Changes to take into account new entry in logbook_shuttle table in DAQ db.
[u/mrichter/AliRoot.git] / SHUTTLE / AliShuttle.cxx
CommitLineData
73abe331 1/**************************************************************************
2 * Copyright(c) 1998-1999, ALICE Experiment at CERN, All rights reserved. *
3 * *
4 * Author: The ALICE Off-line Project. *
5 * Contributors are mentioned in the code where appropriate. *
6 * *
7 * Permission to use, copy, modify and distribute this software and its *
8 * documentation strictly for non-commercial purposes is hereby granted *
9 * without fee, provided that the above copyright notice appears in all *
10 * copies and that both the copyright notice and this permission notice *
11 * appear in the supporting documentation. The authors make no claims *
12 * about the suitability of this software for any purpose. It is *
13 * provided "as is" without express or implied warranty. *
14 **************************************************************************/
15
ff3781ad 16/* $Id$ */
73abe331 17
18//
19// This class is the main manager for AliShuttle.
20// It organizes the data retrieval from DCS and call the
b948db8d 21// interface methods of AliPreprocessor.
73abe331 22// For every detector in AliShuttleConfgi (see AliShuttleConfig),
23// data for its set of aliases is retrieved. If there is registered
b948db8d 24// AliPreprocessor for this detector then it will be used
25// accroding to the schema (see AliPreprocessor).
26// If there isn't registered AliPreprocessor than the retrieved
73abe331 27// data is stored automatically to the undelying AliCDBStorage.
28// For detSpec is used the alias name.
29//
30
31#include "AliShuttle.h"
32
33#include "AliCDBManager.h"
34#include "AliCDBStorage.h"
35#include "AliCDBId.h"
84090f85 36#include "AliCDBRunRange.h"
37#include "AliCDBPath.h"
5164a766 38#include "AliCDBEntry.h"
73abe331 39#include "AliShuttleConfig.h"
eba76848 40#include "DCSClient/AliDCSClient.h"
73abe331 41#include "AliLog.h"
b948db8d 42#include "AliPreprocessor.h"
5164a766 43#include "AliShuttleStatus.h"
2bb7b766 44#include "AliShuttleLogbookEntry.h"
73abe331 45
57f50b3c 46#include <TSystem.h>
58bc3020 47#include <TObject.h>
b948db8d 48#include <TString.h>
57f50b3c 49#include <TTimeStamp.h>
73abe331 50#include <TObjString.h>
57f50b3c 51#include <TSQLServer.h>
52#include <TSQLResult.h>
53#include <TSQLRow.h>
cb343cfd 54#include <TMutex.h>
9827400b 55#include <TSystemDirectory.h>
56#include <TSystemFile.h>
a986b218 57#include <TFile.h>
9827400b 58#include <TGrid.h>
59#include <TGridResult.h>
503ff24f 60#include <TMap.h>
73abe331 61
e7f62f16 62#include <TMonaLisaWriter.h>
63
5164a766 64#include <fstream>
65
cb343cfd 66#include <sys/types.h>
67#include <sys/wait.h>
68
1bd183ba 69#include <signal.h>
70
73abe331 71ClassImp(AliShuttle)
72
b948db8d 73//______________________________________________________________________________________________
74AliShuttle::AliShuttle(const AliShuttleConfig* config,
75 UInt_t timeout, Int_t retries):
4f0ab988 76fConfig(config),
77fTimeout(timeout), fRetries(retries),
78fPreprocessorMap(),
2bb7b766 79fLogbookEntry(0),
eba76848 80fCurrentDetector(),
926cbc0e 81fFirstProcessing(0),
339adafa 82fFXSError(-1),
85a80aa9 83fStatusEntry(0),
cb343cfd 84fMonitoringMutex(0),
eba76848 85fLastActionTime(0),
e7f62f16 86fLastAction(),
9827400b 87fMonaLisa(0),
88fTestMode(kNone),
ffa29e93 89fReadTestMode(kFALSE),
90fOutputRedirected(kFALSE)
73abe331 91{
92 //
93 // config: AliShuttleConfig used
73abe331 94 // timeout: timeout used for AliDCSClient connection
95 // retries: the number of retries in case of connection error.
96 //
97
57f50b3c 98 if (!fConfig->IsValid()) AliFatal("********** !!!!! Invalid configuration !!!!! **********");
be48e3ea 99 for(int iSys=0;iSys<4;iSys++) {
57f50b3c 100 fServer[iSys]=0;
be48e3ea 101 if (iSys < 3)
2c15234c 102 fFXSlist[iSys].SetOwner(kTRUE);
57f50b3c 103 }
2bb7b766 104 fPreprocessorMap.SetOwner(kTRUE);
be48e3ea 105
106 for (UInt_t iDet=0; iDet<NDetectors(); iDet++)
107 fFirstUnprocessed[iDet] = kFALSE;
108
cb343cfd 109 fMonitoringMutex = new TMutex();
58bc3020 110}
111
b948db8d 112//______________________________________________________________________________________________
57f50b3c 113AliShuttle::~AliShuttle()
58bc3020 114{
9827400b 115 //
116 // destructor
117 //
58bc3020 118
b948db8d 119 fPreprocessorMap.DeleteAll();
be48e3ea 120 for(int iSys=0;iSys<4;iSys++)
57f50b3c 121 if(fServer[iSys]) {
122 fServer[iSys]->Close();
123 delete fServer[iSys];
eba76848 124 fServer[iSys] = 0;
57f50b3c 125 }
2bb7b766 126
127 if (fStatusEntry){
128 delete fStatusEntry;
129 fStatusEntry = 0;
130 }
cb343cfd 131
132 if (fMonitoringMutex)
133 {
134 delete fMonitoringMutex;
135 fMonitoringMutex = 0;
136 }
73abe331 137}
138
b948db8d 139//______________________________________________________________________________________________
57f50b3c 140void AliShuttle::RegisterPreprocessor(AliPreprocessor* preprocessor)
58bc3020 141{
73abe331 142 //
b948db8d 143 // Registers new AliPreprocessor.
73abe331 144 // It uses GetName() for indentificator of the pre processor.
145 // The pre processor is registered it there isn't any other
146 // with the same identificator (GetName()).
147 //
148
eba76848 149 const char* detName = preprocessor->GetName();
150 if(GetDetPos(detName) < 0)
151 AliFatal(Form("********** !!!!! Invalid detector name: %s !!!!! **********", detName));
152
153 if (fPreprocessorMap.GetValue(detName)) {
154 AliWarning(Form("AliPreprocessor %s is already registered!", detName));
73abe331 155 return;
156 }
157
eba76848 158 fPreprocessorMap.Add(new TObjString(detName), preprocessor);
73abe331 159}
b948db8d 160//______________________________________________________________________________________________
3301427a 161Bool_t AliShuttle::Store(const AliCDBPath& path, TObject* object,
84090f85 162 AliCDBMetaData* metaData, Int_t validityStart, Bool_t validityInfinite)
73abe331 163{
9827400b 164 // Stores a CDB object in the storage for offline reconstruction. Objects that are not needed for
165 // offline reconstruction, but should be stored anyway (e.g. for debugging) should NOT be stored
166 // using this function. Use StoreReferenceData instead!
167 // It calls StoreLocally function which temporarily stores the data locally; when the preprocessor
168 // finishes the data are transferred to the main storage (Grid).
b948db8d 169
3301427a 170 return StoreLocally(fgkLocalCDB, path, object, metaData, validityStart, validityInfinite);
84090f85 171}
172
173//______________________________________________________________________________________________
3301427a 174Bool_t AliShuttle::StoreReferenceData(const AliCDBPath& path, TObject* object, AliCDBMetaData* metaData)
84090f85 175{
9827400b 176 // Stores a CDB object in the storage for reference data. This objects will not be available during
177 // offline reconstrunction. Use this function for reference data only!
178 // It calls StoreLocally function which temporarily stores the data locally; when the preprocessor
179 // finishes the data are transferred to the main storage (Grid).
85a80aa9 180
3301427a 181 return StoreLocally(fgkLocalRefStorage, path, object, metaData);
85a80aa9 182}
183
184//______________________________________________________________________________________________
3301427a 185Bool_t AliShuttle::StoreLocally(const TString& localUri,
85a80aa9 186 const AliCDBPath& path, TObject* object, AliCDBMetaData* metaData,
187 Int_t validityStart, Bool_t validityInfinite)
188{
9827400b 189 // Store object temporarily in local storage. Parameters are passed by Store and StoreReferenceData functions.
190 // when the preprocessor finishes the data are transferred to the main storage (Grid).
191 // The parameters are:
192 // 1) Uri of the backup storage (Local)
193 // 2) the object's path.
194 // 3) the object to be stored
195 // 4) the metaData to be associated with the object
196 // 5) the validity start run number w.r.t. the current run,
197 // if the data is valid only for this run leave the default 0
198 // 6) specifies if the calibration data is valid for infinity (this means until updated),
199 // typical for calibration runs, the default is kFALSE
200 //
201 // returns 0 if fail, 1 otherwise
84090f85 202
9827400b 203 if (fTestMode & kErrorStorage)
204 {
205 Log(fCurrentDetector, "StoreLocally - In TESTMODE - Simulating error while storing locally");
206 return kFALSE;
207 }
208
3301427a 209 const char* cdbType = (localUri == fgkLocalCDB) ? "CDB" : "Reference";
2bb7b766 210
85a80aa9 211 Int_t firstRun = GetCurrentRun() - validityStart;
84090f85 212 if(firstRun < 0) {
9827400b 213 AliWarning("First valid run happens to be less than 0! Setting it to 0.");
84090f85 214 firstRun=0;
215 }
216
217 Int_t lastRun = -1;
218 if(validityInfinite) {
219 lastRun = AliCDBRunRange::Infinity();
220 } else {
221 lastRun = GetCurrentRun();
222 }
223
3301427a 224 // Version is set to current run, it will be used later to transfer data to Grid
225 AliCDBId id(path, firstRun, lastRun, GetCurrentRun(), -1);
2bb7b766 226
227 if(! dynamic_cast<TObjString*> (metaData->GetProperty("RunUsed(TObjString)"))){
228 TObjString runUsed = Form("%d", GetCurrentRun());
9e080f92 229 metaData->SetProperty("RunUsed(TObjString)", runUsed.Clone());
2bb7b766 230 }
84090f85 231
3301427a 232 Bool_t result = kFALSE;
84090f85 233
3301427a 234 if (!(AliCDBManager::Instance()->GetStorage(localUri))) {
235 Log("SHUTTLE", Form("StoreLocally - Cannot activate local %s storage", cdbType));
84090f85 236 } else {
3301427a 237 result = AliCDBManager::Instance()->GetStorage(localUri)
84090f85 238 ->Put(object, id, metaData);
239 }
240
241 if(!result) {
242
9827400b 243 Log(fCurrentDetector, Form("StoreLocally - Can't store object <%s>!", id.ToString().Data()));
3301427a 244 }
2bb7b766 245
3301427a 246 return result;
247}
84090f85 248
3301427a 249//______________________________________________________________________________________________
250Bool_t AliShuttle::StoreOCDB()
251{
9827400b 252 //
253 // Called when preprocessor ends successfully or when previous storage attempt failed (kStoreError status)
254 // Calls underlying StoreOCDB(const char*) function twice, for OCDB and Reference storage.
255 // Then calls StoreRefFilesToGrid to store reference files.
256 //
257
5e993b6f 258 UpdateShuttleStatus(AliShuttleStatus::kStoreStarted);
56d3e9e4 259
9827400b 260 if (fTestMode & kErrorGrid)
261 {
262 Log("SHUTTLE", "StoreOCDB - In TESTMODE - Simulating error while storing in the Grid");
263 Log(fCurrentDetector, "StoreOCDB - In TESTMODE - Simulating error while storing in the Grid");
264 return kFALSE;
265 }
266
c88ad5db 267 Log("SHUTTLE","StoreOCDB - Storing OCDB data ...");
5e993b6f 268 Int_t resultCDB = StoreOCDB(fgkMainCDB);
86aa42c3 269
c88ad5db 270 Log("SHUTTLE","StoreOCDB - Storing reference data ...");
5e993b6f 271 Int_t resultRef = StoreOCDB(fgkMainRefStorage);
9827400b 272
c88ad5db 273 Log("SHUTTLE","StoreOCDB - Storing reference files ...");
274 Bool_t resultRefFiles = CopyFilesToGrid("reference");
275
276 Bool_t resultMetadata = kTRUE;
277 if(fCurrentDetector == "GRP")
278 {
7802e884 279 Log("SHUTTLE","StoreOCDB - Storing Run Metadata file ...");
c88ad5db 280 resultMetadata = CopyFilesToGrid("metadata");
281 }
9827400b 282
5e993b6f 283 Int_t storeResult = 0;
284
285 if (resultCDB < 0 || resultRef < 0 || resultRefFiles == kFALSE || resultMetadata == kFALSE)
286 storeResult = -1;
287 else if (resultCDB > 0 || resultRef > 0)
288 storeResult = 1;
289
290 if (storeResult < 0)
291 {
292 Log("SHUTTLE",
293 Form("\t\t\t****** run %d - %s: STORAGE ERROR ******",
294 GetCurrentRun(), fCurrentDetector.Data()));
295 UpdateShuttleStatus(AliShuttleStatus::kStoreError);
296 }
297 else if (storeResult > 0)
298 {
299 Log("SHUTTLE",
300 Form("\t\t\t****** run %d - %s: STORAGE DELAYED ******",
301 GetCurrentRun(), fCurrentDetector.Data()));
302 UpdateShuttleStatus(AliShuttleStatus::kStoreDelayed);
303 }
304 else if (storeResult == 0)
305 {
306 Log("SHUTTLE",
307 Form("\t\t\t****** run %d - %s: DONE ******",
308 GetCurrentRun(), fCurrentDetector.Data()));
309 UpdateShuttleStatus(AliShuttleStatus::kDone);
310 UpdateShuttleLogbook(fCurrentDetector, "DONE");
311 }
312
313 return (storeResult == 0);
3301427a 314}
315
316//______________________________________________________________________________________________
5e993b6f 317Int_t AliShuttle::StoreOCDB(const TString& gridURI)
3301427a 318{
319 //
320 // Called by StoreOCDB(), performs actual storage to the main OCDB and reference storages (Grid)
321 //
5e993b6f 322 // Return code:
323 // -2 initialization error
324 // -1 storage error
325 // 0 success
326 // 1 storage delayed (e.g. previous unprocessed runs)
327 //
3301427a 328
329 TObjArray* gridIds=0;
330
331 Bool_t result = kTRUE;
5e993b6f 332 Bool_t delayed = kFALSE;
3301427a 333
334 const char* type = 0;
335 TString localURI;
336 if(gridURI == fgkMainCDB) {
337 type = "OCDB";
338 localURI = fgkLocalCDB;
339 } else if(gridURI == fgkMainRefStorage) {
340 type = "reference";
341 localURI = fgkLocalRefStorage;
342 } else {
343 AliError(Form("Invalid storage URI: %s", gridURI.Data()));
5e993b6f 344 return -2;
3301427a 345 }
346
347 AliCDBManager* man = AliCDBManager::Instance();
348
349 AliCDBStorage *gridSto = man->GetStorage(gridURI);
350 if(!gridSto) {
351 Log("SHUTTLE",
352 Form("StoreOCDB - cannot activate main %s storage", type));
5e993b6f 353 return -2;
3301427a 354 }
355
356 gridIds = gridSto->GetQueryCDBList();
357
358 // get objects previously stored in local CDB
359 AliCDBStorage *localSto = man->GetStorage(localURI);
360 if(!localSto) {
361 Log("SHUTTLE",
362 Form("StoreOCDB - cannot activate local %s storage", type));
5e993b6f 363 return -2;
3301427a 364 }
365 AliCDBPath aPath(GetOfflineDetName(fCurrentDetector.Data()),"*","*");
366 // Local objects were stored with current run as Grid version!
367 TList* localEntries = localSto->GetAll(aPath.GetPath(), GetCurrentRun(), GetCurrentRun());
368 localEntries->SetOwner(1);
369
370 // loop on local stored objects
371 TIter localIter(localEntries);
372 AliCDBEntry *aLocEntry = 0;
373 while((aLocEntry = dynamic_cast<AliCDBEntry*> (localIter.Next()))){
374 aLocEntry->SetOwner(1);
375 AliCDBId aLocId = aLocEntry->GetId();
376 aLocEntry->SetVersion(-1);
377 aLocEntry->SetSubVersion(-1);
6c0c9948 378
379 Log(fCurrentDetector.Data(), Form("Attempting to store %s", aLocId.ToString().Data()));
3301427a 380
381 // If local object is valid up to infinity we store it only if it is
382 // the first unprocessed run!
383 if (aLocId.GetLastRun() == AliCDBRunRange::Infinity() &&
384 !fFirstUnprocessed[GetDetPos(fCurrentDetector)])
385 {
386 Log("SHUTTLE", Form("StoreOCDB - %s: object %s has validity infinite but "
387 "there are previous unprocessed runs!",
388 fCurrentDetector.Data(), aLocId.GetPath().Data()));
5e993b6f 389 Log(fCurrentDetector.Data(), Form("StoreOCDB - %s: object %s has validity infinite but "
390 "there are previous unprocessed runs!",
391 fCurrentDetector.Data(), aLocId.GetPath().Data()));
392 delayed = kTRUE;
3301427a 393 continue;
394 }
395
396 // loop on Grid valid Id's
397 Bool_t store = kTRUE;
398 TIter gridIter(gridIds);
399 AliCDBId* aGridId = 0;
6c0c9948 400 while ((aGridId = dynamic_cast<AliCDBId*> (gridIter.Next()))) {
401 if (aGridId->GetPath() != aLocId.GetPath())
402 continue;
3301427a 403 // skip all objects valid up to infinity
6c0c9948 404 if (aGridId->GetLastRun() == AliCDBRunRange::Infinity())
405 continue;
406
3301427a 407 // if we get here, it means there's already some more recent object stored on Grid!
6c0c9948 408 Log(fCurrentDetector.Data(),
409 Form("StoreOCDB - A more recent object already exists in %s storage: <%s>",
410 type, aGridId->ToString().Data()));
411
3301427a 412 store = kFALSE;
413 break;
414 }
415
6c0c9948 416 Bool_t storeOk = kFALSE;
417 if (store)
418 {
419 Log(fCurrentDetector.Data(), Form("Prechecks succeeded. Ready to store %s", aLocId.ToString().Data()));
420 storeOk = gridSto->Put(aLocEntry);
421 if (storeOk) {
3301427a 422 Log("SHUTTLE",
6c0c9948 423 Form("StoreOCDB - Object <%s> successfully put into %s storage",
424 aLocId.ToString().Data(), type));
2d9019b4 425 Log(fCurrentDetector.Data(),
426 Form("StoreOCDB - Object <%s> successfully put into %s storage",
6c0c9948 427 aLocId.ToString().Data(), type));
428 } else {
429 Log("SHUTTLE",
430 Form("StoreOCDB - Grid %s storage of object <%s> failed",
431 type, aLocId.ToString().Data()));
432 Log(fCurrentDetector.Data(),
433 Form("StoreOCDB - Grid %s storage of object <%s> failed",
434 type, aLocId.ToString().Data()));
435 result = kFALSE;
3301427a 436 }
6c0c9948 437 }
438
439 if (!store || storeOk) {
440 // removing local file...
3301427a 441 TString filename;
442 localSto->IdToFilename(aLocId, filename);
c88ad5db 443 Log("SHUTTLE", Form("StoreOCDB - Removing local file %s", filename.Data()));
3301427a 444 RemoveFile(filename.Data());
b948db8d 445 }
446 }
3301427a 447 localEntries->Clear();
2bb7b766 448
5e993b6f 449 Int_t returnCode = 0;
450
451 if (result == kFALSE)
452 returnCode = -1;
453 else if (delayed != kFALSE)
454 returnCode = 1;
455
456 Log("SHUTTLE", Form("StoreOCDB - Returning with %d (result = %d, delayed = %d)", returnCode, result, delayed));
457 Log(fCurrentDetector.Data(), Form("StoreOCDB - Returning with %d (result = %d, delayed = %d)", returnCode, result, delayed));
458
459 return returnCode;
3301427a 460}
461
546242fb 462//______________________________________________________________________________________________
463Bool_t AliShuttle::CleanReferenceStorage(const char* detector)
464{
2d9019b4 465 // clears the directory used to store reference files of a given subdetector
546242fb 466
467 AliCDBManager* man = AliCDBManager::Instance();
468 AliCDBStorage* sto = man->GetStorage(fgkLocalRefStorage);
2d9019b4 469 TString localBaseFolder = sto->GetBaseFolder();
470
471 TString targetDir = GetRefFilePrefix(localBaseFolder.Data(), detector);
472
d524ade6 473 Log("SHUTTLE", Form("CleanReferenceStorage - Cleaning %s", targetDir.Data()));
2d9019b4 474
475 TString begin;
476 begin.Form("%d_", GetCurrentRun());
477
478 TSystemDirectory* baseDir = new TSystemDirectory("/", targetDir);
479 if (!baseDir)
480 return kTRUE;
481
482 TList* dirList = baseDir->GetListOfFiles();
483 delete baseDir;
484
485 if (!dirList) return kTRUE;
486
487 if (dirList->GetEntries() < 3)
488 {
489 delete dirList;
490 return kTRUE;
491 }
492
493 Int_t nDirs = 0, nDel = 0;
494 TIter dirIter(dirList);
495 TSystemFile* entry = 0;
546242fb 496
2d9019b4 497 Bool_t success = kTRUE;
546242fb 498
2d9019b4 499 while ((entry = dynamic_cast<TSystemFile*> (dirIter.Next())))
500 {
501 if (entry->IsDirectory())
502 continue;
503
504 TString fileName(entry->GetName());
505 if (!fileName.BeginsWith(begin))
506 continue;
507
508 nDirs++;
509
510 // delete file
511 Int_t result = gSystem->Unlink(fileName.Data());
512
513 if (result)
514 {
d524ade6 515 Log("SHUTTLE", Form("CleanReferenceStorage - Could not delete file %s!", fileName.Data()));
2d9019b4 516 success = kFALSE;
517 } else {
518 nDel++;
519 }
520 }
521
522 if(nDirs > 0)
523 Log("SHUTTLE", Form("CleanReferenceStorage - %d (over %d) reference files in folder %s were deleted.",
524 nDel, nDirs, targetDir.Data()));
525
526
527 delete dirList;
528 return success;
529
530
531
532
533
546242fb 534
535 Int_t result = gSystem->GetPathInfo(targetDir, 0, (Long64_t*) 0, 0, 0);
536 if (result == 0)
537 {
538 // delete directory
d524ade6 539 result = gSystem->Exec(Form("rm -rf %s", targetDir.Data()));
546242fb 540 if (result != 0)
541 {
d524ade6 542 Log("SHUTTLE", Form("CleanReferenceStorage - Could not clean directory %s", targetDir.Data()));
546242fb 543 return kFALSE;
544 }
545 }
546
547 result = gSystem->mkdir(targetDir, kTRUE);
548 if (result != 0)
549 {
c88ad5db 550 Log("SHUTTLE", Form("CleanReferenceStorage - Error creating base directory %s", targetDir.Data()));
546242fb 551 return kFALSE;
552 }
553
554 return kTRUE;
555}
556
9827400b 557//______________________________________________________________________________________________
558Bool_t AliShuttle::StoreReferenceFile(const char* detector, const char* localFile, const char* gridFileName)
559{
560 //
3c2a21c8 561 // Stores reference file directly (without opening it). This function stores the file locally.
9827400b 562 //
3c2a21c8 563 // The file is stored under the following location:
564 // <base folder of local reference storage>/<DET>/<RUN#>_<gridFileName>
565 // where <gridFileName> is the second parameter given to the function
566 //
9827400b 567
568 if (fTestMode & kErrorStorage)
569 {
570 Log(fCurrentDetector, "StoreReferenceFile - In TESTMODE - Simulating error while storing locally");
571 return kFALSE;
572 }
573
574 AliCDBManager* man = AliCDBManager::Instance();
575 AliCDBStorage* sto = man->GetStorage(fgkLocalRefStorage);
576
577 TString localBaseFolder = sto->GetBaseFolder();
578
d524ade6 579 TString target = GetRefFilePrefix(localBaseFolder.Data(), detector);
580 target.Append(Form("/%d_%s", GetCurrentRun(), gridFileName));
9827400b 581
d524ade6 582 return CopyFileLocally(localFile, target);
c88ad5db 583}
584
585//______________________________________________________________________________________________
586Bool_t AliShuttle::StoreRunMetadataFile(const char* localFile, const char* gridFileName)
587{
588 //
589 // Stores Run metadata file to the Grid, in the run folder
590 //
591 // Only GRP can call this function.
592
593 if (fTestMode & kErrorStorage)
594 {
595 Log(fCurrentDetector, "StoreRunMetaDataFile - In TESTMODE - Simulating error while storing locally");
596 return kFALSE;
597 }
598
599 AliCDBManager* man = AliCDBManager::Instance();
600 AliCDBStorage* sto = man->GetStorage(fgkLocalRefStorage);
601
602 TString localBaseFolder = sto->GetBaseFolder();
603
604 // Build Run level folder
955d2abc 605 // folder = /alice/data/year/lhcPeriod/runNb/raw
c88ad5db 606
c88ad5db 607
675f64cd 608 TString lhcPeriod = GetLHCPeriod();
c88ad5db 609 if (lhcPeriod.Length() == 0)
610 {
611 Log("SHUTTLE","StoreRunMetaDataFile - LHCPeriod not found in logbook!");
612 return 0;
613 }
a1c5723b 614
615 // TODO partitions with one detector only write data into LHCperiod_DET
2694f253 616 TString partition = GetRunParameter("detector");
a1c5723b 617
618 if (partition.Length() > 0 && partition != "ALICE")
619 {
620 lhcPeriod.Append(Form("_%s", partition.Data()));
621 Log(fCurrentDetector, Form("Run data tags merged file will be written in %s",
622 lhcPeriod.Data()));
623 }
8884d71e 624
b8cf5e0f 625 TString target = Form("%s/GRP/RunMetadata%s%d/%s/%09d/raw/%s",
626 localBaseFolder.Data(), fConfig->GetAlienPath(), GetCurrentYear(),
d524ade6 627 lhcPeriod.Data(), GetCurrentRun(), gridFileName);
c88ad5db 628
d524ade6 629 return CopyFileLocally(localFile, target);
c88ad5db 630}
631
632//______________________________________________________________________________________________
d524ade6 633Bool_t AliShuttle::CopyFileLocally(const char* localFile, const TString& target)
c88ad5db 634{
635 //
636 // Stores file locally. Called by StoreReferenceFile and StoreRunMetadataFile
d524ade6 637 // Files are temporarily stored in the local reference storage. When the preprocessor
638 // finishes, the Shuttle calls CopyFilesToGrid to transfer the files to AliEn
639 // (in reference or run level folders)
c88ad5db 640 //
641
d524ade6 642 TString targetDir(target(0, target.Last('/')));
643
644 //try to open base dir folder, if it does not exist
2d9019b4 645 void* dir = gSystem->OpenDirectory(targetDir.Data());
646 if (dir == NULL) {
647 if (gSystem->mkdir(targetDir.Data(), kTRUE)) {
a1c5723b 648 Log("SHUTTLE", Form("CopyFileLocally - Can't open directory <%s>", targetDir.Data()));
2d9019b4 649 return kFALSE;
650 }
651
652 } else {
653 gSystem->FreeDirectory(dir);
654 }
9827400b 655
7d43a416 656 Int_t result = 0;
657
658 result = gSystem->GetPathInfo(localFile, 0, (Long64_t*) 0, 0, 0);
9827400b 659 if (result)
660 {
a1c5723b 661 Log("SHUTTLE", Form("CopyFileLocally - %s does not exist", localFile));
546242fb 662 return kFALSE;
9827400b 663 }
546242fb 664
7d43a416 665 result = gSystem->GetPathInfo(target, 0, (Long64_t*) 0, 0, 0);
666 if (!result)
667 {
a1c5723b 668 Log("SHUTTLE", Form("CopyFileLocally - target file %s already exist, removing...", target.Data()));
7d43a416 669 if (gSystem->Unlink(target.Data()))
670 {
a1c5723b 671 Log("SHUTTLE", Form("CopyFileLocally - Could not remove existing target file %s!", target.Data()));
7d43a416 672 return kFALSE;
673 }
674 }
675
9827400b 676 result = gSystem->CopyFile(localFile, target);
677
678 if (result == 0)
679 {
a1c5723b 680 Log("SHUTTLE", Form("CopyFileLocally - File %s stored locally to %s", localFile, target.Data()));
9827400b 681 return kTRUE;
682 }
683 else
684 {
a1c5723b 685 Log("SHUTTLE", Form("CopyFileLocally - Could not store file %s to %s! Error code = %d",
546242fb 686 localFile, target.Data(), result));
9827400b 687 return kFALSE;
688 }
c88ad5db 689
690
691
9827400b 692}
693
694//______________________________________________________________________________________________
c88ad5db 695Bool_t AliShuttle::CopyFilesToGrid(const char* type)
9827400b 696{
697 //
c88ad5db 698 // Transfers local files to the Grid. Local files can be reference files
699 // or run metadata file (from GRP only).
9827400b 700 //
c88ad5db 701 // According to the type (ref, metadata) the files are stored under the following location:
702 // ref --> <base folder of reference storage>/<DET>/<RUN#>_<gridFileName>
703 // metadata --> <run data folder>/<MetadataFileName>
86aa42c3 704 //
c88ad5db 705
9827400b 706 AliCDBManager* man = AliCDBManager::Instance();
707 AliCDBStorage* sto = man->GetStorage(fgkLocalRefStorage);
708 if (!sto)
709 return kFALSE;
710 TString localBaseFolder = sto->GetBaseFolder();
9827400b 711
c88ad5db 712 TString dir;
713 TString alienDir;
9827400b 714 TString begin;
9827400b 715
c88ad5db 716 if (strcmp(type, "reference") == 0)
717 {
718 dir = GetRefFilePrefix(localBaseFolder.Data(), fCurrentDetector.Data());
719 AliCDBStorage* gridSto = man->GetStorage(fgkMainRefStorage);
720 if (!gridSto)
721 return kFALSE;
722 TString gridBaseFolder = gridSto->GetBaseFolder();
723 alienDir = GetRefFilePrefix(gridBaseFolder.Data(), fCurrentDetector.Data());
724 begin = Form("%d_", GetCurrentRun());
725 }
726 else if (strcmp(type, "metadata") == 0)
727 {
c88ad5db 728
675f64cd 729 TString lhcPeriod = GetLHCPeriod();
c88ad5db 730
731 if (lhcPeriod.Length() == 0)
732 {
733 Log("SHUTTLE","CopyFilesToGrid - LHCPeriod not found in logbook!");
734 return 0;
735 }
736
a1c5723b 737 // TODO partitions with one detector only write data into LHCperiod_DET
2694f253 738 TString partition = GetRunParameter("detector");
a1c5723b 739
740 if (partition.Length() > 0 && partition != "ALICE")
741 {
742 lhcPeriod.Append(Form("_%s", partition.Data()));
743 }
744
b8cf5e0f 745 dir = Form("%s/GRP/RunMetadata%s%d/%s/%09d/raw",
746 localBaseFolder.Data(), fConfig->GetAlienPath(), GetCurrentYear(),
c88ad5db 747 lhcPeriod.Data(), GetCurrentRun());
b8cf5e0f 748 alienDir = dir(dir.Index(fConfig->GetAlienPath()), dir.Length());
675f64cd 749
c88ad5db 750 begin = "";
751 }
752 else
753 {
754 Log("SHUTTLE", "CopyFilesToGrid - Unexpected: type label must be reference or metadata!");
755 return kFALSE;
756 }
757
9827400b 758 TSystemDirectory* baseDir = new TSystemDirectory("/", dir);
3d8bc902 759 if (!baseDir)
760 return kTRUE;
761
2d9019b4 762 TList* dirList = baseDir->GetListOfFiles();
763 delete baseDir;
764
765 if (!dirList) return kTRUE;
766
767 if (dirList->GetEntries() < 3)
3d8bc902 768 {
2d9019b4 769 delete dirList;
9827400b 770 return kTRUE;
3d8bc902 771 }
2d9019b4 772
546242fb 773 if (!gGrid)
774 {
c88ad5db 775 Log("SHUTTLE", "CopyFilesToGrid - Connection to Grid failed: Cannot continue!");
2d9019b4 776 delete dirList;
546242fb 777 return kFALSE;
778 }
779
2d9019b4 780 Int_t nDirs = 0, nTransfer = 0;
781 TIter dirIter(dirList);
782 TSystemFile* entry = 0;
783
9827400b 784 Bool_t success = kTRUE;
3d8bc902 785 Bool_t first = kTRUE;
9827400b 786
2d9019b4 787 while ((entry = dynamic_cast<TSystemFile*> (dirIter.Next())))
788 {
9827400b 789 if (entry->IsDirectory())
790 continue;
791
792 TString fileName(entry->GetName());
793 if (!fileName.BeginsWith(begin))
794 continue;
795
2d9019b4 796 nDirs++;
797
3d8bc902 798 if (first)
799 {
800 first = kFALSE;
c88ad5db 801 // check that folder exists, otherwise create it
3d8bc902 802 TGridResult* result = gGrid->Ls(alienDir.Data(), "a");
803
804 if (!result)
2d9019b4 805 {
806 delete dirList;
3d8bc902 807 return kFALSE;
2d9019b4 808 }
3d8bc902 809
546242fb 810 if (!result->GetFileName(1)) // TODO: It looks like element 0 is always 0!!
3d8bc902 811 {
675f64cd 812 // TODO It does not work currently! Bug in TAliEn::Mkdir
813 // TODO Manually fixed in local root v5-16-00
c88ad5db 814 if (!gGrid->Mkdir(alienDir.Data(),"-p",0))
3d8bc902 815 {
c88ad5db 816 Log("SHUTTLE", Form("CopyFilesToGrid - Cannot create directory %s",
3d8bc902 817 alienDir.Data()));
2d9019b4 818 delete dirList;
3d8bc902 819 return kFALSE;
546242fb 820 } else {
c88ad5db 821 Log("SHUTTLE",Form("CopyFilesToGrid - Folder %s created", alienDir.Data()));
3d8bc902 822 }
823
546242fb 824 } else {
c88ad5db 825 Log("SHUTTLE",Form("CopyFilesToGrid - Folder %s found", alienDir.Data()));
3d8bc902 826 }
827 }
828
9827400b 829 TString fullLocalPath;
830 fullLocalPath.Form("%s/%s", dir.Data(), fileName.Data());
831
832 TString fullGridPath;
833 fullGridPath.Form("alien://%s/%s", alienDir.Data(), fileName.Data());
834
a986b218 835 Bool_t result = TFile::Cp(fullLocalPath, fullGridPath);
9827400b 836
837 if (result)
838 {
c88ad5db 839 Log("SHUTTLE", Form("CopyFilesToGrid - Copying local file %s to %s succeeded!",
840 fullLocalPath.Data(), fullGridPath.Data()));
9827400b 841 RemoveFile(fullLocalPath);
2d9019b4 842 nTransfer++;
9827400b 843 }
844 else
845 {
c88ad5db 846 Log("SHUTTLE", Form("CopyFilesToGrid - Copying local file %s to %s FAILED!",
847 fullLocalPath.Data(), fullGridPath.Data()));
9827400b 848 success = kFALSE;
849 }
850 }
2d9019b4 851
c88ad5db 852 Log("SHUTTLE", Form("CopyFilesToGrid - %d (over %d) files in folder %s copied to Grid.",
853 nTransfer, nDirs, dir.Data()));
2d9019b4 854
855
856 delete dirList;
9827400b 857 return success;
858}
859
2d9019b4 860//______________________________________________________________________________________________
861const char* AliShuttle::GetRefFilePrefix(const char* base, const char* detector)
862{
863 //
864 // Get folder name of reference files
865 //
866
867 TString offDetStr(GetOfflineDetName(detector));
868 TString dir;
869 if (offDetStr == "ITS" || offDetStr == "MUON" || offDetStr == "PHOS")
870 {
871 dir.Form("%s/%s/%s", base, offDetStr.Data(), detector);
872 } else {
873 dir.Form("%s/%s", base, offDetStr.Data());
874 }
875
876 return dir.Data();
877
878
879}
c88ad5db 880
3301427a 881//______________________________________________________________________________________________
882void AliShuttle::CleanLocalStorage(const TString& uri)
883{
9827400b 884 //
885 // Called in case the preprocessor is declared failed. Remove remaining objects from the local storages.
886 //
3301427a 887
888 const char* type = 0;
889 if(uri == fgkLocalCDB) {
890 type = "OCDB";
891 } else if(uri == fgkLocalRefStorage) {
546242fb 892 type = "Reference";
3301427a 893 } else {
894 AliError(Form("Invalid storage URI: %s", uri.Data()));
895 return;
896 }
897
898 AliCDBManager* man = AliCDBManager::Instance();
b948db8d 899
3301427a 900 // open local storage
901 AliCDBStorage *localSto = man->GetStorage(uri);
902 if(!localSto) {
903 Log("SHUTTLE",
904 Form("CleanLocalStorage - cannot activate local %s storage", type));
905 return;
906 }
907
908 TString filename(Form("%s/%s/*/Run*_v%d_s*.root",
546242fb 909 localSto->GetBaseFolder().Data(), GetOfflineDetName(fCurrentDetector.Data()), GetCurrentRun()));
3301427a 910
c88ad5db 911 AliDebug(2, Form("filename = %s", filename.Data()));
3301427a 912
c88ad5db 913 Log("SHUTTLE", Form("Removing remaining local files for run %d and detector %s ...",
3301427a 914 GetCurrentRun(), fCurrentDetector.Data()));
915
916 RemoveFile(filename.Data());
917
918}
919
920//______________________________________________________________________________________________
921void AliShuttle::RemoveFile(const char* filename)
922{
9827400b 923 //
924 // removes local file
925 //
3301427a 926
927 TString command(Form("rm -f %s", filename));
928
929 Int_t result = gSystem->Exec(command.Data());
930 if(result != 0)
931 {
932 Log("SHUTTLE", Form("RemoveFile - %s: Cannot remove file %s!",
933 fCurrentDetector.Data(), filename));
934 }
73abe331 935}
936
b948db8d 937//______________________________________________________________________________________________
5164a766 938AliShuttleStatus* AliShuttle::ReadShuttleStatus()
939{
9827400b 940 //
941 // Reads the AliShuttleStatus from the CDB
942 //
5164a766 943
2bb7b766 944 if (fStatusEntry){
945 delete fStatusEntry;
946 fStatusEntry = 0;
947 }
5164a766 948
3ebbf3c4 949 Int_t path1 = GetCurrentRun()/10000;
10a5a932 950 fStatusEntry = AliCDBManager::Instance()->GetStorage(GetLocalCDB())
3ebbf3c4 951 ->Get(Form("/SHUTTLE/%s/%d", fCurrentDetector.Data(), path1), GetCurrentRun());
5164a766 952
2bb7b766 953 if (!fStatusEntry) return 0;
954 fStatusEntry->SetOwner(1);
5164a766 955
2bb7b766 956 AliShuttleStatus* status = dynamic_cast<AliShuttleStatus*> (fStatusEntry->GetObject());
957 if (!status) {
958 AliError("Invalid object stored to CDB!");
959 return 0;
960 }
5164a766 961
2bb7b766 962 return status;
5164a766 963}
964
965//______________________________________________________________________________________________
7bfb2090 966Bool_t AliShuttle::WriteShuttleStatus(AliShuttleStatus* status)
5164a766 967{
9827400b 968 //
969 // writes the status for one subdetector
970 //
2bb7b766 971
972 if (fStatusEntry){
973 delete fStatusEntry;
974 fStatusEntry = 0;
975 }
5164a766 976
2bb7b766 977 Int_t run = GetCurrentRun();
3ebbf3c4 978 Int_t path1 = run/10000;
979 TString path1_string = Form("%d",path1);
5164a766 980
3ebbf3c4 981 AliCDBId id(AliCDBPath("SHUTTLE", fCurrentDetector, path1_string), run, run);
5164a766 982
2bb7b766 983 fStatusEntry = new AliCDBEntry(status, id, new AliCDBMetaData);
984 fStatusEntry->SetOwner(1);
5164a766 985
2bb7b766 986 UInt_t result = AliCDBManager::Instance()->GetStorage(fgkLocalCDB)->Put(fStatusEntry);
7bfb2090 987
2bb7b766 988 if (!result) {
3301427a 989 Log("SHUTTLE", Form("WriteShuttleStatus - Failed for %s, run %d",
990 fCurrentDetector.Data(), run));
2bb7b766 991 return kFALSE;
992 }
e7f62f16 993
8e828d39 994 SendMLDetInfo();
7bfb2090 995
2bb7b766 996 return kTRUE;
5164a766 997}
998
999//______________________________________________________________________________________________
1000void AliShuttle::UpdateShuttleStatus(AliShuttleStatus::Status newStatus, Bool_t increaseCount)
1001{
9827400b 1002 //
1003 // changes the AliShuttleStatus for the given detector and run to the given status
1004 //
5164a766 1005
2bb7b766 1006 if (!fStatusEntry){
1007 AliError("UNEXPECTED: fStatusEntry empty");
1008 return;
1009 }
5164a766 1010
2bb7b766 1011 AliShuttleStatus* status = dynamic_cast<AliShuttleStatus*> (fStatusEntry->GetObject());
5164a766 1012
2bb7b766 1013 if (!status){
c88ad5db 1014 Log("SHUTTLE", "UpdateShuttleStatus - UNEXPECTED: status could not be read from current CDB entry");
2bb7b766 1015 return;
1016 }
5164a766 1017
2c15234c 1018 TString actionStr = Form("UpdateShuttleStatus - %s: Changing state from %s to %s",
eba76848 1019 fCurrentDetector.Data(),
36c99a6a 1020 status->GetStatusName(),
eba76848 1021 status->GetStatusName(newStatus));
cb343cfd 1022 Log("SHUTTLE", actionStr);
1023 SetLastAction(actionStr);
5164a766 1024
2bb7b766 1025 status->SetStatus(newStatus);
1026 if (increaseCount) status->IncreaseCount();
5164a766 1027
2bb7b766 1028 AliCDBManager::Instance()->GetStorage(fgkLocalCDB)->Put(fStatusEntry);
e7f62f16 1029
8e828d39 1030 SendMLDetInfo();
5164a766 1031}
e7f62f16 1032
1033//______________________________________________________________________________________________
8e828d39 1034void AliShuttle::SendMLDetInfo()
e7f62f16 1035{
1036 //
1037 // sends ML information about the current status of the current detector being processed
1038 //
1039
1040 AliShuttleStatus* status = dynamic_cast<AliShuttleStatus*> (fStatusEntry->GetObject());
1041
1042 if (!status){
8e828d39 1043 Log("SHUTTLE", "SendMLDetInfo - UNEXPECTED: status could not be read from current CDB entry");
e7f62f16 1044 return;
1045 }
1046
1047 TMonaLisaText mlStatus(Form("%s_status", fCurrentDetector.Data()), status->GetStatusName());
1048 TMonaLisaValue mlRetryCount(Form("%s_count", fCurrentDetector.Data()), status->GetCount());
1049
1050 TList mlList;
1051 mlList.Add(&mlStatus);
1052 mlList.Add(&mlRetryCount);
1053
ee6f7523 1054 TString mlID;
1055 mlID.Form("%d", GetCurrentRun());
1056 fMonaLisa->SendParameters(&mlList, mlID);
e7f62f16 1057}
1058
5164a766 1059//______________________________________________________________________________________________
1060Bool_t AliShuttle::ContinueProcessing()
1061{
9827400b 1062 // this function reads the AliShuttleStatus information from CDB and
1063 // checks if the processing should be continued
1064 // if yes it returns kTRUE and updates the AliShuttleStatus with nextStatus
2bb7b766 1065
926cbc0e 1066 if (!fConfig->HostProcessDetector(fCurrentDetector))
1067 return kFALSE;
57c1a579 1068
1069 AliPreprocessor* aPreprocessor =
1070 dynamic_cast<AliPreprocessor*> (fPreprocessorMap.GetValue(fCurrentDetector));
1071 if (!aPreprocessor)
1072 {
c88ad5db 1073 Log("SHUTTLE", Form("ContinueProcessing - %s: no preprocessor registered", fCurrentDetector.Data()));
57c1a579 1074 return kFALSE;
1075 }
1076
2bb7b766 1077 AliShuttleLogbookEntry::Status entryStatus =
eba76848 1078 fLogbookEntry->GetDetectorStatus(fCurrentDetector);
2bb7b766 1079
5e993b6f 1080 if (entryStatus != AliShuttleLogbookEntry::kUnprocessed) {
c88ad5db 1081 Log("SHUTTLE", Form("ContinueProcessing - %s is %s",
2bb7b766 1082 fCurrentDetector.Data(),
1083 fLogbookEntry->GetDetectorStatusName(entryStatus)));
1084 return kFALSE;
1085 }
1086
1087 // if we get here, according to Shuttle logbook subdetector is in UNPROCESSED state
be48e3ea 1088
1089 // check if current run is first unprocessed run for current detector
1090 if (fConfig->StrictRunOrder(fCurrentDetector) &&
1091 !fFirstUnprocessed[GetDetPos(fCurrentDetector)])
1092 {
86aa42c3 1093 if (fTestMode == kNone)
1094 {
c88ad5db 1095 Log("SHUTTLE", Form("ContinueProcessing - %s requires strict run ordering"
1096 " but this is not the first unprocessed run!"));
86aa42c3 1097 return kFALSE;
1098 }
1099 else
1100 {
c88ad5db 1101 Log("SHUTTLE", Form("ContinueProcessing - In TESTMODE - "
1102 "Although %s requires strict run ordering "
1103 "and this is not the first unprocessed run, "
1104 "the SHUTTLE continues"));
86aa42c3 1105 }
be48e3ea 1106 }
1107
926cbc0e 1108 // Is the subdetector processed first time for this run?
1109 fFirstProcessing = kFALSE;
1110
2bb7b766 1111 AliShuttleStatus* status = ReadShuttleStatus();
1112 if (!status) {
1113 // first time
1114 Log("SHUTTLE", Form("ContinueProcessing - %s: Processing first time",
1115 fCurrentDetector.Data()));
1116 status = new AliShuttleStatus(AliShuttleStatus::kStarted);
926cbc0e 1117 fFirstProcessing = kTRUE;
2bb7b766 1118 return WriteShuttleStatus(status);
1119 }
1120
ffa25f20 1121 // The following case shouldn't happen if Shuttle Logbook was correctly updated.
2bb7b766 1122 // If it happens it may mean Logbook updating failed... let's do it now!
1123 if (status->GetStatus() == AliShuttleStatus::kDone ||
ffa25f20 1124 status->GetStatus() == AliShuttleStatus::kFailed ||
1125 status->GetStatus() == AliShuttleStatus::kSkipped) {
2bb7b766 1126 Log("SHUTTLE", Form("ContinueProcessing - %s is already %s. Updating Shuttle Logbook",
1127 fCurrentDetector.Data(),
1128 status->GetStatusName(status->GetStatus())));
ffa25f20 1129
1130 if (status->GetStatus() == AliShuttleStatus::kSkipped)
1131 {
1132 UpdateShuttleLogbook(fCurrentDetector.Data(), "DONE");
1133 }
1134 else
1135 UpdateShuttleLogbook(fCurrentDetector.Data(), status->GetStatusName(status->GetStatus()));
1136
2bb7b766 1137 return kFALSE;
1138 }
1139
5e993b6f 1140 if (status->GetStatus() == AliShuttleStatus::kStoreStarted || status->GetStatus() == AliShuttleStatus::kStoreDelayed ||status->GetStatus() == AliShuttleStatus::kStoreError) {
2bb7b766 1141 Log("SHUTTLE",
c88ad5db 1142 Form("ContinueProcessing - %s: Grid storage of one or more "
1143 "objects failed. Trying again now",
2bb7b766 1144 fCurrentDetector.Data()));
5e993b6f 1145 StoreOCDB();
2bb7b766 1146 return kFALSE;
1147 }
1148
1149 // if we get here, there is a restart
57c1a579 1150 Bool_t cont = kFALSE;
2bb7b766 1151
1152 // abort conditions
cb343cfd 1153 if (status->GetCount() >= fConfig->GetMaxRetries()) {
57c1a579 1154 Log("SHUTTLE", Form("ContinueProcessing - %s failed %d times in status %s - "
1155 "Updating Shuttle Logbook", fCurrentDetector.Data(),
2bb7b766 1156 status->GetCount(), status->GetStatusName()));
1157 UpdateShuttleLogbook(fCurrentDetector.Data(), "FAILED");
e7f62f16 1158 UpdateShuttleStatus(AliShuttleStatus::kFailed);
3301427a 1159
1160 // there may still be objects in local OCDB and reference storage
1161 // and FXS databases may be not updated: do it now!
9827400b 1162
1163 // TODO Currently disabled, we want to keep files in case of failure!
1164 // CleanLocalStorage(fgkLocalCDB);
1165 // CleanLocalStorage(fgkLocalRefStorage);
1166 // UpdateTableFailCase();
1167
1168 // Send mail to detector expert!
c88ad5db 1169 Log("SHUTTLE", Form("ContinueProcessing - Sending mail to %s expert...",
926cbc0e 1170 fCurrentDetector.Data()));
339adafa 1171 if (!SendMail(kPPEMail))
9827400b 1172 Log("SHUTTLE", Form("ContinueProcessing - Could not send mail to %s expert",
926cbc0e 1173 fCurrentDetector.Data()));
3301427a 1174
57c1a579 1175 } else {
1176 Log("SHUTTLE", Form("ContinueProcessing - %s: restarting. "
1177 "Aborted before with %s. Retry number %d.", fCurrentDetector.Data(),
1178 status->GetStatusName(), status->GetCount()));
9827400b 1179 Bool_t increaseCount = kTRUE;
c88ad5db 1180 if (status->GetStatus() == AliShuttleStatus::kDCSError ||
8da81210 1181 status->GetStatus() == AliShuttleStatus::kDCSStarted ||
1182 status->GetStatus() == AliShuttleStatus::kFXSError)
c88ad5db 1183 increaseCount = kFALSE;
675f64cd 1184
9827400b 1185 UpdateShuttleStatus(AliShuttleStatus::kStarted, increaseCount);
57c1a579 1186 cont = kTRUE;
2bb7b766 1187 }
1188
57c1a579 1189 return cont;
5164a766 1190}
1191
8e828d39 1192//______________________________________________________________________________________________
1193void AliShuttle::SendMLRunInfo(const char* status)
1194{
1195 //
1196 // Send information about this run to ML
1197
1198 TMonaLisaText mlStatus("SHUTTLE_status", status);
1199 TString runType(fLogbookEntry->GetRunType());
1200 if (strlen(fLogbookEntry->GetRunParameter("log")) > 0){
1201
1202 runType += "(";
1203 runType += fLogbookEntry->GetRunParameter("log");
1204 runType += ")";
1205 }
728290a4 1206 if (fLogbookEntry->GetDATestMode()){
1207 runType += " (DATest)";
1208 }
8e828d39 1209 TMonaLisaText mlRunType("SHUTTLE_runtype", runType);
1210
1211 TList mlList;
1212 mlList.Add(&mlStatus);
1213 mlList.Add(&mlRunType);
1214
1215 TString mlID;
1216 mlID.Form("%d", GetCurrentRun());
1217 fMonaLisa->SendParameters(&mlList, mlID);
1218}
1219
ef83773f 1220//______________________________________________________________________________________________
1221Int_t AliShuttle::GetMem(Int_t pid)
1222{
1223 // invokes ps to get the memory consumption of the process <pid>
1224 // returns -1 in case of error
1225
1226 TString checkStr;
1227 checkStr.Form("ps -o vsize --pid %d | tail -n 1", pid);
1228 FILE* pipe = gSystem->OpenPipe(checkStr, "r");
1229 if (!pipe)
1230 {
1231 Log("SHUTTLE", Form("Process - Error: "
1232 "Could not open pipe to %s", checkStr.Data()));
1233 return -1;
1234 }
1235
1236 char buffer[100];
1237 if (!fgets(buffer, 100, pipe))
1238 {
1239 Log("SHUTTLE", "Process - Error: ps did not return anything");
1240 gSystem->ClosePipe(pipe);
1241 return -1;
1242 }
1243 gSystem->ClosePipe(pipe);
1244
1245 //Log("SHUTTLE", Form("ps returned %s", buffer));
1246
1247 Int_t mem = 0;
1248 if ((sscanf(buffer, "%d\n", &mem) != 1) || !mem)
1249 {
1250 Log("SHUTTLE", "Process - Error: Could not parse output of ps");
1251 return -1;
1252 }
1253
1254 return mem;
1255}
1256
5164a766 1257//______________________________________________________________________________________________
2bb7b766 1258Bool_t AliShuttle::Process(AliShuttleLogbookEntry* entry)
58bc3020 1259{
73abe331 1260 //
b948db8d 1261 // Makes data retrieval for all detectors in the configuration.
2bb7b766 1262 // entry: Shuttle logbook entry, contains run paramenters and status of detectors
1263 // (Unprocessed, Inactive, Failed or Done).
d477ad88 1264 // Returns kFALSE in case of error occured and kTRUE otherwise
73abe331 1265 //
1266
9827400b 1267 if (!entry) return kFALSE;
2bb7b766 1268
1269 fLogbookEntry = entry;
1270
c88ad5db 1271 Log("SHUTTLE", Form("\t\t\t^*^*^*^*^*^*^*^*^*^*^*^* run %d: START ^*^*^*^*^*^*^*^*^*^*^*^*",
9827400b 1272 GetCurrentRun()));
2bb7b766 1273
339adafa 1274 CountOpenRuns();
1275
8e828d39 1276 // Send the information to ML
1277 SendMLRunInfo("Processing");
3301427a 1278
9827400b 1279 if (fLogbookEntry->IsDone())
1280 {
1281 Log("SHUTTLE","Process - Shuttle is already DONE. Updating logbook");
1282 UpdateShuttleLogbook("shuttle_done");
1283 fLogbookEntry = 0;
1284 return kTRUE;
1285 }
1286
1287 // read test mode if flag is set
1288 if (fReadTestMode)
1289 {
3d8bc902 1290 fTestMode = kNone;
9827400b 1291 TString logEntry(entry->GetRunParameter("log"));
1292 //printf("log entry = %s\n", logEntry.Data());
1293 TString searchStr("Testmode: ");
1294 Int_t pos = logEntry.Index(searchStr.Data());
1295 //printf("%d\n", pos);
1296 if (pos >= 0)
1297 {
1298 TSubString subStr = logEntry(pos + searchStr.Length(), logEntry.Length());
1299 //printf("%s\n", subStr.String().Data());
1300 TString newStr(subStr.Data());
1301 TObjArray* token = newStr.Tokenize(' ');
1302 if (token)
1303 {
1304 //token->Print();
1305 TObjString* tmpStr = dynamic_cast<TObjString*> (token->First());
1306 if (tmpStr)
1307 {
1308 Int_t testMode = tmpStr->String().Atoi();
1309 if (testMode > 0)
1310 {
c88ad5db 1311 Log("SHUTTLE", Form("Process - Enabling test mode %d", testMode));
9827400b 1312 SetTestMode((TestMode) testMode);
1313 }
1314 }
1315 delete token;
1316 }
1317 }
1318 }
c88ad5db 1319
eba76848 1320 fLogbookEntry->Print("all");
57f50b3c 1321
1322 // Initialization
d477ad88 1323 Bool_t hasError = kFALSE;
5164a766 1324
728290a4 1325 // Set the CDB and Reference folders according to the year
1326
1327 // build cdb paths (repeat each time, run might be a DATest run)
1328 if (!fLogbookEntry->GetDATestMode()){
1329 fgkMainCDB.Form("alien://folder=%s%d/OCDB?user=alidaq?cacheFold=/tmp/OCDBCache",
1330 fConfig->GetAlienPath(), GetCurrentYear());
1331
1332 fgkMainRefStorage.Form("alien://folder=%s%d/Reference?user=alidaq?cacheFold=/tmp/OCDBCache",
1333 fConfig->GetAlienPath(), GetCurrentYear());
1334 }
1335 else {
1336 fgkMainCDB.Form("alien://folder=%s%d/DATest/OCDB?user=alidaq?cacheFold=/tmp/OCDBCache",
1337 fConfig->GetAlienPath(), GetCurrentYear());
1338
1339 fgkMainRefStorage.Form("alien://folder=%s%d/DATest/Reference?user=alidaq?cacheFold=/tmp/OCDBCache",
1340 fConfig->GetAlienPath(), GetCurrentYear());
1341 }
1342
1343 AliDebug(2,Form("Main CDB storage = %s",fgkMainCDB.Data()));
1344 AliDebug(2,Form("Main Reference storage = %s",fgkMainRefStorage.Data()));
1345
57f50b3c 1346 // Loop on detectors in the configuration
b948db8d 1347 TIter iter(fConfig->GetDetectors());
2bb7b766 1348 TObjString* aDetector = 0;
b948db8d 1349
4508ab8b 1350 Bool_t first = kTRUE;
1351
be48e3ea 1352 while ((aDetector = (TObjString*) iter.Next()))
1353 {
7bfb2090 1354 fCurrentDetector = aDetector->String();
5164a766 1355
5e993b6f 1356 if (ContinueProcessing() == kFALSE)
1357 continue;
4508ab8b 1358
1359 if (first)
1360 {
1361 // only read QueryCDB when needed and only once
1362 AliCDBStorage *mainCDBSto = AliCDBManager::Instance()->GetStorage(fgkMainCDB);
1363 if(mainCDBSto) mainCDBSto->QueryCDB(GetCurrentRun());
1364 AliCDBStorage *mainRefSto = AliCDBManager::Instance()->GetStorage(fgkMainRefStorage);
1365 if(mainRefSto) mainRefSto->QueryCDB(GetCurrentRun());
1366 first = kFALSE;
1367 }
9e080f92 1368
c88ad5db 1369 Log("SHUTTLE", Form("\t\t\t****** run %d - %s: START ******",
2bb7b766 1370 GetCurrentRun(), aDetector->GetName()));
1371
9d733021 1372 for(Int_t iSys=0;iSys<3;iSys++) fFXSCalled[iSys]=kFALSE;
ef83773f 1373
1374 Int_t initialMem = GetMem(getpid());
1375 Log("SHUTTLE", Form("Memory consumption before forking is %d", initialMem));
9d733021 1376
c88ad5db 1377 Log(fCurrentDetector.Data(), "Process - Starting processing");
85a80aa9 1378
be48e3ea 1379 Int_t pid = fork();
1380
1381 if (pid < 0)
1382 {
c88ad5db 1383 Log("SHUTTLE", "Process - ERROR: Forking failed");
be48e3ea 1384 }
1385 else if (pid > 0)
1386 {
1387 // parent
c88ad5db 1388 Log("SHUTTLE", Form("Process - In parent process of %d - %s: Starting monitoring",
be48e3ea 1389 GetCurrentRun(), aDetector->GetName()));
1390
1391 Long_t begin = time(0);
1392
1393 int status; // to be used with waitpid, on purpose an int (not Int_t)!
1394 while (waitpid(pid, &status, WNOHANG) == 0)
1395 {
1396 Long_t expiredTime = time(0) - begin;
1397
1398 if (expiredTime > fConfig->GetPPTimeOut())
1399 {
56d3e9e4 1400 TString logMsg;
1401 AliShuttleStatus *currentStatus = ReadShuttleStatus();
1402 AliShuttleStatus::Status newStatus = AliShuttleStatus::kInvalid;
1403
1404 if (currentStatus->GetStatus() <= AliShuttleStatus::kPPDone)
1405 {
1406 // in case pp not yet done set status to kPPTimeOut
1407
1408 logMsg.Form("Process - Process of %s timed out. Run time: %d seconds. Killing...",
1409 fCurrentDetector.Data(), expiredTime);
1410 newStatus = AliShuttleStatus::kPPTimeOut;
1411 }
1412 else if (currentStatus->GetStatus() == AliShuttleStatus::kStoreStarted)
1413 {
1414 // in case the pp goes in TimeOut while storing the objects in the OCDB
1415 // set status to kStoreError
1416
1417 logMsg.Form("Process - Process of %s timed out while storing the OCDB object. Run time: %d seconds. Killing... and setting status to StoreError.",
1418 fCurrentDetector.Data(), expiredTime);
1419 newStatus = AliShuttleStatus::kStoreError;
1420 }
1421 else
1422 {
1423 // in other cases don't change the status
1424
1425 logMsg.Form("Process - Process of %s timed out in status = %s. Run time: %d seconds. Killing... without changing the status",
1426 fCurrentDetector.Data(), currentStatus->GetStatusName(), expiredTime);
1427 }
1428
1429 Log("SHUTTLE", logMsg);
1430 Log(fCurrentDetector, logMsg);
be48e3ea 1431
1432 kill(pid, 9);
1433
56d3e9e4 1434 if (newStatus != AliShuttleStatus::kInvalid)
1435 UpdateShuttleStatus(newStatus);
be48e3ea 1436 hasError = kTRUE;
1437
1438 gSystem->Sleep(1000);
1439 }
1440 else
1441 {
be48e3ea 1442 gSystem->Sleep(1000);
9827400b 1443
ef83773f 1444 Int_t mem = GetMem(pid);
1445
1446 if (mem < 0)
9827400b 1447 continue;
9827400b 1448
ef83773f 1449 mem -= initialMem;
1450 if (mem < 0)
1451 mem = 0;
9827400b 1452
1453 if (expiredTime % 60 == 0)
ee6f7523 1454 {
c88ad5db 1455 Log("SHUTTLE", Form("Process - %s: Checking process. "
1456 "Run time: %d seconds - Memory consumption: %d KB",
1457 fCurrentDetector.Data(), expiredTime, mem));
ee6f7523 1458 SendAlive();
1459 }
9827400b 1460
1461 if (mem > fConfig->GetPPMaxMem())
1462 {
1463 TString tmp;
c88ad5db 1464 tmp.Form("Process - Process exceeds maximum allowed memory "
1465 "(%d KB > %d KB). Killing...",
9827400b 1466 mem, fConfig->GetPPMaxMem());
1467 Log("SHUTTLE", tmp);
1468 Log(fCurrentDetector, tmp);
1469
1470 kill(pid, 9);
1471
1472 UpdateShuttleStatus(AliShuttleStatus::kPPOutOfMemory);
1473 hasError = kTRUE;
1474
1475 gSystem->Sleep(1000);
1476 }
be48e3ea 1477 }
1478 }
1479
c88ad5db 1480 Log("SHUTTLE", Form("Process - In parent process of %d - %s: Client has terminated.",
be48e3ea 1481 GetCurrentRun(), aDetector->GetName()));
1482
1483 if (WIFEXITED(status))
1484 {
1485 Int_t returnCode = WEXITSTATUS(status);
1486
c88ad5db 1487 Log("SHUTTLE", Form("Process - %s: the return code is %d", fCurrentDetector.Data(),
3301427a 1488 returnCode));
be48e3ea 1489
9827400b 1490 if (returnCode == 0) hasError = kTRUE;
be48e3ea 1491 }
1492 }
1493 else if (pid == 0)
1494 {
26b1866e 1495 // child
1496 Log("SHUTTLE", Form("Process - In child process of %d - %s", GetCurrentRun(),
c88ad5db 1497 aDetector->GetName()));
be48e3ea 1498
c88ad5db 1499 Log("SHUTTLE", Form("Process - Redirecting output to %s log",fCurrentDetector.Data()));
ffa29e93 1500
546242fb 1501 if ((freopen(GetLogFileName(fCurrentDetector), "a", stdout)) == 0)
ffa29e93 1502 {
c88ad5db 1503 Log("SHUTTLE", "Process - Could not freopen stdout");
ffa29e93 1504 }
1505 else
1506 {
1507 fOutputRedirected = kTRUE;
1508 if ((dup2(fileno(stdout), fileno(stderr))) < 0)
c88ad5db 1509 Log("SHUTTLE", "Process - Could not redirect stderr");
ffa29e93 1510
1511 }
1512
5bac2bde 1513 TString wd = gSystem->WorkingDirectory();
3ebbf3c4 1514 Int_t dir_lev1 = GetCurrentRun()/10000;
1515 TString tmpDir = Form("%s/%d/%d/%s_process", GetShuttleTempDir(),
1516 dir_lev1, GetCurrentRun(), fCurrentDetector.Data());
5bac2bde 1517
d524ade6 1518 Int_t result = gSystem->GetPathInfo(tmpDir.Data(), 0, (Long64_t*) 0, 0, 0);
1519 if (!result) // temp dir already exists!
1520 {
1521 Log(fCurrentDetector.Data(),
1522 Form("Process - %s dir already exists! Removing...", tmpDir.Data()));
1523 gSystem->Exec(Form("rm -rf %s",tmpDir.Data()));
675f64cd 1524 }
1525
1526 if (gSystem->mkdir(tmpDir.Data(), 1))
1527 {
1528 Log(fCurrentDetector.Data(), "Process - could not make temp directory!!");
1529 gSystem->Exit(1);
d524ade6 1530 }
1531
1532 if (!gSystem->ChangeDirectory(tmpDir.Data()))
1533 {
1534 Log(fCurrentDetector.Data(), "Process - could not change directory!!");
1535 gSystem->Exit(1);
1536 }
5bac2bde 1537
ffa25f20 1538 Int_t success = ProcessCurrentDetector();
26b1866e 1539
5bac2bde 1540 gSystem->ChangeDirectory(wd.Data());
b1d18693 1541
ffa25f20 1542 if (success == 1) // Preprocessor finished successfully!
9827400b 1543 {
fb40ead4 1544 // remove temporary folder or DCS map
1545 if (!fConfig->KeepTempFolder())
1546 {
1547 gSystem->Exec(Form("rm -rf %s",tmpDir.Data()));
1548 } else if (!fConfig->KeepDCSMap())
1549 {
1550 gSystem->Exec(Form("rm -f %s/DCSMap.root",tmpDir.Data()));
1551 }
b1d18693 1552
3301427a 1553 // Update time_processed field in FXS DB
1554 if (UpdateTable() == kFALSE)
5bac2bde 1555 Log("SHUTTLE", Form("Process - %s: Could not update FXS databases!",
1556 fCurrentDetector.Data()));
3301427a 1557
1558 // Transfer the data from local storage to main storage (Grid)
3301427a 1559 if (StoreOCDB() == kFALSE)
9827400b 1560 success = kFALSE;
5e993b6f 1561 }
ffa25f20 1562 else if (success == 0)
c88ad5db 1563 {
1564 Log("SHUTTLE",
1565 Form("\t\t\t****** run %d - %s: PP ERROR ******",
1566 GetCurrentRun(), aDetector->GetName()));
be48e3ea 1567 }
1568
4b95672b 1569 for (UInt_t iSys=0; iSys<3; iSys++)
1570 {
1571 if (fFXSCalled[iSys]) fFXSlist[iSys].Clear();
1572 }
1573
c88ad5db 1574 Log("SHUTTLE", Form("Process - Client process of %d - %s is exiting now with %d.",
9827400b 1575 GetCurrentRun(), aDetector->GetName(), success));
be48e3ea 1576
1577 // the client exits here
9827400b 1578 gSystem->Exit(success);
be48e3ea 1579
1580 AliError("We should never get here!!!");
1581 }
7bfb2090 1582 }
5164a766 1583
c88ad5db 1584 Log("SHUTTLE", Form("\t\t\t^*^*^*^*^*^*^*^*^*^*^*^* run %d: FINISH ^*^*^*^*^*^*^*^*^*^*^*^*",
2bb7b766 1585 GetCurrentRun()));
1586
1587 //check if shuttle is done for this run, if so update logbook
1588 TObjArray checkEntryArray;
1589 checkEntryArray.SetOwner(1);
9e080f92 1590 TString whereClause = Form("where run=%d", GetCurrentRun());
b0e53b15 1591 if (!QueryShuttleLogbook(whereClause.Data(), checkEntryArray) ||
1592 checkEntryArray.GetEntries() == 0) {
9e080f92 1593 Log("SHUTTLE", Form("Process - Warning: Cannot check status of run %d on Shuttle logbook!",
1594 GetCurrentRun()));
1595 return hasError == kFALSE;
1596 }
b948db8d 1597
9e080f92 1598 AliShuttleLogbookEntry* checkEntry = dynamic_cast<AliShuttleLogbookEntry*>
1599 (checkEntryArray.At(0));
2bb7b766 1600
9e080f92 1601 if (checkEntry)
1602 {
1603 if (checkEntry->IsDone())
be48e3ea 1604 {
9e080f92 1605 Log("SHUTTLE","Process - Shuttle is DONE. Updating logbook");
1606 UpdateShuttleLogbook("shuttle_done");
1607 }
1608 else
1609 {
1610 for (UInt_t iDet=0; iDet<NDetectors(); iDet++)
be48e3ea 1611 {
9e080f92 1612 if (checkEntry->GetDetectorStatus(iDet) == AliShuttleLogbookEntry::kUnprocessed)
be48e3ea 1613 {
9e080f92 1614 AliDebug(2, Form("Run %d: setting %s as \"not first time unprocessed\"",
1615 checkEntry->GetRun(), GetDetName(iDet)));
1616 fFirstUnprocessed[iDet] = kFALSE;
be48e3ea 1617 }
1618 }
8e828d39 1619 SendMLRunInfo("Pending");
2bb7b766 1620 }
1621 }
1622
1623 fLogbookEntry = 0;
85a80aa9 1624
a7160fe9 1625 return hasError == kFALSE;
73abe331 1626}
1627
b948db8d 1628//______________________________________________________________________________________________
ffa25f20 1629Int_t AliShuttle::ProcessCurrentDetector()
73abe331 1630{
1631 //
2bb7b766 1632 // Makes data retrieval just for a specific detector (fCurrentDetector).
73abe331 1633 // Threre should be a configuration for this detector.
73abe331 1634
1d172743 1635 Log("SHUTTLE", Form("ProcessCurrentDetector - Retrieving values for %s, run %d",
1636 fCurrentDetector.Data(), GetCurrentRun()));
73abe331 1637
d524ade6 1638 TString wd = gSystem->WorkingDirectory();
1639
2d9019b4 1640 if (!CleanReferenceStorage(fCurrentDetector.Data()))
ffa25f20 1641 return 0;
d524ade6 1642
1643 gSystem->ChangeDirectory(wd.Data());
1644
3301427a 1645 // call preprocessor
1646 AliPreprocessor* aPreprocessor =
1647 dynamic_cast<AliPreprocessor*> (fPreprocessorMap.GetValue(fCurrentDetector));
1648
8579c6fb 1649 // check if the preprocessor wants to process this run type
1650 if (aPreprocessor->ProcessRunType() == kFALSE)
ffa25f20 1651 {
1652 UpdateShuttleStatus(AliShuttleStatus::kSkipped);
1653 UpdateShuttleLogbook(fCurrentDetector, "DONE");
bd23d48b 1654 if (!UpdateTableSkippedCase(fCurrentDetector.Data()))
1655 {
1656 AliError(Form("Could not update FXS tables for run %d !", GetCurrentRun()));
1657 }
ffa25f20 1658 Log(fCurrentDetector, Form("ProcessCurrentDetector - %s preprocessor is not interested in this run type", fCurrentDetector.Data()));
1659
1660 return 2;
1661 }
1662
1663 TMap* dcsMap = new TMap();
1664
3301427a 1665 aPreprocessor->Initialize(GetCurrentRun(), GetCurrentStartTime(), GetCurrentEndTime());
1666
1667 Bool_t processDCS = aPreprocessor->ProcessDCS();
d477ad88 1668
651fdaab 1669 if (!processDCS)
1670 {
1d172743 1671 Log(fCurrentDetector, "ProcessCurrentDetector -"
1672 " The preprocessor requested to skip the retrieval of DCS values");
651fdaab 1673 }
8b739301 1674 else if (fTestMode & kSkipDCS)
2c15234c 1675 {
1d172743 1676 Log(fCurrentDetector, "ProcessCurrentDetector - In TESTMODE: Skipping DCS processing");
9827400b 1677 }
1678 else if (fTestMode & kErrorDCS)
1679 {
1d172743 1680 Log(fCurrentDetector, "ProcessCurrentDetector - In TESTMODE: Simulating DCS error");
3d8bc902 1681 UpdateShuttleStatus(AliShuttleStatus::kDCSStarted);
9827400b 1682 UpdateShuttleStatus(AliShuttleStatus::kDCSError);
1d172743 1683 delete dcsMap;
ffa25f20 1684 return 0;
2c15234c 1685 } else {
3301427a 1686
1687 UpdateShuttleStatus(AliShuttleStatus::kDCSStarted);
1688
1d172743 1689 // Query DCS archive
1690 Int_t nServers = fConfig->GetNServers(fCurrentDetector);
a038aa70 1691
1d172743 1692 for (int iServ=0; iServ<nServers; iServ++)
2c15234c 1693 {
1d172743 1694
1695 TString host(fConfig->GetDCSHost(fCurrentDetector, iServ));
1696 Int_t port = fConfig->GetDCSPort(fCurrentDetector, iServ);
542b6cc8 1697 Int_t multiSplit = fConfig->GetMultiSplit(fCurrentDetector, iServ);
1698
1790d4b7 1699 Log(fCurrentDetector, Form("ProcessCurrentDetector -"
1700 " Querying DCS Amanda server %s:%d (%d of %d)",
1701 host.Data(), port, iServ+1, nServers));
1d172743 1702
1703 TMap* aliasMap = 0;
1704 TMap* dpMap = 0;
26b1866e 1705
1d172743 1706 if (fConfig->GetDCSAliases(fCurrentDetector, iServ)->GetEntries() > 0)
2c15234c 1707 {
5ed060b1 1708 Log(fCurrentDetector, Form("Querying %d DCS aliases", fConfig->GetDCSAliases(fCurrentDetector, iServ)->GetEntries()));
1d172743 1709 aliasMap = GetValueSet(host, port,
542b6cc8 1710 fConfig->GetDCSAliases(fCurrentDetector, iServ),
1711 kAlias, multiSplit);
1d172743 1712 if (!aliasMap)
1713 {
1714 Log(fCurrentDetector,
1715 Form("ProcessCurrentDetector -"
675f64cd 1716 " Error retrieving DCS aliases from server %s."
1717 " Sending mail to DCS experts!", host.Data()));
1d172743 1718 UpdateShuttleStatus(AliShuttleStatus::kDCSError);
675f64cd 1719
339adafa 1720 if (!SendMail(kDCSEMail))
fb40ead4 1721 Log("SHUTTLE", Form("ProcessCurrentDetector - "
95d4aaef 1722 "Could not send mail to DCS experts!"));
675f64cd 1723
a038aa70 1724 delete dcsMap;
ffa25f20 1725 return 0;
1d172743 1726 }
2c15234c 1727 }
a038aa70 1728
1d172743 1729 if (fConfig->GetDCSDataPoints(fCurrentDetector, iServ)->GetEntries() > 0)
a038aa70 1730 {
5ed060b1 1731 Log(fCurrentDetector, Form("Querying %d DCS data points", fConfig->GetDCSDataPoints(fCurrentDetector, iServ)->GetEntries()));
1d172743 1732 dpMap = GetValueSet(host, port,
542b6cc8 1733 fConfig->GetDCSDataPoints(fCurrentDetector, iServ),
1734 kDP, multiSplit);
1d172743 1735 if (!dpMap)
1736 {
1737 Log(fCurrentDetector,
1738 Form("ProcessCurrentDetector -"
675f64cd 1739 " Error retrieving DCS data points from server %s."
1740 " Sending mail to DCS experts!", host.Data()));
1d172743 1741 UpdateShuttleStatus(AliShuttleStatus::kDCSError);
675f64cd 1742
339adafa 1743 if (!SendMail(kDCSEMail))
fb40ead4 1744 Log("SHUTTLE", Form("ProcessCurrentDetector - "
95d4aaef 1745 "Could not send mail to DCS experts!"));
675f64cd 1746
1d172743 1747 if (aliasMap) delete aliasMap;
1748 delete dcsMap;
ffa25f20 1749 return 0;
1d172743 1750 }
a038aa70 1751 }
1d172743 1752
1753 // merge aliasMap and dpMap into dcsMap
1754 if(aliasMap) {
1755 TIter iter(aliasMap);
a038aa70 1756 TObjString* key = 0;
1757 while ((key = (TObjString*) iter.Next()))
1d172743 1758 dcsMap->Add(key, aliasMap->GetValue(key->String()));
1759
1760 aliasMap->SetOwner(kFALSE);
1761 delete aliasMap;
1762 }
1763
1764 if(dpMap) {
1765 TIter iter(dpMap);
1766 TObjString* key = 0;
1767 while ((key = (TObjString*) iter.Next()))
1768 dcsMap->Add(key, dpMap->GetValue(key->String()));
1769
1770 dpMap->SetOwner(kFALSE);
1771 delete dpMap;
a038aa70 1772 }
73abe331 1773 }
1774 }
dc25836b 1775
b1d18693 1776 // save map into file, to help debugging in case of preprocessor error
fbc112e3 1777 TFile* f = TFile::Open("DCSMap.root","recreate");
b1d18693 1778 f->cd();
1779 dcsMap->Write("DCSMap", TObject::kSingleKey);
1780 f->Close();
fbc112e3 1781 delete f;
b1d18693 1782
2bb7b766 1783 // DCS Archive DB processing successful. Call Preprocessor!
85a80aa9 1784 UpdateShuttleStatus(AliShuttleStatus::kPPStarted);
a7160fe9 1785
339adafa 1786 fFXSError = -1; // this variable is kTRUE after ::Process if an FXS error occured
8da81210 1787
a038aa70 1788 UInt_t returnValue = aPreprocessor->Process(dcsMap);
8da81210 1789
339adafa 1790 if (fFXSError!=-1) {
8da81210 1791 UpdateShuttleStatus(AliShuttleStatus::kFXSError);
339adafa 1792 SendMail(kFXSEMail, fFXSError);
8da81210 1793 dcsMap->DeleteAll();
1794 delete dcsMap;
ffa25f20 1795 return 0;
8da81210 1796 }
b948db8d 1797
3301427a 1798 if (returnValue > 0) // Preprocessor error!
1799 {
c88ad5db 1800 Log(fCurrentDetector, Form("ProcessCurrentDetector - "
1801 "Preprocessor failed. Process returned %d.", returnValue));
cb343cfd 1802 UpdateShuttleStatus(AliShuttleStatus::kPPError);
a038aa70 1803 dcsMap->DeleteAll();
1804 delete dcsMap;
ffa25f20 1805 return 0;
9827400b 1806 }
1807
1808 // preprocessor ok!
1809 UpdateShuttleStatus(AliShuttleStatus::kPPDone);
1810 Log(fCurrentDetector, Form("ProcessCurrentDetector - %s preprocessor returned success",
1811 fCurrentDetector.Data()));
b948db8d 1812
a038aa70 1813 dcsMap->DeleteAll();
1814 delete dcsMap;
b948db8d 1815
ffa25f20 1816 return 1;
2bb7b766 1817}
1818
fbc112e3 1819//______________________________________________________________________________________________
1820void AliShuttle::CountOpenRuns()
1821{
1822 // Query DAQ's Shuttle logbook and sends the number of open runs to ML
1823
982278d6 1824 SendAlive();
1825
fbc112e3 1826 // check connection, in case connect
1827 if (!Connect(3))
1828 return;
1829
1830 TString sqlQuery;
1831 sqlQuery = Form("select count(*) from %s where shuttle_done=0", fConfig->GetShuttlelbTable());
1832
1833 TSQLResult* aResult = fServer[3]->Query(sqlQuery);
1834 if (!aResult) {
1835 AliError(Form("Can't execute query <%s>!", sqlQuery.Data()));
1836 return;
1837 }
1838
1839 AliDebug(2,Form("Query = %s", sqlQuery.Data()));
1840
1841 if (aResult->GetRowCount() == 0) {
1842 AliError(Form("No result for query %s received", sqlQuery.Data()));
1843 return;
1844 }
1845
1846 if (aResult->GetFieldCount() != 1) {
1847 AliError(Form("Invalid field count for query %s received", sqlQuery.Data()));
1848 return;
1849 }
1850
1851 TSQLRow* aRow = aResult->Next();
1852 if (!aRow) {
1853 AliError(Form("Could not receive result of query %s", sqlQuery.Data()));
1854 return;
1855 }
1856
1857 TString result(aRow->GetField(0), aRow->GetFieldLength(0));
1858 Int_t count = result.Atoi();
1859
1860 Log("SHUTTLE", Form("%d unprocessed runs", count));
1861
1862 delete aRow;
1863 delete aResult;
1864
1865 TMonaLisaValue mlStatus("SHUTTLE_openruns", count);
1866
1867 TList mlList;
1868 mlList.Add(&mlStatus);
1869
1870 fMonaLisa->SendParameters(&mlList, "__PROCESSINGINFO__");
1871}
1872
2bb7b766 1873//______________________________________________________________________________________________
1874Bool_t AliShuttle::QueryShuttleLogbook(const char* whereClause,
1875 TObjArray& entries)
1876{
9827400b 1877 // Query DAQ's Shuttle logbook and fills detector status object.
1878 // Call QueryRunParameters to query DAQ logbook for run parameters.
1879 //
2bb7b766 1880
fc5a4708 1881 entries.SetOwner(1);
1882
2bb7b766 1883 // check connection, in case connect
fbc112e3 1884 if (!Connect(3)) return kFALSE;
2bb7b766 1885
1886 TString sqlQuery;
441b0e9c 1887 sqlQuery = Form("select * from %s %s order by run", fConfig->GetShuttlelbTable(), whereClause);
2bb7b766 1888
be48e3ea 1889 TSQLResult* aResult = fServer[3]->Query(sqlQuery);
2bb7b766 1890 if (!aResult) {
1891 AliError(Form("Can't execute query <%s>!", sqlQuery.Data()));
1892 return kFALSE;
1893 }
1894
fc5a4708 1895 AliDebug(2,Form("Query = %s", sqlQuery.Data()));
1896
2bb7b766 1897 if(aResult->GetRowCount() == 0) {
c88ad5db 1898 Log("SHUTTLE", "No entries in Shuttle Logbook match request");
9827400b 1899 delete aResult;
1900 return kTRUE;
2bb7b766 1901 }
1902
1903 // TODO Check field count!
171c6702 1904 const UInt_t nCols = 25;
2bb7b766 1905 if (aResult->GetFieldCount() != (Int_t) nCols) {
c88ad5db 1906 Log("SHUTTLE", "Invalid SQL result field number!");
2bb7b766 1907 delete aResult;
1908 return kFALSE;
1909 }
1910
2bb7b766 1911 TSQLRow* aRow;
1912 while ((aRow = aResult->Next())) {
1913 TString runString(aRow->GetField(0), aRow->GetFieldLength(0));
1914 Int_t run = runString.Atoi();
1915
eba76848 1916 AliShuttleLogbookEntry *entry = QueryRunParameters(run);
1917 if (!entry)
1918 continue;
2bb7b766 1919
728290a4 1920 // DA test mode flag
1921 TString daTestModeString(aRow->GetField(2), aRow->GetFieldLength(2)); // field 2 = DA test mode flag
1922 Bool_t daTestMode = (Bool_t)daTestModeString.Atoi();
1923 entry->SetDATestMode(daTestMode);
1924
2bb7b766 1925 // loop on detectors
eba76848 1926 for(UInt_t ii = 0; ii < nCols; ii++)
1927 entry->SetDetectorStatus(aResult->GetFieldName(ii), aRow->GetField(ii));
2bb7b766 1928
eba76848 1929 entries.AddLast(entry);
2bb7b766 1930 delete aRow;
1931 }
1932
2bb7b766 1933 delete aResult;
1934 return kTRUE;
1935}
1936
1937//______________________________________________________________________________________________
eba76848 1938AliShuttleLogbookEntry* AliShuttle::QueryRunParameters(Int_t run)
2bb7b766 1939{
eba76848 1940 //
1941 // Retrieve run parameters written in the DAQ logbook and sets them into AliShuttleLogbookEntry object
1942 //
2bb7b766 1943
1944 // check connection, in case connect
be48e3ea 1945 if (!Connect(3))
eba76848 1946 return 0;
2bb7b766 1947
1948 TString sqlQuery;
2c15234c 1949 sqlQuery.Form("select * from %s where run=%d", fConfig->GetDAQlbTable(), run);
2bb7b766 1950
be48e3ea 1951 TSQLResult* aResult = fServer[3]->Query(sqlQuery);
2bb7b766 1952 if (!aResult) {
c88ad5db 1953 Log("SHUTTLE", Form("Can't execute query <%s>!", sqlQuery.Data()));
eba76848 1954 return 0;
2bb7b766 1955 }
1956
eba76848 1957 if (aResult->GetRowCount() == 0) {
2bb7b766 1958 Log("SHUTTLE", Form("QueryRunParameters - No entry in DAQ Logbook for run %d. Skipping", run));
1959 delete aResult;
eba76848 1960 return 0;
2bb7b766 1961 }
1962
eba76848 1963 if (aResult->GetRowCount() > 1) {
c88ad5db 1964 Log("SHUTTLE", Form("QueryRunParameters - UNEXPECTED: "
1965 "more than one entry in DAQ Logbook for run %d!", run));
2bb7b766 1966 delete aResult;
eba76848 1967 return 0;
2bb7b766 1968 }
1969
eba76848 1970 TSQLRow* aRow = aResult->Next();
1971 if (!aRow)
1972 {
c88ad5db 1973 Log("SHUTTLE", Form("QueryRunParameters - Could not retrieve row for run %d. Skipping", run));
eba76848 1974 delete aResult;
1975 return 0;
1976 }
2bb7b766 1977
eba76848 1978 AliShuttleLogbookEntry* entry = new AliShuttleLogbookEntry(run);
2bb7b766 1979
eba76848 1980 for (Int_t ii = 0; ii < aResult->GetFieldCount(); ii++)
1981 entry->SetRunParameter(aResult->GetFieldName(ii), aRow->GetField(ii));
2bb7b766 1982
c9db68c7 1983 delete aRow;
1984 delete aResult;
1985
eba76848 1986 UInt_t startTime = entry->GetStartTime();
1987 UInt_t endTime = entry->GetEndTime();
89bcf3b4 1988 Bool_t ecsSuccess = entry->GetECSSuccess();
c9db68c7 1989
b0e53b15 1990 TString totEventsStr = entry->GetRunParameter("totalEvents");
1991 Int_t totEvents = totEventsStr.Atoi();
c9db68c7 1992
af93e5d7 1993 UInt_t now = time(0);
1994 // TODO make this a configuration parameter
2eb98462 1995 Int_t dcsDelay = fConfig->GetDCSDelay()+fConfig->GetDCSQueryOffset();
af93e5d7 1996
8e828d39 1997 // runs are accepted if they have ecsSuccess set or more than 1 event
af93e5d7 1998 if (startTime != 0 && endTime != 0 && endTime > startTime && (totEvents > 1 || ecsSuccess) && (endTime < now - dcsDelay))
89bcf3b4 1999 {
8e828d39 2000 if (ecsSuccess == kFALSE)
2001 Log("SHUTTLE", Form("Processing run %d although in status ECS failure, Reason: %s", run, entry->GetRunParameter("eor_reason")));
2002 return entry;
89bcf3b4 2003 }
8e828d39 2004
2005 Bool_t skip = kFALSE;
2006
f8523731 2007 if (endTime != 0 && endTime >= now - dcsDelay)
89bcf3b4 2008 {
f8523731 2009 Log("SHUTTLE", Form("Skipping run %d for now, because DCS buffer time is not yet expired", run));
af93e5d7 2010 }
f8523731 2011 else if (totEvents <= 1)
af93e5d7 2012 {
f8523731 2013 Log("SHUTTLE", Form("QueryRunParameters - Run %d has 1 event or less - Skipping!", run));
2014 skip = kTRUE;
af93e5d7 2015 }
89bcf3b4 2016 else
2017 {
2018 Log("SHUTTLE", Form("QueryRunParameters - Invalid parameters for Run %d: "
2019 "startTime = %d, endTime = %d. Skipping (Shuttle won't be marked as DONE)!",
2020 run, startTime, endTime));
2021 }
8e828d39 2022
2023 if (skip)
2024 {
2025 Log("SHUTTLE", Form("Marking SHUTTLE skipped for run %d", run));
2026 fLogbookEntry = entry;
2027 if (!UpdateShuttleLogbook("shuttle_skipped"))
2028 {
2029 AliError(Form("Could not update logbook for run %d !", run));
2030 }
bd23d48b 2031 if (!UpdateTableSkippedCase("ALL"))
2032 {
2033 AliError(Form("Could not update FXS tables for run %d !", run));
2034 }
8e828d39 2035 fLogbookEntry = 0;
2036 }
c9db68c7 2037
2038 delete entry;
2039 return 0;
2bb7b766 2040}
2041
a038aa70 2042//______________________________________________________________________________________________
2043TMap* AliShuttle::GetValueSet(const char* host, Int_t port, const TSeqCollection* entries,
542b6cc8 2044 DCSType type, Int_t multiSplit)
a038aa70 2045{
2046 // Retrieve all "entry" data points from the DCS server
2047 // host, port: TSocket connection parameters
2048 // entries: list of name of the alias or data point
2049 // type: kAlias or kDP
2050 // returns TMap of values, 0 when failure
542b6cc8 2051
2052 AliDCSClient client(host, port, fTimeout, fRetries, multiSplit);
b41b252a 2053
a038aa70 2054 TMap* result = 0;
b41b252a 2055 if (type == kAlias)
a038aa70 2056 {
5f3fe74e 2057 //result = client.GetAliasValues(entries, GetCurrentStartTime()-offset,
2058 // GetCurrentEndTime()+offset);
eff647d5 2059 result = client.GetAliasValues(entries, GetStartTimeDCSQuery(),
2060 GetEndTimeDCSQuery());
b41b252a 2061 }
2062 else if (type == kDP)
2063 {
5f3fe74e 2064 //result = client.GetDPValues(entries, GetCurrentStartTime()-offset,
2065 // GetCurrentEndTime()+offset);
eff647d5 2066 result = client.GetDPValues(entries, GetStartTimeDCSQuery(),
2067 GetEndTimeDCSQuery());
b41b252a 2068 }
a038aa70 2069
b41b252a 2070 if (result == 0)
2071 {
2072 Log(fCurrentDetector.Data(), Form("GetValueSet - Can't get entries! Reason: %s",
1790d4b7 2073 client.GetErrorString(client.GetResultErrorCode())));
2074 if (client.GetResultErrorCode() == AliDCSClient::fgkServerError)
2075 Log(fCurrentDetector.Data(), Form("GetValueSet - Server error code: %s",
2076 client.GetServerError().Data()));
a038aa70 2077
b41b252a 2078 return 0;
a038aa70 2079 }
b41b252a 2080
a038aa70 2081 return result;
2082}
b41b252a 2083
b948db8d 2084//______________________________________________________________________________________________
57f50b3c 2085const char* AliShuttle::GetFile(Int_t system, const char* detector,
2086 const char* id, const char* source)
b948db8d 2087{
9827400b 2088 // Get calibration file from file exchange servers
2089 // First queris the FXS database for the file name, using the run, detector, id and source info
2090 // then calls RetrieveFile(filename) for actual copy to local disk
2091 // run: current run being processed (given by Logbook entry fLogbookEntry)
2092 // detector: the Preprocessor name
2093 // id: provided as a parameter by the Preprocessor
2094 // source: provided by the Preprocessor through GetFileSources function
2095
2096 // check if test mode should simulate a FXS error
2097 if (fTestMode & kErrorFXSFiles)
2098 {
2099 Log(detector, Form("GetFile - In TESTMODE - Simulating error while connecting to %s FXS", GetSystemName(system)));
2100 return 0;
2101 }
2102
57f50b3c 2103 // check connection, in case connect
9d733021 2104 if (!Connect(system))
eba76848 2105 {
9d733021 2106 Log(detector, Form("GetFile - Couldn't connect to %s FXS database", GetSystemName(system)));
339adafa 2107 fFXSError = system;
57f50b3c 2108 return 0;
2109 }
2110
2111 // Query preparation
9d733021 2112 TString sourceName(source);
d386d623 2113 Int_t nFields = 3;
2114 TString sqlQueryStart = Form("select filePath,size,fileChecksum from %s where",
2115 fConfig->GetFXSdbTable(system));
2116 TString whereClause = Form("run=%d and detector=\"%s\" and fileId=\"%s\"",
2117 GetCurrentRun(), detector, id);
2118
9d733021 2119 if (system == kDAQ)
2120 {
d386d623 2121 whereClause += Form(" and DAQsource=\"%s\"", source);
57f50b3c 2122 }
9d733021 2123 else if (system == kDCS)
eba76848 2124 {
9d733021 2125 sourceName="none";
57f50b3c 2126 }
9d733021 2127 else if (system == kHLT)
9e080f92 2128 {
d386d623 2129 whereClause += Form(" and DDLnumbers=\"%s\"", source);
9e080f92 2130 }
2131
9e080f92 2132 TString sqlQuery = Form("%s %s", sqlQueryStart.Data(), whereClause.Data());
2133
2134 AliDebug(2, Form("SQL query: \n%s",sqlQuery.Data()));
2135
2136 // Query execution
2137 TSQLResult* aResult = 0;
9d733021 2138 aResult = dynamic_cast<TSQLResult*> (fServer[system]->Query(sqlQuery));
9e080f92 2139 if (!aResult) {
8da81210 2140 Log(detector, Form("GetFile - Can't execute SQL query to %s database for: id = %s, source = %s",
9d733021 2141 GetSystemName(system), id, sourceName.Data()));
339adafa 2142 fFXSError = system;
9e080f92 2143 return 0;
2144 }
2145
8da81210 2146 if (aResult->GetRowCount() == 0)
9e080f92 2147 {
2148 Log(detector,
8da81210 2149 Form("GetFile - No entry in %s FXS db for: id = %s, source = %s",
9d733021 2150 GetSystemName(system), id, sourceName.Data()));
9e080f92 2151 delete aResult;
2152 return 0;
2153 }
2bb7b766 2154
9e080f92 2155 if (aResult->GetRowCount() > 1) {
2156 Log(detector,
8da81210 2157 Form("GetFile - More than one entry in %s FXS db for: id = %s, source = %s",
9d733021 2158 GetSystemName(system), id, sourceName.Data()));
339adafa 2159 fFXSError = system;
9e080f92 2160 delete aResult;
2161 return 0;
2162 }
2163
9d733021 2164 if (aResult->GetFieldCount() != nFields) {
9e080f92 2165 Log(detector,
9d733021 2166 Form("GetFileName - Wrong field count in %s FXS db for: id = %s, source = %s",
2167 GetSystemName(system), id, sourceName.Data()));
339adafa 2168 fFXSError = system;
9e080f92 2169 delete aResult;
2170 return 0;
2171 }
2172
2173 TSQLRow* aRow = dynamic_cast<TSQLRow*> (aResult->Next());
2174
2175 if (!aRow){
8da81210 2176 Log(detector, Form("GetFile - Empty set result in %s FXS db from query: id = %s, source = %s",
9d733021 2177 GetSystemName(system), id, sourceName.Data()));
339adafa 2178 fFXSError = system;
9e080f92 2179 delete aResult;
2180 return 0;
2181 }
2182
2183 TString filePath(aRow->GetField(0), aRow->GetFieldLength(0));
2184 TString fileSize(aRow->GetField(1), aRow->GetFieldLength(1));
d386d623 2185 TString fileChecksum(aRow->GetField(2), aRow->GetFieldLength(2));
9e080f92 2186
2187 delete aResult;
2188 delete aRow;
2189
d386d623 2190 AliDebug(2, Form("filePath = %s; size = %s, fileChecksum = %s",
2191 filePath.Data(), fileSize.Data(), fileChecksum.Data()));
9e080f92 2192
9e080f92 2193 // retrieved file is renamed to make it unique
3ebbf3c4 2194 Int_t dir_lev1 = GetCurrentRun()/10000;
2195 TString localFileName = Form("%s/%d/%d/%s_process/%s_%s_%d_%s_%s.shuttle",
2196 GetShuttleTempDir(), dir_lev1, GetCurrentRun(), detector,
d524ade6 2197 GetSystemName(system), detector, GetCurrentRun(),
2198 id, sourceName.Data());
3ebbf3c4 2199 Log("SHUTTLE",Form("file from FXS = %s",localFileName.Data()));
9d733021 2200
9e080f92 2201
9d733021 2202 // file retrieval from FXS
4b95672b 2203 UInt_t nRetries = 0;
2204 UInt_t maxRetries = 3;
2205 Bool_t result = kFALSE;
2206
2207 // copy!! if successful TSystem::Exec returns 0
8da81210 2208 while (nRetries++ < maxRetries) {
4b95672b 2209 AliDebug(2, Form("Trying to copy file. Retry # %d", nRetries));
2210 result = RetrieveFile(system, filePath.Data(), localFileName.Data());
8da81210 2211 if (!result)
4b95672b 2212 {
8da81210 2213 Log(detector, Form("GetFile - Copy of file %s from %s FXS failed",
9d733021 2214 filePath.Data(), GetSystemName(system)));
4b95672b 2215 continue;
4f0749a8 2216 }
9e080f92 2217
c98417a1 2218 if (fileSize.Length()>0)
2219 {
2220 // compare filesize of local file with the one stored in the FXS DB
926cbc0e 2221 Long_t size = -1;
2222 Int_t sizeComp = gSystem->GetPathInfo(localFileName.Data(), 0, &size, 0, 0);
c98417a1 2223
926cbc0e 2224 if (sizeComp != 0 || size != fileSize.Atoi())
c98417a1 2225 {
8da81210 2226 Log(detector, Form("GetFile - size of file %s does not match with local copy!",
c98417a1 2227 filePath.Data()));
2228 result = kFALSE;
2229 continue;
2230 }
2231
2232 } else {
2233 Log(fCurrentDetector, Form("GetFile - size of file %s not set in %s database, skipping comparison",
2234 filePath.Data(), GetSystemName(system)));
2235 }
2236
d386d623 2237 if (fileChecksum.Length()>0)
4b95672b 2238 {
2239 // compare md5sum of local file with the one stored in the FXS DB
95d4aaef 2240 if(fileChecksum.Contains(' ')) fileChecksum.Resize(fileChecksum.First(' '));
339adafa 2241 Int_t md5Comp = gSystem->Exec(Form("md5sum %s |grep %s > /dev/null 2> /dev/null",
d524ade6 2242 localFileName.Data(), fileChecksum.Data()));
9e080f92 2243
4b95672b 2244 if (md5Comp != 0)
2245 {
8da81210 2246 Log(detector, Form("GetFile - md5sum of file %s does not match with local copy!",
4b95672b 2247 filePath.Data()));
2248 result = kFALSE;
2249 continue;
2250 }
d386d623 2251 } else {
2252 Log(fCurrentDetector, Form("GetFile - md5sum of file %s not set in %s database, skipping comparison",
2253 filePath.Data(), GetSystemName(system)));
9d733021 2254 }
4b95672b 2255 if (result) break;
9e080f92 2256 }
2257
8da81210 2258 if (!result)
2259 {
339adafa 2260 fFXSError = system;
8da81210 2261 return 0;
2262 }
4b95672b 2263
9d733021 2264 fFXSCalled[system]=kTRUE;
2265 TObjString *fileParams = new TObjString(Form("%s#!?!#%s", id, sourceName.Data()));
2266 fFXSlist[system].Add(fileParams);
9e080f92 2267
675f64cd 2268 static TString staticLocalFileName;
2269 staticLocalFileName.Form("%s", localFileName.Data());
2270
c88ad5db 2271 Log(fCurrentDetector, Form("GetFile - Retrieved file with id %s and "
2272 "source %s from %s to %s", id, source,
d524ade6 2273 GetSystemName(system), localFileName.Data()));
675f64cd 2274
d524ade6 2275 return staticLocalFileName.Data();
2bb7b766 2276}
2277
2278//______________________________________________________________________________________________
9d733021 2279Bool_t AliShuttle::RetrieveFile(UInt_t system, const char* fxsFileName, const char* localFileName)
9e080f92 2280{
9827400b 2281 //
2282 // Copies file from FXS to local Shuttle machine
2283 //
2bb7b766 2284
9e080f92 2285 // check temp directory: trying to cd to temp; if it does not exist, create it
d524ade6 2286 AliDebug(2, Form("Copy file %s from %s FXS into %s",
2287 GetSystemName(system), fxsFileName, localFileName));
2288
2289 TString tmpDir(localFileName);
2290
2291 tmpDir = tmpDir(0,tmpDir.Last('/'));
9e080f92 2292
d524ade6 2293 Int_t noDir = gSystem->GetPathInfo(tmpDir.Data(), 0, (Long64_t*) 0, 0, 0);
2294 if (noDir) // temp dir does not exists!
2295 {
2296 if (gSystem->mkdir(tmpDir.Data(), 1))
2297 {
2298 Log(fCurrentDetector.Data(), "RetrieveFile - could not make temp directory!!");
9e080f92 2299 return kFALSE;
2300 }
9e080f92 2301 }
2302
339adafa 2303 TString command = Form("scp -oPort=%d -2 %s@%s:%s/%s %s",
9d733021 2304 fConfig->GetFXSPort(system),
2305 fConfig->GetFXSUser(system),
2306 fConfig->GetFXSHost(system),
339adafa 2307 fConfig->GetFXSBaseFolder(system),
9d733021 2308 fxsFileName,
9e080f92 2309 localFileName);
2310
2311 AliDebug(2, Form("%s",command.Data()));
2312
4b95672b 2313 Bool_t result = (gSystem->Exec(command.Data()) == 0);
9e080f92 2314
4b95672b 2315 return result;
9e080f92 2316}
2317
2318//______________________________________________________________________________________________
9d733021 2319TList* AliShuttle::GetFileSources(Int_t system, const char* detector, const char* id)
2320{
9827400b 2321 //
2322 // Get sources producing the condition file Id from file exchange servers
4a33bdd9 2323 // if id is NULL all sources are returned (distinct)
9827400b 2324 //
1bcd28db 2325
8884d71e 2326 if (id)
2327 {
2328 Log(detector, Form("GetFileSources - Querying %s FXS for files with id %s produced by %s", GetSystemName(system), id, detector));
2329 } else {
2330 Log(detector, Form("GetFileSources - Querying %s FXS for files produced by %s", GetSystemName(system), detector));
2331 }
9827400b 2332
2333 // check if test mode should simulate a FXS error
2334 if (fTestMode & kErrorFXSSources)
2335 {
2336 Log(detector, Form("GetFileSources - In TESTMODE - Simulating error while connecting to %s FXS", GetSystemName(system)));
2337 return 0;
2338 }
2339
9d733021 2340 if (system == kDCS)
2341 {
c88ad5db 2342 Log(detector, "GetFileSources - WARNING: DCS system has only one source of data!");
6297b37d 2343 TList *list = new TList();
2344 list->SetOwner(1);
2345 list->Add(new TObjString(" "));
2346 return list;
9d733021 2347 }
9e080f92 2348
2349 // check connection, in case connect
9d733021 2350 if (!Connect(system))
2351 {
4a33bdd9 2352 Log(detector, Form("GetFileSources - Couldn't connect to %s FXS database", GetSystemName(system)));
339adafa 2353 fFXSError = system;
9d733021 2354 return NULL;
9e080f92 2355 }
2356
9d733021 2357 TString sourceName = 0;
2358 if (system == kDAQ)
2359 {
2360 sourceName = "DAQsource";
2361 } else if (system == kHLT)
2362 {
2363 sourceName = "DDLnumbers";
2364 }
2365
4a33bdd9 2366 TString sqlQueryStart = Form("select distinct %s from %s where", sourceName.Data(), fConfig->GetFXSdbTable(system));
2367 TString whereClause = Form("run=%d and detector=\"%s\"",
2368 GetCurrentRun(), detector);
2369 if (id)
2370 whereClause += Form(" and fileId=\"%s\"", id);
9e080f92 2371 TString sqlQuery = Form("%s %s", sqlQueryStart.Data(), whereClause.Data());
2372
2373 AliDebug(2, Form("SQL query: \n%s",sqlQuery.Data()));
2374
2375 // Query execution
2376 TSQLResult* aResult;
9d733021 2377 aResult = fServer[system]->Query(sqlQuery);
9e080f92 2378 if (!aResult) {
9d733021 2379 Log(detector, Form("GetFileSources - Can't execute SQL query to %s database for id: %s",
2380 GetSystemName(system), id));
339adafa 2381 fFXSError = system;
9e080f92 2382 return 0;
2383 }
2384
86aa42c3 2385 TList *list = new TList();
2386 list->SetOwner(1);
2387
9d733021 2388 if (aResult->GetRowCount() == 0)
2389 {
9e080f92 2390 Log(detector,
9d733021 2391 Form("GetFileSources - No entry in %s FXS table for id: %s", GetSystemName(system), id));
9e080f92 2392 delete aResult;
86aa42c3 2393 return list;
9e080f92 2394 }
2395
1bcd28db 2396 Log(detector, Form("GetFileSources - Found %d sources", aResult->GetRowCount()));
9e080f92 2397
1bcd28db 2398 TSQLRow* aRow;
9d733021 2399 while ((aRow = aResult->Next()))
2400 {
9e080f92 2401
9d733021 2402 TString source(aRow->GetField(0), aRow->GetFieldLength(0));
2403 AliDebug(2, Form("%s = %s", sourceName.Data(), source.Data()));
2404 list->Add(new TObjString(source));
9e080f92 2405 delete aRow;
2406 }
9d733021 2407
9e080f92 2408 delete aResult;
2409
2410 return list;
2bb7b766 2411}
2412
4a33bdd9 2413//______________________________________________________________________________________________
2414TList* AliShuttle::GetFileIDs(Int_t system, const char* detector, const char* source)
2415{
2416 //
2417 // Get all ids of condition files produced by a given source from file exchange servers
2418 //
2419
1bcd28db 2420 Log(detector, Form("GetFileIDs - Retrieving ids with source %s with %s", source, GetSystemName(system)));
2421
4a33bdd9 2422 // check if test mode should simulate a FXS error
2423 if (fTestMode & kErrorFXSSources)
2424 {
2425 Log(detector, Form("GetFileIDs - In TESTMODE - Simulating error while connecting to %s FXS", GetSystemName(system)));
2426 return 0;
2427 }
2428
2429 // check connection, in case connect
2430 if (!Connect(system))
2431 {
2432 Log(detector, Form("GetFileIDs - Couldn't connect to %s FXS database", GetSystemName(system)));
2433 return NULL;
2434 }
2435
2436 TString sourceName = 0;
2437 if (system == kDAQ)
2438 {
2439 sourceName = "DAQsource";
2440 } else if (system == kHLT)
2441 {
2442 sourceName = "DDLnumbers";
2443 }
2444
2445 TString sqlQueryStart = Form("select fileId from %s where", fConfig->GetFXSdbTable(system));
2446 TString whereClause = Form("run=%d and detector=\"%s\"",
2447 GetCurrentRun(), detector);
2448 if (sourceName.Length() > 0 && source)
2449 whereClause += Form(" and %s=\"%s\"", sourceName.Data(), source);
2450 TString sqlQuery = Form("%s %s", sqlQueryStart.Data(), whereClause.Data());
2451
2452 AliDebug(2, Form("SQL query: \n%s",sqlQuery.Data()));
2453
2454 // Query execution
2455 TSQLResult* aResult;
2456 aResult = fServer[system]->Query(sqlQuery);
2457 if (!aResult) {
2458 Log(detector, Form("GetFileIDs - Can't execute SQL query to %s database for source: %s",
2459 GetSystemName(system), source));
2460 return 0;
2461 }
2462
2463 TList *list = new TList();
2464 list->SetOwner(1);
2465
2466 if (aResult->GetRowCount() == 0)
2467 {
2468 Log(detector,
2469 Form("GetFileIDs - No entry in %s FXS table for source: %s", GetSystemName(system), source));
2470 delete aResult;
2471 return list;
2472 }
2473
1bcd28db 2474 Log(detector, Form("GetFileIDs - Found %d ids", aResult->GetRowCount()));
2475
4a33bdd9 2476 TSQLRow* aRow;
2477
2478 while ((aRow = aResult->Next()))
2479 {
2480
2481 TString id(aRow->GetField(0), aRow->GetFieldLength(0));
2482 AliDebug(2, Form("fileId = %s", id.Data()));
2483 list->Add(new TObjString(id));
2484 delete aRow;
2485 }
2486
2487 delete aResult;
2488
2489 return list;
2490}
2491
2bb7b766 2492//______________________________________________________________________________________________
9d733021 2493Bool_t AliShuttle::Connect(Int_t system)
2bb7b766 2494{
9827400b 2495 // Connect to MySQL Server of the system's FXS MySQL databases
2496 // DAQ Logbook, Shuttle Logbook and DAQ FXS db are on the same host
2497 //
57f50b3c 2498
9d733021 2499 // check connection: if already connected return
3a4ddf0e 2500 if(fServer[system] && fServer[system]->IsConnected()) {
2501 // ping the server --> automatic reconnection should occur if it was broken but the
2502 // server is still alive
2503 fServer[system]->Ping();
2504 return kTRUE;
2505 }
57f50b3c 2506
9d733021 2507 TString dbHost, dbUser, dbPass, dbName;
57f50b3c 2508
9d733021 2509 if (system < 3) // FXS db servers
2510 {
2511 dbHost = Form("mysql://%s:%d", fConfig->GetFXSdbHost(system), fConfig->GetFXSdbPort(system));
2512 dbUser = fConfig->GetFXSdbUser(system);
2513 dbPass = fConfig->GetFXSdbPass(system);
2514 dbName = fConfig->GetFXSdbName(system);
2515 } else { // Run & Shuttle logbook servers
2516 // TODO Will the Shuttle logbook server be the same as the Run logbook server ???
2517 dbHost = Form("mysql://%s:%d", fConfig->GetDAQlbHost(), fConfig->GetDAQlbPort());
2518 dbUser = fConfig->GetDAQlbUser();
2519 dbPass = fConfig->GetDAQlbPass();
2520 dbName = fConfig->GetDAQlbDB();
2521 }
57f50b3c 2522
9d733021 2523 fServer[system] = TSQLServer::Connect(dbHost.Data(), dbUser.Data(), dbPass.Data());
2524 if (!fServer[system] || !fServer[system]->IsConnected()) {
2525 if(system < 3)
2526 {
2527 AliError(Form("Can't establish connection to FXS database for %s",
2528 AliShuttleInterface::GetSystemName(system)));
2529 } else {
2530 AliError("Can't establish connection to Run logbook.");
57f50b3c 2531 }
9d733021 2532 if(fServer[system]) delete fServer[system];
2533 return kFALSE;
2bb7b766 2534 }
57f50b3c 2535
9d733021 2536 // Get tables
2537 TSQLResult* aResult=0;
2538 switch(system){
2539 case kDAQ:
2540 aResult = fServer[kDAQ]->GetTables(dbName.Data());
2541 break;
2542 case kDCS:
2543 aResult = fServer[kDCS]->GetTables(dbName.Data());
2544 break;
2545 case kHLT:
2546 aResult = fServer[kHLT]->GetTables(dbName.Data());
2547 break;
2548 default:
2549 aResult = fServer[3]->GetTables(dbName.Data());
2550 break;
2551 }
2552
2553 delete aResult;
2bb7b766 2554 return kTRUE;
2555}
57f50b3c 2556
9e080f92 2557//______________________________________________________________________________________________
9d733021 2558Bool_t AliShuttle::UpdateTable()
9e080f92 2559{
9827400b 2560 //
2561 // Update FXS table filling time_processed field in all rows corresponding to current run and detector
2562 //
9e080f92 2563
9d733021 2564 Bool_t result = kTRUE;
9e080f92 2565
9d733021 2566 for (UInt_t system=0; system<3; system++)
2567 {
2568 if(!fFXSCalled[system]) continue;
9e080f92 2569
9d733021 2570 // check connection, in case connect
2571 if (!Connect(system))
2572 {
2573 Log(fCurrentDetector, Form("UpdateTable - Couldn't connect to %s FXS database", GetSystemName(system)));
2574 result = kFALSE;
2575 continue;
9e080f92 2576 }
9e080f92 2577
9d733021 2578 TTimeStamp now; // now
2579
2580 // Loop on FXS list entries
2581 TIter iter(&fFXSlist[system]);
2582 TObjString *aFXSentry=0;
2583 while ((aFXSentry = dynamic_cast<TObjString*> (iter.Next())))
2584 {
2585 TString aFXSentrystr = aFXSentry->String();
2586 TObjArray *aFXSarray = aFXSentrystr.Tokenize("#!?!#");
2587 if (!aFXSarray || aFXSarray->GetEntries() != 2 )
2588 {
2589 Log(fCurrentDetector, Form("UpdateTable - error updating %s FXS entry. Check string: <%s>",
2590 GetSystemName(system), aFXSentrystr.Data()));
2591 if(aFXSarray) delete aFXSarray;
2592 result = kFALSE;
2593 continue;
2594 }
2595 const char* fileId = ((TObjString*) aFXSarray->At(0))->GetName();
2596 const char* source = ((TObjString*) aFXSarray->At(1))->GetName();
2597
2598 TString whereClause;
2599 if (system == kDAQ)
2600 {
2601 whereClause = Form("where run=%d and detector=\"%s\" and fileId=\"%s\" and DAQsource=\"%s\";",
2602 GetCurrentRun(), fCurrentDetector.Data(), fileId, source);
2603 }
2604 else if (system == kDCS)
2605 {
2606 whereClause = Form("where run=%d and detector=\"%s\" and fileId=\"%s\";",
2607 GetCurrentRun(), fCurrentDetector.Data(), fileId);
2608 }
2609 else if (system == kHLT)
2610 {
2611 whereClause = Form("where run=%d and detector=\"%s\" and fileId=\"%s\" and DDLnumbers=\"%s\";",
2612 GetCurrentRun(), fCurrentDetector.Data(), fileId, source);
2613 }
2614
2615 delete aFXSarray;
9e080f92 2616
9d733021 2617 TString sqlQuery = Form("update %s set time_processed=%d %s", fConfig->GetFXSdbTable(system),
2618 now.GetSec(), whereClause.Data());
9e080f92 2619
9d733021 2620 AliDebug(2, Form("SQL query: \n%s",sqlQuery.Data()));
9e080f92 2621
9d733021 2622 // Query execution
2623 TSQLResult* aResult;
2624 aResult = dynamic_cast<TSQLResult*> (fServer[system]->Query(sqlQuery));
2625 if (!aResult)
2626 {
2627 Log(fCurrentDetector, Form("UpdateTable - %s db: can't execute SQL query <%s>",
2628 GetSystemName(system), sqlQuery.Data()));
2629 result = kFALSE;
2630 continue;
2631 }
2632 delete aResult;
9e080f92 2633 }
9e080f92 2634 }
2635
9d733021 2636 return result;
9e080f92 2637}
57f50b3c 2638
bd23d48b 2639//_______________________________________________________________________________
2640Bool_t AliShuttle::UpdateTableSkippedCase(const char* detector)
2641{
2642 //
2643 // Update FXS table filling time_processed field in all rows corresponding to current run and detector
2644 // if detector = "ALL" update all detectors
2645 //
2646
2647 Bool_t result = kTRUE;
2648
7a782903 2649 TString detName(detector);
2650
bd23d48b 2651 for (UInt_t system=0; system<3; system++)
2652 {
2653
2654 // check connection, in case connect
2655 if (!Connect(system))
2656 {
2657 Log(fCurrentDetector, Form("UpdateTableSkippedCase - Couldn't connect to %s FXS database", GetSystemName(system)));
2658 result = kFALSE;
2659 continue;
2660 }
2661
2662 TTimeStamp now; // now
2663
2664 // Loop on FXS list entries
2665 TIter iter(&fFXSlist[system]);
2666
2667 TString whereClause;
7a782903 2668 if (detName == "ALL") whereClause = Form("where run=%d and time_processed IS NULL;",GetCurrentRun());
bd23d48b 2669 else whereClause = Form("where run=%d and detector=\"%s\" and time_processed IS NULL;",GetCurrentRun(), detector);
2670
2671 Log("SHUTTLE",Form(" whereClause = %s ",whereClause.Data()));
2672
2673 TString sqlQuery = Form("update %s set time_processed=%d %s", fConfig->GetFXSdbTable(system),
2674 now.GetSec(), whereClause.Data());
2675
2676 AliDebug(2, Form("SQL query: \n%s",sqlQuery.Data()));
2677
2678 // Query execution
2679 TSQLResult* aResult;
2680 aResult = dynamic_cast<TSQLResult*> (fServer[system]->Query(sqlQuery));
2681 if (!aResult)
2682 {
2683 Log("SHUTTLE", Form("UpdateTableSkippedCase - %s db: can't execute SQL query <%s>",
2684 GetSystemName(system), sqlQuery.Data()));
2685 result = kFALSE;
2686 continue;
2687 }
2688 delete aResult;
2689
2690 }
2691
2692 return result;
2693}
3301427a 2694//______________________________________________________________________________________________
2695Bool_t AliShuttle::UpdateTableFailCase()
2696{
9827400b 2697 // Update FXS table filling time_processed field in all rows corresponding to current run and detector
2698 // this is called in case the preprocessor is declared failed for the current run, because
2699 // the fields are updated only in case of success
3301427a 2700
2701 Bool_t result = kTRUE;
2702
2703 for (UInt_t system=0; system<3; system++)
2704 {
2705 // check connection, in case connect
2706 if (!Connect(system))
2707 {
2708 Log(fCurrentDetector, Form("UpdateTableFailCase - Couldn't connect to %s FXS database",
2709 GetSystemName(system)));
2710 result = kFALSE;
2711 continue;
2712 }
2713
2714 TTimeStamp now; // now
2715
2716 // Loop on FXS list entries
2717
2718 TString whereClause = Form("where run=%d and detector=\"%s\";",
2719 GetCurrentRun(), fCurrentDetector.Data());
2720
2721
2722 TString sqlQuery = Form("update %s set time_processed=%d %s", fConfig->GetFXSdbTable(system),
2723 now.GetSec(), whereClause.Data());
2724
2725 AliDebug(2, Form("SQL query: \n%s",sqlQuery.Data()));
2726
2727 // Query execution
2728 TSQLResult* aResult;
2729 aResult = dynamic_cast<TSQLResult*> (fServer[system]->Query(sqlQuery));
2730 if (!aResult)
2731 {
2732 Log(fCurrentDetector, Form("UpdateTableFailCase - %s db: can't execute SQL query <%s>",
2733 GetSystemName(system), sqlQuery.Data()));
2734 result = kFALSE;
2735 continue;
2736 }
2737 delete aResult;
2738 }
2739
2740 return result;
2741}
2742
2bb7b766 2743//______________________________________________________________________________________________
2744Bool_t AliShuttle::UpdateShuttleLogbook(const char* detector, const char* status)
2745{
e7f62f16 2746 //
2747 // Update Shuttle logbook filling detector or shuttle_done column
2748 // ex. of usage: UpdateShuttleLogbook("PHOS", "DONE") or UpdateShuttleLogbook("shuttle_done")
2749 //
57f50b3c 2750
2bb7b766 2751 // check connection, in case connect
be48e3ea 2752 if(!Connect(3)){
2bb7b766 2753 Log("SHUTTLE", "UpdateShuttleLogbook - Couldn't connect to DAQ Logbook.");
2754 return kFALSE;
57f50b3c 2755 }
2756
2bb7b766 2757 TString detName(detector);
2758 TString setClause;
8e828d39 2759 if (detName == "shuttle_done" || detName == "shuttle_skipped")
e7f62f16 2760 {
2bb7b766 2761 setClause = "set shuttle_done=1";
8e828d39 2762
ee6f7523 2763 if (detName == "shuttle_done")
b0e53b15 2764 {
8e828d39 2765 if (TouchFile() != kTRUE)
2156c755 2766 {
2767 SendMLRunInfo("Pending");
503ff24f 2768 return kFALSE;
2156c755 2769 }
8e828d39 2770
2771 SendMLRunInfo("Done");
b0e53b15 2772 }
8e828d39 2773 else
2774 SendMLRunInfo("Skipped");
2775 }
2776 else {
2bb7b766 2777 TString statusStr(status);
2778 if(statusStr.Contains("done", TString::kIgnoreCase) ||
2779 statusStr.Contains("failed", TString::kIgnoreCase)){
eba76848 2780 setClause = Form("set %s=\"%s\"", detector, status);
2bb7b766 2781 } else {
2782 Log("SHUTTLE",
2783 Form("UpdateShuttleLogbook - Invalid status <%s> for detector %s",
2784 status, detector));
2785 return kFALSE;
2786 }
2787 }
57f50b3c 2788
2bb7b766 2789 TString whereClause = Form("where run=%d", GetCurrentRun());
2790
441b0e9c 2791 TString sqlQuery = Form("update %s %s %s",
2792 fConfig->GetShuttlelbTable(), setClause.Data(), whereClause.Data());
57f50b3c 2793
2bb7b766 2794 AliDebug(2, Form("SQL query: \n%s",sqlQuery.Data()));
2795
2796 // Query execution
2797 TSQLResult* aResult;
be48e3ea 2798 aResult = dynamic_cast<TSQLResult*> (fServer[3]->Query(sqlQuery));
2bb7b766 2799 if (!aResult) {
2800 Log("SHUTTLE", Form("UpdateShuttleLogbook - Can't execute query <%s>", sqlQuery.Data()));
2801 return kFALSE;
57f50b3c 2802 }
2bb7b766 2803 delete aResult;
57f50b3c 2804
2805 return kTRUE;
2806}
2807
2808//______________________________________________________________________________________________
2bb7b766 2809Int_t AliShuttle::GetCurrentRun() const
2810{
9827400b 2811 //
2812 // Get current run from logbook entry
2813 //
57f50b3c 2814
2bb7b766 2815 return fLogbookEntry ? fLogbookEntry->GetRun() : -1;
57f50b3c 2816}
2817
2818//______________________________________________________________________________________________
2bb7b766 2819UInt_t AliShuttle::GetCurrentStartTime() const
2820{
9827400b 2821 //
2822 // get current start time
2823 //
57f50b3c 2824
2bb7b766 2825 return fLogbookEntry ? fLogbookEntry->GetStartTime() : 0;
57f50b3c 2826}
2827
2828//______________________________________________________________________________________________
2bb7b766 2829UInt_t AliShuttle::GetCurrentEndTime() const
2830{
9827400b 2831 //
2832 // get current end time from logbook entry
2833 //
57f50b3c 2834
2bb7b766 2835 return fLogbookEntry ? fLogbookEntry->GetEndTime() : 0;
57f50b3c 2836}
5f3fe74e 2837//______________________________________________________________________________________________
2838UInt_t AliShuttle::GetCurrentTimeCreated() const
2839{
2840 //
2841 // get current end time from logbook entry
2842 //
2843
2844 return fLogbookEntry ? fLogbookEntry->GetTimeCreated() : 0;
2845}
57f50b3c 2846
675f64cd 2847//______________________________________________________________________________________________
2848UInt_t AliShuttle::GetCurrentYear() const
2849{
2850 //
2851 // Get current year from logbook entry
2852 //
2853
2854 if (!fLogbookEntry) return 0;
2855
2856 TTimeStamp startTime(GetCurrentStartTime());
2857 TString year = Form("%d",startTime.GetDate());
2858 year = year(0,4);
2859
2860 return year.Atoi();
2861}
2862
2863//______________________________________________________________________________________________
2864const char* AliShuttle::GetLHCPeriod() const
2865{
2866 //
2867 // Get current LHC period from logbook entry
2868 //
2869
2870 if (!fLogbookEntry) return 0;
2871
2872 return fLogbookEntry->GetRunParameter("LHCperiod");
2873}
2874
b948db8d 2875//______________________________________________________________________________________________
2876void AliShuttle::Log(const char* detector, const char* message)
2877{
9827400b 2878 //
2879 // Fill log string with a message
2880 //
7d4cf768 2881
3ebbf3c4 2882 TString logRunDir = GetShuttleLogDir();
2883 if (GetCurrentRun() >=0) {
2884 Int_t logDir_lev1 = GetCurrentRun()/10000;
2885 logRunDir += Form("/%d/%d", logDir_lev1, GetCurrentRun());
2886 }
7d4cf768 2887 void* dir = gSystem->OpenDirectory(logRunDir.Data());
84090f85 2888 if (dir == NULL) {
7d4cf768 2889 if (gSystem->mkdir(logRunDir.Data(), kTRUE)) {
36c99a6a 2890 AliError(Form("Can't open directory <%s>", GetShuttleLogDir()));
84090f85 2891 return;
2892 }
b948db8d 2893
84090f85 2894 } else {
2895 gSystem->FreeDirectory(dir);
2896 }
b948db8d 2897
0b025a52 2898 TString toLog = Form("%s UTC (%d): %s - ", TTimeStamp(time(0)).AsString("s"), getpid(), detector);
e7f62f16 2899 if (GetCurrentRun() >= 0)
2900 toLog += Form("run %d - ", GetCurrentRun());
2bb7b766 2901 toLog += Form("%s", message);
2902
84090f85 2903 AliInfo(toLog.Data());
ffa29e93 2904
2905 // if we redirect the log output already to the file, leave here
2906 if (fOutputRedirected && strcmp(detector, "SHUTTLE") != 0)
2907 return;
b948db8d 2908
ffa29e93 2909 TString fileName = GetLogFileName(detector);
e7f62f16 2910
84090f85 2911 gSystem->ExpandPathName(fileName);
2912
2913 ofstream logFile;
2914 logFile.open(fileName, ofstream::out | ofstream::app);
2915
2916 if (!logFile.is_open()) {
2917 AliError(Form("Could not open file %s", fileName.Data()));
2918 return;
2919 }
7bfb2090 2920
84090f85 2921 logFile << toLog.Data() << "\n";
b948db8d 2922
84090f85 2923 logFile.close();
b948db8d 2924}
2bb7b766 2925
ffa29e93 2926//______________________________________________________________________________________________
2927TString AliShuttle::GetLogFileName(const char* detector) const
2928{
2929 //
2930 // returns the name of the log file for a given sub detector
2931 //
2932
2933 TString fileName;
2934
2935 if (GetCurrentRun() >= 0)
7d4cf768 2936 {
3ebbf3c4 2937 Int_t logDir_lev1 = GetCurrentRun()/10000;
2938 fileName.Form("%s/%d/%d/%s.log", GetShuttleLogDir(), logDir_lev1, GetCurrentRun(),
2939 detector);
7d4cf768 2940 } else {
ffa29e93 2941 fileName.Form("%s/%s.log", GetShuttleLogDir(), detector);
7d4cf768 2942 }
ffa29e93 2943
2944 return fileName;
2945}
2946
ee6f7523 2947//______________________________________________________________________________________________
2948void AliShuttle::SendAlive()
2949{
2950 // sends alive message to ML
2951
2952 TMonaLisaText mlStatus("SHUTTLE_status", "Alive");
2953
2954 TList mlList;
2955 mlList.Add(&mlStatus);
2956
2957 fMonaLisa->SendParameters(&mlList, "__PROCESSINGINFO__");
2958}
2959
2bb7b766 2960//______________________________________________________________________________________________
2961Bool_t AliShuttle::Collect(Int_t run)
2962{
9827400b 2963 //
2964 // Collects conditions data for all UNPROCESSED run written to DAQ LogBook in case of run = -1 (default)
2965 // If a dedicated run is given this run is processed
2966 //
2967 // In operational mode, this is the Shuttle function triggered by the EOR signal.
2968 //
2bb7b766 2969
eba76848 2970 if (run == -1)
2971 Log("SHUTTLE","Collect - Shuttle called. Collecting conditions data for unprocessed runs");
2972 else
2973 Log("SHUTTLE", Form("Collect - Shuttle called. Collecting conditions data for run %d", run));
cb343cfd 2974
2975 SetLastAction("Starting");
2bb7b766 2976
ee6f7523 2977 // create ML instance
2978 if (!fMonaLisa)
2979 fMonaLisa = new TMonaLisaWriter(fConfig->GetMonitorHost(), fConfig->GetMonitorTable());
2980
fbc112e3 2981 CountOpenRuns();
ee6f7523 2982
2bb7b766 2983 TString whereClause("where shuttle_done=0");
eba76848 2984 if (run != -1)
2985 whereClause += Form(" and run=%d", run);
2bb7b766 2986
2987 TObjArray shuttleLogbookEntries;
be48e3ea 2988 if (!QueryShuttleLogbook(whereClause, shuttleLogbookEntries))
2989 {
cb343cfd 2990 Log("SHUTTLE", "Collect - Can't retrieve entries from Shuttle logbook");
2bb7b766 2991 return kFALSE;
2992 }
2993
9e080f92 2994 if (shuttleLogbookEntries.GetEntries() == 0)
2995 {
2996 if (run == -1)
2997 Log("SHUTTLE","Collect - Found no UNPROCESSED runs in Shuttle logbook");
2998 else
2999 Log("SHUTTLE", Form("Collect - Run %d is already DONE "
3000 "or it does not exist in Shuttle logbook", run));
3001 return kTRUE;
3002 }
3003
be48e3ea 3004 for (UInt_t iDet=0; iDet<NDetectors(); iDet++)
3005 fFirstUnprocessed[iDet] = kTRUE;
3006
fc5a4708 3007 if (run != -1)
be48e3ea 3008 {
3009 // query Shuttle logbook for earlier runs, check if some detectors are unprocessed,
3010 // flag them into fFirstUnprocessed array
e3ae8d33 3011 TString whereClauseBis(Form("where shuttle_done=0 and run < %d", run));
be48e3ea 3012 TObjArray tmpLogbookEntries;
e3ae8d33 3013 if (!QueryShuttleLogbook(whereClauseBis, tmpLogbookEntries))
be48e3ea 3014 {
3015 Log("SHUTTLE", "Collect - Can't retrieve entries from Shuttle logbook");
3016 return kFALSE;
3017 }
3018
3019 TIter iter(&tmpLogbookEntries);
3020 AliShuttleLogbookEntry* anEntry = 0;
3021 while ((anEntry = dynamic_cast<AliShuttleLogbookEntry*> (iter.Next())))
3022 {
3023 for (UInt_t iDet=0; iDet<NDetectors(); iDet++)
3024 {
3025 if (anEntry->GetDetectorStatus(iDet) == AliShuttleLogbookEntry::kUnprocessed)
3026 {
3027 AliDebug(2, Form("Run %d: setting %s as \"not first time unprocessed\"",
3028 anEntry->GetRun(), GetDetName(iDet)));
3029 fFirstUnprocessed[iDet] = kFALSE;
3030 }
3031 }
3032
3033 }
3034
3035 }
3036
3037 if (!RetrieveConditionsData(shuttleLogbookEntries))
3038 {
cb343cfd 3039 Log("SHUTTLE", "Collect - Process of at least one run failed");
a3ca69fe 3040 CountOpenRuns();
2bb7b766 3041 return kFALSE;
3042 }
3043
36c99a6a 3044 Log("SHUTTLE", "Collect - Requested run(s) successfully processed");
a3ca69fe 3045 CountOpenRuns();
eba76848 3046 return kTRUE;
2bb7b766 3047}
3048
2bb7b766 3049//______________________________________________________________________________________________
3050Bool_t AliShuttle::RetrieveConditionsData(const TObjArray& dateEntries)
3051{
9827400b 3052 //
3053 // Retrieve conditions data for all runs that aren't processed yet
3054 //
2bb7b766 3055
3056 Bool_t hasError = kFALSE;
3057
3058 TIter iter(&dateEntries);
3059 AliShuttleLogbookEntry* anEntry;
3060
3061 while ((anEntry = (AliShuttleLogbookEntry*) iter.Next())){
3062 if (!Process(anEntry)){
3063 hasError = kTRUE;
3064 }
4b95672b 3065
3066 // clean SHUTTLE temp directory
d524ade6 3067 //TString filename = Form("%s/*.shuttle", GetShuttleTempDir());
3068 //RemoveFile(filename.Data());
2bb7b766 3069 }
3070
3071 return hasError == kFALSE;
3072}
cb343cfd 3073
3074//______________________________________________________________________________________________
3075ULong_t AliShuttle::GetTimeOfLastAction() const
3076{
9827400b 3077 //
3078 // Gets time of last action
3079 //
3080
cb343cfd 3081 ULong_t tmp;
36c99a6a 3082
cb343cfd 3083 fMonitoringMutex->Lock();
be48e3ea 3084
cb343cfd 3085 tmp = fLastActionTime;
36c99a6a 3086
cb343cfd 3087 fMonitoringMutex->UnLock();
36c99a6a 3088
cb343cfd 3089 return tmp;
3090}
3091
3092//______________________________________________________________________________________________
3093const TString AliShuttle::GetLastAction() const
3094{
9827400b 3095 //
cb343cfd 3096 // returns a string description of the last action
9827400b 3097 //
cb343cfd 3098
3099 TString tmp;
36c99a6a 3100
cb343cfd 3101 fMonitoringMutex->Lock();
3102
3103 tmp = fLastAction;
3104