]> git.uio.no Git - u/mrichter/AliRoot.git/blame - SHUTTLE/AliShuttle.cxx
git://git.uio.no / u / mrichter / AliRoot.git / blame
? search:
summary | shortlog | log | commit | commitdiff | tree
blob | blame (incremental) | history | HEAD
adding DCS buffer delay
[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);
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 {
279 Log("StoreOCDB - SHUTTLE","Storing Run Metadata file ...");
280 resultMetadata = CopyFilesToGrid("metadata");
281 }
9827400b 282
5e993b6f 283 Int_t storeResult = 0;
284
285 if (resultCDB < 0 || resultRef < 0 || resultRefFiles == kFALSE || resultMetadata == kFALSE)
286 storeResult = -1;
287 else if (resultCDB > 0 || resultRef > 0)
288 storeResult = 1;
289
290 if (storeResult < 0)
291 {
292 Log("SHUTTLE",
293 Form("\t\t\t****** run %d - %s: STORAGE ERROR ******",
294 GetCurrentRun(), fCurrentDetector.Data()));
295 UpdateShuttleStatus(AliShuttleStatus::kStoreError);
296 }
297 else if (storeResult > 0)
298 {
299 Log("SHUTTLE",
300 Form("\t\t\t****** run %d - %s: STORAGE DELAYED ******",
301 GetCurrentRun(), fCurrentDetector.Data()));
302 UpdateShuttleStatus(AliShuttleStatus::kStoreDelayed);
303 }
304 else if (storeResult == 0)
305 {
306 Log("SHUTTLE",
307 Form("\t\t\t****** run %d - %s: DONE ******",
308 GetCurrentRun(), fCurrentDetector.Data()));
309 UpdateShuttleStatus(AliShuttleStatus::kDone);
310 UpdateShuttleLogbook(fCurrentDetector, "DONE");
311 }
312
313 return (storeResult == 0);
3301427a 314}
315
316//______________________________________________________________________________________________
5e993b6f 317Int_t AliShuttle::StoreOCDB(const TString& gridURI)
3301427a 318{
319 //
320 // Called by StoreOCDB(), performs actual storage to the main OCDB and reference storages (Grid)
321 //
5e993b6f 322 // Return code:
323 // -2 initialization error
324 // -1 storage error
325 // 0 success
326 // 1 storage delayed (e.g. previous unprocessed runs)
327 //
3301427a 328
329 TObjArray* gridIds=0;
330
331 Bool_t result = kTRUE;
5e993b6f 332 Bool_t delayed = kFALSE;
3301427a 333
334 const char* type = 0;
335 TString localURI;
336 if(gridURI == fgkMainCDB) {
337 type = "OCDB";
338 localURI = fgkLocalCDB;
339 } else if(gridURI == fgkMainRefStorage) {
340 type = "reference";
341 localURI = fgkLocalRefStorage;
342 } else {
343 AliError(Form("Invalid storage URI: %s", gridURI.Data()));
5e993b6f 344 return -2;
3301427a 345 }
346
347 AliCDBManager* man = AliCDBManager::Instance();
348
349 AliCDBStorage *gridSto = man->GetStorage(gridURI);
350 if(!gridSto) {
351 Log("SHUTTLE",
352 Form("StoreOCDB - cannot activate main %s storage", type));
5e993b6f 353 return -2;
3301427a 354 }
355
356 gridIds = gridSto->GetQueryCDBList();
357
358 // get objects previously stored in local CDB
359 AliCDBStorage *localSto = man->GetStorage(localURI);
360 if(!localSto) {
361 Log("SHUTTLE",
362 Form("StoreOCDB - cannot activate local %s storage", type));
5e993b6f 363 return -2;
3301427a 364 }
365 AliCDBPath aPath(GetOfflineDetName(fCurrentDetector.Data()),"*","*");
366 // Local objects were stored with current run as Grid version!
367 TList* localEntries = localSto->GetAll(aPath.GetPath(), GetCurrentRun(), GetCurrentRun());
368 localEntries->SetOwner(1);
369
370 // loop on local stored objects
371 TIter localIter(localEntries);
372 AliCDBEntry *aLocEntry = 0;
373 while((aLocEntry = dynamic_cast<AliCDBEntry*> (localIter.Next()))){
374 aLocEntry->SetOwner(1);
375 AliCDBId aLocId = aLocEntry->GetId();
376 aLocEntry->SetVersion(-1);
377 aLocEntry->SetSubVersion(-1);
6c0c9948 378
379 Log(fCurrentDetector.Data(), Form("Attempting to store %s", aLocId.ToString().Data()));
3301427a 380
381 // If local object is valid up to infinity we store it only if it is
382 // the first unprocessed run!
383 if (aLocId.GetLastRun() == AliCDBRunRange::Infinity() &&
384 !fFirstUnprocessed[GetDetPos(fCurrentDetector)])
385 {
386 Log("SHUTTLE", Form("StoreOCDB - %s: object %s has validity infinite but "
387 "there are previous unprocessed runs!",
388 fCurrentDetector.Data(), aLocId.GetPath().Data()));
5e993b6f 389 Log(fCurrentDetector.Data(), Form("StoreOCDB - %s: object %s has validity infinite but "
390 "there are previous unprocessed runs!",
391 fCurrentDetector.Data(), aLocId.GetPath().Data()));
392 delayed = kTRUE;
3301427a 393 continue;
394 }
395
396 // loop on Grid valid Id's
397 Bool_t store = kTRUE;
398 TIter gridIter(gridIds);
399 AliCDBId* aGridId = 0;
6c0c9948 400 while ((aGridId = dynamic_cast<AliCDBId*> (gridIter.Next()))) {
401 if (aGridId->GetPath() != aLocId.GetPath())
402 continue;
3301427a 403 // skip all objects valid up to infinity
6c0c9948 404 if (aGridId->GetLastRun() == AliCDBRunRange::Infinity())
405 continue;
406
3301427a 407 // if we get here, it means there's already some more recent object stored on Grid!
6c0c9948 408 Log(fCurrentDetector.Data(),
409 Form("StoreOCDB - A more recent object already exists in %s storage: <%s>",
410 type, aGridId->ToString().Data()));
411
3301427a 412 store = kFALSE;
413 break;
414 }
415
6c0c9948 416 Bool_t storeOk = kFALSE;
417 if (store)
418 {
419 Log(fCurrentDetector.Data(), Form("Prechecks succeeded. Ready to store %s", aLocId.ToString().Data()));
420 storeOk = gridSto->Put(aLocEntry);
421 if (storeOk) {
3301427a 422 Log("SHUTTLE",
6c0c9948 423 Form("StoreOCDB - Object <%s> successfully put into %s storage",
424 aLocId.ToString().Data(), type));
2d9019b4 425 Log(fCurrentDetector.Data(),
426 Form("StoreOCDB - Object <%s> successfully put into %s storage",
6c0c9948 427 aLocId.ToString().Data(), type));
428 } else {
429 Log("SHUTTLE",
430 Form("StoreOCDB - Grid %s storage of object <%s> failed",
431 type, aLocId.ToString().Data()));
432 Log(fCurrentDetector.Data(),
433 Form("StoreOCDB - Grid %s storage of object <%s> failed",
434 type, aLocId.ToString().Data()));
435 result = kFALSE;
3301427a 436 }
6c0c9948 437 }
438
439 if (!store || storeOk) {
440 // removing local file...
3301427a 441 TString filename;
442 localSto->IdToFilename(aLocId, filename);
c88ad5db 443 Log("SHUTTLE", Form("StoreOCDB - Removing local file %s", filename.Data()));
3301427a 444 RemoveFile(filename.Data());
b948db8d 445 }
446 }
3301427a 447 localEntries->Clear();
2bb7b766 448
5e993b6f 449 Int_t returnCode = 0;
450
451 if (result == kFALSE)
452 returnCode = -1;
453 else if (delayed != kFALSE)
454 returnCode = 1;
455
456 Log("SHUTTLE", Form("StoreOCDB - Returning with %d (result = %d, delayed = %d)", returnCode, result, delayed));
457 Log(fCurrentDetector.Data(), Form("StoreOCDB - Returning with %d (result = %d, delayed = %d)", returnCode, result, delayed));
458
459 return returnCode;
3301427a 460}
461
546242fb 462//______________________________________________________________________________________________
463Bool_t AliShuttle::CleanReferenceStorage(const char* detector)
464{
2d9019b4 465 // clears the directory used to store reference files of a given subdetector
546242fb 466
467 AliCDBManager* man = AliCDBManager::Instance();
468 AliCDBStorage* sto = man->GetStorage(fgkLocalRefStorage);
2d9019b4 469 TString localBaseFolder = sto->GetBaseFolder();
470
471 TString targetDir = GetRefFilePrefix(localBaseFolder.Data(), detector);
472
d524ade6 473 Log("SHUTTLE", Form("CleanReferenceStorage - Cleaning %s", targetDir.Data()));
2d9019b4 474
475 TString begin;
476 begin.Form("%d_", GetCurrentRun());
477
478 TSystemDirectory* baseDir = new TSystemDirectory("/", targetDir);
479 if (!baseDir)
480 return kTRUE;
481
482 TList* dirList = baseDir->GetListOfFiles();
483 delete baseDir;
484
485 if (!dirList) return kTRUE;
486
487 if (dirList->GetEntries() < 3)
488 {
489 delete dirList;
490 return kTRUE;
491 }
492
493 Int_t nDirs = 0, nDel = 0;
494 TIter dirIter(dirList);
495 TSystemFile* entry = 0;
546242fb 496
2d9019b4 497 Bool_t success = kTRUE;
546242fb 498
2d9019b4 499 while ((entry = dynamic_cast<TSystemFile*> (dirIter.Next())))
500 {
501 if (entry->IsDirectory())
502 continue;
503
504 TString fileName(entry->GetName());
505 if (!fileName.BeginsWith(begin))
506 continue;
507
508 nDirs++;
509
510 // delete file
511 Int_t result = gSystem->Unlink(fileName.Data());
512
513 if (result)
514 {
d524ade6 515 Log("SHUTTLE", Form("CleanReferenceStorage - Could not delete file %s!", fileName.Data()));
2d9019b4 516 success = kFALSE;
517 } else {
518 nDel++;
519 }
520 }
521
522 if(nDirs > 0)
523 Log("SHUTTLE", Form("CleanReferenceStorage - %d (over %d) reference files in folder %s were deleted.",
524 nDel, nDirs, targetDir.Data()));
525
526
527 delete dirList;
528 return success;
529
530
531
532
533
546242fb 534
535 Int_t result = gSystem->GetPathInfo(targetDir, 0, (Long64_t*) 0, 0, 0);
536 if (result == 0)
537 {
538 // delete directory
d524ade6 539 result = gSystem->Exec(Form("rm -rf %s", targetDir.Data()));
546242fb 540 if (result != 0)
541 {
d524ade6 542 Log("SHUTTLE", Form("CleanReferenceStorage - Could not clean directory %s", targetDir.Data()));
546242fb 543 return kFALSE;
544 }
545 }
546
547 result = gSystem->mkdir(targetDir, kTRUE);
548 if (result != 0)
549 {
c88ad5db 550 Log("SHUTTLE", Form("CleanReferenceStorage - Error creating base directory %s", targetDir.Data()));
546242fb 551 return kFALSE;
552 }
553
554 return kTRUE;
555}
556
9827400b 557//______________________________________________________________________________________________
558Bool_t AliShuttle::StoreReferenceFile(const char* detector, const char* localFile, const char* gridFileName)
559{
560 //
3c2a21c8 561 // Stores reference file directly (without opening it). This function stores the file locally.
9827400b 562 //
3c2a21c8 563 // The file is stored under the following location:
564 // <base folder of local reference storage>/<DET>/<RUN#>_<gridFileName>
565 // where <gridFileName> is the second parameter given to the function
566 //
9827400b 567
568 if (fTestMode & kErrorStorage)
569 {
570 Log(fCurrentDetector, "StoreReferenceFile - In TESTMODE - Simulating error while storing locally");
571 return kFALSE;
572 }
573
574 AliCDBManager* man = AliCDBManager::Instance();
575 AliCDBStorage* sto = man->GetStorage(fgkLocalRefStorage);
576
577 TString localBaseFolder = sto->GetBaseFolder();
578
d524ade6 579 TString target = GetRefFilePrefix(localBaseFolder.Data(), detector);
580 target.Append(Form("/%d_%s", GetCurrentRun(), gridFileName));
9827400b 581
d524ade6 582 return CopyFileLocally(localFile, target);
c88ad5db 583}
584
585//______________________________________________________________________________________________
586Bool_t AliShuttle::StoreRunMetadataFile(const char* localFile, const char* gridFileName)
587{
588 //
589 // Stores Run metadata file to the Grid, in the run folder
590 //
591 // Only GRP can call this function.
592
593 if (fTestMode & kErrorStorage)
594 {
595 Log(fCurrentDetector, "StoreRunMetaDataFile - In TESTMODE - Simulating error while storing locally");
596 return kFALSE;
597 }
598
599 AliCDBManager* man = AliCDBManager::Instance();
600 AliCDBStorage* sto = man->GetStorage(fgkLocalRefStorage);
601
602 TString localBaseFolder = sto->GetBaseFolder();
603
604 // Build Run level folder
955d2abc 605 // folder = /alice/data/year/lhcPeriod/runNb/raw
c88ad5db 606
c88ad5db 607
675f64cd 608 TString lhcPeriod = GetLHCPeriod();
c88ad5db 609 if (lhcPeriod.Length() == 0)
610 {
611 Log("SHUTTLE","StoreRunMetaDataFile - LHCPeriod not found in logbook!");
612 return 0;
613 }
a1c5723b 614
615 // TODO partitions with one detector only write data into LHCperiod_DET
2694f253 616 TString partition = GetRunParameter("detector");
a1c5723b 617
618 if (partition.Length() > 0 && partition != "ALICE")
619 {
620 lhcPeriod.Append(Form("_%s", partition.Data()));
621 Log(fCurrentDetector, Form("Run data tags merged file will be written in %s",
622 lhcPeriod.Data()));
623 }
8884d71e 624
b8cf5e0f 625 TString target = Form("%s/GRP/RunMetadata%s%d/%s/%09d/raw/%s",
626 localBaseFolder.Data(), fConfig->GetAlienPath(), GetCurrentYear(),
d524ade6 627 lhcPeriod.Data(), GetCurrentRun(), gridFileName);
c88ad5db 628
d524ade6 629 return CopyFileLocally(localFile, target);
c88ad5db 630}
631
632//______________________________________________________________________________________________
d524ade6 633Bool_t AliShuttle::CopyFileLocally(const char* localFile, const TString& target)
c88ad5db 634{
635 //
636 // Stores file locally. Called by StoreReferenceFile and StoreRunMetadataFile
d524ade6 637 // Files are temporarily stored in the local reference storage. When the preprocessor
638 // finishes, the Shuttle calls CopyFilesToGrid to transfer the files to AliEn
639 // (in reference or run level folders)
c88ad5db 640 //
641
d524ade6 642 TString targetDir(target(0, target.Last('/')));
643
644 //try to open base dir folder, if it does not exist
2d9019b4 645 void* dir = gSystem->OpenDirectory(targetDir.Data());
646 if (dir == NULL) {
647 if (gSystem->mkdir(targetDir.Data(), kTRUE)) {
a1c5723b 648 Log("SHUTTLE", Form("CopyFileLocally - Can't open directory <%s>", targetDir.Data()));
2d9019b4 649 return kFALSE;
650 }
651
652 } else {
653 gSystem->FreeDirectory(dir);
654 }
9827400b 655
7d43a416 656 Int_t result = 0;
657
658 result = gSystem->GetPathInfo(localFile, 0, (Long64_t*) 0, 0, 0);
9827400b 659 if (result)
660 {
a1c5723b 661 Log("SHUTTLE", Form("CopyFileLocally - %s does not exist", localFile));
546242fb 662 return kFALSE;
9827400b 663 }
546242fb 664
7d43a416 665 result = gSystem->GetPathInfo(target, 0, (Long64_t*) 0, 0, 0);
666 if (!result)
667 {
a1c5723b 668 Log("SHUTTLE", Form("CopyFileLocally - target file %s already exist, removing...", target.Data()));
7d43a416 669 if (gSystem->Unlink(target.Data()))
670 {
a1c5723b 671 Log("SHUTTLE", Form("CopyFileLocally - Could not remove existing target file %s!", target.Data()));
7d43a416 672 return kFALSE;
673 }
674 }
675
9827400b 676 result = gSystem->CopyFile(localFile, target);
677
678 if (result == 0)
679 {
a1c5723b 680 Log("SHUTTLE", Form("CopyFileLocally - File %s stored locally to %s", localFile, target.Data()));
9827400b 681 return kTRUE;
682 }
683 else
684 {
a1c5723b 685 Log("SHUTTLE", Form("CopyFileLocally - Could not store file %s to %s! Error code = %d",
546242fb 686 localFile, target.Data(), result));
9827400b 687 return kFALSE;
688 }
c88ad5db 689
690
691
9827400b 692}
693
694//______________________________________________________________________________________________
c88ad5db 695Bool_t AliShuttle::CopyFilesToGrid(const char* type)
9827400b 696{
697 //
c88ad5db 698 // Transfers local files to the Grid. Local files can be reference files
699 // or run metadata file (from GRP only).
9827400b 700 //
c88ad5db 701 // According to the type (ref, metadata) the files are stored under the following location:
702 // ref --> <base folder of reference storage>/<DET>/<RUN#>_<gridFileName>
703 // metadata --> <run data folder>/<MetadataFileName>
86aa42c3 704 //
c88ad5db 705
9827400b 706 AliCDBManager* man = AliCDBManager::Instance();
707 AliCDBStorage* sto = man->GetStorage(fgkLocalRefStorage);
708 if (!sto)
709 return kFALSE;
710 TString localBaseFolder = sto->GetBaseFolder();
9827400b 711
c88ad5db 712 TString dir;
713 TString alienDir;
9827400b 714 TString begin;
9827400b 715
c88ad5db 716 if (strcmp(type, "reference") == 0)
717 {
718 dir = GetRefFilePrefix(localBaseFolder.Data(), fCurrentDetector.Data());
719 AliCDBStorage* gridSto = man->GetStorage(fgkMainRefStorage);
720 if (!gridSto)
721 return kFALSE;
722 TString gridBaseFolder = gridSto->GetBaseFolder();
723 alienDir = GetRefFilePrefix(gridBaseFolder.Data(), fCurrentDetector.Data());
724 begin = Form("%d_", GetCurrentRun());
725 }
726 else if (strcmp(type, "metadata") == 0)
727 {
c88ad5db 728
675f64cd 729 TString lhcPeriod = GetLHCPeriod();
c88ad5db 730
731 if (lhcPeriod.Length() == 0)
732 {
733 Log("SHUTTLE","CopyFilesToGrid - LHCPeriod not found in logbook!");
734 return 0;
735 }
736
a1c5723b 737 // TODO partitions with one detector only write data into LHCperiod_DET
2694f253 738 TString partition = GetRunParameter("detector");
a1c5723b 739
740 if (partition.Length() > 0 && partition != "ALICE")
741 {
742 lhcPeriod.Append(Form("_%s", partition.Data()));
743 }
744
b8cf5e0f 745 dir = Form("%s/GRP/RunMetadata%s%d/%s/%09d/raw",
746 localBaseFolder.Data(), fConfig->GetAlienPath(), GetCurrentYear(),
c88ad5db 747 lhcPeriod.Data(), GetCurrentRun());
b8cf5e0f 748 alienDir = dir(dir.Index(fConfig->GetAlienPath()), dir.Length());
675f64cd 749
c88ad5db 750 begin = "";
751 }
752 else
753 {
754 Log("SHUTTLE", "CopyFilesToGrid - Unexpected: type label must be reference or metadata!");
755 return kFALSE;
756 }
757
9827400b 758 TSystemDirectory* baseDir = new TSystemDirectory("/", dir);
3d8bc902 759 if (!baseDir)
760 return kTRUE;
761
2d9019b4 762 TList* dirList = baseDir->GetListOfFiles();
763 delete baseDir;
764
765 if (!dirList) return kTRUE;
766
767 if (dirList->GetEntries() < 3)
3d8bc902 768 {
2d9019b4 769 delete dirList;
9827400b 770 return kTRUE;
3d8bc902 771 }
2d9019b4 772
546242fb 773 if (!gGrid)
774 {
c88ad5db 775 Log("SHUTTLE", "CopyFilesToGrid - Connection to Grid failed: Cannot continue!");
2d9019b4 776 delete dirList;
546242fb 777 return kFALSE;
778 }
779
2d9019b4 780 Int_t nDirs = 0, nTransfer = 0;
781 TIter dirIter(dirList);
782 TSystemFile* entry = 0;
783
9827400b 784 Bool_t success = kTRUE;
3d8bc902 785 Bool_t first = kTRUE;
9827400b 786
2d9019b4 787 while ((entry = dynamic_cast<TSystemFile*> (dirIter.Next())))
788 {
9827400b 789 if (entry->IsDirectory())
790 continue;
791
792 TString fileName(entry->GetName());
793 if (!fileName.BeginsWith(begin))
794 continue;
795
2d9019b4 796 nDirs++;
797
3d8bc902 798 if (first)
799 {
800 first = kFALSE;
c88ad5db 801 // check that folder exists, otherwise create it
3d8bc902 802 TGridResult* result = gGrid->Ls(alienDir.Data(), "a");
803
804 if (!result)
2d9019b4 805 {
806 delete dirList;
3d8bc902 807 return kFALSE;
2d9019b4 808 }
3d8bc902 809
546242fb 810 if (!result->GetFileName(1)) // TODO: It looks like element 0 is always 0!!
3d8bc902 811 {
675f64cd 812 // TODO It does not work currently! Bug in TAliEn::Mkdir
813 // TODO Manually fixed in local root v5-16-00
c88ad5db 814 if (!gGrid->Mkdir(alienDir.Data(),"-p",0))
3d8bc902 815 {
c88ad5db 816 Log("SHUTTLE", Form("CopyFilesToGrid - Cannot create directory %s",
3d8bc902 817 alienDir.Data()));
2d9019b4 818 delete dirList;
3d8bc902 819 return kFALSE;
546242fb 820 } else {
c88ad5db 821 Log("SHUTTLE",Form("CopyFilesToGrid - Folder %s created", alienDir.Data()));
3d8bc902 822 }
823
546242fb 824 } else {
c88ad5db 825 Log("SHUTTLE",Form("CopyFilesToGrid - Folder %s found", alienDir.Data()));
3d8bc902 826 }
827 }
828
9827400b 829 TString fullLocalPath;
830 fullLocalPath.Form("%s/%s", dir.Data(), fileName.Data());
831
832 TString fullGridPath;
833 fullGridPath.Form("alien://%s/%s", alienDir.Data(), fileName.Data());
834
a986b218 835 Bool_t result = TFile::Cp(fullLocalPath, fullGridPath);
9827400b 836
837 if (result)
838 {
c88ad5db 839 Log("SHUTTLE", Form("CopyFilesToGrid - Copying local file %s to %s succeeded!",
840 fullLocalPath.Data(), fullGridPath.Data()));
9827400b 841 RemoveFile(fullLocalPath);
2d9019b4 842 nTransfer++;
9827400b 843 }
844 else
845 {
c88ad5db 846 Log("SHUTTLE", Form("CopyFilesToGrid - Copying local file %s to %s FAILED!",
847 fullLocalPath.Data(), fullGridPath.Data()));
9827400b 848 success = kFALSE;
849 }
850 }
2d9019b4 851
c88ad5db 852 Log("SHUTTLE", Form("CopyFilesToGrid - %d (over %d) files in folder %s copied to Grid.",
853 nTransfer, nDirs, dir.Data()));
2d9019b4 854
855
856 delete dirList;
9827400b 857 return success;
858}
859
2d9019b4 860//______________________________________________________________________________________________
861const char* AliShuttle::GetRefFilePrefix(const char* base, const char* detector)
862{
863 //
864 // Get folder name of reference files
865 //
866
867 TString offDetStr(GetOfflineDetName(detector));
868 TString dir;
869 if (offDetStr == "ITS" || offDetStr == "MUON" || offDetStr == "PHOS")
870 {
871 dir.Form("%s/%s/%s", base, offDetStr.Data(), detector);
872 } else {
873 dir.Form("%s/%s", base, offDetStr.Data());
874 }
875
876 return dir.Data();
877
878
879}
c88ad5db 880
3301427a 881//______________________________________________________________________________________________
882void AliShuttle::CleanLocalStorage(const TString& uri)
883{
9827400b 884 //
885 // Called in case the preprocessor is declared failed. Remove remaining objects from the local storages.
886 //
3301427a 887
888 const char* type = 0;
889 if(uri == fgkLocalCDB) {
890 type = "OCDB";
891 } else if(uri == fgkLocalRefStorage) {
546242fb 892 type = "Reference";
3301427a 893 } else {
894 AliError(Form("Invalid storage URI: %s", uri.Data()));
895 return;
896 }
897
898 AliCDBManager* man = AliCDBManager::Instance();
b948db8d 899
3301427a 900 // open local storage
901 AliCDBStorage *localSto = man->GetStorage(uri);
902 if(!localSto) {
903 Log("SHUTTLE",
904 Form("CleanLocalStorage - cannot activate local %s storage", type));
905 return;
906 }
907
908 TString filename(Form("%s/%s/*/Run*_v%d_s*.root",
546242fb 909 localSto->GetBaseFolder().Data(), GetOfflineDetName(fCurrentDetector.Data()), GetCurrentRun()));
3301427a 910
c88ad5db 911 AliDebug(2, Form("filename = %s", filename.Data()));
3301427a 912
c88ad5db 913 Log("SHUTTLE", Form("Removing remaining local files for run %d and detector %s ...",
3301427a 914 GetCurrentRun(), fCurrentDetector.Data()));
915
916 RemoveFile(filename.Data());
917
918}
919
920//______________________________________________________________________________________________
921void AliShuttle::RemoveFile(const char* filename)
922{
9827400b 923 //
924 // removes local file
925 //
3301427a 926
927 TString command(Form("rm -f %s", filename));
928
929 Int_t result = gSystem->Exec(command.Data());
930 if(result != 0)
931 {
932 Log("SHUTTLE", Form("RemoveFile - %s: Cannot remove file %s!",
933 fCurrentDetector.Data(), filename));
934 }
73abe331 935}
936
b948db8d 937//______________________________________________________________________________________________
5164a766 938AliShuttleStatus* AliShuttle::ReadShuttleStatus()
939{
9827400b 940 //
941 // Reads the AliShuttleStatus from the CDB
942 //
5164a766 943
2bb7b766 944 if (fStatusEntry){
945 delete fStatusEntry;
946 fStatusEntry = 0;
947 }
5164a766 948
10a5a932 949 fStatusEntry = AliCDBManager::Instance()->GetStorage(GetLocalCDB())
2bb7b766 950 ->Get(Form("/SHUTTLE/STATUS/%s", fCurrentDetector.Data()), GetCurrentRun());
5164a766 951
2bb7b766 952 if (!fStatusEntry) return 0;
953 fStatusEntry->SetOwner(1);
5164a766 954
2bb7b766 955 AliShuttleStatus* status = dynamic_cast<AliShuttleStatus*> (fStatusEntry->GetObject());
956 if (!status) {
957 AliError("Invalid object stored to CDB!");
958 return 0;
959 }
5164a766 960
2bb7b766 961 return status;
5164a766 962}
963
964//______________________________________________________________________________________________
7bfb2090 965Bool_t AliShuttle::WriteShuttleStatus(AliShuttleStatus* status)
5164a766 966{
9827400b 967 //
968 // writes the status for one subdetector
969 //
2bb7b766 970
971 if (fStatusEntry){
972 delete fStatusEntry;
973 fStatusEntry = 0;
974 }
5164a766 975
2bb7b766 976 Int_t run = GetCurrentRun();
5164a766 977
2bb7b766 978 AliCDBId id(AliCDBPath("SHUTTLE", "STATUS", fCurrentDetector), run, run);
5164a766 979
2bb7b766 980 fStatusEntry = new AliCDBEntry(status, id, new AliCDBMetaData);
981 fStatusEntry->SetOwner(1);
5164a766 982
2bb7b766 983 UInt_t result = AliCDBManager::Instance()->GetStorage(fgkLocalCDB)->Put(fStatusEntry);
7bfb2090 984
2bb7b766 985 if (!result) {
3301427a 986 Log("SHUTTLE", Form("WriteShuttleStatus - Failed for %s, run %d",
987 fCurrentDetector.Data(), run));
2bb7b766 988 return kFALSE;
989 }
e7f62f16 990
8e828d39 991 SendMLDetInfo();
7bfb2090 992
2bb7b766 993 return kTRUE;
5164a766 994}
995
996//______________________________________________________________________________________________
997void AliShuttle::UpdateShuttleStatus(AliShuttleStatus::Status newStatus, Bool_t increaseCount)
998{
9827400b 999 //
1000 // changes the AliShuttleStatus for the given detector and run to the given status
1001 //
5164a766 1002
2bb7b766 1003 if (!fStatusEntry){
1004 AliError("UNEXPECTED: fStatusEntry empty");
1005 return;
1006 }
5164a766 1007
2bb7b766 1008 AliShuttleStatus* status = dynamic_cast<AliShuttleStatus*> (fStatusEntry->GetObject());
5164a766 1009
2bb7b766 1010 if (!status){
c88ad5db 1011 Log("SHUTTLE", "UpdateShuttleStatus - UNEXPECTED: status could not be read from current CDB entry");
2bb7b766 1012 return;
1013 }
5164a766 1014
2c15234c 1015 TString actionStr = Form("UpdateShuttleStatus - %s: Changing state from %s to %s",
eba76848 1016 fCurrentDetector.Data(),
36c99a6a 1017 status->GetStatusName(),
eba76848 1018 status->GetStatusName(newStatus));
cb343cfd 1019 Log("SHUTTLE", actionStr);
1020 SetLastAction(actionStr);
5164a766 1021
2bb7b766 1022 status->SetStatus(newStatus);
1023 if (increaseCount) status->IncreaseCount();
5164a766 1024
2bb7b766 1025 AliCDBManager::Instance()->GetStorage(fgkLocalCDB)->Put(fStatusEntry);
e7f62f16 1026
8e828d39 1027 SendMLDetInfo();
5164a766 1028}
e7f62f16 1029
1030//______________________________________________________________________________________________
8e828d39 1031void AliShuttle::SendMLDetInfo()
e7f62f16 1032{
1033 //
1034 // sends ML information about the current status of the current detector being processed
1035 //
1036
1037 AliShuttleStatus* status = dynamic_cast<AliShuttleStatus*> (fStatusEntry->GetObject());
1038
1039 if (!status){
8e828d39 1040 Log("SHUTTLE", "SendMLDetInfo - UNEXPECTED: status could not be read from current CDB entry");
e7f62f16 1041 return;
1042 }
1043
1044 TMonaLisaText mlStatus(Form("%s_status", fCurrentDetector.Data()), status->GetStatusName());
1045 TMonaLisaValue mlRetryCount(Form("%s_count", fCurrentDetector.Data()), status->GetCount());
1046
1047 TList mlList;
1048 mlList.Add(&mlStatus);
1049 mlList.Add(&mlRetryCount);
1050
ee6f7523 1051 TString mlID;
1052 mlID.Form("%d", GetCurrentRun());
1053 fMonaLisa->SendParameters(&mlList, mlID);
e7f62f16 1054}
1055
5164a766 1056//______________________________________________________________________________________________
1057Bool_t AliShuttle::ContinueProcessing()
1058{
9827400b 1059 // this function reads the AliShuttleStatus information from CDB and
1060 // checks if the processing should be continued
1061 // if yes it returns kTRUE and updates the AliShuttleStatus with nextStatus
2bb7b766 1062
926cbc0e 1063 if (!fConfig->HostProcessDetector(fCurrentDetector))
1064 return kFALSE;
57c1a579 1065
1066 AliPreprocessor* aPreprocessor =
1067 dynamic_cast<AliPreprocessor*> (fPreprocessorMap.GetValue(fCurrentDetector));
1068 if (!aPreprocessor)
1069 {
c88ad5db 1070 Log("SHUTTLE", Form("ContinueProcessing - %s: no preprocessor registered", fCurrentDetector.Data()));
57c1a579 1071 return kFALSE;
1072 }
1073
2bb7b766 1074 AliShuttleLogbookEntry::Status entryStatus =
eba76848 1075 fLogbookEntry->GetDetectorStatus(fCurrentDetector);
2bb7b766 1076
5e993b6f 1077 if (entryStatus != AliShuttleLogbookEntry::kUnprocessed) {
c88ad5db 1078 Log("SHUTTLE", Form("ContinueProcessing - %s is %s",
2bb7b766 1079 fCurrentDetector.Data(),
1080 fLogbookEntry->GetDetectorStatusName(entryStatus)));
1081 return kFALSE;
1082 }
1083
1084 // if we get here, according to Shuttle logbook subdetector is in UNPROCESSED state
be48e3ea 1085
1086 // check if current run is first unprocessed run for current detector
1087 if (fConfig->StrictRunOrder(fCurrentDetector) &&
1088 !fFirstUnprocessed[GetDetPos(fCurrentDetector)])
1089 {
86aa42c3 1090 if (fTestMode == kNone)
1091 {
c88ad5db 1092 Log("SHUTTLE", Form("ContinueProcessing - %s requires strict run ordering"
1093 " but this is not the first unprocessed run!"));
86aa42c3 1094 return kFALSE;
1095 }
1096 else
1097 {
c88ad5db 1098 Log("SHUTTLE", Form("ContinueProcessing - In TESTMODE - "
1099 "Although %s requires strict run ordering "
1100 "and this is not the first unprocessed run, "
1101 "the SHUTTLE continues"));
86aa42c3 1102 }
be48e3ea 1103 }
1104
926cbc0e 1105 // Is the subdetector processed first time for this run?
1106 fFirstProcessing = kFALSE;
1107
2bb7b766 1108 AliShuttleStatus* status = ReadShuttleStatus();
1109 if (!status) {
1110 // first time
1111 Log("SHUTTLE", Form("ContinueProcessing - %s: Processing first time",
1112 fCurrentDetector.Data()));
1113 status = new AliShuttleStatus(AliShuttleStatus::kStarted);
926cbc0e 1114 fFirstProcessing = kTRUE;
2bb7b766 1115 return WriteShuttleStatus(status);
1116 }
1117
ffa25f20 1118 // The following case shouldn't happen if Shuttle Logbook was correctly updated.
2bb7b766 1119 // If it happens it may mean Logbook updating failed... let's do it now!
1120 if (status->GetStatus() == AliShuttleStatus::kDone ||
ffa25f20 1121 status->GetStatus() == AliShuttleStatus::kFailed ||
1122 status->GetStatus() == AliShuttleStatus::kSkipped) {
2bb7b766 1123 Log("SHUTTLE", Form("ContinueProcessing - %s is already %s. Updating Shuttle Logbook",
1124 fCurrentDetector.Data(),
1125 status->GetStatusName(status->GetStatus())));
ffa25f20 1126
1127 if (status->GetStatus() == AliShuttleStatus::kSkipped)
1128 {
1129 UpdateShuttleLogbook(fCurrentDetector.Data(), "DONE");
1130 }
1131 else
1132 UpdateShuttleLogbook(fCurrentDetector.Data(), status->GetStatusName(status->GetStatus()));
1133
2bb7b766 1134 return kFALSE;
1135 }
1136
5e993b6f 1137 if (status->GetStatus() == AliShuttleStatus::kStoreStarted || status->GetStatus() == AliShuttleStatus::kStoreDelayed ||status->GetStatus() == AliShuttleStatus::kStoreError) {
2bb7b766 1138 Log("SHUTTLE",
c88ad5db 1139 Form("ContinueProcessing - %s: Grid storage of one or more "
1140 "objects failed. Trying again now",
2bb7b766 1141 fCurrentDetector.Data()));
5e993b6f 1142 StoreOCDB();
2bb7b766 1143 return kFALSE;
1144 }
1145
1146 // if we get here, there is a restart
57c1a579 1147 Bool_t cont = kFALSE;
2bb7b766 1148
1149 // abort conditions
cb343cfd 1150 if (status->GetCount() >= fConfig->GetMaxRetries()) {
57c1a579 1151 Log("SHUTTLE", Form("ContinueProcessing - %s failed %d times in status %s - "
1152 "Updating Shuttle Logbook", fCurrentDetector.Data(),
2bb7b766 1153 status->GetCount(), status->GetStatusName()));
1154 UpdateShuttleLogbook(fCurrentDetector.Data(), "FAILED");
e7f62f16 1155 UpdateShuttleStatus(AliShuttleStatus::kFailed);
3301427a 1156
1157 // there may still be objects in local OCDB and reference storage
1158 // and FXS databases may be not updated: do it now!
9827400b 1159
1160 // TODO Currently disabled, we want to keep files in case of failure!
1161 // CleanLocalStorage(fgkLocalCDB);
1162 // CleanLocalStorage(fgkLocalRefStorage);
1163 // UpdateTableFailCase();
1164
1165 // Send mail to detector expert!
c88ad5db 1166 Log("SHUTTLE", Form("ContinueProcessing - Sending mail to %s expert...",
926cbc0e 1167 fCurrentDetector.Data()));
339adafa 1168 if (!SendMail(kPPEMail))
9827400b 1169 Log("SHUTTLE", Form("ContinueProcessing - Could not send mail to %s expert",
926cbc0e 1170 fCurrentDetector.Data()));
3301427a 1171
57c1a579 1172 } else {
1173 Log("SHUTTLE", Form("ContinueProcessing - %s: restarting. "
1174 "Aborted before with %s. Retry number %d.", fCurrentDetector.Data(),
1175 status->GetStatusName(), status->GetCount()));
9827400b 1176 Bool_t increaseCount = kTRUE;
c88ad5db 1177 if (status->GetStatus() == AliShuttleStatus::kDCSError ||
8da81210 1178 status->GetStatus() == AliShuttleStatus::kDCSStarted ||
1179 status->GetStatus() == AliShuttleStatus::kFXSError)
c88ad5db 1180 increaseCount = kFALSE;
675f64cd 1181
9827400b 1182 UpdateShuttleStatus(AliShuttleStatus::kStarted, increaseCount);
57c1a579 1183 cont = kTRUE;
2bb7b766 1184 }
1185
57c1a579 1186 return cont;
5164a766 1187}
1188
8e828d39 1189//______________________________________________________________________________________________
1190void AliShuttle::SendMLRunInfo(const char* status)
1191{
1192 //
1193 // Send information about this run to ML
1194
1195 TMonaLisaText mlStatus("SHUTTLE_status", status);
1196 TString runType(fLogbookEntry->GetRunType());
1197 if (strlen(fLogbookEntry->GetRunParameter("log")) > 0){
1198
1199 runType += "(";
1200 runType += fLogbookEntry->GetRunParameter("log");
1201 runType += ")";
1202 }
1203 TMonaLisaText mlRunType("SHUTTLE_runtype", runType);
1204
1205 TList mlList;
1206 mlList.Add(&mlStatus);
1207 mlList.Add(&mlRunType);
1208
1209 TString mlID;
1210 mlID.Form("%d", GetCurrentRun());
1211 fMonaLisa->SendParameters(&mlList, mlID);
1212}
1213
ef83773f 1214//______________________________________________________________________________________________
1215Int_t AliShuttle::GetMem(Int_t pid)
1216{
1217 // invokes ps to get the memory consumption of the process <pid>
1218 // returns -1 in case of error
1219
1220 TString checkStr;
1221 checkStr.Form("ps -o vsize --pid %d | tail -n 1", pid);
1222 FILE* pipe = gSystem->OpenPipe(checkStr, "r");
1223 if (!pipe)
1224 {
1225 Log("SHUTTLE", Form("Process - Error: "
1226 "Could not open pipe to %s", checkStr.Data()));
1227 return -1;
1228 }
1229
1230 char buffer[100];
1231 if (!fgets(buffer, 100, pipe))
1232 {
1233 Log("SHUTTLE", "Process - Error: ps did not return anything");
1234 gSystem->ClosePipe(pipe);
1235 return -1;
1236 }
1237 gSystem->ClosePipe(pipe);
1238
1239 //Log("SHUTTLE", Form("ps returned %s", buffer));
1240
1241 Int_t mem = 0;
1242 if ((sscanf(buffer, "%d\n", &mem) != 1) || !mem)
1243 {
1244 Log("SHUTTLE", "Process - Error: Could not parse output of ps");
1245 return -1;
1246 }
1247
1248 return mem;
1249}
1250
5164a766 1251//______________________________________________________________________________________________
2bb7b766 1252Bool_t AliShuttle::Process(AliShuttleLogbookEntry* entry)
58bc3020 1253{
73abe331 1254 //
b948db8d 1255 // Makes data retrieval for all detectors in the configuration.
2bb7b766 1256 // entry: Shuttle logbook entry, contains run paramenters and status of detectors
1257 // (Unprocessed, Inactive, Failed or Done).
d477ad88 1258 // Returns kFALSE in case of error occured and kTRUE otherwise
73abe331 1259 //
1260
9827400b 1261 if (!entry) return kFALSE;
2bb7b766 1262
1263 fLogbookEntry = entry;
1264
c88ad5db 1265 Log("SHUTTLE", Form("\t\t\t^*^*^*^*^*^*^*^*^*^*^*^* run %d: START ^*^*^*^*^*^*^*^*^*^*^*^*",
9827400b 1266 GetCurrentRun()));
2bb7b766 1267
339adafa 1268 CountOpenRuns();
1269
8e828d39 1270 // Send the information to ML
1271 SendMLRunInfo("Processing");
3301427a 1272
9827400b 1273 if (fLogbookEntry->IsDone())
1274 {
1275 Log("SHUTTLE","Process - Shuttle is already DONE. Updating logbook");
1276 UpdateShuttleLogbook("shuttle_done");
1277 fLogbookEntry = 0;
1278 return kTRUE;
1279 }
1280
1281 // read test mode if flag is set
1282 if (fReadTestMode)
1283 {
3d8bc902 1284 fTestMode = kNone;
9827400b 1285 TString logEntry(entry->GetRunParameter("log"));
1286 //printf("log entry = %s\n", logEntry.Data());
1287 TString searchStr("Testmode: ");
1288 Int_t pos = logEntry.Index(searchStr.Data());
1289 //printf("%d\n", pos);
1290 if (pos >= 0)
1291 {
1292 TSubString subStr = logEntry(pos + searchStr.Length(), logEntry.Length());
1293 //printf("%s\n", subStr.String().Data());
1294 TString newStr(subStr.Data());
1295 TObjArray* token = newStr.Tokenize(' ');
1296 if (token)
1297 {
1298 //token->Print();
1299 TObjString* tmpStr = dynamic_cast<TObjString*> (token->First());
1300 if (tmpStr)
1301 {
1302 Int_t testMode = tmpStr->String().Atoi();
1303 if (testMode > 0)
1304 {
c88ad5db 1305 Log("SHUTTLE", Form("Process - Enabling test mode %d", testMode));
9827400b 1306 SetTestMode((TestMode) testMode);
1307 }
1308 }
1309 delete token;
1310 }
1311 }
1312 }
c88ad5db 1313
eba76848 1314 fLogbookEntry->Print("all");
57f50b3c 1315
1316 // Initialization
d477ad88 1317 Bool_t hasError = kFALSE;
5164a766 1318
675f64cd 1319 // Set the CDB and Reference folders according to the year and LHC period
1320 TString lhcPeriod(GetLHCPeriod());
1321 if (lhcPeriod.Length() == 0)
1322 {
7d4cf768 1323 Log("SHUTTLE","Process - LHCPeriod not found in logbook!");
1324 return 0;
675f64cd 1325 }
1326
1265724d 1327 // build cdb paths (repeat each time, LHCperiod might have changed)
1328 fgkMainCDB.Form("alien://folder=%s%d/%s/OCDB?user=alidaq?cacheFold=/tmp/OCDBCache",
b8cf5e0f 1329 fConfig->GetAlienPath(), GetCurrentYear(), lhcPeriod.Data());
675f64cd 1330
1265724d 1331 fgkMainRefStorage.Form("alien://folder=%s%d/%s/Reference?user=alidaq?cacheFold=/tmp/OCDBCache",
b8cf5e0f 1332 fConfig->GetAlienPath(), GetCurrentYear(), lhcPeriod.Data());
675f64cd 1333
57f50b3c 1334 // Loop on detectors in the configuration
b948db8d 1335 TIter iter(fConfig->GetDetectors());
2bb7b766 1336 TObjString* aDetector = 0;
b948db8d 1337
4508ab8b 1338 Bool_t first = kTRUE;
1339
be48e3ea 1340 while ((aDetector = (TObjString*) iter.Next()))
1341 {
7bfb2090 1342 fCurrentDetector = aDetector->String();
5164a766 1343
5e993b6f 1344 if (ContinueProcessing() == kFALSE)
1345 continue;
4508ab8b 1346
1347 if (first)
1348 {
1349 // only read QueryCDB when needed and only once
1350 AliCDBStorage *mainCDBSto = AliCDBManager::Instance()->GetStorage(fgkMainCDB);
1351 if(mainCDBSto) mainCDBSto->QueryCDB(GetCurrentRun());
1352 AliCDBStorage *mainRefSto = AliCDBManager::Instance()->GetStorage(fgkMainRefStorage);
1353 if(mainRefSto) mainRefSto->QueryCDB(GetCurrentRun());
1354 first = kFALSE;
1355 }
9e080f92 1356
c88ad5db 1357 Log("SHUTTLE", Form("\t\t\t****** run %d - %s: START ******",
2bb7b766 1358 GetCurrentRun(), aDetector->GetName()));
1359
9d733021 1360 for(Int_t iSys=0;iSys<3;iSys++) fFXSCalled[iSys]=kFALSE;
ef83773f 1361
1362 Int_t initialMem = GetMem(getpid());
1363 Log("SHUTTLE", Form("Memory consumption before forking is %d", initialMem));
9d733021 1364
c88ad5db 1365 Log(fCurrentDetector.Data(), "Process - Starting processing");
85a80aa9 1366
be48e3ea 1367 Int_t pid = fork();
1368
1369 if (pid < 0)
1370 {
c88ad5db 1371 Log("SHUTTLE", "Process - ERROR: Forking failed");
be48e3ea 1372 }
1373 else if (pid > 0)
1374 {
1375 // parent
c88ad5db 1376 Log("SHUTTLE", Form("Process - In parent process of %d - %s: Starting monitoring",
be48e3ea 1377 GetCurrentRun(), aDetector->GetName()));
1378
1379 Long_t begin = time(0);
1380
1381 int status; // to be used with waitpid, on purpose an int (not Int_t)!
1382 while (waitpid(pid, &status, WNOHANG) == 0)
1383 {
1384 Long_t expiredTime = time(0) - begin;
1385
1386 if (expiredTime > fConfig->GetPPTimeOut())
1387 {
9827400b 1388 TString tmp;
c88ad5db 1389 tmp.Form("Process - Process of %s time out. "
1390 "Run time: %d seconds. Killing...",
1391 fCurrentDetector.Data(), expiredTime);
9827400b 1392 Log("SHUTTLE", tmp);
1393 Log(fCurrentDetector, tmp);
be48e3ea 1394
1395 kill(pid, 9);
1396
26b1866e 1397 UpdateShuttleStatus(AliShuttleStatus::kPPTimeOut);
be48e3ea 1398 hasError = kTRUE;
1399
1400 gSystem->Sleep(1000);
1401 }
1402 else
1403 {
be48e3ea 1404 gSystem->Sleep(1000);
9827400b 1405
ef83773f 1406 Int_t mem = GetMem(pid);
1407
1408 if (mem < 0)
9827400b 1409 continue;
9827400b 1410
ef83773f 1411 mem -= initialMem;
1412 if (mem < 0)
1413 mem = 0;
9827400b 1414
1415 if (expiredTime % 60 == 0)
ee6f7523 1416 {
c88ad5db 1417 Log("SHUTTLE", Form("Process - %s: Checking process. "
1418 "Run time: %d seconds - Memory consumption: %d KB",
1419 fCurrentDetector.Data(), expiredTime, mem));
ee6f7523 1420 SendAlive();
1421 }
9827400b 1422
1423 if (mem > fConfig->GetPPMaxMem())
1424 {
1425 TString tmp;
c88ad5db 1426 tmp.Form("Process - Process exceeds maximum allowed memory "
1427 "(%d KB > %d KB). Killing...",
9827400b 1428 mem, fConfig->GetPPMaxMem());
1429 Log("SHUTTLE", tmp);
1430 Log(fCurrentDetector, tmp);
1431
1432 kill(pid, 9);
1433
1434 UpdateShuttleStatus(AliShuttleStatus::kPPOutOfMemory);
1435 hasError = kTRUE;
1436
1437 gSystem->Sleep(1000);
1438 }
be48e3ea 1439 }
1440 }
1441
c88ad5db 1442 Log("SHUTTLE", Form("Process - In parent process of %d - %s: Client has terminated.",
be48e3ea 1443 GetCurrentRun(), aDetector->GetName()));
1444
1445 if (WIFEXITED(status))
1446 {
1447 Int_t returnCode = WEXITSTATUS(status);
1448
c88ad5db 1449 Log("SHUTTLE", Form("Process - %s: the return code is %d", fCurrentDetector.Data(),
3301427a 1450 returnCode));
be48e3ea 1451
9827400b 1452 if (returnCode == 0) hasError = kTRUE;
be48e3ea 1453 }
1454 }
1455 else if (pid == 0)
1456 {
26b1866e 1457 // child
1458 Log("SHUTTLE", Form("Process - In child process of %d - %s", GetCurrentRun(),
c88ad5db 1459 aDetector->GetName()));
be48e3ea 1460
c88ad5db 1461 Log("SHUTTLE", Form("Process - Redirecting output to %s log",fCurrentDetector.Data()));
ffa29e93 1462
546242fb 1463 if ((freopen(GetLogFileName(fCurrentDetector), "a", stdout)) == 0)
ffa29e93 1464 {
c88ad5db 1465 Log("SHUTTLE", "Process - Could not freopen stdout");
ffa29e93 1466 }
1467 else
1468 {
1469 fOutputRedirected = kTRUE;
1470 if ((dup2(fileno(stdout), fileno(stderr))) < 0)
c88ad5db 1471 Log("SHUTTLE", "Process - Could not redirect stderr");
ffa29e93 1472
1473 }
1474
5bac2bde 1475 TString wd = gSystem->WorkingDirectory();
675f64cd 1476 TString tmpDir = Form("%s/%s_%d_process", GetShuttleTempDir(),
1477 fCurrentDetector.Data(), GetCurrentRun());
5bac2bde 1478
d524ade6 1479 Int_t result = gSystem->GetPathInfo(tmpDir.Data(), 0, (Long64_t*) 0, 0, 0);
1480 if (!result) // temp dir already exists!
1481 {
1482 Log(fCurrentDetector.Data(),
1483 Form("Process - %s dir already exists! Removing...", tmpDir.Data()));
1484 gSystem->Exec(Form("rm -rf %s",tmpDir.Data()));
675f64cd 1485 }
1486
1487 if (gSystem->mkdir(tmpDir.Data(), 1))
1488 {
1489 Log(fCurrentDetector.Data(), "Process - could not make temp directory!!");
1490 gSystem->Exit(1);
d524ade6 1491 }
1492
1493 if (!gSystem->ChangeDirectory(tmpDir.Data()))
1494 {
1495 Log(fCurrentDetector.Data(), "Process - could not change directory!!");
1496 gSystem->Exit(1);
1497 }
5bac2bde 1498
ffa25f20 1499 Int_t success = ProcessCurrentDetector();
26b1866e 1500
5bac2bde 1501 gSystem->ChangeDirectory(wd.Data());
b1d18693 1502
ffa25f20 1503 if (success == 1) // Preprocessor finished successfully!
9827400b 1504 {
fb40ead4 1505 // remove temporary folder or DCS map
1506 if (!fConfig->KeepTempFolder())
1507 {
1508 gSystem->Exec(Form("rm -rf %s",tmpDir.Data()));
1509 } else if (!fConfig->KeepDCSMap())
1510 {
1511 gSystem->Exec(Form("rm -f %s/DCSMap.root",tmpDir.Data()));
1512 }
b1d18693 1513
3301427a 1514 // Update time_processed field in FXS DB
1515 if (UpdateTable() == kFALSE)
5bac2bde 1516 Log("SHUTTLE", Form("Process - %s: Could not update FXS databases!",
1517 fCurrentDetector.Data()));
3301427a 1518
1519 // Transfer the data from local storage to main storage (Grid)
3301427a 1520 if (StoreOCDB() == kFALSE)
9827400b 1521 success = kFALSE;
5e993b6f 1522 }
ffa25f20 1523 else if (success == 0)
c88ad5db 1524 {
1525 Log("SHUTTLE",
1526 Form("\t\t\t****** run %d - %s: PP ERROR ******",
1527 GetCurrentRun(), aDetector->GetName()));
be48e3ea 1528 }
1529
4b95672b 1530 for (UInt_t iSys=0; iSys<3; iSys++)
1531 {
1532 if (fFXSCalled[iSys]) fFXSlist[iSys].Clear();
1533 }
1534
c88ad5db 1535 Log("SHUTTLE", Form("Process - Client process of %d - %s is exiting now with %d.",
9827400b 1536 GetCurrentRun(), aDetector->GetName(), success));
be48e3ea 1537
1538 // the client exits here
9827400b 1539 gSystem->Exit(success);
be48e3ea 1540
1541 AliError("We should never get here!!!");
1542 }
7bfb2090 1543 }
5164a766 1544
c88ad5db 1545 Log("SHUTTLE", Form("\t\t\t^*^*^*^*^*^*^*^*^*^*^*^* run %d: FINISH ^*^*^*^*^*^*^*^*^*^*^*^*",
2bb7b766 1546 GetCurrentRun()));
1547
1548 //check if shuttle is done for this run, if so update logbook
1549 TObjArray checkEntryArray;
1550 checkEntryArray.SetOwner(1);
9e080f92 1551 TString whereClause = Form("where run=%d", GetCurrentRun());
b0e53b15 1552 if (!QueryShuttleLogbook(whereClause.Data(), checkEntryArray) ||
1553 checkEntryArray.GetEntries() == 0) {
9e080f92 1554 Log("SHUTTLE", Form("Process - Warning: Cannot check status of run %d on Shuttle logbook!",
1555 GetCurrentRun()));
1556 return hasError == kFALSE;
1557 }
b948db8d 1558
9e080f92 1559 AliShuttleLogbookEntry* checkEntry = dynamic_cast<AliShuttleLogbookEntry*>
1560 (checkEntryArray.At(0));
2bb7b766 1561
9e080f92 1562 if (checkEntry)
1563 {
1564 if (checkEntry->IsDone())
be48e3ea 1565 {
9e080f92 1566 Log("SHUTTLE","Process - Shuttle is DONE. Updating logbook");
1567 UpdateShuttleLogbook("shuttle_done");
1568 }
1569 else
1570 {
1571 for (UInt_t iDet=0; iDet<NDetectors(); iDet++)
be48e3ea 1572 {
9e080f92 1573 if (checkEntry->GetDetectorStatus(iDet) == AliShuttleLogbookEntry::kUnprocessed)
be48e3ea 1574 {
9e080f92 1575 AliDebug(2, Form("Run %d: setting %s as \"not first time unprocessed\"",
1576 checkEntry->GetRun(), GetDetName(iDet)));
1577 fFirstUnprocessed[iDet] = kFALSE;
be48e3ea 1578 }
1579 }
8e828d39 1580 SendMLRunInfo("Pending");
2bb7b766 1581 }
1582 }
1583
1584 fLogbookEntry = 0;
85a80aa9 1585
a7160fe9 1586 return hasError == kFALSE;
73abe331 1587}
1588
b948db8d 1589//______________________________________________________________________________________________
ffa25f20 1590Int_t AliShuttle::ProcessCurrentDetector()
73abe331 1591{
1592 //
2bb7b766 1593 // Makes data retrieval just for a specific detector (fCurrentDetector).
73abe331 1594 // Threre should be a configuration for this detector.
73abe331 1595
1d172743 1596 Log("SHUTTLE", Form("ProcessCurrentDetector - Retrieving values for %s, run %d",
1597 fCurrentDetector.Data(), GetCurrentRun()));
73abe331 1598
d524ade6 1599 TString wd = gSystem->WorkingDirectory();
1600
2d9019b4 1601 if (!CleanReferenceStorage(fCurrentDetector.Data()))
ffa25f20 1602 return 0;
d524ade6 1603
1604 gSystem->ChangeDirectory(wd.Data());
1605
3301427a 1606 // call preprocessor
1607 AliPreprocessor* aPreprocessor =
1608 dynamic_cast<AliPreprocessor*> (fPreprocessorMap.GetValue(fCurrentDetector));
1609
8579c6fb 1610 // check if the preprocessor wants to process this run type
1611 if (aPreprocessor->ProcessRunType() == kFALSE)
ffa25f20 1612 {
1613 UpdateShuttleStatus(AliShuttleStatus::kSkipped);
1614 UpdateShuttleLogbook(fCurrentDetector, "DONE");
1615 Log(fCurrentDetector, Form("ProcessCurrentDetector - %s preprocessor is not interested in this run type", fCurrentDetector.Data()));
1616
1617 return 2;
1618 }
1619
1620 TMap* dcsMap = new TMap();
1621
3301427a 1622 aPreprocessor->Initialize(GetCurrentRun(), GetCurrentStartTime(), GetCurrentEndTime());
1623
1624 Bool_t processDCS = aPreprocessor->ProcessDCS();
d477ad88 1625
651fdaab 1626 if (!processDCS)
1627 {
1d172743 1628 Log(fCurrentDetector, "ProcessCurrentDetector -"
1629 " The preprocessor requested to skip the retrieval of DCS values");
651fdaab 1630 }
8b739301 1631 else if (fTestMode & kSkipDCS)
2c15234c 1632 {
1d172743 1633 Log(fCurrentDetector, "ProcessCurrentDetector - In TESTMODE: Skipping DCS processing");
9827400b 1634 }
1635 else if (fTestMode & kErrorDCS)
1636 {
1d172743 1637 Log(fCurrentDetector, "ProcessCurrentDetector - In TESTMODE: Simulating DCS error");
3d8bc902 1638 UpdateShuttleStatus(AliShuttleStatus::kDCSStarted);
9827400b 1639 UpdateShuttleStatus(AliShuttleStatus::kDCSError);
1d172743 1640 delete dcsMap;
ffa25f20 1641 return 0;
2c15234c 1642 } else {
3301427a 1643
1644 UpdateShuttleStatus(AliShuttleStatus::kDCSStarted);
1645
1d172743 1646 // Query DCS archive
1647 Int_t nServers = fConfig->GetNServers(fCurrentDetector);
a038aa70 1648
1d172743 1649 for (int iServ=0; iServ<nServers; iServ++)
2c15234c 1650 {
1d172743 1651
1652 TString host(fConfig->GetDCSHost(fCurrentDetector, iServ));
1653 Int_t port = fConfig->GetDCSPort(fCurrentDetector, iServ);
542b6cc8 1654 Int_t multiSplit = fConfig->GetMultiSplit(fCurrentDetector, iServ);
1655
1790d4b7 1656 Log(fCurrentDetector, Form("ProcessCurrentDetector -"
1657 " Querying DCS Amanda server %s:%d (%d of %d)",
1658 host.Data(), port, iServ+1, nServers));
1d172743 1659
1660 TMap* aliasMap = 0;
1661 TMap* dpMap = 0;
26b1866e 1662
1d172743 1663 if (fConfig->GetDCSAliases(fCurrentDetector, iServ)->GetEntries() > 0)
2c15234c 1664 {
1d172743 1665 aliasMap = GetValueSet(host, port,
542b6cc8 1666 fConfig->GetDCSAliases(fCurrentDetector, iServ),
1667 kAlias, multiSplit);
1d172743 1668 if (!aliasMap)
1669 {
1670 Log(fCurrentDetector,
1671 Form("ProcessCurrentDetector -"
675f64cd 1672 " Error retrieving DCS aliases from server %s."
1673 " Sending mail to DCS experts!", host.Data()));
1d172743 1674 UpdateShuttleStatus(AliShuttleStatus::kDCSError);
675f64cd 1675
339adafa 1676 if (!SendMail(kDCSEMail))
fb40ead4 1677 Log("SHUTTLE", Form("ProcessCurrentDetector - "
95d4aaef 1678 "Could not send mail to DCS experts!"));
675f64cd 1679
a038aa70 1680 delete dcsMap;
ffa25f20 1681 return 0;
1d172743 1682 }
2c15234c 1683 }
a038aa70 1684
1d172743 1685 if (fConfig->GetDCSDataPoints(fCurrentDetector, iServ)->GetEntries() > 0)
a038aa70 1686 {
1d172743 1687 dpMap = GetValueSet(host, port,
542b6cc8 1688 fConfig->GetDCSDataPoints(fCurrentDetector, iServ),
1689 kDP, multiSplit);
1d172743 1690 if (!dpMap)
1691 {
1692 Log(fCurrentDetector,
1693 Form("ProcessCurrentDetector -"
675f64cd 1694 " Error retrieving DCS data points from server %s."
1695 " Sending mail to DCS experts!", host.Data()));
1d172743 1696 UpdateShuttleStatus(AliShuttleStatus::kDCSError);
675f64cd 1697
339adafa 1698 if (!SendMail(kDCSEMail))
fb40ead4 1699 Log("SHUTTLE", Form("ProcessCurrentDetector - "
95d4aaef 1700 "Could not send mail to DCS experts!"));
675f64cd 1701
1d172743 1702 if (aliasMap) delete aliasMap;
1703 delete dcsMap;
ffa25f20 1704 return 0;
1d172743 1705 }
a038aa70 1706 }
1d172743 1707
1708 // merge aliasMap and dpMap into dcsMap
1709 if(aliasMap) {
1710 TIter iter(aliasMap);
a038aa70 1711 TObjString* key = 0;
1712 while ((key = (TObjString*) iter.Next()))
1d172743 1713 dcsMap->Add(key, aliasMap->GetValue(key->String()));
1714
1715 aliasMap->SetOwner(kFALSE);
1716 delete aliasMap;
1717 }
1718
1719 if(dpMap) {
1720 TIter iter(dpMap);
1721 TObjString* key = 0;
1722 while ((key = (TObjString*) iter.Next()))
1723 dcsMap->Add(key, dpMap->GetValue(key->String()));
1724
1725 dpMap->SetOwner(kFALSE);
1726 delete dpMap;
a038aa70 1727 }
73abe331 1728 }
1729 }
dc25836b 1730
b1d18693 1731 // save map into file, to help debugging in case of preprocessor error
fbc112e3 1732 TFile* f = TFile::Open("DCSMap.root","recreate");
b1d18693 1733 f->cd();
1734 dcsMap->Write("DCSMap", TObject::kSingleKey);
1735 f->Close();
fbc112e3 1736 delete f;
b1d18693 1737
2bb7b766 1738 // DCS Archive DB processing successful. Call Preprocessor!
85a80aa9 1739 UpdateShuttleStatus(AliShuttleStatus::kPPStarted);
a7160fe9 1740
339adafa 1741 fFXSError = -1; // this variable is kTRUE after ::Process if an FXS error occured
8da81210 1742
a038aa70 1743 UInt_t returnValue = aPreprocessor->Process(dcsMap);
8da81210 1744
339adafa 1745 if (fFXSError!=-1) {
8da81210 1746 UpdateShuttleStatus(AliShuttleStatus::kFXSError);
339adafa 1747 SendMail(kFXSEMail, fFXSError);
8da81210 1748 dcsMap->DeleteAll();
1749 delete dcsMap;
ffa25f20 1750 return 0;
8da81210 1751 }
b948db8d 1752
3301427a 1753 if (returnValue > 0) // Preprocessor error!
1754 {
c88ad5db 1755 Log(fCurrentDetector, Form("ProcessCurrentDetector - "
1756 "Preprocessor failed. Process returned %d.", returnValue));
cb343cfd 1757 UpdateShuttleStatus(AliShuttleStatus::kPPError);
a038aa70 1758 dcsMap->DeleteAll();
1759 delete dcsMap;
ffa25f20 1760 return 0;
9827400b 1761 }
1762
1763 // preprocessor ok!
1764 UpdateShuttleStatus(AliShuttleStatus::kPPDone);
1765 Log(fCurrentDetector, Form("ProcessCurrentDetector - %s preprocessor returned success",
1766 fCurrentDetector.Data()));
b948db8d 1767
a038aa70 1768 dcsMap->DeleteAll();
1769 delete dcsMap;
b948db8d 1770
ffa25f20 1771 return 1;
2bb7b766 1772}
1773
fbc112e3 1774//______________________________________________________________________________________________
1775void AliShuttle::CountOpenRuns()
1776{
1777 // Query DAQ's Shuttle logbook and sends the number of open runs to ML
1778
1779 // check connection, in case connect
1780 if (!Connect(3))
1781 return;
1782
1783 TString sqlQuery;
1784 sqlQuery = Form("select count(*) from %s where shuttle_done=0", fConfig->GetShuttlelbTable());
1785
1786 TSQLResult* aResult = fServer[3]->Query(sqlQuery);
1787 if (!aResult) {
1788 AliError(Form("Can't execute query <%s>!", sqlQuery.Data()));
1789 return;
1790 }
1791
1792 AliDebug(2,Form("Query = %s", sqlQuery.Data()));
1793
1794 if (aResult->GetRowCount() == 0) {
1795 AliError(Form("No result for query %s received", sqlQuery.Data()));
1796 return;
1797 }
1798
1799 if (aResult->GetFieldCount() != 1) {
1800 AliError(Form("Invalid field count for query %s received", sqlQuery.Data()));
1801 return;
1802 }
1803
1804 TSQLRow* aRow = aResult->Next();
1805 if (!aRow) {
1806 AliError(Form("Could not receive result of query %s", sqlQuery.Data()));
1807 return;
1808 }
1809
1810 TString result(aRow->GetField(0), aRow->GetFieldLength(0));
1811 Int_t count = result.Atoi();
1812
1813 Log("SHUTTLE", Form("%d unprocessed runs", count));
1814
1815 delete aRow;
1816 delete aResult;
1817
1818 TMonaLisaValue mlStatus("SHUTTLE_openruns", count);
1819
1820 TList mlList;
1821 mlList.Add(&mlStatus);
1822
1823 fMonaLisa->SendParameters(&mlList, "__PROCESSINGINFO__");
1824}
1825
2bb7b766 1826//______________________________________________________________________________________________
1827Bool_t AliShuttle::QueryShuttleLogbook(const char* whereClause,
1828 TObjArray& entries)
1829{
9827400b 1830 // Query DAQ's Shuttle logbook and fills detector status object.
1831 // Call QueryRunParameters to query DAQ logbook for run parameters.
1832 //
2bb7b766 1833
fc5a4708 1834 entries.SetOwner(1);
1835
2bb7b766 1836 // check connection, in case connect
fbc112e3 1837 if (!Connect(3)) return kFALSE;
2bb7b766 1838
1839 TString sqlQuery;
441b0e9c 1840 sqlQuery = Form("select * from %s %s order by run", fConfig->GetShuttlelbTable(), whereClause);
2bb7b766 1841
be48e3ea 1842 TSQLResult* aResult = fServer[3]->Query(sqlQuery);
2bb7b766 1843 if (!aResult) {
1844 AliError(Form("Can't execute query <%s>!", sqlQuery.Data()));
1845 return kFALSE;
1846 }
1847
fc5a4708 1848 AliDebug(2,Form("Query = %s", sqlQuery.Data()));
1849
2bb7b766 1850 if(aResult->GetRowCount() == 0) {
c88ad5db 1851 Log("SHUTTLE", "No entries in Shuttle Logbook match request");
9827400b 1852 delete aResult;
1853 return kTRUE;
2bb7b766 1854 }
1855
1856 // TODO Check field count!
db99d43e 1857 const UInt_t nCols = 23;
2bb7b766 1858 if (aResult->GetFieldCount() != (Int_t) nCols) {
c88ad5db 1859 Log("SHUTTLE", "Invalid SQL result field number!");
2bb7b766 1860 delete aResult;
1861 return kFALSE;
1862 }
1863
2bb7b766 1864 TSQLRow* aRow;
1865 while ((aRow = aResult->Next())) {
1866 TString runString(aRow->GetField(0), aRow->GetFieldLength(0));
1867 Int_t run = runString.Atoi();
1868
eba76848 1869 AliShuttleLogbookEntry *entry = QueryRunParameters(run);
1870 if (!entry)
1871 continue;
2bb7b766 1872
1873 // loop on detectors
eba76848 1874 for(UInt_t ii = 0; ii < nCols; ii++)
1875 entry->SetDetectorStatus(aResult->GetFieldName(ii), aRow->GetField(ii));
2bb7b766 1876
eba76848 1877 entries.AddLast(entry);
2bb7b766 1878 delete aRow;
1879 }
1880
2bb7b766 1881 delete aResult;
1882 return kTRUE;
1883}
1884
1885//______________________________________________________________________________________________
eba76848 1886AliShuttleLogbookEntry* AliShuttle::QueryRunParameters(Int_t run)
2bb7b766 1887{
eba76848 1888 //
1889 // Retrieve run parameters written in the DAQ logbook and sets them into AliShuttleLogbookEntry object
1890 //
2bb7b766 1891
1892 // check connection, in case connect
be48e3ea 1893 if (!Connect(3))
eba76848 1894 return 0;
2bb7b766 1895
1896 TString sqlQuery;
2c15234c 1897 sqlQuery.Form("select * from %s where run=%d", fConfig->GetDAQlbTable(), run);
2bb7b766 1898
be48e3ea 1899 TSQLResult* aResult = fServer[3]->Query(sqlQuery);
2bb7b766 1900 if (!aResult) {
c88ad5db 1901 Log("SHUTTLE", Form("Can't execute query <%s>!", sqlQuery.Data()));
eba76848 1902 return 0;
2bb7b766 1903 }
1904
eba76848 1905 if (aResult->GetRowCount() == 0) {
2bb7b766 1906 Log("SHUTTLE", Form("QueryRunParameters - No entry in DAQ Logbook for run %d. Skipping", run));
1907 delete aResult;
eba76848 1908 return 0;
2bb7b766 1909 }
1910
eba76848 1911 if (aResult->GetRowCount() > 1) {
c88ad5db 1912 Log("SHUTTLE", Form("QueryRunParameters - UNEXPECTED: "
1913 "more than one entry in DAQ Logbook for run %d!", run));
2bb7b766 1914 delete aResult;
eba76848 1915 return 0;
2bb7b766 1916 }
1917
eba76848 1918 TSQLRow* aRow = aResult->Next();
1919 if (!aRow)
1920 {
c88ad5db 1921 Log("SHUTTLE", Form("QueryRunParameters - Could not retrieve row for run %d. Skipping", run));
eba76848 1922 delete aResult;
1923 return 0;
1924 }
2bb7b766 1925
eba76848 1926 AliShuttleLogbookEntry* entry = new AliShuttleLogbookEntry(run);
2bb7b766 1927
eba76848 1928 for (Int_t ii = 0; ii < aResult->GetFieldCount(); ii++)
1929 entry->SetRunParameter(aResult->GetFieldName(ii), aRow->GetField(ii));
2bb7b766 1930
c9db68c7 1931 delete aRow;
1932 delete aResult;
1933
eba76848 1934 UInt_t startTime = entry->GetStartTime();
1935 UInt_t endTime = entry->GetEndTime();
89bcf3b4 1936 Bool_t ecsSuccess = entry->GetECSSuccess();
c9db68c7 1937
b0e53b15 1938 TString totEventsStr = entry->GetRunParameter("totalEvents");
1939 Int_t totEvents = totEventsStr.Atoi();
c9db68c7 1940
af93e5d7 1941 UInt_t now = time(0);
1942 // TODO make this a configuration parameter
1943 Int_t dcsDelay = 120;
1944
8e828d39 1945 // runs are accepted if they have ecsSuccess set or more than 1 event
af93e5d7 1946 if (startTime != 0 && endTime != 0 && endTime > startTime && (totEvents > 1 || ecsSuccess) && (endTime < now - dcsDelay))
89bcf3b4 1947 {
8e828d39 1948 if (ecsSuccess == kFALSE)
1949 Log("SHUTTLE", Form("Processing run %d although in status ECS failure, Reason: %s", run, entry->GetRunParameter("eor_reason")));
1950 return entry;
89bcf3b4 1951 }
8e828d39 1952
1953 Bool_t skip = kFALSE;
1954
1955 if (totEvents <= 1)
89bcf3b4 1956 {
8e828d39 1957 Log("SHUTTLE", Form("QueryRunParameters - Run %d has 1 event or less - Skipping!", run));
1958 skip = kTRUE;
af93e5d7 1959 }
1960 else if (endTime != 0 && endTime >= now - dcsDelay)
1961 {
1962 Log("SHUTTLE", Form("Skipping run %d for now, because DCS buffer time is not yet expired", run));
1963 }
89bcf3b4 1964 else
1965 {
1966 Log("SHUTTLE", Form("QueryRunParameters - Invalid parameters for Run %d: "
1967 "startTime = %d, endTime = %d. Skipping (Shuttle won't be marked as DONE)!",
1968 run, startTime, endTime));
1969 }
8e828d39 1970
1971 if (skip)
1972 {
1973 Log("SHUTTLE", Form("Marking SHUTTLE skipped for run %d", run));
1974 fLogbookEntry = entry;
1975 if (!UpdateShuttleLogbook("shuttle_skipped"))
1976 {
1977 AliError(Form("Could not update logbook for run %d !", run));
1978 }
1979 fLogbookEntry = 0;
1980 }
c9db68c7 1981
1982 delete entry;
1983 return 0;
2bb7b766 1984}
1985
a038aa70 1986//______________________________________________________________________________________________
1987TMap* AliShuttle::GetValueSet(const char* host, Int_t port, const TSeqCollection* entries,
542b6cc8 1988 DCSType type, Int_t multiSplit)
a038aa70 1989{
1990 // Retrieve all "entry" data points from the DCS server
1991 // host, port: TSocket connection parameters
1992 // entries: list of name of the alias or data point
1993 // type: kAlias or kDP
1994 // returns TMap of values, 0 when failure
542b6cc8 1995
1996 AliDCSClient client(host, port, fTimeout, fRetries, multiSplit);
b41b252a 1997
a038aa70 1998 TMap* result = 0;
b41b252a 1999 if (type == kAlias)
a038aa70 2000 {
b41b252a 2001 result = client.GetAliasValues(entries, GetCurrentStartTime(),
2002 GetCurrentEndTime());
2003 }
2004 else if (type == kDP)
2005 {
2006 result = client.GetDPValues(entries, GetCurrentStartTime(),
2007 GetCurrentEndTime());
2008 }
a038aa70 2009
b41b252a 2010 if (result == 0)
2011 {
2012 Log(fCurrentDetector.Data(), Form("GetValueSet - Can't get entries! Reason: %s",
1790d4b7 2013 client.GetErrorString(client.GetResultErrorCode())));
2014 if (client.GetResultErrorCode() == AliDCSClient::fgkServerError)
2015 Log(fCurrentDetector.Data(), Form("GetValueSet - Server error code: %s",
2016 client.GetServerError().Data()));
a038aa70 2017
b41b252a 2018 return 0;
a038aa70 2019 }
b41b252a 2020
a038aa70 2021 return result;
2022}
b41b252a 2023
b948db8d 2024//______________________________________________________________________________________________
57f50b3c 2025const char* AliShuttle::GetFile(Int_t system, const char* detector,
2026 const char* id, const char* source)
b948db8d 2027{
9827400b 2028 // Get calibration file from file exchange servers
2029 // First queris the FXS database for the file name, using the run, detector, id and source info
2030 // then calls RetrieveFile(filename) for actual copy to local disk
2031 // run: current run being processed (given by Logbook entry fLogbookEntry)
2032 // detector: the Preprocessor name
2033 // id: provided as a parameter by the Preprocessor
2034 // source: provided by the Preprocessor through GetFileSources function
2035
2036 // check if test mode should simulate a FXS error
2037 if (fTestMode & kErrorFXSFiles)
2038 {
2039 Log(detector, Form("GetFile - In TESTMODE - Simulating error while connecting to %s FXS", GetSystemName(system)));
2040 return 0;
2041 }
2042
57f50b3c 2043 // check connection, in case connect
9d733021 2044 if (!Connect(system))
eba76848 2045 {
9d733021 2046 Log(detector, Form("GetFile - Couldn't connect to %s FXS database", GetSystemName(system)));
339adafa 2047 fFXSError = system;
57f50b3c 2048 return 0;
2049 }
2050
2051 // Query preparation
9d733021 2052 TString sourceName(source);
d386d623 2053 Int_t nFields = 3;
2054 TString sqlQueryStart = Form("select filePath,size,fileChecksum from %s where",
2055 fConfig->GetFXSdbTable(system));
2056 TString whereClause = Form("run=%d and detector=\"%s\" and fileId=\"%s\"",
2057 GetCurrentRun(), detector, id);
2058
9d733021 2059 if (system == kDAQ)
2060 {
d386d623 2061 whereClause += Form(" and DAQsource=\"%s\"", source);
57f50b3c 2062 }
9d733021 2063 else if (system == kDCS)
eba76848 2064 {
9d733021 2065 sourceName="none";
57f50b3c 2066 }
9d733021 2067 else if (system == kHLT)
9e080f92 2068 {
d386d623 2069 whereClause += Form(" and DDLnumbers=\"%s\"", source);
9e080f92 2070 }
2071
9e080f92 2072 TString sqlQuery = Form("%s %s", sqlQueryStart.Data(), whereClause.Data());
2073
2074 AliDebug(2, Form("SQL query: \n%s",sqlQuery.Data()));
2075
2076 // Query execution
2077 TSQLResult* aResult = 0;
9d733021 2078 aResult = dynamic_cast<TSQLResult*> (fServer[system]->Query(sqlQuery));
9e080f92 2079 if (!aResult) {
8da81210 2080 Log(detector, Form("GetFile - Can't execute SQL query to %s database for: id = %s, source = %s",
9d733021 2081 GetSystemName(system), id, sourceName.Data()));
339adafa 2082 fFXSError = system;
9e080f92 2083 return 0;
2084 }
2085
8da81210 2086 if (aResult->GetRowCount() == 0)
9e080f92 2087 {
2088 Log(detector,
8da81210 2089 Form("GetFile - No entry in %s FXS db for: id = %s, source = %s",
9d733021 2090 GetSystemName(system), id, sourceName.Data()));
9e080f92 2091 delete aResult;
2092 return 0;
2093 }
2bb7b766 2094
9e080f92 2095 if (aResult->GetRowCount() > 1) {
2096 Log(detector,
8da81210 2097 Form("GetFile - More than one entry in %s FXS db for: id = %s, source = %s",
9d733021 2098 GetSystemName(system), id, sourceName.Data()));
339adafa 2099 fFXSError = system;
9e080f92 2100 delete aResult;
2101 return 0;
2102 }
2103
9d733021 2104 if (aResult->GetFieldCount() != nFields) {
9e080f92 2105 Log(detector,
9d733021 2106 Form("GetFileName - Wrong field count in %s FXS db for: id = %s, source = %s",
2107 GetSystemName(system), id, sourceName.Data()));
339adafa 2108 fFXSError = system;
9e080f92 2109 delete aResult;
2110 return 0;
2111 }
2112
2113 TSQLRow* aRow = dynamic_cast<TSQLRow*> (aResult->Next());
2114
2115 if (!aRow){
8da81210 2116 Log(detector, Form("GetFile - Empty set result in %s FXS db from query: id = %s, source = %s",
9d733021 2117 GetSystemName(system), id, sourceName.Data()));
339adafa 2118 fFXSError = system;
9e080f92 2119 delete aResult;
2120 return 0;
2121 }
2122
2123 TString filePath(aRow->GetField(0), aRow->GetFieldLength(0));
2124 TString fileSize(aRow->GetField(1), aRow->GetFieldLength(1));
d386d623 2125 TString fileChecksum(aRow->GetField(2), aRow->GetFieldLength(2));
9e080f92 2126
2127 delete aResult;
2128 delete aRow;
2129
d386d623 2130 AliDebug(2, Form("filePath = %s; size = %s, fileChecksum = %s",
2131 filePath.Data(), fileSize.Data(), fileChecksum.Data()));
9e080f92 2132
9e080f92 2133 // retrieved file is renamed to make it unique
675f64cd 2134 TString localFileName = Form("%s/%s_%d_process/%s_%s_%d_%s_%s.shuttle",
2135 GetShuttleTempDir(), detector, GetCurrentRun(),
d524ade6 2136 GetSystemName(system), detector, GetCurrentRun(),
2137 id, sourceName.Data());
9d733021 2138
9e080f92 2139
9d733021 2140 // file retrieval from FXS
4b95672b 2141 UInt_t nRetries = 0;
2142 UInt_t maxRetries = 3;
2143 Bool_t result = kFALSE;
2144
2145 // copy!! if successful TSystem::Exec returns 0
8da81210 2146 while (nRetries++ < maxRetries) {
4b95672b 2147 AliDebug(2, Form("Trying to copy file. Retry # %d", nRetries));
2148 result = RetrieveFile(system, filePath.Data(), localFileName.Data());
8da81210 2149 if (!result)
4b95672b 2150 {
8da81210 2151 Log(detector, Form("GetFile - Copy of file %s from %s FXS failed",
9d733021 2152 filePath.Data(), GetSystemName(system)));
4b95672b 2153 continue;
4f0749a8 2154 }
9e080f92 2155
c98417a1 2156 if (fileSize.Length()>0)
2157 {
2158 // compare filesize of local file with the one stored in the FXS DB
926cbc0e 2159 Long_t size = -1;
2160 Int_t sizeComp = gSystem->GetPathInfo(localFileName.Data(), 0, &size, 0, 0);
c98417a1 2161
926cbc0e 2162 if (sizeComp != 0 || size != fileSize.Atoi())
c98417a1 2163 {
8da81210 2164 Log(detector, Form("GetFile - size of file %s does not match with local copy!",
c98417a1 2165 filePath.Data()));
2166 result = kFALSE;
2167 continue;
2168 }
2169
2170 } else {
2171 Log(fCurrentDetector, Form("GetFile - size of file %s not set in %s database, skipping comparison",
2172 filePath.Data(), GetSystemName(system)));
2173 }
2174
d386d623 2175 if (fileChecksum.Length()>0)
4b95672b 2176 {
2177 // compare md5sum of local file with the one stored in the FXS DB
95d4aaef 2178 if(fileChecksum.Contains(' ')) fileChecksum.Resize(fileChecksum.First(' '));
339adafa 2179 Int_t md5Comp = gSystem->Exec(Form("md5sum %s |grep %s > /dev/null 2> /dev/null",
d524ade6 2180 localFileName.Data(), fileChecksum.Data()));
9e080f92 2181
4b95672b 2182 if (md5Comp != 0)
2183 {
8da81210 2184 Log(detector, Form("GetFile - md5sum of file %s does not match with local copy!",
4b95672b 2185 filePath.Data()));
2186 result = kFALSE;
2187 continue;
2188 }
d386d623 2189 } else {
2190 Log(fCurrentDetector, Form("GetFile - md5sum of file %s not set in %s database, skipping comparison",
2191 filePath.Data(), GetSystemName(system)));
9d733021 2192 }
4b95672b 2193 if (result) break;
9e080f92 2194 }
2195
8da81210 2196 if (!result)
2197 {
339adafa 2198 fFXSError = system;
8da81210 2199 return 0;
2200 }
4b95672b 2201
9d733021 2202 fFXSCalled[system]=kTRUE;
2203 TObjString *fileParams = new TObjString(Form("%s#!?!#%s", id, sourceName.Data()));
2204 fFXSlist[system].Add(fileParams);
9e080f92 2205
675f64cd 2206 static TString staticLocalFileName;
2207 staticLocalFileName.Form("%s", localFileName.Data());
2208
c88ad5db 2209 Log(fCurrentDetector, Form("GetFile - Retrieved file with id %s and "
2210 "source %s from %s to %s", id, source,
d524ade6 2211 GetSystemName(system), localFileName.Data()));
675f64cd 2212
d524ade6 2213 return staticLocalFileName.Data();
2bb7b766 2214}
2215
2216//______________________________________________________________________________________________
9d733021 2217Bool_t AliShuttle::RetrieveFile(UInt_t system, const char* fxsFileName, const char* localFileName)
9e080f92 2218{
9827400b 2219 //
2220 // Copies file from FXS to local Shuttle machine
2221 //
2bb7b766 2222
9e080f92 2223 // check temp directory: trying to cd to temp; if it does not exist, create it
d524ade6 2224 AliDebug(2, Form("Copy file %s from %s FXS into %s",
2225 GetSystemName(system), fxsFileName, localFileName));
2226
2227 TString tmpDir(localFileName);
2228
2229 tmpDir = tmpDir(0,tmpDir.Last('/'));
9e080f92 2230
d524ade6 2231 Int_t noDir = gSystem->GetPathInfo(tmpDir.Data(), 0, (Long64_t*) 0, 0, 0);
2232 if (noDir) // temp dir does not exists!
2233 {
2234 if (gSystem->mkdir(tmpDir.Data(), 1))
2235 {
2236 Log(fCurrentDetector.Data(), "RetrieveFile - could not make temp directory!!");
9e080f92 2237 return kFALSE;
2238 }
9e080f92 2239 }
2240
339adafa 2241 TString command = Form("scp -oPort=%d -2 %s@%s:%s/%s %s",
9d733021 2242 fConfig->GetFXSPort(system),
2243 fConfig->GetFXSUser(system),
2244 fConfig->GetFXSHost(system),
339adafa 2245 fConfig->GetFXSBaseFolder(system),
9d733021 2246 fxsFileName,
9e080f92 2247 localFileName);
2248
2249 AliDebug(2, Form("%s",command.Data()));
2250
4b95672b 2251 Bool_t result = (gSystem->Exec(command.Data()) == 0);
9e080f92 2252
4b95672b 2253 return result;
9e080f92 2254}
2255
2256//______________________________________________________________________________________________
9d733021 2257TList* AliShuttle::GetFileSources(Int_t system, const char* detector, const char* id)
2258{
9827400b 2259 //
2260 // Get sources producing the condition file Id from file exchange servers
4a33bdd9 2261 // if id is NULL all sources are returned (distinct)
9827400b 2262 //
1bcd28db 2263
8884d71e 2264 if (id)
2265 {
2266 Log(detector, Form("GetFileSources - Querying %s FXS for files with id %s produced by %s", GetSystemName(system), id, detector));
2267 } else {
2268 Log(detector, Form("GetFileSources - Querying %s FXS for files produced by %s", GetSystemName(system), detector));
2269 }
9827400b 2270
2271 // check if test mode should simulate a FXS error
2272 if (fTestMode & kErrorFXSSources)
2273 {
2274 Log(detector, Form("GetFileSources - In TESTMODE - Simulating error while connecting to %s FXS", GetSystemName(system)));
2275 return 0;
2276 }
2277
9d733021 2278 if (system == kDCS)
2279 {
c88ad5db 2280 Log(detector, "GetFileSources - WARNING: DCS system has only one source of data!");
6297b37d 2281 TList *list = new TList();
2282 list->SetOwner(1);
2283 list->Add(new TObjString(" "));
2284 return list;
9d733021 2285 }
9e080f92 2286
2287 // check connection, in case connect
9d733021 2288 if (!Connect(system))
2289 {
4a33bdd9 2290 Log(detector, Form("GetFileSources - Couldn't connect to %s FXS database", GetSystemName(system)));
339adafa 2291 fFXSError = system;
9d733021 2292 return NULL;
9e080f92 2293 }
2294
9d733021 2295 TString sourceName = 0;
2296 if (system == kDAQ)
2297 {
2298 sourceName = "DAQsource";
2299 } else if (system == kHLT)
2300 {
2301 sourceName = "DDLnumbers";
2302 }
2303
4a33bdd9 2304 TString sqlQueryStart = Form("select distinct %s from %s where", sourceName.Data(), fConfig->GetFXSdbTable(system));
2305 TString whereClause = Form("run=%d and detector=\"%s\"",
2306 GetCurrentRun(), detector);
2307 if (id)
2308 whereClause += Form(" and fileId=\"%s\"", id);
9e080f92 2309 TString sqlQuery = Form("%s %s", sqlQueryStart.Data(), whereClause.Data());
2310
2311 AliDebug(2, Form("SQL query: \n%s",sqlQuery.Data()));
2312
2313 // Query execution
2314 TSQLResult* aResult;
9d733021 2315 aResult = fServer[system]->Query(sqlQuery);
9e080f92 2316 if (!aResult) {
9d733021 2317 Log(detector, Form("GetFileSources - Can't execute SQL query to %s database for id: %s",
2318 GetSystemName(system), id));
339adafa 2319 fFXSError = system;
9e080f92 2320 return 0;
2321 }
2322
86aa42c3 2323 TList *list = new TList();
2324 list->SetOwner(1);
2325
9d733021 2326 if (aResult->GetRowCount() == 0)
2327 {
9e080f92 2328 Log(detector,
9d733021 2329 Form("GetFileSources - No entry in %s FXS table for id: %s", GetSystemName(system), id));
9e080f92 2330 delete aResult;
86aa42c3 2331 return list;
9e080f92 2332 }
2333
1bcd28db 2334 Log(detector, Form("GetFileSources - Found %d sources", aResult->GetRowCount()));
9e080f92 2335
1bcd28db 2336 TSQLRow* aRow;
9d733021 2337 while ((aRow = aResult->Next()))
2338 {
9e080f92 2339
9d733021 2340 TString source(aRow->GetField(0), aRow->GetFieldLength(0));
2341 AliDebug(2, Form("%s = %s", sourceName.Data(), source.Data()));
2342 list->Add(new TObjString(source));
9e080f92 2343 delete aRow;
2344 }
9d733021 2345
9e080f92 2346 delete aResult;
2347
2348 return list;
2bb7b766 2349}
2350
4a33bdd9 2351//______________________________________________________________________________________________
2352TList* AliShuttle::GetFileIDs(Int_t system, const char* detector, const char* source)
2353{
2354 //
2355 // Get all ids of condition files produced by a given source from file exchange servers
2356 //
2357
1bcd28db 2358 Log(detector, Form("GetFileIDs - Retrieving ids with source %s with %s", source, GetSystemName(system)));
2359
4a33bdd9 2360 // check if test mode should simulate a FXS error
2361 if (fTestMode & kErrorFXSSources)
2362 {
2363 Log(detector, Form("GetFileIDs - In TESTMODE - Simulating error while connecting to %s FXS", GetSystemName(system)));
2364 return 0;
2365 }
2366
2367 // check connection, in case connect
2368 if (!Connect(system))
2369 {
2370 Log(detector, Form("GetFileIDs - Couldn't connect to %s FXS database", GetSystemName(system)));
2371 return NULL;
2372 }
2373
2374 TString sourceName = 0;
2375 if (system == kDAQ)
2376 {
2377 sourceName = "DAQsource";
2378 } else if (system == kHLT)
2379 {
2380 sourceName = "DDLnumbers";
2381 }
2382
2383 TString sqlQueryStart = Form("select fileId from %s where", fConfig->GetFXSdbTable(system));
2384 TString whereClause = Form("run=%d and detector=\"%s\"",
2385 GetCurrentRun(), detector);
2386 if (sourceName.Length() > 0 && source)
2387 whereClause += Form(" and %s=\"%s\"", sourceName.Data(), source);
2388 TString sqlQuery = Form("%s %s", sqlQueryStart.Data(), whereClause.Data());
2389
2390 AliDebug(2, Form("SQL query: \n%s",sqlQuery.Data()));
2391
2392 // Query execution
2393 TSQLResult* aResult;
2394 aResult = fServer[system]->Query(sqlQuery);
2395 if (!aResult) {
2396 Log(detector, Form("GetFileIDs - Can't execute SQL query to %s database for source: %s",
2397 GetSystemName(system), source));
2398 return 0;
2399 }
2400
2401 TList *list = new TList();
2402 list->SetOwner(1);
2403
2404 if (aResult->GetRowCount() == 0)
2405 {
2406 Log(detector,
2407 Form("GetFileIDs - No entry in %s FXS table for source: %s", GetSystemName(system), source));
2408 delete aResult;
2409 return list;
2410 }
2411
1bcd28db 2412 Log(detector, Form("GetFileIDs - Found %d ids", aResult->GetRowCount()));
2413
4a33bdd9 2414 TSQLRow* aRow;
2415
2416 while ((aRow = aResult->Next()))
2417 {
2418
2419 TString id(aRow->GetField(0), aRow->GetFieldLength(0));
2420 AliDebug(2, Form("fileId = %s", id.Data()));
2421 list->Add(new TObjString(id));
2422 delete aRow;
2423 }
2424
2425 delete aResult;
2426
2427 return list;
2428}
2429
2bb7b766 2430//______________________________________________________________________________________________
9d733021 2431Bool_t AliShuttle::Connect(Int_t system)
2bb7b766 2432{
9827400b 2433 // Connect to MySQL Server of the system's FXS MySQL databases
2434 // DAQ Logbook, Shuttle Logbook and DAQ FXS db are on the same host
2435 //
57f50b3c 2436
9d733021 2437 // check connection: if already connected return
2438 if(fServer[system] && fServer[system]->IsConnected()) return kTRUE;
57f50b3c 2439
9d733021 2440 TString dbHost, dbUser, dbPass, dbName;
57f50b3c 2441
9d733021 2442 if (system < 3) // FXS db servers
2443 {
2444 dbHost = Form("mysql://%s:%d", fConfig->GetFXSdbHost(system), fConfig->GetFXSdbPort(system));
2445 dbUser = fConfig->GetFXSdbUser(system);
2446 dbPass = fConfig->GetFXSdbPass(system);
2447 dbName = fConfig->GetFXSdbName(system);
2448 } else { // Run & Shuttle logbook servers
2449 // TODO Will the Shuttle logbook server be the same as the Run logbook server ???
2450 dbHost = Form("mysql://%s:%d", fConfig->GetDAQlbHost(), fConfig->GetDAQlbPort());
2451 dbUser = fConfig->GetDAQlbUser();
2452 dbPass = fConfig->GetDAQlbPass();
2453 dbName = fConfig->GetDAQlbDB();
2454 }
57f50b3c 2455
9d733021 2456 fServer[system] = TSQLServer::Connect(dbHost.Data(), dbUser.Data(), dbPass.Data());
2457 if (!fServer[system] || !fServer[system]->IsConnected()) {
2458 if(system < 3)
2459 {
2460 AliError(Form("Can't establish connection to FXS database for %s",
2461 AliShuttleInterface::GetSystemName(system)));
2462 } else {
2463 AliError("Can't establish connection to Run logbook.");
57f50b3c 2464 }
9d733021 2465 if(fServer[system]) delete fServer[system];
2466 return kFALSE;
2bb7b766 2467 }
57f50b3c 2468
9d733021 2469 // Get tables
2470 TSQLResult* aResult=0;
2471 switch(system){
2472 case kDAQ:
2473 aResult = fServer[kDAQ]->GetTables(dbName.Data());
2474 break;
2475 case kDCS:
2476 aResult = fServer[kDCS]->GetTables(dbName.Data());
2477 break;
2478 case kHLT:
2479 aResult = fServer[kHLT]->GetTables(dbName.Data());
2480 break;
2481 default:
2482 aResult = fServer[3]->GetTables(dbName.Data());
2483 break;
2484 }
2485
2486 delete aResult;
2bb7b766 2487 return kTRUE;
2488}
57f50b3c 2489
9e080f92 2490//______________________________________________________________________________________________
9d733021 2491Bool_t AliShuttle::UpdateTable()
9e080f92 2492{
9827400b 2493 //
2494 // Update FXS table filling time_processed field in all rows corresponding to current run and detector
2495 //
9e080f92 2496
9d733021 2497 Bool_t result = kTRUE;
9e080f92 2498
9d733021 2499 for (UInt_t system=0; system<3; system++)
2500 {
2501 if(!fFXSCalled[system]) continue;
9e080f92 2502
9d733021 2503 // check connection, in case connect
2504 if (!Connect(system))
2505 {
2506 Log(fCurrentDetector, Form("UpdateTable - Couldn't connect to %s FXS database", GetSystemName(system)));
2507 result = kFALSE;
2508 continue;
9e080f92 2509 }
9e080f92 2510
9d733021 2511 TTimeStamp now; // now
2512
2513 // Loop on FXS list entries
2514 TIter iter(&fFXSlist[system]);
2515 TObjString *aFXSentry=0;
2516 while ((aFXSentry = dynamic_cast<TObjString*> (iter.Next())))
2517 {
2518 TString aFXSentrystr = aFXSentry->String();
2519 TObjArray *aFXSarray = aFXSentrystr.Tokenize("#!?!#");
2520 if (!aFXSarray || aFXSarray->GetEntries() != 2 )
2521 {
2522 Log(fCurrentDetector, Form("UpdateTable - error updating %s FXS entry. Check string: <%s>",
2523 GetSystemName(system), aFXSentrystr.Data()));
2524 if(aFXSarray) delete aFXSarray;
2525 result = kFALSE;
2526 continue;
2527 }
2528 const char* fileId = ((TObjString*) aFXSarray->At(0))->GetName();
2529 const char* source = ((TObjString*) aFXSarray->At(1))->GetName();
2530
2531 TString whereClause;
2532 if (system == kDAQ)
2533 {
2534 whereClause = Form("where run=%d and detector=\"%s\" and fileId=\"%s\" and DAQsource=\"%s\";",
2535 GetCurrentRun(), fCurrentDetector.Data(), fileId, source);
2536 }
2537 else if (system == kDCS)
2538 {
2539 whereClause = Form("where run=%d and detector=\"%s\" and fileId=\"%s\";",
2540 GetCurrentRun(), fCurrentDetector.Data(), fileId);
2541 }
2542 else if (system == kHLT)
2543 {
2544 whereClause = Form("where run=%d and detector=\"%s\" and fileId=\"%s\" and DDLnumbers=\"%s\";",
2545 GetCurrentRun(), fCurrentDetector.Data(), fileId, source);
2546 }
2547
2548 delete aFXSarray;
9e080f92 2549
9d733021 2550 TString sqlQuery = Form("update %s set time_processed=%d %s", fConfig->GetFXSdbTable(system),
2551 now.GetSec(), whereClause.Data());
9e080f92 2552
9d733021 2553 AliDebug(2, Form("SQL query: \n%s",sqlQuery.Data()));
9e080f92 2554
9d733021 2555 // Query execution
2556 TSQLResult* aResult;
2557 aResult = dynamic_cast<TSQLResult*> (fServer[system]->Query(sqlQuery));
2558 if (!aResult)
2559 {
2560 Log(fCurrentDetector, Form("UpdateTable - %s db: can't execute SQL query <%s>",
2561 GetSystemName(system), sqlQuery.Data()));
2562 result = kFALSE;
2563 continue;
2564 }
2565 delete aResult;
9e080f92 2566 }
9e080f92 2567 }
2568
9d733021 2569 return result;
9e080f92 2570}
57f50b3c 2571
3301427a 2572//______________________________________________________________________________________________
2573Bool_t AliShuttle::UpdateTableFailCase()
2574{
9827400b 2575 // Update FXS table filling time_processed field in all rows corresponding to current run and detector
2576 // this is called in case the preprocessor is declared failed for the current run, because
2577 // the fields are updated only in case of success
3301427a 2578
2579 Bool_t result = kTRUE;
2580
2581 for (UInt_t system=0; system<3; system++)
2582 {
2583 // check connection, in case connect
2584 if (!Connect(system))
2585 {
2586 Log(fCurrentDetector, Form("UpdateTableFailCase - Couldn't connect to %s FXS database",
2587 GetSystemName(system)));
2588 result = kFALSE;
2589 continue;
2590 }
2591
2592 TTimeStamp now; // now
2593
2594 // Loop on FXS list entries
2595
2596 TString whereClause = Form("where run=%d and detector=\"%s\";",
2597 GetCurrentRun(), fCurrentDetector.Data());
2598
2599
2600 TString sqlQuery = Form("update %s set time_processed=%d %s", fConfig->GetFXSdbTable(system),
2601 now.GetSec(), whereClause.Data());
2602
2603 AliDebug(2, Form("SQL query: \n%s",sqlQuery.Data()));
2604
2605 // Query execution
2606 TSQLResult* aResult;
2607 aResult = dynamic_cast<TSQLResult*> (fServer[system]->Query(sqlQuery));
2608 if (!aResult)
2609 {
2610 Log(fCurrentDetector, Form("UpdateTableFailCase - %s db: can't execute SQL query <%s>",
2611 GetSystemName(system), sqlQuery.Data()));
2612 result = kFALSE;
2613 continue;
2614 }
2615 delete aResult;
2616 }
2617
2618 return result;
2619}
2620
2bb7b766 2621//______________________________________________________________________________________________
2622Bool_t AliShuttle::UpdateShuttleLogbook(const char* detector, const char* status)
2623{
e7f62f16 2624 //
2625 // Update Shuttle logbook filling detector or shuttle_done column
2626 // ex. of usage: UpdateShuttleLogbook("PHOS", "DONE") or UpdateShuttleLogbook("shuttle_done")
2627 //
57f50b3c 2628
2bb7b766 2629 // check connection, in case connect
be48e3ea 2630 if(!Connect(3)){
2bb7b766 2631 Log("SHUTTLE", "UpdateShuttleLogbook - Couldn't connect to DAQ Logbook.");
2632 return kFALSE;
57f50b3c 2633 }
2634
2bb7b766 2635 TString detName(detector);
2636 TString setClause;
8e828d39 2637 if (detName == "shuttle_done" || detName == "shuttle_skipped")
e7f62f16 2638 {
2bb7b766 2639 setClause = "set shuttle_done=1";
8e828d39 2640
ee6f7523 2641 if (detName == "shuttle_done")
b0e53b15 2642 {
8e828d39 2643 if (TouchFile() != kTRUE)
2156c755 2644 {
2645 SendMLRunInfo("Pending");
503ff24f 2646 return kFALSE;
2156c755 2647 }
8e828d39 2648
2649 SendMLRunInfo("Done");
b0e53b15 2650 }
8e828d39 2651 else
2652 SendMLRunInfo("Skipped");
2653 }
2654 else {
2bb7b766 2655 TString statusStr(status);
2656 if(statusStr.Contains("done", TString::kIgnoreCase) ||
2657 statusStr.Contains("failed", TString::kIgnoreCase)){
eba76848 2658 setClause = Form("set %s=\"%s\"", detector, status);
2bb7b766 2659 } else {
2660 Log("SHUTTLE",
2661 Form("UpdateShuttleLogbook - Invalid status <%s> for detector %s",
2662 status, detector));
2663 return kFALSE;
2664 }
2665 }
57f50b3c 2666
2bb7b766 2667 TString whereClause = Form("where run=%d", GetCurrentRun());
2668
441b0e9c 2669 TString sqlQuery = Form("update %s %s %s",
2670 fConfig->GetShuttlelbTable(), setClause.Data(), whereClause.Data());
57f50b3c 2671
2bb7b766 2672 AliDebug(2, Form("SQL query: \n%s",sqlQuery.Data()));
2673
2674 // Query execution
2675 TSQLResult* aResult;
be48e3ea 2676 aResult = dynamic_cast<TSQLResult*> (fServer[3]->Query(sqlQuery));
2bb7b766 2677 if (!aResult) {
2678 Log("SHUTTLE", Form("UpdateShuttleLogbook - Can't execute query <%s>", sqlQuery.Data()));
2679 return kFALSE;
57f50b3c 2680 }
2bb7b766 2681 delete aResult;
57f50b3c 2682
2683 return kTRUE;
2684}
2685
2686//______________________________________________________________________________________________
2bb7b766 2687Int_t AliShuttle::GetCurrentRun() const
2688{
9827400b 2689 //
2690 // Get current run from logbook entry
2691 //
57f50b3c 2692
2bb7b766 2693 return fLogbookEntry ? fLogbookEntry->GetRun() : -1;
57f50b3c 2694}
2695
2696//______________________________________________________________________________________________
2bb7b766 2697UInt_t AliShuttle::GetCurrentStartTime() const
2698{
9827400b 2699 //
2700 // get current start time
2701 //
57f50b3c 2702
2bb7b766 2703 return fLogbookEntry ? fLogbookEntry->GetStartTime() : 0;
57f50b3c 2704}
2705
2706//______________________________________________________________________________________________
2bb7b766 2707UInt_t AliShuttle::GetCurrentEndTime() const
2708{
9827400b 2709 //
2710 // get current end time from logbook entry
2711 //
57f50b3c 2712
2bb7b766 2713 return fLogbookEntry ? fLogbookEntry->GetEndTime() : 0;
57f50b3c 2714}
2715
675f64cd 2716//______________________________________________________________________________________________
2717UInt_t AliShuttle::GetCurrentYear() const
2718{
2719 //
2720 // Get current year from logbook entry
2721 //
2722
2723 if (!fLogbookEntry) return 0;
2724
2725 TTimeStamp startTime(GetCurrentStartTime());
2726 TString year = Form("%d",startTime.GetDate());
2727 year = year(0,4);
2728
2729 return year.Atoi();
2730}
2731
2732//______________________________________________________________________________________________
2733const char* AliShuttle::GetLHCPeriod() const
2734{
2735 //
2736 // Get current LHC period from logbook entry
2737 //
2738
2739 if (!fLogbookEntry) return 0;
2740
2741 return fLogbookEntry->GetRunParameter("LHCperiod");
2742}
2743
b948db8d 2744//______________________________________________________________________________________________
2745void AliShuttle::Log(const char* detector, const char* message)
2746{
9827400b 2747 //
2748 // Fill log string with a message
2749 //
b948db8d 2750
7d4cf768 2751 TString logRunDir = GetShuttleLogDir();
2752 if (GetCurrentRun() >=0)
2753 logRunDir += Form("/%d", GetCurrentRun());
2754
2755 void* dir = gSystem->OpenDirectory(logRunDir.Data());
84090f85 2756 if (dir == NULL) {
7d4cf768 2757 if (gSystem->mkdir(logRunDir.Data(), kTRUE)) {
36c99a6a 2758 AliError(Form("Can't open directory <%s>", GetShuttleLogDir()));
84090f85 2759 return;
2760 }
b948db8d 2761
84090f85 2762 } else {
2763 gSystem->FreeDirectory(dir);
2764 }
b948db8d 2765
0b025a52 2766 TString toLog = Form("%s UTC (%d): %s - ", TTimeStamp(time(0)).AsString("s"), getpid(), detector);
e7f62f16 2767 if (GetCurrentRun() >= 0)
2768 toLog += Form("run %d - ", GetCurrentRun());
2bb7b766 2769 toLog += Form("%s", message);
2770
84090f85 2771 AliInfo(toLog.Data());
ffa29e93 2772
2773 // if we redirect the log output already to the file, leave here
2774 if (fOutputRedirected && strcmp(detector, "SHUTTLE") != 0)
2775 return;
b948db8d 2776
ffa29e93 2777 TString fileName = GetLogFileName(detector);
e7f62f16 2778
84090f85 2779 gSystem->ExpandPathName(fileName);
2780
2781 ofstream logFile;
2782 logFile.open(fileName, ofstream::out | ofstream::app);
2783
2784 if (!logFile.is_open()) {
2785 AliError(Form("Could not open file %s", fileName.Data()));
2786 return;
2787 }
7bfb2090 2788
84090f85 2789 logFile << toLog.Data() << "\n";
b948db8d 2790
84090f85 2791 logFile.close();
b948db8d 2792}
2bb7b766 2793
ffa29e93 2794//______________________________________________________________________________________________
2795TString AliShuttle::GetLogFileName(const char* detector) const
2796{
2797 //
2798 // returns the name of the log file for a given sub detector
2799 //
2800
2801 TString fileName;
2802
2803 if (GetCurrentRun() >= 0)
7d4cf768 2804 {
2805 fileName.Form("%s/%d/%s_%d.log", GetShuttleLogDir(), GetCurrentRun(),
2806 detector, GetCurrentRun());
2807 } else {
ffa29e93 2808 fileName.Form("%s/%s.log", GetShuttleLogDir(), detector);
7d4cf768 2809 }
ffa29e93 2810
2811 return fileName;
2812}
2813
ee6f7523 2814//______________________________________________________________________________________________
2815void AliShuttle::SendAlive()
2816{
2817 // sends alive message to ML
2818
2819 TMonaLisaText mlStatus("SHUTTLE_status", "Alive");
2820
2821 TList mlList;
2822 mlList.Add(&mlStatus);
2823
2824 fMonaLisa->SendParameters(&mlList, "__PROCESSINGINFO__");
2825}
2826
2bb7b766 2827//______________________________________________________________________________________________
2828Bool_t AliShuttle::Collect(Int_t run)
2829{
9827400b 2830 //
2831 // Collects conditions data for all UNPROCESSED run written to DAQ LogBook in case of run = -1 (default)
2832 // If a dedicated run is given this run is processed
2833 //
2834 // In operational mode, this is the Shuttle function triggered by the EOR signal.
2835 //
2bb7b766 2836
eba76848 2837 if (run == -1)
2838 Log("SHUTTLE","Collect - Shuttle called. Collecting conditions data for unprocessed runs");
2839 else
2840 Log("SHUTTLE", Form("Collect - Shuttle called. Collecting conditions data for run %d", run));
cb343cfd 2841
2842 SetLastAction("Starting");
2bb7b766 2843
ee6f7523 2844 // create ML instance
2845 if (!fMonaLisa)
2846 fMonaLisa = new TMonaLisaWriter(fConfig->GetMonitorHost(), fConfig->GetMonitorTable());
2847
ee6f7523 2848 SendAlive();
fbc112e3 2849 CountOpenRuns();
ee6f7523 2850
2bb7b766 2851 TString whereClause("where shuttle_done=0");
eba76848 2852 if (run != -1)
2853 whereClause += Form(" and run=%d", run);
2bb7b766 2854
2855 TObjArray shuttleLogbookEntries;
be48e3ea 2856 if (!QueryShuttleLogbook(whereClause, shuttleLogbookEntries))
2857 {
cb343cfd 2858 Log("SHUTTLE", "Collect - Can't retrieve entries from Shuttle logbook");
2bb7b766 2859 return kFALSE;
2860 }
2861
9e080f92 2862 if (shuttleLogbookEntries.GetEntries() == 0)
2863 {
2864 if (run == -1)
2865 Log("SHUTTLE","Collect - Found no UNPROCESSED runs in Shuttle logbook");
2866 else
2867 Log("SHUTTLE", Form("Collect - Run %d is already DONE "
2868 "or it does not exist in Shuttle logbook", run));
2869 return kTRUE;
2870 }
2871
be48e3ea 2872 for (UInt_t iDet=0; iDet<NDetectors(); iDet++)
2873 fFirstUnprocessed[iDet] = kTRUE;
2874
fc5a4708 2875 if (run != -1)
be48e3ea 2876 {
2877 // query Shuttle logbook for earlier runs, check if some detectors are unprocessed,
2878 // flag them into fFirstUnprocessed array
e3ae8d33 2879 TString whereClauseBis(Form("where shuttle_done=0 and run < %d", run));
be48e3ea 2880 TObjArray tmpLogbookEntries;
e3ae8d33 2881 if (!QueryShuttleLogbook(whereClauseBis, tmpLogbookEntries))
be48e3ea 2882 {
2883 Log("SHUTTLE", "Collect - Can't retrieve entries from Shuttle logbook");
2884 return kFALSE;
2885 }
2886
2887 TIter iter(&tmpLogbookEntries);
2888 AliShuttleLogbookEntry* anEntry = 0;
2889 while ((anEntry = dynamic_cast<AliShuttleLogbookEntry*> (iter.Next())))
2890 {
2891 for (UInt_t iDet=0; iDet<NDetectors(); iDet++)
2892 {
2893 if (anEntry->GetDetectorStatus(iDet) == AliShuttleLogbookEntry::kUnprocessed)
2894 {
2895 AliDebug(2, Form("Run %d: setting %s as \"not first time unprocessed\"",
2896 anEntry->GetRun(), GetDetName(iDet)));
2897 fFirstUnprocessed[iDet] = kFALSE;
2898 }
2899 }
2900
2901 }
2902
2903 }
2904
2905 if (!RetrieveConditionsData(shuttleLogbookEntries))
2906 {
cb343cfd 2907 Log("SHUTTLE", "Collect - Process of at least one run failed");
a3ca69fe 2908 CountOpenRuns();
2bb7b766 2909 return kFALSE;
2910 }
2911
36c99a6a 2912 Log("SHUTTLE", "Collect - Requested run(s) successfully processed");
a3ca69fe 2913 CountOpenRuns();
eba76848 2914 return kTRUE;
2bb7b766 2915}
2916
2bb7b766 2917//______________________________________________________________________________________________
2918Bool_t AliShuttle::RetrieveConditionsData(const TObjArray& dateEntries)
2919{
9827400b 2920 //
2921 // Retrieve conditions data for all runs that aren't processed yet
2922 //
2bb7b766 2923
2924 Bool_t hasError = kFALSE;
2925
2926 TIter iter(&dateEntries);
2927 AliShuttleLogbookEntry* anEntry;
2928
2929 while ((anEntry = (AliShuttleLogbookEntry*) iter.Next())){
2930 if (!Process(anEntry)){
2931 hasError = kTRUE;
2932 }
4b95672b 2933
2934 // clean SHUTTLE temp directory
d524ade6 2935 //TString filename = Form("%s/*.shuttle", GetShuttleTempDir());
2936 //RemoveFile(filename.Data());
2bb7b766 2937 }
2938
2939 return hasError == kFALSE;
2940}
cb343cfd 2941
2942//______________________________________________________________________________________________
2943ULong_t AliShuttle::GetTimeOfLastAction() const
2944{
9827400b 2945 //
2946 // Gets time of last action
2947 //
2948
cb343cfd 2949 ULong_t tmp;
36c99a6a 2950
cb343cfd 2951 fMonitoringMutex->Lock();
be48e3ea 2952
cb343cfd 2953 tmp = fLastActionTime;
36c99a6a 2954
cb343cfd 2955 fMonitoringMutex->UnLock();
36c99a6a 2956
cb343cfd 2957 return tmp;
2958}
2959
2960//______________________________________________________________________________________________
2961const TString AliShuttle::GetLastAction() const
2962{
9827400b 2963 //
cb343cfd 2964 // returns a string description of the last action
9827400b 2965 //
cb343cfd 2966
2967 TString tmp;
36c99a6a 2968
cb343cfd 2969 fMonitoringMutex->Lock();
2970
2971 tmp = fLastAction;
2972
2973 fMonitoringMutex->UnLock();
2974
36c99a6a 2975 return tmp;
cb343cfd 2976}
2977
2978//______________________________________________________________________________________________
2979void AliShuttle::SetLastAction(const char* action)
2980{
9827400b 2981 //
cb343cfd 2982 // updates the monitoring variables
9827400b 2983 //
36c99a6a 2984
cb343cfd 2985 fMonitoringMutex->Lock();
36c99a6a 2986
cb343cfd 2987 fLastAction = action;
2988 fLastActionTime = time(0);
2989
2990 fMonitoringMutex->UnLock();
2991}
eba76848 2992
2993//______________________________________________________________________________________________
2994const char* AliShuttle::GetRunParameter(const char* param)
2995{
9827400b 2996 //
2997 // returns run parameter read from DAQ logbook
2998 //
eba76848 2999
3000 if(!fLogbookEntry) {
3001 AliError("No logbook entry!");
3002 return 0;
3003 }
3004
3005 return fLogbookEntry->GetRunParameter(param);
3006}
57c1a579 3007
d386d623 3008//______________________________________________________________________________________________
9827400b 3009AliCDBEntry* AliShuttle::GetFromOCDB(const char* detector, const AliCDBPath& path)
d386d623 3010{
9827400b 3011 //
3012 // returns object from OCDB valid for current run
3013 //
d386d623 3014
9827400b 3015 if (fTestMode & kErrorOCDB)
3016 {
3017 Log(detector, "GetFromOCDB - In TESTMODE - Simulating error with OCDB");
3018 return 0;
3019 }
3020
d386d623 3021 AliCDBStorage *sto = AliCDBManager::Instance()->GetStorage(fgkMainCDB);
3022 if (!sto)
3023 {
9827400b 3024 Log(detector, "GetFromOCDB - Cannot activate main OCDB for query!");
d386d623 3025 return 0;
3026 }
3027
3028 return dynamic_cast<AliCDBEntry*> (sto->Get(path, GetCurrentRun()));
3029}
3030
57c1a579 3031//______________________________________________________________________________________________
339adafa 3032Bool_t AliShuttle::SendMail(EMailTarget target, Int_t system)
57c1a579 3033{
9827400b 3034 //
3035 // sends a mail to the subdetector expert in case of preprocessor error
3036 //
3037
3038 if (fTestMode != kNone)
3039 return kTRUE;
fb40ead4 3040
926cbc0e 3041 if (!fConfig->SendMail())
3042 return kTRUE;
57c1a579 3043
339adafa 3044 if (target == kDCSEMail || target == kFXSEMail) {
3045 if (!fFirstProcessing)
3046 return kTRUE;
36c99a6a 3047 }
3048
7bdfe1fa 3049 Int_t runMode = (Int_t)fConfig->GetRunMode();
3050 TString tmpStr;
3051 if (runMode == 0) tmpStr = " Nightly Test:";
3052 else tmpStr = " Data Taking:";
fbc112e3 3053 void* dir = gSystem->OpenDirectory(GetShuttleLogDir());
3054 if (dir == NULL)
3055 {
3056 if (gSystem->mkdir(GetShuttleLogDir(), kTRUE))
3057 {
3058 Log("SHUTTLE", Form("SendMail - Can't open directory <%s>", GetShuttleLogDir()));
3059 return kFALSE;
3060 }
3061
3062 } else {
3063 gSystem->FreeDirectory(dir);
3064 }
3065
339adafa 3066 // det experts in to
3067 TString to="";
3835a613 3068 TIter *iterExperts = 0;
339adafa 3069 if (target == kDCSEMail) {
3070 iterExperts = new TIter(fConfig->GetAdmins(AliShuttleConfig::kAmanda));
3071 }
3072 else if (target == kFXSEMail) {
3073 iterExperts = new TIter(fConfig->GetAdmins(system));
3074 }
3835a613 3075 if (iterExperts) {
3076 TObjString *anExpert=0;
3077 while ((anExpert = (TObjString*) iterExperts->Next()))
3078 {
3079 to += Form("%s,", anExpert->GetName());
3080 }
3081 delete iterExperts;
339adafa 3082 }
3835a613 3083
3084 // add subdetector experts
3085 iterExperts = new TIter(fConfig->GetResponsibles(fCurrentDetector));
339adafa 3086 TObjString *anExpert=0;
3087 while ((anExpert = (TObjString*) iterExperts->Next()))
4508ab8b 3088 {
339adafa 3089 to += Form("%s,", anExpert->GetName());
3090 }
3091 delete iterExperts;
3835a613 3092
339adafa 3093 if (to.Length() > 0)
3094 to.Remove(to.Length()-1);
3095 AliDebug(2, Form("to: %s",to.Data()));
4508ab8b 3096
339adafa 3097 if (to.IsNull()) {
3098 Log("SHUTTLE", Form("List of %d responsibles not set!", (Int_t) target));
3099 return kFALSE;
3100 }
3101
3102 // SHUTTLE responsibles in cc
fb40ead4 3103 TString cc="";
3104 TIter iterAdmins(fConfig->GetAdmins(AliShuttleConfig::kGlobal));
3105 TObjString *anAdmin=0;
3106 while ((anAdmin = (TObjString*) iterAdmins.Next()))
3107 {
3108 cc += Form("%s,", anAdmin->GetName());
3109 }
3110 if (cc.Length() > 0)
c19963db 3111 cc.Remove(cc.Length()-1);
fb40ead4 3112 AliDebug(2, Form("cc: %s",to.Data()));
57c1a579 3113
339adafa 3114 // mail body
675f64cd 3115 TString bodyFileName;
3116 bodyFileName.Form("%s/mail.body", GetShuttleLogDir());
3117 gSystem->ExpandPathName(bodyFileName);
3118
3119 ofstream mailBody;
3120 mailBody.open(bodyFileName, ofstream::out);
3121
3122 if (!mailBody.is_open())
3123 {
339adafa 3124 Log("SHUTTLE", Form("Could not open mail body file %s", bodyFileName.Data()));
675f64cd 3125 return kFALSE;
3126 }
3127
675f64cd 3128
339adafa 3129 TString subject;
3130 TString body;
675f64cd 3131
339adafa 3132 if (target == kDCSEMail){
af93e5d7 3133 subject = Form("%s CRITICAL Retrieval of data points for %s FAILED in run %d !",
7bdfe1fa 3134 tmpStr.Data(), fCurrentDetector.Data(), GetCurrentRun());
339adafa 3135 AliDebug(2, Form("subject: %s", subject.Data()));
3136
3137 body = Form("Dear DCS experts, \n\n");
3138 body += Form("SHUTTLE couldn\'t retrieve the data points for detector %s "
3139 "in run %d!!\n\n", fCurrentDetector.Data(), GetCurrentRun());
fb40ead4 3140 }
339adafa 3141 else if (target == kFXSEMail){
af93e5d7 3142 subject = Form("%s CRITICAL FXS communication for %s FAILED in run %d !",
7bdfe1fa 3143 tmpStr.Data(), fCurrentDetector.Data(), GetCurrentRun());
339adafa 3144 AliDebug(2, Form("subject: %s", subject.Data()));
3145 TString sys;
3146 if (system == kDAQ) sys="DAQ";
3147 else if (system == kDCS) sys="DCS";
3148 else if (system == kHLT) sys="HLT";
3149 else return kFALSE;
9c0e2d64 3150 body = Form("Dear %s FXS experts, \n\n",sys.Data());
339adafa 3151 body += Form("SHUTTLE couldn\'t retrieve data from the FXS for detector %s "
9c0e2d64 3152 "in run %d!!\n\n", fCurrentDetector.Data(), GetCurrentRun());
339adafa 3153 }
3154 else {
7bdfe1fa 3155 subject = Form("%s %s Shuttle preprocessor FAILED in run %d (run type = %s)!",
3156 tmpStr.Data(), fCurrentDetector.Data(), GetCurrentRun(), GetRunType());
339adafa 3157 AliDebug(2, Form("subject: %s", subject.Data()));
3158
3159 body = Form("Dear %s expert(s), \n\n", fCurrentDetector.Data());
3160 body += Form("SHUTTLE just detected that your preprocessor "
3161 "failed processing run %d (run type = %s)!!\n\n",
3162 GetCurrentRun(), GetRunType());
3163 }
675f64cd 3164
7d4cf768 3165 body += Form("Please check %s status on the SHUTTLE monitoring page: \n\n",
3166 fCurrentDetector.Data());
b0e53b15 3167 if (fConfig->GetRunMode() == AliShuttleConfig::kTest)
3168 {
8e828d39 3169 body += Form("\thttp://pcalimonitor.cern.ch/shuttle.jsp?time=24 \n\n");
b0e53b15 3170 } else {
8e828d39 3171 body += Form("\thttp://pcalimonitor.cern.ch/shuttle.jsp?instance=PROD&time=24 \n\n");
b0e53b15 3172 }
339adafa 3173
3174
7d4cf768 3175 TString logFolder = "logs";
3176 if (fConfig->GetRunMode() == AliShuttleConfig::kProd)
3177 logFolder += "_PROD";
3178
3179
675f64cd 3180 body += Form("Find the %s log for the current run on \n\n"
7d4cf768 3181 "\thttp://pcalishuttle01.cern.ch:8880/%s/%d/%s_%d.log \n\n",
3182 fCurrentDetector.Data(), logFolder.Data(), GetCurrentRun(),
3183 fCurrentDetector.Data(), GetCurrentRun());
3184 body += Form("The last 10 lines of %s log file are following:\n\n", fCurrentDetector.Data());
675f64cd 3185
3186 AliDebug(2, Form("Body begin: %s", body.Data()));
3187
3188 mailBody << body.Data();
3189 mailBody.close();
3190 mailBody.open(bodyFileName, ofstream::out | ofstream::app);
3191
339adafa 3192 TString logFileName = Form("%s/%d/%s_%d.log", GetShuttleLogDir(),
3193 GetCurrentRun(), fCurrentDetector.Data(), GetCurrentRun());
675f64cd 3194 TString tailCommand = Form("tail -n 10 %s >> %s", logFileName.Data(), bodyFileName.Data());
3195 if (gSystem->Exec(tailCommand.Data()))
3196 {
3197 mailBody << Form("%s log file not found ...\n\n", fCurrentDetector.Data());
3198 }
3199
3200 TString endBody = Form("------------------------------------------------------\n\n");
36c99a6a 3201 endBody += Form("In case of problems please contact the SHUTTLE core team.\n\n");
3202 endBody += "Please do not answer this message directly, it is automatically generated.\n\n";
546242fb 3203 endBody += "Greetings,\n\n \t\t\tthe SHUTTLE\n";
57c1a579 3204
909732f7 3205 AliDebug(2, Form("Body end: %s", endBody.Data()));
57c1a579 3206
3207 mailBody << endBody.Data();
3208
3209 mailBody.close();
3210
3211 // send mail!
3212 TString mailCommand = Form("mail -s \"%s\" -c %s %s < %s",
3213 subject.Data(),
3214 cc.Data(),
3215 to.Data(),
3216 bodyFileName.Data());
909732f7 3217 AliDebug(2, Form("mail command: %s", mailCommand.Data()));
57c1a579 3218
3219 Bool_t result = gSystem->Exec(mailCommand.Data());
3220
3221 return result == 0;
3222}
d386d623 3223
441b0e9c 3224//______________________________________________________________________________________________
9827400b 3225const char* AliShuttle::GetRunType()
441b0e9c 3226{
9827400b 3227 //
3228 // returns run type read from "run type" logbook
3229 //
441b0e9c 3230
3231 if(!fLogbookEntry) {
3232 AliError("No logbook entry!");
3233 return 0;
3234 }
3235
9827400b 3236 return fLogbookEntry->GetRunType();
441b0e9c 3237}
3238
4859271b 3239//______________________________________________________________________________________________
3240Bool_t AliShuttle::GetHLTStatus()
3241{
3242 // Return HLT status (ON=1 OFF=0)
af76b20d 3243 // Converts the HLT status from the mode string read in the run logbook (not just a bool)
4859271b 3244
3245 if(!fLogbookEntry) {
3246 AliError("No logbook entry!");
3247 return 0;
3248 }
3249
af76b20d 3250 // TODO implement when HLTMode is inserted in run logbook
b4ebc2d5 3251 TString hltMode = fLogbookEntry->GetRunParameter("HLTmode");
af76b20d 3252 TSubString firstChar = hltMode(0,1);
3253 AliDebug(2,Form("First char = %s ",firstChar.Data()));
3254 if (firstChar == "A") {
3255 return kFALSE;
3256 }
3257 else if ((firstChar == "B") || (firstChar == "C") || (firstChar == "D") || (firstChar == "E")) {
3258 return kTRUE;
3259 }
3260 else {
3261 Log("SHUTTLE","Unexpected HLT mode! Returning 0....");
3262 return kFALSE;
3263 }
4859271b 3264}
ca8ea066 3265
3266//______________________________________________________________________________________________
3267const char* AliShuttle::GetTriggerConfiguration()
3268{
3269 // Receives the trigger configuration from the DAQ logbook for the current run
3270
3271 // check connection, if needed reconnect
3272 if (!Connect(3))
3273 return 0;
3274
3275 TString sqlQuery;
3276 sqlQuery.Form("SELECT configFile FROM logbook_trigger_config WHERE run = %d", GetCurrentRun());
3277 TSQLResult* result = fServer[3]->Query(sqlQuery);
3278 if (!result)
3279 {
3280 Log("SHUTTLE", Form("ERROR: Can't execute query <%s>!", sqlQuery.Data()));
3281 return 0;
3282 }
3283
3284 if (result->GetRowCount() == 0)
3285 {
3286 Log("SHUTTLE", "ERROR: Trigger configuration not found in logbook_trigger_config");
3287 delete result;
3288 return 0;
3289 }
3290
3291 TSQLRow* row = result->Next();
3292 if (!row)
3293 {
3294 Log("SHUTTLE", "ERROR: Could not receive logbook_trigger_config data");
3295 delete result;
3296 return 0;
3297 }
3298
3299 // static, so that pointer remains valid when it is returned to the calling class
3300 static TString triggerConfig(row->GetField(0));
3301
3302 delete row;
3303 row = 0;
3304
3305 delete result;
3306 result = 0;
3307
3308 Log("SHUTTLE", Form("Found trigger configuration: %s", triggerConfig.Data()));
3309
3310 return triggerConfig;
3311}
3312
d386d623 3313//______________________________________________________________________________________________
3314void AliShuttle::SetShuttleTempDir(const char* tmpDir)
3315{
9827400b 3316 //
3317 // sets Shuttle temp directory
3318 //
d386d623 3319
3320 fgkShuttleTempDir = gSystem->ExpandPathName(tmpDir);
3321}
3322
3323//______________________________________________________________________________________________
3324void AliShuttle::SetShuttleLogDir(const char* logDir)
3325{
9827400b 3326 //
3327 // sets Shuttle log directory
3328 //
d386d623 3329
3330 fgkShuttleLogDir = gSystem->ExpandPathName(logDir);
3331}
503ff24f 3332//______________________________________________________________________________________________
3333Bool_t AliShuttle::TouchFile()
3334{
3335 //
3336 // touching a file on the grid if run has been DONE
3337 //
3338
3339 if (!gGrid)
3340 {
3341 Log("SHUTTLE",Form("No TGrid connection estabilished!"));
3342 Log("SHUTTLE",Form("Could not touch file for run %i",GetCurrentRun()));
3343 return kFALSE;
3344 }
3345
b8cf5e0f 3346 TString dir;
3347 dir.Form("%s%d/%s/SHUTTLE_DONE", fConfig->GetAlienPath(), GetCurrentYear(), GetLHCPeriod());
3348 // checking whether directory for touch command exists
3349 TString commandLs;
3350 commandLs.Form("ls %s",dir.Data());
3351 TGridResult *resultLs = dynamic_cast<TGridResult*>(gGrid->Command(commandLs));
3352 if (!resultLs){
3353 Log("SHUTTLE",Form("No result for %s command, returning without touching",commandLs.Data()));
3354 return kFALSE;
3355 }
3356 TMap *mapLs = dynamic_cast<TMap*>(resultLs->At(0));
3357 if (!mapLs){
3358 Log("SHUTTLE",Form("No map for %s command, returning without touching",commandLs.Data()));
97d38a5a 3359 delete resultLs;
3360 resultLs = 0x0;
b8cf5e0f 3361 return kFALSE;
3362 }
3363 TObjString *valueLsPath = dynamic_cast<TObjString*>(mapLs->GetValue("path"));
3364 if (!valueLsPath || (TString)(valueLsPath->GetString()).CompareTo(dir)!=1){
3365 Log("SHUTTLE",Form("No directory %s found, creating it",dir.Data()));
3366
3367 // creating the directory
3368
3369 Bool_t boolMkdir = gGrid->Mkdir(dir.Data());
3370 if (!boolMkdir) {
3371 Log("SHUTTLE",Form("Impossible to create dir %s in alien catalogue for run %i!",dir.Data(),GetCurrentRun()));
97d38a5a 3372 delete resultLs;
3373 resultLs = 0x0;
b8cf5e0f 3374 return kFALSE;
3375 }
3376 Log("SHUTTLE",Form("Directory %s successfully created in alien catalogue for run %i",dir.Data(),GetCurrentRun()));
3377 }
3378 else {
3379 Log("SHUTTLE",Form("Directory %s correctly found for run %i",dir.Data(),GetCurrentRun()));
3380 }
3381
97d38a5a 3382 delete resultLs;
3383 resultLs = 0x0;
3384
503ff24f 3385 TString command;
b8cf5e0f 3386 command.Form("touch %s/%i", dir.Data(), GetCurrentRun());
e660ff60 3387 Log("SHUTTLE", Form("Creating entry in file catalog: %s", command.Data()));
503ff24f 3388 TGridResult *resultTouch = dynamic_cast<TGridResult*>(gGrid->Command(command));
b8cf5e0f 3389 if (!resultTouch){
3390 Log("SHUTTLE",Form("No result for touching command, returning without touching for run %i",GetCurrentRun()));
3391 return kFALSE;
503ff24f 3392 }
b8cf5e0f 3393 TMap *mapTouch = dynamic_cast<TMap*>(resultTouch->At(0));
3394 if (!mapTouch){
3395 Log("SHUTTLE",Form("No map for touching command, returning without touching for run %i",GetCurrentRun()));
97d38a5a 3396 delete resultTouch;
3397 resultTouch = 0x0;
b8cf5e0f 3398 return kFALSE;
3399 }
3400 TObjString *valueTouch = dynamic_cast<TObjString*>(mapTouch->GetValue("__result__"));
3401 if (!valueTouch){
3402 Log("SHUTTLE",Form("No value for \"__result__\" key set in the map for touching command, returning without touching for run %i",GetCurrentRun()));
97d38a5a 3403 delete resultTouch;
3404 resultTouch = 0x0;
b8cf5e0f 3405 return kFALSE;
503ff24f 3406 }
b8cf5e0f 3407 if (valueTouch->GetString()!="1"){
3408 Log("SHUTTLE",Form("Failing the touching command, returning without touching for run %i",GetCurrentRun()));
97d38a5a 3409 delete resultTouch;
3410 resultTouch = 0x0;
b8cf5e0f 3411 return kFALSE;
3412 }
97d38a5a 3413 delete resultTouch;
3414 resultTouch = 0x0;
b8cf5e0f 3415 return kTRUE;
503ff24f 3416}
3417
3418