Getting in sync with base class
[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
462//______________________________________________________________________________________________
546242fb 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
557//______________________________________________________________________________________________
9827400b 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
860//______________________________________________________________________________________________
2d9019b4 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
2d9019b4 881//______________________________________________________________________________________________
3301427a 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
1192//______________________________________________________________________________________________
8e828d39 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 }
1206 TMonaLisaText mlRunType("SHUTTLE_runtype", runType);
1207
1208 TList mlList;
1209 mlList.Add(&mlStatus);
1210 mlList.Add(&mlRunType);
1211
1212 TString mlID;
1213 mlID.Form("%d", GetCurrentRun());
1214 fMonaLisa->SendParameters(&mlList, mlID);
1215}
1216
1217//______________________________________________________________________________________________
ef83773f 1218Int_t AliShuttle::GetMem(Int_t pid)
1219{
1220 // invokes ps to get the memory consumption of the process <pid>
1221 // returns -1 in case of error
1222
1223 TString checkStr;
1224 checkStr.Form("ps -o vsize --pid %d | tail -n 1", pid);
1225 FILE* pipe = gSystem->OpenPipe(checkStr, "r");
1226 if (!pipe)
1227 {
1228 Log("SHUTTLE", Form("Process - Error: "
1229 "Could not open pipe to %s", checkStr.Data()));
1230 return -1;
1231 }
1232
1233 char buffer[100];
1234 if (!fgets(buffer, 100, pipe))
1235 {
1236 Log("SHUTTLE", "Process - Error: ps did not return anything");
1237 gSystem->ClosePipe(pipe);
1238 return -1;
1239 }
1240 gSystem->ClosePipe(pipe);
1241
1242 //Log("SHUTTLE", Form("ps returned %s", buffer));
1243
1244 Int_t mem = 0;
1245 if ((sscanf(buffer, "%d\n", &mem) != 1) || !mem)
1246 {
1247 Log("SHUTTLE", "Process - Error: Could not parse output of ps");
1248 return -1;
1249 }
1250
1251 return mem;
1252}
1253
1254//______________________________________________________________________________________________
2bb7b766 1255Bool_t AliShuttle::Process(AliShuttleLogbookEntry* entry)
58bc3020 1256{
73abe331 1257 //
b948db8d 1258 // Makes data retrieval for all detectors in the configuration.
2bb7b766 1259 // entry: Shuttle logbook entry, contains run paramenters and status of detectors
1260 // (Unprocessed, Inactive, Failed or Done).
d477ad88 1261 // Returns kFALSE in case of error occured and kTRUE otherwise
73abe331 1262 //
1263
9827400b 1264 if (!entry) return kFALSE;
2bb7b766 1265
1266 fLogbookEntry = entry;
1267
c88ad5db 1268 Log("SHUTTLE", Form("\t\t\t^*^*^*^*^*^*^*^*^*^*^*^* run %d: START ^*^*^*^*^*^*^*^*^*^*^*^*",
9827400b 1269 GetCurrentRun()));
2bb7b766 1270
339adafa 1271 CountOpenRuns();
1272
8e828d39 1273 // Send the information to ML
1274 SendMLRunInfo("Processing");
3301427a 1275
9827400b 1276 if (fLogbookEntry->IsDone())
1277 {
1278 Log("SHUTTLE","Process - Shuttle is already DONE. Updating logbook");
1279 UpdateShuttleLogbook("shuttle_done");
1280 fLogbookEntry = 0;
1281 return kTRUE;
1282 }
1283
1284 // read test mode if flag is set
1285 if (fReadTestMode)
1286 {
3d8bc902 1287 fTestMode = kNone;
9827400b 1288 TString logEntry(entry->GetRunParameter("log"));
1289 //printf("log entry = %s\n", logEntry.Data());
1290 TString searchStr("Testmode: ");
1291 Int_t pos = logEntry.Index(searchStr.Data());
1292 //printf("%d\n", pos);
1293 if (pos >= 0)
1294 {
1295 TSubString subStr = logEntry(pos + searchStr.Length(), logEntry.Length());
1296 //printf("%s\n", subStr.String().Data());
1297 TString newStr(subStr.Data());
1298 TObjArray* token = newStr.Tokenize(' ');
1299 if (token)
1300 {
1301 //token->Print();
1302 TObjString* tmpStr = dynamic_cast<TObjString*> (token->First());
1303 if (tmpStr)
1304 {
1305 Int_t testMode = tmpStr->String().Atoi();
1306 if (testMode > 0)
1307 {
c88ad5db 1308 Log("SHUTTLE", Form("Process - Enabling test mode %d", testMode));
9827400b 1309 SetTestMode((TestMode) testMode);
1310 }
1311 }
1312 delete token;
1313 }
1314 }
1315 }
c88ad5db 1316
eba76848 1317 fLogbookEntry->Print("all");
57f50b3c 1318
1319 // Initialization
d477ad88 1320 Bool_t hasError = kFALSE;
5164a766 1321
675f64cd 1322 // Set the CDB and Reference folders according to the year and LHC period
1323 TString lhcPeriod(GetLHCPeriod());
1324 if (lhcPeriod.Length() == 0)
1325 {
7d4cf768 1326 Log("SHUTTLE","Process - LHCPeriod not found in logbook!");
1327 return 0;
675f64cd 1328 }
1329
1265724d 1330 // build cdb paths (repeat each time, LHCperiod might have changed)
1331 fgkMainCDB.Form("alien://folder=%s%d/%s/OCDB?user=alidaq?cacheFold=/tmp/OCDBCache",
b8cf5e0f 1332 fConfig->GetAlienPath(), GetCurrentYear(), lhcPeriod.Data());
675f64cd 1333
1265724d 1334 fgkMainRefStorage.Form("alien://folder=%s%d/%s/Reference?user=alidaq?cacheFold=/tmp/OCDBCache",
b8cf5e0f 1335 fConfig->GetAlienPath(), GetCurrentYear(), lhcPeriod.Data());
675f64cd 1336
57f50b3c 1337 // Loop on detectors in the configuration
b948db8d 1338 TIter iter(fConfig->GetDetectors());
2bb7b766 1339 TObjString* aDetector = 0;
b948db8d 1340
4508ab8b 1341 Bool_t first = kTRUE;
1342
be48e3ea 1343 while ((aDetector = (TObjString*) iter.Next()))
1344 {
7bfb2090 1345 fCurrentDetector = aDetector->String();
5164a766 1346
5e993b6f 1347 if (ContinueProcessing() == kFALSE)
1348 continue;
4508ab8b 1349
1350 if (first)
1351 {
1352 // only read QueryCDB when needed and only once
1353 AliCDBStorage *mainCDBSto = AliCDBManager::Instance()->GetStorage(fgkMainCDB);
1354 if(mainCDBSto) mainCDBSto->QueryCDB(GetCurrentRun());
1355 AliCDBStorage *mainRefSto = AliCDBManager::Instance()->GetStorage(fgkMainRefStorage);
1356 if(mainRefSto) mainRefSto->QueryCDB(GetCurrentRun());
1357 first = kFALSE;
1358 }
9e080f92 1359
c88ad5db 1360 Log("SHUTTLE", Form("\t\t\t****** run %d - %s: START ******",
2bb7b766 1361 GetCurrentRun(), aDetector->GetName()));
1362
9d733021 1363 for(Int_t iSys=0;iSys<3;iSys++) fFXSCalled[iSys]=kFALSE;
ef83773f 1364
1365 Int_t initialMem = GetMem(getpid());
1366 Log("SHUTTLE", Form("Memory consumption before forking is %d", initialMem));
9d733021 1367
c88ad5db 1368 Log(fCurrentDetector.Data(), "Process - Starting processing");
85a80aa9 1369
be48e3ea 1370 Int_t pid = fork();
1371
1372 if (pid < 0)
1373 {
c88ad5db 1374 Log("SHUTTLE", "Process - ERROR: Forking failed");
be48e3ea 1375 }
1376 else if (pid > 0)
1377 {
1378 // parent
c88ad5db 1379 Log("SHUTTLE", Form("Process - In parent process of %d - %s: Starting monitoring",
be48e3ea 1380 GetCurrentRun(), aDetector->GetName()));
1381
1382 Long_t begin = time(0);
1383
1384 int status; // to be used with waitpid, on purpose an int (not Int_t)!
1385 while (waitpid(pid, &status, WNOHANG) == 0)
1386 {
1387 Long_t expiredTime = time(0) - begin;
1388
1389 if (expiredTime > fConfig->GetPPTimeOut())
1390 {
56d3e9e4 1391 TString logMsg;
1392 AliShuttleStatus *currentStatus = ReadShuttleStatus();
1393 AliShuttleStatus::Status newStatus = AliShuttleStatus::kInvalid;
1394
1395 if (currentStatus->GetStatus() <= AliShuttleStatus::kPPDone)
1396 {
1397 // in case pp not yet done set status to kPPTimeOut
1398
1399 logMsg.Form("Process - Process of %s timed out. Run time: %d seconds. Killing...",
1400 fCurrentDetector.Data(), expiredTime);
1401 newStatus = AliShuttleStatus::kPPTimeOut;
1402 }
1403 else if (currentStatus->GetStatus() == AliShuttleStatus::kStoreStarted)
1404 {
1405 // in case the pp goes in TimeOut while storing the objects in the OCDB
1406 // set status to kStoreError
1407
1408 logMsg.Form("Process - Process of %s timed out while storing the OCDB object. Run time: %d seconds. Killing... and setting status to StoreError.",
1409 fCurrentDetector.Data(), expiredTime);
1410 newStatus = AliShuttleStatus::kStoreError;
1411 }
1412 else
1413 {
1414 // in other cases don't change the status
1415
1416 logMsg.Form("Process - Process of %s timed out in status = %s. Run time: %d seconds. Killing... without changing the status",
1417 fCurrentDetector.Data(), currentStatus->GetStatusName(), expiredTime);
1418 }
1419
1420 Log("SHUTTLE", logMsg);
1421 Log(fCurrentDetector, logMsg);
be48e3ea 1422
1423 kill(pid, 9);
1424
56d3e9e4 1425 if (newStatus != AliShuttleStatus::kInvalid)
1426 UpdateShuttleStatus(newStatus);
be48e3ea 1427 hasError = kTRUE;
1428
1429 gSystem->Sleep(1000);
1430 }
1431 else
1432 {
be48e3ea 1433 gSystem->Sleep(1000);
9827400b 1434
ef83773f 1435 Int_t mem = GetMem(pid);
1436
1437 if (mem < 0)
9827400b 1438 continue;
9827400b 1439
ef83773f 1440 mem -= initialMem;
1441 if (mem < 0)
1442 mem = 0;
9827400b 1443
1444 if (expiredTime % 60 == 0)
ee6f7523 1445 {
c88ad5db 1446 Log("SHUTTLE", Form("Process - %s: Checking process. "
1447 "Run time: %d seconds - Memory consumption: %d KB",
1448 fCurrentDetector.Data(), expiredTime, mem));
ee6f7523 1449 SendAlive();
1450 }
9827400b 1451
1452 if (mem > fConfig->GetPPMaxMem())
1453 {
1454 TString tmp;
c88ad5db 1455 tmp.Form("Process - Process exceeds maximum allowed memory "
1456 "(%d KB > %d KB). Killing...",
9827400b 1457 mem, fConfig->GetPPMaxMem());
1458 Log("SHUTTLE", tmp);
1459 Log(fCurrentDetector, tmp);
1460
1461 kill(pid, 9);
1462
1463 UpdateShuttleStatus(AliShuttleStatus::kPPOutOfMemory);
1464 hasError = kTRUE;
1465
1466 gSystem->Sleep(1000);
1467 }
be48e3ea 1468 }
1469 }
1470
c88ad5db 1471 Log("SHUTTLE", Form("Process - In parent process of %d - %s: Client has terminated.",
be48e3ea 1472 GetCurrentRun(), aDetector->GetName()));
1473
1474 if (WIFEXITED(status))
1475 {
1476 Int_t returnCode = WEXITSTATUS(status);
1477
c88ad5db 1478 Log("SHUTTLE", Form("Process - %s: the return code is %d", fCurrentDetector.Data(),
3301427a 1479 returnCode));
be48e3ea 1480
9827400b 1481 if (returnCode == 0) hasError = kTRUE;
be48e3ea 1482 }
1483 }
1484 else if (pid == 0)
1485 {
26b1866e 1486 // child
1487 Log("SHUTTLE", Form("Process - In child process of %d - %s", GetCurrentRun(),
c88ad5db 1488 aDetector->GetName()));
be48e3ea 1489
c88ad5db 1490 Log("SHUTTLE", Form("Process - Redirecting output to %s log",fCurrentDetector.Data()));
ffa29e93 1491
546242fb 1492 if ((freopen(GetLogFileName(fCurrentDetector), "a", stdout)) == 0)
ffa29e93 1493 {
c88ad5db 1494 Log("SHUTTLE", "Process - Could not freopen stdout");
ffa29e93 1495 }
1496 else
1497 {
1498 fOutputRedirected = kTRUE;
1499 if ((dup2(fileno(stdout), fileno(stderr))) < 0)
c88ad5db 1500 Log("SHUTTLE", "Process - Could not redirect stderr");
ffa29e93 1501
1502 }
1503
5bac2bde 1504 TString wd = gSystem->WorkingDirectory();
3ebbf3c4 1505 Int_t dir_lev1 = GetCurrentRun()/10000;
1506 TString tmpDir = Form("%s/%d/%d/%s_process", GetShuttleTempDir(),
1507 dir_lev1, GetCurrentRun(), fCurrentDetector.Data());
5bac2bde 1508
d524ade6 1509 Int_t result = gSystem->GetPathInfo(tmpDir.Data(), 0, (Long64_t*) 0, 0, 0);
1510 if (!result) // temp dir already exists!
1511 {
1512 Log(fCurrentDetector.Data(),
1513 Form("Process - %s dir already exists! Removing...", tmpDir.Data()));
1514 gSystem->Exec(Form("rm -rf %s",tmpDir.Data()));
675f64cd 1515 }
1516
1517 if (gSystem->mkdir(tmpDir.Data(), 1))
1518 {
1519 Log(fCurrentDetector.Data(), "Process - could not make temp directory!!");
1520 gSystem->Exit(1);
d524ade6 1521 }
1522
1523 if (!gSystem->ChangeDirectory(tmpDir.Data()))
1524 {
1525 Log(fCurrentDetector.Data(), "Process - could not change directory!!");
1526 gSystem->Exit(1);
1527 }
5bac2bde 1528
ffa25f20 1529 Int_t success = ProcessCurrentDetector();
26b1866e 1530
5bac2bde 1531 gSystem->ChangeDirectory(wd.Data());
b1d18693 1532
ffa25f20 1533 if (success == 1) // Preprocessor finished successfully!
9827400b 1534 {
fb40ead4 1535 // remove temporary folder or DCS map
1536 if (!fConfig->KeepTempFolder())
1537 {
1538 gSystem->Exec(Form("rm -rf %s",tmpDir.Data()));
1539 } else if (!fConfig->KeepDCSMap())
1540 {
1541 gSystem->Exec(Form("rm -f %s/DCSMap.root",tmpDir.Data()));
1542 }
b1d18693 1543
3301427a 1544 // Update time_processed field in FXS DB
1545 if (UpdateTable() == kFALSE)
5bac2bde 1546 Log("SHUTTLE", Form("Process - %s: Could not update FXS databases!",
1547 fCurrentDetector.Data()));
3301427a 1548
1549 // Transfer the data from local storage to main storage (Grid)
3301427a 1550 if (StoreOCDB() == kFALSE)
9827400b 1551 success = kFALSE;
5e993b6f 1552 }
ffa25f20 1553 else if (success == 0)
c88ad5db 1554 {
1555 Log("SHUTTLE",
1556 Form("\t\t\t****** run %d - %s: PP ERROR ******",
1557 GetCurrentRun(), aDetector->GetName()));
be48e3ea 1558 }
1559
4b95672b 1560 for (UInt_t iSys=0; iSys<3; iSys++)
1561 {
1562 if (fFXSCalled[iSys]) fFXSlist[iSys].Clear();
1563 }
1564
c88ad5db 1565 Log("SHUTTLE", Form("Process - Client process of %d - %s is exiting now with %d.",
9827400b 1566 GetCurrentRun(), aDetector->GetName(), success));
be48e3ea 1567
1568 // the client exits here
9827400b 1569 gSystem->Exit(success);
be48e3ea 1570
1571 AliError("We should never get here!!!");
1572 }
7bfb2090 1573 }
5164a766 1574
c88ad5db 1575 Log("SHUTTLE", Form("\t\t\t^*^*^*^*^*^*^*^*^*^*^*^* run %d: FINISH ^*^*^*^*^*^*^*^*^*^*^*^*",
2bb7b766 1576 GetCurrentRun()));
1577
1578 //check if shuttle is done for this run, if so update logbook
1579 TObjArray checkEntryArray;
1580 checkEntryArray.SetOwner(1);
9e080f92 1581 TString whereClause = Form("where run=%d", GetCurrentRun());
b0e53b15 1582 if (!QueryShuttleLogbook(whereClause.Data(), checkEntryArray) ||
1583 checkEntryArray.GetEntries() == 0) {
9e080f92 1584 Log("SHUTTLE", Form("Process - Warning: Cannot check status of run %d on Shuttle logbook!",
1585 GetCurrentRun()));
1586 return hasError == kFALSE;
1587 }
b948db8d 1588
9e080f92 1589 AliShuttleLogbookEntry* checkEntry = dynamic_cast<AliShuttleLogbookEntry*>
1590 (checkEntryArray.At(0));
2bb7b766 1591
9e080f92 1592 if (checkEntry)
1593 {
1594 if (checkEntry->IsDone())
be48e3ea 1595 {
9e080f92 1596 Log("SHUTTLE","Process - Shuttle is DONE. Updating logbook");
1597 UpdateShuttleLogbook("shuttle_done");
1598 }
1599 else
1600 {
1601 for (UInt_t iDet=0; iDet<NDetectors(); iDet++)
be48e3ea 1602 {
9e080f92 1603 if (checkEntry->GetDetectorStatus(iDet) == AliShuttleLogbookEntry::kUnprocessed)
be48e3ea 1604 {
9e080f92 1605 AliDebug(2, Form("Run %d: setting %s as \"not first time unprocessed\"",
1606 checkEntry->GetRun(), GetDetName(iDet)));
1607 fFirstUnprocessed[iDet] = kFALSE;
be48e3ea 1608 }
1609 }
8e828d39 1610 SendMLRunInfo("Pending");
2bb7b766 1611 }
1612 }
1613
1614 fLogbookEntry = 0;
85a80aa9 1615
a7160fe9 1616 return hasError == kFALSE;
73abe331 1617}
1618
b948db8d 1619//______________________________________________________________________________________________
ffa25f20 1620Int_t AliShuttle::ProcessCurrentDetector()
73abe331 1621{
1622 //
2bb7b766 1623 // Makes data retrieval just for a specific detector (fCurrentDetector).
73abe331 1624 // Threre should be a configuration for this detector.
73abe331 1625
1d172743 1626 Log("SHUTTLE", Form("ProcessCurrentDetector - Retrieving values for %s, run %d",
1627 fCurrentDetector.Data(), GetCurrentRun()));
73abe331 1628
d524ade6 1629 TString wd = gSystem->WorkingDirectory();
1630
2d9019b4 1631 if (!CleanReferenceStorage(fCurrentDetector.Data()))
ffa25f20 1632 return 0;
d524ade6 1633
1634 gSystem->ChangeDirectory(wd.Data());
1635
3301427a 1636 // call preprocessor
1637 AliPreprocessor* aPreprocessor =
1638 dynamic_cast<AliPreprocessor*> (fPreprocessorMap.GetValue(fCurrentDetector));
1639
8579c6fb 1640 // check if the preprocessor wants to process this run type
1641 if (aPreprocessor->ProcessRunType() == kFALSE)
ffa25f20 1642 {
1643 UpdateShuttleStatus(AliShuttleStatus::kSkipped);
1644 UpdateShuttleLogbook(fCurrentDetector, "DONE");
bd23d48b 1645 if (!UpdateTableSkippedCase(fCurrentDetector.Data()))
1646 {
1647 AliError(Form("Could not update FXS tables for run %d !", GetCurrentRun()));
1648 }
ffa25f20 1649 Log(fCurrentDetector, Form("ProcessCurrentDetector - %s preprocessor is not interested in this run type", fCurrentDetector.Data()));
1650
1651 return 2;
1652 }
1653
1654 TMap* dcsMap = new TMap();
1655
3301427a 1656 aPreprocessor->Initialize(GetCurrentRun(), GetCurrentStartTime(), GetCurrentEndTime());
1657
1658 Bool_t processDCS = aPreprocessor->ProcessDCS();
d477ad88 1659
651fdaab 1660 if (!processDCS)
1661 {
1d172743 1662 Log(fCurrentDetector, "ProcessCurrentDetector -"
1663 " The preprocessor requested to skip the retrieval of DCS values");
651fdaab 1664 }
8b739301 1665 else if (fTestMode & kSkipDCS)
2c15234c 1666 {
1d172743 1667 Log(fCurrentDetector, "ProcessCurrentDetector - In TESTMODE: Skipping DCS processing");
9827400b 1668 }
1669 else if (fTestMode & kErrorDCS)
1670 {
1d172743 1671 Log(fCurrentDetector, "ProcessCurrentDetector - In TESTMODE: Simulating DCS error");
3d8bc902 1672 UpdateShuttleStatus(AliShuttleStatus::kDCSStarted);
9827400b 1673 UpdateShuttleStatus(AliShuttleStatus::kDCSError);
1d172743 1674 delete dcsMap;
ffa25f20 1675 return 0;
2c15234c 1676 } else {
3301427a 1677
1678 UpdateShuttleStatus(AliShuttleStatus::kDCSStarted);
1679
1d172743 1680 // Query DCS archive
1681 Int_t nServers = fConfig->GetNServers(fCurrentDetector);
a038aa70 1682
1d172743 1683 for (int iServ=0; iServ<nServers; iServ++)
2c15234c 1684 {
1d172743 1685
1686 TString host(fConfig->GetDCSHost(fCurrentDetector, iServ));
1687 Int_t port = fConfig->GetDCSPort(fCurrentDetector, iServ);
542b6cc8 1688 Int_t multiSplit = fConfig->GetMultiSplit(fCurrentDetector, iServ);
1689
1790d4b7 1690 Log(fCurrentDetector, Form("ProcessCurrentDetector -"
1691 " Querying DCS Amanda server %s:%d (%d of %d)",
1692 host.Data(), port, iServ+1, nServers));
1d172743 1693
1694 TMap* aliasMap = 0;
1695 TMap* dpMap = 0;
26b1866e 1696
1d172743 1697 if (fConfig->GetDCSAliases(fCurrentDetector, iServ)->GetEntries() > 0)
2c15234c 1698 {
5ed060b1 1699 Log(fCurrentDetector, Form("Querying %d DCS aliases", fConfig->GetDCSAliases(fCurrentDetector, iServ)->GetEntries()));
1d172743 1700 aliasMap = GetValueSet(host, port,
542b6cc8 1701 fConfig->GetDCSAliases(fCurrentDetector, iServ),
1702 kAlias, multiSplit);
1d172743 1703 if (!aliasMap)
1704 {
1705 Log(fCurrentDetector,
1706 Form("ProcessCurrentDetector -"
675f64cd 1707 " Error retrieving DCS aliases from server %s."
1708 " Sending mail to DCS experts!", host.Data()));
1d172743 1709 UpdateShuttleStatus(AliShuttleStatus::kDCSError);
675f64cd 1710
339adafa 1711 if (!SendMail(kDCSEMail))
fb40ead4 1712 Log("SHUTTLE", Form("ProcessCurrentDetector - "
95d4aaef 1713 "Could not send mail to DCS experts!"));
675f64cd 1714
a038aa70 1715 delete dcsMap;
ffa25f20 1716 return 0;
1d172743 1717 }
2c15234c 1718 }
a038aa70 1719
1d172743 1720 if (fConfig->GetDCSDataPoints(fCurrentDetector, iServ)->GetEntries() > 0)
a038aa70 1721 {
5ed060b1 1722 Log(fCurrentDetector, Form("Querying %d DCS data points", fConfig->GetDCSDataPoints(fCurrentDetector, iServ)->GetEntries()));
1d172743 1723 dpMap = GetValueSet(host, port,
542b6cc8 1724 fConfig->GetDCSDataPoints(fCurrentDetector, iServ),
1725 kDP, multiSplit);
1d172743 1726 if (!dpMap)
1727 {
1728 Log(fCurrentDetector,
1729 Form("ProcessCurrentDetector -"
675f64cd 1730 " Error retrieving DCS data points from server %s."
1731 " Sending mail to DCS experts!", host.Data()));
1d172743 1732 UpdateShuttleStatus(AliShuttleStatus::kDCSError);
675f64cd 1733
339adafa 1734 if (!SendMail(kDCSEMail))
fb40ead4 1735 Log("SHUTTLE", Form("ProcessCurrentDetector - "
95d4aaef 1736 "Could not send mail to DCS experts!"));
675f64cd 1737
1d172743 1738 if (aliasMap) delete aliasMap;
1739 delete dcsMap;
ffa25f20 1740 return 0;
1d172743 1741 }
a038aa70 1742 }
1d172743 1743
1744 // merge aliasMap and dpMap into dcsMap
1745 if(aliasMap) {
1746 TIter iter(aliasMap);
a038aa70 1747 TObjString* key = 0;
1748 while ((key = (TObjString*) iter.Next()))
1d172743 1749 dcsMap->Add(key, aliasMap->GetValue(key->String()));
1750
1751 aliasMap->SetOwner(kFALSE);
1752 delete aliasMap;
1753 }
1754
1755 if(dpMap) {
1756 TIter iter(dpMap);
1757 TObjString* key = 0;
1758 while ((key = (TObjString*) iter.Next()))
1759 dcsMap->Add(key, dpMap->GetValue(key->String()));
1760
1761 dpMap->SetOwner(kFALSE);
1762 delete dpMap;
a038aa70 1763 }
73abe331 1764 }
1765 }
dc25836b 1766
b1d18693 1767 // save map into file, to help debugging in case of preprocessor error
fbc112e3 1768 TFile* f = TFile::Open("DCSMap.root","recreate");
b1d18693 1769 f->cd();
1770 dcsMap->Write("DCSMap", TObject::kSingleKey);
1771 f->Close();
fbc112e3 1772 delete f;
b1d18693 1773
2bb7b766 1774 // DCS Archive DB processing successful. Call Preprocessor!
85a80aa9 1775 UpdateShuttleStatus(AliShuttleStatus::kPPStarted);
a7160fe9 1776
339adafa 1777 fFXSError = -1; // this variable is kTRUE after ::Process if an FXS error occured
8da81210 1778
a038aa70 1779 UInt_t returnValue = aPreprocessor->Process(dcsMap);
8da81210 1780
339adafa 1781 if (fFXSError!=-1) {
8da81210 1782 UpdateShuttleStatus(AliShuttleStatus::kFXSError);
339adafa 1783 SendMail(kFXSEMail, fFXSError);
8da81210 1784 dcsMap->DeleteAll();
1785 delete dcsMap;
ffa25f20 1786 return 0;
8da81210 1787 }
b948db8d 1788
3301427a 1789 if (returnValue > 0) // Preprocessor error!
1790 {
c88ad5db 1791 Log(fCurrentDetector, Form("ProcessCurrentDetector - "
1792 "Preprocessor failed. Process returned %d.", returnValue));
cb343cfd 1793 UpdateShuttleStatus(AliShuttleStatus::kPPError);
a038aa70 1794 dcsMap->DeleteAll();
1795 delete dcsMap;
ffa25f20 1796 return 0;
9827400b 1797 }
1798
1799 // preprocessor ok!
1800 UpdateShuttleStatus(AliShuttleStatus::kPPDone);
1801 Log(fCurrentDetector, Form("ProcessCurrentDetector - %s preprocessor returned success",
1802 fCurrentDetector.Data()));
b948db8d 1803
a038aa70 1804 dcsMap->DeleteAll();
1805 delete dcsMap;
b948db8d 1806
ffa25f20 1807 return 1;
2bb7b766 1808}
1809
1810//______________________________________________________________________________________________
fbc112e3 1811void AliShuttle::CountOpenRuns()
1812{
1813 // Query DAQ's Shuttle logbook and sends the number of open runs to ML
1814
982278d6 1815 SendAlive();
1816
fbc112e3 1817 // check connection, in case connect
1818 if (!Connect(3))
1819 return;
1820
1821 TString sqlQuery;
1822 sqlQuery = Form("select count(*) from %s where shuttle_done=0", fConfig->GetShuttlelbTable());
1823
1824 TSQLResult* aResult = fServer[3]->Query(sqlQuery);
1825 if (!aResult) {
1826 AliError(Form("Can't execute query <%s>!", sqlQuery.Data()));
1827 return;
1828 }
1829
1830 AliDebug(2,Form("Query = %s", sqlQuery.Data()));
1831
1832 if (aResult->GetRowCount() == 0) {
1833 AliError(Form("No result for query %s received", sqlQuery.Data()));
1834 return;
1835 }
1836
1837 if (aResult->GetFieldCount() != 1) {
1838 AliError(Form("Invalid field count for query %s received", sqlQuery.Data()));
1839 return;
1840 }
1841
1842 TSQLRow* aRow = aResult->Next();
1843 if (!aRow) {
1844 AliError(Form("Could not receive result of query %s", sqlQuery.Data()));
1845 return;
1846 }
1847
1848 TString result(aRow->GetField(0), aRow->GetFieldLength(0));
1849 Int_t count = result.Atoi();
1850
1851 Log("SHUTTLE", Form("%d unprocessed runs", count));
1852
1853 delete aRow;
1854 delete aResult;
1855
1856 TMonaLisaValue mlStatus("SHUTTLE_openruns", count);
1857
1858 TList mlList;
1859 mlList.Add(&mlStatus);
1860
1861 fMonaLisa->SendParameters(&mlList, "__PROCESSINGINFO__");
1862}
1863
1864//______________________________________________________________________________________________
2bb7b766 1865Bool_t AliShuttle::QueryShuttleLogbook(const char* whereClause,
1866 TObjArray& entries)
1867{
9827400b 1868 // Query DAQ's Shuttle logbook and fills detector status object.
1869 // Call QueryRunParameters to query DAQ logbook for run parameters.
1870 //
2bb7b766 1871
fc5a4708 1872 entries.SetOwner(1);
1873
2bb7b766 1874 // check connection, in case connect
fbc112e3 1875 if (!Connect(3)) return kFALSE;
2bb7b766 1876
1877 TString sqlQuery;
441b0e9c 1878 sqlQuery = Form("select * from %s %s order by run", fConfig->GetShuttlelbTable(), whereClause);
2bb7b766 1879
be48e3ea 1880 TSQLResult* aResult = fServer[3]->Query(sqlQuery);
2bb7b766 1881 if (!aResult) {
1882 AliError(Form("Can't execute query <%s>!", sqlQuery.Data()));
1883 return kFALSE;
1884 }
1885
fc5a4708 1886 AliDebug(2,Form("Query = %s", sqlQuery.Data()));
1887
2bb7b766 1888 if(aResult->GetRowCount() == 0) {
c88ad5db 1889 Log("SHUTTLE", "No entries in Shuttle Logbook match request");
9827400b 1890 delete aResult;
1891 return kTRUE;
2bb7b766 1892 }
1893
1894 // TODO Check field count!
db99d43e 1895 const UInt_t nCols = 23;
2bb7b766 1896 if (aResult->GetFieldCount() != (Int_t) nCols) {
c88ad5db 1897 Log("SHUTTLE", "Invalid SQL result field number!");
2bb7b766 1898 delete aResult;
1899 return kFALSE;
1900 }
1901
2bb7b766 1902 TSQLRow* aRow;
1903 while ((aRow = aResult->Next())) {
1904 TString runString(aRow->GetField(0), aRow->GetFieldLength(0));
1905 Int_t run = runString.Atoi();
1906
eba76848 1907 AliShuttleLogbookEntry *entry = QueryRunParameters(run);
1908 if (!entry)
1909 continue;
2bb7b766 1910
1911 // loop on detectors
eba76848 1912 for(UInt_t ii = 0; ii < nCols; ii++)
1913 entry->SetDetectorStatus(aResult->GetFieldName(ii), aRow->GetField(ii));
2bb7b766 1914
eba76848 1915 entries.AddLast(entry);
2bb7b766 1916 delete aRow;
1917 }
1918
2bb7b766 1919 delete aResult;
1920 return kTRUE;
1921}
1922
1923//______________________________________________________________________________________________
eba76848 1924AliShuttleLogbookEntry* AliShuttle::QueryRunParameters(Int_t run)
2bb7b766 1925{
eba76848 1926 //
1927 // Retrieve run parameters written in the DAQ logbook and sets them into AliShuttleLogbookEntry object
1928 //
2bb7b766 1929
1930 // check connection, in case connect
be48e3ea 1931 if (!Connect(3))
eba76848 1932 return 0;
2bb7b766 1933
1934 TString sqlQuery;
2c15234c 1935 sqlQuery.Form("select * from %s where run=%d", fConfig->GetDAQlbTable(), run);
2bb7b766 1936
be48e3ea 1937 TSQLResult* aResult = fServer[3]->Query(sqlQuery);
2bb7b766 1938 if (!aResult) {
c88ad5db 1939 Log("SHUTTLE", Form("Can't execute query <%s>!", sqlQuery.Data()));
eba76848 1940 return 0;
2bb7b766 1941 }
1942
eba76848 1943 if (aResult->GetRowCount() == 0) {
2bb7b766 1944 Log("SHUTTLE", Form("QueryRunParameters - No entry in DAQ Logbook for run %d. Skipping", run));
1945 delete aResult;
eba76848 1946 return 0;
2bb7b766 1947 }
1948
eba76848 1949 if (aResult->GetRowCount() > 1) {
c88ad5db 1950 Log("SHUTTLE", Form("QueryRunParameters - UNEXPECTED: "
1951 "more than one entry in DAQ Logbook for run %d!", run));
2bb7b766 1952 delete aResult;
eba76848 1953 return 0;
2bb7b766 1954 }
1955
eba76848 1956 TSQLRow* aRow = aResult->Next();
1957 if (!aRow)
1958 {
c88ad5db 1959 Log("SHUTTLE", Form("QueryRunParameters - Could not retrieve row for run %d. Skipping", run));
eba76848 1960 delete aResult;
1961 return 0;
1962 }
2bb7b766 1963
eba76848 1964 AliShuttleLogbookEntry* entry = new AliShuttleLogbookEntry(run);
2bb7b766 1965
eba76848 1966 for (Int_t ii = 0; ii < aResult->GetFieldCount(); ii++)
1967 entry->SetRunParameter(aResult->GetFieldName(ii), aRow->GetField(ii));
2bb7b766 1968
c9db68c7 1969 delete aRow;
1970 delete aResult;
1971
eba76848 1972 UInt_t startTime = entry->GetStartTime();
1973 UInt_t endTime = entry->GetEndTime();
89bcf3b4 1974 Bool_t ecsSuccess = entry->GetECSSuccess();
c9db68c7 1975
b0e53b15 1976 TString totEventsStr = entry->GetRunParameter("totalEvents");
1977 Int_t totEvents = totEventsStr.Atoi();
c9db68c7 1978
af93e5d7 1979 UInt_t now = time(0);
1980 // TODO make this a configuration parameter
2eb98462 1981 Int_t dcsDelay = fConfig->GetDCSDelay()+fConfig->GetDCSQueryOffset();
af93e5d7 1982
8e828d39 1983 // runs are accepted if they have ecsSuccess set or more than 1 event
af93e5d7 1984 if (startTime != 0 && endTime != 0 && endTime > startTime && (totEvents > 1 || ecsSuccess) && (endTime < now - dcsDelay))
89bcf3b4 1985 {
8e828d39 1986 if (ecsSuccess == kFALSE)
1987 Log("SHUTTLE", Form("Processing run %d although in status ECS failure, Reason: %s", run, entry->GetRunParameter("eor_reason")));
1988 return entry;
89bcf3b4 1989 }
8e828d39 1990
1991 Bool_t skip = kFALSE;
1992
f8523731 1993 if (endTime != 0 && endTime >= now - dcsDelay)
89bcf3b4 1994 {
f8523731 1995 Log("SHUTTLE", Form("Skipping run %d for now, because DCS buffer time is not yet expired", run));
af93e5d7 1996 }
f8523731 1997 else if (totEvents <= 1)
af93e5d7 1998 {
f8523731 1999 Log("SHUTTLE", Form("QueryRunParameters - Run %d has 1 event or less - Skipping!", run));
2000 skip = kTRUE;
af93e5d7 2001 }
89bcf3b4 2002 else
2003 {
2004 Log("SHUTTLE", Form("QueryRunParameters - Invalid parameters for Run %d: "
2005 "startTime = %d, endTime = %d. Skipping (Shuttle won't be marked as DONE)!",
2006 run, startTime, endTime));
2007 }
8e828d39 2008
2009 if (skip)
2010 {
2011 Log("SHUTTLE", Form("Marking SHUTTLE skipped for run %d", run));
2012 fLogbookEntry = entry;
2013 if (!UpdateShuttleLogbook("shuttle_skipped"))
2014 {
2015 AliError(Form("Could not update logbook for run %d !", run));
2016 }
bd23d48b 2017 if (!UpdateTableSkippedCase("ALL"))
2018 {
2019 AliError(Form("Could not update FXS tables for run %d !", run));
2020 }
8e828d39 2021 fLogbookEntry = 0;
2022 }
c9db68c7 2023
2024 delete entry;
2025 return 0;
2bb7b766 2026}
2027
2028//______________________________________________________________________________________________
a038aa70 2029TMap* AliShuttle::GetValueSet(const char* host, Int_t port, const TSeqCollection* entries,
542b6cc8 2030 DCSType type, Int_t multiSplit)
a038aa70 2031{
2032 // Retrieve all "entry" data points from the DCS server
2033 // host, port: TSocket connection parameters
2034 // entries: list of name of the alias or data point
2035 // type: kAlias or kDP
2036 // returns TMap of values, 0 when failure
542b6cc8 2037
2038 AliDCSClient client(host, port, fTimeout, fRetries, multiSplit);
b41b252a 2039
a038aa70 2040 TMap* result = 0;
b41b252a 2041 if (type == kAlias)
a038aa70 2042 {
5f3fe74e 2043 //result = client.GetAliasValues(entries, GetCurrentStartTime()-offset,
2044 // GetCurrentEndTime()+offset);
eff647d5 2045 result = client.GetAliasValues(entries, GetStartTimeDCSQuery(),
2046 GetEndTimeDCSQuery());
b41b252a 2047 }
2048 else if (type == kDP)
2049 {
5f3fe74e 2050 //result = client.GetDPValues(entries, GetCurrentStartTime()-offset,
2051 // GetCurrentEndTime()+offset);
eff647d5 2052 result = client.GetDPValues(entries, GetStartTimeDCSQuery(),
2053 GetEndTimeDCSQuery());
b41b252a 2054 }
a038aa70 2055
b41b252a 2056 if (result == 0)
2057 {
2058 Log(fCurrentDetector.Data(), Form("GetValueSet - Can't get entries! Reason: %s",
1790d4b7 2059 client.GetErrorString(client.GetResultErrorCode())));
2060 if (client.GetResultErrorCode() == AliDCSClient::fgkServerError)
2061 Log(fCurrentDetector.Data(), Form("GetValueSet - Server error code: %s",
2062 client.GetServerError().Data()));
a038aa70 2063
b41b252a 2064 return 0;
a038aa70 2065 }
b41b252a 2066
a038aa70 2067 return result;
2068}
b41b252a 2069
a038aa70 2070//______________________________________________________________________________________________
57f50b3c 2071const char* AliShuttle::GetFile(Int_t system, const char* detector,
2072 const char* id, const char* source)
b948db8d 2073{
9827400b 2074 // Get calibration file from file exchange servers
2075 // First queris the FXS database for the file name, using the run, detector, id and source info
2076 // then calls RetrieveFile(filename) for actual copy to local disk
2077 // run: current run being processed (given by Logbook entry fLogbookEntry)
2078 // detector: the Preprocessor name
2079 // id: provided as a parameter by the Preprocessor
2080 // source: provided by the Preprocessor through GetFileSources function
2081
2082 // check if test mode should simulate a FXS error
2083 if (fTestMode & kErrorFXSFiles)
2084 {
2085 Log(detector, Form("GetFile - In TESTMODE - Simulating error while connecting to %s FXS", GetSystemName(system)));
2086 return 0;
2087 }
2088
57f50b3c 2089 // check connection, in case connect
9d733021 2090 if (!Connect(system))
eba76848 2091 {
9d733021 2092 Log(detector, Form("GetFile - Couldn't connect to %s FXS database", GetSystemName(system)));
339adafa 2093 fFXSError = system;
57f50b3c 2094 return 0;
2095 }
2096
2097 // Query preparation
9d733021 2098 TString sourceName(source);
d386d623 2099 Int_t nFields = 3;
2100 TString sqlQueryStart = Form("select filePath,size,fileChecksum from %s where",
2101 fConfig->GetFXSdbTable(system));
2102 TString whereClause = Form("run=%d and detector=\"%s\" and fileId=\"%s\"",
2103 GetCurrentRun(), detector, id);
2104
9d733021 2105 if (system == kDAQ)
2106 {
d386d623 2107 whereClause += Form(" and DAQsource=\"%s\"", source);
57f50b3c 2108 }
9d733021 2109 else if (system == kDCS)
eba76848 2110 {
9d733021 2111 sourceName="none";
57f50b3c 2112 }
9d733021 2113 else if (system == kHLT)
9e080f92 2114 {
d386d623 2115 whereClause += Form(" and DDLnumbers=\"%s\"", source);
9e080f92 2116 }
2117
9e080f92 2118 TString sqlQuery = Form("%s %s", sqlQueryStart.Data(), whereClause.Data());
2119
2120 AliDebug(2, Form("SQL query: \n%s",sqlQuery.Data()));
2121
2122 // Query execution
2123 TSQLResult* aResult = 0;
9d733021 2124 aResult = dynamic_cast<TSQLResult*> (fServer[system]->Query(sqlQuery));
9e080f92 2125 if (!aResult) {
8da81210 2126 Log(detector, Form("GetFile - Can't execute SQL query to %s database for: id = %s, source = %s",
9d733021 2127 GetSystemName(system), id, sourceName.Data()));
339adafa 2128 fFXSError = system;
9e080f92 2129 return 0;
2130 }
2131
8da81210 2132 if (aResult->GetRowCount() == 0)
9e080f92 2133 {
2134 Log(detector,
8da81210 2135 Form("GetFile - No entry in %s FXS db for: id = %s, source = %s",
9d733021 2136 GetSystemName(system), id, sourceName.Data()));
9e080f92 2137 delete aResult;
2138 return 0;
2139 }
2bb7b766 2140
9e080f92 2141 if (aResult->GetRowCount() > 1) {
2142 Log(detector,
8da81210 2143 Form("GetFile - More than one entry in %s FXS db for: id = %s, source = %s",
9d733021 2144 GetSystemName(system), id, sourceName.Data()));
339adafa 2145 fFXSError = system;
9e080f92 2146 delete aResult;
2147 return 0;
2148 }
2149
9d733021 2150 if (aResult->GetFieldCount() != nFields) {
9e080f92 2151 Log(detector,
9d733021 2152 Form("GetFileName - Wrong field count in %s FXS db for: id = %s, source = %s",
2153 GetSystemName(system), id, sourceName.Data()));
339adafa 2154 fFXSError = system;
9e080f92 2155 delete aResult;
2156 return 0;
2157 }
2158
2159 TSQLRow* aRow = dynamic_cast<TSQLRow*> (aResult->Next());
2160
2161 if (!aRow){
8da81210 2162 Log(detector, Form("GetFile - Empty set result in %s FXS db from query: id = %s, source = %s",
9d733021 2163 GetSystemName(system), id, sourceName.Data()));
339adafa 2164 fFXSError = system;
9e080f92 2165 delete aResult;
2166 return 0;
2167 }
2168
2169 TString filePath(aRow->GetField(0), aRow->GetFieldLength(0));
2170 TString fileSize(aRow->GetField(1), aRow->GetFieldLength(1));
d386d623 2171 TString fileChecksum(aRow->GetField(2), aRow->GetFieldLength(2));
9e080f92 2172
2173 delete aResult;
2174 delete aRow;
2175
d386d623 2176 AliDebug(2, Form("filePath = %s; size = %s, fileChecksum = %s",
2177 filePath.Data(), fileSize.Data(), fileChecksum.Data()));
9e080f92 2178
9e080f92 2179 // retrieved file is renamed to make it unique
3ebbf3c4 2180 Int_t dir_lev1 = GetCurrentRun()/10000;
2181 TString localFileName = Form("%s/%d/%d/%s_process/%s_%s_%d_%s_%s.shuttle",
2182 GetShuttleTempDir(), dir_lev1, GetCurrentRun(), detector,
d524ade6 2183 GetSystemName(system), detector, GetCurrentRun(),
2184 id, sourceName.Data());
3ebbf3c4 2185 Log("SHUTTLE",Form("file from FXS = %s",localFileName.Data()));
9d733021 2186
9e080f92 2187
9d733021 2188 // file retrieval from FXS
4b95672b 2189 UInt_t nRetries = 0;
2190 UInt_t maxRetries = 3;
2191 Bool_t result = kFALSE;
2192
2193 // copy!! if successful TSystem::Exec returns 0
8da81210 2194 while (nRetries++ < maxRetries) {
4b95672b 2195 AliDebug(2, Form("Trying to copy file. Retry # %d", nRetries));
2196 result = RetrieveFile(system, filePath.Data(), localFileName.Data());
8da81210 2197 if (!result)
4b95672b 2198 {
8da81210 2199 Log(detector, Form("GetFile - Copy of file %s from %s FXS failed",
9d733021 2200 filePath.Data(), GetSystemName(system)));
4b95672b 2201 continue;
4f0749a8 2202 }
9e080f92 2203
c98417a1 2204 if (fileSize.Length()>0)
2205 {
2206 // compare filesize of local file with the one stored in the FXS DB
926cbc0e 2207 Long_t size = -1;
2208 Int_t sizeComp = gSystem->GetPathInfo(localFileName.Data(), 0, &size, 0, 0);
c98417a1 2209
926cbc0e 2210 if (sizeComp != 0 || size != fileSize.Atoi())
c98417a1 2211 {
8da81210 2212 Log(detector, Form("GetFile - size of file %s does not match with local copy!",
c98417a1 2213 filePath.Data()));
2214 result = kFALSE;
2215 continue;
2216 }
2217
2218 } else {
2219 Log(fCurrentDetector, Form("GetFile - size of file %s not set in %s database, skipping comparison",
2220 filePath.Data(), GetSystemName(system)));
2221 }
2222
d386d623 2223 if (fileChecksum.Length()>0)
4b95672b 2224 {
2225 // compare md5sum of local file with the one stored in the FXS DB
95d4aaef 2226 if(fileChecksum.Contains(' ')) fileChecksum.Resize(fileChecksum.First(' '));
339adafa 2227 Int_t md5Comp = gSystem->Exec(Form("md5sum %s |grep %s > /dev/null 2> /dev/null",
d524ade6 2228 localFileName.Data(), fileChecksum.Data()));
9e080f92 2229
4b95672b 2230 if (md5Comp != 0)
2231 {
8da81210 2232 Log(detector, Form("GetFile - md5sum of file %s does not match with local copy!",
4b95672b 2233 filePath.Data()));
2234 result = kFALSE;
2235 continue;
2236 }
d386d623 2237 } else {
2238 Log(fCurrentDetector, Form("GetFile - md5sum of file %s not set in %s database, skipping comparison",
2239 filePath.Data(), GetSystemName(system)));
9d733021 2240 }
4b95672b 2241 if (result) break;
9e080f92 2242 }
2243
8da81210 2244 if (!result)
2245 {
339adafa 2246 fFXSError = system;
8da81210 2247 return 0;
2248 }
4b95672b 2249
9d733021 2250 fFXSCalled[system]=kTRUE;
2251 TObjString *fileParams = new TObjString(Form("%s#!?!#%s", id, sourceName.Data()));
2252 fFXSlist[system].Add(fileParams);
9e080f92 2253
675f64cd 2254 static TString staticLocalFileName;
2255 staticLocalFileName.Form("%s", localFileName.Data());
2256
c88ad5db 2257 Log(fCurrentDetector, Form("GetFile - Retrieved file with id %s and "
2258 "source %s from %s to %s", id, source,
d524ade6 2259 GetSystemName(system), localFileName.Data()));
675f64cd 2260
d524ade6 2261 return staticLocalFileName.Data();
2bb7b766 2262}
2263
2264//______________________________________________________________________________________________
9d733021 2265Bool_t AliShuttle::RetrieveFile(UInt_t system, const char* fxsFileName, const char* localFileName)
9e080f92 2266{
9827400b 2267 //
2268 // Copies file from FXS to local Shuttle machine
2269 //
2bb7b766 2270
9e080f92 2271 // check temp directory: trying to cd to temp; if it does not exist, create it
d524ade6 2272 AliDebug(2, Form("Copy file %s from %s FXS into %s",
2273 GetSystemName(system), fxsFileName, localFileName));
2274
2275 TString tmpDir(localFileName);
2276
2277 tmpDir = tmpDir(0,tmpDir.Last('/'));
9e080f92 2278
d524ade6 2279 Int_t noDir = gSystem->GetPathInfo(tmpDir.Data(), 0, (Long64_t*) 0, 0, 0);
2280 if (noDir) // temp dir does not exists!
2281 {
2282 if (gSystem->mkdir(tmpDir.Data(), 1))
2283 {
2284 Log(fCurrentDetector.Data(), "RetrieveFile - could not make temp directory!!");
9e080f92 2285 return kFALSE;
2286 }
9e080f92 2287 }
2288
339adafa 2289 TString command = Form("scp -oPort=%d -2 %s@%s:%s/%s %s",
9d733021 2290 fConfig->GetFXSPort(system),
2291 fConfig->GetFXSUser(system),
2292 fConfig->GetFXSHost(system),
339adafa 2293 fConfig->GetFXSBaseFolder(system),
9d733021 2294 fxsFileName,
9e080f92 2295 localFileName);
2296
2297 AliDebug(2, Form("%s",command.Data()));
2298
4b95672b 2299 Bool_t result = (gSystem->Exec(command.Data()) == 0);
9e080f92 2300
4b95672b 2301 return result;
9e080f92 2302}
2303
2304//______________________________________________________________________________________________
9d733021 2305TList* AliShuttle::GetFileSources(Int_t system, const char* detector, const char* id)
2306{
9827400b 2307 //
2308 // Get sources producing the condition file Id from file exchange servers
4a33bdd9 2309 // if id is NULL all sources are returned (distinct)
9827400b 2310 //
1bcd28db 2311
8884d71e 2312 if (id)
2313 {
2314 Log(detector, Form("GetFileSources - Querying %s FXS for files with id %s produced by %s", GetSystemName(system), id, detector));
2315 } else {
2316 Log(detector, Form("GetFileSources - Querying %s FXS for files produced by %s", GetSystemName(system), detector));
2317 }
9827400b 2318
2319 // check if test mode should simulate a FXS error
2320 if (fTestMode & kErrorFXSSources)
2321 {
2322 Log(detector, Form("GetFileSources - In TESTMODE - Simulating error while connecting to %s FXS", GetSystemName(system)));
2323 return 0;
2324 }
2325
9d733021 2326 if (system == kDCS)
2327 {
c88ad5db 2328 Log(detector, "GetFileSources - WARNING: DCS system has only one source of data!");
6297b37d 2329 TList *list = new TList();
2330 list->SetOwner(1);
2331 list->Add(new TObjString(" "));
2332 return list;
9d733021 2333 }
9e080f92 2334
2335 // check connection, in case connect
9d733021 2336 if (!Connect(system))
2337 {
4a33bdd9 2338 Log(detector, Form("GetFileSources - Couldn't connect to %s FXS database", GetSystemName(system)));
339adafa 2339 fFXSError = system;
9d733021 2340 return NULL;
9e080f92 2341 }
2342
9d733021 2343 TString sourceName = 0;
2344 if (system == kDAQ)
2345 {
2346 sourceName = "DAQsource";
2347 } else if (system == kHLT)
2348 {
2349 sourceName = "DDLnumbers";
2350 }
2351
4a33bdd9 2352 TString sqlQueryStart = Form("select distinct %s from %s where", sourceName.Data(), fConfig->GetFXSdbTable(system));
2353 TString whereClause = Form("run=%d and detector=\"%s\"",
2354 GetCurrentRun(), detector);
2355 if (id)
2356 whereClause += Form(" and fileId=\"%s\"", id);
9e080f92 2357 TString sqlQuery = Form("%s %s", sqlQueryStart.Data(), whereClause.Data());
2358
2359 AliDebug(2, Form("SQL query: \n%s",sqlQuery.Data()));
2360
2361 // Query execution
2362 TSQLResult* aResult;
9d733021 2363 aResult = fServer[system]->Query(sqlQuery);
9e080f92 2364 if (!aResult) {
9d733021 2365 Log(detector, Form("GetFileSources - Can't execute SQL query to %s database for id: %s",
2366 GetSystemName(system), id));
339adafa 2367 fFXSError = system;
9e080f92 2368 return 0;
2369 }
2370
86aa42c3 2371 TList *list = new TList();
2372 list->SetOwner(1);
2373
9d733021 2374 if (aResult->GetRowCount() == 0)
2375 {
9e080f92 2376 Log(detector,
9d733021 2377 Form("GetFileSources - No entry in %s FXS table for id: %s", GetSystemName(system), id));
9e080f92 2378 delete aResult;
86aa42c3 2379 return list;
9e080f92 2380 }
2381
1bcd28db 2382 Log(detector, Form("GetFileSources - Found %d sources", aResult->GetRowCount()));
9e080f92 2383
1bcd28db 2384 TSQLRow* aRow;
9d733021 2385 while ((aRow = aResult->Next()))
2386 {
9e080f92 2387
9d733021 2388 TString source(aRow->GetField(0), aRow->GetFieldLength(0));
2389 AliDebug(2, Form("%s = %s", sourceName.Data(), source.Data()));
2390 list->Add(new TObjString(source));
9e080f92 2391 delete aRow;
2392 }
9d733021 2393
9e080f92 2394 delete aResult;
2395
2396 return list;
2bb7b766 2397}
2398
2399//______________________________________________________________________________________________
4a33bdd9 2400TList* AliShuttle::GetFileIDs(Int_t system, const char* detector, const char* source)
2401{
2402 //
2403 // Get all ids of condition files produced by a given source from file exchange servers
2404 //
2405
1bcd28db 2406 Log(detector, Form("GetFileIDs - Retrieving ids with source %s with %s", source, GetSystemName(system)));
2407
4a33bdd9 2408 // check if test mode should simulate a FXS error
2409 if (fTestMode & kErrorFXSSources)
2410 {
2411 Log(detector, Form("GetFileIDs - In TESTMODE - Simulating error while connecting to %s FXS", GetSystemName(system)));
2412 return 0;
2413 }
2414
2415 // check connection, in case connect
2416 if (!Connect(system))
2417 {
2418 Log(detector, Form("GetFileIDs - Couldn't connect to %s FXS database", GetSystemName(system)));
2419 return NULL;
2420 }
2421
2422 TString sourceName = 0;
2423 if (system == kDAQ)
2424 {
2425 sourceName = "DAQsource";
2426 } else if (system == kHLT)
2427 {
2428 sourceName = "DDLnumbers";
2429 }
2430
2431 TString sqlQueryStart = Form("select fileId from %s where", fConfig->GetFXSdbTable(system));
2432 TString whereClause = Form("run=%d and detector=\"%s\"",
2433 GetCurrentRun(), detector);
2434 if (sourceName.Length() > 0 && source)
2435 whereClause += Form(" and %s=\"%s\"", sourceName.Data(), source);
2436 TString sqlQuery = Form("%s %s", sqlQueryStart.Data(), whereClause.Data());
2437
2438 AliDebug(2, Form("SQL query: \n%s",sqlQuery.Data()));
2439
2440 // Query execution
2441 TSQLResult* aResult;
2442 aResult = fServer[system]->Query(sqlQuery);
2443 if (!aResult) {
2444 Log(detector, Form("GetFileIDs - Can't execute SQL query to %s database for source: %s",
2445 GetSystemName(system), source));
2446 return 0;
2447 }
2448
2449 TList *list = new TList();
2450 list->SetOwner(1);
2451
2452 if (aResult->GetRowCount() == 0)
2453 {
2454 Log(detector,
2455 Form("GetFileIDs - No entry in %s FXS table for source: %s", GetSystemName(system), source));
2456 delete aResult;
2457 return list;
2458 }
2459
1bcd28db 2460 Log(detector, Form("GetFileIDs - Found %d ids", aResult->GetRowCount()));
2461
4a33bdd9 2462 TSQLRow* aRow;
2463
2464 while ((aRow = aResult->Next()))
2465 {
2466
2467 TString id(aRow->GetField(0), aRow->GetFieldLength(0));
2468 AliDebug(2, Form("fileId = %s", id.Data()));
2469 list->Add(new TObjString(id));
2470 delete aRow;
2471 }
2472
2473 delete aResult;
2474
2475 return list;
2476}
2477
2478//______________________________________________________________________________________________
9d733021 2479Bool_t AliShuttle::Connect(Int_t system)
2bb7b766 2480{
9827400b 2481 // Connect to MySQL Server of the system's FXS MySQL databases
2482 // DAQ Logbook, Shuttle Logbook and DAQ FXS db are on the same host
2483 //
57f50b3c 2484
9d733021 2485 // check connection: if already connected return
2486 if(fServer[system] && fServer[system]->IsConnected()) return kTRUE;
57f50b3c 2487
9d733021 2488 TString dbHost, dbUser, dbPass, dbName;
57f50b3c 2489
9d733021 2490 if (system < 3) // FXS db servers
2491 {
2492 dbHost = Form("mysql://%s:%d", fConfig->GetFXSdbHost(system), fConfig->GetFXSdbPort(system));
2493 dbUser = fConfig->GetFXSdbUser(system);
2494 dbPass = fConfig->GetFXSdbPass(system);
2495 dbName = fConfig->GetFXSdbName(system);
2496 } else { // Run & Shuttle logbook servers
2497 // TODO Will the Shuttle logbook server be the same as the Run logbook server ???
2498 dbHost = Form("mysql://%s:%d", fConfig->GetDAQlbHost(), fConfig->GetDAQlbPort());
2499 dbUser = fConfig->GetDAQlbUser();
2500 dbPass = fConfig->GetDAQlbPass();
2501 dbName = fConfig->GetDAQlbDB();
2502 }
57f50b3c 2503
9d733021 2504 fServer[system] = TSQLServer::Connect(dbHost.Data(), dbUser.Data(), dbPass.Data());
2505 if (!fServer[system] || !fServer[system]->IsConnected()) {
2506 if(system < 3)
2507 {
2508 AliError(Form("Can't establish connection to FXS database for %s",
2509 AliShuttleInterface::GetSystemName(system)));
2510 } else {
2511 AliError("Can't establish connection to Run logbook.");
57f50b3c 2512 }
9d733021 2513 if(fServer[system]) delete fServer[system];
2514 return kFALSE;
2bb7b766 2515 }
57f50b3c 2516
9d733021 2517 // Get tables
2518 TSQLResult* aResult=0;
2519 switch(system){
2520 case kDAQ:
2521 aResult = fServer[kDAQ]->GetTables(dbName.Data());
2522 break;
2523 case kDCS:
2524 aResult = fServer[kDCS]->GetTables(dbName.Data());
2525 break;
2526 case kHLT:
2527 aResult = fServer[kHLT]->GetTables(dbName.Data());
2528 break;
2529 default:
2530 aResult = fServer[3]->GetTables(dbName.Data());
2531 break;
2532 }
2533
2534 delete aResult;
2bb7b766 2535 return kTRUE;
2536}
57f50b3c 2537
9e080f92 2538//______________________________________________________________________________________________
9d733021 2539Bool_t AliShuttle::UpdateTable()
9e080f92 2540{
9827400b 2541 //
2542 // Update FXS table filling time_processed field in all rows corresponding to current run and detector
2543 //
9e080f92 2544
9d733021 2545 Bool_t result = kTRUE;
9e080f92 2546
9d733021 2547 for (UInt_t system=0; system<3; system++)
2548 {
2549 if(!fFXSCalled[system]) continue;
9e080f92 2550
9d733021 2551 // check connection, in case connect
2552 if (!Connect(system))
2553 {
2554 Log(fCurrentDetector, Form("UpdateTable - Couldn't connect to %s FXS database", GetSystemName(system)));
2555 result = kFALSE;
2556 continue;
9e080f92 2557 }
9e080f92 2558
9d733021 2559 TTimeStamp now; // now
2560
2561 // Loop on FXS list entries
2562 TIter iter(&fFXSlist[system]);
2563 TObjString *aFXSentry=0;
2564 while ((aFXSentry = dynamic_cast<TObjString*> (iter.Next())))
2565 {
2566 TString aFXSentrystr = aFXSentry->String();
2567 TObjArray *aFXSarray = aFXSentrystr.Tokenize("#!?!#");
2568 if (!aFXSarray || aFXSarray->GetEntries() != 2 )
2569 {
2570 Log(fCurrentDetector, Form("UpdateTable - error updating %s FXS entry. Check string: <%s>",
2571 GetSystemName(system), aFXSentrystr.Data()));
2572 if(aFXSarray) delete aFXSarray;
2573 result = kFALSE;
2574 continue;
2575 }
2576 const char* fileId = ((TObjString*) aFXSarray->At(0))->GetName();
2577 const char* source = ((TObjString*) aFXSarray->At(1))->GetName();
2578
2579 TString whereClause;
2580 if (system == kDAQ)
2581 {
2582 whereClause = Form("where run=%d and detector=\"%s\" and fileId=\"%s\" and DAQsource=\"%s\";",
2583 GetCurrentRun(), fCurrentDetector.Data(), fileId, source);
2584 }
2585 else if (system == kDCS)
2586 {
2587 whereClause = Form("where run=%d and detector=\"%s\" and fileId=\"%s\";",
2588 GetCurrentRun(), fCurrentDetector.Data(), fileId);
2589 }
2590 else if (system == kHLT)
2591 {
2592 whereClause = Form("where run=%d and detector=\"%s\" and fileId=\"%s\" and DDLnumbers=\"%s\";",
2593 GetCurrentRun(), fCurrentDetector.Data(), fileId, source);
2594 }
2595
2596 delete aFXSarray;
9e080f92 2597
9d733021 2598 TString sqlQuery = Form("update %s set time_processed=%d %s", fConfig->GetFXSdbTable(system),
2599 now.GetSec(), whereClause.Data());
9e080f92 2600
9d733021 2601 AliDebug(2, Form("SQL query: \n%s",sqlQuery.Data()));
9e080f92 2602
9d733021 2603 // Query execution
2604 TSQLResult* aResult;
2605 aResult = dynamic_cast<TSQLResult*> (fServer[system]->Query(sqlQuery));
2606 if (!aResult)
2607 {
2608 Log(fCurrentDetector, Form("UpdateTable - %s db: can't execute SQL query <%s>",
2609 GetSystemName(system), sqlQuery.Data()));
2610 result = kFALSE;
2611 continue;
2612 }
2613 delete aResult;
9e080f92 2614 }
9e080f92 2615 }
2616
9d733021 2617 return result;
9e080f92 2618}
57f50b3c 2619
bd23d48b 2620//_______________________________________________________________________________
2621Bool_t AliShuttle::UpdateTableSkippedCase(const char* detector)
2622{
2623 //
2624 // Update FXS table filling time_processed field in all rows corresponding to current run and detector
2625 // if detector = "ALL" update all detectors
2626 //
2627
2628 Bool_t result = kTRUE;
2629
2630 for (UInt_t system=0; system<3; system++)
2631 {
2632
2633 // check connection, in case connect
2634 if (!Connect(system))
2635 {
2636 Log(fCurrentDetector, Form("UpdateTableSkippedCase - Couldn't connect to %s FXS database", GetSystemName(system)));
2637 result = kFALSE;
2638 continue;
2639 }
2640
2641 TTimeStamp now; // now
2642
2643 // Loop on FXS list entries
2644 TIter iter(&fFXSlist[system]);
2645
2646 TString whereClause;
2647 if (detector == "ALL") whereClause = Form("where run=%d and time_processed IS NULL;",GetCurrentRun());
2648 else whereClause = Form("where run=%d and detector=\"%s\" and time_processed IS NULL;",GetCurrentRun(), detector);
2649
2650 Log("SHUTTLE",Form(" whereClause = %s ",whereClause.Data()));
2651
2652 TString sqlQuery = Form("update %s set time_processed=%d %s", fConfig->GetFXSdbTable(system),
2653 now.GetSec(), whereClause.Data());
2654
2655 AliDebug(2, Form("SQL query: \n%s",sqlQuery.Data()));
2656
2657 // Query execution
2658 TSQLResult* aResult;
2659 aResult = dynamic_cast<TSQLResult*> (fServer[system]->Query(sqlQuery));
2660 if (!aResult)
2661 {
2662 Log("SHUTTLE", Form("UpdateTableSkippedCase - %s db: can't execute SQL query <%s>",
2663 GetSystemName(system), sqlQuery.Data()));
2664 result = kFALSE;
2665 continue;
2666 }
2667 delete aResult;
2668
2669 }
2670
2671 return result;
2672}
2bb7b766 2673//______________________________________________________________________________________________
3301427a 2674Bool_t AliShuttle::UpdateTableFailCase()
2675{
9827400b 2676 // Update FXS table filling time_processed field in all rows corresponding to current run and detector
2677 // this is called in case the preprocessor is declared failed for the current run, because
2678 // the fields are updated only in case of success
3301427a 2679
2680 Bool_t result = kTRUE;
2681
2682 for (UInt_t system=0; system<3; system++)
2683 {
2684 // check connection, in case connect
2685 if (!Connect(system))
2686 {
2687 Log(fCurrentDetector, Form("UpdateTableFailCase - Couldn't connect to %s FXS database",
2688 GetSystemName(system)));
2689 result = kFALSE;
2690 continue;
2691 }
2692
2693 TTimeStamp now; // now
2694
2695 // Loop on FXS list entries
2696
2697 TString whereClause = Form("where run=%d and detector=\"%s\";",
2698 GetCurrentRun(), fCurrentDetector.Data());
2699
2700
2701 TString sqlQuery = Form("update %s set time_processed=%d %s", fConfig->GetFXSdbTable(system),
2702 now.GetSec(), whereClause.Data());
2703
2704 AliDebug(2, Form("SQL query: \n%s",sqlQuery.Data()));
2705
2706 // Query execution
2707 TSQLResult* aResult;
2708 aResult = dynamic_cast<TSQLResult*> (fServer[system]->Query(sqlQuery));
2709 if (!aResult)
2710 {
2711 Log(fCurrentDetector, Form("UpdateTableFailCase - %s db: can't execute SQL query <%s>",
2712 GetSystemName(system), sqlQuery.Data()));
2713 result = kFALSE;
2714 continue;
2715 }
2716 delete aResult;
2717 }
2718
2719 return result;
2720}
2721
2722//______________________________________________________________________________________________
2bb7b766 2723Bool_t AliShuttle::UpdateShuttleLogbook(const char* detector, const char* status)
2724{
e7f62f16 2725 //
2726 // Update Shuttle logbook filling detector or shuttle_done column
2727 // ex. of usage: UpdateShuttleLogbook("PHOS", "DONE") or UpdateShuttleLogbook("shuttle_done")
2728 //
57f50b3c 2729
2bb7b766 2730 // check connection, in case connect
be48e3ea 2731 if(!Connect(3)){
2bb7b766 2732 Log("SHUTTLE", "UpdateShuttleLogbook - Couldn't connect to DAQ Logbook.");
2733 return kFALSE;
57f50b3c 2734 }
2735
2bb7b766 2736 TString detName(detector);
2737 TString setClause;
8e828d39 2738 if (detName == "shuttle_done" || detName == "shuttle_skipped")
e7f62f16 2739 {
2bb7b766 2740 setClause = "set shuttle_done=1";
8e828d39 2741
ee6f7523 2742 if (detName == "shuttle_done")
b0e53b15 2743 {
8e828d39 2744 if (TouchFile() != kTRUE)
2156c755 2745 {
2746 SendMLRunInfo("Pending");
503ff24f 2747 return kFALSE;
2156c755 2748 }
8e828d39 2749
2750 SendMLRunInfo("Done");
b0e53b15 2751 }
8e828d39 2752 else
2753 SendMLRunInfo("Skipped");
2754 }
2755 else {
2bb7b766 2756 TString statusStr(status);
2757 if(statusStr.Contains("done", TString::kIgnoreCase) ||
2758 statusStr.Contains("failed", TString::kIgnoreCase)){
eba76848 2759 setClause = Form("set %s=\"%s\"", detector, status);
2bb7b766 2760 } else {
2761 Log("SHUTTLE",
2762 Form("UpdateShuttleLogbook - Invalid status <%s> for detector %s",
2763 status, detector));
2764 return kFALSE;
2765 }
2766 }
57f50b3c 2767
2bb7b766 2768 TString whereClause = Form("where run=%d", GetCurrentRun());
2769
441b0e9c 2770 TString sqlQuery = Form("update %s %s %s",
2771 fConfig->GetShuttlelbTable(), setClause.Data(), whereClause.Data());
57f50b3c 2772
2bb7b766 2773 AliDebug(2, Form("SQL query: \n%s",sqlQuery.Data()));
2774
2775 // Query execution
2776 TSQLResult* aResult;
be48e3ea 2777 aResult = dynamic_cast<TSQLResult*> (fServer[3]->Query(sqlQuery));
2bb7b766 2778 if (!aResult) {
2779 Log("SHUTTLE", Form("UpdateShuttleLogbook - Can't execute query <%s>", sqlQuery.Data()));
2780 return kFALSE;
57f50b3c 2781 }
2bb7b766 2782 delete aResult;
57f50b3c 2783
2784 return kTRUE;
2785}
2786
2787//______________________________________________________________________________________________
2bb7b766 2788Int_t AliShuttle::GetCurrentRun() const
2789{
9827400b 2790 //
2791 // Get current run from logbook entry
2792 //
57f50b3c 2793
2bb7b766 2794 return fLogbookEntry ? fLogbookEntry->GetRun() : -1;
57f50b3c 2795}
2796
2797//______________________________________________________________________________________________
2bb7b766 2798UInt_t AliShuttle::GetCurrentStartTime() const
2799{
9827400b 2800 //
2801 // get current start time
2802 //
57f50b3c 2803
2bb7b766 2804 return fLogbookEntry ? fLogbookEntry->GetStartTime() : 0;
57f50b3c 2805}
2806
2807//______________________________________________________________________________________________
2bb7b766 2808UInt_t AliShuttle::GetCurrentEndTime() const
2809{
9827400b 2810 //
2811 // get current end time from logbook entry
2812 //
57f50b3c 2813
2bb7b766 2814 return fLogbookEntry ? fLogbookEntry->GetEndTime() : 0;
57f50b3c 2815}
5f3fe74e 2816//______________________________________________________________________________________________
2817UInt_t AliShuttle::GetCurrentTimeCreated() const
2818{
2819 //
2820 // get current end time from logbook entry
2821 //
2822
2823 return fLogbookEntry ? fLogbookEntry->GetTimeCreated() : 0;
2824}
57f50b3c 2825
2826//______________________________________________________________________________________________
675f64cd 2827UInt_t AliShuttle::GetCurrentYear() const
2828{
2829 //
2830 // Get current year from logbook entry
2831 //
2832
2833 if (!fLogbookEntry) return 0;
2834
2835 TTimeStamp startTime(GetCurrentStartTime());
2836 TString year = Form("%d",startTime.GetDate());
2837 year = year(0,4);
2838
2839 return year.Atoi();
2840}
2841
2842//______________________________________________________________________________________________
2843const char* AliShuttle::GetLHCPeriod() const
2844{
2845 //
2846 // Get current LHC period from logbook entry
2847 //
2848
2849 if (!fLogbookEntry) return 0;
2850
2851 return fLogbookEntry->GetRunParameter("LHCperiod");
2852}
2853
2854//______________________________________________________________________________________________
b948db8d 2855void AliShuttle::Log(const char* detector, const char* message)
2856{
9827400b 2857 //
2858 // Fill log string with a message
2859 //
7d4cf768 2860
3ebbf3c4 2861 TString logRunDir = GetShuttleLogDir();
2862 if (GetCurrentRun() >=0) {
2863 Int_t logDir_lev1 = GetCurrentRun()/10000;
2864 logRunDir += Form("/%d/%d", logDir_lev1, GetCurrentRun());
2865 }
7d4cf768 2866 void* dir = gSystem->OpenDirectory(logRunDir.Data());
84090f85 2867 if (dir == NULL) {
7d4cf768 2868 if (gSystem->mkdir(logRunDir.Data(), kTRUE)) {
36c99a6a 2869 AliError(Form("Can't open directory <%s>", GetShuttleLogDir()));
84090f85 2870 return;
2871 }
b948db8d 2872
84090f85 2873 } else {
2874 gSystem->FreeDirectory(dir);
2875 }
b948db8d 2876
0b025a52 2877 TString toLog = Form("%s UTC (%d): %s - ", TTimeStamp(time(0)).AsString("s"), getpid(), detector);
e7f62f16 2878 if (GetCurrentRun() >= 0)
2879 toLog += Form("run %d - ", GetCurrentRun());
2bb7b766 2880 toLog += Form("%s", message);
2881
84090f85 2882 AliInfo(toLog.Data());
ffa29e93 2883
2884 // if we redirect the log output already to the file, leave here
2885 if (fOutputRedirected && strcmp(detector, "SHUTTLE") != 0)
2886 return;
b948db8d 2887
ffa29e93 2888 TString fileName = GetLogFileName(detector);
e7f62f16 2889
84090f85 2890 gSystem->ExpandPathName(fileName);
2891
2892 ofstream logFile;
2893 logFile.open(fileName, ofstream::out | ofstream::app);
2894
2895 if (!logFile.is_open()) {
2896 AliError(Form("Could not open file %s", fileName.Data()));
2897 return;
2898 }
7bfb2090 2899
84090f85 2900 logFile << toLog.Data() << "\n";
b948db8d 2901
84090f85 2902 logFile.close();
b948db8d 2903}
2bb7b766 2904
2bb7b766 2905//______________________________________________________________________________________________
ffa29e93 2906TString AliShuttle::GetLogFileName(const char* detector) const
2907{
2908 //
2909 // returns the name of the log file for a given sub detector
2910 //
2911
2912 TString fileName;
2913
2914 if (GetCurrentRun() >= 0)
7d4cf768 2915 {
3ebbf3c4 2916 Int_t logDir_lev1 = GetCurrentRun()/10000;
2917 fileName.Form("%s/%d/%d/%s.log", GetShuttleLogDir(), logDir_lev1, GetCurrentRun(),
2918 detector);
7d4cf768 2919 } else {
ffa29e93 2920 fileName.Form("%s/%s.log", GetShuttleLogDir(), detector);
7d4cf768 2921 }
ffa29e93 2922
2923 return fileName;
2924}
2925
2926//______________________________________________________________________________________________
ee6f7523 2927void AliShuttle::SendAlive()
2928{
2929 // sends alive message to ML
2930
2931 TMonaLisaText mlStatus("SHUTTLE_status", "Alive");
2932
2933 TList mlList;
2934 mlList.Add(&mlStatus);
2935
2936 fMonaLisa->SendParameters(&mlList, "__PROCESSINGINFO__");
2937}
2938
2939//______________________________________________________________________________________________
2bb7b766 2940Bool_t AliShuttle::Collect(Int_t run)
2941{
9827400b 2942 //
2943 // Collects conditions data for all UNPROCESSED run written to DAQ LogBook in case of run = -1 (default)
2944 // If a dedicated run is given this run is processed
2945 //
2946 // In operational mode, this is the Shuttle function triggered by the EOR signal.
2947 //
2bb7b766 2948
eba76848 2949 if (run == -1)
2950 Log("SHUTTLE","Collect - Shuttle called. Collecting conditions data for unprocessed runs");
2951 else
2952 Log("SHUTTLE", Form("Collect - Shuttle called. Collecting conditions data for run %d", run));
cb343cfd 2953
2954 SetLastAction("Starting");
2bb7b766 2955
ee6f7523 2956 // create ML instance
2957 if (!fMonaLisa)
2958 fMonaLisa = new TMonaLisaWriter(fConfig->GetMonitorHost(), fConfig->GetMonitorTable());
2959
fbc112e3 2960 CountOpenRuns();
ee6f7523 2961
2bb7b766 2962 TString whereClause("where shuttle_done=0");
eba76848 2963 if (run != -1)
2964 whereClause += Form(" and run=%d", run);
2bb7b766 2965
2966 TObjArray shuttleLogbookEntries;
be48e3ea 2967 if (!QueryShuttleLogbook(whereClause, shuttleLogbookEntries))
2968 {
cb343cfd 2969 Log("SHUTTLE", "Collect - Can't retrieve entries from Shuttle logbook");
2bb7b766 2970 return kFALSE;
2971 }
2972
9e080f92 2973 if (shuttleLogbookEntries.GetEntries() == 0)
2974 {
2975 if (run == -1)
2976 Log("SHUTTLE","Collect - Found no UNPROCESSED runs in Shuttle logbook");
2977 else
2978 Log("SHUTTLE", Form("Collect - Run %d is already DONE "
2979 "or it does not exist in Shuttle logbook", run));
2980 return kTRUE;
2981 }
2982
be48e3ea 2983 for (UInt_t iDet=0; iDet<NDetectors(); iDet++)
2984 fFirstUnprocessed[iDet] = kTRUE;
2985
fc5a4708 2986 if (run != -1)
be48e3ea 2987 {
2988 // query Shuttle logbook for earlier runs, check if some detectors are unprocessed,
2989 // flag them into fFirstUnprocessed array
e3ae8d33 2990 TString whereClauseBis(Form("where shuttle_done=0 and run < %d", run));
be48e3ea 2991 TObjArray tmpLogbookEntries;
e3ae8d33 2992 if (!QueryShuttleLogbook(whereClauseBis, tmpLogbookEntries))
be48e3ea 2993 {
2994 Log("SHUTTLE", "Collect - Can't retrieve entries from Shuttle logbook");
2995 return kFALSE;
2996 }
2997
2998 TIter iter(&tmpLogbookEntries);
2999 AliShuttleLogbookEntry* anEntry = 0;
3000 while ((anEntry = dynamic_cast<AliShuttleLogbookEntry*> (iter.Next())))
3001 {
3002 for (UInt_t iDet=0; iDet<NDetectors(); iDet++)
3003 {
3004 if (anEntry->GetDetectorStatus(iDet) == AliShuttleLogbookEntry::kUnprocessed)
3005 {
3006 AliDebug(2, Form("Run %d: setting %s as \"not first time unprocessed\"",
3007 anEntry->GetRun(), GetDetName(iDet)));
3008 fFirstUnprocessed[iDet] = kFALSE;
3009 }
3010 }
3011
3012 }
3013
3014 }
3015
3016 if (!RetrieveConditionsData(shuttleLogbookEntries))
3017 {
cb343cfd 3018 Log("SHUTTLE", "Collect - Process of at least one run failed");
a3ca69fe 3019 CountOpenRuns();
2bb7b766 3020 return kFALSE;
3021 }
3022
36c99a6a 3023 Log("SHUTTLE", "Collect - Requested run(s) successfully processed");
a3ca69fe 3024 CountOpenRuns();
eba76848 3025 return kTRUE;
2bb7b766 3026}
3027
2bb7b766 3028//______________________________________________________________________________________________
3029Bool_t AliShuttle::RetrieveConditionsData(const TObjArray& dateEntries)
3030{
9827400b 3031 //
3032 // Retrieve conditions data for all runs that aren't processed yet
3033 //
2bb7b766 3034
3035 Bool_t hasError = kFALSE;
3036
3037 TIter iter(&dateEntries);
3038 AliShuttleLogbookEntry* anEntry;
3039
3040 while ((anEntry = (AliShuttleLogbookEntry*) iter.Next())){
3041 if (!Process(anEntry)){
3042 hasError = kTRUE;
3043 }
4b95672b 3044
3045 // clean SHUTTLE temp directory
d524ade6 3046 //TString filename = Form("%s/*.shuttle", GetShuttleTempDir());
3047 //RemoveFile(filename.Data());
2bb7b766 3048 }
3049
3050 return hasError == kFALSE;
3051}
cb343cfd 3052
3053//______________________________________________________________________________________________
3054ULong_t AliShuttle::GetTimeOfLastAction() const
3055{
9827400b 3056 //
3057 // Gets time of last action
3058 //
3059
cb343cfd 3060 ULong_t tmp;
36c99a6a 3061
cb343cfd 3062 fMonitoringMutex->Lock();
be48e3ea 3063
cb343cfd 3064 tmp = fLastActionTime;
36c99a6a 3065
cb343cfd 3066 fMonitoringMutex->UnLock();
36c99a6a 3067
cb343cfd 3068 return tmp;
3069}
3070
3071//______________________________________________________________________________________________
3072const TString AliShuttle::GetLastAction() const
3073{
9827400b 3074 //
cb343cfd 3075 // returns a string description of the last action
9827400b 3076 //
cb343cfd 3077
3078 TString tmp;
36c99a6a 3079
cb343cfd 3080 fMonitoringMutex->Lock();
3081
3082 tmp = fLastAction;
3083
3084 fMonitoringMutex->UnLock();
3085
36c99a6a 3086 return tmp;
cb343cfd 3087}
3088
3089//______________________________________________________________________________________________
3090void AliShuttle::SetLastAction(const char* action)
3091{
9827400b 3092 //
cb343cfd 3093 // updates the monitoring variables
9827400b 3094 //
36c99a6a 3095
cb343cfd 3096 fMonitoringMutex->Lock();
36c99a6a 3097
cb343cfd 3098 fLastAction = action;
3099 fLastActionTime = time(0);
3100