* provided "as is" without express or implied warranty. *
**************************************************************************/
-/*
-$Log$
-Revision 1.71 2007/12/12 14:56:14 jgrosseo
-sending shuttle_ignore to ML also in case of 0 events
-
-Revision 1.70 2007/12/12 13:45:35 acolla
-Monalisa started in Collect() function. Alive message to monitor is sent at each Collect and every minute during preprocessor processing.
-
-Revision 1.69 2007/12/12 10:06:29 acolla
-in AliShuttle.cxx: SHUTTLE logbook is updated in case of invalid run times:
-
-time_start==0 && time_end==0
-
-logbook is NOT updated if time_start != 0 && time_end == 0, because it may mean that the run is still ongoing.
-
-Revision 1.68 2007/12/11 10:15:17 acolla
-Added marking SHUTTLE=DONE for invalid runs
-(invalid start time or end time) and runs with totalEvents < 1
-
-Revision 1.67 2007/12/07 19:14:36 acolla
-in AliShuttleTrigger:
-
-Added automatic collection of new runs on a regular time basis (settable from the configuration)
-
-in AliShuttleConfig: new members
-
-- triggerWait: time to wait for DIM trigger (s) before starting automatic collection of new runs
-- mode: run mode (test, prod) -> used to build log folder (logs or logs_PROD)
-
-in AliShuttle:
-
-- logs now stored in logs/#RUN/DET_#RUN.log
-
-Revision 1.66 2007/12/05 10:45:19 jgrosseo
-changed order of arguments to TMonaLisaWriter
-
-Revision 1.65 2007/11/26 16:58:37 acolla
-Monalisa configuration added: host and table name
-
-Revision 1.64 2007/11/13 16:15:47 acolla
-DCS map is stored in a file in the temp folder where the detector is processed.
-If the preprocessor fails, the temp folder is not removed. This will help the debugging of the problem.
-
-Revision 1.63 2007/11/02 10:53:16 acolla
-Protection added to AliShuttle::CopyFileLocally
-
-Revision 1.62 2007/10/31 18:23:13 acolla
-Furter developement on the Shuttle:
-
-- Shuttle now connects to the Grid as alidaq. The OCDB and Reference folders
-are now built from /alice/data, e.g.:
-/alice/data/2007/LHC07a/OCDB
-
-the year and LHC period are taken from the Shuttle.
-Raw metadata files are stored by GRP to:
-/alice/data/2007/LHC07a/<runNb>/Raw/RunMetadata.root
-
-- Shuttle sends a mail to DCS experts each time DP retrieval fails.
-
-Revision 1.61 2007/10/30 20:33:51 acolla
-Improved managing of temporary folders, which weren't correctly handled.
-Resolved bug introduced in StoreReferenceFile, which caused SPD preprocessor fail.
-
-Revision 1.60 2007/10/29 18:06:16 acolla
-
-New function StoreRunMetadataFile added to preprocessor and Shuttle interface
-This function can be used by GRP only. It stores raw data tags merged file to the
-raw data folder (e.g. /alice/data/2008/LHC08a/000099999/Raw).
-
-KNOWN ISSUES:
-
-1. Shuttle cannot write to /alice/data/ because it belongs to alidaq. Tag file is stored in /alice/simulation/... for the time being.
-2. Due to a bug in TAlien::Mkdir, the creation of a folder in recursive mode (-p option) does not work. The problem
-has been corrected in the root package on the Shuttle machine.
-
-Revision 1.59 2007/10/05 12:40:55 acolla
-
-Result error code added to AliDCSClient data members (it was "lost" with the new implementation of TMap* GetAliasValues and GetDPValues).
-
-Revision 1.58 2007/09/28 15:27:40 acolla
-
-AliDCSClient "multiSplit" option added in the DCS configuration
-in AliDCSMessage: variable MAX_BODY_SIZE set to 500000
-
-Revision 1.57 2007/09/27 16:53:13 acolla
-Detectors can have more than one AMANDA server. SHUTTLE queries the servers sequentially,
-merges the dcs aliases/DPs in one TMap and sends it to the preprocessor.
-
-Revision 1.56 2007/09/14 16:46:14 jgrosseo
-1) Connect and Close are called before and after each query, so one can
-keep the same AliDCSClient object.
-2) The splitting of a query is moved to GetDPValues/GetAliasValues.
-3) Splitting interval can be specified in constructor
-
-Revision 1.55 2007/08/06 12:26:40 acolla
-Function Bool_t GetHLTStatus added to preprocessor. It returns the status of HLT
-read from the run logbook.
-
-Revision 1.54 2007/07/12 09:51:25 jgrosseo
-removed duplicated log message in GetFile
-
-Revision 1.53 2007/07/12 09:26:28 jgrosseo
-updating hlt fxs base path
-
-Revision 1.52 2007/07/12 08:06:45 jgrosseo
-adding log messages in getfile... functions
-adding not implemented copy constructor in alishuttleconfigholder
-
-Revision 1.51 2007/07/03 17:24:52 acolla
-root moved to v5-16-00. TFileMerger->Cp moved to TFile::Cp.
-
-Revision 1.50 2007/07/02 17:19:32 acolla
-preprocessor is run in a temp directory that is removed when process is finished.
-
-Revision 1.49 2007/06/29 10:45:06 acolla
-Number of columns in MySql Shuttle logbook increased by one (HLT added)
-
-Revision 1.48 2007/06/21 13:06:19 acolla
-GetFileSources returns dummy list with 1 source if system=DCS (better than
-returning error as it was)
-
-Revision 1.47 2007/06/19 17:28:56 acolla
-HLT updated; missing map bug removed.
-
-Revision 1.46 2007/06/09 13:01:09 jgrosseo
-Switching to retrieval of several DCS DPs at a time (multiDPrequest)
-
-Revision 1.45 2007/05/30 06:35:20 jgrosseo
-Adding functionality to the Shuttle/TestShuttle:
-o) Function to retrieve list of sources from a given system (GetFileSources with id=0)
-o) Function to retrieve list of IDs for a given source (GetFileIDs)
-These functions are needed for dealing with the tag files that are saved for the GRP preprocessor
-Example code has been added to the TestProcessor in TestShuttle
-
-Revision 1.44 2007/05/11 16:09:32 acolla
-Reference files for ITS, MUON and PHOS are now stored in OfflineDetName/OnlineDetName/run_...
-example: ITS/SPD/100_filename.root
-
-Revision 1.43 2007/05/10 09:59:51 acolla
-Various bug fixes in StoreRefFilesToGrid; Cleaning of reference storage before processing detector (CleanReferenceStorage)
-
-Revision 1.42 2007/05/03 08:01:39 jgrosseo
-typo in last commit :-(
-
-Revision 1.41 2007/05/03 08:00:48 jgrosseo
-fixing log message when pp want to skip dcs value retrieval
-
-Revision 1.40 2007/04/27 07:06:48 jgrosseo
-GetFileSources returns empty list in case of no files, but successful query
-No mails sent in testmode
-
-Revision 1.39 2007/04/17 12:43:57 acolla
-Correction in StoreOCDB; change of text in mail to detector expert
-
-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
-
-Revision 1.27 2007/01/30 17:52:42 jgrosseo
-adding monalisa monitoring
-
-Revision 1.26 2007/01/23 19:20:03 acolla
-Removed old ldif files, added TOF, MCH ldif files. Added some options in
-AliShuttleConfig::Print. Added in Ali Shuttle: SetShuttleTempDir and
-SetShuttleLogDir
-
-Revision 1.25 2007/01/15 19:13:52 acolla
-Moved some AliInfo to AliDebug in SendMail function
-
-Revision 1.21 2006/12/07 08:51:26 jgrosseo
-update (alberto):
-table, db names in ldap configuration
-added GRP preprocessor
-DCS data can also be retrieved by data point
-
-Revision 1.20 2006/11/16 16:16:48 jgrosseo
-introducing strict run ordering flag
-removed giving preprocessor name to preprocessor, they have to know their name themselves ;-)
-
-Revision 1.19 2006/11/06 14:23:04 jgrosseo
-major update (Alberto)
-o) reading of run parameters from the logbook
-o) online offline naming conversion
-o) standalone DCSclient package
-
-Revision 1.18 2006/10/20 15:22:59 jgrosseo
-o) Adding time out to the execution of the preprocessors: The Shuttle forks and the parent process monitors the child
-o) Merging Collect, CollectAll, CollectNew function
-o) Removing implementation of empty copy constructors (declaration still there!)
-
-Revision 1.17 2006/10/05 16:20:55 jgrosseo
-adapting to new CDB classes
-
-Revision 1.16 2006/10/05 15:46:26 jgrosseo
-applying to the new interface
-
-Revision 1.15 2006/10/02 16:38:39 jgrosseo
-update (alberto):
-fixed memory leaks
-storing of objects that failed to be stored to the grid before
-interfacing of shuttle status table in daq system
-
-Revision 1.14 2006/08/29 09:16:05 jgrosseo
-small update
-
-Revision 1.13 2006/08/15 10:50:00 jgrosseo
-effc++ corrections (alberto)
-
-Revision 1.12 2006/08/08 14:19:29 jgrosseo
-Update to shuttle classes (Alberto)
-
-- Possibility to set the full object's path in the Preprocessor's and
-Shuttle's Store functions
-- Possibility to extend the object's run validity in the same classes
-("startValidity" and "validityInfinite" parameters)
-- Implementation of the StoreReferenceData function to store reference
-data in a dedicated CDB storage.
-
-Revision 1.11 2006/07/21 07:37:20 jgrosseo
-last run is stored after each run
-
-Revision 1.10 2006/07/20 09:54:40 jgrosseo
-introducing status management: The processing per subdetector is divided into several steps,
-after each step the status is stored on disk. If the system crashes in any of the steps the Shuttle
-can keep track of the number of failures and skips further processing after a certain threshold is
-exceeded. These thresholds can be configured in LDAP.
-
-Revision 1.9 2006/07/19 10:09:55 jgrosseo
-new configuration, accesst to DAQ FES (Alberto)
-
-Revision 1.8 2006/07/11 12:44:36 jgrosseo
-adding parameters for extended validity range of data produced by preprocessor
-
-Revision 1.7 2006/07/10 14:37:09 jgrosseo
-small fix + todo comment
-
-Revision 1.6 2006/07/10 13:01:41 jgrosseo
-enhanced storing of last sucessfully processed run (alberto)
-
-Revision 1.5 2006/07/04 14:59:57 jgrosseo
-revision of AliDCSValue: Removed wrapper classes, reduced storage size per value by factor 2
-
-Revision 1.4 2006/06/12 09:11:16 jgrosseo
-coding conventions (Alberto)
-
-Revision 1.3 2006/06/06 14:26:40 jgrosseo
-o) removed files that were moved to STEER
-o) shuttle updated to follow the new interface (Alberto)
-
-Revision 1.2 2006/03/07 07:52:34 hristov
-New version (B.Yordanov)
-
-Revision 1.6 2005/11/19 17:19:14 byordano
-RetrieveDATEEntries and RetrieveConditionsData added
-
-Revision 1.5 2005/11/19 11:09:27 byordano
-AliShuttle declaration added
-
-Revision 1.4 2005/11/17 17:47:34 byordano
-TList changed to TObjArray
-
-Revision 1.3 2005/11/17 14:43:23 byordano
-import to local CVS
-
-Revision 1.1.1.1 2005/10/28 07:33:58 hristov
-Initial import as subdirectory in AliRoot
-
-Revision 1.2 2005/09/13 08:41:15 byordano
-default startTime endTime added
-
-Revision 1.4 2005/08/30 09:13:02 byordano
-some docs added
-
-Revision 1.3 2005/08/29 21:15:47 byordano
-some docs added
-
-*/
+/* $Id$ */
//
// This class is the main manager for AliShuttle.
#include <TFile.h>
#include <TGrid.h>
#include <TGridResult.h>
+#include <TMap.h>
#include <TMonaLisaWriter.h>
#include <sys/types.h>
#include <sys/wait.h>
+#include <signal.h>
+
ClassImp(AliShuttle)
//______________________________________________________________________________________________
fPreprocessorMap(),
fLogbookEntry(0),
fCurrentDetector(),
+fFirstProcessing(0),
+fFXSError(-1),
fStatusEntry(0),
fMonitoringMutex(0),
fLastActionTime(0),
//
// returns 0 if fail, 1 otherwise
+
if (fTestMode & kErrorStorage)
{
Log(fCurrentDetector, "StoreLocally - In TESTMODE - Simulating error while storing locally");
if (!(AliCDBManager::Instance()->GetStorage(localUri))) {
Log("SHUTTLE", Form("StoreLocally - Cannot activate local %s storage", cdbType));
} else {
+ Int_t logLevel = AliLog::GetGlobalLogLevel();
+ AliLog::SetGlobalLogLevel(AliLog::kError);
result = AliCDBManager::Instance()->GetStorage(localUri)
->Put(object, id, metaData);
+ AliLog::SetGlobalLogLevel((AliLog::EType_t)logLevel);
}
if(!result) {
Log(fCurrentDetector, Form("StoreLocally - Can't store object <%s>!", id.ToString().Data()));
}
+
return result;
}
// Then calls StoreRefFilesToGrid to store reference files.
//
+ UpdateShuttleStatus(AliShuttleStatus::kStoreStarted);
+
if (fTestMode & kErrorGrid)
{
Log("SHUTTLE", "StoreOCDB - In TESTMODE - Simulating error while storing in the Grid");
}
Log("SHUTTLE","StoreOCDB - Storing OCDB data ...");
- Bool_t resultCDB = StoreOCDB(fgkMainCDB);
+ Int_t resultCDB = StoreOCDB(fgkMainCDB);
Log("SHUTTLE","StoreOCDB - Storing reference data ...");
- Bool_t resultRef = StoreOCDB(fgkMainRefStorage);
+ Int_t resultRef = StoreOCDB(fgkMainRefStorage);
Log("SHUTTLE","StoreOCDB - Storing reference files ...");
Bool_t resultRefFiles = CopyFilesToGrid("reference");
Bool_t resultMetadata = kTRUE;
if(fCurrentDetector == "GRP")
{
- Log("StoreOCDB - SHUTTLE","Storing Run Metadata file ...");
+ Log("SHUTTLE","StoreOCDB - Storing Run Metadata file ...");
resultMetadata = CopyFilesToGrid("metadata");
}
- return resultCDB && resultRef && resultRefFiles && resultMetadata;
+ Int_t storeResult = 0;
+
+ if (resultCDB < 0 || resultRef < 0 || resultRefFiles == kFALSE || resultMetadata == kFALSE)
+ storeResult = -1;
+ else if (resultCDB > 0 || resultRef > 0)
+ storeResult = 1;
+
+ if (storeResult < 0)
+ {
+ Log("SHUTTLE",
+ Form("\t\t\t****** run %d - %s: STORAGE ERROR ******",
+ GetCurrentRun(), fCurrentDetector.Data()));
+ UpdateShuttleStatus(AliShuttleStatus::kStoreError);
+ }
+ else if (storeResult > 0)
+ {
+ Log("SHUTTLE",
+ Form("\t\t\t****** run %d - %s: STORAGE DELAYED ******",
+ GetCurrentRun(), fCurrentDetector.Data()));
+ UpdateShuttleStatus(AliShuttleStatus::kStoreDelayed);
+ }
+ else if (storeResult == 0)
+ {
+ Log("SHUTTLE",
+ Form("\t\t\t****** run %d - %s: DONE ******",
+ GetCurrentRun(), fCurrentDetector.Data()));
+ UpdateShuttleStatus(AliShuttleStatus::kDone);
+ UpdateShuttleLogbook(fCurrentDetector, "DONE");
+ }
+
+ return (storeResult == 0);
}
//______________________________________________________________________________________________
-Bool_t AliShuttle::StoreOCDB(const TString& gridURI)
+Int_t AliShuttle::StoreOCDB(const TString& gridURI)
{
//
// Called by StoreOCDB(), performs actual storage to the main OCDB and reference storages (Grid)
//
+ // Return code:
+ // -2 initialization error
+ // -1 storage error
+ // 0 success
+ // 1 storage delayed (e.g. previous unprocessed runs)
+ //
TObjArray* gridIds=0;
Bool_t result = kTRUE;
+ Bool_t delayed = kFALSE;
const char* type = 0;
TString localURI;
localURI = fgkLocalRefStorage;
} else {
AliError(Form("Invalid storage URI: %s", gridURI.Data()));
- return kFALSE;
+ return -2;
}
AliCDBManager* man = AliCDBManager::Instance();
if(!gridSto) {
Log("SHUTTLE",
Form("StoreOCDB - cannot activate main %s storage", type));
- return kFALSE;
+ return -2;
}
gridIds = gridSto->GetQueryCDBList();
if(!localSto) {
Log("SHUTTLE",
Form("StoreOCDB - cannot activate local %s storage", type));
- return kFALSE;
+ return -2;
}
AliCDBPath aPath(GetOfflineDetName(fCurrentDetector.Data()),"*","*");
// Local objects were stored with current run as Grid version!
AliCDBId aLocId = aLocEntry->GetId();
aLocEntry->SetVersion(-1);
aLocEntry->SetSubVersion(-1);
+
+ Log(fCurrentDetector.Data(), Form("Attempting to store %s", aLocId.ToString().Data()));
// If local object is valid up to infinity we store it only if it is
// the first unprocessed run!
Log("SHUTTLE", Form("StoreOCDB - %s: object %s has validity infinite but "
"there are previous unprocessed runs!",
fCurrentDetector.Data(), aLocId.GetPath().Data()));
+ Log(fCurrentDetector.Data(), Form("StoreOCDB - %s: object %s has validity infinite but "
+ "there are previous unprocessed runs!",
+ fCurrentDetector.Data(), aLocId.GetPath().Data()));
+ delayed = kTRUE;
continue;
}
Bool_t store = kTRUE;
TIter gridIter(gridIds);
AliCDBId* aGridId = 0;
- while((aGridId = dynamic_cast<AliCDBId*> (gridIter.Next()))){
- if(aGridId->GetPath() != aLocId.GetPath()) continue;
+ while ((aGridId = dynamic_cast<AliCDBId*> (gridIter.Next()))) {
+ if (aGridId->GetPath() != aLocId.GetPath())
+ continue;
// skip all objects valid up to infinity
- if(aGridId->GetLastRun() == AliCDBRunRange::Infinity()) continue;
+ if (aGridId->GetLastRun() == AliCDBRunRange::Infinity())
+ continue;
+
// if we get here, it means there's already some more recent object stored on Grid!
+ Log(fCurrentDetector.Data(),
+ Form("StoreOCDB - A more recent object already exists in %s storage: <%s>",
+ type, aGridId->ToString().Data()));
+
store = kFALSE;
break;
}
- // If we get here, the file can be stored!
- Bool_t storeOk = gridSto->Put(aLocEntry);
- if(!store || storeOk){
-
- if (!store)
- {
- Log(fCurrentDetector.Data(),
- Form("StoreOCDB - A more recent object already exists in %s storage: <%s>",
- type, aGridId->ToString().Data()));
- } else {
+ Bool_t storeOk = kFALSE;
+ if (store)
+ {
+ Log(fCurrentDetector.Data(), Form("Prechecks succeeded. Ready to store %s", aLocId.ToString().Data()));
+ storeOk = gridSto->Put(aLocEntry);
+ if (storeOk) {
Log("SHUTTLE",
- Form("StoreOCDB - Object <%s> successfully put into %s storage",
- aLocId.ToString().Data(), type));
+ Form("StoreOCDB - Object <%s> successfully put into %s storage",
+ aLocId.ToString().Data(), type));
Log(fCurrentDetector.Data(),
Form("StoreOCDB - Object <%s> successfully put into %s storage",
- aLocId.ToString().Data(), type));
+ aLocId.ToString().Data(), type));
+ } else {
+ Log("SHUTTLE",
+ Form("StoreOCDB - Grid %s storage of object <%s> failed",
+ type, aLocId.ToString().Data()));
+ Log(fCurrentDetector.Data(),
+ Form("StoreOCDB - Grid %s storage of object <%s> failed",
+ type, aLocId.ToString().Data()));
+ result = kFALSE;
}
-
- // removing local filename...
+ }
+
+ if (!store || storeOk) {
+ // removing local file...
TString filename;
localSto->IdToFilename(aLocId, filename);
Log("SHUTTLE", Form("StoreOCDB - Removing local file %s", filename.Data()));
RemoveFile(filename.Data());
- continue;
- } else {
- Log("SHUTTLE",
- Form("StoreOCDB - Grid %s storage of object <%s> failed",
- type, aLocId.ToString().Data()));
- Log(fCurrentDetector.Data(),
- Form("StoreOCDB - Grid %s storage of object <%s> failed",
- type, aLocId.ToString().Data()));
- result = kFALSE;
}
}
localEntries->Clear();
- return result;
+ Int_t returnCode = 0;
+
+ if (result == kFALSE)
+ returnCode = -1;
+ else if (delayed != kFALSE)
+ returnCode = 1;
+
+ Log("SHUTTLE", Form("StoreOCDB - Returning with %d (result = %d, delayed = %d)", returnCode, result, delayed));
+ Log(fCurrentDetector.Data(), Form("StoreOCDB - Returning with %d (result = %d, delayed = %d)", returnCode, result, delayed));
+
+ return returnCode;
}
//______________________________________________________________________________________________
TString localBaseFolder = sto->GetBaseFolder();
// Build Run level folder
- // folder = /alice/data/year/lhcPeriod/runNb/Raw
+ // folder = /alice/data/year/lhcPeriod/runNb/raw
TString lhcPeriod = GetLHCPeriod();
return 0;
}
- TString target = Form("%s/GRP/RunMetadata/alice/data/%d/%s/%09d/Raw/%s",
- localBaseFolder.Data(), GetCurrentYear(),
+ // TODO partitions with one detector only write data into LHCperiod_DET
+ TString partition = GetRunParameter("detector");
+
+ if (partition.Length() > 0 && partition != "ALICE")
+ {
+ lhcPeriod.Append(Form("_%s", partition.Data()));
+ Log(fCurrentDetector, Form("Run data tags merged file will be written in %s",
+ lhcPeriod.Data()));
+ }
+
+ TString target = Form("%s/GRP/RunMetadata%s%d/%s/%09d/raw/%s",
+ localBaseFolder.Data(), fConfig->GetAlienPath(), GetCurrentYear(),
lhcPeriod.Data(), GetCurrentRun(), gridFileName);
return CopyFileLocally(localFile, target);
void* dir = gSystem->OpenDirectory(targetDir.Data());
if (dir == NULL) {
if (gSystem->mkdir(targetDir.Data(), kTRUE)) {
- Log("SHUTTLE", Form("StoreFileLocally - Can't open directory <%s>", targetDir.Data()));
+ Log("SHUTTLE", Form("CopyFileLocally - Can't open directory <%s>", targetDir.Data()));
return kFALSE;
}
result = gSystem->GetPathInfo(localFile, 0, (Long64_t*) 0, 0, 0);
if (result)
{
- Log("SHUTTLE", Form("StoreFileLocally - %s does not exist", localFile));
+ Log("SHUTTLE", Form("CopyFileLocally - %s does not exist", localFile));
return kFALSE;
}
result = gSystem->GetPathInfo(target, 0, (Long64_t*) 0, 0, 0);
if (!result)
{
- Log("SHUTTLE", Form("StoreFileLocally - target file %s already exist, removing...", target.Data()));
+ Log("SHUTTLE", Form("CopyFileLocally - target file %s already exist, removing...", target.Data()));
if (gSystem->Unlink(target.Data()))
{
- Log("SHUTTLE", Form("StoreFileLocally - Could not remove existing target file %s!", target.Data()));
+ Log("SHUTTLE", Form("CopyFileLocally - Could not remove existing target file %s!", target.Data()));
return kFALSE;
}
}
if (result == 0)
{
- Log("SHUTTLE", Form("StoreFileLocally - File %s stored locally to %s", localFile, target.Data()));
+ Log("SHUTTLE", Form("CopyFileLocally - File %s stored locally to %s", localFile, target.Data()));
return kTRUE;
}
else
{
- Log("SHUTTLE", Form("StoreFileLocally - Could not store file %s to %s! Error code = %d",
+ Log("SHUTTLE", Form("CopyFileLocally - Could not store file %s to %s! Error code = %d",
localFile, target.Data(), result));
return kFALSE;
}
return 0;
}
- dir = Form("%s/GRP/RunMetadata/alice/data/%d/%s/%09d/Raw",
- localBaseFolder.Data(), GetCurrentYear(),
+ // TODO partitions with one detector only write data into LHCperiod_DET
+ TString partition = GetRunParameter("detector");
+
+ if (partition.Length() > 0 && partition != "ALICE")
+ {
+ lhcPeriod.Append(Form("_%s", partition.Data()));
+ }
+
+ dir = Form("%s/GRP/RunMetadata%s%d/%s/%09d/raw",
+ localBaseFolder.Data(), fConfig->GetAlienPath(), GetCurrentYear(),
lhcPeriod.Data(), GetCurrentRun());
- alienDir = dir(dir.Index("/alice/data/"), dir.Length());
+ alienDir = dir(dir.Index(fConfig->GetAlienPath()), dir.Length());
begin = "";
}
//
TString offDetStr(GetOfflineDetName(detector));
- TString dir;
+ static TString dir;
if (offDetStr == "ITS" || offDetStr == "MUON" || offDetStr == "PHOS")
{
dir.Form("%s/%s/%s", base, offDetStr.Data(), detector);
}
return dir.Data();
-
-
}
//______________________________________________________________________________________________
fStatusEntry = 0;
}
+ Int_t path1 = GetCurrentRun()/10000;
fStatusEntry = AliCDBManager::Instance()->GetStorage(GetLocalCDB())
- ->Get(Form("/SHUTTLE/STATUS/%s", fCurrentDetector.Data()), GetCurrentRun());
+ ->Get(Form("/SHUTTLE/%s/%d", fCurrentDetector.Data(), path1), GetCurrentRun());
if (!fStatusEntry) return 0;
fStatusEntry->SetOwner(1);
}
Int_t run = GetCurrentRun();
+ Int_t path1 = run/10000;
+ TString path1_string = Form("%d",path1);
- AliCDBId id(AliCDBPath("SHUTTLE", "STATUS", fCurrentDetector), run, run);
+ AliCDBId id(AliCDBPath("SHUTTLE", fCurrentDetector, path1_string), run, run);
fStatusEntry = new AliCDBEntry(status, id, new AliCDBMetaData);
fStatusEntry->SetOwner(1);
+ Int_t logLevel = AliLog::GetGlobalLogLevel();
+ AliLog::SetGlobalLogLevel(AliLog::kError);
+
UInt_t result = AliCDBManager::Instance()->GetStorage(fgkLocalCDB)->Put(fStatusEntry);
- if (!result) {
+ AliLog::SetGlobalLogLevel((AliLog::EType_t)logLevel);
+
+ if (!result) {
Log("SHUTTLE", Form("WriteShuttleStatus - Failed for %s, run %d",
fCurrentDetector.Data(), run));
return kFALSE;
}
- SendMLInfo();
+ SendMLDetInfo();
return kTRUE;
}
status->SetStatus(newStatus);
if (increaseCount) status->IncreaseCount();
+ Int_t logLevel = AliLog::GetGlobalLogLevel();
+ AliLog::SetGlobalLogLevel(AliLog::kError);
+
AliCDBManager::Instance()->GetStorage(fgkLocalCDB)->Put(fStatusEntry);
- SendMLInfo();
+ AliLog::SetGlobalLogLevel((AliLog::EType_t)logLevel);
+
+ SendMLDetInfo();
}
//______________________________________________________________________________________________
-void AliShuttle::SendMLInfo()
+void AliShuttle::SendMLDetInfo()
{
//
// sends ML information about the current status of the current detector being processed
AliShuttleStatus* status = dynamic_cast<AliShuttleStatus*> (fStatusEntry->GetObject());
if (!status){
- Log("SHUTTLE", "SendMLInfo - UNEXPECTED: status could not be read from current CDB entry");
+ Log("SHUTTLE", "SendMLDetInfo - UNEXPECTED: status could not be read from current CDB entry");
return;
}
// checks if the processing should be continued
// if yes it returns kTRUE and updates the AliShuttleStatus with nextStatus
- if (!fConfig->HostProcessDetector(fCurrentDetector)) return kFALSE;
+ if (!fConfig->HostProcessDetector(fCurrentDetector))
+ return kFALSE;
AliPreprocessor* aPreprocessor =
dynamic_cast<AliPreprocessor*> (fPreprocessorMap.GetValue(fCurrentDetector));
AliShuttleLogbookEntry::Status entryStatus =
fLogbookEntry->GetDetectorStatus(fCurrentDetector);
- if(entryStatus != AliShuttleLogbookEntry::kUnprocessed) {
+ if (entryStatus != AliShuttleLogbookEntry::kUnprocessed) {
Log("SHUTTLE", Form("ContinueProcessing - %s is %s",
fCurrentDetector.Data(),
fLogbookEntry->GetDetectorStatusName(entryStatus)));
if (fTestMode == kNone)
{
Log("SHUTTLE", Form("ContinueProcessing - %s requires strict run ordering"
- " but this is not the first unprocessed run!"));
+ " but this is not the first unprocessed run!",fCurrentDetector.Data()));
return kFALSE;
}
else
Log("SHUTTLE", Form("ContinueProcessing - In TESTMODE - "
"Although %s requires strict run ordering "
"and this is not the first unprocessed run, "
- "the SHUTTLE continues"));
+ "the SHUTTLE continues",fCurrentDetector.Data()));
}
}
+ // Is the subdetector processed first time for this run?
+ fFirstProcessing = kFALSE;
+
AliShuttleStatus* status = ReadShuttleStatus();
if (!status) {
// first time
Log("SHUTTLE", Form("ContinueProcessing - %s: Processing first time",
fCurrentDetector.Data()));
status = new AliShuttleStatus(AliShuttleStatus::kStarted);
+ fFirstProcessing = kTRUE;
return WriteShuttleStatus(status);
}
- // The following two cases shouldn't happen if Shuttle Logbook was correctly updated.
+ // The following case shouldn't happen if Shuttle Logbook was correctly updated.
// If it happens it may mean Logbook updating failed... let's do it now!
if (status->GetStatus() == AliShuttleStatus::kDone ||
- status->GetStatus() == AliShuttleStatus::kFailed){
+ status->GetStatus() == AliShuttleStatus::kFailed ||
+ status->GetStatus() == AliShuttleStatus::kSkipped) {
Log("SHUTTLE", Form("ContinueProcessing - %s is already %s. Updating Shuttle Logbook",
fCurrentDetector.Data(),
status->GetStatusName(status->GetStatus())));
- UpdateShuttleLogbook(fCurrentDetector.Data(),
- status->GetStatusName(status->GetStatus()));
+
+ if (status->GetStatus() == AliShuttleStatus::kSkipped)
+ {
+ UpdateShuttleLogbook(fCurrentDetector.Data(), "DONE");
+ }
+ else
+ UpdateShuttleLogbook(fCurrentDetector.Data(), status->GetStatusName(status->GetStatus()));
+
return kFALSE;
}
- if (status->GetStatus() == AliShuttleStatus::kStoreError) {
+ if (status->GetStatus() == AliShuttleStatus::kStoreStarted || status->GetStatus() == AliShuttleStatus::kStoreDelayed ||status->GetStatus() == AliShuttleStatus::kStoreError) {
Log("SHUTTLE",
Form("ContinueProcessing - %s: Grid storage of one or more "
"objects failed. Trying again now",
fCurrentDetector.Data()));
- UpdateShuttleStatus(AliShuttleStatus::kStoreStarted);
- if (StoreOCDB()){
- Log("SHUTTLE", Form("ContinueProcessing - %s: all objects "
- "successfully stored into main storage",
- fCurrentDetector.Data()));
- } else {
- Log("SHUTTLE",
- Form("ContinueProcessing - %s: Grid storage failed again",
- fCurrentDetector.Data()));
- UpdateShuttleStatus(AliShuttleStatus::kStoreError);
- }
+ StoreOCDB();
return kFALSE;
}
// Send mail to detector expert!
Log("SHUTTLE", Form("ContinueProcessing - Sending mail to %s expert...",
- fCurrentDetector.Data()));
- if (!SendMail())
+ fCurrentDetector.Data()));
+ // det experts in to
+ TString to="";
+ TIter *iterExperts = 0;
+ iterExperts = new TIter(fConfig->GetResponsibles(fCurrentDetector));
+ TObjString *anExpert=0;
+ while ((anExpert = (TObjString*) iterExperts->Next()))
+ {
+ to += Form("%s, \n", anExpert->GetName());
+ }
+ delete iterExperts;
+
+ if (to.Length() > 0)
+ to.Remove(to.Length()-3);
+ AliDebug(2, Form("to: %s",to.Data()));
+
+ if (to.IsNull()) {
+ Log("SHUTTLE", Form("List of %s responsibles not set!", fCurrentDetector.Data()));
+ return kFALSE;
+ }
+
+ Log(fCurrentDetector.Data(), Form("ContinueProcessing - Sending mail to %s expert(s):",
+ fCurrentDetector.Data()));
+ Log(fCurrentDetector.Data(), Form("\n%s", to.Data()));
+ if (!SendMail(kPPEMail))
Log("SHUTTLE", Form("ContinueProcessing - Could not send mail to %s expert",
- fCurrentDetector.Data()));
+ fCurrentDetector.Data()));
} else {
Log("SHUTTLE", Form("ContinueProcessing - %s: restarting. "
status->GetStatusName(), status->GetCount()));
Bool_t increaseCount = kTRUE;
if (status->GetStatus() == AliShuttleStatus::kDCSError ||
- status->GetStatus() == AliShuttleStatus::kDCSStarted)
+ status->GetStatus() == AliShuttleStatus::kDCSStarted ||
+ status->GetStatus() == AliShuttleStatus::kFXSError ||
+ status->GetStatus() == AliShuttleStatus::kOCDBError)
increaseCount = kFALSE;
UpdateShuttleStatus(AliShuttleStatus::kStarted, increaseCount);
return cont;
}
+//______________________________________________________________________________________________
+void AliShuttle::SendMLRunInfo(const char* status)
+{
+ //
+ // Send information about this run to ML
+
+ TMonaLisaText mlStatus("SHUTTLE_status", status);
+ TString runType(fLogbookEntry->GetRunType());
+ if (strlen(fLogbookEntry->GetRunParameter("log")) > 0){
+
+ runType += "(";
+ runType += fLogbookEntry->GetRunParameter("log");
+ runType += ")";
+ }
+ if (fLogbookEntry->GetDATestMode()){
+ runType += " (DATest)";
+ }
+ TMonaLisaText mlRunType("SHUTTLE_runtype", runType);
+
+ TList mlList;
+ mlList.Add(&mlStatus);
+ mlList.Add(&mlRunType);
+
+ TString mlID;
+ mlID.Form("%d", GetCurrentRun());
+ fMonaLisa->SendParameters(&mlList, mlID);
+}
+
+//______________________________________________________________________________________________
+Int_t AliShuttle::GetMem(Int_t pid)
+{
+ // invokes ps to get the memory consumption of the process <pid>
+ // returns -1 in case of error
+
+ TString checkStr;
+ checkStr.Form("ps -o vsize --pid %d | tail -n 1", pid);
+ FILE* pipe = gSystem->OpenPipe(checkStr, "r");
+ if (!pipe)
+ {
+ Log("SHUTTLE", Form("Process - Error: "
+ "Could not open pipe to %s", checkStr.Data()));
+ return -1;
+ }
+
+ char buffer[100];
+ if (!fgets(buffer, 100, pipe))
+ {
+ Log("SHUTTLE", "Process - Error: ps did not return anything");
+ gSystem->ClosePipe(pipe);
+ return -1;
+ }
+ gSystem->ClosePipe(pipe);
+
+ //Log("SHUTTLE", Form("ps returned %s", buffer));
+
+ Int_t mem = 0;
+ if ((sscanf(buffer, "%d\n", &mem) != 1) || !mem)
+ {
+ Log("SHUTTLE", "Process - Error: Could not parse output of ps");
+ return -1;
+ }
+
+ return mem;
+}
+
//______________________________________________________________________________________________
Bool_t AliShuttle::Process(AliShuttleLogbookEntry* entry)
{
Log("SHUTTLE", Form("\t\t\t^*^*^*^*^*^*^*^*^*^*^*^* run %d: START ^*^*^*^*^*^*^*^*^*^*^*^*",
GetCurrentRun()));
+ CountOpenRuns();
+
// Send the information to ML
- TMonaLisaText mlStatus("SHUTTLE_status", "Processing");
- TMonaLisaText mlRunType("SHUTTLE_runtype", Form("%s (%s)", entry->GetRunType(), entry->GetRunParameter("log")));
-
- TList mlList;
- mlList.Add(&mlStatus);
- mlList.Add(&mlRunType);
-
- TString mlID;
- mlID.Form("%d", GetCurrentRun());
- fMonaLisa->SendParameters(&mlList, mlID);
+ SendMLRunInfo("Processing");
if (fLogbookEntry->IsDone())
{
// Initialization
Bool_t hasError = kFALSE;
- // Set the CDB and Reference folders according to the year and LHC period
- TString lhcPeriod(GetLHCPeriod());
- if (lhcPeriod.Length() == 0)
- {
- Log("SHUTTLE","Process - LHCPeriod not found in logbook!");
- return 0;
- }
-
- if (fgkMainCDB.Length() == 0)
- fgkMainCDB = Form("alien://folder=/alice/data/%d/%s/OCDB?user=alidaq?cacheFold=/tmp/OCDBCache",
- GetCurrentYear(), lhcPeriod.Data());
-
- if (fgkMainRefStorage.Length() == 0)
- fgkMainRefStorage = Form("alien://folder=/alice/data/%d/%s/Reference?user=alidaq?cacheFold=/tmp/OCDBCache",
- GetCurrentYear(), lhcPeriod.Data());
-
- AliCDBStorage *mainCDBSto = AliCDBManager::Instance()->GetStorage(fgkMainCDB);
- if(mainCDBSto) mainCDBSto->QueryCDB(GetCurrentRun());
- AliCDBStorage *mainRefSto = AliCDBManager::Instance()->GetStorage(fgkMainRefStorage);
- if(mainRefSto) mainRefSto->QueryCDB(GetCurrentRun());
+ // Set the CDB and Reference folders according to the year
+
+ // build cdb paths (repeat each time, run might be a DATest run)
+ if (!fLogbookEntry->GetDATestMode()){
+ fgkMainCDB.Form("alien://folder=%s%d/OCDB?user=alidaq?cacheFold=/tmp/OCDBCache",
+ fConfig->GetAlienPath(), GetCurrentYear());
+
+ fgkMainRefStorage.Form("alien://folder=%s%d/Reference?user=alidaq?cacheFold=/tmp/OCDBCache",
+ fConfig->GetAlienPath(), GetCurrentYear());
+ }
+ else {
+ fgkMainCDB.Form("alien://folder=%s%d/DATest/OCDB?user=alidaq?cacheFold=/tmp/OCDBCache",
+ fConfig->GetAlienPath(), GetCurrentYear());
+
+ fgkMainRefStorage.Form("alien://folder=%s%d/DATest/Reference?user=alidaq?cacheFold=/tmp/OCDBCache",
+ fConfig->GetAlienPath(), GetCurrentYear());
+ }
+
+ AliDebug(2,Form("Main CDB storage = %s",fgkMainCDB.Data()));
+ AliDebug(2,Form("Main Reference storage = %s",fgkMainRefStorage.Data()));
// Loop on detectors in the configuration
TIter iter(fConfig->GetDetectors());
TObjString* aDetector = 0;
+ Bool_t first = kTRUE;
+
while ((aDetector = (TObjString*) iter.Next()))
{
fCurrentDetector = aDetector->String();
- if (ContinueProcessing() == kFALSE) continue;
+ if (ContinueProcessing() == kFALSE)
+ continue;
+
+ if (first)
+ {
+ // only read QueryCDB when needed and only once
+ AliCDBStorage *mainCDBSto = AliCDBManager::Instance()->GetStorage(fgkMainCDB);
+ if(mainCDBSto) mainCDBSto->QueryCDB(GetCurrentRun());
+ AliCDBStorage *mainRefSto = AliCDBManager::Instance()->GetStorage(fgkMainRefStorage);
+ if(mainRefSto) mainRefSto->QueryCDB(GetCurrentRun());
+ first = kFALSE;
+ }
Log("SHUTTLE", Form("\t\t\t****** run %d - %s: START ******",
GetCurrentRun(), aDetector->GetName()));
for(Int_t iSys=0;iSys<3;iSys++) fFXSCalled[iSys]=kFALSE;
+
+ Int_t initialMem = GetMem(getpid());
+ Log("SHUTTLE", Form("Memory consumption before forking is %d", initialMem));
Log(fCurrentDetector.Data(), "Process - Starting processing");
if (expiredTime > fConfig->GetPPTimeOut())
{
- TString tmp;
- tmp.Form("Process - Process of %s time out. "
- "Run time: %d seconds. Killing...",
- fCurrentDetector.Data(), expiredTime);
- Log("SHUTTLE", tmp);
- Log(fCurrentDetector, tmp);
+ TString logMsg;
+ AliShuttleStatus *currentStatus = ReadShuttleStatus();
+ AliShuttleStatus::Status newStatus = AliShuttleStatus::kInvalid;
+
+ if (currentStatus->GetStatus() == AliShuttleStatus::kDCSStarted)
+ {
+ // in case the pp goes in TimeOut while retrieving the DCS DPs
+ // set status to kDCSError
+
+ logMsg.Form("Process - Process of %s timed out while retrieving the DCS DataPoints. Run time: %ld seconds. Killing... and setting status to DCSError.",
+ fCurrentDetector.Data(), expiredTime);
+ newStatus = AliShuttleStatus::kDCSError;
+ }
+ else if (currentStatus->GetStatus() <= AliShuttleStatus::kPPDone)
+ {
+ // in case pp not yet done set status to kPPTimeOut
+
+ logMsg.Form("Process - Process of %s timed out. Run time: %ld seconds. Killing...",
+ fCurrentDetector.Data(), expiredTime);
+ newStatus = AliShuttleStatus::kPPTimeOut;
+ }
+ else if (currentStatus->GetStatus() == AliShuttleStatus::kStoreStarted)
+ {
+ // in case the pp goes in TimeOut while storing the objects in the OCDB
+ // set status to kStoreError
+
+ logMsg.Form("Process - Process of %s timed out while storing the OCDB object. Run time: %ld seconds. Killing... and setting status to StoreError.",
+ fCurrentDetector.Data(), expiredTime);
+ newStatus = AliShuttleStatus::kStoreError;
+ }
+ else
+ {
+ // in other cases don't change the status
+
+ logMsg.Form("Process - Process of %s timed out in status = %s. Run time: %ld seconds. Killing... without changing the status",
+ fCurrentDetector.Data(), currentStatus->GetStatusName(), expiredTime);
+ }
+
+ Log("SHUTTLE", logMsg);
+ Log(fCurrentDetector, logMsg);
kill(pid, 9);
- UpdateShuttleStatus(AliShuttleStatus::kPPTimeOut);
+ if (newStatus != AliShuttleStatus::kInvalid)
+ UpdateShuttleStatus(newStatus);
hasError = kTRUE;
gSystem->Sleep(1000);
{
gSystem->Sleep(1000);
- TString checkStr;
- checkStr.Form("ps -o vsize --pid %d | tail -n 1", pid);
- FILE* pipe = gSystem->OpenPipe(checkStr, "r");
- if (!pipe)
- {
- Log("SHUTTLE", Form("Process - Error: "
- "Could not open pipe to %s", checkStr.Data()));
+ Int_t mem = GetMem(pid);
+
+ if (mem < 0)
continue;
- }
- char buffer[100];
- if (!fgets(buffer, 100, pipe))
+ mem -= initialMem;
+ if (mem < 0)
+ mem = 0;
+
+ if (expiredTime % 60 == 0)
{
- Log("SHUTTLE", "Process - Error: ps did not return anything");
- gSystem->ClosePipe(pipe);
- continue;
- }
- gSystem->ClosePipe(pipe);
-
- //Log("SHUTTLE", Form("ps returned %s", buffer));
-
- Int_t mem = 0;
- if ((sscanf(buffer, "%d\n", &mem) != 1) || !mem)
- {
- Log("SHUTTLE", "Process - Error: Could not parse output of ps");
- continue;
- }
-
- if (expiredTime % 60 == 0)
- {
- Log("SHUTTLE", Form("Process - %s: Checking process. "
- "Run time: %d seconds - Memory consumption: %d KB",
- fCurrentDetector.Data(), expiredTime, mem));
- SendAlive();
+ Log("SHUTTLE", Form("Process - %s: Checking process. "
+ "Run time: %ld seconds - Memory consumption: %d KB",
+ fCurrentDetector.Data(), expiredTime, mem));
+ SendAlive();
}
if (mem > fConfig->GetPPMaxMem())
}
else if (pid == 0)
{
- // client
- Log("SHUTTLE", Form("Process - In client process of %d - %s", GetCurrentRun(),
+ // child
+ Log("SHUTTLE", Form("Process - In child process of %d - %s", GetCurrentRun(),
aDetector->GetName()));
Log("SHUTTLE", Form("Process - Redirecting output to %s log",fCurrentDetector.Data()));
Log("SHUTTLE", "Process - Could not redirect stderr");
}
+
+ Log("SHUTTLE", "Executing TGrid::Connect");
+ TGrid::Connect("alien://");
TString wd = gSystem->WorkingDirectory();
- TString tmpDir = Form("%s/%s_%d_process", GetShuttleTempDir(),
- fCurrentDetector.Data(), GetCurrentRun());
+ Int_t dir_lev1 = GetCurrentRun()/10000;
+ TString tmpDir = Form("%s/%d/%d/%s_process", GetShuttleTempDir(),
+ dir_lev1, GetCurrentRun(), fCurrentDetector.Data());
Int_t result = gSystem->GetPathInfo(tmpDir.Data(), 0, (Long64_t*) 0, 0, 0);
if (!result) // temp dir already exists!
gSystem->Exit(1);
}
- Bool_t success = ProcessCurrentDetector();
-
+ Int_t success = ProcessCurrentDetector();
+
gSystem->ChangeDirectory(wd.Data());
- if (success) // Preprocessor finished successfully!
+ if (success == 1) // Preprocessor finished successfully!
{
- // remove temporary folder
- gSystem->Exec(Form("rm -rf %s",tmpDir.Data()));
+ // remove temporary folder or DCS map
+ if (!fConfig->KeepTempFolder())
+ {
+ gSystem->Exec(Form("rm -rf %s",tmpDir.Data()));
+ } else if (!fConfig->KeepDCSMap())
+ {
+ gSystem->Exec(Form("rm -f %s/DCSMap.root",tmpDir.Data()));
+ }
// Update time_processed field in FXS DB
if (UpdateTable() == kFALSE)
fCurrentDetector.Data()));
// Transfer the data from local storage to main storage (Grid)
- UpdateShuttleStatus(AliShuttleStatus::kStoreStarted);
if (StoreOCDB() == kFALSE)
- {
- Log("SHUTTLE",
- Form("\t\t\t****** run %d - %s: STORAGE ERROR ******",
- GetCurrentRun(), aDetector->GetName()));
- UpdateShuttleStatus(AliShuttleStatus::kStoreError);
success = kFALSE;
- } else {
- Log("SHUTTLE",
- Form("\t\t\t****** run %d - %s: DONE ******",
- GetCurrentRun(), aDetector->GetName()));
- UpdateShuttleStatus(AliShuttleStatus::kDone);
- UpdateShuttleLogbook(fCurrentDetector, "DONE");
- }
- } else
+ }
+ else if (success == 0)
{
Log("SHUTTLE",
- Form("\t\t\t****** run %d - %s: PP ERROR ******",
+ Form("\t\t\t****** run %d - %s: ERROR ******",
GetCurrentRun(), aDetector->GetName()));
}
fFirstUnprocessed[iDet] = kFALSE;
}
}
+ SendMLRunInfo("Pending");
}
}
}
//______________________________________________________________________________________________
-Bool_t AliShuttle::ProcessCurrentDetector()
+Int_t AliShuttle::ProcessCurrentDetector()
{
//
// Makes data retrieval just for a specific detector (fCurrentDetector).
TString wd = gSystem->WorkingDirectory();
if (!CleanReferenceStorage(fCurrentDetector.Data()))
- return kFALSE;
+ return 0;
gSystem->ChangeDirectory(wd.Data());
- TMap* dcsMap = new TMap();
-
// call preprocessor
AliPreprocessor* aPreprocessor =
dynamic_cast<AliPreprocessor*> (fPreprocessorMap.GetValue(fCurrentDetector));
+ // check if the preprocessor wants to process this run type
+ if (aPreprocessor->ProcessRunType() == kFALSE)
+ {
+ UpdateShuttleStatus(AliShuttleStatus::kSkipped);
+ UpdateShuttleLogbook(fCurrentDetector, "DONE");
+ if (!UpdateTableSkippedCase(fCurrentDetector.Data()))
+ {
+ AliError(Form("Could not update FXS tables for run %d !", GetCurrentRun()));
+ }
+ Log(fCurrentDetector, Form("ProcessCurrentDetector - %s preprocessor is not interested in this run type", fCurrentDetector.Data()));
+
+ return 2;
+ }
+
+ // checking if OCDB is reachable
+ AliCDBEntry* testEntry = GetFromOCDB("SHUTTLE","GRP/CTP/DummyConfig");
+ if (!testEntry){
+ // OCDB is not accessible, going in OCDBError for current detector
+ AliError("OCDB Test entry not accessible");
+ UpdateShuttleStatus(AliShuttleStatus::kOCDBError);
+ return 0;
+ }
+
+ TMap* dcsMap = new TMap();
+
aPreprocessor->Initialize(GetCurrentRun(), GetCurrentStartTime(), GetCurrentEndTime());
Bool_t processDCS = aPreprocessor->ProcessDCS();
UpdateShuttleStatus(AliShuttleStatus::kDCSStarted);
UpdateShuttleStatus(AliShuttleStatus::kDCSError);
delete dcsMap;
- return kFALSE;
+ return 0;
} else {
UpdateShuttleStatus(AliShuttleStatus::kDCSStarted);
TMap* aliasMap = 0;
TMap* dpMap = 0;
-
+
if (fConfig->GetDCSAliases(fCurrentDetector, iServ)->GetEntries() > 0)
{
+ Log(fCurrentDetector, Form("Querying %d DCS aliases", fConfig->GetDCSAliases(fCurrentDetector, iServ)->GetEntries()));
aliasMap = GetValueSet(host, port,
fConfig->GetDCSAliases(fCurrentDetector, iServ),
kAlias, multiSplit);
" Sending mail to DCS experts!", host.Data()));
UpdateShuttleStatus(AliShuttleStatus::kDCSError);
- if (!SendMailToDCS())
- Log("SHUTTLE", Form("ProcessCurrentDetector - Could not send mail to DCS experts!"));
+ if (!SendMail(kDCSEMail))
+ Log("SHUTTLE", Form("ProcessCurrentDetector - "
+ "Could not send mail to DCS experts!"));
delete dcsMap;
- return kFALSE;
+ return 0;
}
}
if (fConfig->GetDCSDataPoints(fCurrentDetector, iServ)->GetEntries() > 0)
{
+ Log(fCurrentDetector, Form("Querying %d DCS data points", fConfig->GetDCSDataPoints(fCurrentDetector, iServ)->GetEntries()));
dpMap = GetValueSet(host, port,
fConfig->GetDCSDataPoints(fCurrentDetector, iServ),
kDP, multiSplit);
" Sending mail to DCS experts!", host.Data()));
UpdateShuttleStatus(AliShuttleStatus::kDCSError);
- if (!SendMailToDCS())
- Log("SHUTTLE", Form("ProcessCurrentDetector - Could not send mail to DCS experts!"));
+ if (!SendMail(kDCSEMail))
+ Log("SHUTTLE", Form("ProcessCurrentDetector - "
+ "Could not send mail to DCS experts!"));
if (aliasMap) delete aliasMap;
delete dcsMap;
- return kFALSE;
+ return 0;
}
}
// DCS Archive DB processing successful. Call Preprocessor!
UpdateShuttleStatus(AliShuttleStatus::kPPStarted);
+ fFXSError = -1; // this variable is kTRUE after ::Process if an FXS error occured
+
UInt_t returnValue = aPreprocessor->Process(dcsMap);
+
+ if (fFXSError!=-1) {
+ UpdateShuttleStatus(AliShuttleStatus::kFXSError);
+ SendMail(kFXSEMail, fFXSError);
+ dcsMap->DeleteAll();
+ delete dcsMap;
+ return 0;
+ }
if (returnValue > 0) // Preprocessor error!
{
UpdateShuttleStatus(AliShuttleStatus::kPPError);
dcsMap->DeleteAll();
delete dcsMap;
- return kFALSE;
+ return 0;
}
// preprocessor ok!
dcsMap->DeleteAll();
delete dcsMap;
- return kTRUE;
+ return 1;
+}
+
+//______________________________________________________________________________________________
+void AliShuttle::CountOpenRuns()
+{
+ // Query DAQ's Shuttle logbook and sends the number of open runs to ML
+
+ SendAlive();
+
+ // check connection, in case connect
+ if (!Connect(3))
+ return;
+
+ TString sqlQuery;
+ sqlQuery = Form("select count(*) from %s where shuttle_done=0", fConfig->GetShuttlelbTable());
+
+ TSQLResult* aResult = fServer[3]->Query(sqlQuery);
+ if (!aResult) {
+ AliError(Form("Can't execute query <%s>!", sqlQuery.Data()));
+ return;
+ }
+
+ AliDebug(2,Form("Query = %s", sqlQuery.Data()));
+
+ if (aResult->GetRowCount() == 0) {
+ AliError(Form("No result for query %s received", sqlQuery.Data()));
+ return;
+ }
+
+ if (aResult->GetFieldCount() != 1) {
+ AliError(Form("Invalid field count for query %s received", sqlQuery.Data()));
+ return;
+ }
+
+ TSQLRow* aRow = aResult->Next();
+ if (!aRow) {
+ AliError(Form("Could not receive result of query %s", sqlQuery.Data()));
+ return;
+ }
+
+ TString result(aRow->GetField(0), aRow->GetFieldLength(0));
+ Int_t count = result.Atoi();
+
+ Log("SHUTTLE", Form("%d unprocessed runs", count));
+
+ delete aRow;
+ delete aResult;
+
+ TMonaLisaValue mlStatus("SHUTTLE_openruns", count);
+
+ TList mlList;
+ mlList.Add(&mlStatus);
+
+ fMonaLisa->SendParameters(&mlList, "__PROCESSINGINFO__");
}
//______________________________________________________________________________________________
entries.SetOwner(1);
// check connection, in case connect
- if(!Connect(3)) return kFALSE;
+ if (!Connect(3)) return kFALSE;
TString sqlQuery;
sqlQuery = Form("select * from %s %s order by run", fConfig->GetShuttlelbTable(), whereClause);
}
// TODO Check field count!
- const UInt_t nCols = 23;
+ const UInt_t nCols = 26;
if (aResult->GetFieldCount() != (Int_t) nCols) {
Log("SHUTTLE", "Invalid SQL result field number!");
delete aResult;
if (!entry)
continue;
+ // DA test mode flag
+ TString daTestModeString(aRow->GetField(2), aRow->GetFieldLength(2)); // field 2 = DA test mode flag
+ Bool_t daTestMode = (Bool_t)daTestModeString.Atoi();
+ entry->SetDATestMode(daTestMode);
+
// loop on detectors
for(UInt_t ii = 0; ii < nCols; ii++)
entry->SetDetectorStatus(aResult->GetFieldName(ii), aRow->GetField(ii));
for (Int_t ii = 0; ii < aResult->GetFieldCount(); ii++)
entry->SetRunParameter(aResult->GetFieldName(ii), aRow->GetField(ii));
+ delete aRow;
+ delete aResult;
+
UInt_t startTime = entry->GetStartTime();
UInt_t endTime = entry->GetEndTime();
+ Bool_t ecsSuccess = entry->GetECSSuccess();
+ TString runType = entry->GetRunType();
+ TString tmpdaqstartTime = entry->GetRunParameter("DAQ_time_start");
+ TString recordingFlagString = entry->GetRunParameter("GDCmStreamRecording");
+ UInt_t recordingFlag = recordingFlagString.Atoi();
+ UInt_t daqstartTime = tmpdaqstartTime.Atoi();
+
+ UInt_t now = time(0);
+ Int_t dcsDelay = fConfig->GetDCSDelay()+fConfig->GetDCSQueryOffset();
-// if (!startTime || !endTime || startTime > endTime)
-// {
-// Log("SHUTTLE",
-// Form("QueryRunParameters - Invalid parameters for Run %d: startTime = %d, endTime = %d. Skipping!",
-// run, startTime, endTime));
-//
-// Log("SHUTTLE", Form("Marking SHUTTLE done for run %d", run));
-// fLogbookEntry = entry;
-// if (!UpdateShuttleLogbook("shuttle_done"))
-// {
-// AliError(Form("Could not update logbook for run %d !", run));
-// }
-// fLogbookEntry = 0;
-//
-// delete entry;
-// delete aRow;
-// delete aResult;
-// return 0;
-// }
-
- if (!startTime)
- {
- Log("SHUTTLE",
- Form("QueryRunParameters - Invalid parameters for Run %d: "
- "startTime = %d, endTime = %d. Skipping!",
- run, startTime, endTime));
-
- Log("SHUTTLE", Form("Marking SHUTTLE done for run %d", run));
- fLogbookEntry = entry;
- if (!UpdateShuttleLogbook("shuttle_ignored"))
- {
- AliError(Form("Could not update logbook for run %d !", run));
- }
- fLogbookEntry = 0;
+ Bool_t skip = kFALSE;
- delete entry;
- delete aRow;
- delete aResult;
- return 0;
+ // runs are processed if
+ // a) runType is PHYSICS and ecsSuccess is set
+ // b) runType is not PHYSICS and (ecsSuccess is set or DAQ_time_start is non-0)
+ // effectively this means that all runs are processed that started properly (ecsSucess behaviour is different for PHYSICS and non-PHYSICS runs (check with ECS!)
+ if (startTime != 0 && endTime != 0) {
+ if (endTime > startTime) {
+ if (endTime >= now - dcsDelay) {
+ Log("SHUTTLE", Form("Skipping run %d for now, because DCS buffer time is not yet expired", run));
+ } else {
+ if ((runType == "PHYSICS" || runType == "STANDALONE") && recordingFlag == 0){
+ Log("SHUTTLE", Form("QueryRunParameters - Run type for run %d is %s but the recording is OFF - Skipping!", run, runType.Data()));
+ skip = kTRUE;
+ }
+ else {
+ if (runType == "PHYSICS") {
+ if (ecsSuccess) {
+ return entry;
+ } else {
+ Log("SHUTTLE", Form("QueryRunParameters - Run type for run %d is PHYSICS but ECS success flag not set (Reason = %s) - Skipping!", run, entry->GetRunParameter("eor_reason")));
+ skip = kTRUE;
+ }
+ } else {
+ if (ecsSuccess || daqstartTime > 0) {
+ if (ecsSuccess == kFALSE)
+ Log("SHUTTLE", Form("Processing run %d although in status ECS failure (Reason: %s), since run type != PHYSICS and DAQ_time_start != 0", run, entry->GetRunParameter("eor_reason")));
+ return entry;
+ } else {
+ Log("SHUTTLE", Form("QueryRunParameters - Run type for run %d is %s, ECS success flag was not set (Reason = %s) and DAQ_time_start was NULL - Skipping!", run, runType.Data(), entry->GetRunParameter("eor_reason")));
+ skip = kTRUE;
+ }
+ }
+ }
+ }
+ } else {
+ Log("SHUTTLE", Form("QueryRunParameters - Invalid parameters for run %d: startTime equal to endTime: %d %d - Skipping!", run, startTime, endTime));
+ skip = kTRUE;
+ }
+ } else {
+ Log("SHUTTLE", Form("QueryRunParameters - Invalid parameters for Run %d: "
+ "startTime = %d, endTime = %d. Skipping (Shuttle won't be marked as DONE)!",
+ run, startTime, endTime));
}
- if (startTime && !endTime)
+ if (skip)
{
- // TODO Here we don't mark SHUTTLE done, because this may mean
- //the run is still ongoing!!
- Log("SHUTTLE",
- Form("QueryRunParameters - Invalid parameters for Run %d: "
- "startTime = %d, endTime = %d. Skipping (Shuttle won't be marked as DONE)!",
- run, startTime, endTime));
-
- //Log("SHUTTLE", Form("Marking SHUTTLE done for run %d", run));
- //fLogbookEntry = entry;
- //if (!UpdateShuttleLogbook("shuttle_done"))
- //{
- // AliError(Form("Could not update logbook for run %d !", run));
- //}
- //fLogbookEntry = 0;
-
- delete entry;
- delete aRow;
- delete aResult;
- return 0;
- }
-
- if (startTime && endTime && (startTime > endTime))
- {
- Log("SHUTTLE",
- Form("QueryRunParameters - Invalid parameters for Run %d: "
- "startTime = %d, endTime = %d. Skipping!",
- run, startTime, endTime));
-
- Log("SHUTTLE", Form("Marking SHUTTLE done for run %d", run));
- fLogbookEntry = entry;
- if (!UpdateShuttleLogbook("shuttle_ignored"))
+ Log("SHUTTLE", Form("Marking SHUTTLE skipped for run %d", run));
+ fLogbookEntry = entry;
+ if (!UpdateShuttleLogbook("shuttle_skipped"))
{
AliError(Form("Could not update logbook for run %d !", run));
}
- fLogbookEntry = 0;
-
- delete entry;
- delete aRow;
- delete aResult;
- return 0;
- }
-
- TString totEventsStr = entry->GetRunParameter("totalEvents");
- Int_t totEvents = totEventsStr.Atoi();
- if (totEvents < 1)
- {
- Log("SHUTTLE",
- Form("QueryRunParameters - Run %d has 0 events - Skipping!", run));
-
- Log("SHUTTLE", Form("Marking SHUTTLE done for run %d", run));
- fLogbookEntry = entry;
- if (!UpdateShuttleLogbook("shuttle_ignored"))
+ if (!UpdateTableSkippedCase("ALL"))
{
- AliError(Form("Could not update logbook for run %d !", run));
+ AliError(Form("Could not update FXS tables for run %d !", run));
}
fLogbookEntry = 0;
-
- delete entry;
- delete aRow;
- delete aResult;
- return 0;
}
-
- delete aRow;
- delete aResult;
-
- return entry;
+
+ delete entry;
+ return 0;
}
//______________________________________________________________________________________________
TMap* result = 0;
if (type == kAlias)
{
- result = client.GetAliasValues(entries, GetCurrentStartTime(),
- GetCurrentEndTime());
+ //result = client.GetAliasValues(entries, GetCurrentStartTime()-offset,
+ // GetCurrentEndTime()+offset);
+ result = client.GetAliasValues(entries, GetStartTimeDCSQuery(),
+ GetEndTimeDCSQuery());
}
else if (type == kDP)
{
- result = client.GetDPValues(entries, GetCurrentStartTime(),
- GetCurrentEndTime());
+ //result = client.GetDPValues(entries, GetCurrentStartTime()-offset,
+ // GetCurrentEndTime()+offset);
+ result = client.GetDPValues(entries, GetStartTimeDCSQuery(),
+ GetEndTimeDCSQuery());
}
if (result == 0)
if (!Connect(system))
{
Log(detector, Form("GetFile - Couldn't connect to %s FXS database", GetSystemName(system)));
+ fFXSError = system;
return 0;
}
else if (system == kHLT)
{
whereClause += Form(" and DDLnumbers=\"%s\"", source);
- nFields = 3;
}
TString sqlQuery = Form("%s %s", sqlQueryStart.Data(), whereClause.Data());
TSQLResult* aResult = 0;
aResult = dynamic_cast<TSQLResult*> (fServer[system]->Query(sqlQuery));
if (!aResult) {
- Log(detector, Form("GetFileName - Can't execute SQL query to %s database for: id = %s, source = %s",
+ Log(detector, Form("GetFile - Can't execute SQL query to %s database for: id = %s, source = %s",
GetSystemName(system), id, sourceName.Data()));
+ fFXSError = system;
return 0;
}
- if(aResult->GetRowCount() == 0)
+ if (aResult->GetRowCount() == 0)
{
Log(detector,
- Form("GetFileName - No entry in %s FXS db for: id = %s, source = %s",
+ Form("GetFile - No entry in %s FXS db for: id = %s, source = %s",
GetSystemName(system), id, sourceName.Data()));
delete aResult;
return 0;
if (aResult->GetRowCount() > 1) {
Log(detector,
- Form("GetFileName - More than one entry in %s FXS db for: id = %s, source = %s",
+ Form("GetFile - More than one entry in %s FXS db for: id = %s, source = %s",
GetSystemName(system), id, sourceName.Data()));
+ fFXSError = system;
delete aResult;
return 0;
}
Log(detector,
Form("GetFileName - Wrong field count in %s FXS db for: id = %s, source = %s",
GetSystemName(system), id, sourceName.Data()));
+ fFXSError = system;
delete aResult;
return 0;
}
TSQLRow* aRow = dynamic_cast<TSQLRow*> (aResult->Next());
if (!aRow){
- Log(detector, Form("GetFileName - Empty set result in %s FXS db from query: id = %s, source = %s",
+ Log(detector, Form("GetFile - Empty set result in %s FXS db from query: id = %s, source = %s",
GetSystemName(system), id, sourceName.Data()));
+ fFXSError = system;
delete aResult;
return 0;
}
filePath.Data(), fileSize.Data(), fileChecksum.Data()));
// retrieved file is renamed to make it unique
- TString localFileName = Form("%s/%s_%d_process/%s_%s_%d_%s_%s.shuttle",
- GetShuttleTempDir(), detector, GetCurrentRun(),
+ Int_t dir_lev1 = GetCurrentRun()/10000;
+ TString localFileName = Form("%s/%d/%d/%s_process/%s_%s_%d_%s_%s.shuttle",
+ GetShuttleTempDir(), dir_lev1, GetCurrentRun(), detector,
GetSystemName(system), detector, GetCurrentRun(),
id, sourceName.Data());
+ Log("SHUTTLE",Form("file from FXS = %s",localFileName.Data()));
// file retrieval from FXS
Bool_t result = kFALSE;
// copy!! if successful TSystem::Exec returns 0
- while(nRetries++ < maxRetries) {
+ while (nRetries++ < maxRetries) {
AliDebug(2, Form("Trying to copy file. Retry # %d", nRetries));
result = RetrieveFile(system, filePath.Data(), localFileName.Data());
- if(!result)
+ if (!result)
{
- Log(detector, Form("GetFileName - Copy of file %s from %s FXS failed",
+ Log(detector, Form("GetFile - Copy of file %s from %s FXS failed",
filePath.Data(), GetSystemName(system)));
continue;
}
+ if (fileSize.Length()>0)
+ {
+ // compare filesize of local file with the one stored in the FXS DB
+ Long_t size = -1;
+ Int_t sizeComp = gSystem->GetPathInfo(localFileName.Data(), 0, &size, 0, 0);
+
+ if (sizeComp != 0 || size != fileSize.Atoi())
+ {
+ Log(detector, Form("GetFile - size of file %s does not match with local copy!",
+ filePath.Data()));
+ result = kFALSE;
+ continue;
+ }
+
+ } else {
+ Log(fCurrentDetector, Form("GetFile - size of file %s not set in %s database, skipping comparison",
+ filePath.Data(), GetSystemName(system)));
+ }
+
if (fileChecksum.Length()>0)
{
// compare md5sum of local file with the one stored in the FXS DB
- Int_t md5Comp = gSystem->Exec(Form("md5sum %s |grep %s 2>&1 > /dev/null",
+ if(fileChecksum.Contains(' ')) fileChecksum.Resize(fileChecksum.First(' '));
+ Int_t md5Comp = gSystem->Exec(Form("md5sum %s |grep %s > /dev/null 2> /dev/null",
localFileName.Data(), fileChecksum.Data()));
if (md5Comp != 0)
{
- Log(detector, Form("GetFileName - md5sum of file %s does not match with local copy!",
+ Log(detector, Form("GetFile - md5sum of file %s does not match with local copy!",
filePath.Data()));
result = kFALSE;
continue;
if (result) break;
}
- if(!result) return 0;
+ if (!result)
+ {
+ fFXSError = system;
+ return 0;
+ }
fFXSCalled[system]=kTRUE;
TObjString *fileParams = new TObjString(Form("%s#!?!#%s", id, sourceName.Data()));
}
}
- TString baseFXSFolder;
- if (system == kDAQ)
- {
- baseFXSFolder = "FES/";
- }
- else if (system == kDCS)
- {
- baseFXSFolder = "";
- }
- else if (system == kHLT)
- {
- baseFXSFolder = "/opt/FXS/";
- }
-
-
- TString command = Form("scp -oPort=%d -2 %s@%s:%s%s %s",
+ TString command = Form("scp -oPort=%d -2 %s@%s:%s/%s %s",
fConfig->GetFXSPort(system),
fConfig->GetFXSUser(system),
fConfig->GetFXSHost(system),
- baseFXSFolder.Data(),
+ fConfig->GetFXSBaseFolder(system),
fxsFileName,
localFileName);
// if id is NULL all sources are returned (distinct)
//
- Log(detector, Form("GetFileSources - Retrieving sources with id %s from %s", id, GetSystemName(system)));
+ if (id)
+ {
+ Log(detector, Form("GetFileSources - Querying %s FXS for files with id %s produced by %s", GetSystemName(system), id, detector));
+ } else {
+ Log(detector, Form("GetFileSources - Querying %s FXS for files produced by %s", GetSystemName(system), detector));
+ }
// check if test mode should simulate a FXS error
if (fTestMode & kErrorFXSSources)
if (!Connect(system))
{
Log(detector, Form("GetFileSources - Couldn't connect to %s FXS database", GetSystemName(system)));
+ fFXSError = system;
return NULL;
}
- TString sourceName = 0;
+ TString sourceName = "";
if (system == kDAQ)
{
sourceName = "DAQsource";
if (!aResult) {
Log(detector, Form("GetFileSources - Can't execute SQL query to %s database for id: %s",
GetSystemName(system), id));
+ fFXSError = system;
return 0;
}
return NULL;
}
- TString sourceName = 0;
+ TString sourceName = "";
if (system == kDAQ)
{
sourceName = "DAQsource";
//
// check connection: if already connected return
- if(fServer[system] && fServer[system]->IsConnected()) return kTRUE;
+
+ if(fServer[system] && fServer[system]->IsConnected()) {
+ // ping the server
+ if (fServer[system]->PingVerify()==kTRUE){ // connection is still alive
+ return kTRUE;
+ }
+ else{
+ AliWarning(Form("Connection got lost to FXS database for %s. Closing and reconnecting.",
+ AliShuttleInterface::GetSystemName(system)));
+ fServer[system]->Close();
+ delete fServer[system];
+ fServer[system] = 0x0;
+ }
+ }
TString dbHost, dbUser, dbPass, dbName;
delete aFXSarray;
- TString sqlQuery = Form("update %s set time_processed=%d %s", fConfig->GetFXSdbTable(system),
- now.GetSec(), whereClause.Data());
+ TString sqlQuery = Form("update %s set time_processed=%ld %s", fConfig->GetFXSdbTable(system),
+ (ULong_t)now.GetSec(), whereClause.Data());
AliDebug(2, Form("SQL query: \n%s",sqlQuery.Data()));
return result;
}
+//_______________________________________________________________________________
+Bool_t AliShuttle::UpdateTableSkippedCase(const char* detector)
+{
+ //
+ // Update FXS table filling time_processed field in all rows corresponding to current run and detector
+ // if detector = "ALL" update all detectors
+ //
+
+ Bool_t result = kTRUE;
+
+ TString detName(detector);
+
+ for (UInt_t system=0; system<3; system++)
+ {
+
+ // check connection, in case connect
+ if (!Connect(system))
+ {
+ Log(fCurrentDetector, Form("UpdateTableSkippedCase - Couldn't connect to %s FXS database", GetSystemName(system)));
+ result = kFALSE;
+ continue;
+ }
+
+ TTimeStamp now; // now
+
+ // Loop on FXS list entries
+ TIter iter(&fFXSlist[system]);
+
+ TString whereClause;
+ if (detName == "ALL") whereClause = Form("where run=%d and time_processed IS NULL;",GetCurrentRun());
+ else whereClause = Form("where run=%d and detector=\"%s\" and time_processed IS NULL;",GetCurrentRun(), detector);
+
+ //Log("SHUTTLE",Form(" whereClause = %s ",whereClause.Data()));
+
+ TString sqlQuery = Form("update %s set time_processed=%ld %s", fConfig->GetFXSdbTable(system),
+ (ULong_t)now.GetSec(), whereClause.Data());
+
+ AliDebug(2, Form("SQL query: \n%s",sqlQuery.Data()));
+
+ // Query execution
+ TSQLResult* aResult;
+ aResult = dynamic_cast<TSQLResult*> (fServer[system]->Query(sqlQuery));
+ if (!aResult)
+ {
+ Log("SHUTTLE", Form("UpdateTableSkippedCase - %s db: can't execute SQL query <%s>",
+ GetSystemName(system), sqlQuery.Data()));
+ result = kFALSE;
+ continue;
+ }
+ delete aResult;
+
+ }
+
+ return result;
+}
//______________________________________________________________________________________________
Bool_t AliShuttle::UpdateTableFailCase()
{
GetCurrentRun(), fCurrentDetector.Data());
- TString sqlQuery = Form("update %s set time_processed=%d %s", fConfig->GetFXSdbTable(system),
- now.GetSec(), whereClause.Data());
+ TString sqlQuery = Form("update %s set time_processed=%ld %s", fConfig->GetFXSdbTable(system),
+ (ULong_t)now.GetSec(), whereClause.Data());
AliDebug(2, Form("SQL query: \n%s",sqlQuery.Data()));
TString detName(detector);
TString setClause;
- if (detName == "shuttle_done" || detName == "shuttle_ignored")
+ if (detName == "shuttle_done" || detName == "shuttle_skipped")
{
setClause = "set shuttle_done=1";
-
+
if (detName == "shuttle_done")
{
- // Send the information to ML
- TMonaLisaText mlStatus("SHUTTLE_status", "Done");
-
- TList mlList;
- mlList.Add(&mlStatus);
-
- TString mlID;
- mlID.Form("%d", GetCurrentRun());
- fMonaLisa->SendParameters(&mlList, mlID);
+ if (TouchFile() != kTRUE)
+ {
+ SendMLRunInfo("Pending");
+ return kFALSE;
+ }
+
+ SendMLRunInfo("Done");
}
- } else {
+ else
+ SendMLRunInfo("Skipped");
+ }
+ else {
TString statusStr(status);
if(statusStr.Contains("done", TString::kIgnoreCase) ||
statusStr.Contains("failed", TString::kIgnoreCase)){
return fLogbookEntry ? fLogbookEntry->GetEndTime() : 0;
}
-
//______________________________________________________________________________________________
UInt_t AliShuttle::GetCurrentYear() const
{
}
//______________________________________________________________________________________________
-void AliShuttle::Log(const char* detector, const char* message)
+void AliShuttle::Log(const char* detector, const char* message, UInt_t level)
{
//
// Fill log string with a message
//
-
- TString logRunDir = GetShuttleLogDir();
- if (GetCurrentRun() >=0)
- logRunDir += Form("/%d", GetCurrentRun());
+ TString logRunDir = GetShuttleLogDir();
+ if (GetCurrentRun() >=0) {
+ Int_t logDir_lev1 = GetCurrentRun()/10000;
+ logRunDir += Form("/%d/%d", logDir_lev1, GetCurrentRun());
+ }
void* dir = gSystem->OpenDirectory(logRunDir.Data());
if (dir == NULL) {
if (gSystem->mkdir(logRunDir.Data(), kTRUE)) {
gSystem->FreeDirectory(dir);
}
- TString toLog = Form("%s (%d): %s - ", TTimeStamp(time(0)).AsString("s"), getpid(), detector);
+ TString toLog = Form("%s UTC (%d): %s - ", TTimeStamp(time(0)).AsString("s"), getpid(), detector);
if (GetCurrentRun() >= 0)
toLog += Form("run %d - ", GetCurrentRun());
toLog += Form("%s", message);
- AliInfo(toLog.Data());
+ AliLog::Message(level, toLog, MODULENAME(), ClassName(), FUNCTIONNAME(), __FILE__, __LINE__);
// if we redirect the log output already to the file, leave here
if (fOutputRedirected && strcmp(detector, "SHUTTLE") != 0)
if (GetCurrentRun() >= 0)
{
- fileName.Form("%s/%d/%s_%d.log", GetShuttleLogDir(), GetCurrentRun(),
- detector, GetCurrentRun());
+ Int_t logDir_lev1 = GetCurrentRun()/10000;
+ fileName.Form("%s/%d/%d/%s.log", GetShuttleLogDir(), logDir_lev1, GetCurrentRun(),
+ detector);
} else {
fileName.Form("%s/%s.log", GetShuttleLogDir(), detector);
}
if (!fMonaLisa)
fMonaLisa = new TMonaLisaWriter(fConfig->GetMonitorHost(), fConfig->GetMonitorTable());
-
- SendAlive();
+ CountOpenRuns();
TString whereClause("where shuttle_done=0");
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));
+ TString whereClauseBis(Form("where shuttle_done=0 and run < %d", run));
TObjArray tmpLogbookEntries;
- if (!QueryShuttleLogbook(whereClause, tmpLogbookEntries))
+ if (!QueryShuttleLogbook(whereClauseBis, tmpLogbookEntries))
{
Log("SHUTTLE", "Collect - Can't retrieve entries from Shuttle logbook");
return kFALSE;
if (!RetrieveConditionsData(shuttleLogbookEntries))
{
Log("SHUTTLE", "Collect - Process of at least one run failed");
+ CountOpenRuns();
return kFALSE;
}
Log("SHUTTLE", "Collect - Requested run(s) successfully processed");
+ CountOpenRuns();
return kTRUE;
}
}
//______________________________________________________________________________________________
-Bool_t AliShuttle::SendMail()
+Bool_t AliShuttle::SendMail(EMailTarget target, Int_t system)
{
//
// sends a mail to the subdetector expert in case of preprocessor error
if (fTestMode != kNone)
return kTRUE;
+
+ if (!fConfig->SendMail())
+ return kTRUE;
+ if (target == kDCSEMail || target == kFXSEMail) {
+ if (!fFirstProcessing)
+ return kTRUE;
+ }
+
+ Int_t runMode = (Int_t)fConfig->GetRunMode();
+ TString tmpStr;
+ if (runMode == 0) tmpStr = " Nightly Test:";
+ else tmpStr = " Data Taking:";
void* dir = gSystem->OpenDirectory(GetShuttleLogDir());
if (dir == NULL)
{
gSystem->FreeDirectory(dir);
}
- TString bodyFileName;
- bodyFileName.Form("%s/mail.body", GetShuttleLogDir());
- gSystem->ExpandPathName(bodyFileName);
-
- ofstream mailBody;
- mailBody.open(bodyFileName, ofstream::out);
-
- if (!mailBody.is_open())
- {
- Log("SHUTTLE", Form("Could not open mail body file %s", bodyFileName.Data()));
- return kFALSE;
- }
-
+ // det experts in to
TString to="";
- TIter iterExperts(fConfig->GetResponsibles(fCurrentDetector));
+ TIter *iterExperts = 0;
+ if (target == kDCSEMail) {
+ iterExperts = new TIter(fConfig->GetAdmins(AliShuttleConfig::kAmanda));
+ }
+ else if (target == kFXSEMail) {
+ iterExperts = new TIter(fConfig->GetAdmins(system));
+ }
+ if (iterExperts) {
+ TObjString *anExpert=0;
+ while ((anExpert = (TObjString*) iterExperts->Next()))
+ {
+ to += Form("%s,", anExpert->GetName());
+ }
+ delete iterExperts;
+ }
+
+ // add subdetector experts
+ iterExperts = new TIter(fConfig->GetResponsibles(fCurrentDetector));
TObjString *anExpert=0;
- while ((anExpert = (TObjString*) iterExperts.Next()))
+ while ((anExpert = (TObjString*) iterExperts->Next()))
{
to += Form("%s,", anExpert->GetName());
}
- to.Remove(to.Length()-1);
+ delete iterExperts;
+
+ if (to.Length() > 0)
+ to.Remove(to.Length()-1);
AliDebug(2, Form("to: %s",to.Data()));
if (to.IsNull()) {
- Log("SHUTTLE", "List of detector responsibles not yet set!");
+ Log("SHUTTLE", Form("List of %d responsibles not set!", (Int_t) target));
return kFALSE;
}
- TString cc="alberto.colla@cern.ch";
-
- TString subject = Form("%s Shuttle preprocessor FAILED in run %d (run type = %s)!",
- fCurrentDetector.Data(), GetCurrentRun(), GetRunType());
- 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 "
- "failed processing run %d (run type = %s)!!\n\n",
- GetCurrentRun(), GetRunType());
- body += Form("Please check %s status on the SHUTTLE monitoring page: \n\n",
- fCurrentDetector.Data());
- if (fConfig->GetRunMode() == AliShuttleConfig::kTest)
+ // SHUTTLE responsibles in cc
+ TString cc="";
+ TIter iterAdmins(fConfig->GetAdmins(AliShuttleConfig::kGlobal));
+ TObjString *anAdmin=0;
+ while ((anAdmin = (TObjString*) iterAdmins.Next()))
{
- body += Form("\thttp://pcalimonitor.cern.ch:8889/shuttle.jsp?time=168 \n\n");
- } else {
- body += Form("\thttp://pcalimonitor.cern.ch/shuttle.jsp?instance=PROD?time=168 \n\n");
- }
-
-
- TString logFolder = "logs";
- if (fConfig->GetRunMode() == AliShuttleConfig::kProd)
- logFolder += "_PROD";
-
-
- body += Form("Find the %s log for the current run on \n\n"
- "\thttp://pcalishuttle01.cern.ch:8880/%s/%d/%s_%d.log \n\n",
- fCurrentDetector.Data(), logFolder.Data(), GetCurrentRun(),
- fCurrentDetector.Data(), GetCurrentRun());
- body += Form("The last 10 lines of %s log file are following:\n\n", fCurrentDetector.Data());
-
- AliDebug(2, Form("Body begin: %s", body.Data()));
-
- mailBody << body.Data();
- mailBody.close();
- mailBody.open(bodyFileName, ofstream::out | ofstream::app);
-
- TString logFileName = Form("%s/%d/%s_%d.log", GetShuttleLogDir(),
- GetCurrentRun(), fCurrentDetector.Data(), GetCurrentRun());
- TString tailCommand = Form("tail -n 10 %s >> %s", logFileName.Data(), bodyFileName.Data());
- if (gSystem->Exec(tailCommand.Data()))
- {
- mailBody << Form("%s log file not found ...\n\n", fCurrentDetector.Data());
- }
-
- TString endBody = Form("------------------------------------------------------\n\n");
- endBody += Form("In case of problems please contact the SHUTTLE core team.\n\n");
- endBody += "Please do not answer this message directly, it is automatically generated.\n\n";
- endBody += "Greetings,\n\n \t\t\tthe SHUTTLE\n";
-
- 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;
-}
-
-//______________________________________________________________________________________________
-Bool_t AliShuttle::SendMailToDCS()
-{
- //
- // sends a mail to the DCS experts in case of DCS error
- //
-
- if (fTestMode != kNone)
- return kTRUE;
-
- void* dir = gSystem->OpenDirectory(GetShuttleLogDir());
- if (dir == NULL)
- {
- if (gSystem->mkdir(GetShuttleLogDir(), kTRUE))
- {
- Log("SHUTTLE", Form("SendMailToDCS - Can't open directory <%s>", GetShuttleLogDir()));
- return kFALSE;
- }
-
- } else {
- gSystem->FreeDirectory(dir);
+ cc += Form("%s,", anAdmin->GetName());
}
+ if (cc.Length() > 0)
+ cc.Remove(cc.Length()-1);
+ AliDebug(2, Form("cc: %s",to.Data()));
+ // mail body
TString bodyFileName;
bodyFileName.Form("%s/mail.body", GetShuttleLogDir());
gSystem->ExpandPathName(bodyFileName);
if (!mailBody.is_open())
{
- Log("SHUTTLE", Form("SendMailToDCS - Could not open mail body file %s", bodyFileName.Data()));
+ Log("SHUTTLE", Form("Could not open mail body file %s", bodyFileName.Data()));
return kFALSE;
}
- TString to="Vladimir.Fekete@cern.ch, Svetozar.Kapusta@cern.ch";
- //TString to="alberto.colla@cern.ch";
- AliDebug(2, Form("to: %s",to.Data()));
- if (to.IsNull()) {
- Log("SHUTTLE", "List of detector responsibles not yet set!");
- return kFALSE;
- }
+ TString subject;
+ TString body;
- TString cc="alberto.colla@cern.ch";
-
- TString subject = Form("Retrieval of data points for %s FAILED in run %d !",
- fCurrentDetector.Data(), GetCurrentRun());
- AliDebug(2, Form("subject: %s", subject.Data()));
+ if (target == kDCSEMail){
+ subject = Form("%s CRITICAL Retrieval of data points for %s FAILED in run %d !",
+ tmpStr.Data(), fCurrentDetector.Data(), GetCurrentRun());
+ AliDebug(2, Form("subject: %s", subject.Data()));
+
+ body = Form("Dear DCS experts, \n\n");
+ body += Form("SHUTTLE couldn\'t retrieve the data points for detector %s "
+ "in run %d!!\n\n", fCurrentDetector.Data(), GetCurrentRun());
+ }
+ else if (target == kFXSEMail){
+ subject = Form("%s CRITICAL FXS communication for %s FAILED in run %d !",
+ tmpStr.Data(), fCurrentDetector.Data(), GetCurrentRun());
+ AliDebug(2, Form("subject: %s", subject.Data()));
+ TString sys;
+ if (system == kDAQ) sys="DAQ";
+ else if (system == kDCS) sys="DCS";
+ else if (system == kHLT) sys="HLT";
+ else return kFALSE;
+ body = Form("Dear %s FXS experts, \n\n",sys.Data());
+ body += Form("SHUTTLE couldn\'t retrieve data from the FXS for detector %s "
+ "in run %d!!\n\n", fCurrentDetector.Data(), GetCurrentRun());
+ body += Form("The contacted server was:\nDB: %s\nFXS:%s\n\n", fConfig->GetFXSdbHost(system), fConfig->GetFXSHost(system));
+ }
+ else {
+ subject = Form("%s %s Shuttle preprocessor FAILED in run %d (run type = %s)!",
+ tmpStr.Data(), fCurrentDetector.Data(), GetCurrentRun(), GetRunType());
+ AliDebug(2, Form("subject: %s", subject.Data()));
+
+ body = Form("Dear %s expert(s), \n\n", fCurrentDetector.Data());
+ body += Form("SHUTTLE just detected that your preprocessor "
+ "failed processing run %d (run type = %s)!!\n\n",
+ GetCurrentRun(), GetRunType());
+ }
- TString body = Form("Dear DCS experts, \n\n");
- body += Form("SHUTTLE couldn\'t retrieve the data points for detector %s "
- "in run %d!!\n\n", fCurrentDetector.Data(), GetCurrentRun());
body += Form("Please check %s status on the SHUTTLE monitoring page: \n\n",
fCurrentDetector.Data());
if (fConfig->GetRunMode() == AliShuttleConfig::kTest)
{
- body += Form("\thttp://pcalimonitor.cern.ch:8889/shuttle.jsp?time=168 \n\n");
+ body += Form("\thttp://pcalimonitor.cern.ch/shuttle.jsp?time=24 \n\n");
} else {
- body += Form("\thttp://pcalimonitor.cern.ch/shuttle.jsp?instance=PROD?time=168 \n\n");
+ body += Form("\thttp://pcalimonitor.cern.ch/shuttle.jsp?instance=PROD&time=24 \n\n");
}
-
+
+
TString logFolder = "logs";
if (fConfig->GetRunMode() == AliShuttleConfig::kProd)
logFolder += "_PROD";
body += Form("Find the %s log for the current run on \n\n"
- "\thttp://pcalishuttle01.cern.ch:8880/%s/%d/%s_%d.log \n\n",
- fCurrentDetector.Data(), logFolder.Data(), GetCurrentRun(),
- fCurrentDetector.Data(), GetCurrentRun());
- body += Form("The last 10 lines of %s log file are following:\n\n", fCurrentDetector.Data());
+ "\thttp://pcalishuttle02.cern.ch/%s/%d/%d/%s.log \n\n",
+ fCurrentDetector.Data(), logFolder.Data(), GetCurrentRun()/10000,
+ GetCurrentRun(), fCurrentDetector.Data());
+ body += Form("The last 15 lines of %s log file are following:\n\n", fCurrentDetector.Data());
AliDebug(2, Form("Body begin: %s", body.Data()));
mailBody.close();
mailBody.open(bodyFileName, ofstream::out | ofstream::app);
- TString logFileName = Form("%s/%d/%s_%d.log", GetShuttleLogDir(), GetCurrentRun(),
- fCurrentDetector.Data(), GetCurrentRun());
- TString tailCommand = Form("tail -n 10 %s >> %s", logFileName.Data(), bodyFileName.Data());
+ TString logFileName = Form("%s/%d/%d/%s.log", GetShuttleLogDir(),
+ GetCurrentRun()/10000, GetCurrentRun(), fCurrentDetector.Data());
+ TString tailCommand = Form("tail -n 15 %s >> %s", logFileName.Data(), bodyFileName.Data());
if (gSystem->Exec(tailCommand.Data()))
{
mailBody << Form("%s log file not found ...\n\n", fCurrentDetector.Data());
return result == 0;
}
-
//______________________________________________________________________________________________
const char* AliShuttle::GetRunType()
{
Bool_t AliShuttle::GetHLTStatus()
{
// Return HLT status (ON=1 OFF=0)
- // Converts the HLT status from the status string read in the run logbook (not just a bool)
+ // Converts the HLT status from the mode string read in the run logbook (not just a bool)
if(!fLogbookEntry) {
AliError("No logbook entry!");
return 0;
}
- // TODO implement when HLTStatus is inserted in run logbook
- //TString hltStatus = fLogbookEntry->GetRunParameter("HLTStatus");
- //if(hltStatus == "OFF") {return kFALSE};
+ // TODO implement when HLTMode is inserted in run logbook
+ TString hltMode = fLogbookEntry->GetRunParameter("HLTmode");
+ TSubString firstChar = hltMode(0,1);
+ AliDebug(2,Form("First char = %s ",firstChar.Data()));
+ if (firstChar == "A") {
+ return kFALSE;
+ }
+ else if ((firstChar == "B") || (firstChar == "C") || (firstChar == "D") || (firstChar == "E")) {
+ return kTRUE;
+ }
+ else {
+ Log("SHUTTLE","Unexpected HLT mode! Returning 0....");
+ return kFALSE;
+ }
+}
+
+//______________________________________________________________________________________________
+const char* AliShuttle::GetTriggerConfiguration()
+{
+ // Receives the trigger configuration from the DAQ logbook for the current run
+
+ // check connection, if needed reconnect
+ if (!Connect(3))
+ return 0;
- return kTRUE;
+ TString sqlQuery;
+ sqlQuery.Form("SELECT configFile FROM logbook_trigger_config WHERE run = %d", GetCurrentRun());
+ TSQLResult* result = fServer[3]->Query(sqlQuery);
+ if (!result)
+ {
+ Log("SHUTTLE", Form("ERROR: Can't execute query <%s>!", sqlQuery.Data()));
+ return 0;
+ }
+
+ if (result->GetRowCount() == 0)
+ {
+ Log("SHUTTLE", "WARNING: Trigger configuration not found in logbook_trigger_config");
+ delete result;
+ return 0;
+ }
+
+ TSQLRow* row = result->Next();
+ if (!row)
+ {
+ Log("SHUTTLE", "ERROR: Could not receive logbook_trigger_config data");
+ delete result;
+ return 0;
+ }
+
+ // static, so that pointer remains valid when it is returned to the calling class
+ static TString triggerConfig(row->GetField(0));
+
+ delete row;
+ row = 0;
+
+ delete result;
+ result = 0;
+
+ Log("SHUTTLE", Form("Found trigger configuration: %s", triggerConfig.Data()));
+
+ return triggerConfig;
+}
+
+//______________________________________________________________________________________________
+const char* AliShuttle::GetCTPTimeParams()
+{
+ // Receives the CTP time parameters from the DAQ logbook for the current run
+
+ // check connection, if needed reconnect
+ if (!Connect(3))
+ return 0;
+
+ TString sqlQuery;
+ sqlQuery.Form("SELECT alignmentFile FROM logbook_trigger_config WHERE run = %d", GetCurrentRun());
+ TSQLResult* result = fServer[3]->Query(sqlQuery);
+ if (!result)
+ {
+ Log("SHUTTLE", Form("ERROR: Can't execute query <%s>!", sqlQuery.Data()));
+ return 0;
+ }
+
+ if (result->GetRowCount() == 0)
+ {
+ Log("SHUTTLE", "WARNING: CTP time params not found in logbook_trigger_config");
+ delete result;
+ return 0;
+ }
+
+ TSQLRow* row = result->Next();
+ if (!row)
+ {
+ Log("SHUTTLE", "ERROR: Could not receive logbook_trigger_config data");
+ delete result;
+ return 0;
+ }
+
+ // static, so that pointer remains valid when it is returned to the calling class
+ static TString triggerTimeParams(row->GetField(0));
+
+ delete row;
+ row = 0;
+
+ delete result;
+ result = 0;
+
+ Log("SHUTTLE", Form("Found trigger time parameters: %s", triggerTimeParams.Data()));
+
+ return triggerTimeParams;
+}
+
+//______________________________________________________________________________________________
+const char* AliShuttle::GetTriggerDetectorMask()
+{
+ // Receives the trigger detector mask from DAQ logbook
+
+ // check connection, if needed reconnect
+ if (!Connect(3))
+ return 0;
+
+ TString sqlQuery;
+ sqlQuery.Form("SELECT BIN(BIT_OR(inputDetectorMask)) from logbook_trigger_clusters WHERE run = %d;", GetCurrentRun());
+ TSQLResult* result = fServer[3]->Query(sqlQuery);
+ if (!result)
+ {
+ Log("SHUTTLE", Form("ERROR: Can't execute query <%s>!", sqlQuery.Data()));
+ return 0;
+ }
+
+ if (result->GetRowCount() == 0)
+ {
+ Log("SHUTTLE", "ERROR: Trigger Detector Mask not found in logbook_trigger_clusters");
+ delete result;
+ return 0;
+ }
+
+ TSQLRow* row = result->Next();
+ if (!row)
+ {
+ Log("SHUTTLE", "ERROR: Could not receive logbook_trigger_clusters data");
+ delete result;
+ return 0;
+ }
+
+ // static, so that pointer remains valid when it is returned to the calling class
+ static TString triggerDetectorMask(row->GetField(0));
+
+ delete row;
+ row = 0;
+
+ delete result;
+ result = 0;
+
+ Log("SHUTTLE", Form("Found Trigger Detector Mask: %s", triggerDetectorMask.Data()));
+
+ return triggerDetectorMask;
}
//______________________________________________________________________________________________
fgkShuttleLogDir = gSystem->ExpandPathName(logDir);
}
+//______________________________________________________________________________________________
+Bool_t AliShuttle::TouchFile()
+{
+ //
+ // touching a file on the grid if run has been DONE
+ //
+
+ if (!gGrid)
+ {
+ Log("SHUTTLE",Form("No TGrid connection estabilished!"));
+ Log("SHUTTLE",Form("Could not touch file for run %i",GetCurrentRun()));
+ return kFALSE;
+ }
+
+ TString dir;
+ dir.Form("%s%d/SHUTTLE_DONE", fConfig->GetAlienPath(), GetCurrentYear());
+ // checking whether directory for touch command exists
+ TString commandLs;
+ commandLs.Form("ls %s",dir.Data());
+ TGridResult *resultLs = dynamic_cast<TGridResult*>(gGrid->Command(commandLs));
+ if (!resultLs){
+ Log("SHUTTLE",Form("No result for %s command, returning without touching",commandLs.Data()));
+ return kFALSE;
+ }
+ TMap *mapLs = dynamic_cast<TMap*>(resultLs->At(0));
+ if (!mapLs){
+ Log("SHUTTLE",Form("No map for %s command, returning without touching",commandLs.Data()));
+ delete resultLs;
+ resultLs = 0x0;
+ return kFALSE;
+ }
+ TObjString *valueLsPath = dynamic_cast<TObjString*>(mapLs->GetValue("path"));
+ if (!valueLsPath || (valueLsPath->GetString()).CompareTo(dir)!=1){
+ Log("SHUTTLE",Form("No directory %s found, creating it",dir.Data()));
+
+ // creating the directory
+
+ Bool_t boolMkdir = gGrid->Mkdir(dir.Data());
+ if (!boolMkdir) {
+ Log("SHUTTLE",Form("Impossible to create dir %s in alien catalogue for run %i!",dir.Data(),GetCurrentRun()));
+ delete resultLs;
+ resultLs = 0x0;
+ return kFALSE;
+ }
+ Log("SHUTTLE",Form("Directory %s successfully created in alien catalogue for run %i",dir.Data(),GetCurrentRun()));
+ }
+ else {
+ Log("SHUTTLE",Form("Directory %s correctly found for run %i",dir.Data(),GetCurrentRun()));
+ }
+
+ delete resultLs;
+ resultLs = 0x0;
+
+ TString command;
+ command.Form("touch %s/%i", dir.Data(), GetCurrentRun());
+ Log("SHUTTLE", Form("Creating entry in file catalog: %s", command.Data()));
+ TGridResult *resultTouch = dynamic_cast<TGridResult*>(gGrid->Command(command));
+ if (!resultTouch){
+ Log("SHUTTLE",Form("No result for touching command, returning without touching for run %i",GetCurrentRun()));
+ return kFALSE;
+ }
+ TMap *mapTouch = dynamic_cast<TMap*>(resultTouch->At(0));
+ if (!mapTouch){
+ Log("SHUTTLE",Form("No map for touching command, returning without touching for run %i",GetCurrentRun()));
+ delete resultTouch;
+ resultTouch = 0x0;
+ return kFALSE;
+ }
+ TObjString *valueTouch = dynamic_cast<TObjString*>(mapTouch->GetValue("__result__"));
+ if (!valueTouch){
+ Log("SHUTTLE",Form("No value for \"__result__\" key set in the map for touching command, returning without touching for run %i",GetCurrentRun()));
+ delete resultTouch;
+ resultTouch = 0x0;
+ return kFALSE;
+ }
+ if (valueTouch->GetString()!="1"){
+ Log("SHUTTLE",Form("Failing the touching command, returning without touching for run %i",GetCurrentRun()));
+ delete resultTouch;
+ resultTouch = 0x0;
+ return kFALSE;
+ }
+ delete resultTouch;
+ resultTouch = 0x0;
+ Log("SHUTTLE", "Sucessfully touched the file");
+ return kTRUE;
+}
+//______________________________________________________________________________________________
+UInt_t AliShuttle::GetStartTimeDCSQuery()
+{
+ // Return Start Time for the DCS query
+ //
+ // The call is delegated to AliShuttleInterface
+
+ return GetCurrentStartTime()-fConfig->GetDCSQueryOffset();
+}
+//______________________________________________________________________________________________
+UInt_t AliShuttle::GetEndTimeDCSQuery()
+{
+ // Return End Time for the DCS query
+ //
+ // The call is delegated to AliShuttleInterface
+
+ return GetCurrentEndTime()+fConfig->GetDCSQueryOffset();
+}
+//______________________________________________________________________________________________
+void AliShuttle::SendMLFromDet(const char* value)
+{
+ //
+ // Sending an information coming from the current detector to ML
+ //
+
+ TMonaLisaText mlText(Form("%s_RunCondition", fCurrentDetector.Data()), value);
+
+ TList mlList;
+ mlList.Add(&mlText);
+
+ TString mlID;
+ mlID.Form("%d", GetCurrentRun());
+ fMonaLisa->SendParameters(&mlList, mlID);
+
+ return;
+}
+//______________________________________________________________________________________________
+TString* AliShuttle::GetLTUConfig(const char* det)
+{
+ //
+ // Getting ltuFineDelay1, ltuFineDelay2, ltuBCDelay for detector det from logbook_detectors table in logbook
+ //
+
+ if (!Connect(3))
+ return 0;
+
+ TString sqlQuery;
+ sqlQuery.Form("select LTUFineDelay1, LTUFineDelay2, LTUBCDelayAdd from logbook_detectors WHERE run_number = %d and detector = \"%s\";", GetCurrentRun(),det);
+
+ TSQLResult* result = fServer[3]->Query(sqlQuery);
+ if (!result){
+ Log("SHUTTLE","ERROR: No result found for the LTU configuration query");
+ return 0x0;
+ }
+ if (result->GetRowCount() == 0){
+ Log("SHUTTLE",Form("ERROR: LTU configuration not found in logbook_detectors for detector %s, returning null pointer",det));
+ delete result;
+ return 0x0;
+ }
+ if (result->GetFieldCount() != 3){
+ Log("SHUTTLE",Form("ERROR: not all the required fields are there for the LTU configuration for detector %s (only %d found), returning a null pointer",det, result->GetFieldCount()));
+ delete result;
+ return 0x0;
+ }
+ TSQLRow* row = result->Next();
+ if (!row){
+ Printf("ERROR: Could not receive logbook_detectors data, returning null pointer");
+ delete result;
+ return 0x0;
+ }
+ TString* ltuConfigString = new TString[3];
+
+ ltuConfigString[0] = row->GetField(0);
+ ltuConfigString[1] = row->GetField(1);
+ ltuConfigString[2] = row->GetField(2);
+
+ return ltuConfigString;
+
+}
git://git.uio.no / u / mrichter / AliRoot.git / blobdiff