/*
$Log$
+Revision 1.30 2007/02/13 11:23:21 acolla
+Moved getters and setters of Shuttle's main OCDB/Reference, local
+OCDB/Reference, temp and log folders to AliShuttleInterface
+
+Revision 1.27 2007/01/30 17:52:42 jgrosseo
+adding monalisa monitoring
+
+Revision 1.26 2007/01/23 19:20:03 acolla
+Removed old ldif files, added TOF, MCH ldif files. Added some options in
+AliShuttleConfig::Print. Added in Ali Shuttle: SetShuttleTempDir and
+SetShuttleLogDir
+
+Revision 1.25 2007/01/15 19:13:52 acolla
+Moved some AliInfo to AliDebug in SendMail function
+
+Revision 1.21 2006/12/07 08:51:26 jgrosseo
+update (alberto):
+table, db names in ldap configuration
+added GRP preprocessor
+DCS data can also be retrieved by data point
+
+Revision 1.20 2006/11/16 16:16:48 jgrosseo
+introducing strict run ordering flag
+removed giving preprocessor name to preprocessor, they have to know their name themselves ;-)
+
+Revision 1.19 2006/11/06 14:23:04 jgrosseo
+major update (Alberto)
+o) reading of run parameters from the logbook
+o) online offline naming conversion
+o) standalone DCSclient package
+
+Revision 1.18 2006/10/20 15:22:59 jgrosseo
+o) Adding time out to the execution of the preprocessors: The Shuttle forks and the parent process monitors the child
+o) Merging Collect, CollectAll, CollectNew function
+o) Removing implementation of empty copy constructors (declaration still there!)
+
+Revision 1.17 2006/10/05 16:20:55 jgrosseo
+adapting to new CDB classes
+
+Revision 1.16 2006/10/05 15:46:26 jgrosseo
+applying to the new interface
+
+Revision 1.15 2006/10/02 16:38:39 jgrosseo
+update (alberto):
+fixed memory leaks
+storing of objects that failed to be stored to the grid before
+interfacing of shuttle status table in daq system
+
+Revision 1.14 2006/08/29 09:16:05 jgrosseo
+small update
+
+Revision 1.13 2006/08/15 10:50:00 jgrosseo
+effc++ corrections (alberto)
+
Revision 1.12 2006/08/08 14:19:29 jgrosseo
Update to shuttle classes (Alberto)
#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 <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"};
+ClassImp(AliShuttle)
-const char* AliShuttle::fgkDetectorCode[AliShuttle::fgkNDetectors] = {"SPD", "SDD", "SSD", "TPC", "TRD", "TOF",
- "PHS", "CPV", "HMP", "EMC", "MCH", "MTR", "FMD", "ZDC", "PMD", "T00", "V00"};
+Bool_t AliShuttle::fgkProcessDCS(kTRUE);
//______________________________________________________________________________________________
AliShuttle::AliShuttle(const AliShuttleConfig* config,
fConfig(config),
fTimeout(timeout), fRetries(retries),
fPreprocessorMap(),
-fCurrentRun(-1),
-fCurrentStartTime(0), fCurrentEndTime(0),
-fCurrentDetector(""),
-fStatusEntry(0)
+fLogbookEntry(0),
+fCurrentDetector(),
+fStatusEntry(0),
+fGridError(kFALSE),
+fMonitoringMutex(0),
+fLastActionTime(0),
+fLastAction(),
+fMonaLisa(0)
{
//
// 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)
-
-{
-// 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();
}
//______________________________________________________________________________________________
// 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,
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!
- //
-
- // The parameters are
- // 1) the object's path.
- // 2) the object to be stored
- // 3) the metaData to be associated with the object
- // 4) the validity start run number w.r.t. the current run,
- // if the data is valid only for this run leave the default 0
- // 5) 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) CDB storage
- // 2 if stored in backup (Local) CDB storage
-
-
- Int_t firstRun = GetCurrentRun() - validityStart;
- if(firstRun < 0) {
- AliError("First valid run happens to be less than 0! Setting it to 0...");
- firstRun=0;
- }
-
- Int_t lastRun = -1;
- if(validityInfinite) {
- lastRun = AliCDBRunRange::Infinity();
- } else {
- lastRun = GetCurrentRun();
- }
-
- AliCDBId id(path, firstRun, lastRun);
-
- UInt_t result = 0;
-
- if (!(AliCDBManager::Instance()->GetStorage(fgkMainCDB))) {
- Log(fCurrentDetector, "Cannot activate main CDB storage!");
- } else {
- result = (UInt_t) AliCDBManager::Instance()->GetStorage(fgkMainCDB)
- ->Put(object, id, metaData);
- }
-
- if(!result) {
-
- Log(fCurrentDetector,
- "Error while storing object in main storage: it will go to local storage!");
-
- result = AliCDBManager::Instance()->GetStorage(fgkLocalCDB)
- ->Put(object, id, metaData);
+ // It calls WriteToCDB function which perform actual storage
- if(result) {
- result = 2;
- }else{
- Log(fCurrentDetector, "Can't store data!");
- }
- }
- return result;
+ return WriteToCDB(fgkMainCDB, fgkLocalCDB, path, object,
+ metaData, validityStart, validityInfinite);
}
//______________________________________________________________________________________________
-UInt_t AliShuttle::StoreReferenceData(const AliCDBPath& path, TObject* object,
- AliCDBMetaData* metaData, Int_t validityStart, Bool_t validityInfinite)
+UInt_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);
- // The parameters are
- // 1) the object's path.
- // 2) the object to be stored
- // 3) the metaData to be associated with the object
- // 4) the validity start run number w.r.t. the current run,
+}
+
+//______________________________________________________________________________________________
+UInt_t AliShuttle::WriteToCDB(const char* mainUri, const char* 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
- // 5) specifies if the calibration data is valid for infinity (this means until updated),
+ // 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) reference storage
- // 2 if stored in backup (Local) reference storage
+ // 1 if stored in main (Grid) storage
+ // 2 if stored in backup (Local) storage
+
+ const char* cdbType = (mainUri == fgkMainCDB) ? "CDB" : "Reference";
- Int_t firstRun = GetCurrentRun() - validityStart;
+ Int_t firstRun = GetCurrentRun() - validityStart;
if(firstRun < 0) {
- AliError("First valid run happens to be less than 0! Setting it to 0...");
+ AliError("First valid run happens to be less than 0! Setting it to 0.");
firstRun=0;
}
lastRun = GetCurrentRun();
}
- AliCDBId id(path, firstRun, lastRun);
+ AliCDBId id(path, firstRun, lastRun, -1, -1);
+
+ if(! dynamic_cast<TObjString*> (metaData->GetProperty("RunUsed(TObjString)"))){
+ TObjString runUsed = Form("%d", GetCurrentRun());
+ metaData->SetProperty("RunUsed(TObjString)", runUsed.Clone());
+ }
UInt_t result = 0;
- if (!(AliCDBManager::Instance()->GetStorage(fgkMainRefStorage))) {
- Log(fCurrentDetector, "Cannot activate main reference storage!");
+ if (!(AliCDBManager::Instance()->GetStorage(mainUri))) {
+ AliError(Form("WriteToCDB - Cannot activate main %s storage", cdbType));
} else {
- result = (UInt_t) AliCDBManager::Instance()->GetStorage(fgkMainRefStorage)
+ result = (UInt_t) AliCDBManager::Instance()->GetStorage(mainUri)
->Put(object, id, metaData);
}
if(!result) {
Log(fCurrentDetector,
- "Error while storing object in main reference storage: it will go to local ref storage!");
+ Form("WriteToCDB - Problem with main %s storage. Putting <%s> into backup storage",
+ cdbType, path.GetPath().Data()));
+
+ // Set Grid version to current run number, to ease retrieval later
+ id.SetVersion(GetCurrentRun());
- result = AliCDBManager::Instance()->GetStorage(fgkLocalRefStorage)
+ result = AliCDBManager::Instance()->GetStorage(localUri)
->Put(object, id, metaData);
if(result) {
result = 2;
+ fGridError = kTRUE;
}else{
- Log(fCurrentDetector, "Can't store reference data!");
+ Log(fCurrentDetector, "WriteToCDB - Can't store data!");
}
}
+
return result;
}
//______________________________________________________________________________________________
AliShuttleStatus* AliShuttle::ReadShuttleStatus()
{
- // Reads the AliShuttleStatus from the CDB
+// Reads the AliShuttleStatus from the CDB
- if (fStatusEntry)
- {
- delete fStatusEntry;
- fStatusEntry = 0;
- }
+ if (fStatusEntry){
+ delete fStatusEntry;
+ fStatusEntry = 0;
+ }
- fStatusEntry = AliCDBManager::Instance()->GetStorage(AliShuttle::GetLocalCDB())
- ->Get(Form("/SHUTTLE/STATUS/%s", fCurrentDetector.Data()), fCurrentRun);
+ fStatusEntry = AliCDBManager::Instance()->GetStorage(GetLocalCDB())
+ ->Get(Form("/SHUTTLE/STATUS/%s", fCurrentDetector.Data()), GetCurrentRun());
- if (!fStatusEntry)
- return 0;
+ if (!fStatusEntry) return 0;
+ fStatusEntry->SetOwner(1);
- TObject* anObject = fStatusEntry->GetObject();
- if (anObject == NULL || anObject->IsA() != AliShuttleStatus::Class())
- {
- AliError("Invalid object stored to CDB!");
- return 0;
- }
+ AliShuttleStatus* status = dynamic_cast<AliShuttleStatus*> (fStatusEntry->GetObject());
+ if (!status) {
+ AliError("Invalid object stored to CDB!");
+ return 0;
+ }
- AliShuttleStatus* status = dynamic_cast<AliShuttleStatus*> (anObject);
- return status;
+ return status;
}
//______________________________________________________________________________________________
Bool_t AliShuttle::WriteShuttleStatus(AliShuttleStatus* status)
{
- // writes the status for one subdetector
+// writes the status for one subdetector
+
+ if (fStatusEntry){
+ delete fStatusEntry;
+ fStatusEntry = 0;
+ }
- if (fStatusEntry)
- {
- delete fStatusEntry;
- fStatusEntry = 0;
- }
+ Int_t run = GetCurrentRun();
- AliCDBId id(AliCDBPath("SHUTTLE", "STATUS", fCurrentDetector), fCurrentRun, fCurrentRun);
+ AliCDBId id(AliCDBPath("SHUTTLE", "STATUS", fCurrentDetector), run, run);
- fStatusEntry = new AliCDBEntry(status, id, new AliCDBMetaData);
+ fStatusEntry = new AliCDBEntry(status, id, new AliCDBMetaData);
+ fStatusEntry->SetOwner(1);
- UInt_t result = AliCDBManager::Instance()->GetStorage(fgkLocalCDB)->Put(fStatusEntry);
+ UInt_t result = AliCDBManager::Instance()->GetStorage(fgkLocalCDB)->Put(fStatusEntry);
- if (!result)
- {
- AliError(Form("WriteShuttleStatus for %s, run %d failed", fCurrentDetector.Data(), fCurrentRun));
- return kFALSE;
- }
+ if (!result) {
+ AliError(Form("WriteShuttleStatus for %s, run %d failed", fCurrentDetector.Data(), run));
+ return kFALSE;
+ }
+
+ SendMLInfo();
- return kTRUE;
+ return kTRUE;
}
//______________________________________________________________________________________________
{
// changes the AliShuttleStatus for the given detector and run to the given status
- if (!fStatusEntry)
- {
- AliError("UNEXPECTED: fStatusEntry empty");
- return;
- }
+ if (!fStatusEntry){
+ AliError("UNEXPECTED: fStatusEntry empty");
+ return;
+ }
+
+ AliShuttleStatus* status = dynamic_cast<AliShuttleStatus*> (fStatusEntry->GetObject());
- TObject* anObject = fStatusEntry->GetObject();
- AliShuttleStatus* status = dynamic_cast<AliShuttleStatus*> (anObject);
+ if (!status){
+ AliError("UNEXPECTED: status could not be read from current CDB entry");
+ return;
+ }
- if (!status)
- {
- AliError("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);
- Log("SHUTTLE", Form("%s: Changing state from %s to %s", fCurrentDetector.Data(),
- status->GetStatusName(), status->GetStatusName(newStatus)));
+ status->SetStatus(newStatus);
+ if (increaseCount) status->IncreaseCount();
- status->SetStatus(newStatus);
- if (increaseCount)
- status->IncreaseCount();
+ AliCDBManager::Instance()->GetStorage(fgkLocalCDB)->Put(fStatusEntry);
- AliCDBManager::Instance()->GetStorage(fgkLocalCDB)->Put(fStatusEntry);
+ SendMLInfo();
}
//______________________________________________________________________________________________
-Bool_t AliShuttle::ContinueProcessing()
+void AliShuttle::SendMLInfo()
{
- // 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;
- }
-
- // if we get here, there is a restart
-
- // abort conditions
- if (status->GetStatus() == AliShuttleStatus::kPPStarted && status->GetCount() >= fConfig->GetMaxPPRetries() ||
- status->GetCount() >= fConfig->GetMaxRetries())
- {
- Log("SHUTTLE", Form("%s, run %d failed too often, %d times, status %s. Skipping processing.",
- fCurrentDetector.Data(), fCurrentRun, status->GetCount(), status->GetStatusName()));
-
- 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;
+ //
+ // sends ML information about the current status of the current detector being processed
+ //
+
+ AliShuttleStatus* status = dynamic_cast<AliShuttleStatus*> (fStatusEntry->GetObject());
+
+ if (!status){
+ AliError("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);
+
+ fMonaLisa->SendParameters(&mlList);
}
//______________________________________________________________________________________________
-Bool_t AliShuttle::Process(Int_t run, UInt_t startTime, UInt_t endTime)
+Bool_t AliShuttle::ContinueProcessing()
{
- //
- // 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
- //
+// 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
- AliInfo(Form("\n\n ^*^*^*^*^*^* Processing run %d ^*^*^*^*^*^*", run));
+ if (!fConfig->HostProcessDetector(fCurrentDetector)) return kFALSE;
- // Initialization
- Bool_t hasError = kFALSE;
- for(Int_t iSys=0;iSys<3;iSys++) fFESCalled[iSys]=kFALSE;
+ AliPreprocessor* aPreprocessor =
+ dynamic_cast<AliPreprocessor*> (fPreprocessorMap.GetValue(fCurrentDetector));
+ if (!aPreprocessor)
+ {
+ AliInfo(Form("%s: no preprocessor registered", fCurrentDetector.Data()));
+ return kFALSE;
+ }
- fCurrentRun = run;
- fCurrentStartTime = startTime;
- fCurrentEndTime = endTime;
+ AliShuttleLogbookEntry::Status entryStatus =
+ fLogbookEntry->GetDetectorStatus(fCurrentDetector);
- // Loop on detectors in the configuration
- TIter iter(fConfig->GetDetectors());
- TObjString* aDetector;
+ if(entryStatus != AliShuttleLogbookEntry::kUnprocessed) {
+ AliInfo(Form("ContinueProcessing - %s is %s",
+ fCurrentDetector.Data(),
+ fLogbookEntry->GetDetectorStatusName(entryStatus)));
+ return kFALSE;
+ }
- while ((aDetector = (TObjString*) iter.Next())) {
- fCurrentDetector = aDetector->String();
+ // if we get here, according to Shuttle logbook subdetector is in UNPROCESSED state
- Bool_t detectorError=kFALSE;
- if (!fConfig->HostProcessDetector(fCurrentDetector)) continue;
+ // check if current run is first unprocessed run for current detector
+ if (fConfig->StrictRunOrder(fCurrentDetector) &&
+ !fFirstUnprocessed[GetDetPos(fCurrentDetector)])
+ {
+ Log("SHUTTLE", Form("ContinueProcessing - %s requires strict run ordering but this is not the first unprocessed run!"));
+ return kFALSE;
+ }
- if (ContinueProcessing() == kFALSE) continue;
+ AliShuttleStatus* status = ReadShuttleStatus();
+ if (!status) {
+ // first time
+ Log("SHUTTLE", Form("ContinueProcessing - %s: Processing first time",
+ fCurrentDetector.Data()));
+ status = new AliShuttleStatus(AliShuttleStatus::kStarted);
+ return WriteShuttleStatus(status);
+ }
- if(!Process()) {
- hasError = kTRUE;
- detectorError=kTRUE;
- continue;
- }
- AliInfo(Form("Process ended successfully for detector %s!",aDetector->GetName()));
+ // The following two cases 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){
+ Log("SHUTTLE", Form("ContinueProcessing - %s is already %s. Updating Shuttle Logbook",
+ fCurrentDetector.Data(),
+ status->GetStatusName(status->GetStatus())));
+ UpdateShuttleLogbook(fCurrentDetector.Data(),
+ status->GetStatusName(status->GetStatus()));
+ return kFALSE;
+ }
- // Process successful: Update time_processed field in FES logbooks!
- if(fFESCalled[kDAQ]) {
- hasError = (UpdateDAQTable() == kFALSE);
- fFESlist[kDAQ].Clear();
+ if (status->GetStatus() == AliShuttleStatus::kStoreFailed) {
+ Log("SHUTTLE",
+ Form("ContinueProcessing - %s: Grid storage of one or more objects failed. Trying again now",
+ fCurrentDetector.Data()));
+ if(TryToStoreAgain()){
+ Log(fCurrentDetector.Data(), "ContinueProcessing - All objects successfully stored into OCDB");
+ UpdateShuttleStatus(AliShuttleStatus::kDone);
+ UpdateShuttleLogbook(fCurrentDetector.Data(), "DONE");
+ } else {
+ Log("SHUTTLE",
+ Form("ContinueProcessing - %s: Grid storage failed again",
+ fCurrentDetector.Data()));
+ // trigger ML information manually because we do not had a status change
+ SendMLInfo();
}
- //if(fFESCalled[kDCS]) {
- // hasError = UpdateDCSTable(aDetector->GetName());
- // fFESlist[kDCS].Clear();
- //}
- //if(fFESCalled[kHLT]) {
- // hasError = UpdateHLTTable(aDetector->GetName());
- // fFESlist[kHLT].Clear();
- //}
+ return kFALSE;
+ }
- UpdateShuttleStatus(AliShuttleStatus::kDone);
+ // 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);
+ } else {
+ Log("SHUTTLE", Form("ContinueProcessing - %s: restarting. "
+ "Aborted before with %s. Retry number %d.", fCurrentDetector.Data(),
+ status->GetStatusName(), status->GetCount()));
+ UpdateShuttleStatus(AliShuttleStatus::kStarted, kTRUE);
+ cont = kTRUE;
}
- fCurrentRun = -1;
- fCurrentStartTime = 0;
- fCurrentEndTime = 0;
+ // Send mail to detector expert!
+ AliInfo(Form("Sending mail to %s expert...", fCurrentDetector.Data()));
+ if (!SendMail())
+ Log("SHUTTLE", Form("ContinueProcessing - Could not send mail to %s expert",
+ fCurrentDetector.Data()));
- return hasError == kFALSE;
+ return cont;
}
//______________________________________________________________________________________________
-Bool_t AliShuttle::Process()
+Bool_t AliShuttle::Process(AliShuttleLogbookEntry* entry)
{
//
- // 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
+ // 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
//
- AliInfo(Form("Retrieving values for %s, run %d", fCurrentDetector.Data(), fCurrentRun));
+ if(!entry) return kFALSE;
- if (!fConfig->HasDetector(fCurrentDetector)) {
- Log(fCurrentDetector, "There isn't any configuration for %s !");
- UpdateShuttleStatus(AliShuttleStatus::kFailed);
- return kFALSE;
+ fLogbookEntry = entry;
+
+ if (fLogbookEntry->IsDone())
+ {
+ Log("SHUTTLE","Process - Shuttle is already DONE. Updating logbook");
+ UpdateShuttleLogbook("shuttle_done");
+ fLogbookEntry = 0;
+ return kTRUE;
}
- UpdateShuttleStatus(AliShuttleStatus::kDCSStarted);
+ // create ML instance that monitors this run
+ fMonaLisa = new TMonaLisaWriter(Form("%d", GetCurrentRun()), "SHUTTLE", "aliendb1.cern.ch");
+ // disable monitoring of other parameters that come e.g. from TFile
+ gMonitoringWriter = 0;
+
+ AliInfo(Form("\n\n \t\t\t^*^*^*^*^*^*^*^*^*^*^*^* run %d: START ^*^*^*^*^*^*^*^*^*^*^*^* \n",
+ GetCurrentRun()));
- TString host(fConfig->GetDCSHost(fCurrentDetector));
- Int_t port = fConfig->GetDCSPort(fCurrentDetector);
+ // Set run type from run type logbook into current fLogbookEntry
+ SetRunType();
- TIter iter(fConfig->GetDCSAliases(fCurrentDetector));
- TObjString* anAlias;
- TMap aliasMap;
+ // Send the information to ML
+ TMonaLisaText mlStatus("SHUTTLE_status", "Processing");
+ TList mlList;
+ mlList.Add(&mlStatus);
+
+ fMonaLisa->SendParameters(&mlList);
+
+ fLogbookEntry->Print("all");
+
+ // Initialization
Bool_t hasError = kFALSE;
- Bool_t result=kFALSE;
-
- 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...");
- result = GetValueSet(host, port, anAlias->String(), valueSet);
- } else {
- AliInfo(Form("Skipping DCS processing. Port = %d",port));
- result = kTRUE;
+
+ AliCDBStorage *mainCDBSto = AliCDBManager::Instance()->GetStorage(fgkMainCDB);
+ if(mainCDBSto) mainCDBSto->QueryCDB(GetCurrentRun());
+ AliCDBStorage *mainRefSto = AliCDBManager::Instance()->GetStorage(fgkMainRefStorage);
+ if(mainRefSto) mainRefSto->QueryCDB(GetCurrentRun());
+
+ // Loop on detectors in the configuration
+ TIter iter(fConfig->GetDetectors());
+ TObjString* aDetector = 0;
+
+ while ((aDetector = (TObjString*) iter.Next()))
+ {
+ fCurrentDetector = aDetector->String();
+
+ if (ContinueProcessing() == kFALSE) continue;
+
+ AliInfo(Form("\n\n \t\t\t****** run %d - %s: START ******",
+ GetCurrentRun(), aDetector->GetName()));
+
+ for(Int_t iSys=0;iSys<3;iSys++) fFXSCalled[iSys]=kFALSE;
+
+ Log(fCurrentDetector.Data(), "Starting processing");
+
+ Int_t pid = fork();
+
+ if (pid < 0)
+ {
+ Log("SHUTTLE", "ERROR: Forking failed");
}
- if(result) {
- aliasMap.Add(anAlias->Clone(), valueSet.Clone());
- }else{
- TString message = Form("Error while retrieving alias %s !",
- anAlias->GetName());
- Log(fCurrentDetector, message.Data());
- hasError = kTRUE;
- break;
+ else if (pid > 0)
+ {
+ // parent
+ AliInfo(Form("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())
+ {
+ Log("SHUTTLE", Form("Process time out. Run time: %d seconds. Killing...",
+ expiredTime));
+
+ kill(pid, 9);
+
+ hasError = kTRUE;
+
+ gSystem->Sleep(1000);
+ }
+ else
+ {
+ if (expiredTime % 60 == 0)
+ Log("SHUTTLE", Form("Checked process. Run time: %d seconds.",
+ expiredTime));
+ gSystem->Sleep(1000);
+ }
+ }
+
+ AliInfo(Form("In parent process of %d - %s: Client has terminated.",
+ GetCurrentRun(), aDetector->GetName()));
+
+ if (WIFEXITED(status))
+ {
+ Int_t returnCode = WEXITSTATUS(status);
+
+ Log("SHUTTLE", Form("The return code is %d", returnCode));
+
+ if (returnCode != 0)
+ hasError = kTRUE;
+ }
+ }
+ else if (pid == 0)
+ {
+ // client
+ AliInfo(Form("In client process of %d - %s", GetCurrentRun(), aDetector->GetName()));
+
+ UInt_t result = ProcessCurrentDetector();
+
+ Int_t returnCode = 0; // will be set to 1 in case of an error
+
+ if (!result)
+ {
+ returnCode = 1;
+ AliInfo(Form("\n \t\t\t****** run %d - %s: PREPROCESSOR ERROR ****** \n\n",
+ GetCurrentRun(), aDetector->GetName()));
+ }
+ else if (result == 2)
+ {
+ AliInfo(Form("\n \t\t\t****** run %d - %s: STORAGE ERROR ****** \n\n",
+ GetCurrentRun(), aDetector->GetName()));
+ } else
+ {
+ AliInfo(Form("\n \t\t\t****** run %d - %s: DONE ****** \n\n",
+ GetCurrentRun(), aDetector->GetName()));
+ }
+
+ if (result > 0)
+ {
+ // Process successful: Update time_processed field in FXS logbooks!
+ if (UpdateTable() == kFALSE) returnCode = 1;
+ }
+
+ for (UInt_t iSys=0; iSys<3; iSys++)
+ {
+ if (fFXSCalled[iSys]) fFXSlist[iSys].Clear();
+ }
+
+ AliInfo(Form("Client process of %d - %s is exiting now with %d.",
+ GetCurrentRun(), aDetector->GetName(), returnCode));
+
+ // the client exits here
+ gSystem->Exit(returnCode);
+
+ AliError("We should never get here!!!");
}
}
- if (hasError)
- {
- UpdateShuttleStatus(AliShuttleStatus::kDCSError);
- return kFALSE;
+ AliInfo(Form("\n\n \t\t\t^*^*^*^*^*^*^*^*^*^*^*^* run %d: FINISH ^*^*^*^*^*^*^*^*^*^*^*^* \n",
+ 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;
}
- UpdateShuttleStatus(AliShuttleStatus::kPPStarted);
+ AliShuttleLogbookEntry* checkEntry = dynamic_cast<AliShuttleLogbookEntry*>
+ (checkEntryArray.At(0));
- AliPreprocessor* aPreprocessor =
- dynamic_cast<AliPreprocessor*> (fPreprocessorMap.GetValue(fCurrentDetector));
- if(aPreprocessor)
+ if (checkEntry)
{
- aPreprocessor->Initialize(fCurrentRun, fCurrentStartTime, fCurrentEndTime);
-
- // TODO Think about what to do in case of "Grid storage error"
- // (-> object put in local backup storage, return 2)
- hasError = (aPreprocessor->Process(&aliasMap) == 0);
- }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");
- hasError = (Store(path, &aliasMap, &metaData) == 0);
- }
-
- if (hasError)
- UpdateShuttleStatus(AliShuttleStatus::kPPError);
- else
- UpdateShuttleStatus(AliShuttleStatus::kPPDone);
-
- aliasMap.Delete();
+ 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;
+ }
+ }
+ }
+ }
+
+ // remove ML instance
+ delete fMonaLisa;
+ fMonaLisa = 0;
+
+ fLogbookEntry = 0;
return hasError == kFALSE;
}
//______________________________________________________________________________________________
-Bool_t AliShuttle::GetValueSet(const char* host, Int_t port, const char* alias,
- TObjArray& valueSet)
+UInt_t AliShuttle::ProcessCurrentDetector()
{
-// 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
+ //
+ // Makes data retrieval just for a specific detector (fCurrentDetector).
+ // Threre should be a configuration for this detector.
- AliDCSClient client(host, port, fTimeout, fRetries);
- if (!client.IsConnected()) {
- return kFALSE;
- }
+ AliInfo(Form("Retrieving values for %s, run %d", fCurrentDetector.Data(), GetCurrentRun()));
+
+ UpdateShuttleStatus(AliShuttleStatus::kDCSStarted);
- Int_t result = client.GetAliasValues(alias,
- GetCurrentStartTime(), GetCurrentEndTime(), valueSet);
+ TMap dcsMap;
+ dcsMap.SetOwner(1);
- if (result < 0) {
- AliError(Form("Can't get '%s'! Reason: %s",
- alias, AliDCSClient::GetErrorString(result)));
+ Bool_t aDCSError = kFALSE;
+ fGridError = kFALSE;
- if (result == AliDCSClient::fgkServerError) {
- AliError(Form("Server error: %s",
- client.GetServerError().Data()));
+ // TODO Test only... I've added a flag that allows to
+ // exclude DCS archive DB query
+ if (!fgkProcessDCS)
+ {
+ AliInfo("Skipping DCS processing!");
+ aDCSError = kFALSE;
+ } else {
+ TString host(fConfig->GetDCSHost(fCurrentDetector));
+ Int_t port = fConfig->GetDCSPort(fCurrentDetector);
+
+ // Retrieval of Aliases
+ TObjString* anAlias = 0;
+ Int_t iAlias = 1;
+ Int_t nTotAliases= ((TMap*)fConfig->GetDCSAliases(fCurrentDetector))->GetEntries();
+ TIter iterAliases(fConfig->GetDCSAliases(fCurrentDetector));
+ while ((anAlias = (TObjString*) iterAliases.Next()))
+ {
+ TObjArray *valueSet = new TObjArray();
+ valueSet->SetOwner(1);
+
+ if (((iAlias-1) % 500) == 0 || iAlias == nTotAliases)
+ AliInfo(Form("Querying DCS archive: alias %s (%d of %d)",
+ anAlias->GetName(), iAlias++, nTotAliases));
+ aDCSError = (GetValueSet(host, port, anAlias->String(), valueSet, kAlias) == 0);
+
+ if(!aDCSError)
+ {
+ dcsMap.Add(anAlias->Clone(), valueSet);
+ } else {
+ Log(fCurrentDetector,
+ Form("ProcessCurrentDetector - Error while retrieving alias %s",
+ anAlias->GetName()));
+ UpdateShuttleStatus(AliShuttleStatus::kDCSError);
+ dcsMap.DeleteAll();
+ return 0;
+ }
}
- return kFALSE;
+ // Retrieval of Data Points
+ TObjString* aDP = 0;
+ Int_t iDP = 0;
+ Int_t nTotDPs= ((TMap*)fConfig->GetDCSDataPoints(fCurrentDetector))->GetEntries();
+ TIter iterDP(fConfig->GetDCSDataPoints(fCurrentDetector));
+ while ((aDP = (TObjString*) iterDP.Next()))
+ {
+ TObjArray *valueSet = new TObjArray();
+ valueSet->SetOwner(1);
+ if (((iDP-1) % 500) == 0 || iDP == nTotDPs)
+ AliInfo(Form("Querying DCS archive: DP %s (%d of %d)",
+ aDP->GetName(), iDP++, nTotDPs));
+ aDCSError = (GetValueSet(host, port, aDP->String(), valueSet, kDP) == 0);
+
+ if(!aDCSError)
+ {
+ dcsMap.Add(aDP->Clone(), valueSet);
+ } else {
+ Log(fCurrentDetector,
+ Form("ProcessCurrentDetector - Error while retrieving data point %s",
+ aDP->GetName()));
+ UpdateShuttleStatus(AliShuttleStatus::kDCSError);
+ dcsMap.DeleteAll();
+ return 0;
+ }
+ }
}
- return kTRUE;
-}
+ // DCS Archive DB processing successful. Call Preprocessor!
+ UpdateShuttleStatus(AliShuttleStatus::kPPStarted);
-//______________________________________________________________________________________________
-const char* AliShuttle::GetFile(Int_t system, const char* detector,
- const char* id, const char* source)
-{
-// Get calibration file from file exchange servers
-// calls specific getter according to system index (kDAQ, kDCS, kHLT)
+ AliPreprocessor* aPreprocessor =
+ dynamic_cast<AliPreprocessor*> (fPreprocessorMap.GetValue(fCurrentDetector));
- 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));
+ aPreprocessor->Initialize(GetCurrentRun(), GetCurrentStartTime(), GetCurrentEndTime());
+ UInt_t aPPResult = aPreprocessor->Process(&dcsMap);
+
+ UInt_t returnValue = 0;
+ if (aPPResult == 0) { // Preprocessor error
+ UpdateShuttleStatus(AliShuttleStatus::kPPError);
+ returnValue = 0;
+ } else if (fGridError == kFALSE) { // process and Grid storage ok!
+ UpdateShuttleStatus(AliShuttleStatus::kDone);
+ UpdateShuttleLogbook(fCurrentDetector, "DONE");
+ Log(fCurrentDetector.Data(),
+ "ProcessCurrentDetector - Preprocessor and Grid storage ended successfully");
+ returnValue = 1;
+ } else { // Grid storage error (process ok, but object put in local storage)
+ UpdateShuttleStatus(AliShuttleStatus::kStoreFailed);
+ returnValue = 2;
}
- return 0;
+ dcsMap.DeleteAll();
+
+ return returnValue;
}
//______________________________________________________________________________________________
-TList* AliShuttle::GetFileSources(Int_t system, const char* detector, const char* id)
+Bool_t AliShuttle::QueryShuttleLogbook(const char* whereClause,
+ TObjArray& entries)
{
-// Get sources producing the condition file Id from file exchange servers
-// calls specific getter according to system index (kDAQ, kDCS, kHLT)
+// Query DAQ's Shuttle logbook and fills detector status object.
+// Call QueryRunParameters to query DAQ logbook for run parameters.
- 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));
- }
+ entries.SetOwner(1);
- return NULL;
-}
+ // check connection, in case connect
+ if(!Connect(3)) return kFALSE;
-//______________________________________________________________________________________________
-Bool_t AliShuttle::Connect(Int_t system){
-// Connect to MySQL Server of the system's FES logbook
+ TString sqlQuery;
+ sqlQuery = Form("select * from %s %s order by run", fConfig->GetShuttlelbTable(), whereClause);
- // check connection: if already connected return
- if(fServer[system] && fServer[system]->IsConnected()) return kTRUE;
+ TSQLResult* aResult = fServer[3]->Query(sqlQuery);
+ if (!aResult) {
+ AliError(Form("Can't execute query <%s>!", sqlQuery.Data()));
+ return kFALSE;
+ }
- TString aFESlbHost= Form("mysql://%s", fConfig->GetFESlbHost(system));
+ AliDebug(2,Form("Query = %s", sqlQuery.Data()));
- fServer[system] = TSQLServer::Connect(aFESlbHost,
- fConfig->GetFESlbUser(system),
- fConfig->GetFESlbPass(system));
- if (!fServer[system] || !fServer[system]->IsConnected()) {
- AliError(Form("Can't establish connection to FES logbook for %s !",fkSystemNames[system]));
+ if(aResult->GetRowCount() == 0) {
+// if(sqlQuery.EndsWith("where shuttle_done=0 order by run")){
+// Log("SHUTTLE", "QueryShuttleLogbook - All runs in Shuttle Logbook are already DONE");
+// delete aResult;
+// return kTRUE;
+// } else {
+ AliInfo("No entries in Shuttle Logbook match request");
+ delete aResult;
+ return kTRUE;
+// }
+ }
+
+ // TODO Check field count!
+ const UInt_t nCols = 22;
+ if (aResult->GetFieldCount() != (Int_t) nCols) {
+ AliError("Invalid SQL result field number!");
+ delete aResult;
return kFALSE;
}
- // Get tables
- // TODO in the configuration should the table name be there too?
- switch(system){
- case kDAQ:
- fServer[kDAQ]->GetTables("REFSYSLOG");
- break;
- case kDCS:
- //fServer[kDCS]->GetTables("REFSYSLOG");
- break;
- case kHLT:
- //fServer[kHLT]->GetTables("REFSYSLOG");
- break;
- default:
- break;
+ 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;
}
+// if(sqlQuery.EndsWith("where shuttle_done=0 order by run"))
+// Log("SHUTTLE", Form("QueryShuttleLogbook - Found %d unprocessed runs in Shuttle Logbook",
+// entries.GetEntriesFast()));
+ 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
+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(kDAQ)){
- Log(detector, "GetDAQFileName: Couldn't connect to DAQ Logbook !");
+ if (!Connect(3))
return 0;
- }
- // 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;
+ sqlQuery.Form("select * from %s where run=%d", fConfig->GetDAQlbTable(), run);
+
+ TSQLResult* aResult = fServer[3]->Query(sqlQuery);
+ if (!aResult) {
+ AliError(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) {
+ AliError(Form("More than one entry in DAQ Logbook for run %d. Skipping", run));
+ delete aResult;
+ return 0;
+ }
+
+ TSQLRow* aRow = aResult->Next();
+ if (!aRow)
+ {
+ AliError(Form("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));
+
+ UInt_t startTime = entry->GetStartTime();
+ UInt_t endTime = entry->GetEndTime();
+
+ if (!startTime || !endTime || startTime > endTime) {
+ Log("SHUTTLE",
+ Form("QueryRunParameters - Invalid parameters for Run %d: startTime = %d, endTime = %d",
+ run, startTime, endTime));
+ delete entry;
+ delete aRow;
+ delete aResult;
+ return 0;
+ }
+
+ delete aRow;
+ delete aResult;
+
+ return entry;
+}
+
+//______________________________________________________________________________________________
+Bool_t AliShuttle::TryToStoreAgain()
+{
+ // Called in case the detector failed to store the object in Grid OCDB
+ // It tries to store the object again, if it does not find more recent and overlapping objects
+ // Calls underlying TryToStoreAgain(const char*) function twice, for OCDB and Reference storage.
+
+ AliInfo("Trying to store OCDB data again...");
+ Bool_t resultCDB = TryToStoreAgain(fgkMainCDB);
+
+ AliInfo("Trying to store reference data again...");
+ Bool_t resultRef = TryToStoreAgain(fgkMainRefStorage);
+
+ return resultCDB && resultRef;
+}
+
+//______________________________________________________________________________________________
+Bool_t AliShuttle::TryToStoreAgain(TString& gridURI)
+{
+ // Called by TryToStoreAgain(), performs actual storage retry
+
+ TObjArray* gridIds=0;
+
+ Bool_t result = kTRUE;
+
+ const char* type = 0;
+ TString backupURI;
+ if(gridURI == fgkMainCDB) {
+ type = "OCDB";
+ backupURI = fgkLocalCDB;
+ } else if(gridURI == fgkMainRefStorage) {
+ type = "reference";
+ backupURI = fgkLocalRefStorage;
+ } else {
+ AliError(Form("Invalid storage URI: %s", gridURI.Data()));
+ return kFALSE;
+ }
+
+ AliCDBManager* man = AliCDBManager::Instance();
+
+ AliCDBStorage *gridSto = man->GetStorage(gridURI);
+ if(!gridSto) {
+ Log(fCurrentDetector.Data(),
+ Form("TryToStoreAgain - cannot activate main %s storage", type));
+ return kFALSE;
+ }
+
+ gridIds = gridSto->GetQueryCDBList();
+
+ // get objects previously stored in local CDB
+ AliCDBStorage *backupSto = man->GetStorage(backupURI);
+ AliCDBPath aPath(GetOfflineDetName(fCurrentDetector.Data()),"*","*");
+ // Local objects were stored with current run as Grid version!
+ TList* localEntries = backupSto->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);
+
+ // loop on Grid valid Id's
+ Bool_t store = kTRUE;
+ TIter gridIter(gridIds);
+ AliCDBId* aGridId = 0;
+ while((aGridId = dynamic_cast<AliCDBId*> (gridIter.Next()))){
+ // If local object is valid up to infinity we store it only if it is
+ // the first unprocessed run!
+ if (aLocId.GetLastRun() == AliCDBRunRange::Infinity())
+ {
+ if (!fFirstUnprocessed[GetDetPos(fCurrentDetector)])
+ {
+ Log(fCurrentDetector.Data(),
+ ("TryToStoreAgain - This object has validity infinite but "
+ "there are previous unprocessed runs!"));
+ continue;
+ } else {
+ break;
+ }
+ }
+ 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;
+ }
+
+ if(!store){
+ Log(fCurrentDetector.Data(),
+ Form("TryToStoreAgain - A more recent object already exists in %s storage: <%s>",
+ type, aGridId->ToString().Data()));
+ // removing local filename...
+ // TODO maybe it's better not to remove it, it was not copied to the Grid!
+ TString filename;
+ backupSto->IdToFilename(aLocId, filename);
+ AliInfo(Form("Removing local file %s", filename.Data()));
+ gSystem->Exec(Form("rm %s",filename.Data()));
+ continue;
+ }
+
+ // If we get here, the file can be stored!
+ Bool_t storeOk = gridSto->Put(aLocEntry);
+ if(storeOk){
+ Log(fCurrentDetector.Data(),
+ Form("TryToStoreAgain - Object <%s> successfully put into %s storage",
+ aLocId.ToString().Data(), type));
+
+ // removing local filename...
+ TString filename;
+ backupSto->IdToFilename(aLocId, filename);
+ AliInfo(Form("Removing local file %s", filename.Data()));
+ gSystem->Exec(Form("rm %s", filename.Data()));
+ continue;
+ } else {
+ Log(fCurrentDetector.Data(),
+ Form("TryToStoreAgain - Grid %s storage of object <%s> failed again",
+ type, aLocId.ToString().Data()));
+ result = kFALSE;
+ }
+ }
+ localEntries->Clear();
+
+ return result;
+}
+
+//______________________________________________________________________________________________
+Bool_t AliShuttle::GetValueSet(const char* host, Int_t port, const char* entry,
+ TObjArray* valueSet, DCSType type)
+{
+// Retrieve all "entry" data points from the DCS server
+// host, port: TSocket connection parameters
+// entry: name of the alias or data point
+// valueSet: array of retrieved AliDCSValue's
+// type: kAlias or kDP
+
+ AliDCSClient client(host, port, fTimeout, fRetries);
+ if (!client.IsConnected())
+ {
+ return kFALSE;
+ }
+
+ Int_t result=0;
+
+ if (type == kAlias)
+ {
+ result = client.GetAliasValues(entry,
+ GetCurrentStartTime(), GetCurrentEndTime(), valueSet);
+ } else
+ if (type == kDP)
+ {
+ result = client.GetDPValues(entry,
+ GetCurrentStartTime(), GetCurrentEndTime(), valueSet);
+ }
+
+ if (result < 0)
+ {
+ Log(fCurrentDetector.Data(), Form("GetValueSet - Can't get '%s'! Reason: %s",
+ entry, AliDCSClient::GetErrorString(result)));
+
+ if (result == AliDCSClient::fgkServerError)
+ {
+ Log(fCurrentDetector.Data(), Form("GetValueSet - Server error: %s",
+ client.GetServerError().Data()));
+ }
+
+ return kFALSE;
+ }
+
+ return kTRUE;
+}
+
+//______________________________________________________________________________________________
+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 connection, in case connect
+ if (!Connect(system))
+ {
+ Log(detector, Form("GetFile - Couldn't connect to %s FXS database", GetSystemName(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);
+ nFields = 3;
+ }
+
TString sqlQuery = Form("%s %s", sqlQueryStart.Data(), whereClause.Data());
AliDebug(2, Form("SQL query: \n%s",sqlQuery.Data()));
// Query execution
- TSQLResult* aResult;
- aResult = fServer[kDAQ]->Query(sqlQuery);
+ TSQLResult* aResult = 0;
+ aResult = dynamic_cast<TSQLResult*> (fServer[system]->Query(sqlQuery));
if (!aResult) {
- Log(detector, Form("Can't execute query <%s>!", sqlQuery.Data()));
+ Log(detector, Form("GetFileName - Can't execute SQL query to %s database for: id = %s, source = %s",
+ GetSystemName(system), id, sourceName.Data()));
return 0;
}
- if (aResult->GetRowCount() == 0) {
+ if(aResult->GetRowCount() == 0)
+ {
Log(detector,
- Form("GetDAQFileName: No result from SQL query <%s>!", sqlQuery.Data()));
+ Form("GetFileName - No entry in %s FXS db for: id = %s, source = %s",
+ GetSystemName(system), id, sourceName.Data()));
delete aResult;
return 0;
}
- if (aResult->GetRowCount() >1) {
+ if (aResult->GetRowCount() > 1) {
Log(detector,
- Form("GetDAQFileName: More than one row resulting from SQL query <%s>!", sqlQuery.Data()));
+ Form("GetFileName - More than one entry in %s FXS db for: id = %s, source = %s",
+ GetSystemName(system), id, sourceName.Data()));
delete aResult;
return 0;
}
- TSQLRow* aRow = aResult->Next();
+ 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()));
+ delete aResult;
+ return 0;
+ }
- if(!aRow){
- Log(detector, Form("GetDAQFileName: Empty set result from query <%s>!", sqlQuery.Data()));
+ TSQLRow* aRow = dynamic_cast<TSQLRow*> (aResult->Next());
+
+ if (!aRow){
+ Log(detector, Form("GetFileName - Empty set result in %s FXS db from query: id = %s, source = %s",
+ GetSystemName(system), id, sourceName.Data()));
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",filePath.Data()));
+ 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_%d_%s_%s.shuttle",
- detector, fCurrentRun, id, source);
+ TString localFileName = Form("%s_%s_%d_%s_%s.shuttle",
+ GetSystemName(system), detector, GetCurrentRun(), id, sourceName.Data());
- // 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()));
+
+ // 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("GetFileName - Copy of file %s from %s FXS failed",
+ filePath.Data(), GetSystemName(system)));
+ continue;
+ } else {
+ AliInfo(Form("File %s copied from %s FXS into %s/%s",
+ filePath.Data(), GetSystemName(system),
+ GetShuttleTempDir(), localFileName.Data()));
+ }
+
+ if (fileChecksum.Length()>0)
+ {
+ // compare md5sum of local file with the one stored in the FXS DB
+ Int_t md5Comp = gSystem->Exec(Form("md5sum %s/%s |grep %s 2>&1 > /dev/null",
+ GetShuttleTempDir(), localFileName.Data(), fileChecksum.Data()));
+
+ if (md5Comp != 0)
+ {
+ Log(detector, Form("GetFileName - 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) return 0;
- fFESCalled[kDAQ]=kTRUE;
- TObjString *fileParams = new TObjString(Form("%s_!?!_%s", id, source));
- fFESlist[kDAQ].Add(fileParams);
+ fFXSCalled[system]=kTRUE;
+ TObjString *fileParams = new TObjString(Form("%s#!?!#%s", id, sourceName.Data()));
+ fFXSlist[system].Add(fileParams);
- return localFileName.Data();
+ static TString fullLocalFileName;
+ fullLocalFileName = TString::Format("%s/%s", GetShuttleTempDir(), localFileName.Data());
+
+ AliInfo(Form("fullLocalFileName = %s", fullLocalFileName.Data()));
+
+ return fullLocalFileName.Data();
}
//______________________________________________________________________________________________
-Bool_t AliShuttle::RetrieveDAQFile(const char* daqFileName, const char* localFileName){
+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 DAQ FES into folder %s and rename it as %s",
- daqFileName,fgkShuttleTempDir, localFileName));
+ AliDebug(2, Form("Copy file %s from %s FXS into %s/%s",
+ GetSystemName(system), fxsFileName, GetShuttleTempDir(), localFileName));
- void* dir = gSystem->OpenDirectory(fgkShuttleTempDir);
+ void* dir = gSystem->OpenDirectory(GetShuttleTempDir());
if (dir == NULL) {
- if (gSystem->mkdir(fgkShuttleTempDir, kTRUE)) {
- AliError(Form("Can't open directory <%s>!", fgkShuttleTempDir));
+ if (gSystem->mkdir(GetShuttleTempDir(), kTRUE)) {
+ AliError(Form("Can't open directory <%s>", GetShuttleTempDir()));
return kFALSE;
}
gSystem->FreeDirectory(dir);
}
- TString baseDAQFESFolder = "DAQ";
- TString command = Form("scp %s@%s:%s/%s %s/%s",
- fConfig->GetFESUser(kDAQ),
- fConfig->GetFESHost(kDAQ),
- baseDAQFESFolder.Data(),
- daqFileName,
- fgkShuttleTempDir,
- localFileName);
+ TString baseFXSFolder;
+ if (system == kDAQ)
+ {
+ baseFXSFolder = "FES/";
+ }
+ else if (system == kDCS)
+ {
+ baseFXSFolder = "";
+ }
+ else if (system == kHLT)
+ {
+ baseFXSFolder = "~/";
+ }
- AliDebug(2, Form("%s",command.Data()));
- UInt_t nRetries = 0;
- UInt_t maxRetries = 3;
+ TString command = Form("scp -oPort=%d -2 %s@%s:%s%s %s/%s",
+ fConfig->GetFXSPort(system),
+ fConfig->GetFXSUser(system),
+ fConfig->GetFXSHost(system),
+ baseFXSFolder.Data(),
+ fxsFileName,
+ GetShuttleTempDir(),
+ localFileName);
- // 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;
- }
+ AliDebug(2, Form("%s",command.Data()));
- return kFALSE;
+ Bool_t result = (gSystem->Exec(command.Data()) == 0);
+ return result;
}
//______________________________________________________________________________________________
-TList* AliShuttle::GetDAQFileSources(const char* detector, const char* id){
-// Retrieves a file from the DCS FES.
+TList* AliShuttle::GetFileSources(Int_t system, const char* detector, const char* id)
+{
+// Get sources producing the condition file Id from file exchange servers
+
+ if (system == kDCS)
+ {
+ AliError("DCS system has only one source of data!");
+ return NULL;
+
+ }
// check connection, in case connect
- if(!Connect(kDAQ)){
- Log(detector, "GetDAQFileName: Couldn't connect to DAQ Logbook !");
- return 0;
+ if (!Connect(system))
+ {
+ Log(detector, Form("GetFile - Couldn't connect to %s FXS database", GetSystemName(system)));
+ return NULL;
}
- // Query preparation
- TString sqlQueryStart = "select DAQsource from logbook_fs where";
+ TString sourceName = 0;
+ if (system == kDAQ)
+ {
+ sourceName = "DAQsource";
+ } else if (system == kHLT)
+ {
+ sourceName = "DDLnumbers";
+ }
+
+ TString sqlQueryStart = Form("select %s from %s where", sourceName.Data(), fConfig->GetFXSdbTable(system));
TString whereClause = Form("run=%d and detector=\"%s\" and fileId=\"%s\"",
- fCurrentRun, GetDetCode(detector), id);
+ GetCurrentRun(), detector, 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[kDAQ]->Query(sqlQuery);
+ aResult = fServer[system]->Query(sqlQuery);
if (!aResult) {
- Log(detector, Form("GetDAQFileSources: Can't execute query <%s>!", sqlQuery.Data()));
+ Log(detector, Form("GetFileSources - Can't execute SQL query to %s database for id: %s",
+ GetSystemName(system), id));
return 0;
}
- if (aResult->GetRowCount() == 0) {
+ if (aResult->GetRowCount() == 0)
+ {
Log(detector,
- Form("GetDAQFileSources: No result from SQL query <%s>!", sqlQuery.Data()));
+ Form("GetFileSources - No entry in %s FXS table for id: %s", GetSystemName(system), id));
delete aResult;
return 0;
}
TList *list = new TList();
list->SetOwner(1);
- while((aRow = aResult->Next())){
+ while ((aRow = aResult->Next()))
+ {
- TString daqSource(aRow->GetField(0), aRow->GetFieldLength(0));
- AliDebug(2, Form("daqSource = %s", daqSource.Data()));
- list->Add(new TObjString(daqSource));
+ 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;
+}
+
+//______________________________________________________________________________________________
+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 dbHost, dbUser, dbPass, dbName;
+
+ 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()) {
+ 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
+ TSQLResult* aResult=0;
+ switch(system){
+ case kDAQ:
+ aResult = fServer[kDAQ]->GetTables(dbName.Data());
+ break;
+ case kDCS:
+ aResult = fServer[kDCS]->GetTables(dbName.Data());
+ break;
+ case kHLT:
+ 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::UpdateDAQTable(){
-// Update DAQ table filling time_processed field in all rows corresponding to current run and detector
+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(kDAQ)){
- Log(fCurrentDetector, "UpdateDAQTable: Couldn't connect to DAQ Logbook !");
+ if(!Connect(3)){
+ Log("SHUTTLE", "UpdateShuttleLogbook - Couldn't connect to DAQ Logbook.");
return kFALSE;
}
- TTimeStamp now; // now
+ TString detName(detector);
+ TString setClause;
+ if(detName == "shuttle_done")
+ {
+ setClause = "set shuttle_done=1";
+
+ // Send the information to ML
+ TMonaLisaText mlStatus("SHUTTLE_status", "Done");
- // 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;
+ TList mlList;
+ mlList.Add(&mlStatus);
+
+ fMonaLisa->SendParameters(&mlList);
+ } 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;
}
- 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;
+ TString whereClause = Form("where run=%d", GetCurrentRun());
- TString sqlQuery = Form("update logbook_fs set time_processed=%d %s", now.GetSec(), whereClause.Data());
+ TString sqlQuery = Form("update %s %s %s",
+ fConfig->GetShuttlelbTable(), setClause.Data(), whereClause.Data());
- AliDebug(2, Form("SQL query: \n%s",sqlQuery.Data()));
+ AliDebug(2, Form("SQL query: \n%s",sqlQuery.Data()));
- // 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;
+ // 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;
- // 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()));
+ return kTRUE;
+}
- aResult = dynamic_cast<TSQLResult*> (fServer[kDAQ]->Query(sqlQuery));
- if (!aResult) {
- AliWarning("Can't check result!");
- continue;
+//______________________________________________________________________________________________
+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;
+}
+
+//______________________________________________________________________________________________
+void AliShuttle::Log(const char* detector, const char* message)
+{
+// Fill log string with a message
+
+ void* dir = gSystem->OpenDirectory(GetShuttleLogDir());
+ if (dir == NULL) {
+ if (gSystem->mkdir(GetShuttleLogDir(), kTRUE)) {
+ AliError(Form("Can't open directory <%s>", GetShuttleLogDir()));
+ return;
}
- if (aResult->GetRowCount() == 0) {
- Log(fCurrentDetector,
- Form("GetDAQFileName: No result from SQL query <%s>!", sqlQuery.Data()));
- delete aResult;
- //return 0;
+ } else {
+ gSystem->FreeDirectory(dir);
}
- if (aResult->GetRowCount() >1) {
- Log(fCurrentDetector,
- Form("GetDAQFileName: More than one row resulting from SQL query <%s>!", sqlQuery.Data()));
- delete aResult;
- //return 0;
+ TString toLog = Form("%s (%d): %s - ", TTimeStamp(time(0)).AsString("s"), getpid(), detector);
+ if (GetCurrentRun() >= 0)
+ toLog += Form("run %d - ", GetCurrentRun());
+ toLog += Form("%s", message);
+
+ AliInfo(toLog.Data());
+
+ TString fileName;
+ if (GetCurrentRun() >= 0)
+ fileName.Form("%s/%s_%d.log", GetShuttleLogDir(), detector, GetCurrentRun());
+ else
+ fileName.Form("%s/%s.log", GetShuttleLogDir(), 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();
+}
+
+//______________________________________________________________________________________________
+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));
+
+ SetLastAction("Starting");
+
+ TString whereClause("where shuttle_done=0");
+ if (run != -1)
+ whereClause += Form(" and run=%d", run);
+
+ TObjArray shuttleLogbookEntries;
+ if (!QueryShuttleLogbook(whereClause, shuttleLogbookEntries))
+ {
+ Log("SHUTTLE", "Collect - Can't retrieve entries from Shuttle logbook");
+ return kFALSE;
}
- 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;
+ 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;
+ }
+
+ for (UInt_t iDet=0; iDet<NDetectors(); iDet++)
+ fFirstUnprocessed[iDet] = kTRUE;
+
+ 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;
}
- delete aResult;
+ 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;
+ }
+ }
+ }
+
+ }
+
+ if (!RetrieveConditionsData(shuttleLogbookEntries))
+ {
+ Log("SHUTTLE", "Collect - Process of at least one run failed");
+ return kFALSE;
}
+ Log("SHUTTLE", "Collect - Requested run(s) successfully processed");
return kTRUE;
}
//______________________________________________________________________________________________
-const char* AliShuttle::GetDCSFileName(const char* /*detector*/, const char* /*id*/, const char* /*source*/){
-// Retrieves a file from the DCS FES.
+Bool_t AliShuttle::RetrieveConditionsData(const TObjArray& dateEntries)
+{
+// Retrieve conditions data for all runs that aren't processed yet
-return "You're in DCS";
+ Bool_t hasError = kFALSE;
+
+ TIter iter(&dateEntries);
+ AliShuttleLogbookEntry* anEntry;
+
+ while ((anEntry = (AliShuttleLogbookEntry*) iter.Next())){
+ if (!Process(anEntry)){
+ hasError = kTRUE;
+ }
+
+ // clean SHUTTLE temp directory
+ TString command = Form("rm -f %s/*.shuttle", GetShuttleTempDir());
+ gSystem->Exec(command.Data());
+ }
+ return hasError == kFALSE;
}
//______________________________________________________________________________________________
-TList* AliShuttle::GetDCSFileSources(const char* /*detector*/, const char* /*id*/){
-// Retrieves a file from the DCS FES.
+ULong_t AliShuttle::GetTimeOfLastAction() const
+{
+ ULong_t tmp;
+
+ fMonitoringMutex->Lock();
+
+ tmp = fLastActionTime;
-return NULL;
+ fMonitoringMutex->UnLock();
+ return tmp;
}
//______________________________________________________________________________________________
-const char* AliShuttle::GetHLTFileName(const char* /*detector*/, const char* /*id*/, const char* /*source*/){
-// Retrieves a file from the HLT FES.
+const TString AliShuttle::GetLastAction() const
+{
+ // returns a string description of the last action
+
+ TString tmp;
-return "You're in HLT";
+ fMonitoringMutex->Lock();
+
+ tmp = fLastAction;
+
+ fMonitoringMutex->UnLock();
+ return tmp;
}
//______________________________________________________________________________________________
-TList* AliShuttle::GetHLTFileSources(const char* /*detector*/, const char* /*id*/){
-// Retrieves a file from the HLT FES.
+void AliShuttle::SetLastAction(const char* action)
+{
+ // updates the monitoring variables
-return NULL;
+ fMonitoringMutex->Lock();
+ fLastAction = action;
+ fLastActionTime = time(0);
+
+ fMonitoringMutex->UnLock();
}
//______________________________________________________________________________________________
-const char* AliShuttle::GetDetCode(const char* detector){
-// Return detector code
+const char* AliShuttle::GetRunParameter(const char* param)
+{
+// returns run parameter read from DAQ logbook
- for(int iDet=0; iDet < fgkNDetectors; iDet++){
- if(!strcmp(fgkDetectorName[iDet], detector)) return fgkDetectorCode[iDet];
+ if(!fLogbookEntry) {
+ AliError("No logbook entry!");
+ return 0;
}
- return 0;
+ return fLogbookEntry->GetRunParameter(param);
}
//______________________________________________________________________________________________
-void AliShuttle::Log(const char* detector, const char* message)
+AliCDBEntry* AliShuttle::GetFromOCDB(const AliCDBPath& path)
{
-// Fill log string with a message
+// returns obiect from OCDB valid for current run
- void* dir = gSystem->OpenDirectory(fgkShuttleLogDir);
- if (dir == NULL) {
- if (gSystem->mkdir(fgkShuttleLogDir, kTRUE)) {
- AliError(Form("Can't open directory <%s>!", fgkShuttleTempDir));
- return;
+ AliCDBStorage *sto = AliCDBManager::Instance()->GetStorage(fgkMainCDB);
+ if (!sto)
+ {
+ Log("SHUTTLE", "GetFromOCDB - Cannot activate main OCDB for query!");
+ return 0;
+ }
+
+ return dynamic_cast<AliCDBEntry*> (sto->Get(path, GetCurrentRun()));
+}
+
+//______________________________________________________________________________________________
+Bool_t AliShuttle::SendMail()
+{
+// sends a mail to the subdetector expert in case of preprocessor error
+
+ void* dir = gSystem->OpenDirectory(GetShuttleLogDir());
+ if (dir == NULL)
+ {
+ if (gSystem->mkdir(GetShuttleLogDir(), kTRUE))
+ {
+ AliError(Form("Can't open directory <%s>", GetShuttleLogDir()));
+ return kFALSE;
}
} else {
gSystem->FreeDirectory(dir);
}
- TString toLog = Form("%s: %s, run %d - %s", TTimeStamp(time(0)).AsString(),
- detector, GetCurrentRun(), message);
- AliInfo(toLog.Data());
+ TString bodyFileName;
+ bodyFileName.Form("%s/mail.body", GetShuttleLogDir());
+ gSystem->ExpandPathName(bodyFileName);
- TString fileName;
- fileName.Form("%s/%s.log", fgkShuttleLogDir, detector);
- gSystem->ExpandPathName(fileName);
+ ofstream mailBody;
+ mailBody.open(bodyFileName, ofstream::out);
- ofstream logFile;
- logFile.open(fileName, ofstream::out | ofstream::app);
-
- if (!logFile.is_open()) {
- AliError(Form("Could not open file %s", fileName.Data()));
- return;
+ if (!mailBody.is_open())
+ {
+ AliError(Form("Could not open mail body file %s", bodyFileName.Data()));
+ return kFALSE;
}
- logFile << toLog.Data() << "\n";
+ TString to="";
+ TIter iterExperts(fConfig->GetResponsibles(fCurrentDetector));
+ TObjString *anExpert=0;
+ while ((anExpert = (TObjString*) iterExperts.Next()))
+ {
+ to += Form("%s,", anExpert->GetName());
+ }
+ to.Remove(to.Length()-1);
+ AliDebug(2, Form("to: %s",to.Data()));
- logFile.close();
+ // TODO this will be removed...
+ if (to.Contains("not_yet_set")) {
+ AliInfo("List of detector responsibles not yet set!");
+ return kFALSE;
+ }
+
+ TString cc="alberto.colla@cern.ch";
+
+ TString subject = Form("%s Shuttle preprocessor error in run %d !",
+ fCurrentDetector.Data(), GetCurrentRun());
+ AliDebug(2, Form("subject: %s", subject.Data()));
+
+ TString body = Form("Dear %s expert(s), \n\n", fCurrentDetector.Data());
+ body += Form("SHUTTLE just detected that your preprocessor "
+ "exited with ERROR state in run %d!!\n\n", GetCurrentRun());
+ body += Form("Please check %s status on the web page asap!\n\n", fCurrentDetector.Data());
+ body += Form("The last 10 lines of %s log file are following:\n\n");
+
+ AliDebug(2, Form("Body begin: %s", body.Data()));
+
+ mailBody << body.Data();
+ mailBody.close();
+ mailBody.open(bodyFileName, ofstream::out | ofstream::app);
+
+ TString logFileName = Form("%s/%s_%d.log", GetShuttleLogDir(), 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());
+ }
+
+ TString endBody = Form("------------------------------------------------------\n\n");
+ endBody += Form("In case of problems please contact the SHUTTLE core team.\n\n");
+ endBody += "Please do not answer this message directly, it is automatically generated.\n\n";
+ endBody += "Sincerely yours,\n\n \t\t\tthe SHUTTLE\n";
+
+ AliDebug(2, Form("Body end: %s", endBody.Data()));
+
+ mailBody << endBody.Data();
+
+ mailBody.close();
+
+ // 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()));
+
+ Bool_t result = gSystem->Exec(mailCommand.Data());
+
+ return result == 0;
+}
+
+//______________________________________________________________________________________________
+void AliShuttle::SetRunType()
+{
+// Gets run type from logbook and fills current Shuttle logbook entry
+
+ // check connection, in case connect
+ if(!Connect(3)){
+ Log("SHUTTLE", "GetRunType - Couldn't connect to DAQ Logbook.");
+ return;
+ }
+
+ TString sqlQuery = Form("select detector,run_type from %s where run_number=%d",
+ fConfig->GetRunTypelbTable(), GetCurrentRun());
+
+ 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("GetRunType - Can't execute query <%s>", sqlQuery.Data()));
+ return;
+ }
+
+ TSQLRow* aRow;
+ while ((aRow = aResult->Next())) {
+ TString det(aRow->GetField(0), aRow->GetFieldLength(0));
+ TString runType(aRow->GetField(1), aRow->GetFieldLength(1));
+
+ fLogbookEntry->SetRunType(det, runType);
+ delete aRow;
+ }
+
+ delete aResult;
+
+ return;
+
+}
+
+//______________________________________________________________________________________________
+const char* AliShuttle::GetRunType(const char* detCode)
+{
+// returns run type read from "run type" logbook
+
+ if(!fLogbookEntry) {
+ AliError("No logbook entry!");
+ return 0;
+ }
+
+ return fLogbookEntry->GetRunType(detCode);
+}
+
+//______________________________________________________________________________________________
+void AliShuttle::SetShuttleTempDir(const char* tmpDir)
+{
+// sets Shuttle temp directory
+
+ fgkShuttleTempDir = gSystem->ExpandPathName(tmpDir);
+}
+
+//______________________________________________________________________________________________
+void AliShuttle::SetShuttleLogDir(const char* logDir)
+{
+// sets Shuttle log directory
+
+ fgkShuttleLogDir = gSystem->ExpandPathName(logDir);
}