* provided "as is" without express or implied warranty. *
**************************************************************************/
-/*
-$Log$
-Revision 1.37 2007/04/10 16:53:14 jgrosseo
-redirecting sub detector stdout, stderr to sub detector log file
-
-Revision 1.35 2007/04/04 16:26:38 acolla
-1. Re-organization of function calls in TestPreprocessor to make it more meaningful.
-2. Added missing dependency in test preprocessors.
-3. in AliShuttle.cxx: processing time and memory consumption info on a single line.
-
-Revision 1.34 2007/04/04 10:33:36 jgrosseo
-1) Storing of files to the Grid is now done _after_ your preprocessors succeeded. This is transparent, which means that you can still use the same functions (Store, StoreReferenceData) to store files to the Grid. However, the Shuttle first stores them locally and transfers them after the preprocessor finished. The return code of these two functions has changed from UInt_t to Bool_t which gives you the success of the storing.
-In case of an error with the Grid, the Shuttle will retry the storing later, the preprocessor does not need to be run again.
-
-2) The meaning of the return code of the preprocessor has changed. 0 is now success and any other value means failure. This value is stored in the log and you can use it to keep details about the error condition.
-
-3) New function StoreReferenceFile to _directly_ store a file (without opening it) to the reference storage.
-
-4) The memory usage of the preprocessor is monitored. If it exceeds 2 GB it is terminated.
-
-5) New function AliPreprocessor::ProcessDCS(). If you do not need to have DCS data in all cases, you can skip the processing by implemting this function and returning kFALSE under certain conditions. E.g. if there is a certain run type.
-If you always need DCS data (like before), you do not need to implement it.
-
-6) The run type has been added to the monitoring page
-
-Revision 1.33 2007/04/03 13:56:01 acolla
-Grid Storage at the end of preprocessing. Added virtual method to disable DCS query according to the
-run type.
-
-Revision 1.32 2007/02/28 10:41:56 acolla
-Run type field added in SHUTTLE framework. Run type is read from "run type" logbook and retrieved by
-AliPreprocessor::GetRunType() function.
-Added some ldap definition files.
-
-Revision 1.30 2007/02/13 11:23:21 acolla
-Moved getters and setters of Shuttle's main OCDB/Reference, local
-OCDB/Reference, temp and log folders to AliShuttleInterface
-
-Revision 1.27 2007/01/30 17:52:42 jgrosseo
-adding monalisa monitoring
-
-Revision 1.26 2007/01/23 19:20:03 acolla
-Removed old ldif files, added TOF, MCH ldif files. Added some options in
-AliShuttleConfig::Print. Added in Ali Shuttle: SetShuttleTempDir and
-SetShuttleLogDir
-
-Revision 1.25 2007/01/15 19:13:52 acolla
-Moved some AliInfo to AliDebug in SendMail function
-
-Revision 1.21 2006/12/07 08:51:26 jgrosseo
-update (alberto):
-table, db names in ldap configuration
-added GRP preprocessor
-DCS data can also be retrieved by data point
-
-Revision 1.20 2006/11/16 16:16:48 jgrosseo
-introducing strict run ordering flag
-removed giving preprocessor name to preprocessor, they have to know their name themselves ;-)
-
-Revision 1.19 2006/11/06 14:23:04 jgrosseo
-major update (Alberto)
-o) reading of run parameters from the logbook
-o) online offline naming conversion
-o) standalone DCSclient package
-
-Revision 1.18 2006/10/20 15:22:59 jgrosseo
-o) Adding time out to the execution of the preprocessors: The Shuttle forks and the parent process monitors the child
-o) Merging Collect, CollectAll, CollectNew function
-o) Removing implementation of empty copy constructors (declaration still there!)
-
-Revision 1.17 2006/10/05 16:20:55 jgrosseo
-adapting to new CDB classes
-
-Revision 1.16 2006/10/05 15:46:26 jgrosseo
-applying to the new interface
-
-Revision 1.15 2006/10/02 16:38:39 jgrosseo
-update (alberto):
-fixed memory leaks
-storing of objects that failed to be stored to the grid before
-interfacing of shuttle status table in daq system
-
-Revision 1.14 2006/08/29 09:16:05 jgrosseo
-small update
-
-Revision 1.13 2006/08/15 10:50:00 jgrosseo
-effc++ corrections (alberto)
-
-Revision 1.12 2006/08/08 14:19:29 jgrosseo
-Update to shuttle classes (Alberto)
-
-- Possibility to set the full object's path in the Preprocessor's and
-Shuttle's Store functions
-- Possibility to extend the object's run validity in the same classes
-("startValidity" and "validityInfinite" parameters)
-- Implementation of the StoreReferenceData function to store reference
-data in a dedicated CDB storage.
-
-Revision 1.11 2006/07/21 07:37:20 jgrosseo
-last run is stored after each run
-
-Revision 1.10 2006/07/20 09:54:40 jgrosseo
-introducing status management: The processing per subdetector is divided into several steps,
-after each step the status is stored on disk. If the system crashes in any of the steps the Shuttle
-can keep track of the number of failures and skips further processing after a certain threshold is
-exceeded. These thresholds can be configured in LDAP.
-
-Revision 1.9 2006/07/19 10:09:55 jgrosseo
-new configuration, accesst to DAQ FES (Alberto)
-
-Revision 1.8 2006/07/11 12:44:36 jgrosseo
-adding parameters for extended validity range of data produced by preprocessor
-
-Revision 1.7 2006/07/10 14:37:09 jgrosseo
-small fix + todo comment
-
-Revision 1.6 2006/07/10 13:01:41 jgrosseo
-enhanced storing of last sucessfully processed run (alberto)
-
-Revision 1.5 2006/07/04 14:59:57 jgrosseo
-revision of AliDCSValue: Removed wrapper classes, reduced storage size per value by factor 2
-
-Revision 1.4 2006/06/12 09:11:16 jgrosseo
-coding conventions (Alberto)
-
-Revision 1.3 2006/06/06 14:26:40 jgrosseo
-o) removed files that were moved to STEER
-o) shuttle updated to follow the new interface (Alberto)
-
-Revision 1.2 2006/03/07 07:52:34 hristov
-New version (B.Yordanov)
-
-Revision 1.6 2005/11/19 17:19:14 byordano
-RetrieveDATEEntries and RetrieveConditionsData added
-
-Revision 1.5 2005/11/19 11:09:27 byordano
-AliShuttle declaration added
-
-Revision 1.4 2005/11/17 17:47:34 byordano
-TList changed to TObjArray
-
-Revision 1.3 2005/11/17 14:43:23 byordano
-import to local CVS
-
-Revision 1.1.1.1 2005/10/28 07:33:58 hristov
-Initial import as subdirectory in AliRoot
-
-Revision 1.2 2005/09/13 08:41:15 byordano
-default startTime endTime added
-
-Revision 1.4 2005/08/30 09:13:02 byordano
-some docs added
-
-Revision 1.3 2005/08/29 21:15:47 byordano
-some docs added
-
-*/
+/* $Id$ */
//
// This class is the main manager for AliShuttle.
#include <TMutex.h>
#include <TSystemDirectory.h>
#include <TSystemFile.h>
-#include <TFileMerger.h>
+#include <TFile.h>
#include <TGrid.h>
#include <TGridResult.h>
+#include <TMap.h>
#include <TMonaLisaWriter.h>
#include <sys/types.h>
#include <sys/wait.h>
+#include <signal.h>
+
ClassImp(AliShuttle)
//______________________________________________________________________________________________
fPreprocessorMap(),
fLogbookEntry(0),
fCurrentDetector(),
+fFirstProcessing(0),
+fFXSError(-1),
fStatusEntry(0),
fMonitoringMutex(0),
fLastActionTime(0),
// Then calls StoreRefFilesToGrid to store reference files.
//
+ UpdateShuttleStatus(AliShuttleStatus::kStoreStarted);
+
if (fTestMode & kErrorGrid)
{
Log("SHUTTLE", "StoreOCDB - In TESTMODE - Simulating error while storing in the Grid");
return kFALSE;
}
- AliInfo("Storing OCDB data ...");
- Bool_t resultCDB = StoreOCDB(fgkMainCDB);
+ Log("SHUTTLE","StoreOCDB - Storing OCDB data ...");
+ Int_t resultCDB = StoreOCDB(fgkMainCDB);
- AliInfo("Storing reference data ...");
- Bool_t resultRef = StoreOCDB(fgkMainRefStorage);
+ Log("SHUTTLE","StoreOCDB - Storing reference data ...");
+ Int_t resultRef = StoreOCDB(fgkMainRefStorage);
+
+ Log("SHUTTLE","StoreOCDB - Storing reference files ...");
+ Bool_t resultRefFiles = CopyFilesToGrid("reference");
+
+ Bool_t resultMetadata = kTRUE;
+ if(fCurrentDetector == "GRP")
+ {
+ Log("SHUTTLE","StoreOCDB - Storing Run Metadata file ...");
+ resultMetadata = CopyFilesToGrid("metadata");
+ }
- AliInfo("Storing reference files ...");
- Bool_t resultRefFiles = StoreRefFilesToGrid();
+ Int_t storeResult = 0;
- return resultCDB && resultRef && resultRefFiles;
+ if (resultCDB < 0 || resultRef < 0 || resultRefFiles == kFALSE || resultMetadata == kFALSE)
+ storeResult = -1;
+ else if (resultCDB > 0 || resultRef > 0)
+ storeResult = 1;
+
+ if (storeResult < 0)
+ {
+ Log("SHUTTLE",
+ Form("\t\t\t****** run %d - %s: STORAGE ERROR ******",
+ GetCurrentRun(), fCurrentDetector.Data()));
+ UpdateShuttleStatus(AliShuttleStatus::kStoreError);
+ }
+ else if (storeResult > 0)
+ {
+ Log("SHUTTLE",
+ Form("\t\t\t****** run %d - %s: STORAGE DELAYED ******",
+ GetCurrentRun(), fCurrentDetector.Data()));
+ UpdateShuttleStatus(AliShuttleStatus::kStoreDelayed);
+ }
+ else if (storeResult == 0)
+ {
+ Log("SHUTTLE",
+ Form("\t\t\t****** run %d - %s: DONE ******",
+ GetCurrentRun(), fCurrentDetector.Data()));
+ UpdateShuttleStatus(AliShuttleStatus::kDone);
+ UpdateShuttleLogbook(fCurrentDetector, "DONE");
+ }
+
+ return (storeResult == 0);
}
//______________________________________________________________________________________________
-Bool_t AliShuttle::StoreOCDB(const TString& gridURI)
+Int_t AliShuttle::StoreOCDB(const TString& gridURI)
{
//
// Called by StoreOCDB(), performs actual storage to the main OCDB and reference storages (Grid)
//
+ // Return code:
+ // -2 initialization error
+ // -1 storage error
+ // 0 success
+ // 1 storage delayed (e.g. previous unprocessed runs)
+ //
TObjArray* gridIds=0;
Bool_t result = kTRUE;
+ Bool_t delayed = kFALSE;
const char* type = 0;
TString localURI;
localURI = fgkLocalRefStorage;
} else {
AliError(Form("Invalid storage URI: %s", gridURI.Data()));
- return kFALSE;
+ return -2;
}
AliCDBManager* man = AliCDBManager::Instance();
if(!gridSto) {
Log("SHUTTLE",
Form("StoreOCDB - cannot activate main %s storage", type));
- return kFALSE;
+ return -2;
}
gridIds = gridSto->GetQueryCDBList();
if(!localSto) {
Log("SHUTTLE",
Form("StoreOCDB - cannot activate local %s storage", type));
- return kFALSE;
+ return -2;
}
AliCDBPath aPath(GetOfflineDetName(fCurrentDetector.Data()),"*","*");
// Local objects were stored with current run as Grid version!
AliCDBId aLocId = aLocEntry->GetId();
aLocEntry->SetVersion(-1);
aLocEntry->SetSubVersion(-1);
+
+ Log(fCurrentDetector.Data(), Form("Attempting to store %s", aLocId.ToString().Data()));
// If local object is valid up to infinity we store it only if it is
// the first unprocessed run!
Log("SHUTTLE", Form("StoreOCDB - %s: object %s has validity infinite but "
"there are previous unprocessed runs!",
fCurrentDetector.Data(), aLocId.GetPath().Data()));
+ Log(fCurrentDetector.Data(), Form("StoreOCDB - %s: object %s has validity infinite but "
+ "there are previous unprocessed runs!",
+ fCurrentDetector.Data(), aLocId.GetPath().Data()));
+ delayed = kTRUE;
continue;
}
Bool_t store = kTRUE;
TIter gridIter(gridIds);
AliCDBId* aGridId = 0;
- while((aGridId = dynamic_cast<AliCDBId*> (gridIter.Next()))){
- if(aGridId->GetPath() != aLocId.GetPath()) continue;
+ while ((aGridId = dynamic_cast<AliCDBId*> (gridIter.Next()))) {
+ if (aGridId->GetPath() != aLocId.GetPath())
+ continue;
// skip all objects valid up to infinity
- if(aGridId->GetLastRun() == AliCDBRunRange::Infinity()) continue;
+ if (aGridId->GetLastRun() == AliCDBRunRange::Infinity())
+ continue;
+
// if we get here, it means there's already some more recent object stored on Grid!
+ Log(fCurrentDetector.Data(),
+ Form("StoreOCDB - A more recent object already exists in %s storage: <%s>",
+ type, aGridId->ToString().Data()));
+
store = kFALSE;
break;
}
- // If we get here, the file can be stored!
- Bool_t storeOk = gridSto->Put(aLocEntry);
- if(!store || storeOk){
-
- if (!store)
- {
- Log(fCurrentDetector.Data(),
- Form("StoreOCDB - A more recent object already exists in %s storage: <%s>",
- type, aGridId->ToString().Data()));
- } else {
+ Bool_t storeOk = kFALSE;
+ if (store)
+ {
+ Log(fCurrentDetector.Data(), Form("Prechecks succeeded. Ready to store %s", aLocId.ToString().Data()));
+ storeOk = gridSto->Put(aLocEntry);
+ if (storeOk) {
Log("SHUTTLE",
+ Form("StoreOCDB - Object <%s> successfully put into %s storage",
+ aLocId.ToString().Data(), type));
+ Log(fCurrentDetector.Data(),
Form("StoreOCDB - Object <%s> successfully put into %s storage",
- aLocId.ToString().Data(), type));
+ aLocId.ToString().Data(), type));
+ } else {
+ Log("SHUTTLE",
+ Form("StoreOCDB - Grid %s storage of object <%s> failed",
+ type, aLocId.ToString().Data()));
+ Log(fCurrentDetector.Data(),
+ Form("StoreOCDB - Grid %s storage of object <%s> failed",
+ type, aLocId.ToString().Data()));
+ result = kFALSE;
}
-
- // removing local filename...
+ }
+
+ if (!store || storeOk) {
+ // removing local file...
TString filename;
localSto->IdToFilename(aLocId, filename);
- AliInfo(Form("Removing local file %s", filename.Data()));
+ Log("SHUTTLE", Form("StoreOCDB - Removing local file %s", filename.Data()));
RemoveFile(filename.Data());
- continue;
- } else {
- Log("SHUTTLE",
- Form("StoreOCDB - Grid %s storage of object <%s> failed",
- type, aLocId.ToString().Data()));
- result = kFALSE;
}
}
localEntries->Clear();
- return result;
+ Int_t returnCode = 0;
+
+ if (result == kFALSE)
+ returnCode = -1;
+ else if (delayed != kFALSE)
+ returnCode = 1;
+
+ Log("SHUTTLE", Form("StoreOCDB - Returning with %d (result = %d, delayed = %d)", returnCode, result, delayed));
+ Log(fCurrentDetector.Data(), Form("StoreOCDB - Returning with %d (result = %d, delayed = %d)", returnCode, result, delayed));
+
+ return returnCode;
+}
+
+//______________________________________________________________________________________________
+Bool_t AliShuttle::CleanReferenceStorage(const char* detector)
+{
+ // clears the directory used to store reference files of a given subdetector
+
+ AliCDBManager* man = AliCDBManager::Instance();
+ AliCDBStorage* sto = man->GetStorage(fgkLocalRefStorage);
+ TString localBaseFolder = sto->GetBaseFolder();
+
+ TString targetDir = GetRefFilePrefix(localBaseFolder.Data(), detector);
+
+ Log("SHUTTLE", Form("CleanReferenceStorage - Cleaning %s", targetDir.Data()));
+
+ TString begin;
+ begin.Form("%d_", GetCurrentRun());
+
+ TSystemDirectory* baseDir = new TSystemDirectory("/", targetDir);
+ if (!baseDir)
+ return kTRUE;
+
+ TList* dirList = baseDir->GetListOfFiles();
+ delete baseDir;
+
+ if (!dirList) return kTRUE;
+
+ if (dirList->GetEntries() < 3)
+ {
+ delete dirList;
+ return kTRUE;
+ }
+
+ Int_t nDirs = 0, nDel = 0;
+ TIter dirIter(dirList);
+ TSystemFile* entry = 0;
+
+ Bool_t success = kTRUE;
+
+ while ((entry = dynamic_cast<TSystemFile*> (dirIter.Next())))
+ {
+ if (entry->IsDirectory())
+ continue;
+
+ TString fileName(entry->GetName());
+ if (!fileName.BeginsWith(begin))
+ continue;
+
+ nDirs++;
+
+ // delete file
+ Int_t result = gSystem->Unlink(fileName.Data());
+
+ if (result)
+ {
+ Log("SHUTTLE", Form("CleanReferenceStorage - Could not delete file %s!", fileName.Data()));
+ success = kFALSE;
+ } else {
+ nDel++;
+ }
+ }
+
+ if(nDirs > 0)
+ Log("SHUTTLE", Form("CleanReferenceStorage - %d (over %d) reference files in folder %s were deleted.",
+ nDel, nDirs, targetDir.Data()));
+
+
+ delete dirList;
+ return success;
+
+
+
+
+
+
+ Int_t result = gSystem->GetPathInfo(targetDir, 0, (Long64_t*) 0, 0, 0);
+ if (result == 0)
+ {
+ // delete directory
+ result = gSystem->Exec(Form("rm -rf %s", targetDir.Data()));
+ if (result != 0)
+ {
+ Log("SHUTTLE", Form("CleanReferenceStorage - Could not clean directory %s", targetDir.Data()));
+ return kFALSE;
+ }
+ }
+
+ result = gSystem->mkdir(targetDir, kTRUE);
+ if (result != 0)
+ {
+ Log("SHUTTLE", Form("CleanReferenceStorage - Error creating base directory %s", targetDir.Data()));
+ return kFALSE;
+ }
+
+ return kTRUE;
}
//______________________________________________________________________________________________
TString localBaseFolder = sto->GetBaseFolder();
- TString targetDir;
- targetDir.Form("%s/%s", localBaseFolder.Data(), detector);
+ TString target = GetRefFilePrefix(localBaseFolder.Data(), detector);
+ target.Append(Form("/%d_%s", GetCurrentRun(), gridFileName));
+
+ return CopyFileLocally(localFile, target);
+}
+
+//______________________________________________________________________________________________
+Bool_t AliShuttle::StoreRunMetadataFile(const char* localFile, const char* gridFileName)
+{
+ //
+ // Stores Run metadata file to the Grid, in the run folder
+ //
+ // Only GRP can call this function.
+
+ if (fTestMode & kErrorStorage)
+ {
+ Log(fCurrentDetector, "StoreRunMetaDataFile - In TESTMODE - Simulating error while storing locally");
+ return kFALSE;
+ }
+
+ AliCDBManager* man = AliCDBManager::Instance();
+ AliCDBStorage* sto = man->GetStorage(fgkLocalRefStorage);
+
+ TString localBaseFolder = sto->GetBaseFolder();
+
+ // Build Run level folder
+ // folder = /alice/data/year/lhcPeriod/runNb/raw
+
+
+ TString lhcPeriod = GetLHCPeriod();
+ if (lhcPeriod.Length() == 0)
+ {
+ Log("SHUTTLE","StoreRunMetaDataFile - LHCPeriod not found in logbook!");
+ return 0;
+ }
+
+ // TODO partitions with one detector only write data into LHCperiod_DET
+ TString partition = GetRunParameter("detector");
+
+ if (partition.Length() > 0 && partition != "ALICE")
+ {
+ lhcPeriod.Append(Form("_%s", partition.Data()));
+ Log(fCurrentDetector, Form("Run data tags merged file will be written in %s",
+ lhcPeriod.Data()));
+ }
+
+ TString target = Form("%s/GRP/RunMetadata%s%d/%s/%09d/raw/%s",
+ localBaseFolder.Data(), fConfig->GetAlienPath(), GetCurrentYear(),
+ lhcPeriod.Data(), GetCurrentRun(), gridFileName);
+
+ return CopyFileLocally(localFile, target);
+}
+
+//______________________________________________________________________________________________
+Bool_t AliShuttle::CopyFileLocally(const char* localFile, const TString& target)
+{
+ //
+ // Stores file locally. Called by StoreReferenceFile and StoreRunMetadataFile
+ // Files are temporarily stored in the local reference storage. When the preprocessor
+ // finishes, the Shuttle calls CopyFilesToGrid to transfer the files to AliEn
+ // (in reference or run level folders)
+ //
- TString target;
- target.Form("%s/%d_%s", targetDir.Data(), GetCurrentRun(), gridFileName);
+ TString targetDir(target(0, target.Last('/')));
- Int_t result = gSystem->GetPathInfo(targetDir, 0, (Long64_t*) 0, 0, 0);
+ //try to open base dir folder, if it does not exist
+ void* dir = gSystem->OpenDirectory(targetDir.Data());
+ if (dir == NULL) {
+ if (gSystem->mkdir(targetDir.Data(), kTRUE)) {
+ Log("SHUTTLE", Form("CopyFileLocally - Can't open directory <%s>", targetDir.Data()));
+ return kFALSE;
+ }
+
+ } else {
+ gSystem->FreeDirectory(dir);
+ }
+
+ Int_t result = 0;
+
+ result = gSystem->GetPathInfo(localFile, 0, (Long64_t*) 0, 0, 0);
if (result)
{
- result = gSystem->mkdir(targetDir, kTRUE);
- if (result != 0)
+ Log("SHUTTLE", Form("CopyFileLocally - %s does not exist", localFile));
+ return kFALSE;
+ }
+
+ result = gSystem->GetPathInfo(target, 0, (Long64_t*) 0, 0, 0);
+ if (!result)
+ {
+ Log("SHUTTLE", Form("CopyFileLocally - target file %s already exist, removing...", target.Data()));
+ if (gSystem->Unlink(target.Data()))
{
- Log("SHUTTLE", Form("StoreReferenceFile - Error creating base directory %s", targetDir.Data()));
+ Log("SHUTTLE", Form("CopyFileLocally - Could not remove existing target file %s!", target.Data()));
return kFALSE;
}
- }
-
+ }
+
result = gSystem->CopyFile(localFile, target);
if (result == 0)
{
- Log("SHUTTLE", Form("StoreReferenceFile - Stored file %s locally to %s", localFile, target.Data()));
+ Log("SHUTTLE", Form("CopyFileLocally - File %s stored locally to %s", localFile, target.Data()));
return kTRUE;
}
else
{
- Log("SHUTTLE", Form("StoreReferenceFile - Storing file %s locally to %s failed", localFile, target.Data()));
+ Log("SHUTTLE", Form("CopyFileLocally - Could not store file %s to %s! Error code = %d",
+ localFile, target.Data(), result));
return kFALSE;
}
+
+
+
}
//______________________________________________________________________________________________
-Bool_t AliShuttle::StoreRefFilesToGrid()
+Bool_t AliShuttle::CopyFilesToGrid(const char* type)
{
//
- // Transfers the reference file to the Grid.
+ // Transfers local files to the Grid. Local files can be reference files
+ // or run metadata file (from GRP only).
//
- // The file is stored under the following location:
- // <base folder of reference storage>/<DET>/<RUN#>_<gridFileName>
- // where <gridFileName> is the second parameter given to the function
- //
-
+ // According to the type (ref, metadata) the files are stored under the following location:
+ // ref --> <base folder of reference storage>/<DET>/<RUN#>_<gridFileName>
+ // metadata --> <run data folder>/<MetadataFileName>
+ //
+
AliCDBManager* man = AliCDBManager::Instance();
AliCDBStorage* sto = man->GetStorage(fgkLocalRefStorage);
if (!sto)
return kFALSE;
TString localBaseFolder = sto->GetBaseFolder();
-
- TString dir;
- dir.Form("%s/%s", localBaseFolder.Data(), GetOfflineDetName(fCurrentDetector));
- AliCDBStorage* gridSto = man->GetStorage(fgkMainRefStorage);
- if (!gridSto)
- return kFALSE;
- TString gridBaseFolder = gridSto->GetBaseFolder();
+ TString dir;
TString alienDir;
- alienDir.Form("%s%s", gridBaseFolder.Data(), GetOfflineDetName(fCurrentDetector));
+ TString begin;
- if (!gGrid)
- return kFALSE;
+ if (strcmp(type, "reference") == 0)
+ {
+ dir = GetRefFilePrefix(localBaseFolder.Data(), fCurrentDetector.Data());
+ AliCDBStorage* gridSto = man->GetStorage(fgkMainRefStorage);
+ if (!gridSto)
+ return kFALSE;
+ TString gridBaseFolder = gridSto->GetBaseFolder();
+ alienDir = GetRefFilePrefix(gridBaseFolder.Data(), fCurrentDetector.Data());
+ begin = Form("%d_", GetCurrentRun());
+ }
+ else if (strcmp(type, "metadata") == 0)
+ {
+
+ TString lhcPeriod = GetLHCPeriod();
- TString begin;
- begin.Form("%d_", GetCurrentRun());
+ if (lhcPeriod.Length() == 0)
+ {
+ Log("SHUTTLE","CopyFilesToGrid - LHCPeriod not found in logbook!");
+ return 0;
+ }
+
+ // TODO partitions with one detector only write data into LHCperiod_DET
+ TString partition = GetRunParameter("detector");
+ if (partition.Length() > 0 && partition != "ALICE")
+ {
+ lhcPeriod.Append(Form("_%s", partition.Data()));
+ }
+
+ dir = Form("%s/GRP/RunMetadata%s%d/%s/%09d/raw",
+ localBaseFolder.Data(), fConfig->GetAlienPath(), GetCurrentYear(),
+ lhcPeriod.Data(), GetCurrentRun());
+ alienDir = dir(dir.Index(fConfig->GetAlienPath()), dir.Length());
+
+ begin = "";
+ }
+ else
+ {
+ Log("SHUTTLE", "CopyFilesToGrid - Unexpected: type label must be reference or metadata!");
+ return kFALSE;
+ }
+
TSystemDirectory* baseDir = new TSystemDirectory("/", dir);
if (!baseDir)
return kTRUE;
- TList* dirList = baseDir->GetListOfFiles();
- if (!dirList)
+ TList* dirList = baseDir->GetListOfFiles();
+ delete baseDir;
+
+ if (!dirList) return kTRUE;
+
+ if (dirList->GetEntries() < 3)
{
- delete baseDir;
+ delete dirList;
return kTRUE;
}
-
- Int_t nDirs = dirList->GetEntries();
+
+ if (!gGrid)
+ {
+ Log("SHUTTLE", "CopyFilesToGrid - Connection to Grid failed: Cannot continue!");
+ delete dirList;
+ return kFALSE;
+ }
+ Int_t nDirs = 0, nTransfer = 0;
+ TIter dirIter(dirList);
+ TSystemFile* entry = 0;
+
Bool_t success = kTRUE;
Bool_t first = kTRUE;
- for (Int_t iDir=0; iDir<nDirs; ++iDir)
- {
- TSystemFile* entry = dynamic_cast<TSystemFile*> (dirList->At(iDir));
- if (!entry)
- continue;
-
+ while ((entry = dynamic_cast<TSystemFile*> (dirIter.Next())))
+ {
if (entry->IsDirectory())
continue;
if (!fileName.BeginsWith(begin))
continue;
+ nDirs++;
+
if (first)
{
first = kFALSE;
- // check that DET folder exists, otherwise create it
+ // check that folder exists, otherwise create it
TGridResult* result = gGrid->Ls(alienDir.Data(), "a");
if (!result)
+ {
+ delete dirList;
return kFALSE;
+ }
- if (!result->GetFileName(0))
+ if (!result->GetFileName(1)) // TODO: It looks like element 0 is always 0!!
{
- if (!gGrid->Mkdir(alienDir.Data(),"",0))
+ // TODO It does not work currently! Bug in TAliEn::Mkdir
+ // TODO Manually fixed in local root v5-16-00
+ if (!gGrid->Mkdir(alienDir.Data(),"-p",0))
{
- Log("SHUTTLE", Form("StoreRefFilesToGrid - Cannot create directory %s",
+ Log("SHUTTLE", Form("CopyFilesToGrid - Cannot create directory %s",
alienDir.Data()));
- delete baseDir;
+ delete dirList;
return kFALSE;
+ } else {
+ Log("SHUTTLE",Form("CopyFilesToGrid - Folder %s created", alienDir.Data()));
}
+ } else {
+ Log("SHUTTLE",Form("CopyFilesToGrid - Folder %s found", alienDir.Data()));
}
}
TString fullGridPath;
fullGridPath.Form("alien://%s/%s", alienDir.Data(), fileName.Data());
- Log("SHUTTLE", Form("StoreRefFilesToGrid - Copying local file %s to %s", fullLocalPath.Data(), fullGridPath.Data()));
-
- TFileMerger fileMerger;
- Bool_t result = fileMerger.Cp(fullLocalPath, fullGridPath);
+ Bool_t result = TFile::Cp(fullLocalPath, fullGridPath);
if (result)
{
- Log("SHUTTLE", Form("StoreRefFilesToGrid - Copying local file %s to %s succeeded", fullLocalPath.Data(), fullGridPath.Data()));
+ Log("SHUTTLE", Form("CopyFilesToGrid - Copying local file %s to %s succeeded!",
+ fullLocalPath.Data(), fullGridPath.Data()));
RemoveFile(fullLocalPath);
+ nTransfer++;
}
else
{
- Log("SHUTTLE", Form("StoreRefFilesToGrid - Copying local file %s to %s failed", fullLocalPath.Data(), fullGridPath.Data()));
+ Log("SHUTTLE", Form("CopyFilesToGrid - Copying local file %s to %s FAILED!",
+ fullLocalPath.Data(), fullGridPath.Data()));
success = kFALSE;
}
}
-
- delete baseDir;
-
+
+ Log("SHUTTLE", Form("CopyFilesToGrid - %d (over %d) files in folder %s copied to Grid.",
+ nTransfer, nDirs, dir.Data()));
+
+
+ delete dirList;
return success;
}
//______________________________________________________________________________________________
-void AliShuttle::CleanLocalStorage(const TString& uri)
+const char* AliShuttle::GetRefFilePrefix(const char* base, const char* detector)
{
//
- // Called in case the preprocessor is declared failed. Remove remaining objects from the local storages.
+ // Get folder name of reference files
//
- const char* type = 0;
- if(uri == fgkLocalCDB) {
- type = "OCDB";
- } else if(uri == fgkLocalRefStorage) {
- type = "reference";
+ TString offDetStr(GetOfflineDetName(detector));
+ TString dir;
+ if (offDetStr == "ITS" || offDetStr == "MUON" || offDetStr == "PHOS")
+ {
+ dir.Form("%s/%s/%s", base, offDetStr.Data(), detector);
+ } else {
+ dir.Form("%s/%s", base, offDetStr.Data());
+ }
+
+ return dir.Data();
+
+
+}
+
+//______________________________________________________________________________________________
+void AliShuttle::CleanLocalStorage(const TString& uri)
+{
+ //
+ // Called in case the preprocessor is declared failed. Remove remaining objects from the local storages.
+ //
+
+ const char* type = 0;
+ if(uri == fgkLocalCDB) {
+ type = "OCDB";
+ } else if(uri == fgkLocalRefStorage) {
+ type = "Reference";
} else {
AliError(Form("Invalid storage URI: %s", uri.Data()));
return;
}
TString filename(Form("%s/%s/*/Run*_v%d_s*.root",
- localSto->GetBaseFolder().Data(), fCurrentDetector.Data(), GetCurrentRun()));
+ localSto->GetBaseFolder().Data(), GetOfflineDetName(fCurrentDetector.Data()), GetCurrentRun()));
- AliInfo(Form("filename = %s", filename.Data()));
+ AliDebug(2, Form("filename = %s", filename.Data()));
- AliInfo(Form("Removing remaining local files from run %d and detector %s ...",
+ Log("SHUTTLE", Form("Removing remaining local files for run %d and detector %s ...",
GetCurrentRun(), fCurrentDetector.Data()));
RemoveFile(filename.Data());
fStatusEntry = 0;
}
+ Int_t path1 = GetCurrentRun()/10000;
fStatusEntry = AliCDBManager::Instance()->GetStorage(GetLocalCDB())
- ->Get(Form("/SHUTTLE/STATUS/%s", fCurrentDetector.Data()), GetCurrentRun());
+ ->Get(Form("/SHUTTLE/%s/%d", fCurrentDetector.Data(), path1), GetCurrentRun());
if (!fStatusEntry) return 0;
fStatusEntry->SetOwner(1);
}
Int_t run = GetCurrentRun();
+ Int_t path1 = run/10000;
+ TString path1_string = Form("%d",path1);
- AliCDBId id(AliCDBPath("SHUTTLE", "STATUS", fCurrentDetector), run, run);
+ AliCDBId id(AliCDBPath("SHUTTLE", fCurrentDetector, path1_string), run, run);
fStatusEntry = new AliCDBEntry(status, id, new AliCDBMetaData);
fStatusEntry->SetOwner(1);
return kFALSE;
}
- SendMLInfo();
+ SendMLDetInfo();
return kTRUE;
}
AliShuttleStatus* status = dynamic_cast<AliShuttleStatus*> (fStatusEntry->GetObject());
if (!status){
- Log("SHUTTLE", "UNEXPECTED: status could not be read from current CDB entry");
+ Log("SHUTTLE", "UpdateShuttleStatus - UNEXPECTED: status could not be read from current CDB entry");
return;
}
AliCDBManager::Instance()->GetStorage(fgkLocalCDB)->Put(fStatusEntry);
- SendMLInfo();
+ SendMLDetInfo();
}
//______________________________________________________________________________________________
-void AliShuttle::SendMLInfo()
+void AliShuttle::SendMLDetInfo()
{
//
// sends ML information about the current status of the current detector being processed
AliShuttleStatus* status = dynamic_cast<AliShuttleStatus*> (fStatusEntry->GetObject());
if (!status){
- Log("SHUTTLE", "SendMLInfo - UNEXPECTED: status could not be read from current CDB entry");
+ Log("SHUTTLE", "SendMLDetInfo - UNEXPECTED: status could not be read from current CDB entry");
return;
}
mlList.Add(&mlStatus);
mlList.Add(&mlRetryCount);
- fMonaLisa->SendParameters(&mlList);
+ TString mlID;
+ mlID.Form("%d", GetCurrentRun());
+ fMonaLisa->SendParameters(&mlList, mlID);
}
//______________________________________________________________________________________________
// checks if the processing should be continued
// if yes it returns kTRUE and updates the AliShuttleStatus with nextStatus
- if (!fConfig->HostProcessDetector(fCurrentDetector)) return kFALSE;
+ if (!fConfig->HostProcessDetector(fCurrentDetector))
+ return kFALSE;
AliPreprocessor* aPreprocessor =
dynamic_cast<AliPreprocessor*> (fPreprocessorMap.GetValue(fCurrentDetector));
if (!aPreprocessor)
{
- AliInfo(Form("%s: no preprocessor registered", fCurrentDetector.Data()));
+ Log("SHUTTLE", Form("ContinueProcessing - %s: no preprocessor registered", fCurrentDetector.Data()));
return kFALSE;
}
AliShuttleLogbookEntry::Status entryStatus =
fLogbookEntry->GetDetectorStatus(fCurrentDetector);
- if(entryStatus != AliShuttleLogbookEntry::kUnprocessed) {
- AliInfo(Form("ContinueProcessing - %s is %s",
+ if (entryStatus != AliShuttleLogbookEntry::kUnprocessed) {
+ Log("SHUTTLE", Form("ContinueProcessing - %s is %s",
fCurrentDetector.Data(),
fLogbookEntry->GetDetectorStatusName(entryStatus)));
return kFALSE;
if (fConfig->StrictRunOrder(fCurrentDetector) &&
!fFirstUnprocessed[GetDetPos(fCurrentDetector)])
{
- Log("SHUTTLE", Form("ContinueProcessing - %s requires strict run ordering but this is not the first unprocessed run!"));
- return kFALSE;
+ if (fTestMode == kNone)
+ {
+ Log("SHUTTLE", Form("ContinueProcessing - %s requires strict run ordering"
+ " but this is not the first unprocessed run!"));
+ return kFALSE;
+ }
+ else
+ {
+ Log("SHUTTLE", Form("ContinueProcessing - In TESTMODE - "
+ "Although %s requires strict run ordering "
+ "and this is not the first unprocessed run, "
+ "the SHUTTLE continues"));
+ }
}
+ // Is the subdetector processed first time for this run?
+ fFirstProcessing = kFALSE;
+
AliShuttleStatus* status = ReadShuttleStatus();
if (!status) {
// first time
Log("SHUTTLE", Form("ContinueProcessing - %s: Processing first time",
fCurrentDetector.Data()));
status = new AliShuttleStatus(AliShuttleStatus::kStarted);
+ fFirstProcessing = kTRUE;
return WriteShuttleStatus(status);
}
- // The following two cases shouldn't happen if Shuttle Logbook was correctly updated.
+ // The following case shouldn't happen if Shuttle Logbook was correctly updated.
// If it happens it may mean Logbook updating failed... let's do it now!
if (status->GetStatus() == AliShuttleStatus::kDone ||
- status->GetStatus() == AliShuttleStatus::kFailed){
+ status->GetStatus() == AliShuttleStatus::kFailed ||
+ status->GetStatus() == AliShuttleStatus::kSkipped) {
Log("SHUTTLE", Form("ContinueProcessing - %s is already %s. Updating Shuttle Logbook",
fCurrentDetector.Data(),
status->GetStatusName(status->GetStatus())));
- UpdateShuttleLogbook(fCurrentDetector.Data(),
- status->GetStatusName(status->GetStatus()));
+
+ if (status->GetStatus() == AliShuttleStatus::kSkipped)
+ {
+ UpdateShuttleLogbook(fCurrentDetector.Data(), "DONE");
+ }
+ else
+ UpdateShuttleLogbook(fCurrentDetector.Data(), status->GetStatusName(status->GetStatus()));
+
return kFALSE;
}
- if (status->GetStatus() == AliShuttleStatus::kStoreError) {
+ if (status->GetStatus() == AliShuttleStatus::kStoreStarted || status->GetStatus() == AliShuttleStatus::kStoreDelayed ||status->GetStatus() == AliShuttleStatus::kStoreError) {
Log("SHUTTLE",
- Form("ContinueProcessing - %s: Grid storage of one or more objects failed. Trying again now",
- fCurrentDetector.Data()));
- UpdateShuttleStatus(AliShuttleStatus::kStoreStarted);
- if (StoreOCDB()){
- Log("SHUTTLE", Form("ContinueProcessing - %s: all objects successfully stored into main storage",
+ Form("ContinueProcessing - %s: Grid storage of one or more "
+ "objects failed. Trying again now",
fCurrentDetector.Data()));
- UpdateShuttleStatus(AliShuttleStatus::kDone);
- UpdateShuttleLogbook(fCurrentDetector.Data(), "DONE");
- } else {
- Log("SHUTTLE",
- Form("ContinueProcessing - %s: Grid storage failed again",
- fCurrentDetector.Data()));
- UpdateShuttleStatus(AliShuttleStatus::kStoreError);
- }
+ StoreOCDB();
return kFALSE;
}
// UpdateTableFailCase();
// Send mail to detector expert!
- AliInfo(Form("Sending mail to %s expert...", fCurrentDetector.Data()));
- if (!SendMail())
+ Log("SHUTTLE", Form("ContinueProcessing - Sending mail to %s expert...",
+ fCurrentDetector.Data()));
+ if (!SendMail(kPPEMail))
Log("SHUTTLE", Form("ContinueProcessing - Could not send mail to %s expert",
- fCurrentDetector.Data()));
+ fCurrentDetector.Data()));
} else {
Log("SHUTTLE", Form("ContinueProcessing - %s: restarting. "
"Aborted before with %s. Retry number %d.", fCurrentDetector.Data(),
status->GetStatusName(), status->GetCount()));
Bool_t increaseCount = kTRUE;
- if (status->GetStatus() == AliShuttleStatus::kDCSError || status->GetStatus() == AliShuttleStatus::kDCSStarted)
- increaseCount = kFALSE;
+ if (status->GetStatus() == AliShuttleStatus::kDCSError ||
+ status->GetStatus() == AliShuttleStatus::kDCSStarted ||
+ status->GetStatus() == AliShuttleStatus::kFXSError)
+ increaseCount = kFALSE;
+
UpdateShuttleStatus(AliShuttleStatus::kStarted, increaseCount);
cont = kTRUE;
}
return cont;
}
+//______________________________________________________________________________________________
+void AliShuttle::SendMLRunInfo(const char* status)
+{
+ //
+ // Send information about this run to ML
+
+ TMonaLisaText mlStatus("SHUTTLE_status", status);
+ TString runType(fLogbookEntry->GetRunType());
+ if (strlen(fLogbookEntry->GetRunParameter("log")) > 0){
+
+ runType += "(";
+ runType += fLogbookEntry->GetRunParameter("log");
+ runType += ")";
+ }
+ if (fLogbookEntry->GetDATestMode()){
+ runType += " (DATest)";
+ }
+ TMonaLisaText mlRunType("SHUTTLE_runtype", runType);
+
+ TList mlList;
+ mlList.Add(&mlStatus);
+ mlList.Add(&mlRunType);
+
+ TString mlID;
+ mlID.Form("%d", GetCurrentRun());
+ fMonaLisa->SendParameters(&mlList, mlID);
+}
+
+//______________________________________________________________________________________________
+Int_t AliShuttle::GetMem(Int_t pid)
+{
+ // invokes ps to get the memory consumption of the process <pid>
+ // returns -1 in case of error
+
+ TString checkStr;
+ checkStr.Form("ps -o vsize --pid %d | tail -n 1", pid);
+ FILE* pipe = gSystem->OpenPipe(checkStr, "r");
+ if (!pipe)
+ {
+ Log("SHUTTLE", Form("Process - Error: "
+ "Could not open pipe to %s", checkStr.Data()));
+ return -1;
+ }
+
+ char buffer[100];
+ if (!fgets(buffer, 100, pipe))
+ {
+ Log("SHUTTLE", "Process - Error: ps did not return anything");
+ gSystem->ClosePipe(pipe);
+ return -1;
+ }
+ gSystem->ClosePipe(pipe);
+
+ //Log("SHUTTLE", Form("ps returned %s", buffer));
+
+ Int_t mem = 0;
+ if ((sscanf(buffer, "%d\n", &mem) != 1) || !mem)
+ {
+ Log("SHUTTLE", "Process - Error: Could not parse output of ps");
+ return -1;
+ }
+
+ return mem;
+}
+
//______________________________________________________________________________________________
Bool_t AliShuttle::Process(AliShuttleLogbookEntry* entry)
{
fLogbookEntry = entry;
- AliInfo(Form("\n\n \t\t\t^*^*^*^*^*^*^*^*^*^*^*^* run %d: START ^*^*^*^*^*^*^*^*^*^*^*^* \n",
+ Log("SHUTTLE", Form("\t\t\t^*^*^*^*^*^*^*^*^*^*^*^* run %d: START ^*^*^*^*^*^*^*^*^*^*^*^*",
GetCurrentRun()));
- // create ML instance that monitors this run
- fMonaLisa = new TMonaLisaWriter(Form("%d", GetCurrentRun()), "SHUTTLE", "aliendb1.cern.ch");
- // disable monitoring of other parameters that come e.g. from TFile
- gMonitoringWriter = 0;
-
+ CountOpenRuns();
+
// Send the information to ML
- TMonaLisaText mlStatus("SHUTTLE_status", "Processing");
- TMonaLisaText mlRunType("SHUTTLE_runtype", Form("%s (%s)", entry->GetRunType(), entry->GetRunParameter("log")));
-
- TList mlList;
- mlList.Add(&mlStatus);
- mlList.Add(&mlRunType);
-
- fMonaLisa->SendParameters(&mlList);
+ SendMLRunInfo("Processing");
if (fLogbookEntry->IsDone())
{
Int_t testMode = tmpStr->String().Atoi();
if (testMode > 0)
{
- Log("SHUTTLE", Form("Enabling test mode %d", testMode));
+ Log("SHUTTLE", Form("Process - Enabling test mode %d", testMode));
SetTestMode((TestMode) testMode);
}
}
}
}
}
-
- Log("SHUTTLE", Form("The test mode flag is %d", (Int_t) fTestMode));
-
+
fLogbookEntry->Print("all");
// Initialization
Bool_t hasError = kFALSE;
- AliCDBStorage *mainCDBSto = AliCDBManager::Instance()->GetStorage(fgkMainCDB);
- if(mainCDBSto) mainCDBSto->QueryCDB(GetCurrentRun());
- AliCDBStorage *mainRefSto = AliCDBManager::Instance()->GetStorage(fgkMainRefStorage);
- if(mainRefSto) mainRefSto->QueryCDB(GetCurrentRun());
+ // Set the CDB and Reference folders according to the year
+
+ // build cdb paths (repeat each time, run might be a DATest run)
+ if (!fLogbookEntry->GetDATestMode()){
+ fgkMainCDB.Form("alien://folder=%s%d/OCDB?user=alidaq?cacheFold=/tmp/OCDBCache",
+ fConfig->GetAlienPath(), GetCurrentYear());
+
+ fgkMainRefStorage.Form("alien://folder=%s%d/Reference?user=alidaq?cacheFold=/tmp/OCDBCache",
+ fConfig->GetAlienPath(), GetCurrentYear());
+ }
+ else {
+ fgkMainCDB.Form("alien://folder=%s%d/DATest/OCDB?user=alidaq?cacheFold=/tmp/OCDBCache",
+ fConfig->GetAlienPath(), GetCurrentYear());
+
+ fgkMainRefStorage.Form("alien://folder=%s%d/DATest/Reference?user=alidaq?cacheFold=/tmp/OCDBCache",
+ fConfig->GetAlienPath(), GetCurrentYear());
+ }
+
+ AliDebug(2,Form("Main CDB storage = %s",fgkMainCDB.Data()));
+ AliDebug(2,Form("Main Reference storage = %s",fgkMainRefStorage.Data()));
// Loop on detectors in the configuration
TIter iter(fConfig->GetDetectors());
TObjString* aDetector = 0;
+ Bool_t first = kTRUE;
+
while ((aDetector = (TObjString*) iter.Next()))
{
fCurrentDetector = aDetector->String();
- if (ContinueProcessing() == kFALSE) continue;
+ if (ContinueProcessing() == kFALSE)
+ continue;
+
+ if (first)
+ {
+ // only read QueryCDB when needed and only once
+ AliCDBStorage *mainCDBSto = AliCDBManager::Instance()->GetStorage(fgkMainCDB);
+ if(mainCDBSto) mainCDBSto->QueryCDB(GetCurrentRun());
+ AliCDBStorage *mainRefSto = AliCDBManager::Instance()->GetStorage(fgkMainRefStorage);
+ if(mainRefSto) mainRefSto->QueryCDB(GetCurrentRun());
+ first = kFALSE;
+ }
- AliInfo(Form("\n\n \t\t\t****** run %d - %s: START ******",
+ Log("SHUTTLE", Form("\t\t\t****** run %d - %s: START ******",
GetCurrentRun(), aDetector->GetName()));
for(Int_t iSys=0;iSys<3;iSys++) fFXSCalled[iSys]=kFALSE;
+
+ Int_t initialMem = GetMem(getpid());
+ Log("SHUTTLE", Form("Memory consumption before forking is %d", initialMem));
- Log(fCurrentDetector.Data(), "Starting processing");
+ Log(fCurrentDetector.Data(), "Process - Starting processing");
Int_t pid = fork();
if (pid < 0)
{
- Log("SHUTTLE", "ERROR: Forking failed");
+ Log("SHUTTLE", "Process - ERROR: Forking failed");
}
else if (pid > 0)
{
// parent
- AliInfo(Form("In parent process of %d - %s: Starting monitoring",
+ Log("SHUTTLE", Form("Process - In parent process of %d - %s: Starting monitoring",
GetCurrentRun(), aDetector->GetName()));
Long_t begin = time(0);
if (expiredTime > fConfig->GetPPTimeOut())
{
- TString tmp;
- tmp.Form("Process of %s time out. Run time: %d seconds. Killing...",
+ TString logMsg;
+ AliShuttleStatus *currentStatus = ReadShuttleStatus();
+ AliShuttleStatus::Status newStatus = AliShuttleStatus::kInvalid;
+
+ if (currentStatus->GetStatus() <= AliShuttleStatus::kPPDone)
+ {
+ // in case pp not yet done set status to kPPTimeOut
+
+ logMsg.Form("Process - Process of %s timed out. Run time: %d seconds. Killing...",
fCurrentDetector.Data(), expiredTime);
- Log("SHUTTLE", tmp);
- Log(fCurrentDetector, tmp);
+ newStatus = AliShuttleStatus::kPPTimeOut;
+ }
+ else if (currentStatus->GetStatus() == AliShuttleStatus::kStoreStarted)
+ {
+ // in case the pp goes in TimeOut while storing the objects in the OCDB
+ // set status to kStoreError
+
+ logMsg.Form("Process - Process of %s timed out while storing the OCDB object. Run time: %d seconds. Killing... and setting status to StoreError.",
+ fCurrentDetector.Data(), expiredTime);
+ newStatus = AliShuttleStatus::kStoreError;
+ }
+ else
+ {
+ // in other cases don't change the status
+
+ logMsg.Form("Process - Process of %s timed out in status = %s. Run time: %d seconds. Killing... without changing the status",
+ fCurrentDetector.Data(), currentStatus->GetStatusName(), expiredTime);
+ }
+
+ Log("SHUTTLE", logMsg);
+ Log(fCurrentDetector, logMsg);
kill(pid, 9);
- UpdateShuttleStatus(AliShuttleStatus::kPPTimeOut);
+ if (newStatus != AliShuttleStatus::kInvalid)
+ UpdateShuttleStatus(newStatus);
hasError = kTRUE;
gSystem->Sleep(1000);
{
gSystem->Sleep(1000);
- TString checkStr;
- checkStr.Form("ps -o vsize --pid %d | tail -n 1", pid);
- FILE* pipe = gSystem->OpenPipe(checkStr, "r");
- if (!pipe)
- {
- Log("SHUTTLE", Form("Error: Could not open pipe to %s", checkStr.Data()));
+ Int_t mem = GetMem(pid);
+
+ if (mem < 0)
continue;
- }
- char buffer[100];
- if (!fgets(buffer, 100, pipe))
- {
- Log("SHUTTLE", "Error: ps did not return anything");
- gSystem->ClosePipe(pipe);
- continue;
- }
- gSystem->ClosePipe(pipe);
+ mem -= initialMem;
+ if (mem < 0)
+ mem = 0;
- //Log("SHUTTLE", Form("ps returned %s", buffer));
-
- Int_t mem = 0;
- if ((sscanf(buffer, "%d\n", &mem) != 1) || !mem)
+ if (expiredTime % 60 == 0)
{
- Log("SHUTTLE", "Error: Could not parse output of ps");
- continue;
+ Log("SHUTTLE", Form("Process - %s: Checking process. "
+ "Run time: %d seconds - Memory consumption: %d KB",
+ fCurrentDetector.Data(), expiredTime, mem));
+ SendAlive();
}
- if (expiredTime % 60 == 0)
- Log("SHUTTLE", Form("%s: Checking process. Run time: %d seconds - Memory consumption: %d KB",
- fCurrentDetector.Data(), expiredTime, mem));
-
if (mem > fConfig->GetPPMaxMem())
{
TString tmp;
- tmp.Form("Process exceeds maximum allowed memory (%d KB > %d KB). Killing...",
+ tmp.Form("Process - Process exceeds maximum allowed memory "
+ "(%d KB > %d KB). Killing...",
mem, fConfig->GetPPMaxMem());
Log("SHUTTLE", tmp);
Log(fCurrentDetector, tmp);
}
}
- AliInfo(Form("In parent process of %d - %s: Client has terminated.",
+ Log("SHUTTLE", Form("Process - In parent process of %d - %s: Client has terminated.",
GetCurrentRun(), aDetector->GetName()));
if (WIFEXITED(status))
{
Int_t returnCode = WEXITSTATUS(status);
- Log("SHUTTLE", Form("%s: the return code is %d", fCurrentDetector.Data(),
+ Log("SHUTTLE", Form("Process - %s: the return code is %d", fCurrentDetector.Data(),
returnCode));
if (returnCode == 0) hasError = kTRUE;
}
else if (pid == 0)
{
- // client
- AliInfo(Form("In client process of %d - %s", GetCurrentRun(), aDetector->GetName()));
+ // child
+ Log("SHUTTLE", Form("Process - In child process of %d - %s", GetCurrentRun(),
+ aDetector->GetName()));
- AliInfo("Redirecting output...");
+ Log("SHUTTLE", Form("Process - Redirecting output to %s log",fCurrentDetector.Data()));
- if ((freopen(GetLogFileName(fCurrentDetector), "w", stdout)) == 0)
+ if ((freopen(GetLogFileName(fCurrentDetector), "a", stdout)) == 0)
{
- Log("SHUTTLE", "Could not freopen stdout");
+ Log("SHUTTLE", "Process - Could not freopen stdout");
}
else
{
fOutputRedirected = kTRUE;
if ((dup2(fileno(stdout), fileno(stderr))) < 0)
- Log("SHUTTLE", "Could not redirect stderr");
+ Log("SHUTTLE", "Process - Could not redirect stderr");
}
- Bool_t success = ProcessCurrentDetector();
- if (success) // Preprocessor finished successfully!
+ TString wd = gSystem->WorkingDirectory();
+ Int_t dir_lev1 = GetCurrentRun()/10000;
+ TString tmpDir = Form("%s/%d/%d/%s_process", GetShuttleTempDir(),
+ dir_lev1, GetCurrentRun(), fCurrentDetector.Data());
+
+ Int_t result = gSystem->GetPathInfo(tmpDir.Data(), 0, (Long64_t*) 0, 0, 0);
+ if (!result) // temp dir already exists!
+ {
+ Log(fCurrentDetector.Data(),
+ Form("Process - %s dir already exists! Removing...", tmpDir.Data()));
+ gSystem->Exec(Form("rm -rf %s",tmpDir.Data()));
+ }
+
+ if (gSystem->mkdir(tmpDir.Data(), 1))
+ {
+ Log(fCurrentDetector.Data(), "Process - could not make temp directory!!");
+ gSystem->Exit(1);
+ }
+
+ if (!gSystem->ChangeDirectory(tmpDir.Data()))
+ {
+ Log(fCurrentDetector.Data(), "Process - could not change directory!!");
+ gSystem->Exit(1);
+ }
+
+ Int_t success = ProcessCurrentDetector();
+
+ gSystem->ChangeDirectory(wd.Data());
+
+ if (success == 1) // Preprocessor finished successfully!
{
+ // remove temporary folder or DCS map
+ if (!fConfig->KeepTempFolder())
+ {
+ gSystem->Exec(Form("rm -rf %s",tmpDir.Data()));
+ } else if (!fConfig->KeepDCSMap())
+ {
+ gSystem->Exec(Form("rm -f %s/DCSMap.root",tmpDir.Data()));
+ }
+
// Update time_processed field in FXS DB
if (UpdateTable() == kFALSE)
- Log("SHUTTLE", Form("Process - %s: Could not update FXS databases!"));
+ Log("SHUTTLE", Form("Process - %s: Could not update FXS databases!",
+ fCurrentDetector.Data()));
// Transfer the data from local storage to main storage (Grid)
- UpdateShuttleStatus(AliShuttleStatus::kStoreStarted);
if (StoreOCDB() == kFALSE)
- {
- AliInfo(Form("\n \t\t\t****** run %d - %s: STORAGE ERROR ****** \n\n",
- GetCurrentRun(), aDetector->GetName()));
- UpdateShuttleStatus(AliShuttleStatus::kStoreError);
success = kFALSE;
- } else {
- AliInfo(Form("\n \t\t\t****** run %d - %s: DONE ****** \n\n",
- GetCurrentRun(), aDetector->GetName()));
- UpdateShuttleStatus(AliShuttleStatus::kDone);
- UpdateShuttleLogbook(fCurrentDetector, "DONE");
- }
+ }
+ else if (success == 0)
+ {
+ Log("SHUTTLE",
+ Form("\t\t\t****** run %d - %s: PP ERROR ******",
+ GetCurrentRun(), aDetector->GetName()));
}
for (UInt_t iSys=0; iSys<3; iSys++)
if (fFXSCalled[iSys]) fFXSlist[iSys].Clear();
}
- AliInfo(Form("Client process of %d - %s is exiting now with %d.",
+ Log("SHUTTLE", Form("Process - Client process of %d - %s is exiting now with %d.",
GetCurrentRun(), aDetector->GetName(), success));
// the client exits here
}
}
- AliInfo(Form("\n\n \t\t\t^*^*^*^*^*^*^*^*^*^*^*^* run %d: FINISH ^*^*^*^*^*^*^*^*^*^*^*^* \n",
+ Log("SHUTTLE", Form("\t\t\t^*^*^*^*^*^*^*^*^*^*^*^* run %d: FINISH ^*^*^*^*^*^*^*^*^*^*^*^*",
GetCurrentRun()));
//check if shuttle is done for this run, if so update logbook
TObjArray checkEntryArray;
checkEntryArray.SetOwner(1);
TString whereClause = Form("where run=%d", GetCurrentRun());
- if (!QueryShuttleLogbook(whereClause.Data(), checkEntryArray) || checkEntryArray.GetEntries() == 0) {
+ if (!QueryShuttleLogbook(whereClause.Data(), checkEntryArray) ||
+ checkEntryArray.GetEntries() == 0) {
Log("SHUTTLE", Form("Process - Warning: Cannot check status of run %d on Shuttle logbook!",
GetCurrentRun()));
return hasError == kFALSE;
fFirstUnprocessed[iDet] = kFALSE;
}
}
+ SendMLRunInfo("Pending");
}
}
- // remove ML instance
- delete fMonaLisa;
- fMonaLisa = 0;
-
fLogbookEntry = 0;
return hasError == kFALSE;
}
//______________________________________________________________________________________________
-Bool_t AliShuttle::ProcessCurrentDetector()
+Int_t AliShuttle::ProcessCurrentDetector()
{
//
// Makes data retrieval just for a specific detector (fCurrentDetector).
// Threre should be a configuration for this detector.
- AliInfo(Form("Retrieving values for %s, run %d", fCurrentDetector.Data(), GetCurrentRun()));
-
- TMap dcsMap;
- dcsMap.SetOwner(1);
-
- Bool_t aDCSError = kFALSE;
+ Log("SHUTTLE", Form("ProcessCurrentDetector - Retrieving values for %s, run %d",
+ fCurrentDetector.Data(), GetCurrentRun()));
+ TString wd = gSystem->WorkingDirectory();
+
+ if (!CleanReferenceStorage(fCurrentDetector.Data()))
+ return 0;
+
+ gSystem->ChangeDirectory(wd.Data());
+
// call preprocessor
AliPreprocessor* aPreprocessor =
dynamic_cast<AliPreprocessor*> (fPreprocessorMap.GetValue(fCurrentDetector));
+ // check if the preprocessor wants to process this run type
+ if (aPreprocessor->ProcessRunType() == kFALSE)
+ {
+ UpdateShuttleStatus(AliShuttleStatus::kSkipped);
+ UpdateShuttleLogbook(fCurrentDetector, "DONE");
+ if (!UpdateTableSkippedCase(fCurrentDetector.Data()))
+ {
+ AliError(Form("Could not update FXS tables for run %d !", GetCurrentRun()));
+ }
+ Log(fCurrentDetector, Form("ProcessCurrentDetector - %s preprocessor is not interested in this run type", fCurrentDetector.Data()));
+
+ return 2;
+ }
+
+ TMap* dcsMap = new TMap();
+
aPreprocessor->Initialize(GetCurrentRun(), GetCurrentStartTime(), GetCurrentEndTime());
Bool_t processDCS = aPreprocessor->ProcessDCS();
- if (!processDCS || (fTestMode & kSkipDCS))
+ if (!processDCS)
+ {
+ Log(fCurrentDetector, "ProcessCurrentDetector -"
+ " The preprocessor requested to skip the retrieval of DCS values");
+ }
+ else if (fTestMode & kSkipDCS)
{
- Log(fCurrentDetector, "In TESTMODE - Skipping DCS processing!");
+ Log(fCurrentDetector, "ProcessCurrentDetector - In TESTMODE: Skipping DCS processing");
}
else if (fTestMode & kErrorDCS)
{
- Log(fCurrentDetector, "In TESTMODE - Simulating DCS error");
+ Log(fCurrentDetector, "ProcessCurrentDetector - In TESTMODE: Simulating DCS error");
UpdateShuttleStatus(AliShuttleStatus::kDCSStarted);
UpdateShuttleStatus(AliShuttleStatus::kDCSError);
- return kFALSE;
+ delete dcsMap;
+ return 0;
} else {
UpdateShuttleStatus(AliShuttleStatus::kDCSStarted);
- TString host(fConfig->GetDCSHost(fCurrentDetector));
- Int_t port = fConfig->GetDCSPort(fCurrentDetector);
-
- // Retrieval of Aliases
- TObjString* anAlias = 0;
- Int_t iAlias = 1;
- Int_t nTotAliases= ((TMap*)fConfig->GetDCSAliases(fCurrentDetector))->GetEntries();
- TIter iterAliases(fConfig->GetDCSAliases(fCurrentDetector));
- while ((anAlias = (TObjString*) iterAliases.Next()))
+ // Query DCS archive
+ Int_t nServers = fConfig->GetNServers(fCurrentDetector);
+
+ for (int iServ=0; iServ<nServers; iServ++)
{
- TObjArray *valueSet = new TObjArray();
- valueSet->SetOwner(1);
+
+ TString host(fConfig->GetDCSHost(fCurrentDetector, iServ));
+ Int_t port = fConfig->GetDCSPort(fCurrentDetector, iServ);
+ Int_t multiSplit = fConfig->GetMultiSplit(fCurrentDetector, iServ);
- if (((iAlias-1) % 500) == 0 || iAlias == nTotAliases)
- AliInfo(Form("Querying DCS archive: alias %s (%d of %d)",
- anAlias->GetName(), iAlias++, nTotAliases));
- aDCSError = (GetValueSet(host, port, anAlias->String(), valueSet, kAlias) == 0);
+ Log(fCurrentDetector, Form("ProcessCurrentDetector -"
+ " Querying DCS Amanda server %s:%d (%d of %d)",
+ host.Data(), port, iServ+1, nServers));
+
+ TMap* aliasMap = 0;
+ TMap* dpMap = 0;
- if(!aDCSError)
+ if (fConfig->GetDCSAliases(fCurrentDetector, iServ)->GetEntries() > 0)
{
- dcsMap.Add(anAlias->Clone(), valueSet);
- } else {
- Log(fCurrentDetector,
- Form("ProcessCurrentDetector - Error while retrieving alias %s",
- anAlias->GetName()));
- UpdateShuttleStatus(AliShuttleStatus::kDCSError);
- dcsMap.DeleteAll();
- return kFALSE;
- }
- }
+ Log(fCurrentDetector, Form("Querying %d DCS aliases", fConfig->GetDCSAliases(fCurrentDetector, iServ)->GetEntries()));
+ aliasMap = GetValueSet(host, port,
+ fConfig->GetDCSAliases(fCurrentDetector, iServ),
+ kAlias, multiSplit);
+ if (!aliasMap)
+ {
+ Log(fCurrentDetector,
+ Form("ProcessCurrentDetector -"
+ " Error retrieving DCS aliases from server %s."
+ " Sending mail to DCS experts!", host.Data()));
+ UpdateShuttleStatus(AliShuttleStatus::kDCSError);
+
+ if (!SendMail(kDCSEMail))
+ Log("SHUTTLE", Form("ProcessCurrentDetector - "
+ "Could not send mail to DCS experts!"));
- // Retrieval of Data Points
- TObjString* aDP = 0;
- Int_t iDP = 0;
- Int_t nTotDPs= ((TMap*)fConfig->GetDCSDataPoints(fCurrentDetector))->GetEntries();
- TIter iterDP(fConfig->GetDCSDataPoints(fCurrentDetector));
- while ((aDP = (TObjString*) iterDP.Next()))
- {
- TObjArray *valueSet = new TObjArray();
- valueSet->SetOwner(1);
- if (((iDP-1) % 500) == 0 || iDP == nTotDPs)
- AliInfo(Form("Querying DCS archive: DP %s (%d of %d)",
- aDP->GetName(), iDP++, nTotDPs));
- aDCSError = (GetValueSet(host, port, aDP->String(), valueSet, kDP) == 0);
-
- if(!aDCSError)
+ delete dcsMap;
+ return 0;
+ }
+ }
+
+ if (fConfig->GetDCSDataPoints(fCurrentDetector, iServ)->GetEntries() > 0)
{
- dcsMap.Add(aDP->Clone(), valueSet);
- } else {
- Log(fCurrentDetector,
- Form("ProcessCurrentDetector - Error while retrieving data point %s",
- aDP->GetName()));
- UpdateShuttleStatus(AliShuttleStatus::kDCSError);
- dcsMap.DeleteAll();
- return kFALSE;
+ Log(fCurrentDetector, Form("Querying %d DCS data points", fConfig->GetDCSDataPoints(fCurrentDetector, iServ)->GetEntries()));
+ dpMap = GetValueSet(host, port,
+ fConfig->GetDCSDataPoints(fCurrentDetector, iServ),
+ kDP, multiSplit);
+ if (!dpMap)
+ {
+ Log(fCurrentDetector,
+ Form("ProcessCurrentDetector -"
+ " Error retrieving DCS data points from server %s."
+ " Sending mail to DCS experts!", host.Data()));
+ UpdateShuttleStatus(AliShuttleStatus::kDCSError);
+
+ if (!SendMail(kDCSEMail))
+ Log("SHUTTLE", Form("ProcessCurrentDetector - "
+ "Could not send mail to DCS experts!"));
+
+ if (aliasMap) delete aliasMap;
+ delete dcsMap;
+ return 0;
+ }
+ }
+
+ // merge aliasMap and dpMap into dcsMap
+ if(aliasMap) {
+ TIter iter(aliasMap);
+ TObjString* key = 0;
+ while ((key = (TObjString*) iter.Next()))
+ dcsMap->Add(key, aliasMap->GetValue(key->String()));
+
+ aliasMap->SetOwner(kFALSE);
+ delete aliasMap;
+ }
+
+ if(dpMap) {
+ TIter iter(dpMap);
+ TObjString* key = 0;
+ while ((key = (TObjString*) iter.Next()))
+ dcsMap->Add(key, dpMap->GetValue(key->String()));
+
+ dpMap->SetOwner(kFALSE);
+ delete dpMap;
}
}
}
-
+
+ // save map into file, to help debugging in case of preprocessor error
+ TFile* f = TFile::Open("DCSMap.root","recreate");
+ f->cd();
+ dcsMap->Write("DCSMap", TObject::kSingleKey);
+ f->Close();
+ delete f;
+
// DCS Archive DB processing successful. Call Preprocessor!
UpdateShuttleStatus(AliShuttleStatus::kPPStarted);
- UInt_t returnValue = aPreprocessor->Process(&dcsMap);
+ fFXSError = -1; // this variable is kTRUE after ::Process if an FXS error occured
+
+ UInt_t returnValue = aPreprocessor->Process(dcsMap);
+
+ if (fFXSError!=-1) {
+ UpdateShuttleStatus(AliShuttleStatus::kFXSError);
+ SendMail(kFXSEMail, fFXSError);
+ dcsMap->DeleteAll();
+ delete dcsMap;
+ return 0;
+ }
if (returnValue > 0) // Preprocessor error!
{
- Log(fCurrentDetector, Form("Preprocessor failed. Process returned %d.", returnValue));
+ Log(fCurrentDetector, Form("ProcessCurrentDetector - "
+ "Preprocessor failed. Process returned %d.", returnValue));
UpdateShuttleStatus(AliShuttleStatus::kPPError);
- dcsMap.DeleteAll();
- return kFALSE;
+ dcsMap->DeleteAll();
+ delete dcsMap;
+ return 0;
}
// preprocessor ok!
Log(fCurrentDetector, Form("ProcessCurrentDetector - %s preprocessor returned success",
fCurrentDetector.Data()));
- dcsMap.DeleteAll();
+ dcsMap->DeleteAll();
+ delete dcsMap;
- return kTRUE;
+ return 1;
+}
+
+//______________________________________________________________________________________________
+void AliShuttle::CountOpenRuns()
+{
+ // Query DAQ's Shuttle logbook and sends the number of open runs to ML
+
+ SendAlive();
+
+ // check connection, in case connect
+ if (!Connect(3))
+ return;
+
+ TString sqlQuery;
+ sqlQuery = Form("select count(*) from %s where shuttle_done=0", fConfig->GetShuttlelbTable());
+
+ TSQLResult* aResult = fServer[3]->Query(sqlQuery);
+ if (!aResult) {
+ AliError(Form("Can't execute query <%s>!", sqlQuery.Data()));
+ return;
+ }
+
+ AliDebug(2,Form("Query = %s", sqlQuery.Data()));
+
+ if (aResult->GetRowCount() == 0) {
+ AliError(Form("No result for query %s received", sqlQuery.Data()));
+ return;
+ }
+
+ if (aResult->GetFieldCount() != 1) {
+ AliError(Form("Invalid field count for query %s received", sqlQuery.Data()));
+ return;
+ }
+
+ TSQLRow* aRow = aResult->Next();
+ if (!aRow) {
+ AliError(Form("Could not receive result of query %s", sqlQuery.Data()));
+ return;
+ }
+
+ TString result(aRow->GetField(0), aRow->GetFieldLength(0));
+ Int_t count = result.Atoi();
+
+ Log("SHUTTLE", Form("%d unprocessed runs", count));
+
+ delete aRow;
+ delete aResult;
+
+ TMonaLisaValue mlStatus("SHUTTLE_openruns", count);
+
+ TList mlList;
+ mlList.Add(&mlStatus);
+
+ fMonaLisa->SendParameters(&mlList, "__PROCESSINGINFO__");
}
//______________________________________________________________________________________________
entries.SetOwner(1);
// check connection, in case connect
- if(!Connect(3)) return kFALSE;
+ if (!Connect(3)) return kFALSE;
TString sqlQuery;
sqlQuery = Form("select * from %s %s order by run", fConfig->GetShuttlelbTable(), whereClause);
AliDebug(2,Form("Query = %s", sqlQuery.Data()));
if(aResult->GetRowCount() == 0) {
- AliInfo("No entries in Shuttle Logbook match request");
+ Log("SHUTTLE", "No entries in Shuttle Logbook match request");
delete aResult;
return kTRUE;
}
// TODO Check field count!
- const UInt_t nCols = 22;
+ const UInt_t nCols = 25;
if (aResult->GetFieldCount() != (Int_t) nCols) {
- AliError("Invalid SQL result field number!");
+ Log("SHUTTLE", "Invalid SQL result field number!");
delete aResult;
return kFALSE;
}
if (!entry)
continue;
+ // DA test mode flag
+ TString daTestModeString(aRow->GetField(2), aRow->GetFieldLength(2)); // field 2 = DA test mode flag
+ Bool_t daTestMode = (Bool_t)daTestModeString.Atoi();
+ entry->SetDATestMode(daTestMode);
+
// loop on detectors
for(UInt_t ii = 0; ii < nCols; ii++)
entry->SetDetectorStatus(aResult->GetFieldName(ii), aRow->GetField(ii));
TSQLResult* aResult = fServer[3]->Query(sqlQuery);
if (!aResult) {
- AliError(Form("Can't execute query <%s>!", sqlQuery.Data()));
+ Log("SHUTTLE", Form("Can't execute query <%s>!", sqlQuery.Data()));
return 0;
}
}
if (aResult->GetRowCount() > 1) {
- AliError(Form("More than one entry in DAQ Logbook for run %d. Skipping", run));
+ Log("SHUTTLE", Form("QueryRunParameters - UNEXPECTED: "
+ "more than one entry in DAQ Logbook for run %d!", run));
delete aResult;
return 0;
}
TSQLRow* aRow = aResult->Next();
if (!aRow)
{
- AliError(Form("Could not retrieve row for run %d. Skipping", run));
+ Log("SHUTTLE", Form("QueryRunParameters - Could not retrieve row for run %d. Skipping", run));
delete aResult;
return 0;
}
for (Int_t ii = 0; ii < aResult->GetFieldCount(); ii++)
entry->SetRunParameter(aResult->GetFieldName(ii), aRow->GetField(ii));
+ delete aRow;
+ delete aResult;
+
UInt_t startTime = entry->GetStartTime();
UInt_t endTime = entry->GetEndTime();
-
- if (!startTime || !endTime || startTime > endTime) {
- Log("SHUTTLE",
- Form("QueryRunParameters - Invalid parameters for Run %d: startTime = %d, endTime = %d",
- run, startTime, endTime));
- delete entry;
- delete aRow;
- delete aResult;
- return 0;
+ Bool_t ecsSuccess = entry->GetECSSuccess();
+
+ TString totEventsStr = entry->GetRunParameter("totalEvents");
+ Int_t totEvents = totEventsStr.Atoi();
+
+ UInt_t now = time(0);
+ // TODO make this a configuration parameter
+ Int_t dcsDelay = fConfig->GetDCSDelay()+fConfig->GetDCSQueryOffset();
+
+ // runs are accepted if they have ecsSuccess set or more than 1 event
+ if (startTime != 0 && endTime != 0 && endTime > startTime && (totEvents > 1 || ecsSuccess) && (endTime < now - dcsDelay))
+ {
+ if (ecsSuccess == kFALSE)
+ Log("SHUTTLE", Form("Processing run %d although in status ECS failure, Reason: %s", run, entry->GetRunParameter("eor_reason")));
+ return entry;
}
- delete aRow;
- delete aResult;
-
- return entry;
+ Bool_t skip = kFALSE;
+
+ if (endTime != 0 && endTime >= now - dcsDelay)
+ {
+ Log("SHUTTLE", Form("Skipping run %d for now, because DCS buffer time is not yet expired", run));
+ }
+ else if (totEvents <= 1)
+ {
+ Log("SHUTTLE", Form("QueryRunParameters - Run %d has 1 event or less - Skipping!", run));
+ skip = kTRUE;
+ }
+ else
+ {
+ Log("SHUTTLE", Form("QueryRunParameters - Invalid parameters for Run %d: "
+ "startTime = %d, endTime = %d. Skipping (Shuttle won't be marked as DONE)!",
+ run, startTime, endTime));
+ }
+
+ if (skip)
+ {
+ Log("SHUTTLE", Form("Marking SHUTTLE skipped for run %d", run));
+ fLogbookEntry = entry;
+ if (!UpdateShuttleLogbook("shuttle_skipped"))
+ {
+ AliError(Form("Could not update logbook for run %d !", run));
+ }
+ if (!UpdateTableSkippedCase("ALL"))
+ {
+ AliError(Form("Could not update FXS tables for run %d !", run));
+ }
+ fLogbookEntry = 0;
+ }
+
+ delete entry;
+ return 0;
}
//______________________________________________________________________________________________
-Bool_t AliShuttle::GetValueSet(const char* host, Int_t port, const char* entry,
- TObjArray* valueSet, DCSType type)
+TMap* AliShuttle::GetValueSet(const char* host, Int_t port, const TSeqCollection* entries,
+ DCSType type, Int_t multiSplit)
{
// Retrieve all "entry" data points from the DCS server
// host, port: TSocket connection parameters
- // entry: name of the alias or data point
- // valueSet: array of retrieved AliDCSValue's
+ // entries: list of name of the alias or data point
// type: kAlias or kDP
+ // returns TMap of values, 0 when failure
+
+ AliDCSClient client(host, port, fTimeout, fRetries, multiSplit);
- AliDCSClient client(host, port, fTimeout, fRetries);
- if (!client.IsConnected())
- {
- return kFALSE;
- }
-
- Int_t result=0;
-
+ TMap* result = 0;
if (type == kAlias)
{
- result = client.GetAliasValues(entry,
- GetCurrentStartTime(), GetCurrentEndTime(), valueSet);
- } else
- if (type == kDP)
+ //result = client.GetAliasValues(entries, GetCurrentStartTime()-offset,
+ // GetCurrentEndTime()+offset);
+ result = client.GetAliasValues(entries, GetStartTimeDCSQuery(),
+ GetEndTimeDCSQuery());
+ }
+ else if (type == kDP)
{
- result = client.GetDPValues(entry,
- GetCurrentStartTime(), GetCurrentEndTime(), valueSet);
+ //result = client.GetDPValues(entries, GetCurrentStartTime()-offset,
+ // GetCurrentEndTime()+offset);
+ result = client.GetDPValues(entries, GetStartTimeDCSQuery(),
+ GetEndTimeDCSQuery());
}
- if (result < 0)
+ if (result == 0)
{
- Log(fCurrentDetector.Data(), Form("GetValueSet - Can't get '%s'! Reason: %s",
- entry, AliDCSClient::GetErrorString(result)));
-
- if (result == AliDCSClient::fgkServerError)
- {
- Log(fCurrentDetector.Data(), Form("GetValueSet - Server error: %s",
+ Log(fCurrentDetector.Data(), Form("GetValueSet - Can't get entries! Reason: %s",
+ client.GetErrorString(client.GetResultErrorCode())));
+ if (client.GetResultErrorCode() == AliDCSClient::fgkServerError)
+ Log(fCurrentDetector.Data(), Form("GetValueSet - Server error code: %s",
client.GetServerError().Data()));
- }
- return kFALSE;
+ return 0;
}
-
- return kTRUE;
+
+ return result;
}
//______________________________________________________________________________________________
if (!Connect(system))
{
Log(detector, Form("GetFile - Couldn't connect to %s FXS database", GetSystemName(system)));
+ fFXSError = system;
return 0;
}
else if (system == kHLT)
{
whereClause += Form(" and DDLnumbers=\"%s\"", source);
- nFields = 3;
}
TString sqlQuery = Form("%s %s", sqlQueryStart.Data(), whereClause.Data());
TSQLResult* aResult = 0;
aResult = dynamic_cast<TSQLResult*> (fServer[system]->Query(sqlQuery));
if (!aResult) {
- Log(detector, Form("GetFileName - Can't execute SQL query to %s database for: id = %s, source = %s",
+ Log(detector, Form("GetFile - Can't execute SQL query to %s database for: id = %s, source = %s",
GetSystemName(system), id, sourceName.Data()));
+ fFXSError = system;
return 0;
}
- if(aResult->GetRowCount() == 0)
+ if (aResult->GetRowCount() == 0)
{
Log(detector,
- Form("GetFileName - No entry in %s FXS db for: id = %s, source = %s",
+ Form("GetFile - No entry in %s FXS db for: id = %s, source = %s",
GetSystemName(system), id, sourceName.Data()));
delete aResult;
return 0;
if (aResult->GetRowCount() > 1) {
Log(detector,
- Form("GetFileName - More than one entry in %s FXS db for: id = %s, source = %s",
+ Form("GetFile - More than one entry in %s FXS db for: id = %s, source = %s",
GetSystemName(system), id, sourceName.Data()));
+ fFXSError = system;
delete aResult;
return 0;
}
Log(detector,
Form("GetFileName - Wrong field count in %s FXS db for: id = %s, source = %s",
GetSystemName(system), id, sourceName.Data()));
+ fFXSError = system;
delete aResult;
return 0;
}
TSQLRow* aRow = dynamic_cast<TSQLRow*> (aResult->Next());
if (!aRow){
- Log(detector, Form("GetFileName - Empty set result in %s FXS db from query: id = %s, source = %s",
+ Log(detector, Form("GetFile - Empty set result in %s FXS db from query: id = %s, source = %s",
GetSystemName(system), id, sourceName.Data()));
+ fFXSError = system;
delete aResult;
return 0;
}
filePath.Data(), fileSize.Data(), fileChecksum.Data()));
// retrieved file is renamed to make it unique
- TString localFileName = Form("%s_%s_%d_%s_%s.shuttle",
- GetSystemName(system), detector, GetCurrentRun(), id, sourceName.Data());
+ Int_t dir_lev1 = GetCurrentRun()/10000;
+ TString localFileName = Form("%s/%d/%d/%s_process/%s_%s_%d_%s_%s.shuttle",
+ GetShuttleTempDir(), dir_lev1, GetCurrentRun(), detector,
+ GetSystemName(system), detector, GetCurrentRun(),
+ id, sourceName.Data());
+ Log("SHUTTLE",Form("file from FXS = %s",localFileName.Data()));
// file retrieval from FXS
Bool_t result = kFALSE;
// copy!! if successful TSystem::Exec returns 0
- while(nRetries++ < maxRetries) {
+ while (nRetries++ < maxRetries) {
AliDebug(2, Form("Trying to copy file. Retry # %d", nRetries));
result = RetrieveFile(system, filePath.Data(), localFileName.Data());
- if(!result)
+ if (!result)
{
- Log(detector, Form("GetFileName - Copy of file %s from %s FXS failed",
+ Log(detector, Form("GetFile - Copy of file %s from %s FXS failed",
filePath.Data(), GetSystemName(system)));
continue;
+ }
+
+ if (fileSize.Length()>0)
+ {
+ // compare filesize of local file with the one stored in the FXS DB
+ Long_t size = -1;
+ Int_t sizeComp = gSystem->GetPathInfo(localFileName.Data(), 0, &size, 0, 0);
+
+ if (sizeComp != 0 || size != fileSize.Atoi())
+ {
+ Log(detector, Form("GetFile - size of file %s does not match with local copy!",
+ filePath.Data()));
+ result = kFALSE;
+ continue;
+ }
+
} else {
- AliInfo(Form("File %s copied from %s FXS into %s/%s",
- filePath.Data(), GetSystemName(system),
- GetShuttleTempDir(), localFileName.Data()));
- }
+ Log(fCurrentDetector, Form("GetFile - size of file %s not set in %s database, skipping comparison",
+ filePath.Data(), GetSystemName(system)));
+ }
if (fileChecksum.Length()>0)
{
// compare md5sum of local file with the one stored in the FXS DB
- Int_t md5Comp = gSystem->Exec(Form("md5sum %s/%s |grep %s 2>&1 > /dev/null",
- GetShuttleTempDir(), localFileName.Data(), fileChecksum.Data()));
+ if(fileChecksum.Contains(' ')) fileChecksum.Resize(fileChecksum.First(' '));
+ Int_t md5Comp = gSystem->Exec(Form("md5sum %s |grep %s > /dev/null 2> /dev/null",
+ localFileName.Data(), fileChecksum.Data()));
if (md5Comp != 0)
{
- Log(detector, Form("GetFileName - md5sum of file %s does not match with local copy!",
+ Log(detector, Form("GetFile - md5sum of file %s does not match with local copy!",
filePath.Data()));
result = kFALSE;
continue;
if (result) break;
}
- if(!result) return 0;
+ if (!result)
+ {
+ fFXSError = system;
+ return 0;
+ }
fFXSCalled[system]=kTRUE;
TObjString *fileParams = new TObjString(Form("%s#!?!#%s", id, sourceName.Data()));
fFXSlist[system].Add(fileParams);
- static TString fullLocalFileName;
- fullLocalFileName = TString::Format("%s/%s", GetShuttleTempDir(), localFileName.Data());
-
- AliInfo(Form("fullLocalFileName = %s", fullLocalFileName.Data()));
-
- return fullLocalFileName.Data();
-
+ static TString staticLocalFileName;
+ staticLocalFileName.Form("%s", localFileName.Data());
+
+ Log(fCurrentDetector, Form("GetFile - Retrieved file with id %s and "
+ "source %s from %s to %s", id, source,
+ GetSystemName(system), localFileName.Data()));
+
+ return staticLocalFileName.Data();
}
//______________________________________________________________________________________________
//
// check temp directory: trying to cd to temp; if it does not exist, create it
- AliDebug(2, Form("Copy file %s from %s FXS into %s/%s",
- GetSystemName(system), fxsFileName, GetShuttleTempDir(), localFileName));
+ AliDebug(2, Form("Copy file %s from %s FXS into %s",
+ GetSystemName(system), fxsFileName, localFileName));
+
+ TString tmpDir(localFileName);
+
+ tmpDir = tmpDir(0,tmpDir.Last('/'));
- void* dir = gSystem->OpenDirectory(GetShuttleTempDir());
- if (dir == NULL) {
- if (gSystem->mkdir(GetShuttleTempDir(), kTRUE)) {
- AliError(Form("Can't open directory <%s>", GetShuttleTempDir()));
+ Int_t noDir = gSystem->GetPathInfo(tmpDir.Data(), 0, (Long64_t*) 0, 0, 0);
+ if (noDir) // temp dir does not exists!
+ {
+ if (gSystem->mkdir(tmpDir.Data(), 1))
+ {
+ Log(fCurrentDetector.Data(), "RetrieveFile - could not make temp directory!!");
return kFALSE;
}
+ }
+
+ TString command = Form("scp -oPort=%d -2 %s@%s:%s/%s %s",
+ fConfig->GetFXSPort(system),
+ fConfig->GetFXSUser(system),
+ fConfig->GetFXSHost(system),
+ fConfig->GetFXSBaseFolder(system),
+ fxsFileName,
+ localFileName);
+
+ AliDebug(2, Form("%s",command.Data()));
+
+ Bool_t result = (gSystem->Exec(command.Data()) == 0);
+ return result;
+}
+
+//______________________________________________________________________________________________
+TList* AliShuttle::GetFileSources(Int_t system, const char* detector, const char* id)
+{
+ //
+ // Get sources producing the condition file Id from file exchange servers
+ // if id is NULL all sources are returned (distinct)
+ //
+
+ if (id)
+ {
+ Log(detector, Form("GetFileSources - Querying %s FXS for files with id %s produced by %s", GetSystemName(system), id, detector));
} else {
- gSystem->FreeDirectory(dir);
+ Log(detector, Form("GetFileSources - Querying %s FXS for files produced by %s", GetSystemName(system), detector));
+ }
+
+ // check if test mode should simulate a FXS error
+ if (fTestMode & kErrorFXSSources)
+ {
+ Log(detector, Form("GetFileSources - In TESTMODE - Simulating error while connecting to %s FXS", GetSystemName(system)));
+ return 0;
}
- TString baseFXSFolder;
- if (system == kDAQ)
+ if (system == kDCS)
{
- baseFXSFolder = "FES/";
+ Log(detector, "GetFileSources - WARNING: DCS system has only one source of data!");
+ TList *list = new TList();
+ list->SetOwner(1);
+ list->Add(new TObjString(" "));
+ return list;
}
- else if (system == kDCS)
+
+ // check connection, in case connect
+ if (!Connect(system))
{
- baseFXSFolder = "";
+ Log(detector, Form("GetFileSources - Couldn't connect to %s FXS database", GetSystemName(system)));
+ fFXSError = system;
+ return NULL;
}
- else if (system == kHLT)
+
+ TString sourceName = 0;
+ if (system == kDAQ)
{
- baseFXSFolder = "~/";
+ sourceName = "DAQsource";
+ } else if (system == kHLT)
+ {
+ sourceName = "DDLnumbers";
}
+ TString sqlQueryStart = Form("select distinct %s from %s where", sourceName.Data(), fConfig->GetFXSdbTable(system));
+ TString whereClause = Form("run=%d and detector=\"%s\"",
+ GetCurrentRun(), detector);
+ if (id)
+ whereClause += Form(" and fileId=\"%s\"", id);
+ TString sqlQuery = Form("%s %s", sqlQueryStart.Data(), whereClause.Data());
- TString command = Form("scp -oPort=%d -2 %s@%s:%s%s %s/%s",
- fConfig->GetFXSPort(system),
- fConfig->GetFXSUser(system),
- fConfig->GetFXSHost(system),
- baseFXSFolder.Data(),
- fxsFileName,
- GetShuttleTempDir(),
- localFileName);
+ AliDebug(2, Form("SQL query: \n%s",sqlQuery.Data()));
- AliDebug(2, Form("%s",command.Data()));
+ // Query execution
+ TSQLResult* aResult;
+ aResult = fServer[system]->Query(sqlQuery);
+ if (!aResult) {
+ Log(detector, Form("GetFileSources - Can't execute SQL query to %s database for id: %s",
+ GetSystemName(system), id));
+ fFXSError = system;
+ return 0;
+ }
- Bool_t result = (gSystem->Exec(command.Data()) == 0);
+ TList *list = new TList();
+ list->SetOwner(1);
+
+ if (aResult->GetRowCount() == 0)
+ {
+ Log(detector,
+ Form("GetFileSources - No entry in %s FXS table for id: %s", GetSystemName(system), id));
+ delete aResult;
+ return list;
+ }
+
+ Log(detector, Form("GetFileSources - Found %d sources", aResult->GetRowCount()));
+
+ TSQLRow* aRow;
+ while ((aRow = aResult->Next()))
+ {
+
+ TString source(aRow->GetField(0), aRow->GetFieldLength(0));
+ AliDebug(2, Form("%s = %s", sourceName.Data(), source.Data()));
+ list->Add(new TObjString(source));
+ delete aRow;
+ }
+
+ delete aResult;
- return result;
+ return list;
}
//______________________________________________________________________________________________
-TList* AliShuttle::GetFileSources(Int_t system, const char* detector, const char* id)
+TList* AliShuttle::GetFileIDs(Int_t system, const char* detector, const char* source)
{
//
- // Get sources producing the condition file Id from file exchange servers
+ // Get all ids of condition files produced by a given source from file exchange servers
//
+ Log(detector, Form("GetFileIDs - Retrieving ids with source %s with %s", source, GetSystemName(system)));
+
// check if test mode should simulate a FXS error
if (fTestMode & kErrorFXSSources)
{
- Log(detector, Form("GetFileSources - In TESTMODE - Simulating error while connecting to %s FXS", GetSystemName(system)));
+ Log(detector, Form("GetFileIDs - In TESTMODE - Simulating error while connecting to %s FXS", GetSystemName(system)));
return 0;
}
-
- if (system == kDCS)
- {
- AliError("DCS system has only one source of data!");
- return NULL;
- }
-
// check connection, in case connect
if (!Connect(system))
{
- Log(detector, Form("GetFile - Couldn't connect to %s FXS database", GetSystemName(system)));
+ Log(detector, Form("GetFileIDs - Couldn't connect to %s FXS database", GetSystemName(system)));
return NULL;
}
sourceName = "DDLnumbers";
}
- TString sqlQueryStart = Form("select %s from %s where", sourceName.Data(), fConfig->GetFXSdbTable(system));
- TString whereClause = Form("run=%d and detector=\"%s\" and fileId=\"%s\"",
- GetCurrentRun(), detector, id);
+ TString sqlQueryStart = Form("select fileId from %s where", fConfig->GetFXSdbTable(system));
+ TString whereClause = Form("run=%d and detector=\"%s\"",
+ GetCurrentRun(), detector);
+ if (sourceName.Length() > 0 && source)
+ whereClause += Form(" and %s=\"%s\"", sourceName.Data(), source);
TString sqlQuery = Form("%s %s", sqlQueryStart.Data(), whereClause.Data());
AliDebug(2, Form("SQL query: \n%s",sqlQuery.Data()));
TSQLResult* aResult;
aResult = fServer[system]->Query(sqlQuery);
if (!aResult) {
- Log(detector, Form("GetFileSources - Can't execute SQL query to %s database for id: %s",
- GetSystemName(system), id));
+ Log(detector, Form("GetFileIDs - Can't execute SQL query to %s database for source: %s",
+ GetSystemName(system), source));
return 0;
}
+ TList *list = new TList();
+ list->SetOwner(1);
+
if (aResult->GetRowCount() == 0)
{
Log(detector,
- Form("GetFileSources - No entry in %s FXS table for id: %s", GetSystemName(system), id));
+ Form("GetFileIDs - No entry in %s FXS table for source: %s", GetSystemName(system), source));
delete aResult;
- return 0;
+ return list;
}
+ Log(detector, Form("GetFileIDs - Found %d ids", aResult->GetRowCount()));
+
TSQLRow* aRow;
- TList *list = new TList();
- list->SetOwner(1);
while ((aRow = aResult->Next()))
{
- TString source(aRow->GetField(0), aRow->GetFieldLength(0));
- AliDebug(2, Form("%s = %s", sourceName.Data(), source.Data()));
- list->Add(new TObjString(source));
+ TString id(aRow->GetField(0), aRow->GetFieldLength(0));
+ AliDebug(2, Form("fileId = %s", id.Data()));
+ list->Add(new TObjString(id));
delete aRow;
}
//
// check connection: if already connected return
- if(fServer[system] && fServer[system]->IsConnected()) return kTRUE;
+ if(fServer[system] && fServer[system]->IsConnected()) {
+ // ping the server --> automatic reconnection should occur if it was broken but the
+ // server is still alive
+ fServer[system]->Ping();
+ return kTRUE;
+ }
TString dbHost, dbUser, dbPass, dbName;
return result;
}
+//_______________________________________________________________________________
+Bool_t AliShuttle::UpdateTableSkippedCase(const char* detector)
+{
+ //
+ // Update FXS table filling time_processed field in all rows corresponding to current run and detector
+ // if detector = "ALL" update all detectors
+ //
+
+ Bool_t result = kTRUE;
+
+ TString detName(detector);
+
+ for (UInt_t system=0; system<3; system++)
+ {
+
+ // check connection, in case connect
+ if (!Connect(system))
+ {
+ Log(fCurrentDetector, Form("UpdateTableSkippedCase - Couldn't connect to %s FXS database", GetSystemName(system)));
+ result = kFALSE;
+ continue;
+ }
+
+ TTimeStamp now; // now
+
+ // Loop on FXS list entries
+ TIter iter(&fFXSlist[system]);
+
+ TString whereClause;
+ if (detName == "ALL") whereClause = Form("where run=%d and time_processed IS NULL;",GetCurrentRun());
+ else whereClause = Form("where run=%d and detector=\"%s\" and time_processed IS NULL;",GetCurrentRun(), detector);
+
+ Log("SHUTTLE",Form(" whereClause = %s ",whereClause.Data()));
+
+ TString sqlQuery = Form("update %s set time_processed=%d %s", fConfig->GetFXSdbTable(system),
+ now.GetSec(), whereClause.Data());
+
+ AliDebug(2, Form("SQL query: \n%s",sqlQuery.Data()));
+
+ // Query execution
+ TSQLResult* aResult;
+ aResult = dynamic_cast<TSQLResult*> (fServer[system]->Query(sqlQuery));
+ if (!aResult)
+ {
+ Log("SHUTTLE", Form("UpdateTableSkippedCase - %s db: can't execute SQL query <%s>",
+ GetSystemName(system), sqlQuery.Data()));
+ result = kFALSE;
+ continue;
+ }
+ delete aResult;
+
+ }
+
+ return result;
+}
//______________________________________________________________________________________________
Bool_t AliShuttle::UpdateTableFailCase()
{
TString detName(detector);
TString setClause;
- if(detName == "shuttle_done")
+ if (detName == "shuttle_done" || detName == "shuttle_skipped")
{
setClause = "set shuttle_done=1";
-
- // Send the information to ML
- TMonaLisaText mlStatus("SHUTTLE_status", "Done");
-
- TList mlList;
- mlList.Add(&mlStatus);
-
- fMonaLisa->SendParameters(&mlList);
- } else {
+
+ if (detName == "shuttle_done")
+ {
+ if (TouchFile() != kTRUE)
+ {
+ SendMLRunInfo("Pending");
+ return kFALSE;
+ }
+
+ SendMLRunInfo("Done");
+ }
+ else
+ SendMLRunInfo("Skipped");
+ }
+ else {
TString statusStr(status);
if(statusStr.Contains("done", TString::kIgnoreCase) ||
statusStr.Contains("failed", TString::kIgnoreCase)){
return fLogbookEntry ? fLogbookEntry->GetEndTime() : 0;
}
+//______________________________________________________________________________________________
+UInt_t AliShuttle::GetCurrentTimeCreated() const
+{
+ //
+ // get current end time from logbook entry
+ //
+
+ return fLogbookEntry ? fLogbookEntry->GetTimeCreated() : 0;
+}
+
+//______________________________________________________________________________________________
+UInt_t AliShuttle::GetCurrentYear() const
+{
+ //
+ // Get current year from logbook entry
+ //
+
+ if (!fLogbookEntry) return 0;
+
+ TTimeStamp startTime(GetCurrentStartTime());
+ TString year = Form("%d",startTime.GetDate());
+ year = year(0,4);
+
+ return year.Atoi();
+}
+
+//______________________________________________________________________________________________
+const char* AliShuttle::GetLHCPeriod() const
+{
+ //
+ // Get current LHC period from logbook entry
+ //
+
+ if (!fLogbookEntry) return 0;
+
+ return fLogbookEntry->GetRunParameter("LHCperiod");
+}
//______________________________________________________________________________________________
void AliShuttle::Log(const char* detector, const char* message)
//
// Fill log string with a message
//
-
- void* dir = gSystem->OpenDirectory(GetShuttleLogDir());
+
+ TString logRunDir = GetShuttleLogDir();
+ if (GetCurrentRun() >=0) {
+ Int_t logDir_lev1 = GetCurrentRun()/10000;
+ logRunDir += Form("/%d/%d", logDir_lev1, GetCurrentRun());
+ }
+ void* dir = gSystem->OpenDirectory(logRunDir.Data());
if (dir == NULL) {
- if (gSystem->mkdir(GetShuttleLogDir(), kTRUE)) {
+ if (gSystem->mkdir(logRunDir.Data(), kTRUE)) {
AliError(Form("Can't open directory <%s>", GetShuttleLogDir()));
return;
}
gSystem->FreeDirectory(dir);
}
- TString toLog = Form("%s (%d): %s - ", TTimeStamp(time(0)).AsString("s"), getpid(), detector);
+ TString toLog = Form("%s UTC (%d): %s - ", TTimeStamp(time(0)).AsString("s"), getpid(), detector);
if (GetCurrentRun() >= 0)
toLog += Form("run %d - ", GetCurrentRun());
toLog += Form("%s", message);
TString fileName;
if (GetCurrentRun() >= 0)
- fileName.Form("%s/%s_%d.log", GetShuttleLogDir(), detector, GetCurrentRun());
- else
+ {
+ Int_t logDir_lev1 = GetCurrentRun()/10000;
+ fileName.Form("%s/%d/%d/%s.log", GetShuttleLogDir(), logDir_lev1, GetCurrentRun(),
+ detector);
+ } else {
fileName.Form("%s/%s.log", GetShuttleLogDir(), detector);
+ }
return fileName;
}
+//______________________________________________________________________________________________
+void AliShuttle::SendAlive()
+{
+ // sends alive message to ML
+
+ TMonaLisaText mlStatus("SHUTTLE_status", "Alive");
+
+ TList mlList;
+ mlList.Add(&mlStatus);
+
+ fMonaLisa->SendParameters(&mlList, "__PROCESSINGINFO__");
+}
+
//______________________________________________________________________________________________
Bool_t AliShuttle::Collect(Int_t run)
{
SetLastAction("Starting");
+ // create ML instance
+ if (!fMonaLisa)
+ fMonaLisa = new TMonaLisaWriter(fConfig->GetMonitorHost(), fConfig->GetMonitorTable());
+
+ CountOpenRuns();
+
TString whereClause("where shuttle_done=0");
if (run != -1)
whereClause += Form(" and run=%d", run);
{
// query Shuttle logbook for earlier runs, check if some detectors are unprocessed,
// flag them into fFirstUnprocessed array
- TString whereClause(Form("where shuttle_done=0 and run < %d", run));
+ TString whereClauseBis(Form("where shuttle_done=0 and run < %d", run));
TObjArray tmpLogbookEntries;
- if (!QueryShuttleLogbook(whereClause, tmpLogbookEntries))
+ if (!QueryShuttleLogbook(whereClauseBis, tmpLogbookEntries))
{
Log("SHUTTLE", "Collect - Can't retrieve entries from Shuttle logbook");
return kFALSE;
if (!RetrieveConditionsData(shuttleLogbookEntries))
{
Log("SHUTTLE", "Collect - Process of at least one run failed");
+ CountOpenRuns();
return kFALSE;
}
Log("SHUTTLE", "Collect - Requested run(s) successfully processed");
+ CountOpenRuns();
return kTRUE;
}
}
// clean SHUTTLE temp directory
- TString filename = Form("%s/*.shuttle", GetShuttleTempDir());
- RemoveFile(filename.Data());
+ //TString filename = Form("%s/*.shuttle", GetShuttleTempDir());
+ //RemoveFile(filename.Data());
}
return hasError == kFALSE;
}
//______________________________________________________________________________________________
-Bool_t AliShuttle::SendMail()
+Bool_t AliShuttle::SendMail(EMailTarget target, Int_t system)
{
//
// sends a mail to the subdetector expert in case of preprocessor error
if (fTestMode != kNone)
return kTRUE;
+
+ if (!fConfig->SendMail())
+ return kTRUE;
+
+ if (target == kDCSEMail || target == kFXSEMail) {
+ if (!fFirstProcessing)
+ return kTRUE;
+ }
+ Int_t runMode = (Int_t)fConfig->GetRunMode();
+ TString tmpStr;
+ if (runMode == 0) tmpStr = " Nightly Test:";
+ else tmpStr = " Data Taking:";
void* dir = gSystem->OpenDirectory(GetShuttleLogDir());
if (dir == NULL)
{
if (gSystem->mkdir(GetShuttleLogDir(), kTRUE))
{
- AliError(Form("Can't open directory <%s>", GetShuttleLogDir()));
+ Log("SHUTTLE", Form("SendMail - Can't open directory <%s>", GetShuttleLogDir()));
return kFALSE;
}
gSystem->FreeDirectory(dir);
}
+ // det experts in to
+ TString to="";
+ TIter *iterExperts = 0;
+ if (target == kDCSEMail) {
+ iterExperts = new TIter(fConfig->GetAdmins(AliShuttleConfig::kAmanda));
+ }
+ else if (target == kFXSEMail) {
+ iterExperts = new TIter(fConfig->GetAdmins(system));
+ }
+ if (iterExperts) {
+ TObjString *anExpert=0;
+ while ((anExpert = (TObjString*) iterExperts->Next()))
+ {
+ to += Form("%s,", anExpert->GetName());
+ }
+ delete iterExperts;
+ }
+
+ // add subdetector experts
+ iterExperts = new TIter(fConfig->GetResponsibles(fCurrentDetector));
+ TObjString *anExpert=0;
+ while ((anExpert = (TObjString*) iterExperts->Next()))
+ {
+ to += Form("%s,", anExpert->GetName());
+ }
+ delete iterExperts;
+
+ if (to.Length() > 0)
+ to.Remove(to.Length()-1);
+ AliDebug(2, Form("to: %s",to.Data()));
+
+ if (to.IsNull()) {
+ Log("SHUTTLE", Form("List of %d responsibles not set!", (Int_t) target));
+ return kFALSE;
+ }
+
+ // SHUTTLE responsibles in cc
+ TString cc="";
+ TIter iterAdmins(fConfig->GetAdmins(AliShuttleConfig::kGlobal));
+ TObjString *anAdmin=0;
+ while ((anAdmin = (TObjString*) iterAdmins.Next()))
+ {
+ cc += Form("%s,", anAdmin->GetName());
+ }
+ if (cc.Length() > 0)
+ cc.Remove(cc.Length()-1);
+ AliDebug(2, Form("cc: %s",to.Data()));
+
+ // mail body
TString bodyFileName;
bodyFileName.Form("%s/mail.body", GetShuttleLogDir());
gSystem->ExpandPathName(bodyFileName);
if (!mailBody.is_open())
{
- AliError(Form("Could not open mail body file %s", bodyFileName.Data()));
+ Log("SHUTTLE", Form("Could not open mail body file %s", bodyFileName.Data()));
return kFALSE;
}
- TString to="";
- TIter iterExperts(fConfig->GetResponsibles(fCurrentDetector));
- TObjString *anExpert=0;
- while ((anExpert = (TObjString*) iterExperts.Next()))
- {
- to += Form("%s,", anExpert->GetName());
- }
- to.Remove(to.Length()-1);
- AliDebug(2, Form("to: %s",to.Data()));
-
- // TODO this will be removed...
- if (to.Contains("not_yet_set")) {
- AliInfo("List of detector responsibles not yet set!");
- return kFALSE;
- }
- TString cc="alberto.colla@cern.ch";
+ TString subject;
+ TString body;
- TString subject = Form("%s Shuttle preprocessor error in run %d !",
- fCurrentDetector.Data(), GetCurrentRun());
- AliDebug(2, Form("subject: %s", subject.Data()));
+ if (target == kDCSEMail){
+ subject = Form("%s CRITICAL Retrieval of data points for %s FAILED in run %d !",
+ tmpStr.Data(), fCurrentDetector.Data(), GetCurrentRun());
+ AliDebug(2, Form("subject: %s", subject.Data()));
+
+ body = Form("Dear DCS experts, \n\n");
+ body += Form("SHUTTLE couldn\'t retrieve the data points for detector %s "
+ "in run %d!!\n\n", fCurrentDetector.Data(), GetCurrentRun());
+ }
+ else if (target == kFXSEMail){
+ subject = Form("%s CRITICAL FXS communication for %s FAILED in run %d !",
+ tmpStr.Data(), fCurrentDetector.Data(), GetCurrentRun());
+ AliDebug(2, Form("subject: %s", subject.Data()));
+ TString sys;
+ if (system == kDAQ) sys="DAQ";
+ else if (system == kDCS) sys="DCS";
+ else if (system == kHLT) sys="HLT";
+ else return kFALSE;
+ body = Form("Dear %s FXS experts, \n\n",sys.Data());
+ body += Form("SHUTTLE couldn\'t retrieve data from the FXS for detector %s "
+ "in run %d!!\n\n", fCurrentDetector.Data(), GetCurrentRun());
+ }
+ else {
+ subject = Form("%s %s Shuttle preprocessor FAILED in run %d (run type = %s)!",
+ tmpStr.Data(), fCurrentDetector.Data(), GetCurrentRun(), GetRunType());
+ AliDebug(2, Form("subject: %s", subject.Data()));
+
+ body = Form("Dear %s expert(s), \n\n", fCurrentDetector.Data());
+ body += Form("SHUTTLE just detected that your preprocessor "
+ "failed processing run %d (run type = %s)!!\n\n",
+ GetCurrentRun(), GetRunType());
+ }
- TString body = Form("Dear %s expert(s), \n\n", fCurrentDetector.Data());
- body += Form("SHUTTLE just detected that your preprocessor "
- "exited with ERROR state in run %d!!\n\n", GetCurrentRun());
- body += Form("Please check %s status on the web page asap!\n\n", fCurrentDetector.Data());
- body += Form("The last 10 lines of %s log file are following:\n\n");
+ body += Form("Please check %s status on the SHUTTLE monitoring page: \n\n",
+ fCurrentDetector.Data());
+ if (fConfig->GetRunMode() == AliShuttleConfig::kTest)
+ {
+ body += Form("\thttp://pcalimonitor.cern.ch/shuttle.jsp?time=24 \n\n");
+ } else {
+ body += Form("\thttp://pcalimonitor.cern.ch/shuttle.jsp?instance=PROD&time=24 \n\n");
+ }
+
+
+ TString logFolder = "logs";
+ if (fConfig->GetRunMode() == AliShuttleConfig::kProd)
+ logFolder += "_PROD";
+
+
+ body += Form("Find the %s log for the current run on \n\n"
+ "\thttp://pcalishuttle01.cern.ch:8880/%s/%d/%d/%s.log \n\n",
+ fCurrentDetector.Data(), logFolder.Data(), GetCurrentRun()/10000,
+ GetCurrentRun(), fCurrentDetector.Data());
+ body += Form("The last 10 lines of %s log file are following:\n\n", fCurrentDetector.Data());
AliDebug(2, Form("Body begin: %s", body.Data()));
mailBody.close();
mailBody.open(bodyFileName, ofstream::out | ofstream::app);
- TString logFileName = Form("%s/%s_%d.log", GetShuttleLogDir(), fCurrentDetector.Data(), GetCurrentRun());
+ TString logFileName = Form("%s/%d/%d/%s.log", GetShuttleLogDir(),
+ GetCurrentRun()/10000, GetCurrentRun(), fCurrentDetector.Data());
TString tailCommand = Form("tail -n 10 %s >> %s", logFileName.Data(), bodyFileName.Data());
if (gSystem->Exec(tailCommand.Data()))
{
TString endBody = Form("------------------------------------------------------\n\n");
endBody += Form("In case of problems please contact the SHUTTLE core team.\n\n");
endBody += "Please do not answer this message directly, it is automatically generated.\n\n";
- endBody += "Sincerely yours,\n\n \t\t\tthe SHUTTLE\n";
+ endBody += "Greetings,\n\n \t\t\tthe SHUTTLE\n";
AliDebug(2, Form("Body end: %s", endBody.Data()));
return result == 0;
}
-
//______________________________________________________________________________________________
const char* AliShuttle::GetRunType()
{
return fLogbookEntry->GetRunType();
}
+//______________________________________________________________________________________________
+Bool_t AliShuttle::GetHLTStatus()
+{
+ // Return HLT status (ON=1 OFF=0)
+ // Converts the HLT status from the mode string read in the run logbook (not just a bool)
+
+ if(!fLogbookEntry) {
+ AliError("No logbook entry!");
+ return 0;
+ }
+
+ // TODO implement when HLTMode is inserted in run logbook
+ TString hltMode = fLogbookEntry->GetRunParameter("HLTmode");
+ TSubString firstChar = hltMode(0,1);
+ AliDebug(2,Form("First char = %s ",firstChar.Data()));
+ if (firstChar == "A") {
+ return kFALSE;
+ }
+ else if ((firstChar == "B") || (firstChar == "C") || (firstChar == "D") || (firstChar == "E")) {
+ return kTRUE;
+ }
+ else {
+ Log("SHUTTLE","Unexpected HLT mode! Returning 0....");
+ return kFALSE;
+ }
+}
+
+//______________________________________________________________________________________________
+const char* AliShuttle::GetTriggerConfiguration()
+{
+ // Receives the trigger configuration from the DAQ logbook for the current run
+
+ // check connection, if needed reconnect
+ if (!Connect(3))
+ return 0;
+
+ TString sqlQuery;
+ sqlQuery.Form("SELECT configFile FROM logbook_trigger_config WHERE run = %d", GetCurrentRun());
+ TSQLResult* result = fServer[3]->Query(sqlQuery);
+ if (!result)
+ {
+ Log("SHUTTLE", Form("ERROR: Can't execute query <%s>!", sqlQuery.Data()));
+ return 0;
+ }
+
+ if (result->GetRowCount() == 0)
+ {
+ Log("SHUTTLE", "ERROR: Trigger configuration not found in logbook_trigger_config");
+ delete result;
+ return 0;
+ }
+
+ TSQLRow* row = result->Next();
+ if (!row)
+ {
+ Log("SHUTTLE", "ERROR: Could not receive logbook_trigger_config data");
+ delete result;
+ return 0;
+ }
+
+ // static, so that pointer remains valid when it is returned to the calling class
+ static TString triggerConfig(row->GetField(0));
+
+ delete row;
+ row = 0;
+
+ delete result;
+ result = 0;
+
+ Log("SHUTTLE", Form("Found trigger configuration: %s", triggerConfig.Data()));
+
+ return triggerConfig;
+}
+
//______________________________________________________________________________________________
void AliShuttle::SetShuttleTempDir(const char* tmpDir)
{
fgkShuttleLogDir = gSystem->ExpandPathName(logDir);
}
+//______________________________________________________________________________________________
+Bool_t AliShuttle::TouchFile()
+{
+ //
+ // touching a file on the grid if run has been DONE
+ //
+
+ if (!gGrid)
+ {
+ Log("SHUTTLE",Form("No TGrid connection estabilished!"));
+ Log("SHUTTLE",Form("Could not touch file for run %i",GetCurrentRun()));
+ return kFALSE;
+ }
+
+ TString dir;
+ dir.Form("%s%d/SHUTTLE_DONE", fConfig->GetAlienPath(), GetCurrentYear());
+ // checking whether directory for touch command exists
+ TString commandLs;
+ commandLs.Form("ls %s",dir.Data());
+ TGridResult *resultLs = dynamic_cast<TGridResult*>(gGrid->Command(commandLs));
+ if (!resultLs){
+ Log("SHUTTLE",Form("No result for %s command, returning without touching",commandLs.Data()));
+ return kFALSE;
+ }
+ TMap *mapLs = dynamic_cast<TMap*>(resultLs->At(0));
+ if (!mapLs){
+ Log("SHUTTLE",Form("No map for %s command, returning without touching",commandLs.Data()));
+ delete resultLs;
+ resultLs = 0x0;
+ return kFALSE;
+ }
+ TObjString *valueLsPath = dynamic_cast<TObjString*>(mapLs->GetValue("path"));
+ if (!valueLsPath || (TString)(valueLsPath->GetString()).CompareTo(dir)!=1){
+ Log("SHUTTLE",Form("No directory %s found, creating it",dir.Data()));
+
+ // creating the directory
+
+ Bool_t boolMkdir = gGrid->Mkdir(dir.Data());
+ if (!boolMkdir) {
+ Log("SHUTTLE",Form("Impossible to create dir %s in alien catalogue for run %i!",dir.Data(),GetCurrentRun()));
+ delete resultLs;
+ resultLs = 0x0;
+ return kFALSE;
+ }
+ Log("SHUTTLE",Form("Directory %s successfully created in alien catalogue for run %i",dir.Data(),GetCurrentRun()));
+ }
+ else {
+ Log("SHUTTLE",Form("Directory %s correctly found for run %i",dir.Data(),GetCurrentRun()));
+ }
+
+ delete resultLs;
+ resultLs = 0x0;
+
+ TString command;
+ command.Form("touch %s/%i", dir.Data(), GetCurrentRun());
+ Log("SHUTTLE", Form("Creating entry in file catalog: %s", command.Data()));
+ TGridResult *resultTouch = dynamic_cast<TGridResult*>(gGrid->Command(command));
+ if (!resultTouch){
+ Log("SHUTTLE",Form("No result for touching command, returning without touching for run %i",GetCurrentRun()));
+ return kFALSE;
+ }
+ TMap *mapTouch = dynamic_cast<TMap*>(resultTouch->At(0));
+ if (!mapTouch){
+ Log("SHUTTLE",Form("No map for touching command, returning without touching for run %i",GetCurrentRun()));
+ delete resultTouch;
+ resultTouch = 0x0;
+ return kFALSE;
+ }
+ TObjString *valueTouch = dynamic_cast<TObjString*>(mapTouch->GetValue("__result__"));
+ if (!valueTouch){
+ Log("SHUTTLE",Form("No value for \"__result__\" key set in the map for touching command, returning without touching for run %i",GetCurrentRun()));
+ delete resultTouch;
+ resultTouch = 0x0;
+ return kFALSE;
+ }
+ if (valueTouch->GetString()!="1"){
+ Log("SHUTTLE",Form("Failing the touching command, returning without touching for run %i",GetCurrentRun()));
+ delete resultTouch;
+ resultTouch = 0x0;
+ return kFALSE;
+ }
+ delete resultTouch;
+ resultTouch = 0x0;
+ Log("SHUTTLE", "Sucessfully touched the file");
+ return kTRUE;
+}
+//______________________________________________________________________________________________
+UInt_t AliShuttle::GetStartTimeDCSQuery()
+{
+ // Return Start Time for the DCS query
+ //
+ // The call is delegated to AliShuttleInterface
+
+ return GetCurrentTimeCreated()-fConfig->GetDCSQueryOffset();
+}
+//______________________________________________________________________________________________
+UInt_t AliShuttle::GetEndTimeDCSQuery()
+{
+ // Return End Time for the DCS query
+ //
+ // The call is delegated to AliShuttleInterface
+
+ return GetCurrentEndTime()+fConfig->GetDCSQueryOffset();
+}
+
git://git.uio.no / u / mrichter / AliRoot.git / blobdiff