* provided "as is" without express or implied warranty. *
**************************************************************************/
-/*
-$Log$
-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 "AliCDBPath.h"
#include "AliCDBEntry.h"
#include "AliShuttleConfig.h"
-#include "AliDCSClient.h"
+#include "DCSClient/AliDCSClient.h"
#include "AliLog.h"
#include "AliPreprocessor.h"
#include "AliShuttleStatus.h"
+#include "AliShuttleLogbookEntry.h"
#include <TSystem.h>
#include <TObject.h>
#include <TSQLServer.h>
#include <TSQLResult.h>
#include <TSQLRow.h>
+#include <TMutex.h>
+#include <TSystemDirectory.h>
+#include <TSystemFile.h>
+#include <TFile.h>
+#include <TGrid.h>
+#include <TGridResult.h>
+#include <TMap.h>
-#include <fstream>
-
-ClassImp(AliShuttle)
-
-TString AliShuttle::fgkMainCDB("alien://DBFolder=ShuttleCDB");
-TString AliShuttle::fgkLocalCDB("local://LocalShuttleCDB");
-TString AliShuttle::fgkMainRefStorage("alien://DBFolder=ShuttleReference");
-TString AliShuttle::fgkLocalRefStorage("local://LocalReferenceStorage");
-
-Bool_t AliShuttle::fgkProcessDCS(kTRUE);
+#include <TMonaLisaWriter.h>
+#include <fstream>
-const char* AliShuttle::fgkShuttleTempDir = gSystem->ExpandPathName("$ALICE_ROOT/SHUTTLE/temp");
-const char* AliShuttle::fgkShuttleLogDir = gSystem->ExpandPathName("$ALICE_ROOT/SHUTTLE/log");
+#include <sys/types.h>
+#include <sys/wait.h>
-const char* AliShuttle::fgkDetectorName[AliShuttle::fgkNDetectors] = {"SPD", "SDD", "SSD", "TPC", "TRD", "TOF",
- "PHOS", "CPV", "RICH", "EMCAL", "MUON_TRK", "MUON_TRG", "FMD", "ZDC", "PMD", "START", "VZERO"};
+#include <signal.h>
-const char* AliShuttle::fgkDetectorCode[AliShuttle::fgkNDetectors] = {"SPD", "SDD", "SSD", "TPC", "TRD", "TOF",
- "PHS", "CPV", "HMP", "EMC", "MCH", "MTR", "FMD", "ZDC", "PMD", "T00", "V00"};
+ClassImp(AliShuttle)
//______________________________________________________________________________________________
AliShuttle::AliShuttle(const AliShuttleConfig* config,
fConfig(config),
fTimeout(timeout), fRetries(retries),
fPreprocessorMap(),
-fCurrentRun(-1),
-fCurrentStartTime(0), fCurrentEndTime(0),
-fCurrentDetector(""),
+fLogbookEntry(0),
+fCurrentDetector(),
+fFirstProcessing(0),
+fFXSError(-1),
fStatusEntry(0),
-fGridError(kFALSE)
+fMonitoringMutex(0),
+fLastActionTime(0),
+fLastAction(),
+fMonaLisa(0),
+fTestMode(kNone),
+fReadTestMode(kFALSE),
+fOutputRedirected(kFALSE)
{
//
// config: AliShuttleConfig used
//
if (!fConfig->IsValid()) AliFatal("********** !!!!! Invalid configuration !!!!! **********");
- for(int iSys=0;iSys<3;iSys++) {
+ for(int iSys=0;iSys<4;iSys++) {
fServer[iSys]=0;
- fFESlist[iSys].SetOwner(kTRUE);
+ if (iSys < 3)
+ fFXSlist[iSys].SetOwner(kTRUE);
}
-}
-
-//______________________________________________________________________
-AliShuttle::AliShuttle(const AliShuttle& /*other*/):
-AliShuttleInterface(),
-fConfig(0),
-fTimeout(0), fRetries(0),
-fPreprocessorMap(),
-fCurrentRun(-1),
-fCurrentStartTime(0), fCurrentEndTime(0),
-fCurrentDetector(""),
-fStatusEntry(0),
-fGridError(kFALSE)
-{
-// copy constructor (not implemented)
-
-}
+ fPreprocessorMap.SetOwner(kTRUE);
-//______________________________________________________________________
-AliShuttle &AliShuttle::operator=(const AliShuttle& /*other*/)
-{
-// assignment operator (not implemented)
+ for (UInt_t iDet=0; iDet<NDetectors(); iDet++)
+ fFirstUnprocessed[iDet] = kFALSE;
-return *this;
+ fMonitoringMutex = new TMutex();
}
//______________________________________________________________________________________________
AliShuttle::~AliShuttle()
{
-// destructor
+ //
+ // destructor
+ //
fPreprocessorMap.DeleteAll();
- for(int iSys=0;iSys<3;iSys++)
+ for(int iSys=0;iSys<4;iSys++)
if(fServer[iSys]) {
fServer[iSys]->Close();
delete fServer[iSys];
+ fServer[iSys] = 0;
}
+
+ if (fStatusEntry){
+ delete fStatusEntry;
+ fStatusEntry = 0;
+ }
+
+ if (fMonitoringMutex)
+ {
+ delete fMonitoringMutex;
+ fMonitoringMutex = 0;
+ }
}
//______________________________________________________________________________________________
// with the same identificator (GetName()).
//
- if (fPreprocessorMap.GetValue(preprocessor->GetName())) {
- AliWarning(Form("AliPreprocessor %s is already registered!",
- preprocessor->GetName()));
+ const char* detName = preprocessor->GetName();
+ if(GetDetPos(detName) < 0)
+ AliFatal(Form("********** !!!!! Invalid detector name: %s !!!!! **********", detName));
+
+ if (fPreprocessorMap.GetValue(detName)) {
+ AliWarning(Form("AliPreprocessor %s is already registered!", detName));
return;
}
- fPreprocessorMap.Add(new TObjString(preprocessor->GetName()), preprocessor);
+ fPreprocessorMap.Add(new TObjString(detName), preprocessor);
}
-
//______________________________________________________________________________________________
-UInt_t AliShuttle::Store(const AliCDBPath& path, TObject* object,
+Bool_t AliShuttle::Store(const AliCDBPath& path, TObject* object,
AliCDBMetaData* metaData, Int_t validityStart, Bool_t validityInfinite)
{
- // Stores a CDB object in the storage for offline reconstruction. Objects that are not needed for
- // offline reconstruction, but should be stored anyway (e.g. for debugging) should NOT be stored
- // using this function. Use StoreReferenceData instead!
- // It calls WriteToCDB function which perform actual storage
-
- return WriteToCDB(fgkMainCDB, fgkLocalCDB, path, object,
- metaData, validityStart, validityInfinite);
+ // Stores a CDB object in the storage for offline reconstruction. Objects that are not needed for
+ // offline reconstruction, but should be stored anyway (e.g. for debugging) should NOT be stored
+ // using this function. Use StoreReferenceData instead!
+ // It calls StoreLocally function which temporarily stores the data locally; when the preprocessor
+ // finishes the data are transferred to the main storage (Grid).
+ return StoreLocally(fgkLocalCDB, path, object, metaData, validityStart, validityInfinite);
}
//______________________________________________________________________________________________
-UInt_t AliShuttle::StoreReferenceData(const AliCDBPath& path, TObject* object,
- AliCDBMetaData* metaData, Int_t validityStart, Bool_t validityInfinite)
+Bool_t AliShuttle::StoreReferenceData(const AliCDBPath& path, TObject* object, AliCDBMetaData* metaData)
{
- // Stores a CDB object in the storage for reference data. This objects will not be available during
- // offline reconstrunction. Use this function for reference data only!
- // It calls WriteToCDB function which perform actual storage
-
- return WriteToCDB(fgkMainRefStorage, fgkLocalRefStorage, path, object,
- metaData, validityStart, validityInfinite);
+ // Stores a CDB object in the storage for reference data. This objects will not be available during
+ // offline reconstrunction. Use this function for reference data only!
+ // It calls StoreLocally function which temporarily stores the data locally; when the preprocessor
+ // finishes the data are transferred to the main storage (Grid).
+ return StoreLocally(fgkLocalRefStorage, path, object, metaData);
}
//______________________________________________________________________________________________
-UInt_t AliShuttle::WriteToCDB(const char* mainUri, const char* localUri,
+Bool_t AliShuttle::StoreLocally(const TString& localUri,
const AliCDBPath& path, TObject* object, AliCDBMetaData* metaData,
Int_t validityStart, Bool_t validityInfinite)
{
- // write object into the CDB. Parameters are passed by Store and StoreReferenceData functions.
- // The parameters are:
- // 1) Uri of the main storage (Grid)
- // 2) Uri of the backup storage (Local)
- // 3) the object's path.
- // 4) the object to be stored
- // 5) the metaData to be associated with the object
- // 6) the validity start run number w.r.t. the current run,
- // if the data is valid only for this run leave the default 0
- // 7) specifies if the calibration data is valid for infinity (this means until updated),
- // typical for calibration runs, the default is kFALSE
- //
- // returns 0 if fail
- // 1 if stored in main (Grid) storage
- // 2 if stored in backup (Local) storage
-
- const char* cdbType = (mainUri == fgkMainCDB) ? "CDB" : "Reference";
+ // Store object temporarily in local storage. Parameters are passed by Store and StoreReferenceData functions.
+ // when the preprocessor finishes the data are transferred to the main storage (Grid).
+ // The parameters are:
+ // 1) Uri of the backup storage (Local)
+ // 2) the object's path.
+ // 3) the object to be stored
+ // 4) the metaData to be associated with the object
+ // 5) the validity start run number w.r.t. the current run,
+ // if the data is valid only for this run leave the default 0
+ // 6) specifies if the calibration data is valid for infinity (this means until updated),
+ // typical for calibration runs, the default is kFALSE
+ //
+ // returns 0 if fail, 1 otherwise
+
+ if (fTestMode & kErrorStorage)
+ {
+ Log(fCurrentDetector, "StoreLocally - In TESTMODE - Simulating error while storing locally");
+ return kFALSE;
+ }
+ const char* cdbType = (localUri == fgkLocalCDB) ? "CDB" : "Reference";
+
Int_t firstRun = GetCurrentRun() - validityStart;
if(firstRun < 0) {
- AliError("First valid run happens to be less than 0! Setting it to 0...");
+ AliWarning("First valid run happens to be less than 0! Setting it to 0.");
firstRun=0;
}
lastRun = GetCurrentRun();
}
- AliCDBId id(path, firstRun, lastRun);
+ // Version is set to current run, it will be used later to transfer data to Grid
+ AliCDBId id(path, firstRun, lastRun, GetCurrentRun(), -1);
+
+ if(! dynamic_cast<TObjString*> (metaData->GetProperty("RunUsed(TObjString)"))){
+ TObjString runUsed = Form("%d", GetCurrentRun());
+ metaData->SetProperty("RunUsed(TObjString)", runUsed.Clone());
+ }
- UInt_t result = 0;
+ Bool_t result = kFALSE;
- if (!(AliCDBManager::Instance()->GetStorage(mainUri))) {
- Log(fCurrentDetector, Form("Cannot activate main %s storage!", cdbType));
+ if (!(AliCDBManager::Instance()->GetStorage(localUri))) {
+ Log("SHUTTLE", Form("StoreLocally - Cannot activate local %s storage", cdbType));
} else {
- result = (UInt_t) AliCDBManager::Instance()->GetStorage(mainUri)
+ result = AliCDBManager::Instance()->GetStorage(localUri)
->Put(object, id, metaData);
}
if(!result) {
- Log(fCurrentDetector,
- Form("Problem with main %s storage. Object will go to local storage!", cdbType));
+ Log(fCurrentDetector, Form("StoreLocally - Can't store object <%s>!", id.ToString().Data()));
+ }
- result = AliCDBManager::Instance()->GetStorage(localUri)
- ->Put(object, id, metaData);
+ return result;
+}
- if(result) {
- result = 2;
- fGridError = kTRUE;
- }else{
- Log(fCurrentDetector, "Can't store data!");
- }
+//______________________________________________________________________________________________
+Bool_t AliShuttle::StoreOCDB()
+{
+ //
+ // Called when preprocessor ends successfully or when previous storage attempt failed (kStoreError status)
+ // Calls underlying StoreOCDB(const char*) function twice, for OCDB and Reference storage.
+ // 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");
+ Log(fCurrentDetector, "StoreOCDB - In TESTMODE - Simulating error while storing in the Grid");
+ return kFALSE;
+ }
+
+ Log("SHUTTLE","StoreOCDB - Storing OCDB data ...");
+ Int_t resultCDB = StoreOCDB(fgkMainCDB);
+
+ 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("StoreOCDB - SHUTTLE","Storing Run Metadata file ...");
+ resultMetadata = CopyFilesToGrid("metadata");
+ }
+
+ Int_t storeResult = 0;
+
+ 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 result;
+ return (storeResult == 0);
}
//______________________________________________________________________________________________
-AliShuttleStatus* AliShuttle::ReadShuttleStatus()
+Int_t AliShuttle::StoreOCDB(const TString& gridURI)
{
- // Reads the AliShuttleStatus from the CDB
-
- if (fStatusEntry)
- {
- delete fStatusEntry;
- fStatusEntry = 0;
- }
+ //
+ // 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)
+ //
- fStatusEntry = AliCDBManager::Instance()->GetStorage(AliShuttle::GetLocalCDB())
- ->Get(Form("/SHUTTLE/STATUS/%s", fCurrentDetector.Data()), fCurrentRun);
+ TObjArray* gridIds=0;
- if (!fStatusEntry)
- return 0;
+ Bool_t result = kTRUE;
+ Bool_t delayed = kFALSE;
- TObject* anObject = fStatusEntry->GetObject();
- if (anObject == NULL || anObject->IsA() != AliShuttleStatus::Class())
- {
- AliError("Invalid object stored to CDB!");
- return 0;
- }
+ const char* type = 0;
+ TString localURI;
+ if(gridURI == fgkMainCDB) {
+ type = "OCDB";
+ localURI = fgkLocalCDB;
+ } else if(gridURI == fgkMainRefStorage) {
+ type = "reference";
+ localURI = fgkLocalRefStorage;
+ } else {
+ AliError(Form("Invalid storage URI: %s", gridURI.Data()));
+ return -2;
+ }
- AliShuttleStatus* status = dynamic_cast<AliShuttleStatus*> (anObject);
- return status;
-}
+ AliCDBManager* man = AliCDBManager::Instance();
-//______________________________________________________________________________________________
-Bool_t AliShuttle::WriteShuttleStatus(AliShuttleStatus* status)
-{
- // writes the status for one subdetector
+ AliCDBStorage *gridSto = man->GetStorage(gridURI);
+ if(!gridSto) {
+ Log("SHUTTLE",
+ Form("StoreOCDB - cannot activate main %s storage", type));
+ return -2;
+ }
- if (fStatusEntry)
- {
- delete fStatusEntry;
- fStatusEntry = 0;
- }
+ gridIds = gridSto->GetQueryCDBList();
- AliCDBId id(AliCDBPath("SHUTTLE", "STATUS", fCurrentDetector), fCurrentRun, fCurrentRun);
+ // get objects previously stored in local CDB
+ AliCDBStorage *localSto = man->GetStorage(localURI);
+ if(!localSto) {
+ Log("SHUTTLE",
+ Form("StoreOCDB - cannot activate local %s storage", type));
+ return -2;
+ }
+ AliCDBPath aPath(GetOfflineDetName(fCurrentDetector.Data()),"*","*");
+ // Local objects were stored with current run as Grid version!
+ TList* localEntries = localSto->GetAll(aPath.GetPath(), GetCurrentRun(), GetCurrentRun());
+ localEntries->SetOwner(1);
+
+ // loop on local stored objects
+ TIter localIter(localEntries);
+ AliCDBEntry *aLocEntry = 0;
+ while((aLocEntry = dynamic_cast<AliCDBEntry*> (localIter.Next()))){
+ aLocEntry->SetOwner(1);
+ AliCDBId aLocId = aLocEntry->GetId();
+ aLocEntry->SetVersion(-1);
+ aLocEntry->SetSubVersion(-1);
+
+ // If local object is valid up to infinity we store it only if it is
+ // the first unprocessed run!
+ if (aLocId.GetLastRun() == AliCDBRunRange::Infinity() &&
+ !fFirstUnprocessed[GetDetPos(fCurrentDetector)])
+ {
+ 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;
+ }
- fStatusEntry = new AliCDBEntry(status, id, new AliCDBMetaData);
+ // loop on Grid valid Id's
+ Bool_t store = kTRUE;
+ TIter gridIter(gridIds);
+ AliCDBId* aGridId = 0;
+ 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 we get here, it means there's already some more recent object stored on Grid!
+ store = kFALSE;
+ break;
+ }
- UInt_t result = AliCDBManager::Instance()->GetStorage(fgkLocalCDB)->Put(fStatusEntry);
+ // 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 {
+ 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));
+ }
+
+ // removing local filename...
+ TString filename;
+ localSto->IdToFilename(aLocId, filename);
+ 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()));
+ Log(fCurrentDetector.Data(),
+ Form("StoreOCDB - Grid %s storage of object <%s> failed",
+ type, aLocId.ToString().Data()));
+ result = kFALSE;
+ }
+ }
+ localEntries->Clear();
- if (!result)
- {
- AliError(Form("WriteShuttleStatus for %s, run %d failed", fCurrentDetector.Data(), fCurrentRun));
- return kFALSE;
- }
+ Int_t returnCode = 0;
+
+ if (result == kFALSE)
+ returnCode = -1;
+ else if (delayed != kFALSE)
+ returnCode = 1;
- return kTRUE;
+ 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;
}
//______________________________________________________________________________________________
-void AliShuttle::UpdateShuttleStatus(AliShuttleStatus::Status newStatus, Bool_t increaseCount)
+Bool_t AliShuttle::CleanReferenceStorage(const char* detector)
{
- // changes the AliShuttleStatus for the given detector and run to the given status
-
- if (!fStatusEntry)
- {
- AliError("UNEXPECTED: fStatusEntry empty");
- return;
- }
+ // 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();
- TObject* anObject = fStatusEntry->GetObject();
- AliShuttleStatus* status = dynamic_cast<AliShuttleStatus*> (anObject);
+ TString targetDir = GetRefFilePrefix(localBaseFolder.Data(), detector);
+
+ Log("SHUTTLE", Form("CleanReferenceStorage - Cleaning %s", targetDir.Data()));
- if (!status)
- {
- AliError("UNEXPECTED: status could not be read from current CDB entry");
- return;
- }
+ 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;
- Log("SHUTTLE", Form("%s: Changing state from %s to %s", fCurrentDetector.Data(),
- status->GetStatusName(), status->GetStatusName(newStatus)));
+ 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++;
+ }
+ }
- status->SetStatus(newStatus);
- if (increaseCount)
- status->IncreaseCount();
+ if(nDirs > 0)
+ Log("SHUTTLE", Form("CleanReferenceStorage - %d (over %d) reference files in folder %s were deleted.",
+ nDel, nDirs, targetDir.Data()));
- AliCDBManager::Instance()->GetStorage(fgkLocalCDB)->Put(fStatusEntry);
-}
+
+ delete dirList;
+ return success;
-//______________________________________________________________________________________________
-Bool_t AliShuttle::ContinueProcessing()
-{
- // this function reads the AliShuttleStatus information from CDB and
- // checks if the processing should be continued
- // if yes it returns kTRUE and updates the AliShuttleStatus with nextStatus
- AliShuttleStatus* status = ReadShuttleStatus();
- if (!status)
- {
- // first time
- Log("SHUTTLE", Form("%s: Processing first time.", fCurrentDetector.Data()));
- status = new AliShuttleStatus(AliShuttleStatus::kStarted);
- return WriteShuttleStatus(status);
- }
- if (status->GetStatus() == AliShuttleStatus::kDone)
- {
- Log("SHUTTLE", Form("%s already done for run %d", fCurrentDetector.Data(), fCurrentRun));
- return kFALSE;
- }
- if (status->GetStatus() == AliShuttleStatus::kFailed)
- {
- Log("SHUTTLE", Form("%s already in failed state for run %d", fCurrentDetector.Data(), fCurrentRun));
- return kFALSE;
- }
- // TODO what to do in case of storage error? currently it does not do anything
- if (status->GetStatus() == AliShuttleStatus::kStoreFailed)
+ Int_t result = gSystem->GetPathInfo(targetDir, 0, (Long64_t*) 0, 0, 0);
+ if (result == 0)
{
- Log("SHUTTLE", Form("%s: Grid storage failed at least once for run %d", fCurrentDetector.Data(), fCurrentRun));
- return kFALSE;
+ // 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;
+ }
}
- // if we get here, there is a restart
-
- // abort conditions
- if (status->GetStatus() == AliShuttleStatus::kPPStarted && status->GetCount() >= fConfig->GetMaxPPRetries() ||
- status->GetCount() >= fConfig->GetMaxRetries())
+ result = gSystem->mkdir(targetDir, kTRUE);
+ if (result != 0)
{
- Log("SHUTTLE", Form("%s, run %d failed too often, %d times, status %s. Skipping processing.",
- fCurrentDetector.Data(), fCurrentRun, status->GetCount(), status->GetStatusName()));
-
+ Log("SHUTTLE", Form("CleanReferenceStorage - Error creating base directory %s", targetDir.Data()));
return kFALSE;
}
-
- Log("SHUTTLE", Form("Restart of %s, run %d. Got stuck before in %s, count %d",
- fCurrentDetector.Data(), fCurrentRun, status->GetStatusName(), status->GetCount()));
-
- UpdateShuttleStatus(AliShuttleStatus::kStarted, kTRUE);
-
+
return kTRUE;
}
//______________________________________________________________________________________________
-Bool_t AliShuttle::Process(Int_t run, UInt_t startTime, UInt_t endTime)
+Bool_t AliShuttle::StoreReferenceFile(const char* detector, const char* localFile, const char* gridFileName)
{
//
- // Makes data retrieval for all detectors in the configuration.
- // run: is the run number used
- // startTime: is the run start time
- // endTime: is the run end time
- // Returns kFALSE in case of error occured and kTRUE otherwise
+ // Stores reference file directly (without opening it). This function stores the file locally.
//
-
- AliInfo(Form("\n\n \t\t\t^*^*^*^*^*^*^*^*^*^*^*^* run %d: START ^*^*^*^*^*^*^*^*^*^*^*^* \n", run));
-
- // Initialization
- Bool_t hasError = kFALSE;
- for(Int_t iSys=0;iSys<3;iSys++) fFESCalled[iSys]=kFALSE;
-
- fCurrentRun = run;
- fCurrentStartTime = startTime;
- fCurrentEndTime = endTime;
-
- // Loop on detectors in the configuration
- TIter iter(fConfig->GetDetectors());
- TObjString* aDetector;
-
- while ((aDetector = (TObjString*) iter.Next())) {
- fCurrentDetector = aDetector->String();
-
- Bool_t detectorError=kFALSE;
- if (!fConfig->HostProcessDetector(fCurrentDetector)) continue;
-
- if (ContinueProcessing() == kFALSE) continue;
-
- AliInfo(Form("\n\n \t\t\t****** %s: START ******", aDetector->GetName()));
-
- if(!Process()) {
- hasError = kTRUE;
- detectorError=kTRUE;
- continue;
- }
- AliInfo(Form("\n \t\t\t****** %s: FINISH ****** \n\n", aDetector->GetName()));
-
- // Process successful: Update time_processed field in FES logbooks!
- if(fFESCalled[kDAQ]) {
- hasError = (UpdateDAQTable() == kFALSE);
- fFESlist[kDAQ].Clear();
- }
- //if(fFESCalled[kDCS]) {
- // hasError = UpdateDCSTable(aDetector->GetName());
- // fFESlist[kDCS].Clear();
- //}
- //if(fFESCalled[kHLT]) {
- // hasError = UpdateHLTTable(aDetector->GetName());
- // fFESlist[kHLT].Clear();
- //}
-
- // UpdateShuttleStatus(AliShuttleStatus::kDone);
+ // The file is stored under the following location:
+ // <base folder of local reference storage>/<DET>/<RUN#>_<gridFileName>
+ // where <gridFileName> is the second parameter given to the function
+ //
+
+ if (fTestMode & kErrorStorage)
+ {
+ Log(fCurrentDetector, "StoreReferenceFile - In TESTMODE - Simulating error while storing locally");
+ return kFALSE;
}
-
- fCurrentRun = -1;
- fCurrentStartTime = 0;
- fCurrentEndTime = 0;
-
- AliInfo(Form("\n\n \t\t\t^*^*^*^*^*^*^*^*^*^*^*^* run %d: FINISH ^*^*^*^*^*^*^*^*^*^*^*^* \n", run));
-
- return hasError == kFALSE;
+
+ AliCDBManager* man = AliCDBManager::Instance();
+ AliCDBStorage* sto = man->GetStorage(fgkLocalRefStorage);
+
+ TString localBaseFolder = sto->GetBaseFolder();
+
+ TString target = GetRefFilePrefix(localBaseFolder.Data(), detector);
+ target.Append(Form("/%d_%s", GetCurrentRun(), gridFileName));
+
+ return CopyFileLocally(localFile, target);
}
//______________________________________________________________________________________________
-Bool_t AliShuttle::Process()
+Bool_t AliShuttle::StoreRunMetadataFile(const char* localFile, const char* gridFileName)
{
//
- // Makes data retrieval just for one specific detector.
- // Threre should be a configuration for this detector.
- // run: is the run number used
- // startTime: is the run start time
- // endTime: is the run end time
- // detector: detector for which the retrieval will be made
- // Returns kFALSE in case of error occured and kTRUE otherwise
+ // Stores Run metadata file to the Grid, in the run folder
//
-
- AliInfo(Form("Retrieving values for %s, run %d", fCurrentDetector.Data(), fCurrentRun));
-
- if (!fConfig->HasDetector(fCurrentDetector)) {
- Log(fCurrentDetector, "There isn't any configuration for %s !");
- UpdateShuttleStatus(AliShuttleStatus::kFailed);
+ // Only GRP can call this function.
+
+ if (fTestMode & kErrorStorage)
+ {
+ Log(fCurrentDetector, "StoreRunMetaDataFile - In TESTMODE - Simulating error while storing locally");
return kFALSE;
}
-
- UpdateShuttleStatus(AliShuttleStatus::kDCSStarted);
-
- TString host(fConfig->GetDCSHost(fCurrentDetector));
- Int_t port = fConfig->GetDCSPort(fCurrentDetector);
-
- TIter iter(fConfig->GetDCSAliases(fCurrentDetector));
- TObjString* anAlias;
- TMap aliasMap;
-
- Bool_t aDCSError = kFALSE;
- fGridError = kFALSE;
- UInt_t aPPResult;
-
- while ((anAlias = (TObjString*) iter.Next())) {
- TObjArray valueSet;
- // TODO Test only... I've added a flag that allows to
- // exclude DCS archive DB query
- if(fgkProcessDCS){
- AliInfo("Querying DCS archive DB data...");
- aDCSError = (GetValueSet(host, port, anAlias->String(), valueSet) == 0);
- } else {
- AliInfo(Form("Skipping DCS processing. Port = %d",port));
- aDCSError = kFALSE;
- }
- if(!aDCSError) {
- aliasMap.Add(anAlias->Clone(), valueSet.Clone());
- }else{
- TString message = Form("Error while retrieving alias %s !",
- anAlias->GetName());
- Log(fCurrentDetector, message.Data());
- aDCSError = kTRUE;
- break;
- }
- }
-
- if (aDCSError)
+
+ 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)
{
- UpdateShuttleStatus(AliShuttleStatus::kDCSError);
- return kFALSE;
+ Log("SHUTTLE","StoreRunMetaDataFile - LHCPeriod not found in logbook!");
+ return 0;
}
-
- UpdateShuttleStatus(AliShuttleStatus::kPPStarted);
-
- AliPreprocessor* aPreprocessor =
- dynamic_cast<AliPreprocessor*> (fPreprocessorMap.GetValue(fCurrentDetector));
- if(aPreprocessor)
- {
- aPreprocessor->Initialize(fCurrentRun, fCurrentStartTime, fCurrentEndTime);
- aPPResult = aPreprocessor->Process(&aliasMap);
- }else{
- // TODO default behaviour?
- AliInfo(Form("No Preprocessor for %s: storing TMap of DP arrays into CDB!", fCurrentDetector.Data()));
- AliCDBMetaData metaData;
- AliDCSValue dcsValue(fCurrentStartTime, fCurrentEndTime);
- metaData.SetResponsible(Form("Duck, Donald"));
- metaData.SetProperty("StartEndTime", &dcsValue);
- metaData.SetComment("Automatically stored by Shuttle!");
- AliCDBPath path(fCurrentDetector,"DCS","Data");
- aPPResult = Store(path, &aliasMap, &metaData);
- }
-
- if (aPPResult == 0) { // Preprocessor error
- UpdateShuttleStatus(AliShuttleStatus::kPPError);
- } else if (fGridError == kFALSE) { // process and Grid storage ok!
- UpdateShuttleStatus(AliShuttleStatus::kDone);
- } else { // Grid storage error (process ok, but object put in local storage)
- UpdateShuttleStatus(AliShuttleStatus::kStoreFailed);
+
+ // 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()));
}
-
- aliasMap.Delete();
-
- return (aPPResult > 0);
+
+ 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::GetValueSet(const char* host, Int_t port, const char* alias,
- TObjArray& valueSet)
+Bool_t AliShuttle::CopyFileLocally(const char* localFile, const TString& target)
{
-// Retrieve all "alias" data points from the DCS server
-// host, port: TSocket connection parameters
-// alias: name of the alias
-// valueSet: array of retrieved AliDCSValue's
+ //
+ // 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 targetDir(target(0, target.Last('/')));
+
+ //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;
+ }
- AliDCSClient client(host, port, fTimeout, fRetries);
- if (!client.IsConnected()) {
+ } else {
+ gSystem->FreeDirectory(dir);
+ }
+
+ Int_t result = 0;
+
+ result = gSystem->GetPathInfo(localFile, 0, (Long64_t*) 0, 0, 0);
+ if (result)
+ {
+ Log("SHUTTLE", Form("CopyFileLocally - %s does not exist", localFile));
return kFALSE;
}
- Int_t result = client.GetAliasValues(alias,
- GetCurrentStartTime(), GetCurrentEndTime(), valueSet);
-
- if (result < 0) {
- AliError(Form("Can't get '%s'! Reason: %s",
- alias, AliDCSClient::GetErrorString(result)));
-
- if (result == AliDCSClient::fgkServerError) {
- AliError(Form("Server error: %s",
- client.GetServerError().Data()));
+ 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("CopyFileLocally - Could not remove existing target file %s!", target.Data()));
+ return kFALSE;
}
+ }
+
+ result = gSystem->CopyFile(localFile, target);
- return kFALSE;
+ if (result == 0)
+ {
+ Log("SHUTTLE", Form("CopyFileLocally - File %s stored locally to %s", localFile, target.Data()));
+ return kTRUE;
}
+ else
+ {
+ Log("SHUTTLE", Form("CopyFileLocally - Could not store file %s to %s! Error code = %d",
+ localFile, target.Data(), result));
+ return kFALSE;
+ }
+
+
- return kTRUE;
}
//______________________________________________________________________________________________
-const char* AliShuttle::GetFile(Int_t system, const char* detector,
- const char* id, const char* source)
+Bool_t AliShuttle::CopyFilesToGrid(const char* type)
{
-// Get calibration file from file exchange servers
-// calls specific getter according to system index (kDAQ, kDCS, kHLT)
-
- switch(system){
- case kDAQ:
- return GetDAQFileName(detector, id, source);
- break;
- case kDCS:
- return GetDCSFileName(detector, id, source);
- break;
- case kHLT:
- return GetHLTFileName(detector, id, source);
- break;
- default:
- AliError(Form("No valid system index: %d",system));
+ //
+ // Transfers local files to the Grid. Local files can be reference files
+ // or run metadata file (from GRP only).
+ //
+ // 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;
+ TString alienDir;
+ TString begin;
+
+ 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();
+
+ 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();
+ delete baseDir;
+
+ if (!dirList) return kTRUE;
+
+ if (dirList->GetEntries() < 3)
+ {
+ delete dirList;
+ return kTRUE;
+ }
+
+ 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;
+
+ while ((entry = dynamic_cast<TSystemFile*> (dirIter.Next())))
+ {
+ if (entry->IsDirectory())
+ continue;
+
+ TString fileName(entry->GetName());
+ if (!fileName.BeginsWith(begin))
+ continue;
+
+ nDirs++;
+
+ if (first)
+ {
+ first = kFALSE;
+ // check that folder exists, otherwise create it
+ TGridResult* result = gGrid->Ls(alienDir.Data(), "a");
+
+ if (!result)
+ {
+ delete dirList;
+ return kFALSE;
+ }
+
+ if (!result->GetFileName(1)) // TODO: It looks like element 0 is always 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("CopyFilesToGrid - Cannot create directory %s",
+ alienDir.Data()));
+ 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 fullLocalPath;
+ fullLocalPath.Form("%s/%s", dir.Data(), fileName.Data());
+
+ TString fullGridPath;
+ fullGridPath.Form("alien://%s/%s", alienDir.Data(), fileName.Data());
+
+ Bool_t result = TFile::Cp(fullLocalPath, fullGridPath);
+
+ if (result)
+ {
+ Log("SHUTTLE", Form("CopyFilesToGrid - Copying local file %s to %s succeeded!",
+ fullLocalPath.Data(), fullGridPath.Data()));
+ RemoveFile(fullLocalPath);
+ nTransfer++;
+ }
+ else
+ {
+ Log("SHUTTLE", Form("CopyFilesToGrid - Copying local file %s to %s FAILED!",
+ fullLocalPath.Data(), fullGridPath.Data()));
+ success = kFALSE;
+ }
+ }
+
+ Log("SHUTTLE", Form("CopyFilesToGrid - %d (over %d) files in folder %s copied to Grid.",
+ nTransfer, nDirs, dir.Data()));
+
+
+ delete dirList;
+ return success;
+}
+
+//______________________________________________________________________________________________
+const char* AliShuttle::GetRefFilePrefix(const char* base, const char* detector)
+{
+ //
+ // Get folder name of reference files
+ //
+
+ 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;
+ }
+
+ AliCDBManager* man = AliCDBManager::Instance();
+
+ // open local storage
+ AliCDBStorage *localSto = man->GetStorage(uri);
+ if(!localSto) {
+ Log("SHUTTLE",
+ Form("CleanLocalStorage - cannot activate local %s storage", type));
+ return;
+ }
+
+ TString filename(Form("%s/%s/*/Run*_v%d_s*.root",
+ localSto->GetBaseFolder().Data(), GetOfflineDetName(fCurrentDetector.Data()), GetCurrentRun()));
+
+ AliDebug(2, Form("filename = %s", filename.Data()));
+
+ Log("SHUTTLE", Form("Removing remaining local files for run %d and detector %s ...",
+ GetCurrentRun(), fCurrentDetector.Data()));
+
+ RemoveFile(filename.Data());
+
+}
+
+//______________________________________________________________________________________________
+void AliShuttle::RemoveFile(const char* filename)
+{
+ //
+ // removes local file
+ //
+
+ TString command(Form("rm -f %s", filename));
+
+ Int_t result = gSystem->Exec(command.Data());
+ if(result != 0)
+ {
+ Log("SHUTTLE", Form("RemoveFile - %s: Cannot remove file %s!",
+ fCurrentDetector.Data(), filename));
+ }
+}
+
+//______________________________________________________________________________________________
+AliShuttleStatus* AliShuttle::ReadShuttleStatus()
+{
+ //
+ // Reads the AliShuttleStatus from the CDB
+ //
+
+ if (fStatusEntry){
+ delete fStatusEntry;
+ fStatusEntry = 0;
+ }
+
+ fStatusEntry = AliCDBManager::Instance()->GetStorage(GetLocalCDB())
+ ->Get(Form("/SHUTTLE/STATUS/%s", fCurrentDetector.Data()), GetCurrentRun());
+
+ if (!fStatusEntry) return 0;
+ fStatusEntry->SetOwner(1);
+
+ AliShuttleStatus* status = dynamic_cast<AliShuttleStatus*> (fStatusEntry->GetObject());
+ if (!status) {
+ AliError("Invalid object stored to CDB!");
+ return 0;
+ }
+
+ return status;
+}
+
+//______________________________________________________________________________________________
+Bool_t AliShuttle::WriteShuttleStatus(AliShuttleStatus* status)
+{
+ //
+ // writes the status for one subdetector
+ //
+
+ if (fStatusEntry){
+ delete fStatusEntry;
+ fStatusEntry = 0;
+ }
+
+ Int_t run = GetCurrentRun();
+
+ AliCDBId id(AliCDBPath("SHUTTLE", "STATUS", fCurrentDetector), run, run);
+
+ fStatusEntry = new AliCDBEntry(status, id, new AliCDBMetaData);
+ fStatusEntry->SetOwner(1);
+
+ UInt_t result = AliCDBManager::Instance()->GetStorage(fgkLocalCDB)->Put(fStatusEntry);
+
+ if (!result) {
+ Log("SHUTTLE", Form("WriteShuttleStatus - Failed for %s, run %d",
+ fCurrentDetector.Data(), run));
+ return kFALSE;
+ }
+
+ SendMLInfo();
+
+ return kTRUE;
+}
+
+//______________________________________________________________________________________________
+void AliShuttle::UpdateShuttleStatus(AliShuttleStatus::Status newStatus, Bool_t increaseCount)
+{
+ //
+ // changes the AliShuttleStatus for the given detector and run to the given status
+ //
+
+ if (!fStatusEntry){
+ AliError("UNEXPECTED: fStatusEntry empty");
+ return;
+ }
+
+ AliShuttleStatus* status = dynamic_cast<AliShuttleStatus*> (fStatusEntry->GetObject());
+
+ if (!status){
+ Log("SHUTTLE", "UpdateShuttleStatus - UNEXPECTED: status could not be read from current CDB entry");
+ return;
+ }
+
+ TString actionStr = Form("UpdateShuttleStatus - %s: Changing state from %s to %s",
+ fCurrentDetector.Data(),
+ status->GetStatusName(),
+ status->GetStatusName(newStatus));
+ Log("SHUTTLE", actionStr);
+ SetLastAction(actionStr);
+
+ status->SetStatus(newStatus);
+ if (increaseCount) status->IncreaseCount();
+
+ AliCDBManager::Instance()->GetStorage(fgkLocalCDB)->Put(fStatusEntry);
+
+ SendMLInfo();
+}
+
+//______________________________________________________________________________________________
+void AliShuttle::SendMLInfo()
+{
+ //
+ // 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");
+ return;
+ }
+
+ TMonaLisaText mlStatus(Form("%s_status", fCurrentDetector.Data()), status->GetStatusName());
+ TMonaLisaValue mlRetryCount(Form("%s_count", fCurrentDetector.Data()), status->GetCount());
+
+ TList mlList;
+ mlList.Add(&mlStatus);
+ mlList.Add(&mlRetryCount);
+
+ TString mlID;
+ mlID.Form("%d", GetCurrentRun());
+ fMonaLisa->SendParameters(&mlList, mlID);
+}
+
+//______________________________________________________________________________________________
+Bool_t AliShuttle::ContinueProcessing()
+{
+ // this function reads the AliShuttleStatus information from CDB and
+ // checks if the processing should be continued
+ // if yes it returns kTRUE and updates the AliShuttleStatus with nextStatus
+
+ if (!fConfig->HostProcessDetector(fCurrentDetector))
+ return kFALSE;
+
+ AliPreprocessor* aPreprocessor =
+ dynamic_cast<AliPreprocessor*> (fPreprocessorMap.GetValue(fCurrentDetector));
+ if (!aPreprocessor)
+ {
+ Log("SHUTTLE", Form("ContinueProcessing - %s: no preprocessor registered", fCurrentDetector.Data()));
+ return kFALSE;
+ }
+
+ AliShuttleLogbookEntry::Status entryStatus =
+ fLogbookEntry->GetDetectorStatus(fCurrentDetector);
+
+ if (entryStatus != AliShuttleLogbookEntry::kUnprocessed) {
+ Log("SHUTTLE", Form("ContinueProcessing - %s is %s",
+ fCurrentDetector.Data(),
+ fLogbookEntry->GetDetectorStatusName(entryStatus)));
+ return kFALSE;
+ }
+
+ // if we get here, according to Shuttle logbook subdetector is in UNPROCESSED state
+
+ // check if current run is first unprocessed run for current detector
+ if (fConfig->StrictRunOrder(fCurrentDetector) &&
+ !fFirstUnprocessed[GetDetPos(fCurrentDetector)])
+ {
+ 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 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::kSkipped) {
+ Log("SHUTTLE", Form("ContinueProcessing - %s is already %s. Updating Shuttle Logbook",
+ 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::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()));
+ StoreOCDB();
+ return kFALSE;
+ }
+
+ // if we get here, there is a restart
+ Bool_t cont = kFALSE;
+
+ // abort conditions
+ if (status->GetCount() >= fConfig->GetMaxRetries()) {
+ Log("SHUTTLE", Form("ContinueProcessing - %s failed %d times in status %s - "
+ "Updating Shuttle Logbook", fCurrentDetector.Data(),
+ status->GetCount(), status->GetStatusName()));
+ UpdateShuttleLogbook(fCurrentDetector.Data(), "FAILED");
+ UpdateShuttleStatus(AliShuttleStatus::kFailed);
+
+ // there may still be objects in local OCDB and reference storage
+ // and FXS databases may be not updated: do it now!
+
+ // TODO Currently disabled, we want to keep files in case of failure!
+ // CleanLocalStorage(fgkLocalCDB);
+ // CleanLocalStorage(fgkLocalRefStorage);
+ // UpdateTableFailCase();
+
+ // Send mail to detector expert!
+ 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()));
+
+ } 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 ||
+ status->GetStatus() == AliShuttleStatus::kFXSError)
+ increaseCount = kFALSE;
+
+ UpdateShuttleStatus(AliShuttleStatus::kStarted, increaseCount);
+ cont = kTRUE;
+ }
+
+ return cont;
+}
+
+//______________________________________________________________________________________________
+Bool_t AliShuttle::Process(AliShuttleLogbookEntry* entry)
+{
+ //
+ // Makes data retrieval for all detectors in the configuration.
+ // entry: Shuttle logbook entry, contains run paramenters and status of detectors
+ // (Unprocessed, Inactive, Failed or Done).
+ // Returns kFALSE in case of error occured and kTRUE otherwise
+ //
+
+ if (!entry) return kFALSE;
+
+ fLogbookEntry = entry;
+
+ Log("SHUTTLE", Form("\t\t\t^*^*^*^*^*^*^*^*^*^*^*^* run %d: START ^*^*^*^*^*^*^*^*^*^*^*^*",
+ GetCurrentRun()));
+
+ // Send the information to ML
+ CountOpenRuns();
+
+ TMonaLisaText mlStatus("SHUTTLE_status", "Processing");
+ TString runType(entry->GetRunType());
+ if (strlen(entry->GetRunParameter("log")) > 0){
+
+ runType += "(";
+ runType += entry->GetRunParameter("log");
+ runType += ")";
+ }
+ TMonaLisaText mlRunType("SHUTTLE_runtype", runType);
+
+ TList mlList;
+ mlList.Add(&mlStatus);
+ mlList.Add(&mlRunType);
+
+ TString mlID;
+ mlID.Form("%d", GetCurrentRun());
+ fMonaLisa->SendParameters(&mlList, mlID);
+
+ if (fLogbookEntry->IsDone())
+ {
+ Log("SHUTTLE","Process - Shuttle is already DONE. Updating logbook");
+ UpdateShuttleLogbook("shuttle_done");
+ fLogbookEntry = 0;
+ return kTRUE;
+ }
+
+ // read test mode if flag is set
+ if (fReadTestMode)
+ {
+ fTestMode = kNone;
+ TString logEntry(entry->GetRunParameter("log"));
+ //printf("log entry = %s\n", logEntry.Data());
+ TString searchStr("Testmode: ");
+ Int_t pos = logEntry.Index(searchStr.Data());
+ //printf("%d\n", pos);
+ if (pos >= 0)
+ {
+ TSubString subStr = logEntry(pos + searchStr.Length(), logEntry.Length());
+ //printf("%s\n", subStr.String().Data());
+ TString newStr(subStr.Data());
+ TObjArray* token = newStr.Tokenize(' ');
+ if (token)
+ {
+ //token->Print();
+ TObjString* tmpStr = dynamic_cast<TObjString*> (token->First());
+ if (tmpStr)
+ {
+ Int_t testMode = tmpStr->String().Atoi();
+ if (testMode > 0)
+ {
+ Log("SHUTTLE", Form("Process - Enabling test mode %d", testMode));
+ SetTestMode((TestMode) testMode);
+ }
+ }
+ delete token;
+ }
+ }
+ }
+
+ fLogbookEntry->Print("all");
+
+ // Initialization
+ Bool_t hasError = kFALSE;
+
+ // Set the CDB and Reference folders according to the year and LHC period
+ TString lhcPeriod(GetLHCPeriod());
+ if (lhcPeriod.Length() == 0)
+ {
+ Log("SHUTTLE","Process - LHCPeriod not found in logbook!");
+ return 0;
+ }
+
+ if (fgkMainCDB.Length() == 0)
+ fgkMainCDB = Form("alien://folder=%s%d/%s/OCDB?user=alidaq?cacheFold=/tmp/OCDBCache",
+ fConfig->GetAlienPath(), GetCurrentYear(), lhcPeriod.Data());
+
+ if (fgkMainRefStorage.Length() == 0)
+ fgkMainRefStorage = Form("alien://folder=%s%d/%s/Reference?user=alidaq?cacheFold=/tmp/OCDBCache",
+ fConfig->GetAlienPath(), GetCurrentYear(), lhcPeriod.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 (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;
+ }
+
+ Log("SHUTTLE", Form("\t\t\t****** run %d - %s: START ******",
+ GetCurrentRun(), aDetector->GetName()));
+
+ for(Int_t iSys=0;iSys<3;iSys++) fFXSCalled[iSys]=kFALSE;
+
+ Log(fCurrentDetector.Data(), "Process - Starting processing");
+
+ Int_t pid = fork();
+
+ if (pid < 0)
+ {
+ Log("SHUTTLE", "Process - ERROR: Forking failed");
+ }
+ else if (pid > 0)
+ {
+ // parent
+ Log("SHUTTLE", Form("Process - In parent process of %d - %s: Starting monitoring",
+ GetCurrentRun(), aDetector->GetName()));
+
+ Long_t begin = time(0);
+
+ int status; // to be used with waitpid, on purpose an int (not Int_t)!
+ while (waitpid(pid, &status, WNOHANG) == 0)
+ {
+ Long_t expiredTime = time(0) - begin;
+
+ if (expiredTime > fConfig->GetPPTimeOut())
+ {
+ TString tmp;
+ tmp.Form("Process - Process of %s time out. "
+ "Run time: %d seconds. Killing...",
+ fCurrentDetector.Data(), expiredTime);
+ Log("SHUTTLE", tmp);
+ Log(fCurrentDetector, tmp);
+
+ kill(pid, 9);
+
+ UpdateShuttleStatus(AliShuttleStatus::kPPTimeOut);
+ hasError = kTRUE;
+
+ gSystem->Sleep(1000);
+ }
+ else
+ {
+ 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("Process - Error: "
+ "Could not open pipe to %s", checkStr.Data()));
+ continue;
+ }
+
+ char buffer[100];
+ if (!fgets(buffer, 100, pipe))
+ {
+ Log("SHUTTLE", "Process - Error: ps did not return anything");
+ gSystem->ClosePipe(pipe);
+ continue;
+ }
+ 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");
+ continue;
+ }
+
+ if (expiredTime % 60 == 0)
+ {
+ Log("SHUTTLE", Form("Process - %s: Checking process. "
+ "Run time: %d seconds - Memory consumption: %d KB",
+ fCurrentDetector.Data(), expiredTime, mem));
+ SendAlive();
+ }
+
+ if (mem > fConfig->GetPPMaxMem())
+ {
+ TString tmp;
+ tmp.Form("Process - Process exceeds maximum allowed memory "
+ "(%d KB > %d KB). Killing...",
+ mem, fConfig->GetPPMaxMem());
+ Log("SHUTTLE", tmp);
+ Log(fCurrentDetector, tmp);
+
+ kill(pid, 9);
+
+ UpdateShuttleStatus(AliShuttleStatus::kPPOutOfMemory);
+ hasError = kTRUE;
+
+ gSystem->Sleep(1000);
+ }
+ }
+ }
+
+ 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("Process - %s: the return code is %d", fCurrentDetector.Data(),
+ returnCode));
+
+ if (returnCode == 0) hasError = kTRUE;
+ }
+ }
+ else if (pid == 0)
+ {
+ // child
+ Log("SHUTTLE", Form("Process - In child process of %d - %s", GetCurrentRun(),
+ aDetector->GetName()));
+
+ Log("SHUTTLE", Form("Process - Redirecting output to %s log",fCurrentDetector.Data()));
+
+ if ((freopen(GetLogFileName(fCurrentDetector), "a", stdout)) == 0)
+ {
+ Log("SHUTTLE", "Process - Could not freopen stdout");
+ }
+ else
+ {
+ fOutputRedirected = kTRUE;
+ if ((dup2(fileno(stdout), fileno(stderr))) < 0)
+ Log("SHUTTLE", "Process - Could not redirect stderr");
+
+ }
+
+ TString wd = gSystem->WorkingDirectory();
+ TString tmpDir = Form("%s/%s_%d_process", GetShuttleTempDir(),
+ fCurrentDetector.Data(), GetCurrentRun());
+
+ 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!",
+ fCurrentDetector.Data()));
+
+ // Transfer the data from local storage to main storage (Grid)
+ if (StoreOCDB() == kFALSE)
+ success = kFALSE;
+ }
+ 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();
+ }
+
+ Log("SHUTTLE", Form("Process - Client process of %d - %s is exiting now with %d.",
+ GetCurrentRun(), aDetector->GetName(), success));
+
+ // the client exits here
+ gSystem->Exit(success);
+
+ AliError("We should never get here!!!");
+ }
+ }
+
+ 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) {
+ Log("SHUTTLE", Form("Process - Warning: Cannot check status of run %d on Shuttle logbook!",
+ GetCurrentRun()));
+ return hasError == kFALSE;
+ }
+
+ AliShuttleLogbookEntry* checkEntry = dynamic_cast<AliShuttleLogbookEntry*>
+ (checkEntryArray.At(0));
+
+ if (checkEntry)
+ {
+ if (checkEntry->IsDone())
+ {
+ Log("SHUTTLE","Process - Shuttle is DONE. Updating logbook");
+ UpdateShuttleLogbook("shuttle_done");
+ }
+ else
+ {
+ for (UInt_t iDet=0; iDet<NDetectors(); iDet++)
+ {
+ if (checkEntry->GetDetectorStatus(iDet) == AliShuttleLogbookEntry::kUnprocessed)
+ {
+ AliDebug(2, Form("Run %d: setting %s as \"not first time unprocessed\"",
+ checkEntry->GetRun(), GetDetName(iDet)));
+ fFirstUnprocessed[iDet] = kFALSE;
+ }
+ }
+ TMonaLisaText mlStatusPending("SHUTTLE_status", "Pending");
+ mlList.Clear();
+ mlList.Add(&mlStatusPending);
+ fMonaLisa->SendParameters(&mlList, mlID);
+ }
+ }
+
+ fLogbookEntry = 0;
+
+ return hasError == kFALSE;
+}
+
+//______________________________________________________________________________________________
+Int_t AliShuttle::ProcessCurrentDetector()
+{
+ //
+ // Makes data retrieval just for a specific detector (fCurrentDetector).
+ // Threre should be a configuration for this detector.
+
+ 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");
+ 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)
+ {
+ Log(fCurrentDetector, "ProcessCurrentDetector -"
+ " The preprocessor requested to skip the retrieval of DCS values");
+ }
+ else if (fTestMode & kSkipDCS)
+ {
+ Log(fCurrentDetector, "ProcessCurrentDetector - In TESTMODE: Skipping DCS processing");
+ }
+ else if (fTestMode & kErrorDCS)
+ {
+ Log(fCurrentDetector, "ProcessCurrentDetector - In TESTMODE: Simulating DCS error");
+ UpdateShuttleStatus(AliShuttleStatus::kDCSStarted);
+ UpdateShuttleStatus(AliShuttleStatus::kDCSError);
+ delete dcsMap;
+ return 0;
+ } else {
+
+ UpdateShuttleStatus(AliShuttleStatus::kDCSStarted);
+
+ // Query DCS archive
+ Int_t nServers = fConfig->GetNServers(fCurrentDetector);
+
+ for (int iServ=0; iServ<nServers; iServ++)
+ {
+
+ TString host(fConfig->GetDCSHost(fCurrentDetector, iServ));
+ Int_t port = fConfig->GetDCSPort(fCurrentDetector, iServ);
+ Int_t multiSplit = fConfig->GetMultiSplit(fCurrentDetector, iServ);
+
+ 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 (fConfig->GetDCSAliases(fCurrentDetector, iServ)->GetEntries() > 0)
+ {
+ 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!"));
+
+ delete dcsMap;
+ return 0;
+ }
+ }
+
+ if (fConfig->GetDCSDataPoints(fCurrentDetector, iServ)->GetEntries() > 0)
+ {
+ 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);
+
+ 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("ProcessCurrentDetector - "
+ "Preprocessor failed. Process returned %d.", returnValue));
+ UpdateShuttleStatus(AliShuttleStatus::kPPError);
+ dcsMap->DeleteAll();
+ delete dcsMap;
+ return 0;
+ }
+
+ // preprocessor ok!
+ UpdateShuttleStatus(AliShuttleStatus::kPPDone);
+ Log(fCurrentDetector, Form("ProcessCurrentDetector - %s preprocessor returned success",
+ fCurrentDetector.Data()));
+
+ dcsMap->DeleteAll();
+ delete dcsMap;
+
+ return 1;
+}
+
+//______________________________________________________________________________________________
+void AliShuttle::CountOpenRuns()
+{
+ // Query DAQ's Shuttle logbook and sends the number of open runs to ML
+
+ // 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__");
+}
+
+//______________________________________________________________________________________________
+Bool_t AliShuttle::QueryShuttleLogbook(const char* whereClause,
+ TObjArray& entries)
+{
+ // Query DAQ's Shuttle logbook and fills detector status object.
+ // Call QueryRunParameters to query DAQ logbook for run parameters.
+ //
+
+ entries.SetOwner(1);
+
+ // check connection, in case connect
+ if (!Connect(3)) return kFALSE;
+
+ TString sqlQuery;
+ sqlQuery = Form("select * from %s %s order by run", fConfig->GetShuttlelbTable(), whereClause);
+
+ TSQLResult* aResult = fServer[3]->Query(sqlQuery);
+ if (!aResult) {
+ AliError(Form("Can't execute query <%s>!", sqlQuery.Data()));
+ return kFALSE;
+ }
+
+ AliDebug(2,Form("Query = %s", sqlQuery.Data()));
+
+ if(aResult->GetRowCount() == 0) {
+ Log("SHUTTLE", "No entries in Shuttle Logbook match request");
+ delete aResult;
+ return kTRUE;
+ }
+
+ // TODO Check field count!
+ const UInt_t nCols = 23;
+ if (aResult->GetFieldCount() != (Int_t) nCols) {
+ Log("SHUTTLE", "Invalid SQL result field number!");
+ delete aResult;
+ return kFALSE;
+ }
+
+ TSQLRow* aRow;
+ while ((aRow = aResult->Next())) {
+ TString runString(aRow->GetField(0), aRow->GetFieldLength(0));
+ Int_t run = runString.Atoi();
+
+ AliShuttleLogbookEntry *entry = QueryRunParameters(run);
+ if (!entry)
+ continue;
+
+ // loop on detectors
+ for(UInt_t ii = 0; ii < nCols; ii++)
+ entry->SetDetectorStatus(aResult->GetFieldName(ii), aRow->GetField(ii));
+
+ entries.AddLast(entry);
+ delete aRow;
+ }
+
+ delete aResult;
+ return kTRUE;
+}
+
+//______________________________________________________________________________________________
+AliShuttleLogbookEntry* AliShuttle::QueryRunParameters(Int_t run)
+{
+ //
+ // Retrieve run parameters written in the DAQ logbook and sets them into AliShuttleLogbookEntry object
+ //
+
+ // check connection, in case connect
+ if (!Connect(3))
+ return 0;
+
+ TString sqlQuery;
+ sqlQuery.Form("select * from %s where run=%d", fConfig->GetDAQlbTable(), run);
+
+ TSQLResult* aResult = fServer[3]->Query(sqlQuery);
+ if (!aResult) {
+ Log("SHUTTLE", Form("Can't execute query <%s>!", sqlQuery.Data()));
+ return 0;
+ }
+
+ if (aResult->GetRowCount() == 0) {
+ Log("SHUTTLE", Form("QueryRunParameters - No entry in DAQ Logbook for run %d. Skipping", run));
+ delete aResult;
+ return 0;
+ }
+
+ if (aResult->GetRowCount() > 1) {
+ 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)
+ {
+ Log("SHUTTLE", Form("QueryRunParameters - Could not retrieve row for run %d. Skipping", run));
+ delete aResult;
+ return 0;
+ }
+
+ AliShuttleLogbookEntry* entry = new AliShuttleLogbookEntry(run);
+
+ 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();
+ Bool_t ecsSuccess = entry->GetECSSuccess();
+
+ TString totEventsStr = entry->GetRunParameter("totalEvents");
+ Int_t totEvents = totEventsStr.Atoi();
+
+ if (startTime != 0 && endTime != 0 && endTime > startTime && totEvents > 0 && ecsSuccess)
+ return entry;
+
+ if (ecsSuccess == kFALSE)
+ {
+ Log("SHUTTLE", Form("Skipped run %d due to ECS failure, Reason: %s", run, entry->GetRunParameter("eor_reason")));
+ }
+ else if (totEvents < 1)
+ {
+ Log("SHUTTLE", Form("QueryRunParameters - Run %d has 0 events - Skipping!", run));
+ }
+ 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));
+ }
+
+ //Log("SHUTTLE", Form("Marking SHUTTLE done for run %d", run));
+ //fLogbookEntry = entry;
+ //if (!UpdateShuttleLogbook("shuttle_done"))
+ //{
+ // AliError(Form("Could not update logbook for run %d !", run));
+ //}
+ //fLogbookEntry = 0;
+
+ delete entry;
+ return 0;
+}
+
+//______________________________________________________________________________________________
+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
+ // 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);
+
+ TMap* result = 0;
+ if (type == kAlias)
+ {
+ result = client.GetAliasValues(entries, GetCurrentStartTime(),
+ GetCurrentEndTime());
+ }
+ else if (type == kDP)
+ {
+ result = client.GetDPValues(entries, GetCurrentStartTime(),
+ GetCurrentEndTime());
+ }
+
+ if (result == 0)
+ {
+ 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 0;
+ }
+
+ return result;
+}
+
+//______________________________________________________________________________________________
+const char* AliShuttle::GetFile(Int_t system, const char* detector,
+ const char* id, const char* source)
+{
+ // Get calibration file from file exchange servers
+ // First queris the FXS database for the file name, using the run, detector, id and source info
+ // then calls RetrieveFile(filename) for actual copy to local disk
+ // run: current run being processed (given by Logbook entry fLogbookEntry)
+ // detector: the Preprocessor name
+ // id: provided as a parameter by the Preprocessor
+ // source: provided by the Preprocessor through GetFileSources function
+
+ // check if test mode should simulate a FXS error
+ if (fTestMode & kErrorFXSFiles)
+ {
+ Log(detector, Form("GetFile - In TESTMODE - Simulating error while connecting to %s FXS", GetSystemName(system)));
+ return 0;
+ }
+
+ // check connection, in case connect
+ if (!Connect(system))
+ {
+ Log(detector, Form("GetFile - Couldn't connect to %s FXS database", GetSystemName(system)));
+ fFXSError = system;
+ return 0;
+ }
+
+ // Query preparation
+ TString sourceName(source);
+ Int_t nFields = 3;
+ TString sqlQueryStart = Form("select filePath,size,fileChecksum from %s where",
+ fConfig->GetFXSdbTable(system));
+ TString whereClause = Form("run=%d and detector=\"%s\" and fileId=\"%s\"",
+ GetCurrentRun(), detector, id);
+
+ if (system == kDAQ)
+ {
+ whereClause += Form(" and DAQsource=\"%s\"", source);
+ }
+ else if (system == kDCS)
+ {
+ sourceName="none";
+ }
+ else if (system == kHLT)
+ {
+ whereClause += Form(" and DDLnumbers=\"%s\"", source);
+ }
+
+ TString sqlQuery = Form("%s %s", sqlQueryStart.Data(), whereClause.Data());
+
+ AliDebug(2, Form("SQL query: \n%s",sqlQuery.Data()));
+
+ // Query execution
+ TSQLResult* aResult = 0;
+ aResult = dynamic_cast<TSQLResult*> (fServer[system]->Query(sqlQuery));
+ if (!aResult) {
+ 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)
+ {
+ Log(detector,
+ 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("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;
+ }
+
+ if (aResult->GetFieldCount() != nFields) {
+ 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("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;
+ }
+
+ TString filePath(aRow->GetField(0), aRow->GetFieldLength(0));
+ TString fileSize(aRow->GetField(1), aRow->GetFieldLength(1));
+ TString fileChecksum(aRow->GetField(2), aRow->GetFieldLength(2));
+
+ delete aResult;
+ delete aRow;
+
+ AliDebug(2, Form("filePath = %s; size = %s, fileChecksum = %s",
+ filePath.Data(), fileSize.Data(), fileChecksum.Data()));
+
+ // retrieved file is renamed to make it unique
+ TString localFileName = Form("%s/%s_%d_process/%s_%s_%d_%s_%s.shuttle",
+ GetShuttleTempDir(), detector, GetCurrentRun(),
+ GetSystemName(system), detector, GetCurrentRun(),
+ id, sourceName.Data());
+
+
+ // file retrieval from FXS
+ UInt_t nRetries = 0;
+ UInt_t maxRetries = 3;
+ Bool_t result = kFALSE;
+
+ // copy!! if successful TSystem::Exec returns 0
+ while (nRetries++ < maxRetries) {
+ AliDebug(2, Form("Trying to copy file. Retry # %d", nRetries));
+ result = RetrieveFile(system, filePath.Data(), localFileName.Data());
+ if (!result)
+ {
+ 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 {
+ 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
+ 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("GetFile - md5sum of file %s does not match with local copy!",
+ filePath.Data()));
+ result = kFALSE;
+ continue;
+ }
+ } else {
+ Log(fCurrentDetector, Form("GetFile - md5sum of file %s not set in %s database, skipping comparison",
+ filePath.Data(), GetSystemName(system)));
+ }
+ if (result) break;
+ }
+
+ 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 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();
+}
+
+//______________________________________________________________________________________________
+Bool_t AliShuttle::RetrieveFile(UInt_t system, const char* fxsFileName, const char* localFileName)
+{
+ //
+ // Copies file from FXS to local Shuttle machine
+ //
+
+ // 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",
+ GetSystemName(system), fxsFileName, localFileName));
+
+ TString tmpDir(localFileName);
+
+ tmpDir = tmpDir(0,tmpDir.Last('/'));
+
+ 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 {
+ 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;
+ }
+
+ if (system == kDCS)
+ {
+ 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;
+ }
+
+ // check connection, in case connect
+ if (!Connect(system))
+ {
+ Log(detector, Form("GetFileSources - Couldn't connect to %s FXS database", GetSystemName(system)));
+ fFXSError = system;
+ return NULL;
+ }
+
+ TString sourceName = 0;
+ if (system == kDAQ)
+ {
+ sourceName = "DAQsource";
+ } else if (system == kHLT)
+ {
+ sourceName = "DDLnumbers";
}
- return 0;
+ 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());
+
+ AliDebug(2, Form("SQL query: \n%s",sqlQuery.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;
+ }
+
+ 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 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
-// calls specific getter according to system index (kDAQ, kDCS, kHLT)
+ //
+ // 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)));
- switch(system){
- case kDAQ:
- return GetDAQFileSources(detector, id);
- break;
- case kDCS:
- return GetDCSFileSources(detector, id);
- break;
- case kHLT:
- return GetHLTFileSources(detector, id);
- break;
- default:
- AliError(Form("No valid system index: %d",system));
+ // check if test mode should simulate a FXS error
+ if (fTestMode & kErrorFXSSources)
+ {
+ Log(detector, Form("GetFileIDs - In TESTMODE - Simulating error while connecting to %s FXS", GetSystemName(system)));
+ return 0;
+ }
+
+ // check connection, in case connect
+ if (!Connect(system))
+ {
+ Log(detector, Form("GetFileIDs - Couldn't connect to %s FXS database", GetSystemName(system)));
+ return NULL;
+ }
+
+ TString sourceName = 0;
+ if (system == kDAQ)
+ {
+ sourceName = "DAQsource";
+ } else if (system == kHLT)
+ {
+ sourceName = "DDLnumbers";
+ }
+
+ 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()));
+
+ // Query execution
+ TSQLResult* aResult;
+ aResult = fServer[system]->Query(sqlQuery);
+ if (!aResult) {
+ 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("GetFileIDs - No entry in %s FXS table for source: %s", GetSystemName(system), source));
+ delete aResult;
+ return list;
+ }
+
+ Log(detector, Form("GetFileIDs - Found %d ids", aResult->GetRowCount()));
+
+ TSQLRow* aRow;
+
+ while ((aRow = aResult->Next()))
+ {
+
+ TString id(aRow->GetField(0), aRow->GetFieldLength(0));
+ AliDebug(2, Form("fileId = %s", id.Data()));
+ list->Add(new TObjString(id));
+ delete aRow;
}
- return NULL;
+ delete aResult;
+
+ return list;
}
//______________________________________________________________________________________________
-Bool_t AliShuttle::Connect(Int_t system){
-// Connect to MySQL Server of the system's FES logbook
+Bool_t AliShuttle::Connect(Int_t system)
+{
+ // Connect to MySQL Server of the system's FXS MySQL databases
+ // DAQ Logbook, Shuttle Logbook and DAQ FXS db are on the same host
+ //
// check connection: if already connected return
if(fServer[system] && fServer[system]->IsConnected()) return kTRUE;
- TString aFESlbHost= Form("mysql://%s", fConfig->GetFESlbHost(system));
+ TString dbHost, dbUser, dbPass, dbName;
- fServer[system] = TSQLServer::Connect(aFESlbHost,
- fConfig->GetFESlbUser(system),
- fConfig->GetFESlbPass(system));
+ if (system < 3) // FXS db servers
+ {
+ dbHost = Form("mysql://%s:%d", fConfig->GetFXSdbHost(system), fConfig->GetFXSdbPort(system));
+ dbUser = fConfig->GetFXSdbUser(system);
+ dbPass = fConfig->GetFXSdbPass(system);
+ dbName = fConfig->GetFXSdbName(system);
+ } else { // Run & Shuttle logbook servers
+ // TODO Will the Shuttle logbook server be the same as the Run logbook server ???
+ dbHost = Form("mysql://%s:%d", fConfig->GetDAQlbHost(), fConfig->GetDAQlbPort());
+ dbUser = fConfig->GetDAQlbUser();
+ dbPass = fConfig->GetDAQlbPass();
+ dbName = fConfig->GetDAQlbDB();
+ }
+
+ fServer[system] = TSQLServer::Connect(dbHost.Data(), dbUser.Data(), dbPass.Data());
if (!fServer[system] || !fServer[system]->IsConnected()) {
- AliError(Form("Can't establish connection to FES logbook for %s !",fkSystemNames[system]));
+ if(system < 3)
+ {
+ AliError(Form("Can't establish connection to FXS database for %s",
+ AliShuttleInterface::GetSystemName(system)));
+ } else {
+ AliError("Can't establish connection to Run logbook.");
+ }
+ if(fServer[system]) delete fServer[system];
return kFALSE;
}
// Get tables
- // TODO in the configuration should the table name be there too?
+ TSQLResult* aResult=0;
switch(system){
case kDAQ:
- fServer[kDAQ]->GetTables("REFSYSLOG");
+ aResult = fServer[kDAQ]->GetTables(dbName.Data());
break;
case kDCS:
- //fServer[kDCS]->GetTables("REFSYSLOG");
+ aResult = fServer[kDCS]->GetTables(dbName.Data());
break;
case kHLT:
- //fServer[kHLT]->GetTables("REFSYSLOG");
+ aResult = fServer[kHLT]->GetTables(dbName.Data());
break;
default:
+ aResult = fServer[3]->GetTables(dbName.Data());
break;
}
+ delete aResult;
+ return kTRUE;
+}
+
+//______________________________________________________________________________________________
+Bool_t AliShuttle::UpdateTable()
+{
+ //
+ // Update FXS table filling time_processed field in all rows corresponding to current run and detector
+ //
+
+ Bool_t result = kTRUE;
+
+ for (UInt_t system=0; system<3; system++)
+ {
+ if(!fFXSCalled[system]) continue;
+
+ // check connection, in case connect
+ if (!Connect(system))
+ {
+ Log(fCurrentDetector, Form("UpdateTable - Couldn't connect to %s FXS database", GetSystemName(system)));
+ result = kFALSE;
+ continue;
+ }
+
+ TTimeStamp now; // now
+
+ // Loop on FXS list entries
+ TIter iter(&fFXSlist[system]);
+ TObjString *aFXSentry=0;
+ while ((aFXSentry = dynamic_cast<TObjString*> (iter.Next())))
+ {
+ TString aFXSentrystr = aFXSentry->String();
+ TObjArray *aFXSarray = aFXSentrystr.Tokenize("#!?!#");
+ if (!aFXSarray || aFXSarray->GetEntries() != 2 )
+ {
+ Log(fCurrentDetector, Form("UpdateTable - error updating %s FXS entry. Check string: <%s>",
+ GetSystemName(system), aFXSentrystr.Data()));
+ if(aFXSarray) delete aFXSarray;
+ result = kFALSE;
+ continue;
+ }
+ const char* fileId = ((TObjString*) aFXSarray->At(0))->GetName();
+ const char* source = ((TObjString*) aFXSarray->At(1))->GetName();
+
+ TString whereClause;
+ if (system == kDAQ)
+ {
+ whereClause = Form("where run=%d and detector=\"%s\" and fileId=\"%s\" and DAQsource=\"%s\";",
+ GetCurrentRun(), fCurrentDetector.Data(), fileId, source);
+ }
+ else if (system == kDCS)
+ {
+ whereClause = Form("where run=%d and detector=\"%s\" and fileId=\"%s\";",
+ GetCurrentRun(), fCurrentDetector.Data(), fileId);
+ }
+ else if (system == kHLT)
+ {
+ whereClause = Form("where run=%d and detector=\"%s\" and fileId=\"%s\" and DDLnumbers=\"%s\";",
+ GetCurrentRun(), fCurrentDetector.Data(), fileId, source);
+ }
+
+ delete aFXSarray;
+
+ 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(fCurrentDetector, Form("UpdateTable - %s db: can't execute SQL query <%s>",
+ GetSystemName(system), sqlQuery.Data()));
+ result = kFALSE;
+ continue;
+ }
+ delete aResult;
+ }
+ }
+
+ return result;
+}
+
+//______________________________________________________________________________________________
+Bool_t AliShuttle::UpdateTableFailCase()
+{
+ // Update FXS table filling time_processed field in all rows corresponding to current run and detector
+ // this is called in case the preprocessor is declared failed for the current run, because
+ // the fields are updated only in case of success
+
+ Bool_t result = kTRUE;
+
+ for (UInt_t system=0; system<3; system++)
+ {
+ // check connection, in case connect
+ if (!Connect(system))
+ {
+ Log(fCurrentDetector, Form("UpdateTableFailCase - Couldn't connect to %s FXS database",
+ GetSystemName(system)));
+ result = kFALSE;
+ continue;
+ }
+
+ TTimeStamp now; // now
+
+ // Loop on FXS list entries
+
+ TString whereClause = Form("where run=%d and detector=\"%s\";",
+ GetCurrentRun(), fCurrentDetector.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(fCurrentDetector, Form("UpdateTableFailCase - %s db: can't execute SQL query <%s>",
+ GetSystemName(system), sqlQuery.Data()));
+ result = kFALSE;
+ continue;
+ }
+ delete aResult;
+ }
+
+ return result;
+}
+
+//______________________________________________________________________________________________
+Bool_t AliShuttle::UpdateShuttleLogbook(const char* detector, const char* status)
+{
+ //
+ // Update Shuttle logbook filling detector or shuttle_done column
+ // ex. of usage: UpdateShuttleLogbook("PHOS", "DONE") or UpdateShuttleLogbook("shuttle_done")
+ //
+
+ // check connection, in case connect
+ if(!Connect(3)){
+ Log("SHUTTLE", "UpdateShuttleLogbook - Couldn't connect to DAQ Logbook.");
+ return kFALSE;
+ }
+
+ TString detName(detector);
+ TString setClause;
+ if (detName == "shuttle_done" || detName == "shuttle_ignored")
+ {
+ setClause = "set shuttle_done=1";
+
+ if (detName == "shuttle_done")
+ {
+ if (TouchFile()==kTRUE){
+ //Send the information to ML
+ TMonaLisaText mlStatus("SHUTTLE_status", "Done");
+
+ TList mlList;
+ mlList.Add(&mlStatus);
+
+ TString mlID;
+ mlID.Form("%d", GetCurrentRun());
+ fMonaLisa->SendParameters(&mlList, mlID);
+ }
+ else{
+ return kFALSE;
+ }
+
+ }
+ } else {
+ TString statusStr(status);
+ if(statusStr.Contains("done", TString::kIgnoreCase) ||
+ statusStr.Contains("failed", TString::kIgnoreCase)){
+ setClause = Form("set %s=\"%s\"", detector, status);
+ } else {
+ Log("SHUTTLE",
+ Form("UpdateShuttleLogbook - Invalid status <%s> for detector %s",
+ status, detector));
+ return kFALSE;
+ }
+ }
+
+ TString whereClause = Form("where run=%d", GetCurrentRun());
+
+ TString sqlQuery = Form("update %s %s %s",
+ fConfig->GetShuttlelbTable(), setClause.Data(), whereClause.Data());
+
+ AliDebug(2, Form("SQL query: \n%s",sqlQuery.Data()));
+
+ // Query execution
+ TSQLResult* aResult;
+ aResult = dynamic_cast<TSQLResult*> (fServer[3]->Query(sqlQuery));
+ if (!aResult) {
+ Log("SHUTTLE", Form("UpdateShuttleLogbook - Can't execute query <%s>", sqlQuery.Data()));
+ return kFALSE;
+ }
+ delete aResult;
+
return kTRUE;
}
-//______________________________________________________________________________________________
-const char* AliShuttle::GetDAQFileName(const char* detector, const char* id, const char* source){
-// Retrieves a file from the DAQ FES.
-// First queris the DAQ logbook_fs for the DAQ file name, using the run, detector, id and source info
-// then calls RetrieveDAQFile(DAQfilename) for actual copy to local disk
-// run: current run being processed (fCurrentRun)
-// detector: comes from the Preprocessor name (must be converted into detector code with GetDetCode)
-// id: provided as a parameter by the Preprocessor
-// source: provided by the Preprocessor through GetFileSources function
+//______________________________________________________________________________________________
+Int_t AliShuttle::GetCurrentRun() const
+{
+ //
+ // Get current run from logbook entry
+ //
+
+ return fLogbookEntry ? fLogbookEntry->GetRun() : -1;
+}
+
+//______________________________________________________________________________________________
+UInt_t AliShuttle::GetCurrentStartTime() const
+{
+ //
+ // get current start time
+ //
+
+ return fLogbookEntry ? fLogbookEntry->GetStartTime() : 0;
+}
+
+//______________________________________________________________________________________________
+UInt_t AliShuttle::GetCurrentEndTime() const
+{
+ //
+ // get current end time from logbook entry
+ //
+
+ return fLogbookEntry ? fLogbookEntry->GetEndTime() : 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
+ //
+
+ TString logRunDir = GetShuttleLogDir();
+ if (GetCurrentRun() >=0)
+ logRunDir += Form("/%d", GetCurrentRun());
+
+ void* dir = gSystem->OpenDirectory(logRunDir.Data());
+ if (dir == NULL) {
+ if (gSystem->mkdir(logRunDir.Data(), kTRUE)) {
+ AliError(Form("Can't open directory <%s>", GetShuttleLogDir()));
+ return;
+ }
+
+ } else {
+ gSystem->FreeDirectory(dir);
+ }
+
+ 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);
+
+ AliInfo(toLog.Data());
+
+ // if we redirect the log output already to the file, leave here
+ if (fOutputRedirected && strcmp(detector, "SHUTTLE") != 0)
+ return;
+
+ TString fileName = GetLogFileName(detector);
+
+ gSystem->ExpandPathName(fileName);
+
+ ofstream logFile;
+ logFile.open(fileName, ofstream::out | ofstream::app);
+
+ if (!logFile.is_open()) {
+ AliError(Form("Could not open file %s", fileName.Data()));
+ return;
+ }
+
+ logFile << toLog.Data() << "\n";
+
+ logFile.close();
+}
+
+//______________________________________________________________________________________________
+TString AliShuttle::GetLogFileName(const char* detector) const
+{
+ //
+ // returns the name of the log file for a given sub detector
+ //
+
+ TString fileName;
+
+ if (GetCurrentRun() >= 0)
+ {
+ fileName.Form("%s/%d/%s_%d.log", GetShuttleLogDir(), GetCurrentRun(),
+ detector, GetCurrentRun());
+ } 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)
+{
+ //
+ // Collects conditions data for all UNPROCESSED run written to DAQ LogBook in case of run = -1 (default)
+ // If a dedicated run is given this run is processed
+ //
+ // In operational mode, this is the Shuttle function triggered by the EOR signal.
+ //
+
+ if (run == -1)
+ Log("SHUTTLE","Collect - Shuttle called. Collecting conditions data for unprocessed runs");
+ else
+ Log("SHUTTLE", Form("Collect - Shuttle called. Collecting conditions data for run %d", run));
- // check connection, in case connect
- if(!Connect(kDAQ)){
- Log(detector, "GetDAQFileName: Couldn't connect to DAQ Logbook !");
- return 0;
- }
+ SetLastAction("Starting");
- // Query preparation
- TString sqlQueryStart = "select filePath from logbook_fs where";
- TString whereClause = Form("run=%d and detector=\"%s\" and fileId=\"%s\" and DAQsource=\"%s\"",
- fCurrentRun, GetDetCode(detector), id, source);
- TString sqlQuery = Form("%s %s", sqlQueryStart.Data(), whereClause.Data());
+ // create ML instance
+ if (!fMonaLisa)
+ fMonaLisa = new TMonaLisaWriter(fConfig->GetMonitorHost(), fConfig->GetMonitorTable());
+
+ SendAlive();
+ CountOpenRuns();
- AliDebug(2, Form("SQL query: \n%s",sqlQuery.Data()));
+ TString whereClause("where shuttle_done=0");
+ if (run != -1)
+ whereClause += Form(" and run=%d", run);
- // Query execution
- TSQLResult* aResult;
- aResult = fServer[kDAQ]->Query(sqlQuery);
- if (!aResult) {
- Log(detector, Form("Can't execute query <%s>!", sqlQuery.Data()));
- return 0;
+ TObjArray shuttleLogbookEntries;
+ if (!QueryShuttleLogbook(whereClause, shuttleLogbookEntries))
+ {
+ Log("SHUTTLE", "Collect - Can't retrieve entries from Shuttle logbook");
+ return kFALSE;
}
- if (aResult->GetRowCount() == 0) {
- Log(detector,
- Form("GetDAQFileName: No result from SQL query <%s>!", sqlQuery.Data()));
- delete aResult;
- return 0;
+ if (shuttleLogbookEntries.GetEntries() == 0)
+ {
+ if (run == -1)
+ Log("SHUTTLE","Collect - Found no UNPROCESSED runs in Shuttle logbook");
+ else
+ Log("SHUTTLE", Form("Collect - Run %d is already DONE "
+ "or it does not exist in Shuttle logbook", run));
+ return kTRUE;
}
- if (aResult->GetRowCount() >1) {
- Log(detector,
- Form("GetDAQFileName: More than one row resulting from SQL query <%s>!", sqlQuery.Data()));
- delete aResult;
- return 0;
- }
+ for (UInt_t iDet=0; iDet<NDetectors(); iDet++)
+ fFirstUnprocessed[iDet] = kTRUE;
- TSQLRow* aRow = aResult->Next();
+ if (run != -1)
+ {
+ // 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));
+ TObjArray tmpLogbookEntries;
+ if (!QueryShuttleLogbook(whereClause, tmpLogbookEntries))
+ {
+ Log("SHUTTLE", "Collect - Can't retrieve entries from Shuttle logbook");
+ return kFALSE;
+ }
- if(!aRow){
- Log(detector, Form("GetDAQFileName: Empty set result from query <%s>!", sqlQuery.Data()));
- delete aResult;
- return 0;
- }
+ TIter iter(&tmpLogbookEntries);
+ AliShuttleLogbookEntry* anEntry = 0;
+ while ((anEntry = dynamic_cast<AliShuttleLogbookEntry*> (iter.Next())))
+ {
+ for (UInt_t iDet=0; iDet<NDetectors(); iDet++)
+ {
+ if (anEntry->GetDetectorStatus(iDet) == AliShuttleLogbookEntry::kUnprocessed)
+ {
+ AliDebug(2, Form("Run %d: setting %s as \"not first time unprocessed\"",
+ anEntry->GetRun(), GetDetName(iDet)));
+ fFirstUnprocessed[iDet] = kFALSE;
+ }
+ }
- TString filePath(aRow->GetField(0), aRow->GetFieldLength(0));
+ }
- delete aResult;
+ }
- AliDebug(2, Form("filePath = %s",filePath.Data()));
+ if (!RetrieveConditionsData(shuttleLogbookEntries))
+ {
+ Log("SHUTTLE", "Collect - Process of at least one run failed");
+ CountOpenRuns();
+ return kFALSE;
+ }
- // retrieved file is renamed to make it unique
- TString localFileName = Form("%s_%d_%s_%s.shuttle",
- detector, fCurrentRun, id, source);
+ Log("SHUTTLE", "Collect - Requested run(s) successfully processed");
+ CountOpenRuns();
+ return kTRUE;
+}
- // file retrieval from DAQ FES
- Bool_t result = RetrieveDAQFile(filePath.Data(), localFileName.Data());
- if(!result) {
- Log(detector, Form("copying file %s from DAQ FES failed!", filePath.Data()));
- return 0;
- } else {
- AliInfo(Form("File %s copied from DAQ FES into %s/%s !",
- filePath.Data(), fgkShuttleTempDir, localFileName.Data()));
- }
+//______________________________________________________________________________________________
+Bool_t AliShuttle::RetrieveConditionsData(const TObjArray& dateEntries)
+{
+ //
+ // Retrieve conditions data for all runs that aren't processed yet
+ //
+ Bool_t hasError = kFALSE;
+
+ TIter iter(&dateEntries);
+ AliShuttleLogbookEntry* anEntry;
- fFESCalled[kDAQ]=kTRUE;
- TObjString *fileParams = new TObjString(Form("%s_!?!_%s", id, source));
- fFESlist[kDAQ].Add(fileParams);
+ while ((anEntry = (AliShuttleLogbookEntry*) iter.Next())){
+ if (!Process(anEntry)){
+ hasError = kTRUE;
+ }
- return localFileName.Data();
+ // clean SHUTTLE temp directory
+ //TString filename = Form("%s/*.shuttle", GetShuttleTempDir());
+ //RemoveFile(filename.Data());
+ }
+ return hasError == kFALSE;
}
//______________________________________________________________________________________________
-Bool_t AliShuttle::RetrieveDAQFile(const char* daqFileName, const char* localFileName){
+ULong_t AliShuttle::GetTimeOfLastAction() const
+{
+ //
+ // Gets time of last action
+ //
- // check temp directory: trying to cd to temp; if it does not exist, create it
- AliDebug(2, Form("Copy file %s from DAQ FES into folder %s and rename it as %s",
- daqFileName,fgkShuttleTempDir, localFileName));
+ ULong_t tmp;
- void* dir = gSystem->OpenDirectory(fgkShuttleTempDir);
- if (dir == NULL) {
- if (gSystem->mkdir(fgkShuttleTempDir, kTRUE)) {
- AliError(Form("Can't open directory <%s>!", fgkShuttleTempDir));
- return kFALSE;
- }
+ fMonitoringMutex->Lock();
- } else {
- gSystem->FreeDirectory(dir);
- }
+ tmp = fLastActionTime;
- TString baseDAQFESFolder = "DAQ";
- TString command = Form("scp %s@%s:%s/%s %s/%s",
- fConfig->GetFESUser(kDAQ),
- fConfig->GetFESHost(kDAQ),
- baseDAQFESFolder.Data(),
- daqFileName,
- fgkShuttleTempDir,
- localFileName);
+ fMonitoringMutex->UnLock();
- AliDebug(2, Form("%s",command.Data()));
+ return tmp;
+}
- UInt_t nRetries = 0;
- UInt_t maxRetries = 3;
+//______________________________________________________________________________________________
+const TString AliShuttle::GetLastAction() const
+{
+ //
+ // returns a string description of the last action
+ //
- // copy!! if successful TSystem::Exec returns 0
- while(nRetries++ < maxRetries) {
- AliDebug(2, Form("Trying to copy file. Retry # %d", nRetries));
- if(gSystem->Exec(command.Data()) == 0) return kTRUE;
- }
+ TString tmp;
+
+ fMonitoringMutex->Lock();
+
+ tmp = fLastAction;
+
+ fMonitoringMutex->UnLock();
+
+ return tmp;
+}
+
+//______________________________________________________________________________________________
+void AliShuttle::SetLastAction(const char* action)
+{
+ //
+ // updates the monitoring variables
+ //
- return kFALSE;
+ fMonitoringMutex->Lock();
+ fLastAction = action;
+ fLastActionTime = time(0);
+
+ fMonitoringMutex->UnLock();
}
//______________________________________________________________________________________________
-TList* AliShuttle::GetDAQFileSources(const char* detector, const char* id){
-// Retrieves a file from the DCS FES.
+const char* AliShuttle::GetRunParameter(const char* param)
+{
+ //
+ // returns run parameter read from DAQ logbook
+ //
- // check connection, in case connect
- if(!Connect(kDAQ)){
- Log(detector, "GetDAQFileName: Couldn't connect to DAQ Logbook !");
+ if(!fLogbookEntry) {
+ AliError("No logbook entry!");
return 0;
}
- // Query preparation
- TString sqlQueryStart = "select DAQsource from logbook_fs where";
- TString whereClause = Form("run=%d and detector=\"%s\" and fileId=\"%s\"",
- fCurrentRun, GetDetCode(detector), id);
- TString sqlQuery = Form("%s %s", sqlQueryStart.Data(), whereClause.Data());
+ return fLogbookEntry->GetRunParameter(param);
+}
- AliDebug(2, Form("SQL query: \n%s",sqlQuery.Data()));
+//______________________________________________________________________________________________
+AliCDBEntry* AliShuttle::GetFromOCDB(const char* detector, const AliCDBPath& path)
+{
+ //
+ // returns object from OCDB valid for current run
+ //
- // Query execution
- TSQLResult* aResult;
- aResult = fServer[kDAQ]->Query(sqlQuery);
- if (!aResult) {
- Log(detector, Form("GetDAQFileSources: Can't execute query <%s>!", sqlQuery.Data()));
+ if (fTestMode & kErrorOCDB)
+ {
+ Log(detector, "GetFromOCDB - In TESTMODE - Simulating error with OCDB");
return 0;
}
-
- if (aResult->GetRowCount() == 0) {
- Log(detector,
- Form("GetDAQFileSources: No result from SQL query <%s>!", sqlQuery.Data()));
- delete aResult;
+
+ AliCDBStorage *sto = AliCDBManager::Instance()->GetStorage(fgkMainCDB);
+ if (!sto)
+ {
+ Log(detector, "GetFromOCDB - Cannot activate main OCDB for query!");
return 0;
}
- TSQLRow* aRow;
- TList *list = new TList();
- list->SetOwner(1);
-
- while((aRow = aResult->Next())){
+ return dynamic_cast<AliCDBEntry*> (sto->Get(path, GetCurrentRun()));
+}
- TString daqSource(aRow->GetField(0), aRow->GetFieldLength(0));
- AliDebug(2, Form("daqSource = %s", daqSource.Data()));
- list->Add(new TObjString(daqSource));
+//______________________________________________________________________________________________
+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;
}
- delete aResult;
- return list;
+ void* dir = gSystem->OpenDirectory(GetShuttleLogDir());
+ if (dir == NULL)
+ {
+ if (gSystem->mkdir(GetShuttleLogDir(), kTRUE))
+ {
+ Log("SHUTTLE", Form("SendMail - Can't open directory <%s>", GetShuttleLogDir()));
+ return kFALSE;
+ }
-}
+ } else {
+ gSystem->FreeDirectory(dir);
+ }
-//______________________________________________________________________________________________
-Bool_t AliShuttle::UpdateDAQTable(){
-// Update DAQ table filling time_processed field in all rows corresponding to current run and detector
+ // det experts in to
+ TString to="";
+ TIter *iterExperts;
+ if (target == kDCSEMail) {
+ iterExperts = new TIter(fConfig->GetAdmins(AliShuttleConfig::kAmanda));
+ }
+ else if (target == kFXSEMail) {
+ iterExperts = new TIter(fConfig->GetAdmins(system));
+ }
+ else {
+ 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()));
- // check connection, in case connect
- if(!Connect(kDAQ)){
- Log(fCurrentDetector, "UpdateDAQTable: Couldn't connect to DAQ Logbook !");
+ if (to.IsNull()) {
+ Log("SHUTTLE", Form("List of %d responsibles not set!", (Int_t) target));
return kFALSE;
}
- TTimeStamp now; // now
-
- // Loop on FES list entries
- TIter iter(&fFESlist[kDAQ]);
- TObjString *aFESentry=0;
- while((aFESentry = dynamic_cast<TObjString*> (iter.Next()))){
- TString aFESentrystr = aFESentry->String();
- TObjArray *aFESarray = aFESentrystr.Tokenize("_!?!_");
- if(!aFESarray || aFESarray->GetEntries() != 2 ) {
- Log(fCurrentDetector,Form("UpdateDAQTable: error updating FES entry! string = %s",
- aFESentrystr.Data()));
- if(aFESarray) delete aFESarray;
- return kFALSE;
- }
- const char* fileId = ((TObjString*) aFESarray->At(0))->GetName();
- const char* daqSource = ((TObjString*) aFESarray->At(1))->GetName();
- TString whereClause = Form("where run=%d and detector=\"%s\" and fileId=\"%s\" and DAQsource=\"%s\";",
- fCurrentRun,GetDetCode(fCurrentDetector), fileId, daqSource);
-
- delete aFESarray;
+ // 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()));
- TString sqlQuery = Form("update logbook_fs set time_processed=%d %s", now.GetSec(), whereClause.Data());
+ // mail body
+ TString bodyFileName;
+ bodyFileName.Form("%s/mail.body", GetShuttleLogDir());
+ gSystem->ExpandPathName(bodyFileName);
- AliDebug(2, Form("SQL query: \n%s",sqlQuery.Data()));
+ ofstream mailBody;
+ mailBody.open(bodyFileName, ofstream::out);
- // Query execution
- TSQLResult* aResult;
- aResult = dynamic_cast<TSQLResult*> (fServer[kDAQ]->Query(sqlQuery));
- if (!aResult) {
- Log(fCurrentDetector, Form("UpdateDAQTable: Can't execute query <%s>!", sqlQuery.Data()));
- return kFALSE;
- }
- delete aResult;
+ if (!mailBody.is_open())
+ {
+ Log("SHUTTLE", Form("Could not open mail body file %s", bodyFileName.Data()));
+ return kFALSE;
+ }
- // check result - TODO Is it necessary?
- sqlQuery = Form("select time_processed from logbook_fs %s", whereClause.Data());
- AliDebug(2, Form(" CHECK - SQL query: \n%s",sqlQuery.Data()));
- aResult = dynamic_cast<TSQLResult*> (fServer[kDAQ]->Query(sqlQuery));
- if (!aResult) {
- AliWarning("Can't check result!");
- continue;
- }
+ TString subject;
+ TString body;
- if (aResult->GetRowCount() == 0) {
- Log(fCurrentDetector,
- Form("GetDAQFileName: No result from SQL query <%s>!", sqlQuery.Data()));
- delete aResult;
- //return 0;
+ if (target == kDCSEMail){
+ subject = Form("Retrieval of data points for %s FAILED in run %d !",
+ 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("FXS communication for %s FAILED in run %d !",
+ 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 Shuttle preprocessor FAILED in run %d (run type = %s)!",
+ 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());
}
- if (aResult->GetRowCount() >1) {
- Log(fCurrentDetector,
- Form("GetDAQFileName: More than one row resulting from SQL query <%s>!", sqlQuery.Data()));
- delete aResult;
- //return 0;
+ 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:8889/shuttle.jsp?time=168 \n\n");
+ } else {
+ body += Form("\thttp://pcalimonitor.cern.ch/shuttle.jsp?instance=PROD&time=168 \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/%s_%d.log \n\n",
+ fCurrentDetector.Data(), logFolder.Data(), GetCurrentRun(),
+ fCurrentDetector.Data(), GetCurrentRun());
+ 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 << body.Data();
+ mailBody.close();
+ mailBody.open(bodyFileName, ofstream::out | ofstream::app);
+
+ TString logFileName = Form("%s/%d/%s_%d.log", GetShuttleLogDir(),
+ GetCurrentRun(), fCurrentDetector.Data(), GetCurrentRun());
+ TString tailCommand = Form("tail -n 10 %s >> %s", logFileName.Data(), bodyFileName.Data());
+ if (gSystem->Exec(tailCommand.Data()))
+ {
+ mailBody << Form("%s log file not found ...\n\n", fCurrentDetector.Data());
}
- TSQLRow *row = dynamic_cast<TSQLRow*> (aResult->Next());
- TString processedTimeString(row->GetField(0), row->GetFieldLength(0));
- Int_t processedTime = processedTimeString.Atoi();
- if(processedTime != now.GetSec()){
- Log(fCurrentDetector, Form("UpdateDAQTable: Update table error: processed_time=%d, now=%d !",
- processedTime, now.GetSec()));
- delete aResult;
- return kFALSE;
- }
+ 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 += "Greetings,\n\n \t\t\tthe SHUTTLE\n";
- delete aResult;
+ AliDebug(2, Form("Body end: %s", endBody.Data()));
- }
+ mailBody << endBody.Data();
- return kTRUE;
-}
+ mailBody.close();
-//______________________________________________________________________________________________
-const char* AliShuttle::GetDCSFileName(const char* /*detector*/, const char* /*id*/, const char* /*source*/){
-// Retrieves a file from the DCS FES.
+ // send mail!
+ TString mailCommand = Form("mail -s \"%s\" -c %s %s < %s",
+ subject.Data(),
+ cc.Data(),
+ to.Data(),
+ bodyFileName.Data());
+ AliDebug(2, Form("mail command: %s", mailCommand.Data()));
-return "You're in DCS";
+ Bool_t result = gSystem->Exec(mailCommand.Data());
+ return result == 0;
}
//______________________________________________________________________________________________
-TList* AliShuttle::GetDCSFileSources(const char* /*detector*/, const char* /*id*/){
-// Retrieves a file from the DCS FES.
+const char* AliShuttle::GetRunType()
+{
+ //
+ // returns run type read from "run type" logbook
+ //
-return NULL;
+ if(!fLogbookEntry) {
+ AliError("No logbook entry!");
+ return 0;
+ }
+ return fLogbookEntry->GetRunType();
}
//______________________________________________________________________________________________
-const char* AliShuttle::GetHLTFileName(const char* /*detector*/, const char* /*id*/, const char* /*source*/){
-// Retrieves a file from the HLT FES.
+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)
-return "You're in HLT";
+ 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;
+ }
}
//______________________________________________________________________________________________
-TList* AliShuttle::GetHLTFileSources(const char* /*detector*/, const char* /*id*/){
-// Retrieves a file from the HLT FES.
+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;
-return NULL;
+ 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;
}
//______________________________________________________________________________________________
-const char* AliShuttle::GetDetCode(const char* detector){
-// Return detector code
-
- for(int iDet=0; iDet < fgkNDetectors; iDet++){
- if(!strcmp(fgkDetectorName[iDet], detector)) return fgkDetectorCode[iDet];
- }
+void AliShuttle::SetShuttleTempDir(const char* tmpDir)
+{
+ //
+ // sets Shuttle temp directory
+ //
- return 0;
+ fgkShuttleTempDir = gSystem->ExpandPathName(tmpDir);
}
//______________________________________________________________________________________________
-void AliShuttle::Log(const char* detector, const char* message)
+void AliShuttle::SetShuttleLogDir(const char* logDir)
{
-// Fill log string with a message
-
- void* dir = gSystem->OpenDirectory(fgkShuttleLogDir);
- if (dir == NULL) {
- if (gSystem->mkdir(fgkShuttleLogDir, kTRUE)) {
- AliError(Form("Can't open directory <%s>!", fgkShuttleTempDir));
- return;
- }
+ //
+ // sets Shuttle log directory
+ //
- } else {
- gSystem->FreeDirectory(dir);
+ 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 toLog = Form("%s: %s, run %d - %s", TTimeStamp(time(0)).AsString("s"),
- detector, GetCurrentRun(), message);
- AliInfo(toLog.Data());
+ TString dir;
+ dir.Form("%s%d/%s/SHUTTLE_DONE", fConfig->GetAlienPath(), GetCurrentYear(), GetLHCPeriod());
+ // 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()));
+ 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()));
- TString fileName;
- fileName.Form("%s/%s.log", fgkShuttleLogDir, detector);
- gSystem->ExpandPathName(fileName);
+ // creating the directory
- ofstream logFile;
- logFile.open(fileName, ofstream::out | ofstream::app);
+ 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()));
+ 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()));
+ }
- if (!logFile.is_open()) {
- AliError(Form("Could not open file %s", fileName.Data()));
- return;
- }
+ 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()));
+ 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()));
+ return kFALSE;
+ }
+ if (valueTouch->GetString()!="1"){
+ Log("SHUTTLE",Form("Failing the touching command, returning without touching for run %i",GetCurrentRun()));
+ return kFALSE;
+ }
+ return kTRUE;
+}
- logFile << toLog.Data() << "\n";
- logFile.close();
-}
git://git.uio.no / u / mrichter / AliRoot.git / blobdiff