
1281 if (!fgets(buffer, 100, pipe))
1283 Log("SHUTTLE", "Process - Error: ps did not return anything");
1284 gSystem->ClosePipe(pipe);
1287 gSystem->ClosePipe(pipe);
1289 //Log("SHUTTLE", Form("ps returned %s", buffer));
1292 if ((sscanf(buffer, "%d\n", &mem) != 1) || !mem)
1294 Log("SHUTTLE", "Process - Error: Could not parse output of ps");
1301 //______________________________________________________________________________________________
1302 Bool_t AliShuttle::Process(AliShuttleLogbookEntry* entry)
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
1311 if (!entry) return kFALSE;
1313 fLogbookEntry = entry;
1315 Log("SHUTTLE", Form("\t\t\t^*^*^*^*^*^*^*^*^*^*^*^* run %d: START ^*^*^*^*^*^*^*^*^*^*^*^*",
1320 // Send the information to ML
1321 SendMLRunInfo("Processing");
1323 if (fLogbookEntry->IsDone())
1325 Log("SHUTTLE","Process - Shuttle is already DONE. Updating logbook");
1326 UpdateShuttleLogbook("shuttle_done");
1331 // read test mode if flag is set
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);
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(' ');
1349 TObjString* tmpStr = dynamic_cast<TObjString*> (token->First());
1352 Int_t testMode = tmpStr->String().Atoi();
1355 Log("SHUTTLE", Form("Process - Enabling test mode %d", testMode));
1356 SetTestMode((TestMode) testMode);
1364 fLogbookEntry->Print("all");
1367 Bool_t hasError = kFALSE;
1369 // Set the CDB and Reference folders according to the year
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());
1376 fgkMainRefStorage.Form("alien://folder=%s%d/Reference?user=alidaq?cacheFold=/tmp/OCDBCache",
1377 fConfig->GetAlienPath(), GetCurrentYear());
1380 fgkMainCDB.Form("alien://folder=%s%d/DATest/OCDB?user=alidaq?cacheFold=/tmp/OCDBCache",
1381 fConfig->GetAlienPath(), GetCurrentYear());
1383 fgkMainRefStorage.Form("alien://folder=%s%d/DATest/Reference?user=alidaq?cacheFold=/tmp/OCDBCache",
1384 fConfig->GetAlienPath(), GetCurrentYear());
1387 AliDebug(2,Form("Main CDB storage = %s",fgkMainCDB.Data()));
1388 AliDebug(2,Form("Main Reference storage = %s",fgkMainRefStorage.Data()));
1390 // Loop on detectors in the configuration
1391 TIter iter(fConfig->GetDetectors());
1392 TObjString* aDetector = 0;
1394 Bool_t first = kTRUE;
1396 while ((aDetector = (TObjString*) iter.Next()))
1398 fCurrentDetector = aDetector->String();
1400 if (ContinueProcessing() == kFALSE)
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());
1413 Log("SHUTTLE", Form("\t\t\t****** run %d - %s: START ******",
1414 GetCurrentRun(), aDetector->GetName()));
1416 for(Int_t iSys=0;iSys<4;iSys++) fFXSCalled[iSys]=kFALSE;
1418 Int_t initialMem = GetMem(getpid());
1419 Log("SHUTTLE", Form("Memory consumption before forking is %d", initialMem));
1421 Log(fCurrentDetector.Data(), "Process - Starting processing");
1427 Log("SHUTTLE", "Process - ERROR: Forking failed");
1432 Log("SHUTTLE", Form("Process - In parent process of %d - %s: Starting monitoring",
1433 GetCurrentRun(), aDetector->GetName()));
1435 Long_t begin = time(0);
1437 int status; // to be used with waitpid, on purpose an int (not Int_t)!
1438 while (waitpid(pid, &status, WNOHANG) == 0)
1440 Long_t expiredTime = time(0) - begin;
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)
1448 AliShuttleStatus *currentStatus = ReadShuttleStatus();
1449 AliShuttleStatus::Status newStatus = AliShuttleStatus::kInvalid;
1451 if (currentStatus->GetStatus() == AliShuttleStatus::kDCSStarted)
1453 // in case the pp goes in TimeOut while retrieving the DCS DPs
1454 // set status to kDCSError
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;
1460 else if (currentStatus->GetStatus() <= AliShuttleStatus::kPPDone)
1462 // in case pp not yet done set status to kPPTimeOut
1464 logMsg.Form("Process - Process of %s timed out. Run time: %ld seconds. Killing...",
1465 fCurrentDetector.Data(), expiredTime);
1466 newStatus = AliShuttleStatus::kPPTimeOut;
1468 else if (currentStatus->GetStatus() == AliShuttleStatus::kStoreStarted)
1470 // in case the pp goes in TimeOut while storing the objects in the OCDB
1471 // set status to kStoreError
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;
1479 // in other cases don't change the status
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);
1485 Log("SHUTTLE", logMsg);
1486 Log(fCurrentDetector, logMsg);
1490 if (newStatus != AliShuttleStatus::kInvalid)
1491 UpdateShuttleStatus(newStatus);
1494 gSystem->Sleep(1000);
1498 gSystem->Sleep(1000);
1500 Int_t mem = GetMem(pid);
1509 if (expiredTime % 60 == 0)
1511 Log("SHUTTLE", Form("Process - %s: Checking process. "
1512 "Run time: %ld seconds - Memory consumption: %d KB",
1513 fCurrentDetector.Data(), expiredTime, mem));
1517 if (mem > fConfig->GetPPMaxMem())
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);
1528 UpdateShuttleStatus(AliShuttleStatus::kPPOutOfMemory);
1531 gSystem->Sleep(1000);
1536 Log("SHUTTLE", Form("Process - In parent process of %d - %s: Client has terminated.",
1537 GetCurrentRun(), aDetector->GetName()));
1539 if (WIFEXITED(status))
1541 Int_t returnCode = WEXITSTATUS(status);
1543 Log("SHUTTLE", Form("Process - %s: the return code is %d", fCurrentDetector.Data(),
1546 if (returnCode == 0) hasError = kTRUE;
1552 Log("SHUTTLE", Form("Process - In child process of %d - %s", GetCurrentRun(),
1553 aDetector->GetName()));
1555 Log("SHUTTLE", Form("Process - Redirecting output to %s log",fCurrentDetector.Data()));
1557 if ((freopen(GetLogFileName(fCurrentDetector), "a", stdout)) == 0)
1559 Log("SHUTTLE", "Process - Could not freopen stdout");
1563 fOutputRedirected = kTRUE;
1564 if ((dup2(fileno(stdout), fileno(stderr))) < 0)
1565 Log("SHUTTLE", "Process - Could not redirect stderr");
1569 Log("SHUTTLE", "Executing TGrid::Connect");
1570 TGrid::Connect("alien://");
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());
1577 Int_t result = gSystem->GetPathInfo(tmpDir.Data(), 0, (Long64_t*) 0, 0, 0);
1578 if (!result) // temp dir already exists!
1580 Log(fCurrentDetector.Data(),
1581 Form("Process - %s dir already exists! Removing...", tmpDir.Data()));
1582 gSystem->Exec(Form("rm -rf %s",tmpDir.Data()));
1585 if (gSystem->mkdir(tmpDir.Data(), 1))
1587 Log(fCurrentDetector.Data(), "Process - could not make temp directory!!");
1591 if (!gSystem->ChangeDirectory(tmpDir.Data()))
1593 Log(fCurrentDetector.Data(), "Process - could not change directory!!");
1597 Int_t success = ProcessCurrentDetector();
1599 gSystem->ChangeDirectory(wd.Data());
1601 if (success == 1) // Preprocessor finished successfully!
1603 // remove temporary folder or DCS map
1604 if (!fConfig->KeepTempFolder())
1606 gSystem->Exec(Form("rm -rf %s",tmpDir.Data()));
1607 } else if (!fConfig->KeepDCSMap())
1609 gSystem->Exec(Form("rm -f %s/DCSMap.root",tmpDir.Data()));
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()));
1617 // Transfer the data from local storage to main storage (Grid)
1618 if (StoreOCDB() == kFALSE)
1621 else if (success == 0)
1624 Form("\t\t\t****** run %d - %s: ERROR ******",
1625 GetCurrentRun(), aDetector->GetName()));
1628 for (UInt_t iSys=0; iSys<4; iSys++)
1630 if (fFXSCalled[iSys]) fFXSlist[iSys].Clear();
1633 Log("SHUTTLE", Form("Process - Client process of %d - %s is exiting now with %d.",
1634 GetCurrentRun(), aDetector->GetName(), success));
1636 // the client exits here
1637 gSystem->Exit(success);
1639 AliError("We should never get here!!!");
1643 Log("SHUTTLE", Form("\t\t\t^*^*^*^*^*^*^*^*^*^*^*^* run %d: FINISH ^*^*^*^*^*^*^*^*^*^*^*^*",
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!",
1654 return hasError == kFALSE;
1657 AliShuttleLogbookEntry* checkEntry = dynamic_cast<AliShuttleLogbookEntry*>
1658 (checkEntryArray.At(0));
1662 if (checkEntry->IsDone())
1664 Log("SHUTTLE","Process - Shuttle is DONE. Updating logbook");
1665 UpdateShuttleLogbook("shuttle_done");
1669 for (UInt_t iDet=0; iDet<NDetectors(); iDet++)
1671 if (checkEntry->GetDetectorStatus(iDet) == AliShuttleLogbookEntry::kUnprocessed)
1673 AliDebug(2, Form("Run %d: setting %s as \"not first time unprocessed\"",
1674 checkEntry->GetRun(), GetDetName(iDet)));
1675 fFirstUnprocessed[iDet] = kFALSE;
1678 SendMLRunInfo("Pending");
1684 return hasError == kFALSE;
1687 //______________________________________________________________________________________________
1688 Int_t AliShuttle::ProcessCurrentDetector()
1691 // Makes data retrieval just for a specific detector (fCurrentDetector).
1692 // Threre should be a configuration for this detector.
1694 Log("SHUTTLE", Form("ProcessCurrentDetector - Retrieving values for %s, run %d",
1695 fCurrentDetector.Data(), GetCurrentRun()));
1697 TString wd = gSystem->WorkingDirectory();
1699 if (!CleanReferenceStorage(fCurrentDetector.Data()))
1702 gSystem->ChangeDirectory(wd.Data());
1704 // call preprocessor
1705 AliPreprocessor* aPreprocessor =
1706 dynamic_cast<AliPreprocessor*> (fPreprocessorMap.GetValue(fCurrentDetector));
1708 // check if the preprocessor wants to process this run type
1709 if (aPreprocessor->ProcessRunType() == kFALSE)
1711 UpdateShuttleStatus(AliShuttleStatus::kSkipped);
1712 UpdateShuttleLogbook(fCurrentDetector, "DONE");
1713 if (!UpdateTableSkippedCase(fCurrentDetector.Data()))
1715 AliError(Form("Could not update FXS tables for run %d !", GetCurrentRun()));
1717 Log(fCurrentDetector, Form("ProcessCurrentDetector - %s preprocessor is not interested in this run type", fCurrentDetector.Data()));
1722 // checking if OCDB is reachable
1723 AliCDBEntry* testEntry = GetFromOCDB("SHUTTLE","GRP/CTP/DummyConfig");
1725 // OCDB is not accessible, going in OCDBError for current detector
1726 AliError("OCDB Test entry not accessible");
1727 UpdateShuttleStatus(AliShuttleStatus::kOCDBError);
1731 TMap* dcsMap = new TMap();
1733 aPreprocessor->Initialize(GetCurrentRun(), GetCurrentStartTime(), GetCurrentEndTime());
1735 Bool_t processDCS = aPreprocessor->ProcessDCS();
1739 Log(fCurrentDetector, "ProcessCurrentDetector -"
1740 " The preprocessor requested to skip the retrieval of DCS values");
1742 else if (fTestMode & kSkipDCS)
1744 Log(fCurrentDetector, "ProcessCurrentDetector - In TESTMODE: Skipping DCS processing");
1746 else if (fTestMode & kErrorDCS)
1748 Log(fCurrentDetector, "ProcessCurrentDetector - In TESTMODE: Simulating DCS error");
1749 UpdateShuttleStatus(AliShuttleStatus::kDCSStarted);
1750 UpdateShuttleStatus(AliShuttleStatus::kDCSError);
1755 UpdateShuttleStatus(AliShuttleStatus::kDCSStarted);
1757 // Query DCS archive
1758 Int_t nServers = fConfig->GetNServers(fCurrentDetector);
1760 for (int iServ=0; iServ<nServers; iServ++)
1763 TString host(fConfig->GetDCSHost(fCurrentDetector, iServ));
1764 Int_t port = fConfig->GetDCSPort(fCurrentDetector, iServ);
1765 Int_t multiSplit = fConfig->GetMultiSplit(fCurrentDetector, iServ);
1767 Log(fCurrentDetector, Form("ProcessCurrentDetector -"
1768 " Querying DCS Amanda server %s:%d (%d of %d)",
1769 host.Data(), port, iServ+1, nServers));
1774 if (fConfig->GetDCSAliases(fCurrentDetector, iServ)->GetEntries() > 0)
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);
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);
1788 if (!SendMail(kDCSEMail))
1789 Log("SHUTTLE", Form("ProcessCurrentDetector - "
1790 "Could not send mail to DCS experts!"));
1797 if (fConfig->GetDCSDataPoints(fCurrentDetector, iServ)->GetEntries() > 0)
1799 Log(fCurrentDetector, Form("Querying %d DCS data points", fConfig->GetDCSDataPoints(fCurrentDetector, iServ)->GetEntries()));
1800 dpMap = GetValueSet(host, port,
1801 fConfig->GetDCSDataPoints(fCurrentDetector, iServ),
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);
1811 if (!SendMail(kDCSEMail))
1812 Log("SHUTTLE", Form("ProcessCurrentDetector - "
1813 "Could not send mail to DCS experts!"));
1815 if (aliasMap) delete aliasMap;
1821 // merge aliasMap and dpMap into dcsMap
1823 TIter iter(aliasMap);
1824 TObjString* key = 0;
1825 while ((key = (TObjString*) iter.Next()))
1826 dcsMap->Add(key, aliasMap->GetValue(key->String()));
1828 aliasMap->SetOwner(kFALSE);
1834 TObjString* key = 0;
1835 while ((key = (TObjString*) iter.Next()))
1836 dcsMap->Add(key, dpMap->GetValue(key->String()));
1838 dpMap->SetOwner(kFALSE);
1844 // save map into file, to help debugging in case of preprocessor error
1845 TFile* f = TFile::Open("DCSMap.root","recreate");
1847 dcsMap->Write("DCSMap", TObject::kSingleKey);
1851 // DCS Archive DB processing successful. Call Preprocessor!
1852 UpdateShuttleStatus(AliShuttleStatus::kPPStarted);
1854 fFXSError = -1; // this variable is kTRUE after ::Process if an FXS error occured
1856 UInt_t returnValue = aPreprocessor->Process(dcsMap);
1858 if (fFXSError!=-1) {
1859 UpdateShuttleStatus(AliShuttleStatus::kFXSError);
1860 SendMail(kFXSEMail, fFXSError);
1861 dcsMap->DeleteAll();
1866 if (returnValue > 0) // Preprocessor error!
1868 Log(fCurrentDetector, Form("ProcessCurrentDetector - "
1869 "Preprocessor failed. Process returned %d.", returnValue));
1870 UpdateShuttleStatus(AliShuttleStatus::kPPError);
1871 dcsMap->DeleteAll();
1877 UpdateShuttleStatus(AliShuttleStatus::kPPDone);
1878 Log(fCurrentDetector, Form("ProcessCurrentDetector - %s preprocessor returned success",
1879 fCurrentDetector.Data()));
1881 dcsMap->DeleteAll();
1887 //______________________________________________________________________________________________
1888 void AliShuttle::CountOpenRuns()
1890 // Query DAQ's Shuttle logbook and sends the number of open runs to ML
1894 // check connection, in case connect
1899 sqlQuery = Form("select count(*) from %s where shuttle_done=0", fConfig->GetShuttlelbTable());
1901 TSQLResult* aResult = fServer[4]->Query(sqlQuery);
1903 AliError(Form("Can't execute query <%s>!", sqlQuery.Data()));
1907 AliDebug(2,Form("Query = %s", sqlQuery.Data()));
1909 if (aResult->GetRowCount() == 0) {
1910 AliError(Form("No result for query %s received", sqlQuery.Data()));
1914 if (aResult->GetFieldCount() != 1) {
1915 AliError(Form("Invalid field count for query %s received", sqlQuery.Data()));
1919 TSQLRow* aRow = aResult->Next();
1921 AliError(Form("Could not receive result of query %s", sqlQuery.Data()));
1925 TString result(aRow->GetField(0), aRow->GetFieldLength(0));
1926 Int_t count = result.Atoi();
1928 Log("SHUTTLE", Form("%d unprocessed runs", count));
1933 TMonaLisaValue mlStatus("SHUTTLE_openruns", count);
1936 mlList.Add(&mlStatus);
1938 fMonaLisa->SendParameters(&mlList, "__PROCESSINGINFO__");
1941 //______________________________________________________________________________________________
1942 Bool_t AliShuttle::QueryShuttleLogbook(const char* whereClause,
1945 // Query DAQ's Shuttle logbook and fills detector status object.
1946 // Call QueryRunParameters to query DAQ logbook for run parameters.
1949 entries.SetOwner(1);
1951 // check connection, in case connect
1952 if (!Connect(4)) return kFALSE;
1955 sqlQuery = Form("select * from %s %s order by run", fConfig->GetShuttlelbTable(), whereClause);
1957 TSQLResult* aResult = fServer[4]->Query(sqlQuery);
1959 AliError(Form("Can't execute query <%s>!", sqlQuery.Data()));
1963 AliDebug(2,Form("Query = %s", sqlQuery.Data()));
1965 if(aResult->GetRowCount() == 0) {
1966 Log("SHUTTLE", "No entries in Shuttle Logbook match request");
1971 // TODO Check field count!
1972 const UInt_t nCols = 27;
1973 if (aResult->GetFieldCount() != (Int_t) nCols) {
1974 Log("SHUTTLE", "Invalid SQL result field number!");
1980 while ((aRow = aResult->Next())) {
1981 TString runString(aRow->GetField(0), aRow->GetFieldLength(0));
1982 Int_t run = runString.Atoi();
1984 AliShuttleLogbookEntry *entry = QueryRunParameters(run);
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);
1993 // loop on detectors
1994 for(UInt_t ii = 0; ii < nCols; ii++)
1995 entry->SetDetectorStatus(aResult->GetFieldName(ii), aRow->GetField(ii));
1997 entries.AddLast(entry);
2005 //______________________________________________________________________________________________
2006 AliShuttleLogbookEntry* AliShuttle::QueryRunParameters(Int_t run)
2009 // Retrieve run parameters written in the DAQ logbook and sets them into AliShuttleLogbookEntry object
2012 // check connection, in case connect
2017 sqlQuery.Form("select * from %s where run=%d", fConfig->GetDAQlbTable(), run);
2019 TSQLResult* aResult = fServer[4]->Query(sqlQuery);
2021 Log("SHUTTLE", Form("Can't execute query <%s>!", sqlQuery.Data()));
2025 if (aResult->GetRowCount() == 0) {
2026 Log("SHUTTLE", Form("QueryRunParameters - No entry in DAQ Logbook for run %d. Skipping", run));
2031 if (aResult->GetRowCount() > 1) {
2032 Log("SHUTTLE", Form("QueryRunParameters - UNEXPECTED: "
2033 "more than one entry in DAQ Logbook for run %d!", run));
2038 TSQLRow* aRow = aResult->Next();
2041 Log("SHUTTLE", Form("QueryRunParameters - Could not retrieve row for run %d. Skipping", run));
2046 AliShuttleLogbookEntry* entry = new AliShuttleLogbookEntry(run);
2048 for (Int_t ii = 0; ii < aResult->GetFieldCount(); ii++)
2049 entry->SetRunParameter(aResult->GetFieldName(ii), aRow->GetField(ii));
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();
2063 UInt_t now = time(0);
2064 Int_t dcsDelay = fConfig->GetDCSDelay()+fConfig->GetDCSQueryOffset();
2066 Bool_t skip = kFALSE;
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));
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()));
2082 if (runType == "PHYSICS") {
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")));
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")));
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")));
2102 Log("SHUTTLE", Form("QueryRunParameters - Invalid parameters for run %d: startTime equal to endTime: %d %d - Skipping!", run, startTime, endTime));
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));
2113 Log("SHUTTLE", Form("Marking SHUTTLE skipped for run %d", run));
2114 fLogbookEntry = entry;
2115 if (!UpdateShuttleLogbook("shuttle_skipped"))
2117 AliError(Form("Could not update logbook for run %d !", run));
2119 if (!UpdateTableSkippedCase("ALL"))
2121 AliError(Form("Could not update FXS tables for run %d !", run));
2130 //______________________________________________________________________________________________
2131 TMap* AliShuttle::GetValueSet(const char* host, Int_t port, const TSeqCollection* entries,
2132 DCSType type, Int_t multiSplit)
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
2140 AliDCSClient client(host, port, fTimeout, fRetries, multiSplit);
2143 UInt_t startQuery = GetStartTimeDCSQuery();
2144 UInt_t endQuery = GetEndTimeDCSQuery();
2145 if (fCurrentDetector == "GRP" && (endQuery - startQuery) <= 120) { // enlarging DCS query for GRP when a run is shorter than 2 minutes (i.e. the time of forced archival of GRP DPs)
2146 Log(fCurrentDetector.Data(), Form("GetValueSet: run lasting less than 120 seconds, enlarging DCS window for DPs retrival to 130 s"));
2147 startQuery = endQuery - 130; // we add 130 s to be sure that there is something (even if the archival is forced after 120 s)
2152 //result = client.GetAliasValues(entries, GetCurrentStartTime()-offset,
2153 // GetCurrentEndTime()+offset);
2155 result = client.GetAliasValues(entries, startQuery, endQuery);
2157 else if (type == kDP)
2159 //result = client.GetDPValues(entries, GetCurrentStartTime()-offset,
2160 // GetCurrentEndTime()+offset);
2161 result = client.GetDPValues(entries, startQuery, endQuery);
2166 Log(fCurrentDetector.Data(), Form("GetValueSet - Can't get entries! Reason: %s",
2167 client.GetErrorString(client.GetResultErrorCode())));
2168 if (client.GetResultErrorCode() == AliDCSClient::fgkServerError)
2169 Log(fCurrentDetector.Data(), Form("GetValueSet - Server error code: %s",
2170 client.GetServerError().Data()));
2178 //______________________________________________________________________________________________
2179 const char* AliShuttle::GetFile(Int_t system, const char* detector,
2180 const char* id, const char* source)
2182 // Get calibration file from file exchange servers
2183 // First queris the FXS database for the file name, using the run, detector, id and source info
2184 // then calls RetrieveFile(filename) for actual copy to local disk
2185 // run: current run being processed (given by Logbook entry fLogbookEntry)
2186 // detector: the Preprocessor name
2187 // id: provided as a parameter by the Preprocessor
2188 // source: provided by the Preprocessor through GetFileSources function
2190 // check if test mode should simulate a FXS error
2191 if (fTestMode & kErrorFXSFiles)
2193 Log(detector, Form("GetFile - In TESTMODE - Simulating error while connecting to %s FXS", GetSystemName(system)));
2197 // check connection, in case connect
2198 if (!Connect(system))
2200 Log(detector, Form("GetFile - Couldn't connect to %s FXS database", GetSystemName(system)));
2205 // Query preparation
2206 TString sourceName(source);
2208 TString sqlQueryStart = Form("select filePath,size,fileChecksum from %s where",
2209 fConfig->GetFXSdbTable(system));
2210 TString whereClause = Form("run=%d and detector=\"%s\" and fileId=\"%s\"",
2211 GetCurrentRun(), detector, id);
2213 if (system == kDAQ || system == kDQM)
2215 whereClause += Form(" and DAQsource=\"%s\"", source);
2217 else if (system == kDCS)
2221 else if (system == kHLT)
2223 whereClause += Form(" and DDLnumbers=\"%s\"", source);
2226 TString sqlQuery = Form("%s %s", sqlQueryStart.Data(), whereClause.Data());
2228 AliDebug(2, Form("SQL query: \n%s",sqlQuery.Data()));
2231 TSQLResult* aResult = 0;
2232 aResult = dynamic_cast<TSQLResult*> (fServer[system]->Query(sqlQuery));
2234 Log(detector, Form("GetFile - Can't execute SQL query to %s database for: id = %s, source = %s",
2235 GetSystemName(system), id, sourceName.Data()));
2240 if (aResult->GetRowCount() == 0)
2243 Form("GetFile - No entry in %s FXS db for: id = %s, source = %s",
2244 GetSystemName(system), id, sourceName.Data()));
2249 if (aResult->GetRowCount() > 1) {
2251 Form("GetFile - More than one entry in %s FXS db for: id = %s, source = %s",
2252 GetSystemName(system), id, sourceName.Data()));
2258 if (aResult->GetFieldCount() != nFields) {
2260 Form("GetFileName - Wrong field count in %s FXS db for: id = %s, source = %s",
2261 GetSystemName(system), id, sourceName.Data()));
2267 TSQLRow* aRow = dynamic_cast<TSQLRow*> (aResult->Next());
2270 Log(detector, Form("GetFile - Empty set result in %s FXS db from query: id = %s, source = %s",
2271 GetSystemName(system), id, sourceName.Data()));
2277 TString filePath(aRow->GetField(0), aRow->GetFieldLength(0));
2278 TString fileSize(aRow->GetField(1), aRow->GetFieldLength(1));
2279 TString fileChecksum(aRow->GetField(2), aRow->GetFieldLength(2));
2284 AliDebug(2, Form("filePath = %s; size = %s, fileChecksum = %s",
2285 filePath.Data(), fileSize.Data(), fileChecksum.Data()));
2287 // retrieved file is renamed to make it unique
2288 Int_t dir_lev1 = GetCurrentRun()/10000;
2289 TString localFileName = Form("%s/%d/%d/%s_process/%s_%s_%d_%s_%s.shuttle",
2290 GetShuttleTempDir(), dir_lev1, GetCurrentRun(), detector,
2291 GetSystemName(system), detector, GetCurrentRun(),
2292 id, sourceName.Data());
2293 Log("SHUTTLE",Form("file from FXS = %s",localFileName.Data()));
2296 // file retrieval from FXS
2297 UInt_t nRetries = 0;
2298 UInt_t maxRetries = 3;
2299 Bool_t result = kFALSE;
2301 // copy!! if successful TSystem::Exec returns 0
2302 while (nRetries++ < maxRetries) {
2303 AliDebug(2, Form("Trying to copy file. Retry # %d", nRetries));
2304 result = RetrieveFile(system, filePath.Data(), localFileName.Data());
2307 Log(detector, Form("GetFile - Copy of file %s from %s FXS failed",
2308 filePath.Data(), GetSystemName(system)));
2312 if (fileSize.Length()>0)
2314 // compare filesize of local file with the one stored in the FXS DB
2316 Int_t sizeComp = gSystem->GetPathInfo(localFileName.Data(), 0, &size, 0, 0);
2318 if (sizeComp != 0 || size != fileSize.Atoi())
2320 Log(detector, Form("GetFile - size of file %s does not match with local copy!",
2327 Log(fCurrentDetector, Form("GetFile - size of file %s not set in %s database, skipping comparison",
2328 filePath.Data(), GetSystemName(system)));
2331 if (fileChecksum.Length()>0)
2333 // compare md5sum of local file with the one stored in the FXS DB
2334 if(fileChecksum.Contains(' ')) fileChecksum.Resize(fileChecksum.First(' '));
2335 Int_t md5Comp = gSystem->Exec(Form("md5sum %s |grep %s > /dev/null 2> /dev/null",
2336 localFileName.Data(), fileChecksum.Data()));
2340 Log(detector, Form("GetFile - md5sum of file %s does not match with local copy!",
2346 Log(fCurrentDetector, Form("GetFile - md5sum of file %s not set in %s database, skipping comparison",
2347 filePath.Data(), GetSystemName(system)));
2358 fFXSCalled[system]=kTRUE;
2359 TObjString *fileParams = new TObjString(Form("%s#!?!#%s", id, sourceName.Data()));
2360 fFXSlist[system].Add(fileParams);
2362 static TString staticLocalFileName;
2363 staticLocalFileName.Form("%s", localFileName.Data());
2365 Log(fCurrentDetector, Form("GetFile - Retrieved file with id %s and "
2366 "source %s from %s to %s", id, source,
2367 GetSystemName(system), localFileName.Data()));
2369 return staticLocalFileName.Data();
2372 //______________________________________________________________________________________________
2373 Bool_t AliShuttle::RetrieveFile(UInt_t system, const char* fxsFileName, const char* localFileName)
2376 // Copies file from FXS to local Shuttle machine
2379 // check temp directory: trying to cd to temp; if it does not exist, create it
2380 AliDebug(2, Form("Copy file %s from %s FXS into %s",
2381 GetSystemName(system), fxsFileName, localFileName));
2383 TString tmpDir(localFileName);
2385 tmpDir = tmpDir(0,tmpDir.Last('/'));
2387 Int_t noDir = gSystem->GetPathInfo(tmpDir.Data(), 0, (Long64_t*) 0, 0, 0);
2388 if (noDir) // temp dir does not exists!
2390 if (gSystem->mkdir(tmpDir.Data(), 1))
2392 Log(fCurrentDetector.Data(), "RetrieveFile - could not make temp directory!!");
2397 TString command = Form("scp -oPort=%d -2 %s@%s:%s/%s %s",
2398 fConfig->GetFXSPort(system),
2399 fConfig->GetFXSUser(system),
2400 fConfig->GetFXSHost(system),
2401 fConfig->GetFXSBaseFolder(system),
2405 AliDebug(2, Form("%s",command.Data()));
2407 Bool_t result = (gSystem->Exec(command.Data()) == 0);
2412 //______________________________________________________________________________________________
2413 TList* AliShuttle::GetFileSources(Int_t system, const char* detector, const char* id)
2416 // Get sources producing the condition file Id from file exchange servers
2417 // if id is NULL all sources are returned (distinct)
2422 Log(detector, Form("GetFileSources - Querying %s FXS for files with id %s produced by %s", GetSystemName(system), id, detector));
2424 Log(detector, Form("GetFileSources - Querying %s FXS for files produced by %s", GetSystemName(system), detector));
2427 // check if test mode should simulate a FXS error
2428 if (fTestMode & kErrorFXSSources)
2430 Log(detector, Form("GetFileSources - In TESTMODE - Simulating error while connecting to %s FXS", GetSystemName(system)));
2436 Log(detector, "GetFileSources - WARNING: DCS system has only one source of data!");
2437 TList *list = new TList();
2439 list->Add(new TObjString(" "));
2443 // check connection, in case connect
2444 if (!Connect(system))
2446 Log(detector, Form("GetFileSources - Couldn't connect to %s FXS database", GetSystemName(system)));
2451 TString sourceName = "";
2452 if (system == kDAQ || system == kDQM)
2454 sourceName = "DAQsource";
2455 } else if (system == kHLT)
2457 sourceName = "DDLnumbers";
2460 TString sqlQueryStart = Form("select distinct %s from %s where", sourceName.Data(), fConfig->GetFXSdbTable(system));
2461 TString whereClause = Form("run=%d and detector=\"%s\"",
2462 GetCurrentRun(), detector);
2464 whereClause += Form(" and fileId=\"%s\"", id);
2465 TString sqlQuery = Form("%s %s", sqlQueryStart.Data(), whereClause.Data());
2467 AliDebug(2, Form("SQL query: \n%s",sqlQuery.Data()));
2470 TSQLResult* aResult;
2471 aResult = fServer[system]->Query(sqlQuery);
2473 Log(detector, Form("GetFileSources - Can't execute SQL query to %s database for id: %s",
2474 GetSystemName(system), id));
2479 TList *list = new TList();
2482 if (aResult->GetRowCount() == 0)
2485 Form("GetFileSources - No entry in %s FXS table for id: %s", GetSystemName(system), id));
2490 Log(detector, Form("GetFileSources - Found %d sources", aResult->GetRowCount()));
2493 while ((aRow = aResult->Next()))
2496 TString source(aRow->GetField(0), aRow->GetFieldLength(0));
2497 AliDebug(2, Form("%s = %s", sourceName.Data(), source.Data()));
2498 list->Add(new TObjString(source));
2507 //______________________________________________________________________________________________
2508 TList* AliShuttle::GetFileIDs(Int_t system, const char* detector, const char* source)
2511 // Get all ids of condition files produced by a given source from file exchange servers
2514 Log(detector, Form("GetFileIDs - Retrieving ids with source %s with %s", source, GetSystemName(system)));
2516 // check if test mode should simulate a FXS error
2517 if (fTestMode & kErrorFXSSources)
2519 Log(detector, Form("GetFileIDs - In TESTMODE - Simulating error while connecting to %s FXS", GetSystemName(system)));
2523 // check connection, in case connect
2524 if (!Connect(system))
2526 Log(detector, Form("GetFileIDs - Couldn't connect to %s FXS database", GetSystemName(system)));
2530 TString sourceName = "";
2533 sourceName = "DAQsource";
2534 } else if (system == kHLT)
2536 sourceName = "DDLnumbers";
2539 TString sqlQueryStart = Form("select fileId from %s where", fConfig->GetFXSdbTable(system));
2540 TString whereClause = Form("run=%d and detector=\"%s\"",
2541 GetCurrentRun(), detector);
2542 if (sourceName.Length() > 0 && source)
2543 whereClause += Form(" and %s=\"%s\"", sourceName.Data(), source);
2544 TString sqlQuery = Form("%s %s", sqlQueryStart.Data(), whereClause.Data());
2546 AliDebug(2, Form("SQL query: \n%s",sqlQuery.Data()));
2549 TSQLResult* aResult;
2550 aResult = fServer[system]->Query(sqlQuery);
2552 Log(detector, Form("GetFileIDs - Can't execute SQL query to %s database for source: %s",
2553 GetSystemName(system), source));
2557 TList *list = new TList();
2560 if (aResult->GetRowCount() == 0)
2563 Form("GetFileIDs - No entry in %s FXS table for source: %s", GetSystemName(system), source));
2568 Log(detector, Form("GetFileIDs - Found %d ids", aResult->GetRowCount()));
2572 while ((aRow = aResult->Next()))
2575 TString id(aRow->GetField(0), aRow->GetFieldLength(0));
2576 AliDebug(2, Form("fileId = %s", id.Data()));
2577 list->Add(new TObjString(id));
2586 //______________________________________________________________________________________________
2587 Bool_t AliShuttle::Connect(Int_t system)
2589 // Connect to MySQL Server of the system's FXS MySQL databases
2590 // DAQ Logbook, Shuttle Logbook and DAQ FXS db are on the same host
2593 // check connection: if already connected return
2595 if(fServer[system] && fServer[system]->IsConnected()) {
2597 if (fServer[system]->PingVerify()==kTRUE){ // connection is still alive
2601 AliWarning(Form("Connection got lost to FXS database for %s. Closing and reconnecting.",
2602 AliShuttleInterface::GetSystemName(system)));
2603 fServer[system]->Close();
2604 delete fServer[system];
2605 fServer[system] = 0x0;
2609 TString dbHost, dbUser, dbPass, dbName;
2611 if (system < 4) // FXS db servers
2613 dbHost = Form("mysql://%s:%d", fConfig->GetFXSdbHost(system), fConfig->GetFXSdbPort(system));
2614 dbUser = fConfig->GetFXSdbUser(system);
2615 dbPass = fConfig->GetFXSdbPass(system);
2616 dbName = fConfig->GetFXSdbName(system);
2617 } else { // Run & Shuttle logbook servers
2618 // TODO Will the Shuttle logbook server be the same as the Run logbook server ???
2619 dbHost = Form("mysql://%s:%d", fConfig->GetDAQlbHost(), fConfig->GetDAQlbPort());
2620 dbUser = fConfig->GetDAQlbUser();
2621 dbPass = fConfig->GetDAQlbPass();
2622 dbName = fConfig->GetDAQlbDB();
2625 fServer[system] = TSQLServer::Connect(dbHost.Data(), dbUser.Data(), dbPass.Data());
2626 if (!fServer[system] || !fServer[system]->IsConnected()) {
2629 AliError(Form("Can't establish connection to FXS database for %s",
2630 AliShuttleInterface::GetSystemName(system)));
2632 AliError("Can't establish connection to Run logbook.");
2634 if(fServer[system]) delete fServer[system];
2639 TSQLResult* aResult=0;
2642 aResult = fServer[kDAQ]->GetTables(dbName.Data());
2645 aResult = fServer[kDCS]->GetTables(dbName.Data());
2648 aResult = fServer[kHLT]->GetTables(dbName.Data());
2651 aResult = fServer[kDQM]->GetTables(dbName.Data());
2654 aResult = fServer[4]->GetTables(dbName.Data());
2662 //______________________________________________________________________________________________
2663 Bool_t AliShuttle::UpdateTable()
2666 // Update FXS table filling time_processed field in all rows corresponding to current run and detector
2669 Bool_t result = kTRUE;
2671 for (UInt_t system=0; system<4; system++)
2673 if(!fFXSCalled[system]) continue;
2675 // check connection, in case connect
2676 if (!Connect(system))
2678 Log(fCurrentDetector, Form("UpdateTable - Couldn't connect to %s FXS database", GetSystemName(system)));
2683 TTimeStamp now; // now
2685 // Loop on FXS list entries
2686 TIter iter(&fFXSlist[system]);
2687 TObjString *aFXSentry=0;
2688 while ((aFXSentry = dynamic_cast<TObjString*> (iter.Next())))
2690 TString aFXSentrystr = aFXSentry->String();
2691 TObjArray *aFXSarray = aFXSentrystr.Tokenize("#!?!#");
2692 if (!aFXSarray || aFXSarray->GetEntries() != 2 )
2694 Log(fCurrentDetector, Form("UpdateTable - error updating %s FXS entry. Check string: <%s>",
2695 GetSystemName(system), aFXSentrystr.Data()));
2696 if(aFXSarray) delete aFXSarray;
2700 const char* fileId = ((TObjString*) aFXSarray->At(0))->GetName();
2701 const char* source = ((TObjString*) aFXSarray->At(1))->GetName();
2703 TString whereClause;
2704 if (system == kDAQ || system == kDQM)
2706 whereClause = Form("where run=%d and detector=\"%s\" and fileId=\"%s\" and DAQsource=\"%s\";",
2707 GetCurrentRun(), fCurrentDetector.Data(), fileId, source);
2709 else if (system == kDCS)
2711 whereClause = Form("where run=%d and detector=\"%s\" and fileId=\"%s\";",
2712 GetCurrentRun(), fCurrentDetector.Data(), fileId);
2714 else if (system == kHLT)
2716 whereClause = Form("where run=%d and detector=\"%s\" and fileId=\"%s\" and DDLnumbers=\"%s\";",
2717 GetCurrentRun(), fCurrentDetector.Data(), fileId, source);
2722 TString sqlQuery = Form("update %s set time_processed=%ld %s", fConfig->GetFXSdbTable(system),
2723 (ULong_t)now.GetSec(), whereClause.Data());
2725 AliDebug(2, Form("SQL query: \n%s",sqlQuery.Data()));
2728 TSQLResult* aResult;
2729 aResult = dynamic_cast<TSQLResult*> (fServer[system]->Query(sqlQuery));
2732 Log(fCurrentDetector, Form("UpdateTable - %s db: can't execute SQL query <%s>",
2733 GetSystemName(system), sqlQuery.Data()));
2744 //_______________________________________________________________________________
2745 Bool_t AliShuttle::UpdateTableSkippedCase(const char* detector)
2748 // Update FXS table filling time_processed field in all rows corresponding to current run and detector
2749 // if detector = "ALL" update all detectors
2752 Bool_t result = kTRUE;
2754 TString detName(detector);
2756 for (UInt_t system=0; system<4; system++)
2759 // check connection, in case connect
2760 if (!Connect(system))
2762 Log(fCurrentDetector, Form("UpdateTableSkippedCase - Couldn't connect to %s FXS database", GetSystemName(system)));
2767 TTimeStamp now; // now
2769 // Loop on FXS list entries
2770 TIter iter(&fFXSlist[system]);
2772 TString whereClause;
2773 if (detName == "ALL") whereClause = Form("where run=%d and time_processed IS NULL;",GetCurrentRun());
2774 else whereClause = Form("where run=%d and detector=\"%s\" and time_processed IS NULL;",GetCurrentRun(), detector);
2776 //Log("SHUTTLE",Form(" whereClause = %s ",whereClause.Data()));
2778 TString sqlQuery = Form("update %s set time_processed=%ld %s", fConfig->GetFXSdbTable(system),
2779 (ULong_t)now.GetSec(), whereClause.Data());
2781 AliDebug(2, Form("SQL query: \n%s",sqlQuery.Data()));
2784 TSQLResult* aResult;
2785 aResult = dynamic_cast<TSQLResult*> (fServer[system]->Query(sqlQuery));
2788 Log("SHUTTLE", Form("UpdateTableSkippedCase - %s db: can't execute SQL query <%s>",
2789 GetSystemName(system), sqlQuery.Data()));
2799 //______________________________________________________________________________________________
2800 Bool_t AliShuttle::UpdateTableFailCase()
2802 // Update FXS table filling time_processed field in all rows corresponding to current run and detector
2803 // this is called in case the preprocessor is declared failed for the current run, because
2804 // the fields are updated only in case of success
2806 Bool_t result = kTRUE;
2808 for (UInt_t system=0; system<4; system++)
2810 // check connection, in case connect
2811 if (!Connect(system))
2813 Log(fCurrentDetector, Form("UpdateTableFailCase - Couldn't connect to %s FXS database",
2814 GetSystemName(system)));
2819 TTimeStamp now; // now
2821 // Loop on FXS list entries
2823 TString whereClause = Form("where run=%d and detector=\"%s\";",
2824 GetCurrentRun(), fCurrentDetector.Data());
2827 TString sqlQuery = Form("update %s set time_processed=%ld %s", fConfig->GetFXSdbTable(system),
2828 (ULong_t)now.GetSec(), whereClause.Data());
2830 AliDebug(2, Form("SQL query: \n%s",sqlQuery.Data()));
2833 TSQLResult* aResult;
2834 aResult = dynamic_cast<TSQLResult*> (fServer[system]->Query(sqlQuery));
2837 Log(fCurrentDetector, Form("UpdateTableFailCase - %s db: can't execute SQL query <%s>",
2838 GetSystemName(system), sqlQuery.Data()));
2848 //______________________________________________________________________________________________
2849 Bool_t AliShuttle::UpdateShuttleLogbook(const char* detector, const char* status)
2852 // Update Shuttle logbook filling detector or shuttle_done column
2853 // ex. of usage: UpdateShuttleLogbook("PHOS", "DONE") or UpdateShuttleLogbook("shuttle_done")
2856 // check connection, in case connect
2858 Log("SHUTTLE", "UpdateShuttleLogbook - Couldn't connect to DAQ Logbook.");
2862 TString detName(detector);
2864 if (detName == "shuttle_done" || detName == "shuttle_skipped")
2866 setClause = "set shuttle_done=1";
2868 if (detName == "shuttle_done")
2870 if (TouchFile() != kTRUE)
2872 SendMLRunInfo("Pending");
2876 SendMLRunInfo("Done");
2879 SendMLRunInfo("Skipped");
2882 TString statusStr(status);
2883 if(statusStr.Contains("done", TString::kIgnoreCase) ||
2884 statusStr.Contains("failed", TString::kIgnoreCase)){
2885 setClause = Form("set %s=\"%s\"", detector, status);
2888 Form("UpdateShuttleLogbook - Invalid status <%s> for detector %s",
2894 TString whereClause = Form("where run=%d", GetCurrentRun());
2896 TString sqlQuery = Form("update %s %s %s",
2897 fConfig->GetShuttlelbTable(), setClause.Data(), whereClause.Data());
2899 AliDebug(2, Form("SQL query: \n%s",sqlQuery.Data()));
2902 TSQLResult* aResult;
2903 aResult = dynamic_cast<TSQLResult*> (fServer[4]->Query(sqlQuery));
2905 Log("SHUTTLE", Form("UpdateShuttleLogbook - Can't execute query <%s>", sqlQuery.Data()));
2913 //______________________________________________________________________________________________
2914 Int_t AliShuttle::GetCurrentRun() const
2917 // Get current run from logbook entry
2920 return fLogbookEntry ? fLogbookEntry->GetRun() : -1;
2923 //______________________________________________________________________________________________
2924 UInt_t AliShuttle::GetCurrentStartTime() const
2927 // get current start time
2930 return fLogbookEntry ? fLogbookEntry->GetStartTime() : 0;
2933 //______________________________________________________________________________________________
2934 UInt_t AliShuttle::GetCurrentEndTime() const
2937 // get current end time from logbook entry
2940 return fLogbookEntry ? fLogbookEntry->GetEndTime() : 0;
2942 //______________________________________________________________________________________________
2943 UInt_t AliShuttle::GetCurrentYear() const
2946 // Get current year from logbook entry
2949 if (!fLogbookEntry) return 0;
2951 TTimeStamp startTime(GetCurrentStartTime());
2952 TString year = Form("%d",startTime.GetDate());
2958 //______________________________________________________________________________________________
2959 const char* AliShuttle::GetLHCPeriod() const
2962 // Get current LHC period from logbook entry
2965 if (!fLogbookEntry) return 0;
2967 return fLogbookEntry->GetRunParameter("LHCperiod");
2970 //______________________________________________________________________________________________
2971 void AliShuttle::Log(const char* detector, const char* message, UInt_t level)
2974 // Fill log string with a message
2977 TString logRunDir = GetShuttleLogDir();
2978 if (GetCurrentRun() >=0) {
2979 Int_t logDir_lev1 = GetCurrentRun()/10000;
2980 logRunDir += Form("/%d/%d", logDir_lev1, GetCurrentRun());
2982 void* dir = gSystem->OpenDirectory(logRunDir.Data());
2984 if (gSystem->mkdir(logRunDir.Data(), kTRUE)) {
2985 AliError(Form("Can't open directory <%s>", GetShuttleLogDir()));
2990 gSystem->FreeDirectory(dir);
2993 TString toLog = Form("%s UTC (%d): %s - ", TTimeStamp(time(0)).AsString("s"), getpid(), detector);
2994 if (GetCurrentRun() >= 0)
2995 toLog += Form("run %d - ", GetCurrentRun());
2996 toLog += Form("%s", message);
2998 AliLog::Message(level, toLog, MODULENAME(), ClassName(), FUNCTIONNAME(), __FILE__, __LINE__);
3000 // if we redirect the log output already to the file, leave here
3001 if (fOutputRedirected && strcmp(detector, "SHUTTLE") != 0)
3004 TString fileName = GetLogFileName(detector);
3006 gSystem->ExpandPathName(fileName);
3009 logFile.open(fileName, ofstream::out | ofstream::app);
3011 if (!logFile.is_open()) {
3012 AliError(Form("Could not open file %s", fileName.Data()));
3016 logFile << toLog.Data() << "\n";
3021 //______________________________________________________________________________________________
3022 TString AliShuttle::GetLogFileName(const char* detector) const
3025 // returns the name of the log file for a given sub detector
3030 if (GetCurrentRun() >= 0)
3032 Int_t logDir_lev1 = GetCurrentRun()/10000;
3033 fileName.Form("%s/%d/%d/%s.log", GetShuttleLogDir(), logDir_lev1, GetCurrentRun(),
3036 fileName.Form("%s/%s.log", GetShuttleLogDir(), detector);
3042 //______________________________________________________________________________________________
3043 void AliShuttle::SendAlive()
3045 // sends alive message to ML
3047 TMonaLisaText mlStatus("SHUTTLE_status", "Alive");
3050 mlList.Add(&mlStatus);
3052 fMonaLisa->SendParameters(&mlList, "__PROCESSINGINFO__");
3055 //______________________________________________________________________________________________
3056 Bool_t AliShuttle::Collect(Int_t run)
3059 // Collects conditions data for all UNPROCESSED run written to DAQ LogBook in case of run = -1 (default)
3060 // If a dedicated run is given this run is processed
3062 // In operational mode, this is the Shuttle function triggered by the EOR signal.
3066 Log("SHUTTLE","Collect - Shuttle called. Collecting conditions data for unprocessed runs");
3068 Log("SHUTTLE", Form("Collect - Shuttle called. Collecting conditions data for run %d", run));
3070 SetLastAction("Starting");
3072 // create ML instance
3074 fMonaLisa = new TMonaLisaWriter(fConfig->GetMonitorHost(), fConfig->GetMonitorTable());
3078 TString whereClause("where shuttle_done=0");
3080 whereClause += Form(" and run=%d", run);
3082 TObjArray shuttleLogbookEntries;
3083 if (!QueryShuttleLogbook(whereClause, shuttleLogbookEntries))
3085 Log("SHUTTLE", "Collect - Can't retrieve entries from Shuttle logbook");
3089 if (shuttleLogbookEntries.GetEntries() == 0)
3092 Log("SHUTTLE","Collect - Found no UNPROCESSED runs in Shuttle logbook");
3094 Log("SHUTTLE", Form("Collect - Run %d is already DONE "
3095 "or it does not exist in Shuttle logbook", run));
3099 for (UInt_t iDet=0; iDet<NDetectors(); iDet++)
3100 fFirstUnprocessed[iDet] = kTRUE;
3104 // query Shuttle logbook for earlier runs, check if some detectors are unprocessed,
3105 // flag them into fFirstUnprocessed array
3106 TString whereClauseBis(Form("where shuttle_done=0 and run < %d", run));
3107 TObjArray tmpLogbookEntries;
3108 if (!QueryShuttleLogbook(whereClauseBis, tmpLogbookEntries))
3110 Log("SHUTTLE", "Collect - Can't retrieve entries from Shuttle logbook");
3114 TIter iter(&tmpLogbookEntries);
3115 AliShuttleLogbookEntry* anEntry = 0;
3116 while ((anEntry = dynamic_cast<AliShuttleLogbookEntry*> (iter.Next())))
3118 for (UInt_t iDet=0; iDet<NDetectors(); iDet++)
3120 if (anEntry->GetDetectorStatus(iDet) == AliShuttleLogbookEntry::kUnprocessed)
3122 AliDebug(2, Form("Run %d: setting %s as \"not first time unprocessed\"",
3123 anEntry->GetRun(), GetDetName(iDet)));
3124 fFirstUnprocessed[iDet] = kFALSE;
3132 if (!RetrieveConditionsData(shuttleLogbookEntries))
3134 Log("SHUTTLE", "Collect - Process of at least one run failed");
3139 Log("SHUTTLE", "Collect - Requested run(s) successfully processed");
3144 //______________________________________________________________________________________________
3145 Bool_t AliShuttle::RetrieveConditionsData(const TObjArray& dateEntries)
3148 // Retrieve conditions data for all runs that aren't processed yet
3151 Bool_t hasError = kFALSE;
3153 TIter iter(&dateEntries);
3154 AliShuttleLogbookEntry* anEntry;
3156 while ((anEntry = (AliShuttleLogbookEntry*) iter.Next())){
3157 if (!Process(anEntry)){
3161 // clean SHUTTLE temp directory
3162 //TString filename = Form("%s/*.shuttle", GetShuttleTempDir());
3163 //RemoveFile(filename.Data());
3166 return hasError == kFALSE;
3169 //______________________________________________________________________________________________
3170 ULong_t AliShuttle::GetTimeOfLastAction() const
3173 // Gets time of last action
3178 fMonitoringMutex->Lock();
3180 tmp = fLastActionTime;
3182 fMonitoringMutex->UnLock();
3187 //______________________________________________________________________________________________
3188 const TString AliShuttle::GetLastAction() const
3191 // returns a string description of the last action
3196 fMonitoringMutex->Lock();
3200 fMonitoringMutex->UnLock();
3205 //______________________________________________________________________________________________
3206 void AliShuttle::SetLastAction(const char* action)
3209 // updates the monitoring variables
3212 fMonitoringMutex->Lock();
3214 fLastAction = action;
3215 fLastActionTime = time(0);
3217 fMonitoringMutex->UnLock();
3220 //______________________________________________________________________________________________
3221 const char* AliShuttle::GetRunParameter(const char* param)
3224 // returns run parameter read from DAQ logbook
3227 if(!fLogbookEntry) {
3228 AliError("No logbook entry!");
3232 return fLogbookEntry->GetRunParameter(param);
3235 //______________________________________________________________________________________________
3236 AliCDBEntry* AliShuttle::GetFromOCDB(const char* detector, const AliCDBPath& path)
3239 // returns object from OCDB valid for current run
3242 if (fTestMode & kErrorOCDB)
3244 Log(detector, "GetFromOCDB - In TESTMODE - Simulating error with OCDB");
3248 AliCDBStorage *sto = AliCDBManager::Instance()->GetStorage(fgkMainCDB);
3251 Log(detector, "GetFromOCDB - Cannot activate main OCDB for query!");
3255 return dynamic_cast<AliCDBEntry*> (sto->Get(path, GetCurrentRun()));
3258 //______________________________________________________________________________________________
3259 Bool_t AliShuttle::SendMail(EMailTarget target, Int_t system)
3262 // sends a mail to the subdetector expert in case of preprocessor error
3265 if (fTestMode != kNone)
3268 if (!fConfig->SendMail())
3271 if (target == kDCSEMail || target == kFXSEMail) {
3272 if (!fFirstProcessing)
3276 Int_t runMode = (Int_t)fConfig->GetRunMode();
3278 if (runMode == 0) tmpStr = " Nightly Test:";
3279 else tmpStr = " Data Taking:";
3280 void* dir = gSystem->OpenDirectory(GetShuttleLogDir());
3283 if (gSystem->mkdir(GetShuttleLogDir(), kTRUE))
3285 Log("SHUTTLE", Form("SendMail - Can't open directory <%s>", GetShuttleLogDir()));
3290 gSystem->FreeDirectory(dir);
3293 // det experts in to
3295 TIter *iterExperts = 0;
3296 if (target == kDCSEMail) {
3297 iterExperts = new TIter(fConfig->GetAdmins(AliShuttleConfig::kAmanda));
3299 else if (target == kFXSEMail) {
3300 iterExperts = new TIter(fConfig->GetAdmins(system));
3303 TObjString *anExpert=0;
3304 while ((anExpert = (TObjString*) iterExperts->Next()))
3306 to += Form("%s,", anExpert->GetName());
3311 // add subdetector experts
3312 iterExperts = new TIter(fConfig->GetResponsibles(fCurrentDetector));
3313 TObjString *anExpert=0;
3314 while ((anExpert = (TObjString*) iterExperts->Next()))
3316 to += Form("%s,", anExpert->GetName());
3320 if (to.Length() > 0)
3321 to.Remove(to.Length()-1);
3322 AliDebug(2, Form("to: %s",to.Data()));
3325 Log("SHUTTLE", Form("List of %d responsibles not set!", (Int_t) target));
3329 // SHUTTLE responsibles in cc
3331 TIter iterAdmins(fConfig->GetAdmins(AliShuttleConfig::kGlobal));
3332 TObjString *anAdmin=0;
3333 while ((anAdmin = (TObjString*) iterAdmins.Next()))
3335 cc += Form("%s,", anAdmin->GetName());
3337 if (cc.Length() > 0)
3338 cc.Remove(cc.Length()-1);
3339 AliDebug(2, Form("cc: %s",to.Data()));
3342 TString bodyFileName;
3343 bodyFileName.Form("%s/mail.body", GetShuttleLogDir());
3344 gSystem->ExpandPathName(bodyFileName);
3347 mailBody.open(bodyFileName, ofstream::out);
3349 if (!mailBody.is_open())
3351 Log("SHUTTLE", Form("Could not open mail body file %s", bodyFileName.Data()));
3359 if (target == kDCSEMail){
3360 subject = Form("%s CRITICAL Retrieval of data points for %s FAILED in run %d !",
3361 tmpStr.Data(), fCurrentDetector.Data(), GetCurrentRun());
3362 AliDebug(2, Form("subject: %s", subject.Data()));
3364 body = Form("Dear DCS experts, \n\n");
3365 body += Form("SHUTTLE couldn\'t retrieve the data points for detector %s "
3366 "in run %d!!\n\n", fCurrentDetector.Data(), GetCurrentRun());
3368 else if (target == kFXSEMail){
3369 subject = Form("%s CRITICAL FXS communication for %s FAILED in run %d !",
3370 tmpStr.Data(), fCurrentDetector.Data(), GetCurrentRun());
3371 AliDebug(2, Form("subject: %s", subject.Data()));
3373 if (system == kDAQ) sys="DAQ";
3374 else if (system == kDCS) sys="DCS";
3375 else if (system == kHLT) sys="HLT";
3376 else if (system == kDQM) sys="DQM";
3378 body = Form("Dear %s FXS experts, \n\n",sys.Data());
3379 body += Form("SHUTTLE couldn\'t retrieve data from the FXS for detector %s "
3380 "in run %d!!\n\n", fCurrentDetector.Data(), GetCurrentRun());
3381 body += Form("The contacted server was:\nDB: %s\nFXS:%s\n\n", fConfig->GetFXSdbHost(system), fConfig->GetFXSHost(system));
3384 subject = Form("%s %s Shuttle preprocessor FAILED in run %d (run type = %s)!",
3385 tmpStr.Data(), fCurrentDetector.Data(), GetCurrentRun(), GetRunType());
3386 AliDebug(2, Form("subject: %s", subject.Data()));
3388 body = Form("Dear %s expert(s), \n\n", fCurrentDetector.Data());
3389 body += Form("SHUTTLE just detected that your preprocessor "
3390 "failed processing run %d (run type = %s)!!\n\n",
3391 GetCurrentRun(), GetRunType());
3394 body += Form("Please check %s status on the SHUTTLE monitoring page: \n\n",
3395 fCurrentDetector.Data());
3396 if (fConfig->GetRunMode() == AliShuttleConfig::kTest)
3398 body += Form("\thttp://pcalimonitor.cern.ch/shuttle.jsp?time=24 \n\n");
3400 body += Form("\thttp://pcalimonitor.cern.ch/shuttle.jsp?instance=PROD&time=24 \n\n");
3404 TString logFolder = "logs";
3405 if (fConfig->GetRunMode() == AliShuttleConfig::kProd)
3406 logFolder += "_PROD";
3409 body += Form("Find the %s log for the current run on \n\n"
3410 "\thttp://pcalishuttle02.cern.ch/%s/%d/%d/%s.log \n\n",
3411 fCurrentDetector.Data(), logFolder.Data(), GetCurrentRun()/10000,
3412 GetCurrentRun(), fCurrentDetector.Data());
3413 body += Form("The last 15 lines of %s log file are following:\n\n", fCurrentDetector.Data());
3415 AliDebug(2, Form("Body begin: %s", body.Data()));
3417 mailBody << body.Data();
3419 mailBody.open(bodyFileName, ofstream::out | ofstream::app);
3421 TString logFileName = Form("%s/%d/%d/%s.log", GetShuttleLogDir(),
3422 GetCurrentRun()/10000, GetCurrentRun(), fCurrentDetector.Data());
3423 TString tailCommand = Form("tail -n 15 %s >> %s", logFileName.Data(), bodyFileName.Data());
3424 if (gSystem->Exec(tailCommand.Data()))
3426 mailBody << Form("%s log file not found ...\n\n", fCurrentDetector.Data());
3429 TString endBody = Form("------------------------------------------------------\n\n");
3430 endBody += Form("In case of problems please contact the SHUTTLE core team.\n\n");
3431 endBody += "Please do not answer this message directly, it is automatically generated.\n\n";
3432 endBody += "Greetings,\n\n \t\t\tthe SHUTTLE\n";
3434 AliDebug(2, Form("Body end: %s", endBody.Data()));
3436 mailBody << endBody.Data();
3441 TString mailCommand = Form("mail -s \"%s\" -c %s %s < %s",
3445 bodyFileName.Data());
3446 AliDebug(2, Form("mail command: %s", mailCommand.Data()));
3448 Bool_t result = gSystem->Exec(mailCommand.Data());
3452 //______________________________________________________________________________________________
3453 const char* AliShuttle::GetRunType()
3456 // returns run type read from "run type" logbook
3459 if(!fLogbookEntry) {
3460 AliError("No logbook entry!");
3464 return fLogbookEntry->GetRunType();
3467 //______________________________________________________________________________________________
3468 Bool_t AliShuttle::GetHLTStatus()
3470 // Return HLT status (ON=1 OFF=0)
3471 // Converts the HLT status from the mode string read in the run logbook (not just a bool)
3473 if(!fLogbookEntry) {
3474 AliError("No logbook entry!");
3478 // TODO implement when HLTMode is inserted in run logbook
3479 TString hltMode = fLogbookEntry->GetRunParameter("HLTmode");
3480 TSubString firstChar = hltMode(0,1);
3481 AliDebug(2,Form("First char = %s ",firstChar.Data()));
3482 if (firstChar == "A") {
3485 else if ((firstChar == "B") || (firstChar == "C") || (firstChar == "D") || (firstChar == "E")) {
3489 Log("SHUTTLE","Unexpected HLT mode! Returning 0....");
3494 //______________________________________________________________________________________________
3495 const char* AliShuttle::GetTriggerConfiguration()
3497 // Receives the trigger configuration from the DAQ logbook for the current run
3499 // check connection, if needed reconnect
3504 sqlQuery.Form("SELECT configFile FROM logbook_trigger_config WHERE run = %d", GetCurrentRun());
3505 TSQLResult* result = fServer[4]->Query(sqlQuery);
3508 Log("SHUTTLE", Form("ERROR: Can't execute query <%s>!", sqlQuery.Data()));
3512 if (result->GetRowCount() == 0)
3514 Log("SHUTTLE", "WARNING: Trigger configuration not found in logbook_trigger_config");
3519 TSQLRow* row = result->Next();
3522 Log("SHUTTLE", "ERROR: Could not receive logbook_trigger_config data");
3527 // static, so that pointer remains valid when it is returned to the calling class
3528 static TString triggerConfig(row->GetField(0));
3536 Log("SHUTTLE", Form("Found trigger configuration: %s", triggerConfig.Data()));
3538 return triggerConfig;
3541 //______________________________________________________________________________________________
3542 const char* AliShuttle::GetCTPTimeParams()
3544 // Receives the CTP time parameters from the DAQ logbook for the current run
3546 // check connection, if needed reconnect
3551 sqlQuery.Form("SELECT alignmentFile FROM logbook_trigger_config WHERE run = %d", GetCurrentRun());
3552 TSQLResult* result = fServer[4]->Query(sqlQuery);
3555 Log("SHUTTLE", Form("ERROR: Can't execute query <%s>!", sqlQuery.Data()));
3559 if (result->GetRowCount() == 0)
3561 Log("SHUTTLE", "WARNING: CTP time params not found in logbook_trigger_config");
3566 TSQLRow* row = result->Next();
3569 Log("SHUTTLE", "ERROR: Could not receive logbook_trigger_config data");
3574 // static, so that pointer remains valid when it is returned to the calling class
3575 static TString triggerTimeParams(row->GetField(0));
3583 Log("SHUTTLE", Form("Found trigger time parameters: %s", triggerTimeParams.Data()));
3585 return triggerTimeParams;
3588 //______________________________________________________________________________________________
3589 const char* AliShuttle::GetTriggerDetectorMask()
3591 // Receives the trigger detector mask from DAQ logbook
3593 // check connection, if needed reconnect
3598 sqlQuery.Form("SELECT BIN(BIT_OR(inputDetectorMask)) from logbook_trigger_clusters WHERE run = %d;", GetCurrentRun());
3599 TSQLResult* result = fServer[4]->Query(sqlQuery);
3602 Log("SHUTTLE", Form("ERROR: Can't execute query <%s>!", sqlQuery.Data()));
3606 if (result->GetRowCount() == 0)
3608 Log("SHUTTLE", "ERROR: Trigger Detector Mask not found in logbook_trigger_clusters");
3613 TSQLRow* row = result->Next();
3616 Log("SHUTTLE", "ERROR: Could not receive logbook_trigger_clusters data");
3621 // static, so that pointer remains valid when it is returned to the calling class
3622 static TString triggerDetectorMask(row->GetField(0));
3630 Log("SHUTTLE", Form("Found Trigger Detector Mask: %s", triggerDetectorMask.Data()));
3632 return triggerDetectorMask;
3635 //______________________________________________________________________________________________
3636 void AliShuttle::SetShuttleTempDir(const char* tmpDir)
3639 // sets Shuttle temp directory
3642 fgkShuttleTempDir = gSystem->ExpandPathName(tmpDir);
3645 //______________________________________________________________________________________________
3646 void AliShuttle::SetShuttleLogDir(const char* logDir)
3649 // sets Shuttle log directory
3652 fgkShuttleLogDir = gSystem->ExpandPathName(logDir);
3654 //______________________________________________________________________________________________
3655 Bool_t AliShuttle::TouchFile()
3658 // touching a file on the grid if run has been DONE
3663 Log("SHUTTLE",Form("No TGrid connection estabilished!"));
3664 Log("SHUTTLE",Form("Could not touch file for run %i",GetCurrentRun()));
3669 dir.Form("%s%d/SHUTTLE_DONE", fConfig->GetAlienPath(), GetCurrentYear());
3670 // checking whether directory for touch command exists
3671 TGridResult* resultLs = gGrid->Ls(dir.Data());
3672 if (!resultLs){ // unfortunately we don't get this for ls of a non existing dir
3673 Log("SHUTTLE",Form("No result for \"Ls(\"%s\")\", returning without touching", dir.Data()));
3677 if ( resultLs->GetEntries() == 1 && !resultLs->GetFileName(0) ) {
3678 // this is what we currently get for ls of a non existing dir
3679 Log("SHUTTLE",Form("No directory %s found, creating it",dir.Data()));
3681 // creating the directory
3683 Bool_t boolMkdir = gGrid->Mkdir(dir.Data());
3685 Log("SHUTTLE",Form("Impossible to create dir %s in alien catalogue for run %i!",dir.Data(),GetCurrentRun()));
3690 Log("SHUTTLE",Form("Directory %s successfully created in alien catalogue for run %i",dir.Data(),GetCurrentRun()));
3693 Log("SHUTTLE",Form("Directory %s correctly found for run %i",dir.Data(),GetCurrentRun()));
3699 // Before trying to touch, check that the file is not already there (the touch would fail forever, leaving the run in pending)
3700 TString lsFileCommand;
3701 lsFileCommand.Form("ls %s/%i", dir.Data(), GetCurrentRun());
3702 TGridResult *resultLsFile = dynamic_cast<TGridResult*>(gGrid->Command(lsFileCommand));
3704 Log("SHUTTLE",Form("No result for file ls command, returning without touching for run %i",GetCurrentRun()));
3707 TMap *mapLsFile = dynamic_cast<TMap*>(resultLsFile->At(0));
3709 Log("SHUTTLE",Form("No map for file ls command, returning without touching for run %i",GetCurrentRun()));
3710 delete resultLsFile;
3714 TObjString *valueLsFile = dynamic_cast<TObjString*>(mapLsFile->GetValue("name"));
3716 Log("SHUTTLE",Form("\"name\" key set in the map for file ls command. Touchfile for run %i already there.",GetCurrentRun()));
3717 Log("SHUTTLE", "The file was already there, did not touch it.");
3720 command.Form("touch %s/%i", dir.Data(), GetCurrentRun());
3721 Log("SHUTTLE", Form("Creating entry in file catalog: %s", command.Data()));
3722 TGridResult *resultTouch = dynamic_cast<TGridResult*>(gGrid->Command(command));
3724 Log("SHUTTLE",Form("No result for touching command, returning without touching for run %i",GetCurrentRun()));
3727 TMap *mapTouch = dynamic_cast<TMap*>(resultTouch->At(0));
3729 Log("SHUTTLE",Form("No map for touching command, returning without touching for run %i",GetCurrentRun()));
3734 TObjString *valueTouch = dynamic_cast<TObjString*>(mapTouch->GetValue("__result__"));
3736 Log("SHUTTLE",Form("No value for \"__result__\" key set in the map for touching command, returning without touching for run %i",GetCurrentRun()));
3741 if (valueTouch->GetString()!="1"){
3742 Log("SHUTTLE",Form("Failing the touching command, returning without touching for run %i",GetCurrentRun()));
3747 delete resultLsFile;
3751 Log("SHUTTLE", "Sucessfully touched the file");
3756 //______________________________________________________________________________________________
3757 UInt_t AliShuttle::GetStartTimeDCSQuery()
3759 // Return Start Time for the DCS query
3761 // The call is delegated to AliShuttleInterface
3763 return GetCurrentStartTime()-fConfig->GetDCSQueryOffset();
3765 //______________________________________________________________________________________________
3766 UInt_t AliShuttle::GetEndTimeDCSQuery()
3768 // Return End Time for the DCS query
3770 // The call is delegated to AliShuttleInterface
3772 return GetCurrentEndTime()+fConfig->GetDCSQueryOffset();
3774 //______________________________________________________________________________________________
3775 void AliShuttle::SendMLFromDet(const char* value)
3778 // Sending an information coming from the current detector to ML
3781 TMonaLisaText mlText(Form("%s_RunCondition", fCurrentDetector.Data()), value);
3784 mlList.Add(&mlText);
3787 mlID.Form("%d", GetCurrentRun());
3788 fMonaLisa->SendParameters(&mlList, mlID);
3792 //______________________________________________________________________________________________
3793 TString* AliShuttle::GetLTUConfig(const char* det)
3796 // Getting ltuFineDelay1, ltuFineDelay2, ltuBCDelay for detector det from logbook_detectors table in logbook
3803 sqlQuery.Form("select LTUFineDelay1, LTUFineDelay2, LTUBCDelayAdd from logbook_detectors WHERE run_number = %d and detector = \"%s\";", GetCurrentRun(),det);
3805 TSQLResult* result = fServer[4]->Query(sqlQuery);
3807 Log("SHUTTLE","ERROR: No result found for the LTU configuration query");
3810 if (result->GetRowCount() == 0){
3811 Log("SHUTTLE",Form("ERROR: LTU configuration not found in logbook_detectors for detector %s, returning null pointer",det));
3815 if (result->GetFieldCount() != 3){
3816 Log("SHUTTLE",Form("ERROR: not all the required fields are there for the LTU configuration for detector %s (only %d found), returning a null pointer",det, result->GetFieldCount()));
3820 TSQLRow* row = result->Next();
3822 Printf("ERROR: Could not receive logbook_detectors data, returning null pointer");
3826 TString* ltuConfigString = new TString[3];
3828 ltuConfigString[0] = row->GetField(0);
3829 ltuConfigString[1] = row->GetField(1);
3830 ltuConfigString[2] = row->GetField(2);
3832 return ltuConfigString;