/*
$Log$
+Revision 1.38 2007/04/12 08:26:18 jgrosseo
+updated comment
+
+Revision 1.37 2007/04/10 16:53:14 jgrosseo
+redirecting sub detector stdout, stderr to sub detector log file
+
+Revision 1.35 2007/04/04 16:26:38 acolla
+1. Re-organization of function calls in TestPreprocessor to make it more meaningful.
+2. Added missing dependency in test preprocessors.
+3. in AliShuttle.cxx: processing time and memory consumption info on a single line.
+
+Revision 1.34 2007/04/04 10:33:36 jgrosseo
+1) Storing of files to the Grid is now done _after_ your preprocessors succeeded. This is transparent, which means that you can still use the same functions (Store, StoreReferenceData) to store files to the Grid. However, the Shuttle first stores them locally and transfers them after the preprocessor finished. The return code of these two functions has changed from UInt_t to Bool_t which gives you the success of the storing.
+In case of an error with the Grid, the Shuttle will retry the storing later, the preprocessor does not need to be run again.
+
+2) The meaning of the return code of the preprocessor has changed. 0 is now success and any other value means failure. This value is stored in the log and you can use it to keep details about the error condition.
+
+3) New function StoreReferenceFile to _directly_ store a file (without opening it) to the reference storage.
+
+4) The memory usage of the preprocessor is monitored. If it exceeds 2 GB it is terminated.
+
+5) New function AliPreprocessor::ProcessDCS(). If you do not need to have DCS data in all cases, you can skip the processing by implemting this function and returning kFALSE under certain conditions. E.g. if there is a certain run type.
+If you always need DCS data (like before), you do not need to implement it.
+
+6) The run type has been added to the monitoring page
+
+Revision 1.33 2007/04/03 13:56:01 acolla
+Grid Storage at the end of preprocessing. Added virtual method to disable DCS query according to the
+run type.
+
+Revision 1.32 2007/02/28 10:41:56 acolla
+Run type field added in SHUTTLE framework. Run type is read from "run type" logbook and retrieved by
+AliPreprocessor::GetRunType() function.
+Added some ldap definition files.
+
Revision 1.30 2007/02/13 11:23:21 acolla
Moved getters and setters of Shuttle's main OCDB/Reference, local
OCDB/Reference, temp and log folders to AliShuttleInterface
#include <TSQLResult.h>
#include <TSQLRow.h>
#include <TMutex.h>
+#include <TSystemDirectory.h>
+#include <TSystemFile.h>
+#include <TFileMerger.h>
+#include <TGrid.h>
+#include <TGridResult.h>
#include <TMonaLisaWriter.h>
ClassImp(AliShuttle)
-Bool_t AliShuttle::fgkProcessDCS(kTRUE);
-
//______________________________________________________________________________________________
AliShuttle::AliShuttle(const AliShuttleConfig* config,
UInt_t timeout, Int_t retries):
fLogbookEntry(0),
fCurrentDetector(),
fStatusEntry(0),
-fGridError(kFALSE),
fMonitoringMutex(0),
fLastActionTime(0),
fLastAction(),
-fMonaLisa(0)
+fMonaLisa(0),
+fTestMode(kNone),
+fReadTestMode(kFALSE),
+fOutputRedirected(kFALSE)
{
//
// config: AliShuttleConfig used
//______________________________________________________________________________________________
AliShuttle::~AliShuttle()
{
-// destructor
+ //
+ // destructor
+ //
fPreprocessorMap.DeleteAll();
for(int iSys=0;iSys<4;iSys++)
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)
+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);
+ // 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, -1, -1);
+ // 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))) {
- AliError(Form("WriteToCDB - 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("WriteToCDB - Problem with main %s storage. Putting <%s> into backup storage",
- cdbType, path.GetPath().Data()));
+ Log(fCurrentDetector, Form("StoreLocally - Can't store object <%s>!", id.ToString().Data()));
+ }
- // Set Grid version to current run number, to ease retrieval later
- id.SetVersion(GetCurrentRun());
+ return result;
+}
- result = AliCDBManager::Instance()->GetStorage(localUri)
- ->Put(object, id, metaData);
+//______________________________________________________________________________________________
+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.
+ //
+
+ 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;
+ }
+
+ AliInfo("Storing reference data ...");
+ Bool_t resultRef = StoreOCDB(fgkMainRefStorage);
+
+ AliInfo("Storing reference files ...");
+ Bool_t resultRefFiles = StoreRefFilesToGrid();
+
+ AliInfo("Storing OCDB data ...");
+ Bool_t resultCDB = StoreOCDB(fgkMainCDB);
+
+ return resultCDB && resultRef && resultRefFiles;
+}
+
+//______________________________________________________________________________________________
+Bool_t AliShuttle::StoreOCDB(const TString& gridURI)
+{
+ //
+ // Called by StoreOCDB(), performs actual storage to the main OCDB and reference storages (Grid)
+ //
+
+ TObjArray* gridIds=0;
+
+ Bool_t result = kTRUE;
+ // to check whether all files have been transferred, or some files were left behind
+ // because the run is not first unprocessed
+ Bool_t willDoAgain = kFALSE;
+
+ 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 kFALSE;
+ }
+
+ AliCDBManager* man = AliCDBManager::Instance();
+
+ AliCDBStorage *gridSto = man->GetStorage(gridURI);
+ if(!gridSto) {
+ Log("SHUTTLE",
+ Form("StoreOCDB - cannot activate main %s storage", type));
+ return kFALSE;
+ }
+
+ gridIds = gridSto->GetQueryCDBList();
+
+ // 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 kFALSE;
+ }
+ 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()));
+ willDoAgain=kTRUE;
+ continue;
+ }
+
+ // 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;
+ }
+
+ // If we get here, the file can be stored!
+ Bool_t storeOk = gridSto->Put(aLocEntry);
+ if(!store || storeOk){
- if(result) {
- result = 2;
- fGridError = kTRUE;
- }else{
- Log(fCurrentDetector, "WriteToCDB - Can't store data!");
+ 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));
+ }
+
+ // removing local filename...
+ TString filename;
+ localSto->IdToFilename(aLocId, filename);
+ AliInfo(Form("Removing local file %s", filename.Data()));
+ RemoveFile(filename.Data());
+ continue;
+ } else {
+ Log("SHUTTLE",
+ Form("StoreOCDB - Grid %s storage of object <%s> failed",
+ type, aLocId.ToString().Data()));
+ result = kFALSE;
}
}
+ localEntries->Clear();
+
+ if(result && willDoAgain) {
+ Log(fCurrentDetector.Data(),
+ "Some files have been left on local storage, will try again later!");
+ result = kFALSE;
+ }
return result;
+}
+
+//______________________________________________________________________________________________
+Bool_t AliShuttle::StoreReferenceFile(const char* detector, const char* localFile, const char* gridFileName)
+{
+ //
+ // Stores reference file directly (without opening it). This function stores the file locally.
+ //
+ // 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;
+ }
+
+ AliCDBManager* man = AliCDBManager::Instance();
+ AliCDBStorage* sto = man->GetStorage(fgkLocalRefStorage);
+
+ TString localBaseFolder = sto->GetBaseFolder();
+
+ TString targetDir;
+ targetDir.Form("%s/%s", localBaseFolder.Data(), detector);
+
+ TString target;
+ target.Form("%s/%d_%s", targetDir.Data(), GetCurrentRun(), gridFileName);
+
+ Int_t result = gSystem->GetPathInfo(targetDir, 0, (Long64_t*) 0, 0, 0);
+ if (result)
+ {
+ result = gSystem->mkdir(targetDir, kTRUE);
+ if (result != 0)
+ {
+ Log("SHUTTLE", Form("StoreReferenceFile - Error creating base directory %s", targetDir.Data()));
+ return kFALSE;
+ }
+ }
+
+ result = gSystem->CopyFile(localFile, target);
+ if (result == 0)
+ {
+ Log("SHUTTLE", Form("StoreReferenceFile - Stored file %s locally to %s", localFile, target.Data()));
+ return kTRUE;
+ }
+ else
+ {
+ Log("SHUTTLE", Form("StoreReferenceFile - Storing file %s locally to %s failed", localFile, target.Data()));
+ return kFALSE;
+ }
+}
+
+//______________________________________________________________________________________________
+Bool_t AliShuttle::StoreRefFilesToGrid()
+{
+ //
+ // Transfers the reference file to the Grid.
+ //
+ // The file is stored under the following location:
+ // <base folder of reference storage>/<DET>/<RUN#>_<gridFileName>
+ // where <gridFileName> is the second parameter given to the function
+ //
+
+ AliCDBManager* man = AliCDBManager::Instance();
+ AliCDBStorage* sto = man->GetStorage(fgkLocalRefStorage);
+ if (!sto)
+ return kFALSE;
+ TString localBaseFolder = sto->GetBaseFolder();
+
+ TString dir;
+ dir.Form("%s/%s", localBaseFolder.Data(), GetOfflineDetName(fCurrentDetector));
+
+ AliCDBStorage* gridSto = man->GetStorage(fgkMainRefStorage);
+ if (!gridSto)
+ return kFALSE;
+ TString gridBaseFolder = gridSto->GetBaseFolder();
+ TString alienDir;
+ alienDir.Form("%s%s", gridBaseFolder.Data(), GetOfflineDetName(fCurrentDetector));
+
+ if (!gGrid)
+ return kFALSE;
+
+ TString begin;
+ begin.Form("%d_", GetCurrentRun());
+
+ TSystemDirectory* baseDir = new TSystemDirectory("/", dir);
+ if (!baseDir)
+ return kTRUE;
+
+ TList* dirList = baseDir->GetListOfFiles();
+ if (!dirList)
+ {
+ delete baseDir;
+ return kTRUE;
+ }
+
+ Int_t nDirs = dirList->GetEntries();
+
+ Bool_t success = kTRUE;
+ Bool_t first = kTRUE;
+
+ for (Int_t iDir=0; iDir<nDirs; ++iDir)
+ {
+ TSystemFile* entry = dynamic_cast<TSystemFile*> (dirList->At(iDir));
+ if (!entry)
+ continue;
+
+ if (entry->IsDirectory())
+ continue;
+
+ TString fileName(entry->GetName());
+ if (!fileName.BeginsWith(begin))
+ continue;
+
+ if (first)
+ {
+ first = kFALSE;
+ // check that DET folder exists, otherwise create it
+ TGridResult* result = gGrid->Ls(alienDir.Data(), "a");
+
+ if (!result)
+ return kFALSE;
+
+ if (!result->GetFileName(0))
+ {
+ if (!gGrid->Mkdir(alienDir.Data(),"",0))
+ {
+ Log("SHUTTLE", Form("StoreRefFilesToGrid - Cannot create directory %s",
+ alienDir.Data()));
+ delete baseDir;
+ return kFALSE;
+ }
+
+ }
+ }
+
+ TString fullLocalPath;
+ fullLocalPath.Form("%s/%s", dir.Data(), fileName.Data());
+
+ TString fullGridPath;
+ fullGridPath.Form("alien://%s/%s", alienDir.Data(), fileName.Data());
+
+ Log("SHUTTLE", Form("StoreRefFilesToGrid - Copying local file %s to %s", fullLocalPath.Data(), fullGridPath.Data()));
+
+ TFileMerger fileMerger;
+ Bool_t result = fileMerger.Cp(fullLocalPath, fullGridPath);
+
+ if (result)
+ {
+ Log("SHUTTLE", Form("StoreRefFilesToGrid - Copying local file %s to %s succeeded", fullLocalPath.Data(), fullGridPath.Data()));
+ RemoveFile(fullLocalPath);
+ }
+ else
+ {
+ Log("SHUTTLE", Form("StoreRefFilesToGrid - Copying local file %s to %s failed", fullLocalPath.Data(), fullGridPath.Data()));
+ success = kFALSE;
+ }
+ }
+
+ delete baseDir;
+
+ return success;
+}
+
+//______________________________________________________________________________________________
+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(), fCurrentDetector.Data(), GetCurrentRun()));
+
+ AliInfo(Form("filename = %s", filename.Data()));
+
+ AliInfo(Form("Removing remaining local files from 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
+ //
+ // Reads the AliShuttleStatus from the CDB
+ //
if (fStatusEntry){
delete fStatusEntry;
//______________________________________________________________________________________________
Bool_t AliShuttle::WriteShuttleStatus(AliShuttleStatus* status)
{
-// writes the status for one subdetector
+ //
+ // writes the status for one subdetector
+ //
if (fStatusEntry){
delete fStatusEntry;
UInt_t result = AliCDBManager::Instance()->GetStorage(fgkLocalCDB)->Put(fStatusEntry);
if (!result) {
- AliError(Form("WriteShuttleStatus for %s, run %d failed", fCurrentDetector.Data(), run));
+ Log("SHUTTLE", Form("WriteShuttleStatus - Failed for %s, run %d",
+ fCurrentDetector.Data(), run));
return kFALSE;
}
//______________________________________________________________________________________________
void AliShuttle::UpdateShuttleStatus(AliShuttleStatus::Status newStatus, Bool_t increaseCount)
{
- // changes the AliShuttleStatus for the given detector and run to the given status
+ //
+ // changes the AliShuttleStatus for the given detector and run to the given status
+ //
if (!fStatusEntry){
AliError("UNEXPECTED: fStatusEntry empty");
AliShuttleStatus* status = dynamic_cast<AliShuttleStatus*> (fStatusEntry->GetObject());
if (!status){
- AliError("UNEXPECTED: status could not be read from current CDB entry");
+ Log("SHUTTLE", "UNEXPECTED: status could not be read from current CDB entry");
return;
}
AliShuttleStatus* status = dynamic_cast<AliShuttleStatus*> (fStatusEntry->GetObject());
if (!status){
- AliError("UNEXPECTED: status could not be read from current CDB entry");
+ Log("SHUTTLE", "SendMLInfo - UNEXPECTED: status could not be read from current CDB entry");
return;
}
//______________________________________________________________________________________________
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
+ // 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;
return kFALSE;
}
- if (status->GetStatus() == AliShuttleStatus::kStoreFailed) {
+ if (status->GetStatus() == AliShuttleStatus::kStoreError) {
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::kStoreStarted);
+ if (StoreOCDB()){
+ Log("SHUTTLE", Form("ContinueProcessing - %s: all objects successfully stored into main storage",
+ fCurrentDetector.Data()));
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();
+ UpdateShuttleStatus(AliShuttleStatus::kStoreError);
}
return kFALSE;
}
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!
+ AliInfo(Form("Sending mail to %s expert...", fCurrentDetector.Data()));
+ if (!SendMail())
+ 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()));
- UpdateShuttleStatus(AliShuttleStatus::kStarted, kTRUE);
+ Bool_t increaseCount = kTRUE;
+ if (status->GetStatus() == AliShuttleStatus::kDCSError || status->GetStatus() == AliShuttleStatus::kDCSStarted)
+ increaseCount = kFALSE;
+ UpdateShuttleStatus(AliShuttleStatus::kStarted, increaseCount);
cont = kTRUE;
}
- // 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 cont;
}
// Returns kFALSE in case of error occured and kTRUE otherwise
//
- if(!entry) return kFALSE;
+ if (!entry) return kFALSE;
fLogbookEntry = entry;
- if (fLogbookEntry->IsDone())
- {
- Log("SHUTTLE","Process - Shuttle is already DONE. Updating logbook");
- UpdateShuttleLogbook("shuttle_done");
- fLogbookEntry = 0;
- return kTRUE;
- }
+ AliInfo(Form("\n\n \t\t\t^*^*^*^*^*^*^*^*^*^*^*^* run %d: START ^*^*^*^*^*^*^*^*^*^*^*^* \n",
+ GetCurrentRun()));
// create ML instance that monitors this run
fMonaLisa = new TMonaLisaWriter(Form("%d", GetCurrentRun()), "SHUTTLE", "aliendb1.cern.ch");
// disable monitoring of other parameters that come e.g. from TFile
gMonitoringWriter = 0;
- AliInfo(Form("\n\n \t\t\t^*^*^*^*^*^*^*^*^*^*^*^* run %d: START ^*^*^*^*^*^*^*^*^*^*^*^* \n",
- GetCurrentRun()));
-
- // Set run type from run type logbook into current fLogbookEntry
- SetRunType();
-
// Send the information to ML
TMonaLisaText mlStatus("SHUTTLE_status", "Processing");
+ TMonaLisaText mlRunType("SHUTTLE_runtype", Form("%s (%s)", entry->GetRunType(), entry->GetRunParameter("log")));
TList mlList;
mlList.Add(&mlStatus);
+ mlList.Add(&mlRunType);
fMonaLisa->SendParameters(&mlList);
-
+
+ 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("Enabling test mode %d", testMode));
+ SetTestMode((TestMode) testMode);
+ }
+ }
+ delete token;
+ }
+ }
+ }
+
+ Log("SHUTTLE", Form("The test mode flag is %d", (Int_t) fTestMode));
+
fLogbookEntry->Print("all");
// Initialization
if (expiredTime > fConfig->GetPPTimeOut())
{
- Log("SHUTTLE", Form("Process time out. Run time: %d seconds. Killing...",
- expiredTime));
+ TString tmp;
+ tmp.Form("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
{
- if (expiredTime % 60 == 0)
- Log("SHUTTLE", Form("Checked process. Run time: %d seconds.",
- expiredTime));
gSystem->Sleep(1000);
+
+ TString checkStr;
+ checkStr.Form("ps -o vsize --pid %d | tail -n 1", pid);
+ FILE* pipe = gSystem->OpenPipe(checkStr, "r");
+ if (!pipe)
+ {
+ Log("SHUTTLE", Form("Error: Could not open pipe to %s", checkStr.Data()));
+ continue;
+ }
+
+ char buffer[100];
+ if (!fgets(buffer, 100, pipe))
+ {
+ Log("SHUTTLE", "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", "Error: Could not parse output of ps");
+ continue;
+ }
+
+ if (expiredTime % 60 == 0)
+ Log("SHUTTLE", Form("%s: Checking process. Run time: %d seconds - Memory consumption: %d KB",
+ fCurrentDetector.Data(), expiredTime, mem));
+
+ if (mem > fConfig->GetPPMaxMem())
+ {
+ TString tmp;
+ tmp.Form("Process exceeds maximum allowed memory (%d KB > %d KB). Killing...",
+ mem, fConfig->GetPPMaxMem());
+ Log("SHUTTLE", tmp);
+ Log(fCurrentDetector, tmp);
+
+ kill(pid, 9);
+
+ UpdateShuttleStatus(AliShuttleStatus::kPPOutOfMemory);
+ hasError = kTRUE;
+
+ gSystem->Sleep(1000);
+ }
}
}
{
Int_t returnCode = WEXITSTATUS(status);
- Log("SHUTTLE", Form("The return code is %d", returnCode));
+ Log("SHUTTLE", Form("%s: the return code is %d", fCurrentDetector.Data(),
+ returnCode));
- if (returnCode != 0)
- hasError = kTRUE;
+ 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();
+ AliInfo("Redirecting output...");
- Int_t returnCode = 0; // will be set to 1 in case of an error
-
- if (!result)
+ if ((freopen(GetLogFileName(fCurrentDetector), "w", stdout)) == 0)
{
- returnCode = 1;
- AliInfo(Form("\n \t\t\t****** run %d - %s: PREPROCESSOR ERROR ****** \n\n",
- GetCurrentRun(), aDetector->GetName()));
+ Log("SHUTTLE", "Could not freopen stdout");
}
- else if (result == 2)
+ else
{
- AliInfo(Form("\n \t\t\t****** run %d - %s: STORAGE ERROR ****** \n\n",
+ fOutputRedirected = kTRUE;
+ if ((dup2(fileno(stdout), fileno(stderr))) < 0)
+ Log("SHUTTLE", "Could not redirect stderr");
+
+ }
+
+ Bool_t success = ProcessCurrentDetector();
+ if (success) // Preprocessor finished successfully!
+ {
+ // Update time_processed field in FXS DB
+ if (UpdateTable() == kFALSE)
+ Log("SHUTTLE", Form("Process - %s: Could not update FXS databases!"));
+
+ // Transfer the data from local storage to main storage (Grid)
+ UpdateShuttleStatus(AliShuttleStatus::kStoreStarted);
+ if (StoreOCDB() == kFALSE)
+ {
+ AliInfo(Form("\n \t\t\t****** run %d - %s: STORAGE ERROR ****** \n\n",
GetCurrentRun(), aDetector->GetName()));
- } else
- {
- AliInfo(Form("\n \t\t\t****** run %d - %s: DONE ****** \n\n",
+ UpdateShuttleStatus(AliShuttleStatus::kStoreError);
+ success = kFALSE;
+ } 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;
+ UpdateShuttleStatus(AliShuttleStatus::kDone);
+ UpdateShuttleLogbook(fCurrentDetector, "DONE");
+ }
}
for (UInt_t iSys=0; iSys<3; iSys++)
}
AliInfo(Form("Client process of %d - %s is exiting now with %d.",
- GetCurrentRun(), aDetector->GetName(), returnCode));
+ GetCurrentRun(), aDetector->GetName(), success));
// the client exits here
- gSystem->Exit(returnCode);
+ gSystem->Exit(success);
AliError("We should never get here!!!");
}
}
//______________________________________________________________________________________________
-UInt_t AliShuttle::ProcessCurrentDetector()
+Bool_t AliShuttle::ProcessCurrentDetector()
{
//
// Makes data retrieval just for a specific detector (fCurrentDetector).
AliInfo(Form("Retrieving values for %s, run %d", fCurrentDetector.Data(), GetCurrentRun()));
- UpdateShuttleStatus(AliShuttleStatus::kDCSStarted);
-
TMap dcsMap;
dcsMap.SetOwner(1);
Bool_t aDCSError = kFALSE;
- fGridError = kFALSE;
- // TODO Test only... I've added a flag that allows to
- // exclude DCS archive DB query
- if (!fgkProcessDCS)
+ // call preprocessor
+ AliPreprocessor* aPreprocessor =
+ dynamic_cast<AliPreprocessor*> (fPreprocessorMap.GetValue(fCurrentDetector));
+
+ aPreprocessor->Initialize(GetCurrentRun(), GetCurrentStartTime(), GetCurrentEndTime());
+
+ Bool_t processDCS = aPreprocessor->ProcessDCS();
+
+ if (!processDCS || (fTestMode & kSkipDCS))
+ {
+ Log(fCurrentDetector, "In TESTMODE - Skipping DCS processing!");
+ }
+ else if (fTestMode & kErrorDCS)
{
- AliInfo("Skipping DCS processing!");
- aDCSError = kFALSE;
+ Log(fCurrentDetector, "In TESTMODE - Simulating DCS error");
+ UpdateShuttleStatus(AliShuttleStatus::kDCSStarted);
+ UpdateShuttleStatus(AliShuttleStatus::kDCSError);
+ return kFALSE;
} else {
+
+ UpdateShuttleStatus(AliShuttleStatus::kDCSStarted);
+
TString host(fConfig->GetDCSHost(fCurrentDetector));
Int_t port = fConfig->GetDCSPort(fCurrentDetector);
anAlias->GetName()));
UpdateShuttleStatus(AliShuttleStatus::kDCSError);
dcsMap.DeleteAll();
- return 0;
+ return kFALSE;
}
}
aDP->GetName()));
UpdateShuttleStatus(AliShuttleStatus::kDCSError);
dcsMap.DeleteAll();
- return 0;
+ return kFALSE;
}
}
}
// DCS Archive DB processing successful. Call Preprocessor!
UpdateShuttleStatus(AliShuttleStatus::kPPStarted);
- AliPreprocessor* aPreprocessor =
- dynamic_cast<AliPreprocessor*> (fPreprocessorMap.GetValue(fCurrentDetector));
-
- aPreprocessor->Initialize(GetCurrentRun(), GetCurrentStartTime(), GetCurrentEndTime());
- UInt_t aPPResult = aPreprocessor->Process(&dcsMap);
+ UInt_t returnValue = aPreprocessor->Process(&dcsMap);
- UInt_t returnValue = 0;
- if (aPPResult == 0) { // Preprocessor error
+ if (returnValue > 0) // Preprocessor error!
+ {
+ Log(fCurrentDetector, Form("Preprocessor failed. Process returned %d.", returnValue));
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;
+ dcsMap.DeleteAll();
+ return kFALSE;
}
+
+ // preprocessor ok!
+ UpdateShuttleStatus(AliShuttleStatus::kPPDone);
+ Log(fCurrentDetector, Form("ProcessCurrentDetector - %s preprocessor returned success",
+ fCurrentDetector.Data()));
dcsMap.DeleteAll();
- return returnValue;
+ return kTRUE;
}
//______________________________________________________________________________________________
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.
+ // Query DAQ's Shuttle logbook and fills detector status object.
+ // Call QueryRunParameters to query DAQ logbook for run parameters.
+ //
entries.SetOwner(1);
AliDebug(2,Form("Query = %s", sqlQuery.Data()));
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;
-// }
+ AliInfo("No entries in Shuttle Logbook match request");
+ delete aResult;
+ return kTRUE;
}
// TODO Check field count!
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;
}
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
+ // 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())
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
-
+ // 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))
{
//______________________________________________________________________________________________
Bool_t AliShuttle::RetrieveFile(UInt_t system, const char* fxsFileName, const char* localFileName)
{
-// Copies file from FXS to local Shuttle machine
+ //
+ // 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/%s",
//______________________________________________________________________________________________
TList* AliShuttle::GetFileSources(Int_t system, const char* detector, const char* id)
{
-// Get sources producing the condition file Id from file exchange servers
+ //
+ // Get sources producing the condition file Id from file exchange servers
+ //
+
+ // 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)
{
AliError("DCS system has only one source of data!");
return NULL;
-
}
// check connection, in case connect
//______________________________________________________________________________________________
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
+ // 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;
//______________________________________________________________________________________________
Bool_t AliShuttle::UpdateTable()
{
-// Update FXS table filling time_processed field in all rows corresponding to current run and detector
+ //
+ // Update FXS table filling time_processed field in all rows corresponding to current run and detector
+ //
Bool_t result = kTRUE;
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)
{
//______________________________________________________________________________________________
Int_t AliShuttle::GetCurrentRun() const
{
-// Get current run from logbook entry
+ //
+ // Get current run from logbook entry
+ //
return fLogbookEntry ? fLogbookEntry->GetRun() : -1;
}
//______________________________________________________________________________________________
UInt_t AliShuttle::GetCurrentStartTime() const
{
-// get current start time
+ //
+ // get current start time
+ //
return fLogbookEntry ? fLogbookEntry->GetStartTime() : 0;
}
//______________________________________________________________________________________________
UInt_t AliShuttle::GetCurrentEndTime() const
{
-// get current end time from logbook entry
+ //
+ // 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
+ //
+ // Fill log string with a message
+ //
void* dir = gSystem->OpenDirectory(GetShuttleLogDir());
if (dir == NULL) {
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;
- if (GetCurrentRun() >= 0)
- fileName.Form("%s/%s_%d.log", GetShuttleLogDir(), detector, GetCurrentRun());
- else
- fileName.Form("%s/%s.log", GetShuttleLogDir(), detector);
+ TString fileName = GetLogFileName(detector);
gSystem->ExpandPathName(fileName);
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/%s_%d.log", GetShuttleLogDir(), detector, GetCurrentRun());
+ else
+ fileName.Form("%s/%s.log", GetShuttleLogDir(), detector);
+
+ return fileName;
+}
+
//______________________________________________________________________________________________
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.
-//
+ //
+ // 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");
//______________________________________________________________________________________________
Bool_t AliShuttle::RetrieveConditionsData(const TObjArray& dateEntries)
{
-// Retrieve conditions data for all runs that aren't processed yet
+ //
+ // Retrieve conditions data for all runs that aren't processed yet
+ //
Bool_t hasError = kFALSE;
}
// clean SHUTTLE temp directory
- TString command = Form("rm -f %s/*.shuttle", GetShuttleTempDir());
- gSystem->Exec(command.Data());
+ TString filename = Form("%s/*.shuttle", GetShuttleTempDir());
+ RemoveFile(filename.Data());
}
return hasError == kFALSE;
//______________________________________________________________________________________________
ULong_t AliShuttle::GetTimeOfLastAction() const
{
+ //
+ // Gets time of last action
+ //
+
ULong_t tmp;
fMonitoringMutex->Lock();
//______________________________________________________________________________________________
const TString AliShuttle::GetLastAction() const
{
+ //
// returns a string description of the last action
+ //
TString tmp;
//______________________________________________________________________________________________
void AliShuttle::SetLastAction(const char* action)
{
+ //
// updates the monitoring variables
+ //
fMonitoringMutex->Lock();
//______________________________________________________________________________________________
const char* AliShuttle::GetRunParameter(const char* param)
{
-// returns run parameter read from DAQ logbook
+ //
+ // returns run parameter read from DAQ logbook
+ //
if(!fLogbookEntry) {
AliError("No logbook entry!");
}
//______________________________________________________________________________________________
-AliCDBEntry* AliShuttle::GetFromOCDB(const AliCDBPath& path)
+AliCDBEntry* AliShuttle::GetFromOCDB(const char* detector, const AliCDBPath& path)
{
-// returns obiect from OCDB valid for current run
+ //
+ // returns object from OCDB valid for current run
+ //
+ if (fTestMode & kErrorOCDB)
+ {
+ Log(detector, "GetFromOCDB - In TESTMODE - Simulating error with OCDB");
+ return 0;
+ }
+
AliCDBStorage *sto = AliCDBManager::Instance()->GetStorage(fgkMainCDB);
if (!sto)
{
- Log("SHUTTLE", "GetFromOCDB - Cannot activate main OCDB for query!");
+ Log(detector, "GetFromOCDB - Cannot activate main OCDB for query!");
return 0;
}
//______________________________________________________________________________________________
Bool_t AliShuttle::SendMail()
{
-// sends a mail to the subdetector expert in case of preprocessor error
+ //
+ // sends a mail to the subdetector expert in case of preprocessor error
+ //
+
+ if (fTestMode != kNone)
+ return kTRUE;
void* dir = gSystem->OpenDirectory(GetShuttleLogDir());
if (dir == NULL)
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());
+ "FAILED after %d retries in run %d!!\n\n", fConfig->GetMaxRetries(), 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");
}
//______________________________________________________________________________________________
-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)
+const char* AliShuttle::GetRunType()
{
-// returns run type read from "run type" logbook
+ //
+ // returns run type read from "run type" logbook
+ //
if(!fLogbookEntry) {
AliError("No logbook entry!");
return 0;
}
- return fLogbookEntry->GetRunType(detCode);
+ return fLogbookEntry->GetRunType();
}
//______________________________________________________________________________________________
void AliShuttle::SetShuttleTempDir(const char* tmpDir)
{
-// sets Shuttle temp directory
+ //
+ // sets Shuttle temp directory
+ //
fgkShuttleTempDir = gSystem->ExpandPathName(tmpDir);
}
//______________________________________________________________________________________________
void AliShuttle::SetShuttleLogDir(const char* logDir)
{
-// sets Shuttle log directory
+ //
+ // sets Shuttle log directory
+ //
fgkShuttleLogDir = gSystem->ExpandPathName(logDir);
}
git://git.uio.no / u / mrichter / AliRoot.git / blobdiff