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>
61 #include <TMonaLisaWriter.h>
65 #include <sys/types.h>
72 //______________________________________________________________________________________________
73 AliShuttle::AliShuttle(const AliShuttleConfig* config,
74 UInt_t timeout, Int_t retries):
76 fTimeout(timeout), fRetries(retries),
88 fReadTestMode(kFALSE),
89 fOutputRedirected(kFALSE)
92 // config: AliShuttleConfig used
93 // timeout: timeout used for AliDCSClient connection
94 // retries: the number of retries in case of connection error.
97 if (!fConfig->IsValid()) AliFatal("********** !!!!! Invalid configuration !!!!! **********");
98 for(int iSys=0;iSys<4;iSys++) {
101 fFXSlist[iSys].SetOwner(kTRUE);
103 fPreprocessorMap.SetOwner(kTRUE);
105 for (UInt_t iDet=0; iDet<NDetectors(); iDet++)
106 fFirstUnprocessed[iDet] = kFALSE;
108 fMonitoringMutex = new TMutex();
111 //______________________________________________________________________________________________
112 AliShuttle::~AliShuttle()
118 fPreprocessorMap.DeleteAll();
119 for(int iSys=0;iSys<4;iSys++)
121 fServer[iSys]->Close();
122 delete fServer[iSys];
131 if (fMonitoringMutex)
133 delete fMonitoringMutex;
134 fMonitoringMutex = 0;
138 //______________________________________________________________________________________________
139 void AliShuttle::RegisterPreprocessor(AliPreprocessor* preprocessor)
142 // Registers new AliPreprocessor.
143 // It uses GetName() for indentificator of the pre processor.
144 // The pre processor is registered it there isn't any other
145 // with the same identificator (GetName()).
148 const char* detName = preprocessor->GetName();
149 if(GetDetPos(detName) < 0)
150 AliFatal(Form("********** !!!!! Invalid detector name: %s !!!!! **********", detName));
152 if (fPreprocessorMap.GetValue(detName)) {
153 AliWarning(Form("AliPreprocessor %s is already registered!", detName));
157 fPreprocessorMap.Add(new TObjString(detName), preprocessor);
159 //______________________________________________________________________________________________
160 Bool_t AliShuttle::Store(const AliCDBPath& path, TObject* object,
161 AliCDBMetaData* metaData, Int_t validityStart, Bool_t validityInfinite)
163 // Stores a CDB object in the storage for offline reconstruction. Objects that are not needed for
164 // offline reconstruction, but should be stored anyway (e.g. for debugging) should NOT be stored
165 // using this function. Use StoreReferenceData instead!
166 // It calls StoreLocally function which temporarily stores the data locally; when the preprocessor
167 // finishes the data are transferred to the main storage (Grid).
169 return StoreLocally(fgkLocalCDB, path, object, metaData, validityStart, validityInfinite);
172 //______________________________________________________________________________________________
173 Bool_t AliShuttle::StoreReferenceData(const AliCDBPath& path, TObject* object, AliCDBMetaData* metaData)
175 // Stores a CDB object in the storage for reference data. This objects will not be available during
176 // offline reconstrunction. Use this function for reference data only!
177 // It calls StoreLocally function which temporarily stores the data locally; when the preprocessor
178 // finishes the data are transferred to the main storage (Grid).
180 return StoreLocally(fgkLocalRefStorage, path, object, metaData);
183 //______________________________________________________________________________________________
184 Bool_t AliShuttle::StoreLocally(const TString& localUri,
185 const AliCDBPath& path, TObject* object, AliCDBMetaData* metaData,
186 Int_t validityStart, Bool_t validityInfinite)
188 // Store object temporarily in local storage. Parameters are passed by Store and StoreReferenceData functions.
189 // when the preprocessor finishes the data are transferred to the main storage (Grid).
190 // The parameters are:
191 // 1) Uri of the backup storage (Local)
192 // 2) the object's path.
193 // 3) the object to be stored
194 // 4) the metaData to be associated with the object
195 // 5) the validity start run number w.r.t. the current run,
196 // if the data is valid only for this run leave the default 0
197 // 6) specifies if the calibration data is valid for infinity (this means until updated),
198 // typical for calibration runs, the default is kFALSE
200 // returns 0 if fail, 1 otherwise
202 if (fTestMode & kErrorStorage)
204 Log(fCurrentDetector, "StoreLocally - In TESTMODE - Simulating error while storing locally");
208 const char* cdbType = (localUri == fgkLocalCDB) ? "CDB" : "Reference";
210 Int_t firstRun = GetCurrentRun() - validityStart;
212 AliWarning("First valid run happens to be less than 0! Setting it to 0.");
217 if(validityInfinite) {
218 lastRun = AliCDBRunRange::Infinity();
220 lastRun = GetCurrentRun();
223 // Version is set to current run, it will be used later to transfer data to Grid
224 AliCDBId id(path, firstRun, lastRun, GetCurrentRun(), -1);
226 if(! dynamic_cast<TObjString*> (metaData->GetProperty("RunUsed(TObjString)"))){
227 TObjString runUsed = Form("%d", GetCurrentRun());
228 metaData->SetProperty("RunUsed(TObjString)", runUsed.Clone());
231 Bool_t result = kFALSE;
233 if (!(AliCDBManager::Instance()->GetStorage(localUri))) {
234 Log("SHUTTLE", Form("StoreLocally - Cannot activate local %s storage", cdbType));
236 result = AliCDBManager::Instance()->GetStorage(localUri)
237 ->Put(object, id, metaData);
242 Log(fCurrentDetector, Form("StoreLocally - Can't store object <%s>!", id.ToString().Data()));
248 //______________________________________________________________________________________________
249 Bool_t AliShuttle::StoreOCDB()
252 // Called when preprocessor ends successfully or when previous storage attempt failed (kStoreError status)
253 // Calls underlying StoreOCDB(const char*) function twice, for OCDB and Reference storage.
254 // Then calls StoreRefFilesToGrid to store reference files.
257 UpdateShuttleStatus(AliShuttleStatus::kStoreStarted);
259 if (fTestMode & kErrorGrid)
261 Log("SHUTTLE", "StoreOCDB - In TESTMODE - Simulating error while storing in the Grid");
262 Log(fCurrentDetector, "StoreOCDB - In TESTMODE - Simulating error while storing in the Grid");
266 Log("SHUTTLE","StoreOCDB - Storing OCDB data ...");
267 Int_t resultCDB = StoreOCDB(fgkMainCDB);
269 Log("SHUTTLE","StoreOCDB - Storing reference data ...");
270 Int_t resultRef = StoreOCDB(fgkMainRefStorage);
272 Log("SHUTTLE","StoreOCDB - Storing reference files ...");
273 Bool_t resultRefFiles = CopyFilesToGrid("reference");
275 Bool_t resultMetadata = kTRUE;
276 if(fCurrentDetector == "GRP")
278 Log("StoreOCDB - SHUTTLE","Storing Run Metadata file ...");
279 resultMetadata = CopyFilesToGrid("metadata");
282 Int_t storeResult = 0;
284 if (resultCDB < 0 || resultRef < 0 || resultRefFiles == kFALSE || resultMetadata == kFALSE)
286 else if (resultCDB > 0 || resultRef > 0)
292 Form("\t\t\t****** run %d - %s: STORAGE ERROR ******",
293 GetCurrentRun(), fCurrentDetector.Data()));
294 UpdateShuttleStatus(AliShuttleStatus::kStoreError);
296 else if (storeResult > 0)
299 Form("\t\t\t****** run %d - %s: STORAGE DELAYED ******",
300 GetCurrentRun(), fCurrentDetector.Data()));
301 UpdateShuttleStatus(AliShuttleStatus::kStoreDelayed);
303 else if (storeResult == 0)
306 Form("\t\t\t****** run %d - %s: DONE ******",
307 GetCurrentRun(), fCurrentDetector.Data()));
308 UpdateShuttleStatus(AliShuttleStatus::kDone);
309 UpdateShuttleLogbook(fCurrentDetector, "DONE");
312 return (storeResult == 0);
315 //______________________________________________________________________________________________
316 Int_t AliShuttle::StoreOCDB(const TString& gridURI)
319 // Called by StoreOCDB(), performs actual storage to the main OCDB and reference storages (Grid)
322 // -2 initialization error
325 // 1 storage delayed (e.g. previous unprocessed runs)
328 TObjArray* gridIds=0;
330 Bool_t result = kTRUE;
331 Bool_t delayed = kFALSE;
333 const char* type = 0;
335 if(gridURI == fgkMainCDB) {
337 localURI = fgkLocalCDB;
338 } else if(gridURI == fgkMainRefStorage) {
340 localURI = fgkLocalRefStorage;
342 AliError(Form("Invalid storage URI: %s", gridURI.Data()));
346 AliCDBManager* man = AliCDBManager::Instance();
348 AliCDBStorage *gridSto = man->GetStorage(gridURI);
351 Form("StoreOCDB - cannot activate main %s storage", type));
355 gridIds = gridSto->GetQueryCDBList();
357 // get objects previously stored in local CDB
358 AliCDBStorage *localSto = man->GetStorage(localURI);
361 Form("StoreOCDB - cannot activate local %s storage", type));
364 AliCDBPath aPath(GetOfflineDetName(fCurrentDetector.Data()),"*","*");
365 // Local objects were stored with current run as Grid version!
366 TList* localEntries = localSto->GetAll(aPath.GetPath(), GetCurrentRun(), GetCurrentRun());
367 localEntries->SetOwner(1);
369 // loop on local stored objects
370 TIter localIter(localEntries);
371 AliCDBEntry *aLocEntry = 0;
372 while((aLocEntry = dynamic_cast<AliCDBEntry*> (localIter.Next()))){
373 aLocEntry->SetOwner(1);
374 AliCDBId aLocId = aLocEntry->GetId();
375 aLocEntry->SetVersion(-1);
376 aLocEntry->SetSubVersion(-1);
378 // If local object is valid up to infinity we store it only if it is
379 // the first unprocessed run!
380 if (aLocId.GetLastRun() == AliCDBRunRange::Infinity() &&
381 !fFirstUnprocessed[GetDetPos(fCurrentDetector)])
383 Log("SHUTTLE", Form("StoreOCDB - %s: object %s has validity infinite but "
384 "there are previous unprocessed runs!",
385 fCurrentDetector.Data(), aLocId.GetPath().Data()));
386 Log(fCurrentDetector.Data(), Form("StoreOCDB - %s: object %s has validity infinite but "
387 "there are previous unprocessed runs!",
388 fCurrentDetector.Data(), aLocId.GetPath().Data()));
393 // loop on Grid valid Id's
394 Bool_t store = kTRUE;
395 TIter gridIter(gridIds);
396 AliCDBId* aGridId = 0;
397 while((aGridId = dynamic_cast<AliCDBId*> (gridIter.Next()))){
398 if(aGridId->GetPath() != aLocId.GetPath()) continue;
399 // skip all objects valid up to infinity
400 if(aGridId->GetLastRun() == AliCDBRunRange::Infinity()) continue;
401 // if we get here, it means there's already some more recent object stored on Grid!
406 // If we get here, the file can be stored!
407 Bool_t storeOk = gridSto->Put(aLocEntry);
408 if(!store || storeOk){
412 Log(fCurrentDetector.Data(),
413 Form("StoreOCDB - A more recent object already exists in %s storage: <%s>",
414 type, aGridId->ToString().Data()));
417 Form("StoreOCDB - Object <%s> successfully put into %s storage",
418 aLocId.ToString().Data(), type));
419 Log(fCurrentDetector.Data(),
420 Form("StoreOCDB - Object <%s> successfully put into %s storage",
421 aLocId.ToString().Data(), type));
424 // removing local filename...
426 localSto->IdToFilename(aLocId, filename);
427 Log("SHUTTLE", Form("StoreOCDB - Removing local file %s", filename.Data()));
428 RemoveFile(filename.Data());
432 Form("StoreOCDB - Grid %s storage of object <%s> failed",
433 type, aLocId.ToString().Data()));
434 Log(fCurrentDetector.Data(),
435 Form("StoreOCDB - Grid %s storage of object <%s> failed",
436 type, aLocId.ToString().Data()));
440 localEntries->Clear();
442 Int_t returnCode = 0;
444 if (result == kFALSE)
446 else if (delayed != kFALSE)
449 Log("SHUTTLE", Form("StoreOCDB - Returning with %d (result = %d, delayed = %d)", returnCode, result, delayed));
450 Log(fCurrentDetector.Data(), Form("StoreOCDB - Returning with %d (result = %d, delayed = %d)", returnCode, result, delayed));
455 //______________________________________________________________________________________________
456 Bool_t AliShuttle::CleanReferenceStorage(const char* detector)
458 // clears the directory used to store reference files of a given subdetector
460 AliCDBManager* man = AliCDBManager::Instance();
461 AliCDBStorage* sto = man->GetStorage(fgkLocalRefStorage);
462 TString localBaseFolder = sto->GetBaseFolder();
464 TString targetDir = GetRefFilePrefix(localBaseFolder.Data(), detector);
466 Log("SHUTTLE", Form("CleanReferenceStorage - Cleaning %s", targetDir.Data()));
469 begin.Form("%d_", GetCurrentRun());
471 TSystemDirectory* baseDir = new TSystemDirectory("/", targetDir);
475 TList* dirList = baseDir->GetListOfFiles();
478 if (!dirList) return kTRUE;
480 if (dirList->GetEntries() < 3)
486 Int_t nDirs = 0, nDel = 0;
487 TIter dirIter(dirList);
488 TSystemFile* entry = 0;
490 Bool_t success = kTRUE;
492 while ((entry = dynamic_cast<TSystemFile*> (dirIter.Next())))
494 if (entry->IsDirectory())
497 TString fileName(entry->GetName());
498 if (!fileName.BeginsWith(begin))
504 Int_t result = gSystem->Unlink(fileName.Data());
508 Log("SHUTTLE", Form("CleanReferenceStorage - Could not delete file %s!", fileName.Data()));
516 Log("SHUTTLE", Form("CleanReferenceStorage - %d (over %d) reference files in folder %s were deleted.",
517 nDel, nDirs, targetDir.Data()));
528 Int_t result = gSystem->GetPathInfo(targetDir, 0, (Long64_t*) 0, 0, 0);
532 result = gSystem->Exec(Form("rm -rf %s", targetDir.Data()));
535 Log("SHUTTLE", Form("CleanReferenceStorage - Could not clean directory %s", targetDir.Data()));
540 result = gSystem->mkdir(targetDir, kTRUE);
543 Log("SHUTTLE", Form("CleanReferenceStorage - Error creating base directory %s", targetDir.Data()));
550 //______________________________________________________________________________________________
551 Bool_t AliShuttle::StoreReferenceFile(const char* detector, const char* localFile, const char* gridFileName)
554 // Stores reference file directly (without opening it). This function stores the file locally.
556 // The file is stored under the following location:
557 // <base folder of local reference storage>/<DET>/<RUN#>_<gridFileName>
558 // where <gridFileName> is the second parameter given to the function
561 if (fTestMode & kErrorStorage)
563 Log(fCurrentDetector, "StoreReferenceFile - In TESTMODE - Simulating error while storing locally");
567 AliCDBManager* man = AliCDBManager::Instance();
568 AliCDBStorage* sto = man->GetStorage(fgkLocalRefStorage);
570 TString localBaseFolder = sto->GetBaseFolder();
572 TString target = GetRefFilePrefix(localBaseFolder.Data(), detector);
573 target.Append(Form("/%d_%s", GetCurrentRun(), gridFileName));
575 return CopyFileLocally(localFile, target);
578 //______________________________________________________________________________________________
579 Bool_t AliShuttle::StoreRunMetadataFile(const char* localFile, const char* gridFileName)
582 // Stores Run metadata file to the Grid, in the run folder
584 // Only GRP can call this function.
586 if (fTestMode & kErrorStorage)
588 Log(fCurrentDetector, "StoreRunMetaDataFile - In TESTMODE - Simulating error while storing locally");
592 AliCDBManager* man = AliCDBManager::Instance();
593 AliCDBStorage* sto = man->GetStorage(fgkLocalRefStorage);
595 TString localBaseFolder = sto->GetBaseFolder();
597 // Build Run level folder
598 // folder = /alice/data/year/lhcPeriod/runNb/raw
601 TString lhcPeriod = GetLHCPeriod();
602 if (lhcPeriod.Length() == 0)
604 Log("SHUTTLE","StoreRunMetaDataFile - LHCPeriod not found in logbook!");
608 // TODO partitions with one detector only write data into LHCperiod_DET
609 TString partition = GetRunParameter("detector");
611 if (partition.Length() > 0 && partition != "ALICE")
613 lhcPeriod.Append(Form("_%s", partition.Data()));
614 Log(fCurrentDetector, Form("Run data tags merged file will be written in %s",
618 TString target = Form("%s/GRP/RunMetadata/alice/data/%d/%s/%09d/raw/%s",
619 localBaseFolder.Data(), GetCurrentYear(),
620 lhcPeriod.Data(), GetCurrentRun(), gridFileName);
622 return CopyFileLocally(localFile, target);
625 //______________________________________________________________________________________________
626 Bool_t AliShuttle::CopyFileLocally(const char* localFile, const TString& target)
629 // Stores file locally. Called by StoreReferenceFile and StoreRunMetadataFile
630 // Files are temporarily stored in the local reference storage. When the preprocessor
631 // finishes, the Shuttle calls CopyFilesToGrid to transfer the files to AliEn
632 // (in reference or run level folders)
635 TString targetDir(target(0, target.Last('/')));
637 //try to open base dir folder, if it does not exist
638 void* dir = gSystem->OpenDirectory(targetDir.Data());
640 if (gSystem->mkdir(targetDir.Data(), kTRUE)) {
641 Log("SHUTTLE", Form("CopyFileLocally - Can't open directory <%s>", targetDir.Data()));
646 gSystem->FreeDirectory(dir);
651 result = gSystem->GetPathInfo(localFile, 0, (Long64_t*) 0, 0, 0);
654 Log("SHUTTLE", Form("CopyFileLocally - %s does not exist", localFile));
658 result = gSystem->GetPathInfo(target, 0, (Long64_t*) 0, 0, 0);
661 Log("SHUTTLE", Form("CopyFileLocally - target file %s already exist, removing...", target.Data()));
662 if (gSystem->Unlink(target.Data()))
664 Log("SHUTTLE", Form("CopyFileLocally - Could not remove existing target file %s!", target.Data()));
669 result = gSystem->CopyFile(localFile, target);
673 Log("SHUTTLE", Form("CopyFileLocally - File %s stored locally to %s", localFile, target.Data()));
678 Log("SHUTTLE", Form("CopyFileLocally - Could not store file %s to %s! Error code = %d",
679 localFile, target.Data(), result));
687 //______________________________________________________________________________________________
688 Bool_t AliShuttle::CopyFilesToGrid(const char* type)
691 // Transfers local files to the Grid. Local files can be reference files
692 // or run metadata file (from GRP only).
694 // According to the type (ref, metadata) the files are stored under the following location:
695 // ref --> <base folder of reference storage>/<DET>/<RUN#>_<gridFileName>
696 // metadata --> <run data folder>/<MetadataFileName>
699 AliCDBManager* man = AliCDBManager::Instance();
700 AliCDBStorage* sto = man->GetStorage(fgkLocalRefStorage);
703 TString localBaseFolder = sto->GetBaseFolder();
709 if (strcmp(type, "reference") == 0)
711 dir = GetRefFilePrefix(localBaseFolder.Data(), fCurrentDetector.Data());
712 AliCDBStorage* gridSto = man->GetStorage(fgkMainRefStorage);
715 TString gridBaseFolder = gridSto->GetBaseFolder();
716 alienDir = GetRefFilePrefix(gridBaseFolder.Data(), fCurrentDetector.Data());
717 begin = Form("%d_", GetCurrentRun());
719 else if (strcmp(type, "metadata") == 0)
722 TString lhcPeriod = GetLHCPeriod();
724 if (lhcPeriod.Length() == 0)
726 Log("SHUTTLE","CopyFilesToGrid - LHCPeriod not found in logbook!");
730 // TODO partitions with one detector only write data into LHCperiod_DET
731 TString partition = GetRunParameter("detector");
733 if (partition.Length() > 0 && partition != "ALICE")
735 lhcPeriod.Append(Form("_%s", partition.Data()));
738 dir = Form("%s/GRP/RunMetadata/alice/data/%d/%s/%09d/raw",
739 localBaseFolder.Data(), GetCurrentYear(),
740 lhcPeriod.Data(), GetCurrentRun());
741 alienDir = dir(dir.Index("/alice/data/"), dir.Length());
747 Log("SHUTTLE", "CopyFilesToGrid - Unexpected: type label must be reference or metadata!");
751 TSystemDirectory* baseDir = new TSystemDirectory("/", dir);
755 TList* dirList = baseDir->GetListOfFiles();
758 if (!dirList) return kTRUE;
760 if (dirList->GetEntries() < 3)
768 Log("SHUTTLE", "CopyFilesToGrid - Connection to Grid failed: Cannot continue!");
773 Int_t nDirs = 0, nTransfer = 0;
774 TIter dirIter(dirList);
775 TSystemFile* entry = 0;
777 Bool_t success = kTRUE;
778 Bool_t first = kTRUE;
780 while ((entry = dynamic_cast<TSystemFile*> (dirIter.Next())))
782 if (entry->IsDirectory())
785 TString fileName(entry->GetName());
786 if (!fileName.BeginsWith(begin))
794 // check that folder exists, otherwise create it
795 TGridResult* result = gGrid->Ls(alienDir.Data(), "a");
803 if (!result->GetFileName(1)) // TODO: It looks like element 0 is always 0!!
805 // TODO It does not work currently! Bug in TAliEn::Mkdir
806 // TODO Manually fixed in local root v5-16-00
807 if (!gGrid->Mkdir(alienDir.Data(),"-p",0))
809 Log("SHUTTLE", Form("CopyFilesToGrid - Cannot create directory %s",
814 Log("SHUTTLE",Form("CopyFilesToGrid - Folder %s created", alienDir.Data()));
818 Log("SHUTTLE",Form("CopyFilesToGrid - Folder %s found", alienDir.Data()));
822 TString fullLocalPath;
823 fullLocalPath.Form("%s/%s", dir.Data(), fileName.Data());
825 TString fullGridPath;
826 fullGridPath.Form("alien://%s/%s", alienDir.Data(), fileName.Data());
828 Bool_t result = TFile::Cp(fullLocalPath, fullGridPath);
832 Log("SHUTTLE", Form("CopyFilesToGrid - Copying local file %s to %s succeeded!",
833 fullLocalPath.Data(), fullGridPath.Data()));
834 RemoveFile(fullLocalPath);
839 Log("SHUTTLE", Form("CopyFilesToGrid - Copying local file %s to %s FAILED!",
840 fullLocalPath.Data(), fullGridPath.Data()));
845 Log("SHUTTLE", Form("CopyFilesToGrid - %d (over %d) files in folder %s copied to Grid.",
846 nTransfer, nDirs, dir.Data()));
853 //______________________________________________________________________________________________
854 const char* AliShuttle::GetRefFilePrefix(const char* base, const char* detector)
857 // Get folder name of reference files
860 TString offDetStr(GetOfflineDetName(detector));
862 if (offDetStr == "ITS" || offDetStr == "MUON" || offDetStr == "PHOS")
864 dir.Form("%s/%s/%s", base, offDetStr.Data(), detector);
866 dir.Form("%s/%s", base, offDetStr.Data());
874 //______________________________________________________________________________________________
875 void AliShuttle::CleanLocalStorage(const TString& uri)
878 // Called in case the preprocessor is declared failed. Remove remaining objects from the local storages.
881 const char* type = 0;
882 if(uri == fgkLocalCDB) {
884 } else if(uri == fgkLocalRefStorage) {
887 AliError(Form("Invalid storage URI: %s", uri.Data()));
891 AliCDBManager* man = AliCDBManager::Instance();
893 // open local storage
894 AliCDBStorage *localSto = man->GetStorage(uri);
897 Form("CleanLocalStorage - cannot activate local %s storage", type));
901 TString filename(Form("%s/%s/*/Run*_v%d_s*.root",
902 localSto->GetBaseFolder().Data(), GetOfflineDetName(fCurrentDetector.Data()), GetCurrentRun()));
904 AliDebug(2, Form("filename = %s", filename.Data()));
906 Log("SHUTTLE", Form("Removing remaining local files for run %d and detector %s ...",
907 GetCurrentRun(), fCurrentDetector.Data()));
909 RemoveFile(filename.Data());
913 //______________________________________________________________________________________________
914 void AliShuttle::RemoveFile(const char* filename)
917 // removes local file
920 TString command(Form("rm -f %s", filename));
922 Int_t result = gSystem->Exec(command.Data());
925 Log("SHUTTLE", Form("RemoveFile - %s: Cannot remove file %s!",
926 fCurrentDetector.Data(), filename));
930 //______________________________________________________________________________________________
931 AliShuttleStatus* AliShuttle::ReadShuttleStatus()
934 // Reads the AliShuttleStatus from the CDB
942 fStatusEntry = AliCDBManager::Instance()->GetStorage(GetLocalCDB())
943 ->Get(Form("/SHUTTLE/STATUS/%s", fCurrentDetector.Data()), GetCurrentRun());
945 if (!fStatusEntry) return 0;
946 fStatusEntry->SetOwner(1);
948 AliShuttleStatus* status = dynamic_cast<AliShuttleStatus*> (fStatusEntry->GetObject());
950 AliError("Invalid object stored to CDB!");
957 //______________________________________________________________________________________________
958 Bool_t AliShuttle::WriteShuttleStatus(AliShuttleStatus* status)
961 // writes the status for one subdetector
969 Int_t run = GetCurrentRun();
971 AliCDBId id(AliCDBPath("SHUTTLE", "STATUS", fCurrentDetector), run, run);
973 fStatusEntry = new AliCDBEntry(status, id, new AliCDBMetaData);
974 fStatusEntry->SetOwner(1);
976 UInt_t result = AliCDBManager::Instance()->GetStorage(fgkLocalCDB)->Put(fStatusEntry);
979 Log("SHUTTLE", Form("WriteShuttleStatus - Failed for %s, run %d",
980 fCurrentDetector.Data(), run));
989 //______________________________________________________________________________________________
990 void AliShuttle::UpdateShuttleStatus(AliShuttleStatus::Status newStatus, Bool_t increaseCount)
993 // changes the AliShuttleStatus for the given detector and run to the given status
997 AliError("UNEXPECTED: fStatusEntry empty");
1001 AliShuttleStatus* status = dynamic_cast<AliShuttleStatus*> (fStatusEntry->GetObject());
1004 Log("SHUTTLE", "UpdateShuttleStatus - UNEXPECTED: status could not be read from current CDB entry");
1008 TString actionStr = Form("UpdateShuttleStatus - %s: Changing state from %s to %s",
1009 fCurrentDetector.Data(),
1010 status->GetStatusName(),
1011 status->GetStatusName(newStatus));
1012 Log("SHUTTLE", actionStr);
1013 SetLastAction(actionStr);
1015 status->SetStatus(newStatus);
1016 if (increaseCount) status->IncreaseCount();
1018 AliCDBManager::Instance()->GetStorage(fgkLocalCDB)->Put(fStatusEntry);
1023 //______________________________________________________________________________________________
1024 void AliShuttle::SendMLInfo()
1027 // sends ML information about the current status of the current detector being processed
1030 AliShuttleStatus* status = dynamic_cast<AliShuttleStatus*> (fStatusEntry->GetObject());
1033 Log("SHUTTLE", "SendMLInfo - UNEXPECTED: status could not be read from current CDB entry");
1037 TMonaLisaText mlStatus(Form("%s_status", fCurrentDetector.Data()), status->GetStatusName());
1038 TMonaLisaValue mlRetryCount(Form("%s_count", fCurrentDetector.Data()), status->GetCount());
1041 mlList.Add(&mlStatus);
1042 mlList.Add(&mlRetryCount);
1045 mlID.Form("%d", GetCurrentRun());
1046 fMonaLisa->SendParameters(&mlList, mlID);
1049 //______________________________________________________________________________________________
1050 Bool_t AliShuttle::ContinueProcessing()
1052 // this function reads the AliShuttleStatus information from CDB and
1053 // checks if the processing should be continued
1054 // if yes it returns kTRUE and updates the AliShuttleStatus with nextStatus
1056 if (!fConfig->HostProcessDetector(fCurrentDetector))
1059 AliPreprocessor* aPreprocessor =
1060 dynamic_cast<AliPreprocessor*> (fPreprocessorMap.GetValue(fCurrentDetector));
1063 Log("SHUTTLE", Form("ContinueProcessing - %s: no preprocessor registered", fCurrentDetector.Data()));
1067 AliShuttleLogbookEntry::Status entryStatus =
1068 fLogbookEntry->GetDetectorStatus(fCurrentDetector);
1070 if (entryStatus != AliShuttleLogbookEntry::kUnprocessed) {
1071 Log("SHUTTLE", Form("ContinueProcessing - %s is %s",
1072 fCurrentDetector.Data(),
1073 fLogbookEntry->GetDetectorStatusName(entryStatus)));
1077 // if we get here, according to Shuttle logbook subdetector is in UNPROCESSED state
1079 // check if current run is first unprocessed run for current detector
1080 if (fConfig->StrictRunOrder(fCurrentDetector) &&
1081 !fFirstUnprocessed[GetDetPos(fCurrentDetector)])
1083 if (fTestMode == kNone)
1085 Log("SHUTTLE", Form("ContinueProcessing - %s requires strict run ordering"
1086 " but this is not the first unprocessed run!"));
1091 Log("SHUTTLE", Form("ContinueProcessing - In TESTMODE - "
1092 "Although %s requires strict run ordering "
1093 "and this is not the first unprocessed run, "
1094 "the SHUTTLE continues"));
1098 // Is the subdetector processed first time for this run?
1099 fFirstProcessing = kFALSE;
1101 AliShuttleStatus* status = ReadShuttleStatus();
1104 Log("SHUTTLE", Form("ContinueProcessing - %s: Processing first time",
1105 fCurrentDetector.Data()));
1106 status = new AliShuttleStatus(AliShuttleStatus::kStarted);
1107 fFirstProcessing = kTRUE;
1108 return WriteShuttleStatus(status);
1111 // The following two cases shouldn't happen if Shuttle Logbook was correctly updated.
1112 // If it happens it may mean Logbook updating failed... let's do it now!
1113 if (status->GetStatus() == AliShuttleStatus::kDone ||
1114 status->GetStatus() == AliShuttleStatus::kFailed){
1115 Log("SHUTTLE", Form("ContinueProcessing - %s is already %s. Updating Shuttle Logbook",
1116 fCurrentDetector.Data(),
1117 status->GetStatusName(status->GetStatus())));
1118 UpdateShuttleLogbook(fCurrentDetector.Data(),
1119 status->GetStatusName(status->GetStatus()));
1123 if (status->GetStatus() == AliShuttleStatus::kStoreStarted || status->GetStatus() == AliShuttleStatus::kStoreDelayed ||status->GetStatus() == AliShuttleStatus::kStoreError) {
1125 Form("ContinueProcessing - %s: Grid storage of one or more "
1126 "objects failed. Trying again now",
1127 fCurrentDetector.Data()));
1132 // if we get here, there is a restart
1133 Bool_t cont = kFALSE;
1136 if (status->GetCount() >= fConfig->GetMaxRetries()) {
1137 Log("SHUTTLE", Form("ContinueProcessing - %s failed %d times in status %s - "
1138 "Updating Shuttle Logbook", fCurrentDetector.Data(),
1139 status->GetCount(), status->GetStatusName()));
1140 UpdateShuttleLogbook(fCurrentDetector.Data(), "FAILED");
1141 UpdateShuttleStatus(AliShuttleStatus::kFailed);
1143 // there may still be objects in local OCDB and reference storage
1144 // and FXS databases may be not updated: do it now!
1146 // TODO Currently disabled, we want to keep files in case of failure!
1147 // CleanLocalStorage(fgkLocalCDB);
1148 // CleanLocalStorage(fgkLocalRefStorage);
1149 // UpdateTableFailCase();
1151 // Send mail to detector expert!
1152 Log("SHUTTLE", Form("ContinueProcessing - Sending mail to %s expert...",
1153 fCurrentDetector.Data()));
1154 if (!SendMail(kPPEMail))
1155 Log("SHUTTLE", Form("ContinueProcessing - Could not send mail to %s expert",
1156 fCurrentDetector.Data()));
1159 Log("SHUTTLE", Form("ContinueProcessing - %s: restarting. "
1160 "Aborted before with %s. Retry number %d.", fCurrentDetector.Data(),
1161 status->GetStatusName(), status->GetCount()));
1162 Bool_t increaseCount = kTRUE;
1163 if (status->GetStatus() == AliShuttleStatus::kDCSError ||
1164 status->GetStatus() == AliShuttleStatus::kDCSStarted ||
1165 status->GetStatus() == AliShuttleStatus::kFXSError)
1166 increaseCount = kFALSE;
1168 UpdateShuttleStatus(AliShuttleStatus::kStarted, increaseCount);
1175 //______________________________________________________________________________________________
1176 Bool_t AliShuttle::Process(AliShuttleLogbookEntry* entry)
1179 // Makes data retrieval for all detectors in the configuration.
1180 // entry: Shuttle logbook entry, contains run paramenters and status of detectors
1181 // (Unprocessed, Inactive, Failed or Done).
1182 // Returns kFALSE in case of error occured and kTRUE otherwise
1185 if (!entry) return kFALSE;
1187 fLogbookEntry = entry;
1189 Log("SHUTTLE", Form("\t\t\t^*^*^*^*^*^*^*^*^*^*^*^* run %d: START ^*^*^*^*^*^*^*^*^*^*^*^*",
1192 // Send the information to ML
1195 TMonaLisaText mlStatus("SHUTTLE_status", "Processing");
1196 TString runType(entry->GetRunType());
1197 if (strlen(entry->GetRunParameter("log")) > 0){
1200 runType += entry->GetRunParameter("log");
1203 TMonaLisaText mlRunType("SHUTTLE_runtype", runType);
1206 mlList.Add(&mlStatus);
1207 mlList.Add(&mlRunType);
1210 mlID.Form("%d", GetCurrentRun());
1211 fMonaLisa->SendParameters(&mlList, mlID);
1213 if (fLogbookEntry->IsDone())
1215 Log("SHUTTLE","Process - Shuttle is already DONE. Updating logbook");
1216 UpdateShuttleLogbook("shuttle_done");
1221 // read test mode if flag is set
1225 TString logEntry(entry->GetRunParameter("log"));
1226 //printf("log entry = %s\n", logEntry.Data());
1227 TString searchStr("Testmode: ");
1228 Int_t pos = logEntry.Index(searchStr.Data());
1229 //printf("%d\n", pos);
1232 TSubString subStr = logEntry(pos + searchStr.Length(), logEntry.Length());
1233 //printf("%s\n", subStr.String().Data());
1234 TString newStr(subStr.Data());
1235 TObjArray* token = newStr.Tokenize(' ');
1239 TObjString* tmpStr = dynamic_cast<TObjString*> (token->First());
1242 Int_t testMode = tmpStr->String().Atoi();
1245 Log("SHUTTLE", Form("Process - Enabling test mode %d", testMode));
1246 SetTestMode((TestMode) testMode);
1254 fLogbookEntry->Print("all");
1257 Bool_t hasError = kFALSE;
1259 // Set the CDB and Reference folders according to the year and LHC period
1260 TString lhcPeriod(GetLHCPeriod());
1261 if (lhcPeriod.Length() == 0)
1263 Log("SHUTTLE","Process - LHCPeriod not found in logbook!");
1267 if (fgkMainCDB.Length() == 0)
1268 fgkMainCDB = Form("alien://folder=/alice/data/%d/%s/OCDB?user=alidaq?cacheFold=/tmp/OCDBCache",
1269 GetCurrentYear(), lhcPeriod.Data());
1271 if (fgkMainRefStorage.Length() == 0)
1272 fgkMainRefStorage = Form("alien://folder=/alice/data/%d/%s/Reference?user=alidaq?cacheFold=/tmp/OCDBCache",
1273 GetCurrentYear(), lhcPeriod.Data());
1275 // Loop on detectors in the configuration
1276 TIter iter(fConfig->GetDetectors());
1277 TObjString* aDetector = 0;
1279 Bool_t first = kTRUE;
1281 while ((aDetector = (TObjString*) iter.Next()))
1283 fCurrentDetector = aDetector->String();
1285 if (ContinueProcessing() == kFALSE)
1290 // only read QueryCDB when needed and only once
1291 AliCDBStorage *mainCDBSto = AliCDBManager::Instance()->GetStorage(fgkMainCDB);
1292 if(mainCDBSto) mainCDBSto->QueryCDB(GetCurrentRun());
1293 AliCDBStorage *mainRefSto = AliCDBManager::Instance()->GetStorage(fgkMainRefStorage);
1294 if(mainRefSto) mainRefSto->QueryCDB(GetCurrentRun());
1298 Log("SHUTTLE", Form("\t\t\t****** run %d - %s: START ******",
1299 GetCurrentRun(), aDetector->GetName()));
1301 for(Int_t iSys=0;iSys<3;iSys++) fFXSCalled[iSys]=kFALSE;
1303 Log(fCurrentDetector.Data(), "Process - Starting processing");
1309 Log("SHUTTLE", "Process - ERROR: Forking failed");
1314 Log("SHUTTLE", Form("Process - In parent process of %d - %s: Starting monitoring",
1315 GetCurrentRun(), aDetector->GetName()));
1317 Long_t begin = time(0);
1319 int status; // to be used with waitpid, on purpose an int (not Int_t)!
1320 while (waitpid(pid, &status, WNOHANG) == 0)
1322 Long_t expiredTime = time(0) - begin;
1324 if (expiredTime > fConfig->GetPPTimeOut())
1327 tmp.Form("Process - Process of %s time out. "
1328 "Run time: %d seconds. Killing...",
1329 fCurrentDetector.Data(), expiredTime);
1330 Log("SHUTTLE", tmp);
1331 Log(fCurrentDetector, tmp);
1335 UpdateShuttleStatus(AliShuttleStatus::kPPTimeOut);
1338 gSystem->Sleep(1000);
1342 gSystem->Sleep(1000);
1345 checkStr.Form("ps -o vsize --pid %d | tail -n 1", pid);
1346 FILE* pipe = gSystem->OpenPipe(checkStr, "r");
1349 Log("SHUTTLE", Form("Process - Error: "
1350 "Could not open pipe to %s", checkStr.Data()));
1355 if (!fgets(buffer, 100, pipe))
1357 Log("SHUTTLE", "Process - Error: ps did not return anything");
1358 gSystem->ClosePipe(pipe);
1361 gSystem->ClosePipe(pipe);
1363 //Log("SHUTTLE", Form("ps returned %s", buffer));
1366 if ((sscanf(buffer, "%d\n", &mem) != 1) || !mem)
1368 Log("SHUTTLE", "Process - Error: Could not parse output of ps");
1372 if (expiredTime % 60 == 0)
1374 Log("SHUTTLE", Form("Process - %s: Checking process. "
1375 "Run time: %d seconds - Memory consumption: %d KB",
1376 fCurrentDetector.Data(), expiredTime, mem));
1380 if (mem > fConfig->GetPPMaxMem())
1383 tmp.Form("Process - Process exceeds maximum allowed memory "
1384 "(%d KB > %d KB). Killing...",
1385 mem, fConfig->GetPPMaxMem());
1386 Log("SHUTTLE", tmp);
1387 Log(fCurrentDetector, tmp);
1391 UpdateShuttleStatus(AliShuttleStatus::kPPOutOfMemory);
1394 gSystem->Sleep(1000);
1399 Log("SHUTTLE", Form("Process - In parent process of %d - %s: Client has terminated.",
1400 GetCurrentRun(), aDetector->GetName()));
1402 if (WIFEXITED(status))
1404 Int_t returnCode = WEXITSTATUS(status);
1406 Log("SHUTTLE", Form("Process - %s: the return code is %d", fCurrentDetector.Data(),
1409 if (returnCode == 0) hasError = kTRUE;
1415 Log("SHUTTLE", Form("Process - In client process of %d - %s", GetCurrentRun(),
1416 aDetector->GetName()));
1418 Log("SHUTTLE", Form("Process - Redirecting output to %s log",fCurrentDetector.Data()));
1420 if ((freopen(GetLogFileName(fCurrentDetector), "a", stdout)) == 0)
1422 Log("SHUTTLE", "Process - Could not freopen stdout");
1426 fOutputRedirected = kTRUE;
1427 if ((dup2(fileno(stdout), fileno(stderr))) < 0)
1428 Log("SHUTTLE", "Process - Could not redirect stderr");
1432 TString wd = gSystem->WorkingDirectory();
1433 TString tmpDir = Form("%s/%s_%d_process", GetShuttleTempDir(),
1434 fCurrentDetector.Data(), GetCurrentRun());
1436 Int_t result = gSystem->GetPathInfo(tmpDir.Data(), 0, (Long64_t*) 0, 0, 0);
1437 if (!result) // temp dir already exists!
1439 Log(fCurrentDetector.Data(),
1440 Form("Process - %s dir already exists! Removing...", tmpDir.Data()));
1441 gSystem->Exec(Form("rm -rf %s",tmpDir.Data()));
1444 if (gSystem->mkdir(tmpDir.Data(), 1))
1446 Log(fCurrentDetector.Data(), "Process - could not make temp directory!!");
1450 if (!gSystem->ChangeDirectory(tmpDir.Data()))
1452 Log(fCurrentDetector.Data(), "Process - could not change directory!!");
1456 Bool_t success = ProcessCurrentDetector();
1458 gSystem->ChangeDirectory(wd.Data());
1460 if (success) // Preprocessor finished successfully!
1462 // remove temporary folder or DCS map
1463 if (!fConfig->KeepTempFolder())
1465 gSystem->Exec(Form("rm -rf %s",tmpDir.Data()));
1466 } else if (!fConfig->KeepDCSMap())
1468 gSystem->Exec(Form("rm -f %s/DCSMap.root",tmpDir.Data()));
1471 // Update time_processed field in FXS DB
1472 if (UpdateTable() == kFALSE)
1473 Log("SHUTTLE", Form("Process - %s: Could not update FXS databases!",
1474 fCurrentDetector.Data()));
1476 // Transfer the data from local storage to main storage (Grid)
1477 if (StoreOCDB() == kFALSE)
1483 Form("\t\t\t****** run %d - %s: PP ERROR ******",
1484 GetCurrentRun(), aDetector->GetName()));
1487 for (UInt_t iSys=0; iSys<3; iSys++)
1489 if (fFXSCalled[iSys]) fFXSlist[iSys].Clear();
1492 Log("SHUTTLE", Form("Process - Client process of %d - %s is exiting now with %d.",
1493 GetCurrentRun(), aDetector->GetName(), success));
1495 // the client exits here
1496 gSystem->Exit(success);
1498 AliError("We should never get here!!!");
1502 Log("SHUTTLE", Form("\t\t\t^*^*^*^*^*^*^*^*^*^*^*^* run %d: FINISH ^*^*^*^*^*^*^*^*^*^*^*^*",
1505 //check if shuttle is done for this run, if so update logbook
1506 TObjArray checkEntryArray;
1507 checkEntryArray.SetOwner(1);
1508 TString whereClause = Form("where run=%d", GetCurrentRun());
1509 if (!QueryShuttleLogbook(whereClause.Data(), checkEntryArray) ||
1510 checkEntryArray.GetEntries() == 0) {
1511 Log("SHUTTLE", Form("Process - Warning: Cannot check status of run %d on Shuttle logbook!",
1513 return hasError == kFALSE;
1516 AliShuttleLogbookEntry* checkEntry = dynamic_cast<AliShuttleLogbookEntry*>
1517 (checkEntryArray.At(0));
1521 if (checkEntry->IsDone())
1523 Log("SHUTTLE","Process - Shuttle is DONE. Updating logbook");
1524 UpdateShuttleLogbook("shuttle_done");
1528 for (UInt_t iDet=0; iDet<NDetectors(); iDet++)
1530 if (checkEntry->GetDetectorStatus(iDet) == AliShuttleLogbookEntry::kUnprocessed)
1532 AliDebug(2, Form("Run %d: setting %s as \"not first time unprocessed\"",
1533 checkEntry->GetRun(), GetDetName(iDet)));
1534 fFirstUnprocessed[iDet] = kFALSE;
1537 TMonaLisaText mlStatusPending("SHUTTLE_status", "Pending");
1539 mlList.Add(&mlStatusPending);
1540 fMonaLisa->SendParameters(&mlList, mlID);
1546 return hasError == kFALSE;
1549 //______________________________________________________________________________________________
1550 Bool_t AliShuttle::ProcessCurrentDetector()
1553 // Makes data retrieval just for a specific detector (fCurrentDetector).
1554 // Threre should be a configuration for this detector.
1556 Log("SHUTTLE", Form("ProcessCurrentDetector - Retrieving values for %s, run %d",
1557 fCurrentDetector.Data(), GetCurrentRun()));
1559 TString wd = gSystem->WorkingDirectory();
1561 if (!CleanReferenceStorage(fCurrentDetector.Data()))
1564 gSystem->ChangeDirectory(wd.Data());
1566 TMap* dcsMap = new TMap();
1568 // call preprocessor
1569 AliPreprocessor* aPreprocessor =
1570 dynamic_cast<AliPreprocessor*> (fPreprocessorMap.GetValue(fCurrentDetector));
1572 aPreprocessor->Initialize(GetCurrentRun(), GetCurrentStartTime(), GetCurrentEndTime());
1574 Bool_t processDCS = aPreprocessor->ProcessDCS();
1578 Log(fCurrentDetector, "ProcessCurrentDetector -"
1579 " The preprocessor requested to skip the retrieval of DCS values");
1581 else if (fTestMode & kSkipDCS)
1583 Log(fCurrentDetector, "ProcessCurrentDetector - In TESTMODE: Skipping DCS processing");
1585 else if (fTestMode & kErrorDCS)
1587 Log(fCurrentDetector, "ProcessCurrentDetector - In TESTMODE: Simulating DCS error");
1588 UpdateShuttleStatus(AliShuttleStatus::kDCSStarted);
1589 UpdateShuttleStatus(AliShuttleStatus::kDCSError);
1594 UpdateShuttleStatus(AliShuttleStatus::kDCSStarted);
1596 // Query DCS archive
1597 Int_t nServers = fConfig->GetNServers(fCurrentDetector);
1599 for (int iServ=0; iServ<nServers; iServ++)
1602 TString host(fConfig->GetDCSHost(fCurrentDetector, iServ));
1603 Int_t port = fConfig->GetDCSPort(fCurrentDetector, iServ);
1604 Int_t multiSplit = fConfig->GetMultiSplit(fCurrentDetector, iServ);
1606 Log(fCurrentDetector, Form("ProcessCurrentDetector -"
1607 " Querying DCS Amanda server %s:%d (%d of %d)",
1608 host.Data(), port, iServ+1, nServers));
1613 if (fConfig->GetDCSAliases(fCurrentDetector, iServ)->GetEntries() > 0)
1615 aliasMap = GetValueSet(host, port,
1616 fConfig->GetDCSAliases(fCurrentDetector, iServ),
1617 kAlias, multiSplit);
1620 Log(fCurrentDetector,
1621 Form("ProcessCurrentDetector -"
1622 " Error retrieving DCS aliases from server %s."
1623 " Sending mail to DCS experts!", host.Data()));
1624 UpdateShuttleStatus(AliShuttleStatus::kDCSError);
1626 if (!SendMail(kDCSEMail))
1627 Log("SHUTTLE", Form("ProcessCurrentDetector - "
1628 "Could not send mail to DCS experts!"));
1635 if (fConfig->GetDCSDataPoints(fCurrentDetector, iServ)->GetEntries() > 0)
1637 dpMap = GetValueSet(host, port,
1638 fConfig->GetDCSDataPoints(fCurrentDetector, iServ),
1642 Log(fCurrentDetector,
1643 Form("ProcessCurrentDetector -"
1644 " Error retrieving DCS data points from server %s."
1645 " Sending mail to DCS experts!", host.Data()));
1646 UpdateShuttleStatus(AliShuttleStatus::kDCSError);
1648 if (!SendMail(kDCSEMail))
1649 Log("SHUTTLE", Form("ProcessCurrentDetector - "
1650 "Could not send mail to DCS experts!"));
1652 if (aliasMap) delete aliasMap;
1658 // merge aliasMap and dpMap into dcsMap
1660 TIter iter(aliasMap);
1661 TObjString* key = 0;
1662 while ((key = (TObjString*) iter.Next()))
1663 dcsMap->Add(key, aliasMap->GetValue(key->String()));
1665 aliasMap->SetOwner(kFALSE);
1671 TObjString* key = 0;
1672 while ((key = (TObjString*) iter.Next()))
1673 dcsMap->Add(key, dpMap->GetValue(key->String()));
1675 dpMap->SetOwner(kFALSE);
1681 // save map into file, to help debugging in case of preprocessor error
1682 TFile* f = TFile::Open("DCSMap.root","recreate");
1684 dcsMap->Write("DCSMap", TObject::kSingleKey);
1688 // DCS Archive DB processing successful. Call Preprocessor!
1689 UpdateShuttleStatus(AliShuttleStatus::kPPStarted);
1691 fFXSError = -1; // this variable is kTRUE after ::Process if an FXS error occured
1693 UInt_t returnValue = aPreprocessor->Process(dcsMap);
1695 if (fFXSError!=-1) {
1696 UpdateShuttleStatus(AliShuttleStatus::kFXSError);
1697 SendMail(kFXSEMail, fFXSError);
1698 dcsMap->DeleteAll();
1703 if (returnValue > 0) // Preprocessor error!
1705 Log(fCurrentDetector, Form("ProcessCurrentDetector - "
1706 "Preprocessor failed. Process returned %d.", returnValue));
1707 UpdateShuttleStatus(AliShuttleStatus::kPPError);
1708 dcsMap->DeleteAll();
1714 UpdateShuttleStatus(AliShuttleStatus::kPPDone);
1715 Log(fCurrentDetector, Form("ProcessCurrentDetector - %s preprocessor returned success",
1716 fCurrentDetector.Data()));
1718 dcsMap->DeleteAll();
1724 //______________________________________________________________________________________________
1725 void AliShuttle::CountOpenRuns()
1727 // Query DAQ's Shuttle logbook and sends the number of open runs to ML
1729 // check connection, in case connect
1734 sqlQuery = Form("select count(*) from %s where shuttle_done=0", fConfig->GetShuttlelbTable());
1736 TSQLResult* aResult = fServer[3]->Query(sqlQuery);
1738 AliError(Form("Can't execute query <%s>!", sqlQuery.Data()));
1742 AliDebug(2,Form("Query = %s", sqlQuery.Data()));
1744 if (aResult->GetRowCount() == 0) {
1745 AliError(Form("No result for query %s received", sqlQuery.Data()));
1749 if (aResult->GetFieldCount() != 1) {
1750 AliError(Form("Invalid field count for query %s received", sqlQuery.Data()));
1754 TSQLRow* aRow = aResult->Next();
1756 AliError(Form("Could not receive result of query %s", sqlQuery.Data()));
1760 TString result(aRow->GetField(0), aRow->GetFieldLength(0));
1761 Int_t count = result.Atoi();
1763 Log("SHUTTLE", Form("%d unprocessed runs", count));
1768 TMonaLisaValue mlStatus("SHUTTLE_openruns", count);
1771 mlList.Add(&mlStatus);
1773 fMonaLisa->SendParameters(&mlList, "__PROCESSINGINFO__");
1776 //______________________________________________________________________________________________
1777 Bool_t AliShuttle::QueryShuttleLogbook(const char* whereClause,
1780 // Query DAQ's Shuttle logbook and fills detector status object.
1781 // Call QueryRunParameters to query DAQ logbook for run parameters.
1784 entries.SetOwner(1);
1786 // check connection, in case connect
1787 if (!Connect(3)) return kFALSE;
1790 sqlQuery = Form("select * from %s %s order by run", fConfig->GetShuttlelbTable(), whereClause);
1792 TSQLResult* aResult = fServer[3]->Query(sqlQuery);
1794 AliError(Form("Can't execute query <%s>!", sqlQuery.Data()));
1798 AliDebug(2,Form("Query = %s", sqlQuery.Data()));
1800 if(aResult->GetRowCount() == 0) {
1801 Log("SHUTTLE", "No entries in Shuttle Logbook match request");
1806 // TODO Check field count!
1807 const UInt_t nCols = 23;
1808 if (aResult->GetFieldCount() != (Int_t) nCols) {
1809 Log("SHUTTLE", "Invalid SQL result field number!");
1815 while ((aRow = aResult->Next())) {
1816 TString runString(aRow->GetField(0), aRow->GetFieldLength(0));
1817 Int_t run = runString.Atoi();
1819 AliShuttleLogbookEntry *entry = QueryRunParameters(run);
1823 // loop on detectors
1824 for(UInt_t ii = 0; ii < nCols; ii++)
1825 entry->SetDetectorStatus(aResult->GetFieldName(ii), aRow->GetField(ii));
1827 entries.AddLast(entry);
1835 //______________________________________________________________________________________________
1836 AliShuttleLogbookEntry* AliShuttle::QueryRunParameters(Int_t run)
1839 // Retrieve run parameters written in the DAQ logbook and sets them into AliShuttleLogbookEntry object
1842 // check connection, in case connect
1847 sqlQuery.Form("select * from %s where run=%d", fConfig->GetDAQlbTable(), run);
1849 TSQLResult* aResult = fServer[3]->Query(sqlQuery);
1851 Log("SHUTTLE", Form("Can't execute query <%s>!", sqlQuery.Data()));
1855 if (aResult->GetRowCount() == 0) {
1856 Log("SHUTTLE", Form("QueryRunParameters - No entry in DAQ Logbook for run %d. Skipping", run));
1861 if (aResult->GetRowCount() > 1) {
1862 Log("SHUTTLE", Form("QueryRunParameters - UNEXPECTED: "
1863 "more than one entry in DAQ Logbook for run %d!", run));
1868 TSQLRow* aRow = aResult->Next();
1871 Log("SHUTTLE", Form("QueryRunParameters - Could not retrieve row for run %d. Skipping", run));
1876 AliShuttleLogbookEntry* entry = new AliShuttleLogbookEntry(run);
1878 for (Int_t ii = 0; ii < aResult->GetFieldCount(); ii++)
1879 entry->SetRunParameter(aResult->GetFieldName(ii), aRow->GetField(ii));
1881 UInt_t startTime = entry->GetStartTime();
1882 UInt_t endTime = entry->GetEndTime();
1884 // if (!startTime || !endTime || startTime > endTime)
1887 // Form("QueryRunParameters - Invalid parameters for Run %d: startTime = %d, endTime = %d. Skipping!",
1888 // run, startTime, endTime));
1890 // Log("SHUTTLE", Form("Marking SHUTTLE done for run %d", run));
1891 // fLogbookEntry = entry;
1892 // if (!UpdateShuttleLogbook("shuttle_done"))
1894 // AliError(Form("Could not update logbook for run %d !", run));
1896 // fLogbookEntry = 0;
1907 Form("QueryRunParameters - Invalid parameters for Run %d: "
1908 "startTime = %d, endTime = %d. Skipping!",
1909 run, startTime, endTime));
1911 Log("SHUTTLE", Form("Marking SHUTTLE done for run %d", run));
1912 fLogbookEntry = entry;
1913 if (!UpdateShuttleLogbook("shuttle_ignored"))
1915 AliError(Form("Could not update logbook for run %d !", run));
1925 if (startTime && !endTime)
1927 // TODO Here we don't mark SHUTTLE done, because this may mean
1928 //the run is still ongoing!!
1930 Form("QueryRunParameters - Invalid parameters for Run %d: "
1931 "startTime = %d, endTime = %d. Skipping (Shuttle won't be marked as DONE)!",
1932 run, startTime, endTime));
1934 //Log("SHUTTLE", Form("Marking SHUTTLE done for run %d", run));
1935 //fLogbookEntry = entry;
1936 //if (!UpdateShuttleLogbook("shuttle_done"))
1938 // AliError(Form("Could not update logbook for run %d !", run));
1940 //fLogbookEntry = 0;
1948 if (startTime && endTime && (startTime > endTime))
1951 Form("QueryRunParameters - Invalid parameters for Run %d: "
1952 "startTime = %d, endTime = %d. Skipping!",
1953 run, startTime, endTime));
1955 Log("SHUTTLE", Form("Marking SHUTTLE done for run %d", run));
1956 fLogbookEntry = entry;
1957 if (!UpdateShuttleLogbook("shuttle_ignored"))
1959 AliError(Form("Could not update logbook for run %d !", run));
1969 TString totEventsStr = entry->GetRunParameter("totalEvents");
1970 Int_t totEvents = totEventsStr.Atoi();
1974 Form("QueryRunParameters - Run %d has 0 events - Skipping!", run));
1976 Log("SHUTTLE", Form("Marking SHUTTLE done for run %d", run));
1977 fLogbookEntry = entry;
1978 if (!UpdateShuttleLogbook("shuttle_ignored"))
1980 AliError(Form("Could not update logbook for run %d !", run));
1996 //______________________________________________________________________________________________
1997 TMap* AliShuttle::GetValueSet(const char* host, Int_t port, const TSeqCollection* entries,
1998 DCSType type, Int_t multiSplit)
2000 // Retrieve all "entry" data points from the DCS server
2001 // host, port: TSocket connection parameters
2002 // entries: list of name of the alias or data point
2003 // type: kAlias or kDP
2004 // returns TMap of values, 0 when failure
2006 AliDCSClient client(host, port, fTimeout, fRetries, multiSplit);
2011 result = client.GetAliasValues(entries, GetCurrentStartTime(),
2012 GetCurrentEndTime());
2014 else if (type == kDP)
2016 result = client.GetDPValues(entries, GetCurrentStartTime(),
2017 GetCurrentEndTime());
2022 Log(fCurrentDetector.Data(), Form("GetValueSet - Can't get entries! Reason: %s",
2023 client.GetErrorString(client.GetResultErrorCode())));
2024 if (client.GetResultErrorCode() == AliDCSClient::fgkServerError)
2025 Log(fCurrentDetector.Data(), Form("GetValueSet - Server error code: %s",
2026 client.GetServerError().Data()));
2034 //______________________________________________________________________________________________
2035 const char* AliShuttle::GetFile(Int_t system, const char* detector,
2036 const char* id, const char* source)
2038 // Get calibration file from file exchange servers
2039 // First queris the FXS database for the file name, using the run, detector, id and source info
2040 // then calls RetrieveFile(filename) for actual copy to local disk
2041 // run: current run being processed (given by Logbook entry fLogbookEntry)
2042 // detector: the Preprocessor name
2043 // id: provided as a parameter by the Preprocessor
2044 // source: provided by the Preprocessor through GetFileSources function
2046 // check if test mode should simulate a FXS error
2047 if (fTestMode & kErrorFXSFiles)
2049 Log(detector, Form("GetFile - In TESTMODE - Simulating error while connecting to %s FXS", GetSystemName(system)));
2053 // check connection, in case connect
2054 if (!Connect(system))
2056 Log(detector, Form("GetFile - Couldn't connect to %s FXS database", GetSystemName(system)));
2061 // Query preparation
2062 TString sourceName(source);
2064 TString sqlQueryStart = Form("select filePath,size,fileChecksum from %s where",
2065 fConfig->GetFXSdbTable(system));
2066 TString whereClause = Form("run=%d and detector=\"%s\" and fileId=\"%s\"",
2067 GetCurrentRun(), detector, id);
2071 whereClause += Form(" and DAQsource=\"%s\"", source);
2073 else if (system == kDCS)
2077 else if (system == kHLT)
2079 whereClause += Form(" and DDLnumbers=\"%s\"", source);
2082 TString sqlQuery = Form("%s %s", sqlQueryStart.Data(), whereClause.Data());
2084 AliDebug(2, Form("SQL query: \n%s",sqlQuery.Data()));
2087 TSQLResult* aResult = 0;
2088 aResult = dynamic_cast<TSQLResult*> (fServer[system]->Query(sqlQuery));
2090 Log(detector, Form("GetFile - Can't execute SQL query to %s database for: id = %s, source = %s",
2091 GetSystemName(system), id, sourceName.Data()));
2096 if (aResult->GetRowCount() == 0)
2099 Form("GetFile - No entry in %s FXS db for: id = %s, source = %s",
2100 GetSystemName(system), id, sourceName.Data()));
2105 if (aResult->GetRowCount() > 1) {
2107 Form("GetFile - More than one entry in %s FXS db for: id = %s, source = %s",
2108 GetSystemName(system), id, sourceName.Data()));
2114 if (aResult->GetFieldCount() != nFields) {
2116 Form("GetFileName - Wrong field count in %s FXS db for: id = %s, source = %s",
2117 GetSystemName(system), id, sourceName.Data()));
2123 TSQLRow* aRow = dynamic_cast<TSQLRow*> (aResult->Next());
2126 Log(detector, Form("GetFile - Empty set result in %s FXS db from query: id = %s, source = %s",
2127 GetSystemName(system), id, sourceName.Data()));
2133 TString filePath(aRow->GetField(0), aRow->GetFieldLength(0));
2134 TString fileSize(aRow->GetField(1), aRow->GetFieldLength(1));
2135 TString fileChecksum(aRow->GetField(2), aRow->GetFieldLength(2));
2140 AliDebug(2, Form("filePath = %s; size = %s, fileChecksum = %s",
2141 filePath.Data(), fileSize.Data(), fileChecksum.Data()));
2143 // retrieved file is renamed to make it unique
2144 TString localFileName = Form("%s/%s_%d_process/%s_%s_%d_%s_%s.shuttle",
2145 GetShuttleTempDir(), detector, GetCurrentRun(),
2146 GetSystemName(system), detector, GetCurrentRun(),
2147 id, sourceName.Data());
2150 // file retrieval from FXS
2151 UInt_t nRetries = 0;
2152 UInt_t maxRetries = 3;
2153 Bool_t result = kFALSE;
2155 // copy!! if successful TSystem::Exec returns 0
2156 while (nRetries++ < maxRetries) {
2157 AliDebug(2, Form("Trying to copy file. Retry # %d", nRetries));
2158 result = RetrieveFile(system, filePath.Data(), localFileName.Data());
2161 Log(detector, Form("GetFile - Copy of file %s from %s FXS failed",
2162 filePath.Data(), GetSystemName(system)));
2166 if (fileSize.Length()>0)
2168 // compare filesize of local file with the one stored in the FXS DB
2170 Int_t sizeComp = gSystem->GetPathInfo(localFileName.Data(), 0, &size, 0, 0);
2172 if (sizeComp != 0 || size != fileSize.Atoi())
2174 Log(detector, Form("GetFile - size of file %s does not match with local copy!",
2181 Log(fCurrentDetector, Form("GetFile - size of file %s not set in %s database, skipping comparison",
2182 filePath.Data(), GetSystemName(system)));
2185 if (fileChecksum.Length()>0)
2187 // compare md5sum of local file with the one stored in the FXS DB
2188 if(fileChecksum.Contains(' ')) fileChecksum.Resize(fileChecksum.First(' '));
2189 Int_t md5Comp = gSystem->Exec(Form("md5sum %s |grep %s > /dev/null 2> /dev/null",
2190 localFileName.Data(), fileChecksum.Data()));
2194 Log(detector, Form("GetFile - md5sum of file %s does not match with local copy!",
2200 Log(fCurrentDetector, Form("GetFile - md5sum of file %s not set in %s database, skipping comparison",
2201 filePath.Data(), GetSystemName(system)));
2212 fFXSCalled[system]=kTRUE;
2213 TObjString *fileParams = new TObjString(Form("%s#!?!#%s", id, sourceName.Data()));
2214 fFXSlist[system].Add(fileParams);
2216 static TString staticLocalFileName;
2217 staticLocalFileName.Form("%s", localFileName.Data());
2219 Log(fCurrentDetector, Form("GetFile - Retrieved file with id %s and "
2220 "source %s from %s to %s", id, source,
2221 GetSystemName(system), localFileName.Data()));
2223 return staticLocalFileName.Data();
2226 //______________________________________________________________________________________________
2227 Bool_t AliShuttle::RetrieveFile(UInt_t system, const char* fxsFileName, const char* localFileName)
2230 // Copies file from FXS to local Shuttle machine
2233 // check temp directory: trying to cd to temp; if it does not exist, create it
2234 AliDebug(2, Form("Copy file %s from %s FXS into %s",
2235 GetSystemName(system), fxsFileName, localFileName));
2237 TString tmpDir(localFileName);
2239 tmpDir = tmpDir(0,tmpDir.Last('/'));
2241 Int_t noDir = gSystem->GetPathInfo(tmpDir.Data(), 0, (Long64_t*) 0, 0, 0);
2242 if (noDir) // temp dir does not exists!
2244 if (gSystem->mkdir(tmpDir.Data(), 1))
2246 Log(fCurrentDetector.Data(), "RetrieveFile - could not make temp directory!!");
2251 TString command = Form("scp -oPort=%d -2 %s@%s:%s/%s %s",
2252 fConfig->GetFXSPort(system),
2253 fConfig->GetFXSUser(system),
2254 fConfig->GetFXSHost(system),
2255 fConfig->GetFXSBaseFolder(system),
2259 AliDebug(2, Form("%s",command.Data()));
2261 Bool_t result = (gSystem->Exec(command.Data()) == 0);
2266 //______________________________________________________________________________________________
2267 TList* AliShuttle::GetFileSources(Int_t system, const char* detector, const char* id)
2270 // Get sources producing the condition file Id from file exchange servers
2271 // if id is NULL all sources are returned (distinct)
2276 Log(detector, Form("GetFileSources - Querying %s FXS for files with id %s produced by %s", GetSystemName(system), id, detector));
2278 Log(detector, Form("GetFileSources - Querying %s FXS for files produced by %s", GetSystemName(system), detector));
2281 // check if test mode should simulate a FXS error
2282 if (fTestMode & kErrorFXSSources)
2284 Log(detector, Form("GetFileSources - In TESTMODE - Simulating error while connecting to %s FXS", GetSystemName(system)));
2290 Log(detector, "GetFileSources - WARNING: DCS system has only one source of data!");
2291 TList *list = new TList();
2293 list->Add(new TObjString(" "));
2297 // check connection, in case connect
2298 if (!Connect(system))
2300 Log(detector, Form("GetFileSources - Couldn't connect to %s FXS database", GetSystemName(system)));
2305 TString sourceName = 0;
2308 sourceName = "DAQsource";
2309 } else if (system == kHLT)
2311 sourceName = "DDLnumbers";
2314 TString sqlQueryStart = Form("select distinct %s from %s where", sourceName.Data(), fConfig->GetFXSdbTable(system));
2315 TString whereClause = Form("run=%d and detector=\"%s\"",
2316 GetCurrentRun(), detector);
2318 whereClause += Form(" and fileId=\"%s\"", id);
2319 TString sqlQuery = Form("%s %s", sqlQueryStart.Data(), whereClause.Data());
2321 AliDebug(2, Form("SQL query: \n%s",sqlQuery.Data()));
2324 TSQLResult* aResult;
2325 aResult = fServer[system]->Query(sqlQuery);
2327 Log(detector, Form("GetFileSources - Can't execute SQL query to %s database for id: %s",
2328 GetSystemName(system), id));
2333 TList *list = new TList();
2336 if (aResult->GetRowCount() == 0)
2339 Form("GetFileSources - No entry in %s FXS table for id: %s", GetSystemName(system), id));
2344 Log(detector, Form("GetFileSources - Found %d sources", aResult->GetRowCount()));
2347 while ((aRow = aResult->Next()))
2350 TString source(aRow->GetField(0), aRow->GetFieldLength(0));
2351 AliDebug(2, Form("%s = %s", sourceName.Data(), source.Data()));
2352 list->Add(new TObjString(source));
2361 //______________________________________________________________________________________________
2362 TList* AliShuttle::GetFileIDs(Int_t system, const char* detector, const char* source)
2365 // Get all ids of condition files produced by a given source from file exchange servers
2368 Log(detector, Form("GetFileIDs - Retrieving ids with source %s with %s", source, GetSystemName(system)));
2370 // check if test mode should simulate a FXS error
2371 if (fTestMode & kErrorFXSSources)
2373 Log(detector, Form("GetFileIDs - In TESTMODE - Simulating error while connecting to %s FXS", GetSystemName(system)));
2377 // check connection, in case connect
2378 if (!Connect(system))
2380 Log(detector, Form("GetFileIDs - Couldn't connect to %s FXS database", GetSystemName(system)));
2384 TString sourceName = 0;
2387 sourceName = "DAQsource";
2388 } else if (system == kHLT)
2390 sourceName = "DDLnumbers";
2393 TString sqlQueryStart = Form("select fileId from %s where", fConfig->GetFXSdbTable(system));
2394 TString whereClause = Form("run=%d and detector=\"%s\"",
2395 GetCurrentRun(), detector);
2396 if (sourceName.Length() > 0 && source)
2397 whereClause += Form(" and %s=\"%s\"", sourceName.Data(), source);
2398 TString sqlQuery = Form("%s %s", sqlQueryStart.Data(), whereClause.Data());
2400 AliDebug(2, Form("SQL query: \n%s",sqlQuery.Data()));
2403 TSQLResult* aResult;
2404 aResult = fServer[system]->Query(sqlQuery);
2406 Log(detector, Form("GetFileIDs - Can't execute SQL query to %s database for source: %s",
2407 GetSystemName(system), source));
2411 TList *list = new TList();
2414 if (aResult->GetRowCount() == 0)
2417 Form("GetFileIDs - No entry in %s FXS table for source: %s", GetSystemName(system), source));
2422 Log(detector, Form("GetFileIDs - Found %d ids", aResult->GetRowCount()));
2426 while ((aRow = aResult->Next()))
2429 TString id(aRow->GetField(0), aRow->GetFieldLength(0));
2430 AliDebug(2, Form("fileId = %s", id.Data()));
2431 list->Add(new TObjString(id));
2440 //______________________________________________________________________________________________
2441 Bool_t AliShuttle::Connect(Int_t system)
2443 // Connect to MySQL Server of the system's FXS MySQL databases
2444 // DAQ Logbook, Shuttle Logbook and DAQ FXS db are on the same host
2447 // check connection: if already connected return
2448 if(fServer[system] && fServer[system]->IsConnected()) return kTRUE;
2450 TString dbHost, dbUser, dbPass, dbName;
2452 if (system < 3) // FXS db servers
2454 dbHost = Form("mysql://%s:%d", fConfig->GetFXSdbHost(system), fConfig->GetFXSdbPort(system));
2455 dbUser = fConfig->GetFXSdbUser(system);
2456 dbPass = fConfig->GetFXSdbPass(system);
2457 dbName = fConfig->GetFXSdbName(system);
2458 } else { // Run & Shuttle logbook servers
2459 // TODO Will the Shuttle logbook server be the same as the Run logbook server ???
2460 dbHost = Form("mysql://%s:%d", fConfig->GetDAQlbHost(), fConfig->GetDAQlbPort());
2461 dbUser = fConfig->GetDAQlbUser();
2462 dbPass = fConfig->GetDAQlbPass();
2463 dbName = fConfig->GetDAQlbDB();
2466 fServer[system] = TSQLServer::Connect(dbHost.Data(), dbUser.Data(), dbPass.Data());
2467 if (!fServer[system] || !fServer[system]->IsConnected()) {
2470 AliError(Form("Can't establish connection to FXS database for %s",
2471 AliShuttleInterface::GetSystemName(system)));
2473 AliError("Can't establish connection to Run logbook.");
2475 if(fServer[system]) delete fServer[system];
2480 TSQLResult* aResult=0;
2483 aResult = fServer[kDAQ]->GetTables(dbName.Data());
2486 aResult = fServer[kDCS]->GetTables(dbName.Data());
2489 aResult = fServer[kHLT]->GetTables(dbName.Data());
2492 aResult = fServer[3]->GetTables(dbName.Data());
2500 //______________________________________________________________________________________________
2501 Bool_t AliShuttle::UpdateTable()
2504 // Update FXS table filling time_processed field in all rows corresponding to current run and detector
2507 Bool_t result = kTRUE;
2509 for (UInt_t system=0; system<3; system++)
2511 if(!fFXSCalled[system]) continue;
2513 // check connection, in case connect
2514 if (!Connect(system))
2516 Log(fCurrentDetector, Form("UpdateTable - Couldn't connect to %s FXS database", GetSystemName(system)));
2521 TTimeStamp now; // now
2523 // Loop on FXS list entries
2524 TIter iter(&fFXSlist[system]);
2525 TObjString *aFXSentry=0;
2526 while ((aFXSentry = dynamic_cast<TObjString*> (iter.Next())))
2528 TString aFXSentrystr = aFXSentry->String();
2529 TObjArray *aFXSarray = aFXSentrystr.Tokenize("#!?!#");
2530 if (!aFXSarray || aFXSarray->GetEntries() != 2 )
2532 Log(fCurrentDetector, Form("UpdateTable - error updating %s FXS entry. Check string: <%s>",
2533 GetSystemName(system), aFXSentrystr.Data()));
2534 if(aFXSarray) delete aFXSarray;
2538 const char* fileId = ((TObjString*) aFXSarray->At(0))->GetName();
2539 const char* source = ((TObjString*) aFXSarray->At(1))->GetName();
2541 TString whereClause;
2544 whereClause = Form("where run=%d and detector=\"%s\" and fileId=\"%s\" and DAQsource=\"%s\";",
2545 GetCurrentRun(), fCurrentDetector.Data(), fileId, source);
2547 else if (system == kDCS)
2549 whereClause = Form("where run=%d and detector=\"%s\" and fileId=\"%s\";",
2550 GetCurrentRun(), fCurrentDetector.Data(), fileId);
2552 else if (system == kHLT)
2554 whereClause = Form("where run=%d and detector=\"%s\" and fileId=\"%s\" and DDLnumbers=\"%s\";",
2555 GetCurrentRun(), fCurrentDetector.Data(), fileId, source);
2560 TString sqlQuery = Form("update %s set time_processed=%d %s", fConfig->GetFXSdbTable(system),
2561 now.GetSec(), whereClause.Data());
2563 AliDebug(2, Form("SQL query: \n%s",sqlQuery.Data()));
2566 TSQLResult* aResult;
2567 aResult = dynamic_cast<TSQLResult*> (fServer[system]->Query(sqlQuery));
2570 Log(fCurrentDetector, Form("UpdateTable - %s db: can't execute SQL query <%s>",
2571 GetSystemName(system), sqlQuery.Data()));
2582 //______________________________________________________________________________________________
2583 Bool_t AliShuttle::UpdateTableFailCase()
2585 // Update FXS table filling time_processed field in all rows corresponding to current run and detector
2586 // this is called in case the preprocessor is declared failed for the current run, because
2587 // the fields are updated only in case of success
2589 Bool_t result = kTRUE;
2591 for (UInt_t system=0; system<3; system++)
2593 // check connection, in case connect
2594 if (!Connect(system))
2596 Log(fCurrentDetector, Form("UpdateTableFailCase - Couldn't connect to %s FXS database",
2597 GetSystemName(system)));
2602 TTimeStamp now; // now
2604 // Loop on FXS list entries
2606 TString whereClause = Form("where run=%d and detector=\"%s\";",
2607 GetCurrentRun(), fCurrentDetector.Data());
2610 TString sqlQuery = Form("update %s set time_processed=%d %s", fConfig->GetFXSdbTable(system),
2611 now.GetSec(), whereClause.Data());
2613 AliDebug(2, Form("SQL query: \n%s",sqlQuery.Data()));
2616 TSQLResult* aResult;
2617 aResult = dynamic_cast<TSQLResult*> (fServer[system]->Query(sqlQuery));
2620 Log(fCurrentDetector, Form("UpdateTableFailCase - %s db: can't execute SQL query <%s>",
2621 GetSystemName(system), sqlQuery.Data()));
2631 //______________________________________________________________________________________________
2632 Bool_t AliShuttle::UpdateShuttleLogbook(const char* detector, const char* status)
2635 // Update Shuttle logbook filling detector or shuttle_done column
2636 // ex. of usage: UpdateShuttleLogbook("PHOS", "DONE") or UpdateShuttleLogbook("shuttle_done")
2639 // check connection, in case connect
2641 Log("SHUTTLE", "UpdateShuttleLogbook - Couldn't connect to DAQ Logbook.");
2645 TString detName(detector);
2647 if (detName == "shuttle_done" || detName == "shuttle_ignored")
2649 setClause = "set shuttle_done=1";
2651 if (detName == "shuttle_done")
2653 // Send the information to ML
2654 TMonaLisaText mlStatus("SHUTTLE_status", "Done");
2657 mlList.Add(&mlStatus);
2660 mlID.Form("%d", GetCurrentRun());
2661 fMonaLisa->SendParameters(&mlList, mlID);
2664 TString statusStr(status);
2665 if(statusStr.Contains("done", TString::kIgnoreCase) ||
2666 statusStr.Contains("failed", TString::kIgnoreCase)){
2667 setClause = Form("set %s=\"%s\"", detector, status);
2670 Form("UpdateShuttleLogbook - Invalid status <%s> for detector %s",
2676 TString whereClause = Form("where run=%d", GetCurrentRun());
2678 TString sqlQuery = Form("update %s %s %s",
2679 fConfig->GetShuttlelbTable(), setClause.Data(), whereClause.Data());
2681 AliDebug(2, Form("SQL query: \n%s",sqlQuery.Data()));
2684 TSQLResult* aResult;
2685 aResult = dynamic_cast<TSQLResult*> (fServer[3]->Query(sqlQuery));
2687 Log("SHUTTLE", Form("UpdateShuttleLogbook - Can't execute query <%s>", sqlQuery.Data()));
2695 //______________________________________________________________________________________________
2696 Int_t AliShuttle::GetCurrentRun() const
2699 // Get current run from logbook entry
2702 return fLogbookEntry ? fLogbookEntry->GetRun() : -1;
2705 //______________________________________________________________________________________________
2706 UInt_t AliShuttle::GetCurrentStartTime() const
2709 // get current start time
2712 return fLogbookEntry ? fLogbookEntry->GetStartTime() : 0;
2715 //______________________________________________________________________________________________
2716 UInt_t AliShuttle::GetCurrentEndTime() const
2719 // get current end time from logbook entry
2722 return fLogbookEntry ? fLogbookEntry->GetEndTime() : 0;
2725 //______________________________________________________________________________________________
2726 UInt_t AliShuttle::GetCurrentYear() const
2729 // Get current year from logbook entry
2732 if (!fLogbookEntry) return 0;
2734 TTimeStamp startTime(GetCurrentStartTime());
2735 TString year = Form("%d",startTime.GetDate());
2741 //______________________________________________________________________________________________
2742 const char* AliShuttle::GetLHCPeriod() const
2745 // Get current LHC period from logbook entry
2748 if (!fLogbookEntry) return 0;
2750 return fLogbookEntry->GetRunParameter("LHCperiod");
2753 //______________________________________________________________________________________________
2754 void AliShuttle::Log(const char* detector, const char* message)
2757 // Fill log string with a message
2760 TString logRunDir = GetShuttleLogDir();
2761 if (GetCurrentRun() >=0)
2762 logRunDir += Form("/%d", GetCurrentRun());
2764 void* dir = gSystem->OpenDirectory(logRunDir.Data());
2766 if (gSystem->mkdir(logRunDir.Data(), kTRUE)) {
2767 AliError(Form("Can't open directory <%s>", GetShuttleLogDir()));
2772 gSystem->FreeDirectory(dir);
2775 TString toLog = Form("%s (%d): %s - ", TTimeStamp(time(0)).AsString("s"), getpid(), detector);
2776 if (GetCurrentRun() >= 0)
2777 toLog += Form("run %d - ", GetCurrentRun());
2778 toLog += Form("%s", message);
2780 AliInfo(toLog.Data());
2782 // if we redirect the log output already to the file, leave here
2783 if (fOutputRedirected && strcmp(detector, "SHUTTLE") != 0)
2786 TString fileName = GetLogFileName(detector);
2788 gSystem->ExpandPathName(fileName);
2791 logFile.open(fileName, ofstream::out | ofstream::app);
2793 if (!logFile.is_open()) {
2794 AliError(Form("Could not open file %s", fileName.Data()));
2798 logFile << toLog.Data() << "\n";
2803 //______________________________________________________________________________________________
2804 TString AliShuttle::GetLogFileName(const char* detector) const
2807 // returns the name of the log file for a given sub detector
2812 if (GetCurrentRun() >= 0)
2814 fileName.Form("%s/%d/%s_%d.log", GetShuttleLogDir(), GetCurrentRun(),
2815 detector, GetCurrentRun());
2817 fileName.Form("%s/%s.log", GetShuttleLogDir(), detector);
2823 //______________________________________________________________________________________________
2824 void AliShuttle::SendAlive()
2826 // sends alive message to ML
2828 TMonaLisaText mlStatus("SHUTTLE_status", "Alive");
2831 mlList.Add(&mlStatus);
2833 fMonaLisa->SendParameters(&mlList, "__PROCESSINGINFO__");
2836 //______________________________________________________________________________________________
2837 Bool_t AliShuttle::Collect(Int_t run)
2840 // Collects conditions data for all UNPROCESSED run written to DAQ LogBook in case of run = -1 (default)
2841 // If a dedicated run is given this run is processed
2843 // In operational mode, this is the Shuttle function triggered by the EOR signal.
2847 Log("SHUTTLE","Collect - Shuttle called. Collecting conditions data for unprocessed runs");
2849 Log("SHUTTLE", Form("Collect - Shuttle called. Collecting conditions data for run %d", run));
2851 SetLastAction("Starting");
2853 // create ML instance
2855 fMonaLisa = new TMonaLisaWriter(fConfig->GetMonitorHost(), fConfig->GetMonitorTable());
2860 TString whereClause("where shuttle_done=0");
2862 whereClause += Form(" and run=%d", run);
2864 TObjArray shuttleLogbookEntries;
2865 if (!QueryShuttleLogbook(whereClause, shuttleLogbookEntries))
2867 Log("SHUTTLE", "Collect - Can't retrieve entries from Shuttle logbook");
2871 if (shuttleLogbookEntries.GetEntries() == 0)
2874 Log("SHUTTLE","Collect - Found no UNPROCESSED runs in Shuttle logbook");
2876 Log("SHUTTLE", Form("Collect - Run %d is already DONE "
2877 "or it does not exist in Shuttle logbook", run));
2881 for (UInt_t iDet=0; iDet<NDetectors(); iDet++)
2882 fFirstUnprocessed[iDet] = kTRUE;
2886 // query Shuttle logbook for earlier runs, check if some detectors are unprocessed,
2887 // flag them into fFirstUnprocessed array
2888 TString whereClause(Form("where shuttle_done=0 and run < %d", run));
2889 TObjArray tmpLogbookEntries;
2890 if (!QueryShuttleLogbook(whereClause, tmpLogbookEntries))
2892 Log("SHUTTLE", "Collect - Can't retrieve entries from Shuttle logbook");
2896 TIter iter(&tmpLogbookEntries);
2897 AliShuttleLogbookEntry* anEntry = 0;
2898 while ((anEntry = dynamic_cast<AliShuttleLogbookEntry*> (iter.Next())))
2900 for (UInt_t iDet=0; iDet<NDetectors(); iDet++)
2902 if (anEntry->GetDetectorStatus(iDet) == AliShuttleLogbookEntry::kUnprocessed)
2904 AliDebug(2, Form("Run %d: setting %s as \"not first time unprocessed\"",
2905 anEntry->GetRun(), GetDetName(iDet)));
2906 fFirstUnprocessed[iDet] = kFALSE;
2914 if (!RetrieveConditionsData(shuttleLogbookEntries))
2916 Log("SHUTTLE", "Collect - Process of at least one run failed");
2921 Log("SHUTTLE", "Collect - Requested run(s) successfully processed");
2926 //______________________________________________________________________________________________
2927 Bool_t AliShuttle::RetrieveConditionsData(const TObjArray& dateEntries)
2930 // Retrieve conditions data for all runs that aren't processed yet
2933 Bool_t hasError = kFALSE;
2935 TIter iter(&dateEntries);
2936 AliShuttleLogbookEntry* anEntry;
2938 while ((anEntry = (AliShuttleLogbookEntry*) iter.Next())){
2939 if (!Process(anEntry)){
2943 // clean SHUTTLE temp directory
2944 //TString filename = Form("%s/*.shuttle", GetShuttleTempDir());
2945 //RemoveFile(filename.Data());
2948 return hasError == kFALSE;
2951 //______________________________________________________________________________________________
2952 ULong_t AliShuttle::GetTimeOfLastAction() const
2955 // Gets time of last action
2960 fMonitoringMutex->Lock();
2962 tmp = fLastActionTime;
2964 fMonitoringMutex->UnLock();
2969 //______________________________________________________________________________________________
2970 const TString AliShuttle::GetLastAction() const
2973 // returns a string description of the last action
2978 fMonitoringMutex->Lock();
2982 fMonitoringMutex->UnLock();
2987 //______________________________________________________________________________________________
2988 void AliShuttle::SetLastAction(const char* action)
2991 // updates the monitoring variables
2994 fMonitoringMutex->Lock();
2996 fLastAction = action;
2997 fLastActionTime = time(0);
2999 fMonitoringMutex->UnLock();
3002 //______________________________________________________________________________________________
3003 const char* AliShuttle::GetRunParameter(const char* param)
3006 // returns run parameter read from DAQ logbook
3009 if(!fLogbookEntry) {
3010 AliError("No logbook entry!");
3014 return fLogbookEntry->GetRunParameter(param);
3017 //______________________________________________________________________________________________
3018 AliCDBEntry* AliShuttle::GetFromOCDB(const char* detector, const AliCDBPath& path)
3021 // returns object from OCDB valid for current run
3024 if (fTestMode & kErrorOCDB)
3026 Log(detector, "GetFromOCDB - In TESTMODE - Simulating error with OCDB");
3030 AliCDBStorage *sto = AliCDBManager::Instance()->GetStorage(fgkMainCDB);
3033 Log(detector, "GetFromOCDB - Cannot activate main OCDB for query!");
3037 return dynamic_cast<AliCDBEntry*> (sto->Get(path, GetCurrentRun()));
3040 //______________________________________________________________________________________________
3041 Bool_t AliShuttle::SendMail(EMailTarget target, Int_t system)
3044 // sends a mail to the subdetector expert in case of preprocessor error
3047 if (fTestMode != kNone)
3050 if (!fConfig->SendMail())
3053 if (target == kDCSEMail || target == kFXSEMail) {
3054 if (!fFirstProcessing)
3058 void* dir = gSystem->OpenDirectory(GetShuttleLogDir());
3061 if (gSystem->mkdir(GetShuttleLogDir(), kTRUE))
3063 Log("SHUTTLE", Form("SendMail - Can't open directory <%s>", GetShuttleLogDir()));
3068 gSystem->FreeDirectory(dir);
3071 // det experts in to
3074 if (target == kDCSEMail) {
3075 iterExperts = new TIter(fConfig->GetAdmins(AliShuttleConfig::kAmanda));
3077 else if (target == kFXSEMail) {
3078 iterExperts = new TIter(fConfig->GetAdmins(system));
3081 iterExperts = new TIter(fConfig->GetResponsibles(fCurrentDetector));
3083 TObjString *anExpert=0;
3084 while ((anExpert = (TObjString*) iterExperts->Next()))
3086 to += Form("%s,", anExpert->GetName());
3089 if (to.Length() > 0)
3090 to.Remove(to.Length()-1);
3091 AliDebug(2, Form("to: %s",to.Data()));
3094 Log("SHUTTLE", Form("List of %d responsibles not set!", (Int_t) target));
3098 // SHUTTLE responsibles in cc
3100 TIter iterAdmins(fConfig->GetAdmins(AliShuttleConfig::kGlobal));
3101 TObjString *anAdmin=0;
3102 while ((anAdmin = (TObjString*) iterAdmins.Next()))
3104 cc += Form("%s,", anAdmin->GetName());
3106 if (cc.Length() > 0)
3107 cc.Remove(cc.Length()-1);
3108 AliDebug(2, Form("cc: %s",to.Data()));
3111 TString bodyFileName;
3112 bodyFileName.Form("%s/mail.body", GetShuttleLogDir());
3113 gSystem->ExpandPathName(bodyFileName);
3116 mailBody.open(bodyFileName, ofstream::out);
3118 if (!mailBody.is_open())
3120 Log("SHUTTLE", Form("Could not open mail body file %s", bodyFileName.Data()));
3128 if (target == kDCSEMail){
3129 subject = Form("Retrieval of data points for %s FAILED in run %d !",
3130 fCurrentDetector.Data(), GetCurrentRun());
3131 AliDebug(2, Form("subject: %s", subject.Data()));
3133 body = Form("Dear DCS experts, \n\n");
3134 body += Form("SHUTTLE couldn\'t retrieve the data points for detector %s "
3135 "in run %d!!\n\n", fCurrentDetector.Data(), GetCurrentRun());
3137 else if (target == kFXSEMail){
3138 subject = Form("FXS communication for %s FAILED in run %d !",
3139 fCurrentDetector.Data(), GetCurrentRun());
3140 AliDebug(2, Form("subject: %s", subject.Data()));
3142 if (system == kDAQ) sys="DAQ";
3143 else if (system == kDCS) sys="DCS";
3144 else if (system == kHLT) sys="HLT";
3146 body = Form("Dear %s FXS experts, \n\n",sys.Data());
3147 body += Form("SHUTTLE couldn\'t retrieve data from the FXS for detector %s "
3148 "in run %d!!\n\n", fCurrentDetector.Data(), GetCurrentRun());
3151 subject = Form("%s Shuttle preprocessor FAILED in run %d (run type = %s)!",
3152 fCurrentDetector.Data(), GetCurrentRun(), GetRunType());
3153 AliDebug(2, Form("subject: %s", subject.Data()));
3155 body = Form("Dear %s expert(s), \n\n", fCurrentDetector.Data());
3156 body += Form("SHUTTLE just detected that your preprocessor "
3157 "failed processing run %d (run type = %s)!!\n\n",
3158 GetCurrentRun(), GetRunType());
3161 body += Form("Please check %s status on the SHUTTLE monitoring page: \n\n",
3162 fCurrentDetector.Data());
3163 if (fConfig->GetRunMode() == AliShuttleConfig::kTest)
3165 body += Form("\thttp://pcalimonitor.cern.ch:8889/shuttle.jsp?time=168 \n\n");
3167 body += Form("\thttp://pcalimonitor.cern.ch/shuttle.jsp?instance=PROD&time=168 \n\n");
3171 TString logFolder = "logs";
3172 if (fConfig->GetRunMode() == AliShuttleConfig::kProd)
3173 logFolder += "_PROD";
3176 body += Form("Find the %s log for the current run on \n\n"
3177 "\thttp://pcalishuttle01.cern.ch:8880/%s/%d/%s_%d.log \n\n",
3178 fCurrentDetector.Data(), logFolder.Data(), GetCurrentRun(),
3179 fCurrentDetector.Data(), GetCurrentRun());
3180 body += Form("The last 10 lines of %s log file are following:\n\n", fCurrentDetector.Data());
3182 AliDebug(2, Form("Body begin: %s", body.Data()));
3184 mailBody << body.Data();
3186 mailBody.open(bodyFileName, ofstream::out | ofstream::app);
3188 TString logFileName = Form("%s/%d/%s_%d.log", GetShuttleLogDir(),
3189 GetCurrentRun(), fCurrentDetector.Data(), GetCurrentRun());
3190 TString tailCommand = Form("tail -n 10 %s >> %s", logFileName.Data(), bodyFileName.Data());
3191 if (gSystem->Exec(tailCommand.Data()))
3193 mailBody << Form("%s log file not found ...\n\n", fCurrentDetector.Data());
3196 TString endBody = Form("------------------------------------------------------\n\n");
3197 endBody += Form("In case of problems please contact the SHUTTLE core team.\n\n");
3198 endBody += "Please do not answer this message directly, it is automatically generated.\n\n";
3199 endBody += "Greetings,\n\n \t\t\tthe SHUTTLE\n";
3201 AliDebug(2, Form("Body end: %s", endBody.Data()));
3203 mailBody << endBody.Data();
3208 TString mailCommand = Form("mail -s \"%s\" -c %s %s < %s",
3212 bodyFileName.Data());
3213 AliDebug(2, Form("mail command: %s", mailCommand.Data()));
3215 Bool_t result = gSystem->Exec(mailCommand.Data());
3220 //______________________________________________________________________________________________
3221 const char* AliShuttle::GetRunType()
3224 // returns run type read from "run type" logbook
3227 if(!fLogbookEntry) {
3228 AliError("No logbook entry!");
3232 return fLogbookEntry->GetRunType();
3235 //______________________________________________________________________________________________
3236 Bool_t AliShuttle::GetHLTStatus()
3238 // Return HLT status (ON=1 OFF=0)
3239 // Converts the HLT status from the status string read in the run logbook (not just a bool)
3241 if(!fLogbookEntry) {
3242 AliError("No logbook entry!");
3246 // TODO implement when HLTStatus is inserted in run logbook
3247 //TString hltStatus = fLogbookEntry->GetRunParameter("HLTStatus");
3248 //if(hltStatus == "OFF") {return kFALSE};
3253 //______________________________________________________________________________________________
3254 void AliShuttle::SetShuttleTempDir(const char* tmpDir)
3257 // sets Shuttle temp directory
3260 fgkShuttleTempDir = gSystem->ExpandPathName(tmpDir);
3263 //______________________________________________________________________________________________
3264 void AliShuttle::SetShuttleLogDir(const char* logDir)
3267 // sets Shuttle log directory
3270 fgkShuttleLogDir = gSystem->ExpandPathName(logDir);