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.
31 #include "AliShuttle.h"
33 #include "AliCDBManager.h"
34 #include "AliCDBStorage.h"
36 #include "AliCDBRunRange.h"
37 #include "AliCDBPath.h"
38 #include "AliCDBEntry.h"
39 #include "AliShuttleConfig.h"
40 #include "DCSClient/AliDCSClient.h"
42 #include "AliPreprocessor.h"
43 #include "AliShuttleStatus.h"
44 #include "AliShuttleLogbookEntry.h"
49 #include <TTimeStamp.h>
50 #include <TObjString.h>
51 #include <TSQLServer.h>
52 #include <TSQLResult.h>
55 #include <TSystemDirectory.h>
56 #include <TSystemFile.h>
59 #include <TGridResult.h>
62 #include <TMonaLisaWriter.h>
66 #include <sys/types.h>
73 //______________________________________________________________________________________________
74 AliShuttle::AliShuttle(const AliShuttleConfig* config,
75 UInt_t timeout, Int_t retries):
77 fTimeout(timeout), fRetries(retries),
89 fReadTestMode(kFALSE),
90 fOutputRedirected(kFALSE)
93 // config: AliShuttleConfig used
94 // timeout: timeout used for AliDCSClient connection
95 // retries: the number of retries in case of connection error.
98 if (!fConfig->IsValid()) AliFatal("********** !!!!! Invalid configuration !!!!! **********");
99 for(int iSys=0;iSys<4;iSys++) {
102 fFXSlist[iSys].SetOwner(kTRUE);
104 fPreprocessorMap.SetOwner(kTRUE);
106 for (UInt_t iDet=0; iDet<NDetectors(); iDet++)
107 fFirstUnprocessed[iDet] = kFALSE;
109 fMonitoringMutex = new TMutex();
112 //______________________________________________________________________________________________
113 AliShuttle::~AliShuttle()
119 fPreprocessorMap.DeleteAll();
120 for(int iSys=0;iSys<4;iSys++)
122 fServer[iSys]->Close();
123 delete fServer[iSys];
132 if (fMonitoringMutex)
134 delete fMonitoringMutex;
135 fMonitoringMutex = 0;
139 //______________________________________________________________________________________________
140 void AliShuttle::RegisterPreprocessor(AliPreprocessor* preprocessor)
143 // Registers new AliPreprocessor.
144 // It uses GetName() for indentificator of the pre processor.
145 // The pre processor is registered it there isn't any other
146 // with the same identificator (GetName()).
149 const char* detName = preprocessor->GetName();
150 if(GetDetPos(detName) < 0)
151 AliFatal(Form("********** !!!!! Invalid detector name: %s !!!!! **********", detName));
153 if (fPreprocessorMap.GetValue(detName)) {
154 AliWarning(Form("AliPreprocessor %s is already registered!", detName));
158 fPreprocessorMap.Add(new TObjString(detName), preprocessor);
160 //______________________________________________________________________________________________
161 Bool_t AliShuttle::Store(const AliCDBPath& path, TObject* object,
162 AliCDBMetaData* metaData, Int_t validityStart, Bool_t validityInfinite)
164 // Stores a CDB object in the storage for offline reconstruction. Objects that are not needed for
165 // offline reconstruction, but should be stored anyway (e.g. for debugging) should NOT be stored
166 // using this function. Use StoreReferenceData instead!
167 // It calls StoreLocally function which temporarily stores the data locally; when the preprocessor
168 // finishes the data are transferred to the main storage (Grid).
170 return StoreLocally(fgkLocalCDB, path, object, metaData, validityStart, validityInfinite);
173 //______________________________________________________________________________________________
174 Bool_t AliShuttle::StoreReferenceData(const AliCDBPath& path, TObject* object, AliCDBMetaData* metaData)
176 // Stores a CDB object in the storage for reference data. This objects will not be available during
177 // offline reconstrunction. Use this function for reference data only!
178 // It calls StoreLocally function which temporarily stores the data locally; when the preprocessor
179 // finishes the data are transferred to the main storage (Grid).
181 return StoreLocally(fgkLocalRefStorage, path, object, metaData);
184 //______________________________________________________________________________________________
185 Bool_t AliShuttle::StoreLocally(const TString& localUri,
186 const AliCDBPath& path, TObject* object, AliCDBMetaData* metaData,
187 Int_t validityStart, Bool_t validityInfinite)
189 // Store object temporarily in local storage. Parameters are passed by Store and StoreReferenceData functions.
190 // when the preprocessor finishes the data are transferred to the main storage (Grid).
191 // The parameters are:
192 // 1) Uri of the backup storage (Local)
193 // 2) the object's path.
194 // 3) the object to be stored
195 // 4) the metaData to be associated with the object
196 // 5) the validity start run number w.r.t. the current run,
197 // if the data is valid only for this run leave the default 0
198 // 6) specifies if the calibration data is valid for infinity (this means until updated),
199 // typical for calibration runs, the default is kFALSE
201 // returns 0 if fail, 1 otherwise
203 if (fTestMode & kErrorStorage)
205 Log(fCurrentDetector, "StoreLocally - In TESTMODE - Simulating error while storing locally");
209 const char* cdbType = (localUri == fgkLocalCDB) ? "CDB" : "Reference";
211 Int_t firstRun = GetCurrentRun() - validityStart;
213 AliWarning("First valid run happens to be less than 0! Setting it to 0.");
218 if(validityInfinite) {
219 lastRun = AliCDBRunRange::Infinity();
221 lastRun = GetCurrentRun();
224 // Version is set to current run, it will be used later to transfer data to Grid
225 AliCDBId id(path, firstRun, lastRun, GetCurrentRun(), -1);
227 if(! dynamic_cast<TObjString*> (metaData->GetProperty("RunUsed(TObjString)"))){
228 TObjString runUsed = Form("%d", GetCurrentRun());
229 metaData->SetProperty("RunUsed(TObjString)", runUsed.Clone());
232 Bool_t result = kFALSE;
234 if (!(AliCDBManager::Instance()->GetStorage(localUri))) {
235 Log("SHUTTLE", Form("StoreLocally - Cannot activate local %s storage", cdbType));
237 result = AliCDBManager::Instance()->GetStorage(localUri)
238 ->Put(object, id, metaData);
243 Log(fCurrentDetector, Form("StoreLocally - Can't store object <%s>!", id.ToString().Data()));
249 //______________________________________________________________________________________________
250 Bool_t AliShuttle::StoreOCDB()
253 // Called when preprocessor ends successfully or when previous storage attempt failed (kStoreError status)
254 // Calls underlying StoreOCDB(const char*) function twice, for OCDB and Reference storage.
255 // Then calls StoreRefFilesToGrid to store reference files.
258 UpdateShuttleStatus(AliShuttleStatus::kStoreStarted);
260 if (fTestMode & kErrorGrid)
262 Log("SHUTTLE", "StoreOCDB - In TESTMODE - Simulating error while storing in the Grid");
263 Log(fCurrentDetector, "StoreOCDB - In TESTMODE - Simulating error while storing in the Grid");
267 Log("SHUTTLE","StoreOCDB - Storing OCDB data ...");
268 Int_t resultCDB = StoreOCDB(fgkMainCDB);
270 Log("SHUTTLE","StoreOCDB - Storing reference data ...");
271 Int_t resultRef = StoreOCDB(fgkMainRefStorage);
273 Log("SHUTTLE","StoreOCDB - Storing reference files ...");
274 Bool_t resultRefFiles = CopyFilesToGrid("reference");
276 Bool_t resultMetadata = kTRUE;
277 if(fCurrentDetector == "GRP")
279 Log("SHUTTLE","StoreOCDB - Storing Run Metadata file ...");
280 resultMetadata = CopyFilesToGrid("metadata");
283 Int_t storeResult = 0;
285 if (resultCDB < 0 || resultRef < 0 || resultRefFiles == kFALSE || resultMetadata == kFALSE)
287 else if (resultCDB > 0 || resultRef > 0)
293 Form("\t\t\t****** run %d - %s: STORAGE ERROR ******",
294 GetCurrentRun(), fCurrentDetector.Data()));
295 UpdateShuttleStatus(AliShuttleStatus::kStoreError);
297 else if (storeResult > 0)
300 Form("\t\t\t****** run %d - %s: STORAGE DELAYED ******",
301 GetCurrentRun(), fCurrentDetector.Data()));
302 UpdateShuttleStatus(AliShuttleStatus::kStoreDelayed);
304 else if (storeResult == 0)
307 Form("\t\t\t****** run %d - %s: DONE ******",
308 GetCurrentRun(), fCurrentDetector.Data()));
309 UpdateShuttleStatus(AliShuttleStatus::kDone);
310 UpdateShuttleLogbook(fCurrentDetector, "DONE");
313 return (storeResult == 0);
316 //______________________________________________________________________________________________
317 Int_t AliShuttle::StoreOCDB(const TString& gridURI)
320 // Called by StoreOCDB(), performs actual storage to the main OCDB and reference storages (Grid)
323 // -2 initialization error
326 // 1 storage delayed (e.g. previous unprocessed runs)
329 TObjArray* gridIds=0;
331 Bool_t result = kTRUE;
332 Bool_t delayed = kFALSE;
334 const char* type = 0;
336 if(gridURI == fgkMainCDB) {
338 localURI = fgkLocalCDB;
339 } else if(gridURI == fgkMainRefStorage) {
341 localURI = fgkLocalRefStorage;
343 AliError(Form("Invalid storage URI: %s", gridURI.Data()));
347 AliCDBManager* man = AliCDBManager::Instance();
349 AliCDBStorage *gridSto = man->GetStorage(gridURI);
352 Form("StoreOCDB - cannot activate main %s storage", type));
356 gridIds = gridSto->GetQueryCDBList();
358 // get objects previously stored in local CDB
359 AliCDBStorage *localSto = man->GetStorage(localURI);
362 Form("StoreOCDB - cannot activate local %s storage", type));
365 AliCDBPath aPath(GetOfflineDetName(fCurrentDetector.Data()),"*","*");
366 // Local objects were stored with current run as Grid version!
367 TList* localEntries = localSto->GetAll(aPath.GetPath(), GetCurrentRun(), GetCurrentRun());
368 localEntries->SetOwner(1);
370 // loop on local stored objects
371 TIter localIter(localEntries);
372 AliCDBEntry *aLocEntry = 0;
373 while((aLocEntry = dynamic_cast<AliCDBEntry*> (localIter.Next()))){
374 aLocEntry->SetOwner(1);
375 AliCDBId aLocId = aLocEntry->GetId();
376 aLocEntry->SetVersion(-1);
377 aLocEntry->SetSubVersion(-1);
379 Log(fCurrentDetector.Data(), Form("Attempting to store %s", aLocId.ToString().Data()));
381 // If local object is valid up to infinity we store it only if it is
382 // the first unprocessed run!
383 if (aLocId.GetLastRun() == AliCDBRunRange::Infinity() &&
384 !fFirstUnprocessed[GetDetPos(fCurrentDetector)])
386 Log("SHUTTLE", Form("StoreOCDB - %s: object %s has validity infinite but "
387 "there are previous unprocessed runs!",
388 fCurrentDetector.Data(), aLocId.GetPath().Data()));
389 Log(fCurrentDetector.Data(), Form("StoreOCDB - %s: object %s has validity infinite but "
390 "there are previous unprocessed runs!",
391 fCurrentDetector.Data(), aLocId.GetPath().Data()));
396 // loop on Grid valid Id's
397 Bool_t store = kTRUE;
398 TIter gridIter(gridIds);
399 AliCDBId* aGridId = 0;
400 while ((aGridId = dynamic_cast<AliCDBId*> (gridIter.Next()))) {
401 if (aGridId->GetPath() != aLocId.GetPath())
403 // skip all objects valid up to infinity
404 if (aGridId->GetLastRun() == AliCDBRunRange::Infinity())
407 // if we get here, it means there's already some more recent object stored on Grid!
408 Log(fCurrentDetector.Data(),
409 Form("StoreOCDB - A more recent object already exists in %s storage: <%s>",
410 type, aGridId->ToString().Data()));
416 Bool_t storeOk = kFALSE;
419 Log(fCurrentDetector.Data(), Form("Prechecks succeeded. Ready to store %s", aLocId.ToString().Data()));
420 storeOk = gridSto->Put(aLocEntry);
423 Form("StoreOCDB - Object <%s> successfully put into %s storage",
424 aLocId.ToString().Data(), type));
425 Log(fCurrentDetector.Data(),
426 Form("StoreOCDB - Object <%s> successfully put into %s storage",
427 aLocId.ToString().Data(), type));
430 Form("StoreOCDB - Grid %s storage of object <%s> failed",
431 type, aLocId.ToString().Data()));
432 Log(fCurrentDetector.Data(),
433 Form("StoreOCDB - Grid %s storage of object <%s> failed",
434 type, aLocId.ToString().Data()));
439 if (!store || storeOk) {
440 // removing local file...
442 localSto->IdToFilename(aLocId, filename);
443 Log("SHUTTLE", Form("StoreOCDB - Removing local file %s", filename.Data()));
444 RemoveFile(filename.Data());
447 localEntries->Clear();
449 Int_t returnCode = 0;
451 if (result == kFALSE)
453 else if (delayed != kFALSE)
456 Log("SHUTTLE", Form("StoreOCDB - Returning with %d (result = %d, delayed = %d)", returnCode, result, delayed));
457 Log(fCurrentDetector.Data(), Form("StoreOCDB - Returning with %d (result = %d, delayed = %d)", returnCode, result, delayed));
462 //______________________________________________________________________________________________
463 Bool_t AliShuttle::CleanReferenceStorage(const char* detector)
465 // clears the directory used to store reference files of a given subdetector
467 AliCDBManager* man = AliCDBManager::Instance();
468 AliCDBStorage* sto = man->GetStorage(fgkLocalRefStorage);
469 TString localBaseFolder = sto->GetBaseFolder();
471 TString targetDir = GetRefFilePrefix(localBaseFolder.Data(), detector);
473 Log("SHUTTLE", Form("CleanReferenceStorage - Cleaning %s", targetDir.Data()));
476 begin.Form("%d_", GetCurrentRun());
478 TSystemDirectory* baseDir = new TSystemDirectory("/", targetDir);
482 TList* dirList = baseDir->GetListOfFiles();
485 if (!dirList) return kTRUE;
487 if (dirList->GetEntries() < 3)
493 Int_t nDirs = 0, nDel = 0;
494 TIter dirIter(dirList);
495 TSystemFile* entry = 0;
497 Bool_t success = kTRUE;
499 while ((entry = dynamic_cast<TSystemFile*> (dirIter.Next())))
501 if (entry->IsDirectory())
504 TString fileName(entry->GetName());
505 if (!fileName.BeginsWith(begin))
511 Int_t result = gSystem->Unlink(fileName.Data());
515 Log("SHUTTLE", Form("CleanReferenceStorage - Could not delete file %s!", fileName.Data()));
523 Log("SHUTTLE", Form("CleanReferenceStorage - %d (over %d) reference files in folder %s were deleted.",
524 nDel, nDirs, targetDir.Data()));
535 Int_t result = gSystem->GetPathInfo(targetDir, 0, (Long64_t*) 0, 0, 0);
539 result = gSystem->Exec(Form("rm -rf %s", targetDir.Data()));
542 Log("SHUTTLE", Form("CleanReferenceStorage - Could not clean directory %s", targetDir.Data()));
547 result = gSystem->mkdir(targetDir, kTRUE);
550 Log("SHUTTLE", Form("CleanReferenceStorage - Error creating base directory %s", targetDir.Data()));
557 //______________________________________________________________________________________________
558 Bool_t AliShuttle::StoreReferenceFile(const char* detector, const char* localFile, const char* gridFileName)
561 // Stores reference file directly (without opening it). This function stores the file locally.
563 // The file is stored under the following location:
564 // <base folder of local reference storage>/<DET>/<RUN#>_<gridFileName>
565 // where <gridFileName> is the second parameter given to the function
568 if (fTestMode & kErrorStorage)
570 Log(fCurrentDetector, "StoreReferenceFile - In TESTMODE - Simulating error while storing locally");
574 AliCDBManager* man = AliCDBManager::Instance();
575 AliCDBStorage* sto = man->GetStorage(fgkLocalRefStorage);
577 TString localBaseFolder = sto->GetBaseFolder();
579 TString target = GetRefFilePrefix(localBaseFolder.Data(), detector);
580 target.Append(Form("/%d_%s", GetCurrentRun(), gridFileName));
582 return CopyFileLocally(localFile, target);
585 //______________________________________________________________________________________________
586 Bool_t AliShuttle::StoreRunMetadataFile(const char* localFile, const char* gridFileName)
589 // Stores Run metadata file to the Grid, in the run folder
591 // Only GRP can call this function.
593 if (fTestMode & kErrorStorage)
595 Log(fCurrentDetector, "StoreRunMetaDataFile - In TESTMODE - Simulating error while storing locally");
599 AliCDBManager* man = AliCDBManager::Instance();
600 AliCDBStorage* sto = man->GetStorage(fgkLocalRefStorage);
602 TString localBaseFolder = sto->GetBaseFolder();
604 // Build Run level folder
605 // folder = /alice/data/year/lhcPeriod/runNb/raw
608 TString lhcPeriod = GetLHCPeriod();
609 if (lhcPeriod.Length() == 0)
611 Log("SHUTTLE","StoreRunMetaDataFile - LHCPeriod not found in logbook!");
615 // TODO partitions with one detector only write data into LHCperiod_DET
616 TString partition = GetRunParameter("detector");
618 if (partition.Length() > 0 && partition != "ALICE")
620 lhcPeriod.Append(Form("_%s", partition.Data()));
621 Log(fCurrentDetector, Form("Run data tags merged file will be written in %s",
625 TString target = Form("%s/GRP/RunMetadata%s%d/%s/%09d/raw/%s",
626 localBaseFolder.Data(), fConfig->GetAlienPath(), GetCurrentYear(),
627 lhcPeriod.Data(), GetCurrentRun(), gridFileName);
629 return CopyFileLocally(localFile, target);
632 //______________________________________________________________________________________________
633 Bool_t AliShuttle::CopyFileLocally(const char* localFile, const TString& target)
636 // Stores file locally. Called by StoreReferenceFile and StoreRunMetadataFile
637 // Files are temporarily stored in the local reference storage. When the preprocessor
638 // finishes, the Shuttle calls CopyFilesToGrid to transfer the files to AliEn
639 // (in reference or run level folders)
642 TString targetDir(target(0, target.Last('/')));
644 //try to open base dir folder, if it does not exist
645 void* dir = gSystem->OpenDirectory(targetDir.Data());
647 if (gSystem->mkdir(targetDir.Data(), kTRUE)) {
648 Log("SHUTTLE", Form("CopyFileLocally - Can't open directory <%s>", targetDir.Data()));
653 gSystem->FreeDirectory(dir);
658 result = gSystem->GetPathInfo(localFile, 0, (Long64_t*) 0, 0, 0);
661 Log("SHUTTLE", Form("CopyFileLocally - %s does not exist", localFile));
665 result = gSystem->GetPathInfo(target, 0, (Long64_t*) 0, 0, 0);
668 Log("SHUTTLE", Form("CopyFileLocally - target file %s already exist, removing...", target.Data()));
669 if (gSystem->Unlink(target.Data()))
671 Log("SHUTTLE", Form("CopyFileLocally - Could not remove existing target file %s!", target.Data()));
676 result = gSystem->CopyFile(localFile, target);
680 Log("SHUTTLE", Form("CopyFileLocally - File %s stored locally to %s", localFile, target.Data()));
685 Log("SHUTTLE", Form("CopyFileLocally - Could not store file %s to %s! Error code = %d",
686 localFile, target.Data(), result));
694 //______________________________________________________________________________________________
695 Bool_t AliShuttle::CopyFilesToGrid(const char* type)
698 // Transfers local files to the Grid. Local files can be reference files
699 // or run metadata file (from GRP only).
701 // According to the type (ref, metadata) the files are stored under the following location:
702 // ref --> <base folder of reference storage>/<DET>/<RUN#>_<gridFileName>
703 // metadata --> <run data folder>/<MetadataFileName>
706 AliCDBManager* man = AliCDBManager::Instance();
707 AliCDBStorage* sto = man->GetStorage(fgkLocalRefStorage);
710 TString localBaseFolder = sto->GetBaseFolder();
716 if (strcmp(type, "reference") == 0)
718 dir = GetRefFilePrefix(localBaseFolder.Data(), fCurrentDetector.Data());
719 AliCDBStorage* gridSto = man->GetStorage(fgkMainRefStorage);
722 TString gridBaseFolder = gridSto->GetBaseFolder();
723 alienDir = GetRefFilePrefix(gridBaseFolder.Data(), fCurrentDetector.Data());
724 begin = Form("%d_", GetCurrentRun());
726 else if (strcmp(type, "metadata") == 0)
729 TString lhcPeriod = GetLHCPeriod();
731 if (lhcPeriod.Length() == 0)
733 Log("SHUTTLE","CopyFilesToGrid - LHCPeriod not found in logbook!");
737 // TODO partitions with one detector only write data into LHCperiod_DET
738 TString partition = GetRunParameter("detector");
740 if (partition.Length() > 0 && partition != "ALICE")
742 lhcPeriod.Append(Form("_%s", partition.Data()));
745 dir = Form("%s/GRP/RunMetadata%s%d/%s/%09d/raw",
746 localBaseFolder.Data(), fConfig->GetAlienPath(), GetCurrentYear(),
747 lhcPeriod.Data(), GetCurrentRun());
748 alienDir = dir(dir.Index(fConfig->GetAlienPath()), dir.Length());
754 Log("SHUTTLE", "CopyFilesToGrid - Unexpected: type label must be reference or metadata!");
758 TSystemDirectory* baseDir = new TSystemDirectory("/", dir);
762 TList* dirList = baseDir->GetListOfFiles();
765 if (!dirList) return kTRUE;
767 if (dirList->GetEntries() < 3)
775 Log("SHUTTLE", "CopyFilesToGrid - Connection to Grid failed: Cannot continue!");
780 Int_t nDirs = 0, nTransfer = 0;
781 TIter dirIter(dirList);
782 TSystemFile* entry = 0;
784 Bool_t success = kTRUE;
785 Bool_t first = kTRUE;
787 while ((entry = dynamic_cast<TSystemFile*> (dirIter.Next())))
789 if (entry->IsDirectory())
792 TString fileName(entry->GetName());
793 if (!fileName.BeginsWith(begin))
801 // check that folder exists, otherwise create it
802 TGridResult* result = gGrid->Ls(alienDir.Data(), "a");
810 if (!result->GetFileName(1)) // TODO: It looks like element 0 is always 0!!
812 // TODO It does not work currently! Bug in TAliEn::Mkdir
813 // TODO Manually fixed in local root v5-16-00
814 if (!gGrid->Mkdir(alienDir.Data(),"-p",0))
816 Log("SHUTTLE", Form("CopyFilesToGrid - Cannot create directory %s",
821 Log("SHUTTLE",Form("CopyFilesToGrid - Folder %s created", alienDir.Data()));
825 Log("SHUTTLE",Form("CopyFilesToGrid - Folder %s found", alienDir.Data()));
829 TString fullLocalPath;
830 fullLocalPath.Form("%s/%s", dir.Data(), fileName.Data());
832 TString fullGridPath;
833 fullGridPath.Form("alien://%s/%s", alienDir.Data(), fileName.Data());
835 Bool_t result = TFile::Cp(fullLocalPath, fullGridPath);
839 Log("SHUTTLE", Form("CopyFilesToGrid - Copying local file %s to %s succeeded!",
840 fullLocalPath.Data(), fullGridPath.Data()));
841 RemoveFile(fullLocalPath);
846 Log("SHUTTLE", Form("CopyFilesToGrid - Copying local file %s to %s FAILED!",
847 fullLocalPath.Data(), fullGridPath.Data()));
852 Log("SHUTTLE", Form("CopyFilesToGrid - %d (over %d) files in folder %s copied to Grid.",
853 nTransfer, nDirs, dir.Data()));
860 //______________________________________________________________________________________________
861 const char* AliShuttle::GetRefFilePrefix(const char* base, const char* detector)
864 // Get folder name of reference files
867 TString offDetStr(GetOfflineDetName(detector));
869 if (offDetStr == "ITS" || offDetStr == "MUON" || offDetStr == "PHOS")
871 dir.Form("%s/%s/%s", base, offDetStr.Data(), detector);
873 dir.Form("%s/%s", base, offDetStr.Data());
881 //______________________________________________________________________________________________
882 void AliShuttle::CleanLocalStorage(const TString& uri)
885 // Called in case the preprocessor is declared failed. Remove remaining objects from the local storages.
888 const char* type = 0;
889 if(uri == fgkLocalCDB) {
891 } else if(uri == fgkLocalRefStorage) {
894 AliError(Form("Invalid storage URI: %s", uri.Data()));
898 AliCDBManager* man = AliCDBManager::Instance();
900 // open local storage
901 AliCDBStorage *localSto = man->GetStorage(uri);
904 Form("CleanLocalStorage - cannot activate local %s storage", type));
908 TString filename(Form("%s/%s/*/Run*_v%d_s*.root",
909 localSto->GetBaseFolder().Data(), GetOfflineDetName(fCurrentDetector.Data()), GetCurrentRun()));
911 AliDebug(2, Form("filename = %s", filename.Data()));
913 Log("SHUTTLE", Form("Removing remaining local files for run %d and detector %s ...",
914 GetCurrentRun(), fCurrentDetector.Data()));
916 RemoveFile(filename.Data());
920 //______________________________________________________________________________________________
921 void AliShuttle::RemoveFile(const char* filename)
924 // removes local file
927 TString command(Form("rm -f %s", filename));
929 Int_t result = gSystem->Exec(command.Data());
932 Log("SHUTTLE", Form("RemoveFile - %s: Cannot remove file %s!",
933 fCurrentDetector.Data(), filename));
937 //______________________________________________________________________________________________
938 AliShuttleStatus* AliShuttle::ReadShuttleStatus()
941 // Reads the AliShuttleStatus from the CDB
949 Int_t path1 = GetCurrentRun()/10000;
950 fStatusEntry = AliCDBManager::Instance()->GetStorage(GetLocalCDB())
951 ->Get(Form("/SHUTTLE/%s/%d", fCurrentDetector.Data(), path1), GetCurrentRun());
953 if (!fStatusEntry) return 0;
954 fStatusEntry->SetOwner(1);
956 AliShuttleStatus* status = dynamic_cast<AliShuttleStatus*> (fStatusEntry->GetObject());
958 AliError("Invalid object stored to CDB!");
965 //______________________________________________________________________________________________
966 Bool_t AliShuttle::WriteShuttleStatus(AliShuttleStatus* status)
969 // writes the status for one subdetector
977 Int_t run = GetCurrentRun();
978 Int_t path1 = run/10000;
979 TString path1_string = Form("%d",path1);
981 AliCDBId id(AliCDBPath("SHUTTLE", fCurrentDetector, path1_string), run, run);
983 fStatusEntry = new AliCDBEntry(status, id, new AliCDBMetaData);
984 fStatusEntry->SetOwner(1);
986 UInt_t result = AliCDBManager::Instance()->GetStorage(fgkLocalCDB)->Put(fStatusEntry);
989 Log("SHUTTLE", Form("WriteShuttleStatus - Failed for %s, run %d",
990 fCurrentDetector.Data(), run));
999 //______________________________________________________________________________________________
1000 void AliShuttle::UpdateShuttleStatus(AliShuttleStatus::Status newStatus, Bool_t increaseCount)
1003 // changes the AliShuttleStatus for the given detector and run to the given status
1007 AliError("UNEXPECTED: fStatusEntry empty");
1011 AliShuttleStatus* status = dynamic_cast<AliShuttleStatus*> (fStatusEntry->GetObject());
1014 Log("SHUTTLE", "UpdateShuttleStatus - UNEXPECTED: status could not be read from current CDB entry");
1018 TString actionStr = Form("UpdateShuttleStatus - %s: Changing state from %s to %s",
1019 fCurrentDetector.Data(),
1020 status->GetStatusName(),
1021 status->GetStatusName(newStatus));
1022 Log("SHUTTLE", actionStr);
1023 SetLastAction(actionStr);
1025 status->SetStatus(newStatus);
1026 if (increaseCount) status->IncreaseCount();
1028 AliCDBManager::Instance()->GetStorage(fgkLocalCDB)->Put(fStatusEntry);
1033 //______________________________________________________________________________________________
1034 void AliShuttle::SendMLDetInfo()
1037 // sends ML information about the current status of the current detector being processed
1040 AliShuttleStatus* status = dynamic_cast<AliShuttleStatus*> (fStatusEntry->GetObject());
1043 Log("SHUTTLE", "SendMLDetInfo - UNEXPECTED: status could not be read from current CDB entry");
1047 TMonaLisaText mlStatus(Form("%s_status", fCurrentDetector.Data()), status->GetStatusName());
1048 TMonaLisaValue mlRetryCount(Form("%s_count", fCurrentDetector.Data()), status->GetCount());
1051 mlList.Add(&mlStatus);
1052 mlList.Add(&mlRetryCount);
1055 mlID.Form("%d", GetCurrentRun());
1056 fMonaLisa->SendParameters(&mlList, mlID);
1059 //______________________________________________________________________________________________
1060 Bool_t AliShuttle::ContinueProcessing()
1062 // this function reads the AliShuttleStatus information from CDB and
1063 // checks if the processing should be continued
1064 // if yes it returns kTRUE and updates the AliShuttleStatus with nextStatus
1066 if (!fConfig->HostProcessDetector(fCurrentDetector))
1069 AliPreprocessor* aPreprocessor =
1070 dynamic_cast<AliPreprocessor*> (fPreprocessorMap.GetValue(fCurrentDetector));
1073 Log("SHUTTLE", Form("ContinueProcessing - %s: no preprocessor registered", fCurrentDetector.Data()));
1077 AliShuttleLogbookEntry::Status entryStatus =
1078 fLogbookEntry->GetDetectorStatus(fCurrentDetector);
1080 if (entryStatus != AliShuttleLogbookEntry::kUnprocessed) {
1081 Log("SHUTTLE", Form("ContinueProcessing - %s is %s",
1082 fCurrentDetector.Data(),
1083 fLogbookEntry->GetDetectorStatusName(entryStatus)));
1087 // if we get here, according to Shuttle logbook subdetector is in UNPROCESSED state
1089 // check if current run is first unprocessed run for current detector
1090 if (fConfig->StrictRunOrder(fCurrentDetector) &&
1091 !fFirstUnprocessed[GetDetPos(fCurrentDetector)])
1093 if (fTestMode == kNone)
1095 Log("SHUTTLE", Form("ContinueProcessing - %s requires strict run ordering"
1096 " but this is not the first unprocessed run!"));
1101 Log("SHUTTLE", Form("ContinueProcessing - In TESTMODE - "
1102 "Although %s requires strict run ordering "
1103 "and this is not the first unprocessed run, "
1104 "the SHUTTLE continues"));
1108 // Is the subdetector processed first time for this run?
1109 fFirstProcessing = kFALSE;
1111 AliShuttleStatus* status = ReadShuttleStatus();
1114 Log("SHUTTLE", Form("ContinueProcessing - %s: Processing first time",
1115 fCurrentDetector.Data()));
1116 status = new AliShuttleStatus(AliShuttleStatus::kStarted);
1117 fFirstProcessing = kTRUE;
1118 return WriteShuttleStatus(status);
1121 // The following case shouldn't happen if Shuttle Logbook was correctly updated.
1122 // If it happens it may mean Logbook updating failed... let's do it now!
1123 if (status->GetStatus() == AliShuttleStatus::kDone ||
1124 status->GetStatus() == AliShuttleStatus::kFailed ||
1125 status->GetStatus() == AliShuttleStatus::kSkipped) {
1126 Log("SHUTTLE", Form("ContinueProcessing - %s is already %s. Updating Shuttle Logbook",
1127 fCurrentDetector.Data(),
1128 status->GetStatusName(status->GetStatus())));
1130 if (status->GetStatus() == AliShuttleStatus::kSkipped)
1132 UpdateShuttleLogbook(fCurrentDetector.Data(), "DONE");
1135 UpdateShuttleLogbook(fCurrentDetector.Data(), status->GetStatusName(status->GetStatus()));
1140 if (status->GetStatus() == AliShuttleStatus::kStoreStarted || status->GetStatus() == AliShuttleStatus::kStoreDelayed ||status->GetStatus() == AliShuttleStatus::kStoreError) {
1142 Form("ContinueProcessing - %s: Grid storage of one or more "
1143 "objects failed. Trying again now",
1144 fCurrentDetector.Data()));
1149 // if we get here, there is a restart
1150 Bool_t cont = kFALSE;
1153 if (status->GetCount() >= fConfig->GetMaxRetries()) {
1154 Log("SHUTTLE", Form("ContinueProcessing - %s failed %d times in status %s - "
1155 "Updating Shuttle Logbook", fCurrentDetector.Data(),
1156 status->GetCount(), status->GetStatusName()));
1157 UpdateShuttleLogbook(fCurrentDetector.Data(), "FAILED");
1158 UpdateShuttleStatus(AliShuttleStatus::kFailed);
1160 // there may still be objects in local OCDB and reference storage
1161 // and FXS databases may be not updated: do it now!
1163 // TODO Currently disabled, we want to keep files in case of failure!
1164 // CleanLocalStorage(fgkLocalCDB);
1165 // CleanLocalStorage(fgkLocalRefStorage);
1166 // UpdateTableFailCase();
1168 // Send mail to detector expert!
1169 Log("SHUTTLE", Form("ContinueProcessing - Sending mail to %s expert...",
1170 fCurrentDetector.Data()));
1171 if (!SendMail(kPPEMail))
1172 Log("SHUTTLE", Form("ContinueProcessing - Could not send mail to %s expert",
1173 fCurrentDetector.Data()));
1176 Log("SHUTTLE", Form("ContinueProcessing - %s: restarting. "
1177 "Aborted before with %s. Retry number %d.", fCurrentDetector.Data(),
1178 status->GetStatusName(), status->GetCount()));
1179 Bool_t increaseCount = kTRUE;
1180 if (status->GetStatus() == AliShuttleStatus::kDCSError ||
1181 status->GetStatus() == AliShuttleStatus::kDCSStarted ||
1182 status->GetStatus() == AliShuttleStatus::kFXSError)
1183 increaseCount = kFALSE;
1185 UpdateShuttleStatus(AliShuttleStatus::kStarted, increaseCount);
1192 //______________________________________________________________________________________________
1193 void AliShuttle::SendMLRunInfo(const char* status)
1196 // Send information about this run to ML
1198 TMonaLisaText mlStatus("SHUTTLE_status", status);
1199 TString runType(fLogbookEntry->GetRunType());
1200 if (strlen(fLogbookEntry->GetRunParameter("log")) > 0){
1203 runType += fLogbookEntry->GetRunParameter("log");
1206 if (fLogbookEntry->GetDATestMode()){
1207 runType += " (DATest)";
1209 TMonaLisaText mlRunType("SHUTTLE_runtype", runType);
1212 mlList.Add(&mlStatus);
1213 mlList.Add(&mlRunType);
1216 mlID.Form("%d", GetCurrentRun());
1217 fMonaLisa->SendParameters(&mlList, mlID);
1220 //______________________________________________________________________________________________
1221 Int_t AliShuttle::GetMem(Int_t pid)
1223 // invokes ps to get the memory consumption of the process <pid>
1224 // returns -1 in case of error
1227 checkStr.Form("ps -o vsize --pid %d | tail -n 1", pid);
1228 FILE* pipe = gSystem->OpenPipe(checkStr, "r");
1231 Log("SHUTTLE", Form("Process - Error: "
1232 "Could not open pipe to %s", checkStr.Data()));
1237 if (!fgets(buffer, 100, pipe))
1239 Log("SHUTTLE", "Process - Error: ps did not return anything");
1240 gSystem->ClosePipe(pipe);
1243 gSystem->ClosePipe(pipe);
1245 //Log("SHUTTLE", Form("ps returned %s", buffer));
1248 if ((sscanf(buffer, "%d\n", &mem) != 1) || !mem)
1250 Log("SHUTTLE", "Process - Error: Could not parse output of ps");
1257 //______________________________________________________________________________________________
1258 Bool_t AliShuttle::Process(AliShuttleLogbookEntry* entry)
1261 // Makes data retrieval for all detectors in the configuration.
1262 // entry: Shuttle logbook entry, contains run paramenters and status of detectors
1263 // (Unprocessed, Inactive, Failed or Done).
1264 // Returns kFALSE in case of error occured and kTRUE otherwise
1267 if (!entry) return kFALSE;
1269 fLogbookEntry = entry;
1271 Log("SHUTTLE", Form("\t\t\t^*^*^*^*^*^*^*^*^*^*^*^* run %d: START ^*^*^*^*^*^*^*^*^*^*^*^*",
1276 // Send the information to ML
1277 SendMLRunInfo("Processing");
1279 if (fLogbookEntry->IsDone())
1281 Log("SHUTTLE","Process - Shuttle is already DONE. Updating logbook");
1282 UpdateShuttleLogbook("shuttle_done");
1287 // read test mode if flag is set
1291 TString logEntry(entry->GetRunParameter("log"));
1292 //printf("log entry = %s\n", logEntry.Data());
1293 TString searchStr("Testmode: ");
1294 Int_t pos = logEntry.Index(searchStr.Data());
1295 //printf("%d\n", pos);
1298 TSubString subStr = logEntry(pos + searchStr.Length(), logEntry.Length());
1299 //printf("%s\n", subStr.String().Data());
1300 TString newStr(subStr.Data());
1301 TObjArray* token = newStr.Tokenize(' ');
1305 TObjString* tmpStr = dynamic_cast<TObjString*> (token->First());
1308 Int_t testMode = tmpStr->String().Atoi();
1311 Log("SHUTTLE", Form("Process - Enabling test mode %d", testMode));
1312 SetTestMode((TestMode) testMode);
1320 fLogbookEntry->Print("all");
1323 Bool_t hasError = kFALSE;
1325 // Set the CDB and Reference folders according to the year
1327 // build cdb paths (repeat each time, run might be a DATest run)
1328 if (!fLogbookEntry->GetDATestMode()){
1329 fgkMainCDB.Form("alien://folder=%s%d/OCDB?user=alidaq?cacheFold=/tmp/OCDBCache",
1330 fConfig->GetAlienPath(), GetCurrentYear());
1332 fgkMainRefStorage.Form("alien://folder=%s%d/Reference?user=alidaq?cacheFold=/tmp/OCDBCache",
1333 fConfig->GetAlienPath(), GetCurrentYear());
1336 fgkMainCDB.Form("alien://folder=%s%d/DATest/OCDB?user=alidaq?cacheFold=/tmp/OCDBCache",
1337 fConfig->GetAlienPath(), GetCurrentYear());
1339 fgkMainRefStorage.Form("alien://folder=%s%d/DATest/Reference?user=alidaq?cacheFold=/tmp/OCDBCache",
1340 fConfig->GetAlienPath(), GetCurrentYear());
1343 AliDebug(2,Form("Main CDB storage = %s",fgkMainCDB.Data()));
1344 AliDebug(2,Form("Main Reference storage = %s",fgkMainRefStorage.Data()));
1346 // Loop on detectors in the configuration
1347 TIter iter(fConfig->GetDetectors());
1348 TObjString* aDetector = 0;
1350 Bool_t first = kTRUE;
1352 while ((aDetector = (TObjString*) iter.Next()))
1354 fCurrentDetector = aDetector->String();
1356 if (ContinueProcessing() == kFALSE)
1361 // only read QueryCDB when needed and only once
1362 AliCDBStorage *mainCDBSto = AliCDBManager::Instance()->GetStorage(fgkMainCDB);
1363 if(mainCDBSto) mainCDBSto->QueryCDB(GetCurrentRun());
1364 AliCDBStorage *mainRefSto = AliCDBManager::Instance()->GetStorage(fgkMainRefStorage);
1365 if(mainRefSto) mainRefSto->QueryCDB(GetCurrentRun());
1369 Log("SHUTTLE", Form("\t\t\t****** run %d - %s: START ******",
1370 GetCurrentRun(), aDetector->GetName()));
1372 for(Int_t iSys=0;iSys<3;iSys++) fFXSCalled[iSys]=kFALSE;
1374 Int_t initialMem = GetMem(getpid());
1375 Log("SHUTTLE", Form("Memory consumption before forking is %d", initialMem));
1377 Log(fCurrentDetector.Data(), "Process - Starting processing");
1383 Log("SHUTTLE", "Process - ERROR: Forking failed");
1388 Log("SHUTTLE", Form("Process - In parent process of %d - %s: Starting monitoring",
1389 GetCurrentRun(), aDetector->GetName()));
1391 Long_t begin = time(0);
1393 int status; // to be used with waitpid, on purpose an int (not Int_t)!
1394 while (waitpid(pid, &status, WNOHANG) == 0)
1396 Long_t expiredTime = time(0) - begin;
1398 if (expiredTime > fConfig->GetPPTimeOut())
1401 AliShuttleStatus *currentStatus = ReadShuttleStatus();
1402 AliShuttleStatus::Status newStatus = AliShuttleStatus::kInvalid;
1404 if (currentStatus->GetStatus() <= AliShuttleStatus::kPPDone)
1406 // in case pp not yet done set status to kPPTimeOut
1408 logMsg.Form("Process - Process of %s timed out. Run time: %d seconds. Killing...",
1409 fCurrentDetector.Data(), expiredTime);
1410 newStatus = AliShuttleStatus::kPPTimeOut;
1412 else if (currentStatus->GetStatus() == AliShuttleStatus::kStoreStarted)
1414 // in case the pp goes in TimeOut while storing the objects in the OCDB
1415 // set status to kStoreError
1417 logMsg.Form("Process - Process of %s timed out while storing the OCDB object. Run time: %d seconds. Killing... and setting status to StoreError.",
1418 fCurrentDetector.Data(), expiredTime);
1419 newStatus = AliShuttleStatus::kStoreError;
1423 // in other cases don't change the status
1425 logMsg.Form("Process - Process of %s timed out in status = %s. Run time: %d seconds. Killing... without changing the status",
1426 fCurrentDetector.Data(), currentStatus->GetStatusName(), expiredTime);
1429 Log("SHUTTLE", logMsg);
1430 Log(fCurrentDetector, logMsg);
1434 if (newStatus != AliShuttleStatus::kInvalid)
1435 UpdateShuttleStatus(newStatus);
1438 gSystem->Sleep(1000);
1442 gSystem->Sleep(1000);
1444 Int_t mem = GetMem(pid);
1453 if (expiredTime % 60 == 0)
1455 Log("SHUTTLE", Form("Process - %s: Checking process. "
1456 "Run time: %d seconds - Memory consumption: %d KB",
1457 fCurrentDetector.Data(), expiredTime, mem));
1461 if (mem > fConfig->GetPPMaxMem())
1464 tmp.Form("Process - Process exceeds maximum allowed memory "
1465 "(%d KB > %d KB). Killing...",
1466 mem, fConfig->GetPPMaxMem());
1467 Log("SHUTTLE", tmp);
1468 Log(fCurrentDetector, tmp);
1472 UpdateShuttleStatus(AliShuttleStatus::kPPOutOfMemory);
1475 gSystem->Sleep(1000);
1480 Log("SHUTTLE", Form("Process - In parent process of %d - %s: Client has terminated.",
1481 GetCurrentRun(), aDetector->GetName()));
1483 if (WIFEXITED(status))
1485 Int_t returnCode = WEXITSTATUS(status);
1487 Log("SHUTTLE", Form("Process - %s: the return code is %d", fCurrentDetector.Data(),
1490 if (returnCode == 0) hasError = kTRUE;
1496 Log("SHUTTLE", Form("Process - In child process of %d - %s", GetCurrentRun(),
1497 aDetector->GetName()));
1499 Log("SHUTTLE", Form("Process - Redirecting output to %s log",fCurrentDetector.Data()));
1501 if ((freopen(GetLogFileName(fCurrentDetector), "a", stdout)) == 0)
1503 Log("SHUTTLE", "Process - Could not freopen stdout");
1507 fOutputRedirected = kTRUE;
1508 if ((dup2(fileno(stdout), fileno(stderr))) < 0)
1509 Log("SHUTTLE", "Process - Could not redirect stderr");
1513 TString wd = gSystem->WorkingDirectory();
1514 Int_t dir_lev1 = GetCurrentRun()/10000;
1515 TString tmpDir = Form("%s/%d/%d/%s_process", GetShuttleTempDir(),
1516 dir_lev1, GetCurrentRun(), fCurrentDetector.Data());
1518 Int_t result = gSystem->GetPathInfo(tmpDir.Data(), 0, (Long64_t*) 0, 0, 0);
1519 if (!result) // temp dir already exists!
1521 Log(fCurrentDetector.Data(),
1522 Form("Process - %s dir already exists! Removing...", tmpDir.Data()));
1523 gSystem->Exec(Form("rm -rf %s",tmpDir.Data()));
1526 if (gSystem->mkdir(tmpDir.Data(), 1))
1528 Log(fCurrentDetector.Data(), "Process - could not make temp directory!!");
1532 if (!gSystem->ChangeDirectory(tmpDir.Data()))
1534 Log(fCurrentDetector.Data(), "Process - could not change directory!!");
1538 Int_t success = ProcessCurrentDetector();
1540 gSystem->ChangeDirectory(wd.Data());
1542 if (success == 1) // Preprocessor finished successfully!
1544 // remove temporary folder or DCS map
1545 if (!fConfig->KeepTempFolder())
1547 gSystem->Exec(Form("rm -rf %s",tmpDir.Data()));
1548 } else if (!fConfig->KeepDCSMap())
1550 gSystem->Exec(Form("rm -f %s/DCSMap.root",tmpDir.Data()));
1553 // Update time_processed field in FXS DB
1554 if (UpdateTable() == kFALSE)
1555 Log("SHUTTLE", Form("Process - %s: Could not update FXS databases!",
1556 fCurrentDetector.Data()));
1558 // Transfer the data from local storage to main storage (Grid)
1559 if (StoreOCDB() == kFALSE)
1562 else if (success == 0)
1565 Form("\t\t\t****** run %d - %s: PP ERROR ******",
1566 GetCurrentRun(), aDetector->GetName()));
1569 for (UInt_t iSys=0; iSys<3; iSys++)
1571 if (fFXSCalled[iSys]) fFXSlist[iSys].Clear();
1574 Log("SHUTTLE", Form("Process - Client process of %d - %s is exiting now with %d.",
1575 GetCurrentRun(), aDetector->GetName(), success));
1577 // the client exits here
1578 gSystem->Exit(success);
1580 AliError("We should never get here!!!");
1584 Log("SHUTTLE", Form("\t\t\t^*^*^*^*^*^*^*^*^*^*^*^* run %d: FINISH ^*^*^*^*^*^*^*^*^*^*^*^*",
1587 //check if shuttle is done for this run, if so update logbook
1588 TObjArray checkEntryArray;
1589 checkEntryArray.SetOwner(1);
1590 TString whereClause = Form("where run=%d", GetCurrentRun());
1591 if (!QueryShuttleLogbook(whereClause.Data(), checkEntryArray) ||
1592 checkEntryArray.GetEntries() == 0) {
1593 Log("SHUTTLE", Form("Process - Warning: Cannot check status of run %d on Shuttle logbook!",
1595 return hasError == kFALSE;
1598 AliShuttleLogbookEntry* checkEntry = dynamic_cast<AliShuttleLogbookEntry*>
1599 (checkEntryArray.At(0));
1603 if (checkEntry->IsDone())
1605 Log("SHUTTLE","Process - Shuttle is DONE. Updating logbook");
1606 UpdateShuttleLogbook("shuttle_done");
1610 for (UInt_t iDet=0; iDet<NDetectors(); iDet++)
1612 if (checkEntry->GetDetectorStatus(iDet) == AliShuttleLogbookEntry::kUnprocessed)
1614 AliDebug(2, Form("Run %d: setting %s as \"not first time unprocessed\"",
1615 checkEntry->GetRun(), GetDetName(iDet)));
1616 fFirstUnprocessed[iDet] = kFALSE;
1619 SendMLRunInfo("Pending");
1625 return hasError == kFALSE;
1628 //______________________________________________________________________________________________
1629 Int_t AliShuttle::ProcessCurrentDetector()
1632 // Makes data retrieval just for a specific detector (fCurrentDetector).
1633 // Threre should be a configuration for this detector.
1635 Log("SHUTTLE", Form("ProcessCurrentDetector - Retrieving values for %s, run %d",
1636 fCurrentDetector.Data(), GetCurrentRun()));
1638 TString wd = gSystem->WorkingDirectory();
1640 if (!CleanReferenceStorage(fCurrentDetector.Data()))
1643 gSystem->ChangeDirectory(wd.Data());
1645 // call preprocessor
1646 AliPreprocessor* aPreprocessor =
1647 dynamic_cast<AliPreprocessor*> (fPreprocessorMap.GetValue(fCurrentDetector));
1649 // check if the preprocessor wants to process this run type
1650 if (aPreprocessor->ProcessRunType() == kFALSE)
1652 UpdateShuttleStatus(AliShuttleStatus::kSkipped);
1653 UpdateShuttleLogbook(fCurrentDetector, "DONE");
1654 if (!UpdateTableSkippedCase(fCurrentDetector.Data()))
1656 AliError(Form("Could not update FXS tables for run %d !", GetCurrentRun()));
1658 Log(fCurrentDetector, Form("ProcessCurrentDetector - %s preprocessor is not interested in this run type", fCurrentDetector.Data()));
1663 TMap* dcsMap = new TMap();
1665 aPreprocessor->Initialize(GetCurrentRun(), GetCurrentStartTime(), GetCurrentEndTime());
1667 Bool_t processDCS = aPreprocessor->ProcessDCS();
1671 Log(fCurrentDetector, "ProcessCurrentDetector -"
1672 " The preprocessor requested to skip the retrieval of DCS values");
1674 else if (fTestMode & kSkipDCS)
1676 Log(fCurrentDetector, "ProcessCurrentDetector - In TESTMODE: Skipping DCS processing");
1678 else if (fTestMode & kErrorDCS)
1680 Log(fCurrentDetector, "ProcessCurrentDetector - In TESTMODE: Simulating DCS error");
1681 UpdateShuttleStatus(AliShuttleStatus::kDCSStarted);
1682 UpdateShuttleStatus(AliShuttleStatus::kDCSError);
1687 UpdateShuttleStatus(AliShuttleStatus::kDCSStarted);
1689 // Query DCS archive
1690 Int_t nServers = fConfig->GetNServers(fCurrentDetector);
1692 for (int iServ=0; iServ<nServers; iServ++)
1695 TString host(fConfig->GetDCSHost(fCurrentDetector, iServ));
1696 Int_t port = fConfig->GetDCSPort(fCurrentDetector, iServ);
1697 Int_t multiSplit = fConfig->GetMultiSplit(fCurrentDetector, iServ);
1699 Log(fCurrentDetector, Form("ProcessCurrentDetector -"
1700 " Querying DCS Amanda server %s:%d (%d of %d)",
1701 host.Data(), port, iServ+1, nServers));
1706 if (fConfig->GetDCSAliases(fCurrentDetector, iServ)->GetEntries() > 0)
1708 Log(fCurrentDetector, Form("Querying %d DCS aliases", fConfig->GetDCSAliases(fCurrentDetector, iServ)->GetEntries()));
1709 aliasMap = GetValueSet(host, port,
1710 fConfig->GetDCSAliases(fCurrentDetector, iServ),
1711 kAlias, multiSplit);
1714 Log(fCurrentDetector,
1715 Form("ProcessCurrentDetector -"
1716 " Error retrieving DCS aliases from server %s."
1717 " Sending mail to DCS experts!", host.Data()));
1718 UpdateShuttleStatus(AliShuttleStatus::kDCSError);
1720 if (!SendMail(kDCSEMail))
1721 Log("SHUTTLE", Form("ProcessCurrentDetector - "
1722 "Could not send mail to DCS experts!"));
1729 if (fConfig->GetDCSDataPoints(fCurrentDetector, iServ)->GetEntries() > 0)
1731 Log(fCurrentDetector, Form("Querying %d DCS data points", fConfig->GetDCSDataPoints(fCurrentDetector, iServ)->GetEntries()));
1732 dpMap = GetValueSet(host, port,
1733 fConfig->GetDCSDataPoints(fCurrentDetector, iServ),
1737 Log(fCurrentDetector,
1738 Form("ProcessCurrentDetector -"
1739 " Error retrieving DCS data points from server %s."
1740 " Sending mail to DCS experts!", host.Data()));
1741 UpdateShuttleStatus(AliShuttleStatus::kDCSError);
1743 if (!SendMail(kDCSEMail))
1744 Log("SHUTTLE", Form("ProcessCurrentDetector - "
1745 "Could not send mail to DCS experts!"));
1747 if (aliasMap) delete aliasMap;
1753 // merge aliasMap and dpMap into dcsMap
1755 TIter iter(aliasMap);
1756 TObjString* key = 0;
1757 while ((key = (TObjString*) iter.Next()))
1758 dcsMap->Add(key, aliasMap->GetValue(key->String()));
1760 aliasMap->SetOwner(kFALSE);
1766 TObjString* key = 0;
1767 while ((key = (TObjString*) iter.Next()))
1768 dcsMap->Add(key, dpMap->GetValue(key->String()));
1770 dpMap->SetOwner(kFALSE);
1776 // save map into file, to help debugging in case of preprocessor error
1777 TFile* f = TFile::Open("DCSMap.root","recreate");
1779 dcsMap->Write("DCSMap", TObject::kSingleKey);
1783 // DCS Archive DB processing successful. Call Preprocessor!
1784 UpdateShuttleStatus(AliShuttleStatus::kPPStarted);
1786 fFXSError = -1; // this variable is kTRUE after ::Process if an FXS error occured
1788 UInt_t returnValue = aPreprocessor->Process(dcsMap);
1790 if (fFXSError!=-1) {
1791 UpdateShuttleStatus(AliShuttleStatus::kFXSError);
1792 SendMail(kFXSEMail, fFXSError);
1793 dcsMap->DeleteAll();
1798 if (returnValue > 0) // Preprocessor error!
1800 Log(fCurrentDetector, Form("ProcessCurrentDetector - "
1801 "Preprocessor failed. Process returned %d.", returnValue));
1802 UpdateShuttleStatus(AliShuttleStatus::kPPError);
1803 dcsMap->DeleteAll();
1809 UpdateShuttleStatus(AliShuttleStatus::kPPDone);
1810 Log(fCurrentDetector, Form("ProcessCurrentDetector - %s preprocessor returned success",
1811 fCurrentDetector.Data()));
1813 dcsMap->DeleteAll();
1819 //______________________________________________________________________________________________
1820 void AliShuttle::CountOpenRuns()
1822 // Query DAQ's Shuttle logbook and sends the number of open runs to ML
1826 // check connection, in case connect
1831 sqlQuery = Form("select count(*) from %s where shuttle_done=0", fConfig->GetShuttlelbTable());
1833 TSQLResult* aResult = fServer[3]->Query(sqlQuery);
1835 AliError(Form("Can't execute query <%s>!", sqlQuery.Data()));
1839 AliDebug(2,Form("Query = %s", sqlQuery.Data()));
1841 if (aResult->GetRowCount() == 0) {
1842 AliError(Form("No result for query %s received", sqlQuery.Data()));
1846 if (aResult->GetFieldCount() != 1) {
1847 AliError(Form("Invalid field count for query %s received", sqlQuery.Data()));
1851 TSQLRow* aRow = aResult->Next();
1853 AliError(Form("Could not receive result of query %s", sqlQuery.Data()));
1857 TString result(aRow->GetField(0), aRow->GetFieldLength(0));
1858 Int_t count = result.Atoi();
1860 Log("SHUTTLE", Form("%d unprocessed runs", count));
1865 TMonaLisaValue mlStatus("SHUTTLE_openruns", count);
1868 mlList.Add(&mlStatus);
1870 fMonaLisa->SendParameters(&mlList, "__PROCESSINGINFO__");
1873 //______________________________________________________________________________________________
1874 Bool_t AliShuttle::QueryShuttleLogbook(const char* whereClause,
1877 // Query DAQ's Shuttle logbook and fills detector status object.
1878 // Call QueryRunParameters to query DAQ logbook for run parameters.
1881 entries.SetOwner(1);
1883 // check connection, in case connect
1884 if (!Connect(3)) return kFALSE;
1887 sqlQuery = Form("select * from %s %s order by run", fConfig->GetShuttlelbTable(), whereClause);
1889 TSQLResult* aResult = fServer[3]->Query(sqlQuery);
1891 AliError(Form("Can't execute query <%s>!", sqlQuery.Data()));
1895 AliDebug(2,Form("Query = %s", sqlQuery.Data()));
1897 if(aResult->GetRowCount() == 0) {
1898 Log("SHUTTLE", "No entries in Shuttle Logbook match request");
1903 // TODO Check field count!
1904 const UInt_t nCols = 25;
1905 if (aResult->GetFieldCount() != (Int_t) nCols) {
1906 Log("SHUTTLE", "Invalid SQL result field number!");
1912 while ((aRow = aResult->Next())) {
1913 TString runString(aRow->GetField(0), aRow->GetFieldLength(0));
1914 Int_t run = runString.Atoi();
1916 AliShuttleLogbookEntry *entry = QueryRunParameters(run);
1920 // DA test mode flag
1921 TString daTestModeString(aRow->GetField(2), aRow->GetFieldLength(2)); // field 2 = DA test mode flag
1922 Bool_t daTestMode = (Bool_t)daTestModeString.Atoi();
1923 entry->SetDATestMode(daTestMode);
1925 // loop on detectors
1926 for(UInt_t ii = 0; ii < nCols; ii++)
1927 entry->SetDetectorStatus(aResult->GetFieldName(ii), aRow->GetField(ii));
1929 entries.AddLast(entry);
1937 //______________________________________________________________________________________________
1938 AliShuttleLogbookEntry* AliShuttle::QueryRunParameters(Int_t run)
1941 // Retrieve run parameters written in the DAQ logbook and sets them into AliShuttleLogbookEntry object
1944 // check connection, in case connect
1949 sqlQuery.Form("select * from %s where run=%d", fConfig->GetDAQlbTable(), run);
1951 TSQLResult* aResult = fServer[3]->Query(sqlQuery);
1953 Log("SHUTTLE", Form("Can't execute query <%s>!", sqlQuery.Data()));
1957 if (aResult->GetRowCount() == 0) {
1958 Log("SHUTTLE", Form("QueryRunParameters - No entry in DAQ Logbook for run %d. Skipping", run));
1963 if (aResult->GetRowCount() > 1) {
1964 Log("SHUTTLE", Form("QueryRunParameters - UNEXPECTED: "
1965 "more than one entry in DAQ Logbook for run %d!", run));
1970 TSQLRow* aRow = aResult->Next();
1973 Log("SHUTTLE", Form("QueryRunParameters - Could not retrieve row for run %d. Skipping", run));
1978 AliShuttleLogbookEntry* entry = new AliShuttleLogbookEntry(run);
1980 for (Int_t ii = 0; ii < aResult->GetFieldCount(); ii++)
1981 entry->SetRunParameter(aResult->GetFieldName(ii), aRow->GetField(ii));
1986 UInt_t startTime = entry->GetStartTime();
1987 UInt_t endTime = entry->GetEndTime();
1988 Bool_t ecsSuccess = entry->GetECSSuccess();
1990 TString totEventsStr = entry->GetRunParameter("totalEvents");
1991 Int_t totEvents = totEventsStr.Atoi();
1993 UInt_t now = time(0);
1994 // TODO make this a configuration parameter
1995 Int_t dcsDelay = fConfig->GetDCSDelay()+fConfig->GetDCSQueryOffset();
1997 // runs are accepted if they have ecsSuccess set or more than 1 event
1998 if (startTime != 0 && endTime != 0 && endTime > startTime && (totEvents > 1 || ecsSuccess) && (endTime < now - dcsDelay))
2000 if (ecsSuccess == kFALSE)
2001 Log("SHUTTLE", Form("Processing run %d although in status ECS failure, Reason: %s", run, entry->GetRunParameter("eor_reason")));
2005 Bool_t skip = kFALSE;
2007 if (endTime != 0 && endTime >= now - dcsDelay)
2009 Log("SHUTTLE", Form("Skipping run %d for now, because DCS buffer time is not yet expired", run));
2011 else if (totEvents <= 1)
2013 Log("SHUTTLE", Form("QueryRunParameters - Run %d has 1 event or less - Skipping!", run));
2018 Log("SHUTTLE", Form("QueryRunParameters - Invalid parameters for Run %d: "
2019 "startTime = %d, endTime = %d. Skipping (Shuttle won't be marked as DONE)!",
2020 run, startTime, endTime));
2025 Log("SHUTTLE", Form("Marking SHUTTLE skipped for run %d", run));
2026 fLogbookEntry = entry;
2027 if (!UpdateShuttleLogbook("shuttle_skipped"))
2029 AliError(Form("Could not update logbook for run %d !", run));
2031 if (!UpdateTableSkippedCase("ALL"))
2033 AliError(Form("Could not update FXS tables for run %d !", run));
2042 //______________________________________________________________________________________________
2043 TMap* AliShuttle::GetValueSet(const char* host, Int_t port, const TSeqCollection* entries,
2044 DCSType type, Int_t multiSplit)
2046 // Retrieve all "entry" data points from the DCS server
2047 // host, port: TSocket connection parameters
2048 // entries: list of name of the alias or data point
2049 // type: kAlias or kDP
2050 // returns TMap of values, 0 when failure
2052 AliDCSClient client(host, port, fTimeout, fRetries, multiSplit);
2057 //result = client.GetAliasValues(entries, GetCurrentStartTime()-offset,
2058 // GetCurrentEndTime()+offset);
2059 result = client.GetAliasValues(entries, GetStartTimeDCSQuery(),
2060 GetEndTimeDCSQuery());
2062 else if (type == kDP)
2064 //result = client.GetDPValues(entries, GetCurrentStartTime()-offset,
2065 // GetCurrentEndTime()+offset);
2066 result = client.GetDPValues(entries, GetStartTimeDCSQuery(),
2067 GetEndTimeDCSQuery());
2072 Log(fCurrentDetector.Data(), Form("GetValueSet - Can't get entries! Reason: %s",
2073 client.GetErrorString(client.GetResultErrorCode())));
2074 if (client.GetResultErrorCode() == AliDCSClient::fgkServerError)
2075 Log(fCurrentDetector.Data(), Form("GetValueSet - Server error code: %s",
2076 client.GetServerError().Data()));
2084 //______________________________________________________________________________________________
2085 const char* AliShuttle::GetFile(Int_t system, const char* detector,
2086 const char* id, const char* source)
2088 // Get calibration file from file exchange servers
2089 // First queris the FXS database for the file name, using the run, detector, id and source info
2090 // then calls RetrieveFile(filename) for actual copy to local disk
2091 // run: current run being processed (given by Logbook entry fLogbookEntry)
2092 // detector: the Preprocessor name
2093 // id: provided as a parameter by the Preprocessor
2094 // source: provided by the Preprocessor through GetFileSources function
2096 // check if test mode should simulate a FXS error
2097 if (fTestMode & kErrorFXSFiles)
2099 Log(detector, Form("GetFile - In TESTMODE - Simulating error while connecting to %s FXS", GetSystemName(system)));
2103 // check connection, in case connect
2104 if (!Connect(system))
2106 Log(detector, Form("GetFile - Couldn't connect to %s FXS database", GetSystemName(system)));
2111 // Query preparation
2112 TString sourceName(source);
2114 TString sqlQueryStart = Form("select filePath,size,fileChecksum from %s where",
2115 fConfig->GetFXSdbTable(system));
2116 TString whereClause = Form("run=%d and detector=\"%s\" and fileId=\"%s\"",
2117 GetCurrentRun(), detector, id);
2121 whereClause += Form(" and DAQsource=\"%s\"", source);
2123 else if (system == kDCS)
2127 else if (system == kHLT)
2129 whereClause += Form(" and DDLnumbers=\"%s\"", source);
2132 TString sqlQuery = Form("%s %s", sqlQueryStart.Data(), whereClause.Data());
2134 AliDebug(2, Form("SQL query: \n%s",sqlQuery.Data()));
2137 TSQLResult* aResult = 0;
2138 aResult = dynamic_cast<TSQLResult*> (fServer[system]->Query(sqlQuery));
2140 Log(detector, Form("GetFile - Can't execute SQL query to %s database for: id = %s, source = %s",
2141 GetSystemName(system), id, sourceName.Data()));
2146 if (aResult->GetRowCount() == 0)
2149 Form("GetFile - No entry in %s FXS db for: id = %s, source = %s",
2150 GetSystemName(system), id, sourceName.Data()));
2155 if (aResult->GetRowCount() > 1) {
2157 Form("GetFile - More than one entry in %s FXS db for: id = %s, source = %s",
2158 GetSystemName(system), id, sourceName.Data()));
2164 if (aResult->GetFieldCount() != nFields) {
2166 Form("GetFileName - Wrong field count in %s FXS db for: id = %s, source = %s",
2167 GetSystemName(system), id, sourceName.Data()));
2173 TSQLRow* aRow = dynamic_cast<TSQLRow*> (aResult->Next());
2176 Log(detector, Form("GetFile - Empty set result in %s FXS db from query: id = %s, source = %s",
2177 GetSystemName(system), id, sourceName.Data()));
2183 TString filePath(aRow->GetField(0), aRow->GetFieldLength(0));
2184 TString fileSize(aRow->GetField(1), aRow->GetFieldLength(1));
2185 TString fileChecksum(aRow->GetField(2), aRow->GetFieldLength(2));
2190 AliDebug(2, Form("filePath = %s; size = %s, fileChecksum = %s",
2191 filePath.Data(), fileSize.Data(), fileChecksum.Data()));
2193 // retrieved file is renamed to make it unique
2194 Int_t dir_lev1 = GetCurrentRun()/10000;
2195 TString localFileName = Form("%s/%d/%d/%s_process/%s_%s_%d_%s_%s.shuttle",
2196 GetShuttleTempDir(), dir_lev1, GetCurrentRun(), detector,
2197 GetSystemName(system), detector, GetCurrentRun(),
2198 id, sourceName.Data());
2199 Log("SHUTTLE",Form("file from FXS = %s",localFileName.Data()));
2202 // file retrieval from FXS
2203 UInt_t nRetries = 0;
2204 UInt_t maxRetries = 3;
2205 Bool_t result = kFALSE;
2207 // copy!! if successful TSystem::Exec returns 0
2208 while (nRetries++ < maxRetries) {
2209 AliDebug(2, Form("Trying to copy file. Retry # %d", nRetries));
2210 result = RetrieveFile(system, filePath.Data(), localFileName.Data());
2213 Log(detector, Form("GetFile - Copy of file %s from %s FXS failed",
2214 filePath.Data(), GetSystemName(system)));
2218 if (fileSize.Length()>0)
2220 // compare filesize of local file with the one stored in the FXS DB
2222 Int_t sizeComp = gSystem->GetPathInfo(localFileName.Data(), 0, &size, 0, 0);
2224 if (sizeComp != 0 || size != fileSize.Atoi())
2226 Log(detector, Form("GetFile - size of file %s does not match with local copy!",
2233 Log(fCurrentDetector, Form("GetFile - size of file %s not set in %s database, skipping comparison",
2234 filePath.Data(), GetSystemName(system)));
2237 if (fileChecksum.Length()>0)
2239 // compare md5sum of local file with the one stored in the FXS DB
2240 if(fileChecksum.Contains(' ')) fileChecksum.Resize(fileChecksum.First(' '));
2241 Int_t md5Comp = gSystem->Exec(Form("md5sum %s |grep %s > /dev/null 2> /dev/null",
2242 localFileName.Data(), fileChecksum.Data()));
2246 Log(detector, Form("GetFile - md5sum of file %s does not match with local copy!",
2252 Log(fCurrentDetector, Form("GetFile - md5sum of file %s not set in %s database, skipping comparison",
2253 filePath.Data(), GetSystemName(system)));
2264 fFXSCalled[system]=kTRUE;
2265 TObjString *fileParams = new TObjString(Form("%s#!?!#%s", id, sourceName.Data()));
2266 fFXSlist[system].Add(fileParams);
2268 static TString staticLocalFileName;
2269 staticLocalFileName.Form("%s", localFileName.Data());
2271 Log(fCurrentDetector, Form("GetFile - Retrieved file with id %s and "
2272 "source %s from %s to %s", id, source,
2273 GetSystemName(system), localFileName.Data()));
2275 return staticLocalFileName.Data();
2278 //______________________________________________________________________________________________
2279 Bool_t AliShuttle::RetrieveFile(UInt_t system, const char* fxsFileName, const char* localFileName)
2282 // Copies file from FXS to local Shuttle machine
2285 // check temp directory: trying to cd to temp; if it does not exist, create it
2286 AliDebug(2, Form("Copy file %s from %s FXS into %s",
2287 GetSystemName(system), fxsFileName, localFileName));
2289 TString tmpDir(localFileName);
2291 tmpDir = tmpDir(0,tmpDir.Last('/'));
2293 Int_t noDir = gSystem->GetPathInfo(tmpDir.Data(), 0, (Long64_t*) 0, 0, 0);
2294 if (noDir) // temp dir does not exists!
2296 if (gSystem->mkdir(tmpDir.Data(), 1))
2298 Log(fCurrentDetector.Data(), "RetrieveFile - could not make temp directory!!");
2303 TString command = Form("scp -oPort=%d -2 %s@%s:%s/%s %s",
2304 fConfig->GetFXSPort(system),
2305 fConfig->GetFXSUser(system),
2306 fConfig->GetFXSHost(system),
2307 fConfig->GetFXSBaseFolder(system),
2311 AliDebug(2, Form("%s",command.Data()));
2313 Bool_t result = (gSystem->Exec(command.Data()) == 0);
2318 //______________________________________________________________________________________________
2319 TList* AliShuttle::GetFileSources(Int_t system, const char* detector, const char* id)
2322 // Get sources producing the condition file Id from file exchange servers
2323 // if id is NULL all sources are returned (distinct)
2328 Log(detector, Form("GetFileSources - Querying %s FXS for files with id %s produced by %s", GetSystemName(system), id, detector));
2330 Log(detector, Form("GetFileSources - Querying %s FXS for files produced by %s", GetSystemName(system), detector));
2333 // check if test mode should simulate a FXS error
2334 if (fTestMode & kErrorFXSSources)
2336 Log(detector, Form("GetFileSources - In TESTMODE - Simulating error while connecting to %s FXS", GetSystemName(system)));
2342 Log(detector, "GetFileSources - WARNING: DCS system has only one source of data!");
2343 TList *list = new TList();
2345 list->Add(new TObjString(" "));
2349 // check connection, in case connect
2350 if (!Connect(system))
2352 Log(detector, Form("GetFileSources - Couldn't connect to %s FXS database", GetSystemName(system)));
2357 TString sourceName = 0;
2360 sourceName = "DAQsource";
2361 } else if (system == kHLT)
2363 sourceName = "DDLnumbers";
2366 TString sqlQueryStart = Form("select distinct %s from %s where", sourceName.Data(), fConfig->GetFXSdbTable(system));
2367 TString whereClause = Form("run=%d and detector=\"%s\"",
2368 GetCurrentRun(), detector);
2370 whereClause += Form(" and fileId=\"%s\"", id);
2371 TString sqlQuery = Form("%s %s", sqlQueryStart.Data(), whereClause.Data());
2373 AliDebug(2, Form("SQL query: \n%s",sqlQuery.Data()));
2376 TSQLResult* aResult;
2377 aResult = fServer[system]->Query(sqlQuery);
2379 Log(detector, Form("GetFileSources - Can't execute SQL query to %s database for id: %s",
2380 GetSystemName(system), id));
2385 TList *list = new TList();
2388 if (aResult->GetRowCount() == 0)
2391 Form("GetFileSources - No entry in %s FXS table for id: %s", GetSystemName(system), id));
2396 Log(detector, Form("GetFileSources - Found %d sources", aResult->GetRowCount()));
2399 while ((aRow = aResult->Next()))
2402 TString source(aRow->GetField(0), aRow->GetFieldLength(0));
2403 AliDebug(2, Form("%s = %s", sourceName.Data(), source.Data()));
2404 list->Add(new TObjString(source));
2413 //______________________________________________________________________________________________
2414 TList* AliShuttle::GetFileIDs(Int_t system, const char* detector, const char* source)
2417 // Get all ids of condition files produced by a given source from file exchange servers
2420 Log(detector, Form("GetFileIDs - Retrieving ids with source %s with %s", source, GetSystemName(system)));
2422 // check if test mode should simulate a FXS error
2423 if (fTestMode & kErrorFXSSources)
2425 Log(detector, Form("GetFileIDs - In TESTMODE - Simulating error while connecting to %s FXS", GetSystemName(system)));
2429 // check connection, in case connect
2430 if (!Connect(system))
2432 Log(detector, Form("GetFileIDs - Couldn't connect to %s FXS database", GetSystemName(system)));
2436 TString sourceName = 0;
2439 sourceName = "DAQsource";
2440 } else if (system == kHLT)
2442 sourceName = "DDLnumbers";
2445 TString sqlQueryStart = Form("select fileId from %s where", fConfig->GetFXSdbTable(system));
2446 TString whereClause = Form("run=%d and detector=\"%s\"",
2447 GetCurrentRun(), detector);
2448 if (sourceName.Length() > 0 && source)
2449 whereClause += Form(" and %s=\"%s\"", sourceName.Data(), source);
2450 TString sqlQuery = Form("%s %s", sqlQueryStart.Data(), whereClause.Data());
2452 AliDebug(2, Form("SQL query: \n%s",sqlQuery.Data()));
2455 TSQLResult* aResult;
2456 aResult = fServer[system]->Query(sqlQuery);
2458 Log(detector, Form("GetFileIDs - Can't execute SQL query to %s database for source: %s",
2459 GetSystemName(system), source));
2463 TList *list = new TList();
2466 if (aResult->GetRowCount() == 0)
2469 Form("GetFileIDs - No entry in %s FXS table for source: %s", GetSystemName(system), source));
2474 Log(detector, Form("GetFileIDs - Found %d ids", aResult->GetRowCount()));
2478 while ((aRow = aResult->Next()))
2481 TString id(aRow->GetField(0), aRow->GetFieldLength(0));
2482 AliDebug(2, Form("fileId = %s", id.Data()));
2483 list->Add(new TObjString(id));
2492 //______________________________________________________________________________________________
2493 Bool_t AliShuttle::Connect(Int_t system)
2495 // Connect to MySQL Server of the system's FXS MySQL databases
2496 // DAQ Logbook, Shuttle Logbook and DAQ FXS db are on the same host
2499 // check connection: if already connected return
2500 if(fServer[system] && fServer[system]->IsConnected()) {
2501 // ping the server --> automatic reconnection should occur if it was broken but the
2502 // server is still alive
2503 fServer[system]->Ping();
2507 TString dbHost, dbUser, dbPass, dbName;
2509 if (system < 3) // FXS db servers
2511 dbHost = Form("mysql://%s:%d", fConfig->GetFXSdbHost(system), fConfig->GetFXSdbPort(system));
2512 dbUser = fConfig->GetFXSdbUser(system);
2513 dbPass = fConfig->GetFXSdbPass(system);
2514 dbName = fConfig->GetFXSdbName(system);
2515 } else { // Run & Shuttle logbook servers
2516 // TODO Will the Shuttle logbook server be the same as the Run logbook server ???
2517 dbHost = Form("mysql://%s:%d", fConfig->GetDAQlbHost(), fConfig->GetDAQlbPort());
2518 dbUser = fConfig->GetDAQlbUser();
2519 dbPass = fConfig->GetDAQlbPass();
2520 dbName = fConfig->GetDAQlbDB();
2523 fServer[system] = TSQLServer::Connect(dbHost.Data(), dbUser.Data(), dbPass.Data());
2524 if (!fServer[system] || !fServer[system]->IsConnected()) {
2527 AliError(Form("Can't establish connection to FXS database for %s",
2528 AliShuttleInterface::GetSystemName(system)));
2530 AliError("Can't establish connection to Run logbook.");
2532 if(fServer[system]) delete fServer[system];
2537 TSQLResult* aResult=0;
2540 aResult = fServer[kDAQ]->GetTables(dbName.Data());
2543 aResult = fServer[kDCS]->GetTables(dbName.Data());
2546 aResult = fServer[kHLT]->GetTables(dbName.Data());
2549 aResult = fServer[3]->GetTables(dbName.Data());
2557 //______________________________________________________________________________________________
2558 Bool_t AliShuttle::UpdateTable()
2561 // Update FXS table filling time_processed field in all rows corresponding to current run and detector
2564 Bool_t result = kTRUE;
2566 for (UInt_t system=0; system<3; system++)
2568 if(!fFXSCalled[system]) continue;
2570 // check connection, in case connect
2571 if (!Connect(system))
2573 Log(fCurrentDetector, Form("UpdateTable - Couldn't connect to %s FXS database", GetSystemName(system)));
2578 TTimeStamp now; // now
2580 // Loop on FXS list entries
2581 TIter iter(&fFXSlist[system]);
2582 TObjString *aFXSentry=0;
2583 while ((aFXSentry = dynamic_cast<TObjString*> (iter.Next())))
2585 TString aFXSentrystr = aFXSentry->String();
2586 TObjArray *aFXSarray = aFXSentrystr.Tokenize("#!?!#");
2587 if (!aFXSarray || aFXSarray->GetEntries() != 2 )
2589 Log(fCurrentDetector, Form("UpdateTable - error updating %s FXS entry. Check string: <%s>",
2590 GetSystemName(system), aFXSentrystr.Data()));
2591 if(aFXSarray) delete aFXSarray;
2595 const char* fileId = ((TObjString*) aFXSarray->At(0))->GetName();
2596 const char* source = ((TObjString*) aFXSarray->At(1))->GetName();
2598 TString whereClause;
2601 whereClause = Form("where run=%d and detector=\"%s\" and fileId=\"%s\" and DAQsource=\"%s\";",
2602 GetCurrentRun(), fCurrentDetector.Data(), fileId, source);
2604 else if (system == kDCS)
2606 whereClause = Form("where run=%d and detector=\"%s\" and fileId=\"%s\";",
2607 GetCurrentRun(), fCurrentDetector.Data(), fileId);
2609 else if (system == kHLT)
2611 whereClause = Form("where run=%d and detector=\"%s\" and fileId=\"%s\" and DDLnumbers=\"%s\";",
2612 GetCurrentRun(), fCurrentDetector.Data(), fileId, source);
2617 TString sqlQuery = Form("update %s set time_processed=%d %s", fConfig->GetFXSdbTable(system),
2618 now.GetSec(), whereClause.Data());
2620 AliDebug(2, Form("SQL query: \n%s",sqlQuery.Data()));
2623 TSQLResult* aResult;
2624 aResult = dynamic_cast<TSQLResult*> (fServer[system]->Query(sqlQuery));
2627 Log(fCurrentDetector, Form("UpdateTable - %s db: can't execute SQL query <%s>",
2628 GetSystemName(system), sqlQuery.Data()));
2639 //_______________________________________________________________________________
2640 Bool_t AliShuttle::UpdateTableSkippedCase(const char* detector)
2643 // Update FXS table filling time_processed field in all rows corresponding to current run and detector
2644 // if detector = "ALL" update all detectors
2647 Bool_t result = kTRUE;
2649 TString detName(detector);
2651 for (UInt_t system=0; system<3; system++)
2654 // check connection, in case connect
2655 if (!Connect(system))
2657 Log(fCurrentDetector, Form("UpdateTableSkippedCase - Couldn't connect to %s FXS database", GetSystemName(system)));
2662 TTimeStamp now; // now
2664 // Loop on FXS list entries
2665 TIter iter(&fFXSlist[system]);
2667 TString whereClause;
2668 if (detName == "ALL") whereClause = Form("where run=%d and time_processed IS NULL;",GetCurrentRun());
2669 else whereClause = Form("where run=%d and detector=\"%s\" and time_processed IS NULL;",GetCurrentRun(), detector);
2671 Log("SHUTTLE",Form(" whereClause = %s ",whereClause.Data()));
2673 TString sqlQuery = Form("update %s set time_processed=%d %s", fConfig->GetFXSdbTable(system),
2674 now.GetSec(), whereClause.Data());
2676 AliDebug(2, Form("SQL query: \n%s",sqlQuery.Data()));
2679 TSQLResult* aResult;
2680 aResult = dynamic_cast<TSQLResult*> (fServer[system]->Query(sqlQuery));
2683 Log("SHUTTLE", Form("UpdateTableSkippedCase - %s db: can't execute SQL query <%s>",
2684 GetSystemName(system), sqlQuery.Data()));
2694 //______________________________________________________________________________________________
2695 Bool_t AliShuttle::UpdateTableFailCase()
2697 // Update FXS table filling time_processed field in all rows corresponding to current run and detector
2698 // this is called in case the preprocessor is declared failed for the current run, because
2699 // the fields are updated only in case of success
2701 Bool_t result = kTRUE;
2703 for (UInt_t system=0; system<3; system++)
2705 // check connection, in case connect
2706 if (!Connect(system))
2708 Log(fCurrentDetector, Form("UpdateTableFailCase - Couldn't connect to %s FXS database",
2709 GetSystemName(system)));
2714 TTimeStamp now; // now
2716 // Loop on FXS list entries
2718 TString whereClause = Form("where run=%d and detector=\"%s\";",
2719 GetCurrentRun(), fCurrentDetector.Data());
2722 TString sqlQuery = Form("update %s set time_processed=%d %s", fConfig->GetFXSdbTable(system),
2723 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("UpdateTableFailCase - %s db: can't execute SQL query <%s>",
2733 GetSystemName(system), sqlQuery.Data()));
2743 //______________________________________________________________________________________________
2744 Bool_t AliShuttle::UpdateShuttleLogbook(const char* detector, const char* status)
2747 // Update Shuttle logbook filling detector or shuttle_done column
2748 // ex. of usage: UpdateShuttleLogbook("PHOS", "DONE") or UpdateShuttleLogbook("shuttle_done")
2751 // check connection, in case connect
2753 Log("SHUTTLE", "UpdateShuttleLogbook - Couldn't connect to DAQ Logbook.");
2757 TString detName(detector);
2759 if (detName == "shuttle_done" || detName == "shuttle_skipped")
2761 setClause = "set shuttle_done=1";
2763 if (detName == "shuttle_done")
2765 if (TouchFile() != kTRUE)
2767 SendMLRunInfo("Pending");
2771 SendMLRunInfo("Done");
2774 SendMLRunInfo("Skipped");
2777 TString statusStr(status);
2778 if(statusStr.Contains("done", TString::kIgnoreCase) ||
2779 statusStr.Contains("failed", TString::kIgnoreCase)){
2780 setClause = Form("set %s=\"%s\"", detector, status);
2783 Form("UpdateShuttleLogbook - Invalid status <%s> for detector %s",
2789 TString whereClause = Form("where run=%d", GetCurrentRun());
2791 TString sqlQuery = Form("update %s %s %s",
2792 fConfig->GetShuttlelbTable(), setClause.Data(), whereClause.Data());
2794 AliDebug(2, Form("SQL query: \n%s",sqlQuery.Data()));
2797 TSQLResult* aResult;
2798 aResult = dynamic_cast<TSQLResult*> (fServer[3]->Query(sqlQuery));
2800 Log("SHUTTLE", Form("UpdateShuttleLogbook - Can't execute query <%s>", sqlQuery.Data()));
2808 //______________________________________________________________________________________________
2809 Int_t AliShuttle::GetCurrentRun() const
2812 // Get current run from logbook entry
2815 return fLogbookEntry ? fLogbookEntry->GetRun() : -1;
2818 //______________________________________________________________________________________________
2819 UInt_t AliShuttle::GetCurrentStartTime() const
2822 // get current start time
2825 return fLogbookEntry ? fLogbookEntry->GetStartTime() : 0;
2828 //______________________________________________________________________________________________
2829 UInt_t AliShuttle::GetCurrentEndTime() const
2832 // get current end time from logbook entry
2835 return fLogbookEntry ? fLogbookEntry->GetEndTime() : 0;
2837 //______________________________________________________________________________________________
2838 UInt_t AliShuttle::GetCurrentYear() const
2841 // Get current year from logbook entry
2844 if (!fLogbookEntry) return 0;
2846 TTimeStamp startTime(GetCurrentStartTime());
2847 TString year = Form("%d",startTime.GetDate());
2853 //______________________________________________________________________________________________
2854 const char* AliShuttle::GetLHCPeriod() const
2857 // Get current LHC period from logbook entry
2860 if (!fLogbookEntry) return 0;
2862 return fLogbookEntry->GetRunParameter("LHCperiod");
2865 //______________________________________________________________________________________________
2866 void AliShuttle::Log(const char* detector, const char* message)
2869 // Fill log string with a message
2872 TString logRunDir = GetShuttleLogDir();
2873 if (GetCurrentRun() >=0) {
2874 Int_t logDir_lev1 = GetCurrentRun()/10000;
2875 logRunDir += Form("/%d/%d", logDir_lev1, GetCurrentRun());
2877 void* dir = gSystem->OpenDirectory(logRunDir.Data());
2879 if (gSystem->mkdir(logRunDir.Data(), kTRUE)) {
2880 AliError(Form("Can't open directory <%s>", GetShuttleLogDir()));
2885 gSystem->FreeDirectory(dir);
2888 TString toLog = Form("%s UTC (%d): %s - ", TTimeStamp(time(0)).AsString("s"), getpid(), detector);
2889 if (GetCurrentRun() >= 0)
2890 toLog += Form("run %d - ", GetCurrentRun());
2891 toLog += Form("%s", message);
2893 AliInfo(toLog.Data());
2895 // if we redirect the log output already to the file, leave here
2896 if (fOutputRedirected && strcmp(detector, "SHUTTLE") != 0)
2899 TString fileName = GetLogFileName(detector);
2901 gSystem->ExpandPathName(fileName);
2904 logFile.open(fileName, ofstream::out | ofstream::app);
2906 if (!logFile.is_open()) {
2907 AliError(Form("Could not open file %s", fileName.Data()));
2911 logFile << toLog.Data() << "\n";
2916 //______________________________________________________________________________________________
2917 TString AliShuttle::GetLogFileName(const char* detector) const
2920 // returns the name of the log file for a given sub detector
2925 if (GetCurrentRun() >= 0)
2927 Int_t logDir_lev1 = GetCurrentRun()/10000;
2928 fileName.Form("%s/%d/%d/%s.log", GetShuttleLogDir(), logDir_lev1, GetCurrentRun(),
2931 fileName.Form("%s/%s.log", GetShuttleLogDir(), detector);
2937 //______________________________________________________________________________________________
2938 void AliShuttle::SendAlive()
2940 // sends alive message to ML
2942 TMonaLisaText mlStatus("SHUTTLE_status", "Alive");
2945 mlList.Add(&mlStatus);
2947 fMonaLisa->SendParameters(&mlList, "__PROCESSINGINFO__");
2950 //______________________________________________________________________________________________
2951 Bool_t AliShuttle::Collect(Int_t run)
2954 // Collects conditions data for all UNPROCESSED run written to DAQ LogBook in case of run = -1 (default)
2955 // If a dedicated run is given this run is processed
2957 // In operational mode, this is the Shuttle function triggered by the EOR signal.
2961 Log("SHUTTLE","Collect - Shuttle called. Collecting conditions data for unprocessed runs");
2963 Log("SHUTTLE", Form("Collect - Shuttle called. Collecting conditions data for run %d", run));
2965 SetLastAction("Starting");
2967 // create ML instance
2969 fMonaLisa = new TMonaLisaWriter(fConfig->GetMonitorHost(), fConfig->GetMonitorTable());
2973 TString whereClause("where shuttle_done=0");
2975 whereClause += Form(" and run=%d", run);
2977 TObjArray shuttleLogbookEntries;
2978 if (!QueryShuttleLogbook(whereClause, shuttleLogbookEntries))
2980 Log("SHUTTLE", "Collect - Can't retrieve entries from Shuttle logbook");
2984 if (shuttleLogbookEntries.GetEntries() == 0)
2987 Log("SHUTTLE","Collect - Found no UNPROCESSED runs in Shuttle logbook");
2989 Log("SHUTTLE", Form("Collect - Run %d is already DONE "
2990 "or it does not exist in Shuttle logbook", run));
2994 for (UInt_t iDet=0; iDet<NDetectors(); iDet++)
2995 fFirstUnprocessed[iDet] = kTRUE;
2999 // query Shuttle logbook for earlier runs, check if some detectors are unprocessed,
3000 // flag them into fFirstUnprocessed array
3001 TString whereClauseBis(Form("where shuttle_done=0 and run < %d", run));
3002 TObjArray tmpLogbookEntries;
3003 if (!QueryShuttleLogbook(whereClauseBis, tmpLogbookEntries))
3005 Log("SHUTTLE", "Collect - Can't retrieve entries from Shuttle logbook");
3009 TIter iter(&tmpLogbookEntries);
3010 AliShuttleLogbookEntry* anEntry = 0;
3011 while ((anEntry = dynamic_cast<AliShuttleLogbookEntry*> (iter.Next())))
3013 for (UInt_t iDet=0; iDet<NDetectors(); iDet++)
3015 if (anEntry->GetDetectorStatus(iDet) == AliShuttleLogbookEntry::kUnprocessed)
3017 AliDebug(2, Form("Run %d: setting %s as \"not first time unprocessed\"",
3018 anEntry->GetRun(), GetDetName(iDet)));
3019 fFirstUnprocessed[iDet] = kFALSE;
3027 if (!RetrieveConditionsData(shuttleLogbookEntries))
3029 Log("SHUTTLE", "Collect - Process of at least one run failed");
3034 Log("SHUTTLE", "Collect - Requested run(s) successfully processed");
3039 //______________________________________________________________________________________________
3040 Bool_t AliShuttle::RetrieveConditionsData(const TObjArray& dateEntries)
3043 // Retrieve conditions data for all runs that aren't processed yet
3046 Bool_t hasError = kFALSE;
3048 TIter iter(&dateEntries);
3049 AliShuttleLogbookEntry* anEntry;
3051 while ((anEntry = (AliShuttleLogbookEntry*) iter.Next())){
3052 if (!Process(anEntry)){
3056 // clean SHUTTLE temp directory
3057 //TString filename = Form("%s/*.shuttle", GetShuttleTempDir());
3058 //RemoveFile(filename.Data());
3061 return hasError == kFALSE;
3064 //______________________________________________________________________________________________
3065 ULong_t AliShuttle::GetTimeOfLastAction() const
3068 // Gets time of last action
3073 fMonitoringMutex->Lock();
3075 tmp = fLastActionTime;
3077 fMonitoringMutex->UnLock();
3082 //______________________________________________________________________________________________
3083 const TString AliShuttle::GetLastAction() const
3086 // returns a string description of the last action
3091 fMonitoringMutex->Lock();
3095 fMonitoringMutex->UnLock();
3100 //______________________________________________________________________________________________
3101 void AliShuttle::SetLastAction(const char* action)
3104 // updates the monitoring variables
3107 fMonitoringMutex->Lock();
3109 fLastAction = action;
3110 fLastActionTime = time(0);
3112 fMonitoringMutex->UnLock();
3115 //______________________________________________________________________________________________
3116 const char* AliShuttle::GetRunParameter(const char* param)
3119 // returns run parameter read from DAQ logbook
3122 if(!fLogbookEntry) {
3123 AliError("No logbook entry!");
3127 return fLogbookEntry->GetRunParameter(param);
3130 //______________________________________________________________________________________________
3131 AliCDBEntry* AliShuttle::GetFromOCDB(const char* detector, const AliCDBPath& path)
3134 // returns object from OCDB valid for current run
3137 if (fTestMode & kErrorOCDB)
3139 Log(detector, "GetFromOCDB - In TESTMODE - Simulating error with OCDB");
3143 AliCDBStorage *sto = AliCDBManager::Instance()->GetStorage(fgkMainCDB);
3146 Log(detector, "GetFromOCDB - Cannot activate main OCDB for query!");
3150 return dynamic_cast<AliCDBEntry*> (sto->Get(path, GetCurrentRun()));
3153 //______________________________________________________________________________________________
3154 Bool_t AliShuttle::SendMail(EMailTarget target, Int_t system)
3157 // sends a mail to the subdetector expert in case of preprocessor error
3160 if (fTestMode != kNone)
3163 if (!fConfig->SendMail())
3166 if (target == kDCSEMail || target == kFXSEMail) {
3167 if (!fFirstProcessing)
3171 Int_t runMode = (Int_t)fConfig->GetRunMode();
3173 if (runMode == 0) tmpStr = " Nightly Test:";
3174 else tmpStr = " Data Taking:";
3175 void* dir = gSystem->OpenDirectory(GetShuttleLogDir());
3178 if (gSystem->mkdir(GetShuttleLogDir(), kTRUE))
3180 Log("SHUTTLE", Form("SendMail - Can't open directory <%s>", GetShuttleLogDir()));
3185 gSystem->FreeDirectory(dir);
3188 // det experts in to
3190 TIter *iterExperts = 0;
3191 if (target == kDCSEMail) {
3192 iterExperts = new TIter(fConfig->GetAdmins(AliShuttleConfig::kAmanda));
3194 else if (target == kFXSEMail) {
3195 iterExperts = new TIter(fConfig->GetAdmins(system));
3198 TObjString *anExpert=0;
3199 while ((anExpert = (TObjString*) iterExperts->Next()))
3201 to += Form("%s,", anExpert->GetName());
3206 // add subdetector experts
3207 iterExperts = new TIter(fConfig->GetResponsibles(fCurrentDetector));
3208 TObjString *anExpert=0;
3209 while ((anExpert = (TObjString*) iterExperts->Next()))
3211 to += Form("%s,", anExpert->GetName());
3215 if (to.Length() > 0)
3216 to.Remove(to.Length()-1);
3217 AliDebug(2, Form("to: %s",to.Data()));
3220 Log("SHUTTLE", Form("List of %d responsibles not set!", (Int_t) target));
3224 // SHUTTLE responsibles in cc
3226 TIter iterAdmins(fConfig->GetAdmins(AliShuttleConfig::kGlobal));
3227 TObjString *anAdmin=0;
3228 while ((anAdmin = (TObjString*) iterAdmins.Next()))
3230 cc += Form("%s,", anAdmin->GetName());
3232 if (cc.Length() > 0)
3233 cc.Remove(cc.Length()-1);
3234 AliDebug(2, Form("cc: %s",to.Data()));
3237 TString bodyFileName;
3238 bodyFileName.Form("%s/mail.body", GetShuttleLogDir());
3239 gSystem->ExpandPathName(bodyFileName);
3242 mailBody.open(bodyFileName, ofstream::out);
3244 if (!mailBody.is_open())
3246 Log("SHUTTLE", Form("Could not open mail body file %s", bodyFileName.Data()));
3254 if (target == kDCSEMail){
3255 subject = Form("%s CRITICAL Retrieval of data points for %s FAILED in run %d !",
3256 tmpStr.Data(), fCurrentDetector.Data(), GetCurrentRun());
3257 AliDebug(2, Form("subject: %s", subject.Data()));
3259 body = Form("Dear DCS experts, \n\n");
3260 body += Form("SHUTTLE couldn\'t retrieve the data points for detector %s "
3261 "in run %d!!\n\n", fCurrentDetector.Data(), GetCurrentRun());
3263 else if (target == kFXSEMail){
3264 subject = Form("%s CRITICAL FXS communication for %s FAILED in run %d !",
3265 tmpStr.Data(), fCurrentDetector.Data(), GetCurrentRun());
3266 AliDebug(2, Form("subject: %s", subject.Data()));
3268 if (system == kDAQ) sys="DAQ";
3269 else if (system == kDCS) sys="DCS";
3270 else if (system == kHLT) sys="HLT";
3272 body = Form("Dear %s FXS experts, \n\n",sys.Data());
3273 body += Form("SHUTTLE couldn\'t retrieve data from the FXS for detector %s "
3274 "in run %d!!\n\n", fCurrentDetector.Data(), GetCurrentRun());
3277 subject = Form("%s %s Shuttle preprocessor FAILED in run %d (run type = %s)!",
3278 tmpStr.Data(), fCurrentDetector.Data(), GetCurrentRun(), GetRunType());
3279 AliDebug(2, Form("subject: %s", subject.Data()));
3281 body = Form("Dear %s expert(s), \n\n", fCurrentDetector.Data());
3282 body += Form("SHUTTLE just detected that your preprocessor "
3283 "failed processing run %d (run type = %s)!!\n\n",
3284 GetCurrentRun(), GetRunType());
3287 body += Form("Please check %s status on the SHUTTLE monitoring page: \n\n",
3288 fCurrentDetector.Data());
3289 if (fConfig->GetRunMode() == AliShuttleConfig::kTest)
3291 body += Form("\thttp://pcalimonitor.cern.ch/shuttle.jsp?time=24 \n\n");
3293 body += Form("\thttp://pcalimonitor.cern.ch/shuttle.jsp?instance=PROD&time=24 \n\n");
3297 TString logFolder = "logs";
3298 if (fConfig->GetRunMode() == AliShuttleConfig::kProd)
3299 logFolder += "_PROD";
3302 body += Form("Find the %s log for the current run on \n\n"
3303 "\thttp://pcalishuttle01.cern.ch:8880/%s/%d/%d/%s.log \n\n",
3304 fCurrentDetector.Data(), logFolder.Data(), GetCurrentRun()/10000,
3305 GetCurrentRun(), fCurrentDetector.Data());
3306 body += Form("The last 10 lines of %s log file are following:\n\n", fCurrentDetector.Data());
3308 AliDebug(2, Form("Body begin: %s", body.Data()));
3310 mailBody << body.Data();
3312 mailBody.open(bodyFileName, ofstream::out | ofstream::app);
3314 TString logFileName = Form("%s/%d/%d/%s.log", GetShuttleLogDir(),
3315 GetCurrentRun()/10000, GetCurrentRun(), fCurrentDetector.Data());
3316 TString tailCommand = Form("tail -n 10 %s >> %s", logFileName.Data(), bodyFileName.Data());
3317 if (gSystem->Exec(tailCommand.Data()))
3319 mailBody << Form("%s log file not found ...\n\n", fCurrentDetector.Data());
3322 TString endBody = Form("------------------------------------------------------\n\n");
3323 endBody += Form("In case of problems please contact the SHUTTLE core team.\n\n");
3324 endBody += "Please do not answer this message directly, it is automatically generated.\n\n";
3325 endBody += "Greetings,\n\n \t\t\tthe SHUTTLE\n";
3327 AliDebug(2, Form("Body end: %s", endBody.Data()));
3329 mailBody << endBody.Data();
3334 TString mailCommand = Form("mail -s \"%s\" -c %s %s < %s",
3338 bodyFileName.Data());
3339 AliDebug(2, Form("mail command: %s", mailCommand.Data()));
3341 Bool_t result = gSystem->Exec(mailCommand.Data());
3345 //______________________________________________________________________________________________
3346 const char* AliShuttle::GetRunType()
3349 // returns run type read from "run type" logbook
3352 if(!fLogbookEntry) {
3353 AliError("No logbook entry!");
3357 return fLogbookEntry->GetRunType();
3360 //______________________________________________________________________________________________
3361 Bool_t AliShuttle::GetHLTStatus()
3363 // Return HLT status (ON=1 OFF=0)
3364 // Converts the HLT status from the mode string read in the run logbook (not just a bool)
3366 if(!fLogbookEntry) {
3367 AliError("No logbook entry!");
3371 // TODO implement when HLTMode is inserted in run logbook
3372 TString hltMode = fLogbookEntry->GetRunParameter("HLTmode");
3373 TSubString firstChar = hltMode(0,1);
3374 AliDebug(2,Form("First char = %s ",firstChar.Data()));
3375 if (firstChar == "A") {
3378 else if ((firstChar == "B") || (firstChar == "C") || (firstChar == "D") || (firstChar == "E")) {
3382 Log("SHUTTLE","Unexpected HLT mode! Returning 0....");
3387 //______________________________________________________________________________________________
3388 const char* AliShuttle::GetTriggerConfiguration()
3390 // Receives the trigger configuration from the DAQ logbook for the current run
3392 // check connection, if needed reconnect
3397 sqlQuery.Form("SELECT configFile FROM logbook_trigger_config WHERE run = %d", GetCurrentRun());
3398 TSQLResult* result = fServer[3]->Query(sqlQuery);
3401 Log("SHUTTLE", Form("ERROR: Can't execute query <%s>!", sqlQuery.Data()));
3405 if (result->GetRowCount() == 0)
3407 Log("SHUTTLE", "ERROR: Trigger configuration not found in logbook_trigger_config");
3412 TSQLRow* row = result->Next();
3415 Log("SHUTTLE", "ERROR: Could not receive logbook_trigger_config data");
3420 // static, so that pointer remains valid when it is returned to the calling class
3421 static TString triggerConfig(row->GetField(0));
3429 Log("SHUTTLE", Form("Found trigger configuration: %s", triggerConfig.Data()));
3431 return triggerConfig;
3434 //______________________________________________________________________________________________
3435 void AliShuttle::SetShuttleTempDir(const char* tmpDir)
3438 // sets Shuttle temp directory
3441 fgkShuttleTempDir = gSystem->ExpandPathName(tmpDir);
3444 //______________________________________________________________________________________________
3445 void AliShuttle::SetShuttleLogDir(const char* logDir)
3448 // sets Shuttle log directory
3451 fgkShuttleLogDir = gSystem->ExpandPathName(logDir);
3453 //______________________________________________________________________________________________
3454 Bool_t AliShuttle::TouchFile()
3457 // touching a file on the grid if run has been DONE
3462 Log("SHUTTLE",Form("No TGrid connection estabilished!"));
3463 Log("SHUTTLE",Form("Could not touch file for run %i",GetCurrentRun()));
3468 dir.Form("%s%d/SHUTTLE_DONE", fConfig->GetAlienPath(), GetCurrentYear());
3469 // checking whether directory for touch command exists
3471 commandLs.Form("ls %s",dir.Data());
3472 TGridResult *resultLs = dynamic_cast<TGridResult*>(gGrid->Command(commandLs));
3474 Log("SHUTTLE",Form("No result for %s command, returning without touching",commandLs.Data()));
3477 TMap *mapLs = dynamic_cast<TMap*>(resultLs->At(0));
3479 Log("SHUTTLE",Form("No map for %s command, returning without touching",commandLs.Data()));
3484 TObjString *valueLsPath = dynamic_cast<TObjString*>(mapLs->GetValue("path"));
3485 if (!valueLsPath || (TString)(valueLsPath->GetString()).CompareTo(dir)!=1){
3486 Log("SHUTTLE",Form("No directory %s found, creating it",dir.Data()));
3488 // creating the directory
3490 Bool_t boolMkdir = gGrid->Mkdir(dir.Data());
3492 Log("SHUTTLE",Form("Impossible to create dir %s in alien catalogue for run %i!",dir.Data(),GetCurrentRun()));
3497 Log("SHUTTLE",Form("Directory %s successfully created in alien catalogue for run %i",dir.Data(),GetCurrentRun()));
3500 Log("SHUTTLE",Form("Directory %s correctly found for run %i",dir.Data(),GetCurrentRun()));
3507 command.Form("touch %s/%i", dir.Data(), GetCurrentRun());
3508 Log("SHUTTLE", Form("Creating entry in file catalog: %s", command.Data()));
3509 TGridResult *resultTouch = dynamic_cast<TGridResult*>(gGrid->Command(command));
3511 Log("SHUTTLE",Form("No result for touching command, returning without touching for run %i",GetCurrentRun()));
3514 TMap *mapTouch = dynamic_cast<TMap*>(resultTouch->At(0));
3516 Log("SHUTTLE",Form("No map for touching command, returning without touching for run %i",GetCurrentRun()));
3521 TObjString *valueTouch = dynamic_cast<TObjString*>(mapTouch->GetValue("__result__"));
3523 Log("SHUTTLE",Form("No value for \"__result__\" key set in the map for touching command, returning without touching for run %i",GetCurrentRun()));
3528 if (valueTouch->GetString()!="1"){
3529 Log("SHUTTLE",Form("Failing the touching command, returning without touching for run %i",GetCurrentRun()));
3536 Log("SHUTTLE", "Sucessfully touched the file");
3539 //______________________________________________________________________________________________
3540 UInt_t AliShuttle::GetStartTimeDCSQuery()
3542 // Return Start Time for the DCS query
3544 // The call is delegated to AliShuttleInterface
3546 return GetCurrentStartTime()-fConfig->GetDCSQueryOffset();
3548 //______________________________________________________________________________________________
3549 UInt_t AliShuttle::GetEndTimeDCSQuery()
3551 // Return End Time for the DCS query
3553 // The call is delegated to AliShuttleInterface
3555 return GetCurrentEndTime()+fConfig->GetDCSQueryOffset();