]> git.uio.no Git - u/mrichter/AliRoot.git/blame - SHUTTLE/AliShuttle.cxx
Shadowed data member corrected.
[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);
378
379 // If local object is valid up to infinity we store it only if it is
380 // the first unprocessed run!
381 if (aLocId.GetLastRun() == AliCDBRunRange::Infinity() &&
382 !fFirstUnprocessed[GetDetPos(fCurrentDetector)])
383 {
384 Log("SHUTTLE", Form("StoreOCDB - %s: object %s has validity infinite but "
385 "there are previous unprocessed runs!",
386 fCurrentDetector.Data(), aLocId.GetPath().Data()));
5e993b6f 387 Log(fCurrentDetector.Data(), Form("StoreOCDB - %s: object %s has validity infinite but "
388 "there are previous unprocessed runs!",
389 fCurrentDetector.Data(), aLocId.GetPath().Data()));
390 delayed = kTRUE;
3301427a 391 continue;
392 }
393
394 // loop on Grid valid Id's
395 Bool_t store = kTRUE;
396 TIter gridIter(gridIds);
397 AliCDBId* aGridId = 0;
398 while((aGridId = dynamic_cast<AliCDBId*> (gridIter.Next()))){
399 if(aGridId->GetPath() != aLocId.GetPath()) continue;
400 // skip all objects valid up to infinity
401 if(aGridId->GetLastRun() == AliCDBRunRange::Infinity()) continue;
402 // if we get here, it means there's already some more recent object stored on Grid!
403 store = kFALSE;
404 break;
405 }
406
407 // If we get here, the file can be stored!
408 Bool_t storeOk = gridSto->Put(aLocEntry);
409 if(!store || storeOk){
410
411 if (!store)
412 {
413 Log(fCurrentDetector.Data(),
414 Form("StoreOCDB - A more recent object already exists in %s storage: <%s>",
415 type, aGridId->ToString().Data()));
416 } else {
417 Log("SHUTTLE",
418 Form("StoreOCDB - Object <%s> successfully put into %s storage",
419 aLocId.ToString().Data(), type));
2d9019b4 420 Log(fCurrentDetector.Data(),
421 Form("StoreOCDB - Object <%s> successfully put into %s storage",
422 aLocId.ToString().Data(), type));
3301427a 423 }
84090f85 424
3301427a 425 // removing local filename...
426 TString filename;
427 localSto->IdToFilename(aLocId, filename);
c88ad5db 428 Log("SHUTTLE", Form("StoreOCDB - Removing local file %s", filename.Data()));
3301427a 429 RemoveFile(filename.Data());
430 continue;
431 } else {
432 Log("SHUTTLE",
433 Form("StoreOCDB - Grid %s storage of object <%s> failed",
434 type, aLocId.ToString().Data()));
2d9019b4 435 Log(fCurrentDetector.Data(),
436 Form("StoreOCDB - Grid %s storage of object <%s> failed",
437 type, aLocId.ToString().Data()));
3301427a 438 result = kFALSE;
b948db8d 439 }
440 }
3301427a 441 localEntries->Clear();
2bb7b766 442
5e993b6f 443 Int_t returnCode = 0;
444
445 if (result == kFALSE)
446 returnCode = -1;
447 else if (delayed != kFALSE)
448 returnCode = 1;
449
450 Log("SHUTTLE", Form("StoreOCDB - Returning with %d (result = %d, delayed = %d)", returnCode, result, delayed));
451 Log(fCurrentDetector.Data(), Form("StoreOCDB - Returning with %d (result = %d, delayed = %d)", returnCode, result, delayed));
452
453 return returnCode;
3301427a 454}
455
546242fb 456//______________________________________________________________________________________________
457Bool_t AliShuttle::CleanReferenceStorage(const char* detector)
458{
2d9019b4 459 // clears the directory used to store reference files of a given subdetector
546242fb 460
461 AliCDBManager* man = AliCDBManager::Instance();
462 AliCDBStorage* sto = man->GetStorage(fgkLocalRefStorage);
2d9019b4 463 TString localBaseFolder = sto->GetBaseFolder();
464
465 TString targetDir = GetRefFilePrefix(localBaseFolder.Data(), detector);
466
d524ade6 467 Log("SHUTTLE", Form("CleanReferenceStorage - Cleaning %s", targetDir.Data()));
2d9019b4 468
469 TString begin;
470 begin.Form("%d_", GetCurrentRun());
471
472 TSystemDirectory* baseDir = new TSystemDirectory("/", targetDir);
473 if (!baseDir)
474 return kTRUE;
475
476 TList* dirList = baseDir->GetListOfFiles();
477 delete baseDir;
478
479 if (!dirList) return kTRUE;
480
481 if (dirList->GetEntries() < 3)
482 {
483 delete dirList;
484 return kTRUE;
485 }
486
487 Int_t nDirs = 0, nDel = 0;
488 TIter dirIter(dirList);
489 TSystemFile* entry = 0;
546242fb 490
2d9019b4 491 Bool_t success = kTRUE;
546242fb 492
2d9019b4 493 while ((entry = dynamic_cast<TSystemFile*> (dirIter.Next())))
494 {
495 if (entry->IsDirectory())
496 continue;
497
498 TString fileName(entry->GetName());
499 if (!fileName.BeginsWith(begin))
500 continue;
501
502 nDirs++;
503
504 // delete file
505 Int_t result = gSystem->Unlink(fileName.Data());
506
507 if (result)
508 {
d524ade6 509 Log("SHUTTLE", Form("CleanReferenceStorage - Could not delete file %s!", fileName.Data()));
2d9019b4 510 success = kFALSE;
511 } else {
512 nDel++;
513 }
514 }
515
516 if(nDirs > 0)
517 Log("SHUTTLE", Form("CleanReferenceStorage - %d (over %d) reference files in folder %s were deleted.",
518 nDel, nDirs, targetDir.Data()));
519
520
521 delete dirList;
522 return success;
523
524
525
526
527
546242fb 528
529 Int_t result = gSystem->GetPathInfo(targetDir, 0, (Long64_t*) 0, 0, 0);
530 if (result == 0)
531 {
532 // delete directory
d524ade6 533 result = gSystem->Exec(Form("rm -rf %s", targetDir.Data()));
546242fb 534 if (result != 0)
535 {
d524ade6 536 Log("SHUTTLE", Form("CleanReferenceStorage - Could not clean directory %s", targetDir.Data()));
546242fb 537 return kFALSE;
538 }
539 }
540
541 result = gSystem->mkdir(targetDir, kTRUE);
542 if (result != 0)
543 {
c88ad5db 544 Log("SHUTTLE", Form("CleanReferenceStorage - Error creating base directory %s", targetDir.Data()));
546242fb 545 return kFALSE;
546 }
547
548 return kTRUE;
549}
550
9827400b 551//______________________________________________________________________________________________
552Bool_t AliShuttle::StoreReferenceFile(const char* detector, const char* localFile, const char* gridFileName)
553{
554 //
3c2a21c8 555 // Stores reference file directly (without opening it). This function stores the file locally.
9827400b 556 //
3c2a21c8 557 // The file is stored under the following location:
558 // <base folder of local reference storage>/<DET>/<RUN#>_<gridFileName>
559 // where <gridFileName> is the second parameter given to the function
560 //
9827400b 561
562 if (fTestMode & kErrorStorage)
563 {
564 Log(fCurrentDetector, "StoreReferenceFile - In TESTMODE - Simulating error while storing locally");
565 return kFALSE;
566 }
567
568 AliCDBManager* man = AliCDBManager::Instance();
569 AliCDBStorage* sto = man->GetStorage(fgkLocalRefStorage);
570
571 TString localBaseFolder = sto->GetBaseFolder();
572
d524ade6 573 TString target = GetRefFilePrefix(localBaseFolder.Data(), detector);
574 target.Append(Form("/%d_%s", GetCurrentRun(), gridFileName));
9827400b 575
d524ade6 576 return CopyFileLocally(localFile, target);
c88ad5db 577}
578
579//______________________________________________________________________________________________
580Bool_t AliShuttle::StoreRunMetadataFile(const char* localFile, const char* gridFileName)
581{
582 //
583 // Stores Run metadata file to the Grid, in the run folder
584 //
585 // Only GRP can call this function.
586
587 if (fTestMode & kErrorStorage)
588 {
589 Log(fCurrentDetector, "StoreRunMetaDataFile - In TESTMODE - Simulating error while storing locally");
590 return kFALSE;
591 }
592
593 AliCDBManager* man = AliCDBManager::Instance();
594 AliCDBStorage* sto = man->GetStorage(fgkLocalRefStorage);
595
596 TString localBaseFolder = sto->GetBaseFolder();
597
598 // Build Run level folder
955d2abc 599 // folder = /alice/data/year/lhcPeriod/runNb/raw
c88ad5db 600
c88ad5db 601
675f64cd 602 TString lhcPeriod = GetLHCPeriod();
c88ad5db 603 if (lhcPeriod.Length() == 0)
604 {
605 Log("SHUTTLE","StoreRunMetaDataFile - LHCPeriod not found in logbook!");
606 return 0;
607 }
a1c5723b 608
609 // TODO partitions with one detector only write data into LHCperiod_DET
2694f253 610 TString partition = GetRunParameter("detector");
a1c5723b 611
612 if (partition.Length() > 0 && partition != "ALICE")
613 {
614 lhcPeriod.Append(Form("_%s", partition.Data()));
615 Log(fCurrentDetector, Form("Run data tags merged file will be written in %s",
616 lhcPeriod.Data()));
617 }
8884d71e 618
b8cf5e0f 619 TString target = Form("%s/GRP/RunMetadata%s%d/%s/%09d/raw/%s",
620 localBaseFolder.Data(), fConfig->GetAlienPath(), GetCurrentYear(),
d524ade6 621 lhcPeriod.Data(), GetCurrentRun(), gridFileName);
c88ad5db 622
d524ade6 623 return CopyFileLocally(localFile, target);
c88ad5db 624}
625
626//______________________________________________________________________________________________
d524ade6 627Bool_t AliShuttle::CopyFileLocally(const char* localFile, const TString& target)
c88ad5db 628{
629 //
630 // Stores file locally. Called by StoreReferenceFile and StoreRunMetadataFile
d524ade6 631 // Files are temporarily stored in the local reference storage. When the preprocessor
632 // finishes, the Shuttle calls CopyFilesToGrid to transfer the files to AliEn
633 // (in reference or run level folders)
c88ad5db 634 //
635
d524ade6 636 TString targetDir(target(0, target.Last('/')));
637
638 //try to open base dir folder, if it does not exist
2d9019b4 639 void* dir = gSystem->OpenDirectory(targetDir.Data());
640 if (dir == NULL) {
641 if (gSystem->mkdir(targetDir.Data(), kTRUE)) {
a1c5723b 642 Log("SHUTTLE", Form("CopyFileLocally - Can't open directory <%s>", targetDir.Data()));
2d9019b4 643 return kFALSE;
644 }
645
646 } else {
647 gSystem->FreeDirectory(dir);
648 }
9827400b 649
7d43a416 650 Int_t result = 0;
651
652 result = gSystem->GetPathInfo(localFile, 0, (Long64_t*) 0, 0, 0);
9827400b 653 if (result)
654 {
a1c5723b 655 Log("SHUTTLE", Form("CopyFileLocally - %s does not exist", localFile));
546242fb 656 return kFALSE;
9827400b 657 }
546242fb 658
7d43a416 659 result = gSystem->GetPathInfo(target, 0, (Long64_t*) 0, 0, 0);
660 if (!result)
661 {
a1c5723b 662 Log("SHUTTLE", Form("CopyFileLocally - target file %s already exist, removing...", target.Data()));
7d43a416 663 if (gSystem->Unlink(target.Data()))
664 {
a1c5723b 665 Log("SHUTTLE", Form("CopyFileLocally - Could not remove existing target file %s!", target.Data()));
7d43a416 666 return kFALSE;
667 }
668 }
669
9827400b 670 result = gSystem->CopyFile(localFile, target);
671
672 if (result == 0)
673 {
a1c5723b 674 Log("SHUTTLE", Form("CopyFileLocally - File %s stored locally to %s", localFile, target.Data()));
9827400b 675 return kTRUE;
676 }
677 else
678 {
a1c5723b 679 Log("SHUTTLE", Form("CopyFileLocally - Could not store file %s to %s! Error code = %d",
546242fb 680 localFile, target.Data(), result));
9827400b 681 return kFALSE;
682 }
c88ad5db 683
684
685
9827400b 686}
687
688//______________________________________________________________________________________________
c88ad5db 689Bool_t AliShuttle::CopyFilesToGrid(const char* type)
9827400b 690{
691 //
c88ad5db 692 // Transfers local files to the Grid. Local files can be reference files
693 // or run metadata file (from GRP only).
9827400b 694 //
c88ad5db 695 // According to the type (ref, metadata) the files are stored under the following location:
696 // ref --> <base folder of reference storage>/<DET>/<RUN#>_<gridFileName>
697 // metadata --> <run data folder>/<MetadataFileName>
86aa42c3 698 //
c88ad5db 699
9827400b 700 AliCDBManager* man = AliCDBManager::Instance();
701 AliCDBStorage* sto = man->GetStorage(fgkLocalRefStorage);
702 if (!sto)
703 return kFALSE;
704 TString localBaseFolder = sto->GetBaseFolder();
9827400b 705
c88ad5db 706 TString dir;
707 TString alienDir;
9827400b 708 TString begin;
9827400b 709
c88ad5db 710 if (strcmp(type, "reference") == 0)
711 {
712 dir = GetRefFilePrefix(localBaseFolder.Data(), fCurrentDetector.Data());
713 AliCDBStorage* gridSto = man->GetStorage(fgkMainRefStorage);
714 if (!gridSto)
715 return kFALSE;
716 TString gridBaseFolder = gridSto->GetBaseFolder();
717 alienDir = GetRefFilePrefix(gridBaseFolder.Data(), fCurrentDetector.Data());
718 begin = Form("%d_", GetCurrentRun());
719 }
720 else if (strcmp(type, "metadata") == 0)
721 {
c88ad5db 722
675f64cd 723 TString lhcPeriod = GetLHCPeriod();
c88ad5db 724
725 if (lhcPeriod.Length() == 0)
726 {
727 Log("SHUTTLE","CopyFilesToGrid - LHCPeriod not found in logbook!");
728 return 0;
729 }
730
a1c5723b 731 // TODO partitions with one detector only write data into LHCperiod_DET
2694f253 732 TString partition = GetRunParameter("detector");
a1c5723b 733
734 if (partition.Length() > 0 && partition != "ALICE")
735 {
736 lhcPeriod.Append(Form("_%s", partition.Data()));
737 }
738
b8cf5e0f 739 dir = Form("%s/GRP/RunMetadata%s%d/%s/%09d/raw",
740 localBaseFolder.Data(), fConfig->GetAlienPath(), GetCurrentYear(),
c88ad5db 741 lhcPeriod.Data(), GetCurrentRun());
b8cf5e0f 742 alienDir = dir(dir.Index(fConfig->GetAlienPath()), dir.Length());
675f64cd 743
c88ad5db 744 begin = "";
745 }
746 else
747 {
748 Log("SHUTTLE", "CopyFilesToGrid - Unexpected: type label must be reference or metadata!");
749 return kFALSE;
750 }
751
9827400b 752 TSystemDirectory* baseDir = new TSystemDirectory("/", dir);
3d8bc902 753 if (!baseDir)
754 return kTRUE;
755
2d9019b4 756 TList* dirList = baseDir->GetListOfFiles();
757 delete baseDir;
758
759 if (!dirList) return kTRUE;
760
761 if (dirList->GetEntries() < 3)
3d8bc902 762 {
2d9019b4 763 delete dirList;
9827400b 764 return kTRUE;
3d8bc902 765 }
2d9019b4 766
546242fb 767 if (!gGrid)
768 {
c88ad5db 769 Log("SHUTTLE", "CopyFilesToGrid - Connection to Grid failed: Cannot continue!");
2d9019b4 770 delete dirList;
546242fb 771 return kFALSE;
772 }
773
2d9019b4 774 Int_t nDirs = 0, nTransfer = 0;
775 TIter dirIter(dirList);
776 TSystemFile* entry = 0;
777
9827400b 778 Bool_t success = kTRUE;
3d8bc902 779 Bool_t first = kTRUE;
9827400b 780
2d9019b4 781 while ((entry = dynamic_cast<TSystemFile*> (dirIter.Next())))
782 {
9827400b 783 if (entry->IsDirectory())
784 continue;
785
786 TString fileName(entry->GetName());
787 if (!fileName.BeginsWith(begin))
788 continue;
789
2d9019b4 790 nDirs++;
791
3d8bc902 792 if (first)
793 {
794 first = kFALSE;
c88ad5db 795 // check that folder exists, otherwise create it
3d8bc902 796 TGridResult* result = gGrid->Ls(alienDir.Data(), "a");
797
798 if (!result)
2d9019b4 799 {
800 delete dirList;
3d8bc902 801 return kFALSE;
2d9019b4 802 }
3d8bc902 803
546242fb 804 if (!result->GetFileName(1)) // TODO: It looks like element 0 is always 0!!
3d8bc902 805 {
675f64cd 806 // TODO It does not work currently! Bug in TAliEn::Mkdir
807 // TODO Manually fixed in local root v5-16-00
c88ad5db 808 if (!gGrid->Mkdir(alienDir.Data(),"-p",0))
3d8bc902 809 {
c88ad5db 810 Log("SHUTTLE", Form("CopyFilesToGrid - Cannot create directory %s",
3d8bc902 811 alienDir.Data()));
2d9019b4 812 delete dirList;
3d8bc902 813 return kFALSE;
546242fb 814 } else {
c88ad5db 815 Log("SHUTTLE",Form("CopyFilesToGrid - Folder %s created", alienDir.Data()));
3d8bc902 816 }
817
546242fb 818 } else {
c88ad5db 819 Log("SHUTTLE",Form("CopyFilesToGrid - Folder %s found", alienDir.Data()));
3d8bc902 820 }
821 }
822
9827400b 823 TString fullLocalPath;
824 fullLocalPath.Form("%s/%s", dir.Data(), fileName.Data());
825
826 TString fullGridPath;
827 fullGridPath.Form("alien://%s/%s", alienDir.Data(), fileName.Data());
828
a986b218 829 Bool_t result = TFile::Cp(fullLocalPath, fullGridPath);
9827400b 830
831 if (result)
832 {
c88ad5db 833 Log("SHUTTLE", Form("CopyFilesToGrid - Copying local file %s to %s succeeded!",
834 fullLocalPath.Data(), fullGridPath.Data()));
9827400b 835 RemoveFile(fullLocalPath);
2d9019b4 836 nTransfer++;
9827400b 837 }
838 else
839 {
c88ad5db 840 Log("SHUTTLE", Form("CopyFilesToGrid - Copying local file %s to %s FAILED!",
841 fullLocalPath.Data(), fullGridPath.Data()));
9827400b 842 success = kFALSE;
843 }
844 }
2d9019b4 845
c88ad5db 846 Log("SHUTTLE", Form("CopyFilesToGrid - %d (over %d) files in folder %s copied to Grid.",
847 nTransfer, nDirs, dir.Data()));
2d9019b4 848
849
850 delete dirList;
9827400b 851 return success;
852}
853
2d9019b4 854//______________________________________________________________________________________________
855const char* AliShuttle::GetRefFilePrefix(const char* base, const char* detector)
856{
857 //
858 // Get folder name of reference files
859 //
860
861 TString offDetStr(GetOfflineDetName(detector));
862 TString dir;
863 if (offDetStr == "ITS" || offDetStr == "MUON" || offDetStr == "PHOS")
864 {
865 dir.Form("%s/%s/%s", base, offDetStr.Data(), detector);
866 } else {
867 dir.Form("%s/%s", base, offDetStr.Data());
868 }
869
870 return dir.Data();
871
872
873}
c88ad5db 874
3301427a 875//______________________________________________________________________________________________
876void AliShuttle::CleanLocalStorage(const TString& uri)
877{
9827400b 878 //
879 // Called in case the preprocessor is declared failed. Remove remaining objects from the local storages.
880 //
3301427a 881
882 const char* type = 0;
883 if(uri == fgkLocalCDB) {
884 type = "OCDB";
885 } else if(uri == fgkLocalRefStorage) {
546242fb 886 type = "Reference";
3301427a 887 } else {
888 AliError(Form("Invalid storage URI: %s", uri.Data()));
889 return;
890 }
891
892 AliCDBManager* man = AliCDBManager::Instance();
b948db8d 893
3301427a 894 // open local storage
895 AliCDBStorage *localSto = man->GetStorage(uri);
896 if(!localSto) {
897 Log("SHUTTLE",
898 Form("CleanLocalStorage - cannot activate local %s storage", type));
899 return;
900 }
901
902 TString filename(Form("%s/%s/*/Run*_v%d_s*.root",
546242fb 903 localSto->GetBaseFolder().Data(), GetOfflineDetName(fCurrentDetector.Data()), GetCurrentRun()));
3301427a 904
c88ad5db 905 AliDebug(2, Form("filename = %s", filename.Data()));
3301427a 906
c88ad5db 907 Log("SHUTTLE", Form("Removing remaining local files for run %d and detector %s ...",
3301427a 908 GetCurrentRun(), fCurrentDetector.Data()));
909
910 RemoveFile(filename.Data());
911
912}
913
914//______________________________________________________________________________________________
915void AliShuttle::RemoveFile(const char* filename)
916{
9827400b 917 //
918 // removes local file
919 //
3301427a 920
921 TString command(Form("rm -f %s", filename));
922
923 Int_t result = gSystem->Exec(command.Data());
924 if(result != 0)
925 {
926 Log("SHUTTLE", Form("RemoveFile - %s: Cannot remove file %s!",
927 fCurrentDetector.Data(), filename));
928 }
73abe331 929}
930
b948db8d 931//______________________________________________________________________________________________
5164a766 932AliShuttleStatus* AliShuttle::ReadShuttleStatus()
933{
9827400b 934 //
935 // Reads the AliShuttleStatus from the CDB
936 //
5164a766 937
2bb7b766 938 if (fStatusEntry){
939 delete fStatusEntry;
940 fStatusEntry = 0;
941 }
5164a766 942
10a5a932 943 fStatusEntry = AliCDBManager::Instance()->GetStorage(GetLocalCDB())
2bb7b766 944 ->Get(Form("/SHUTTLE/STATUS/%s", fCurrentDetector.Data()), GetCurrentRun());
5164a766 945
2bb7b766 946 if (!fStatusEntry) return 0;
947 fStatusEntry->SetOwner(1);
5164a766 948
2bb7b766 949 AliShuttleStatus* status = dynamic_cast<AliShuttleStatus*> (fStatusEntry->GetObject());
950 if (!status) {
951 AliError("Invalid object stored to CDB!");
952 return 0;
953 }
5164a766 954
2bb7b766 955 return status;
5164a766 956}
957
958//______________________________________________________________________________________________
7bfb2090 959Bool_t AliShuttle::WriteShuttleStatus(AliShuttleStatus* status)
5164a766 960{
9827400b 961 //
962 // writes the status for one subdetector
963 //
2bb7b766 964
965 if (fStatusEntry){
966 delete fStatusEntry;
967 fStatusEntry = 0;
968 }
5164a766 969
2bb7b766 970 Int_t run = GetCurrentRun();
5164a766 971
2bb7b766 972 AliCDBId id(AliCDBPath("SHUTTLE", "STATUS", fCurrentDetector), run, run);
5164a766 973
2bb7b766 974 fStatusEntry = new AliCDBEntry(status, id, new AliCDBMetaData);
975 fStatusEntry->SetOwner(1);
5164a766 976
2bb7b766 977 UInt_t result = AliCDBManager::Instance()->GetStorage(fgkLocalCDB)->Put(fStatusEntry);
7bfb2090 978
2bb7b766 979 if (!result) {
3301427a 980 Log("SHUTTLE", Form("WriteShuttleStatus - Failed for %s, run %d",
981 fCurrentDetector.Data(), run));
2bb7b766 982 return kFALSE;
983 }
e7f62f16 984
985 SendMLInfo();
7bfb2090 986
2bb7b766 987 return kTRUE;
5164a766 988}
989
990//______________________________________________________________________________________________
991void AliShuttle::UpdateShuttleStatus(AliShuttleStatus::Status newStatus, Bool_t increaseCount)
992{
9827400b 993 //
994 // changes the AliShuttleStatus for the given detector and run to the given status
995 //
5164a766 996
2bb7b766 997 if (!fStatusEntry){
998 AliError("UNEXPECTED: fStatusEntry empty");
999 return;
1000 }
5164a766 1001
2bb7b766 1002 AliShuttleStatus* status = dynamic_cast<AliShuttleStatus*> (fStatusEntry->GetObject());
5164a766 1003
2bb7b766 1004 if (!status){
c88ad5db 1005 Log("SHUTTLE", "UpdateShuttleStatus - UNEXPECTED: status could not be read from current CDB entry");
2bb7b766 1006 return;
1007 }
5164a766 1008
2c15234c 1009 TString actionStr = Form("UpdateShuttleStatus - %s: Changing state from %s to %s",
eba76848 1010 fCurrentDetector.Data(),
36c99a6a 1011 status->GetStatusName(),
eba76848 1012 status->GetStatusName(newStatus));
cb343cfd 1013 Log("SHUTTLE", actionStr);
1014 SetLastAction(actionStr);
5164a766 1015
2bb7b766 1016 status->SetStatus(newStatus);
1017 if (increaseCount) status->IncreaseCount();
5164a766 1018
2bb7b766 1019 AliCDBManager::Instance()->GetStorage(fgkLocalCDB)->Put(fStatusEntry);
e7f62f16 1020
1021 SendMLInfo();
5164a766 1022}
e7f62f16 1023
1024//______________________________________________________________________________________________
1025void AliShuttle::SendMLInfo()
1026{
1027 //
1028 // sends ML information about the current status of the current detector being processed
1029 //
1030
1031 AliShuttleStatus* status = dynamic_cast<AliShuttleStatus*> (fStatusEntry->GetObject());
1032
1033 if (!status){
3301427a 1034 Log("SHUTTLE", "SendMLInfo - UNEXPECTED: status could not be read from current CDB entry");
e7f62f16 1035 return;
1036 }
1037
1038 TMonaLisaText mlStatus(Form("%s_status", fCurrentDetector.Data()), status->GetStatusName());
1039 TMonaLisaValue mlRetryCount(Form("%s_count", fCurrentDetector.Data()), status->GetCount());
1040
1041 TList mlList;
1042 mlList.Add(&mlStatus);
1043 mlList.Add(&mlRetryCount);
1044
ee6f7523 1045 TString mlID;
1046 mlID.Form("%d", GetCurrentRun());
1047 fMonaLisa->SendParameters(&mlList, mlID);
e7f62f16 1048}
1049
5164a766 1050//______________________________________________________________________________________________
1051Bool_t AliShuttle::ContinueProcessing()
1052{
9827400b 1053 // this function reads the AliShuttleStatus information from CDB and
1054 // checks if the processing should be continued
1055 // if yes it returns kTRUE and updates the AliShuttleStatus with nextStatus
2bb7b766 1056
926cbc0e 1057 if (!fConfig->HostProcessDetector(fCurrentDetector))
1058 return kFALSE;
57c1a579 1059
1060 AliPreprocessor* aPreprocessor =
1061 dynamic_cast<AliPreprocessor*> (fPreprocessorMap.GetValue(fCurrentDetector));
1062 if (!aPreprocessor)
1063 {
c88ad5db 1064 Log("SHUTTLE", Form("ContinueProcessing - %s: no preprocessor registered", fCurrentDetector.Data()));
57c1a579 1065 return kFALSE;
1066 }
1067
2bb7b766 1068 AliShuttleLogbookEntry::Status entryStatus =
eba76848 1069 fLogbookEntry->GetDetectorStatus(fCurrentDetector);
2bb7b766 1070
5e993b6f 1071 if (entryStatus != AliShuttleLogbookEntry::kUnprocessed) {
c88ad5db 1072 Log("SHUTTLE", Form("ContinueProcessing - %s is %s",
2bb7b766 1073 fCurrentDetector.Data(),
1074 fLogbookEntry->GetDetectorStatusName(entryStatus)));
1075 return kFALSE;
1076 }
1077
1078 // if we get here, according to Shuttle logbook subdetector is in UNPROCESSED state
be48e3ea 1079
1080 // check if current run is first unprocessed run for current detector
1081 if (fConfig->StrictRunOrder(fCurrentDetector) &&
1082 !fFirstUnprocessed[GetDetPos(fCurrentDetector)])
1083 {
86aa42c3 1084 if (fTestMode == kNone)
1085 {
c88ad5db 1086 Log("SHUTTLE", Form("ContinueProcessing - %s requires strict run ordering"
1087 " but this is not the first unprocessed run!"));
86aa42c3 1088 return kFALSE;
1089 }
1090 else
1091 {
c88ad5db 1092 Log("SHUTTLE", Form("ContinueProcessing - In TESTMODE - "
1093 "Although %s requires strict run ordering "
1094 "and this is not the first unprocessed run, "
1095 "the SHUTTLE continues"));
86aa42c3 1096 }
be48e3ea 1097 }
1098
926cbc0e 1099 // Is the subdetector processed first time for this run?
1100 fFirstProcessing = kFALSE;
1101
2bb7b766 1102 AliShuttleStatus* status = ReadShuttleStatus();
1103 if (!status) {
1104 // first time
1105 Log("SHUTTLE", Form("ContinueProcessing - %s: Processing first time",
1106 fCurrentDetector.Data()));
1107 status = new AliShuttleStatus(AliShuttleStatus::kStarted);
926cbc0e 1108 fFirstProcessing = kTRUE;
2bb7b766 1109 return WriteShuttleStatus(status);
1110 }
1111
ffa25f20 1112 // The following case shouldn't happen if Shuttle Logbook was correctly updated.
2bb7b766 1113 // If it happens it may mean Logbook updating failed... let's do it now!
1114 if (status->GetStatus() == AliShuttleStatus::kDone ||
ffa25f20 1115 status->GetStatus() == AliShuttleStatus::kFailed ||
1116 status->GetStatus() == AliShuttleStatus::kSkipped) {
2bb7b766 1117 Log("SHUTTLE", Form("ContinueProcessing - %s is already %s. Updating Shuttle Logbook",
1118 fCurrentDetector.Data(),
1119 status->GetStatusName(status->GetStatus())));
ffa25f20 1120
1121 if (status->GetStatus() == AliShuttleStatus::kSkipped)
1122 {
1123 UpdateShuttleLogbook(fCurrentDetector.Data(), "DONE");
1124 }
1125 else
1126 UpdateShuttleLogbook(fCurrentDetector.Data(), status->GetStatusName(status->GetStatus()));
1127
2bb7b766 1128 return kFALSE;
1129 }
1130
5e993b6f 1131 if (status->GetStatus() == AliShuttleStatus::kStoreStarted || status->GetStatus() == AliShuttleStatus::kStoreDelayed ||status->GetStatus() == AliShuttleStatus::kStoreError) {
2bb7b766 1132 Log("SHUTTLE",
c88ad5db 1133 Form("ContinueProcessing - %s: Grid storage of one or more "
1134 "objects failed. Trying again now",
2bb7b766 1135 fCurrentDetector.Data()));
5e993b6f 1136 StoreOCDB();
2bb7b766 1137 return kFALSE;
1138 }
1139
1140 // if we get here, there is a restart
57c1a579 1141 Bool_t cont = kFALSE;
2bb7b766 1142
1143 // abort conditions
cb343cfd 1144 if (status->GetCount() >= fConfig->GetMaxRetries()) {
57c1a579 1145 Log("SHUTTLE", Form("ContinueProcessing - %s failed %d times in status %s - "
1146 "Updating Shuttle Logbook", fCurrentDetector.Data(),
2bb7b766 1147 status->GetCount(), status->GetStatusName()));
1148 UpdateShuttleLogbook(fCurrentDetector.Data(), "FAILED");
e7f62f16 1149 UpdateShuttleStatus(AliShuttleStatus::kFailed);
3301427a 1150
1151 // there may still be objects in local OCDB and reference storage
1152 // and FXS databases may be not updated: do it now!
9827400b 1153
1154 // TODO Currently disabled, we want to keep files in case of failure!
1155 // CleanLocalStorage(fgkLocalCDB);
1156 // CleanLocalStorage(fgkLocalRefStorage);
1157 // UpdateTableFailCase();
1158
1159 // Send mail to detector expert!
c88ad5db 1160 Log("SHUTTLE", Form("ContinueProcessing - Sending mail to %s expert...",
926cbc0e 1161 fCurrentDetector.Data()));
339adafa 1162 if (!SendMail(kPPEMail))
9827400b 1163 Log("SHUTTLE", Form("ContinueProcessing - Could not send mail to %s expert",
926cbc0e 1164 fCurrentDetector.Data()));
3301427a 1165
57c1a579 1166 } else {
1167 Log("SHUTTLE", Form("ContinueProcessing - %s: restarting. "
1168 "Aborted before with %s. Retry number %d.", fCurrentDetector.Data(),
1169 status->GetStatusName(), status->GetCount()));
9827400b 1170 Bool_t increaseCount = kTRUE;
c88ad5db 1171 if (status->GetStatus() == AliShuttleStatus::kDCSError ||
8da81210 1172 status->GetStatus() == AliShuttleStatus::kDCSStarted ||
1173 status->GetStatus() == AliShuttleStatus::kFXSError)
c88ad5db 1174 increaseCount = kFALSE;
675f64cd 1175
9827400b 1176 UpdateShuttleStatus(AliShuttleStatus::kStarted, increaseCount);
57c1a579 1177 cont = kTRUE;
2bb7b766 1178 }
1179
57c1a579 1180 return cont;
5164a766 1181}
1182
1183//______________________________________________________________________________________________
2bb7b766 1184Bool_t AliShuttle::Process(AliShuttleLogbookEntry* entry)
58bc3020 1185{
73abe331 1186 //
b948db8d 1187 // Makes data retrieval for all detectors in the configuration.
2bb7b766 1188 // entry: Shuttle logbook entry, contains run paramenters and status of detectors
1189 // (Unprocessed, Inactive, Failed or Done).
d477ad88 1190 // Returns kFALSE in case of error occured and kTRUE otherwise
73abe331 1191 //
1192
9827400b 1193 if (!entry) return kFALSE;
2bb7b766 1194
1195 fLogbookEntry = entry;
1196
c88ad5db 1197 Log("SHUTTLE", Form("\t\t\t^*^*^*^*^*^*^*^*^*^*^*^* run %d: START ^*^*^*^*^*^*^*^*^*^*^*^*",
9827400b 1198 GetCurrentRun()));
2bb7b766 1199
e7f62f16 1200 // Send the information to ML
339adafa 1201 CountOpenRuns();
1202
e7f62f16 1203 TMonaLisaText mlStatus("SHUTTLE_status", "Processing");
339adafa 1204 TString runType(entry->GetRunType());
1205 if (strlen(entry->GetRunParameter("log")) > 0){
1206
1207 runType += "(";
1208 runType += entry->GetRunParameter("log");
1209 runType += ")";
1210 }
1211 TMonaLisaText mlRunType("SHUTTLE_runtype", runType);
e7f62f16 1212
1213 TList mlList;
1214 mlList.Add(&mlStatus);
9827400b 1215 mlList.Add(&mlRunType);
e7f62f16 1216
ee6f7523 1217 TString mlID;
1218 mlID.Form("%d", GetCurrentRun());
1219 fMonaLisa->SendParameters(&mlList, mlID);
3301427a 1220
9827400b 1221 if (fLogbookEntry->IsDone())
1222 {
1223 Log("SHUTTLE","Process - Shuttle is already DONE. Updating logbook");
1224 UpdateShuttleLogbook("shuttle_done");
1225 fLogbookEntry = 0;
1226 return kTRUE;
1227 }
1228
1229 // read test mode if flag is set
1230 if (fReadTestMode)
1231 {
3d8bc902 1232 fTestMode = kNone;
9827400b 1233 TString logEntry(entry->GetRunParameter("log"));
1234 //printf("log entry = %s\n", logEntry.Data());
1235 TString searchStr("Testmode: ");
1236 Int_t pos = logEntry.Index(searchStr.Data());
1237 //printf("%d\n", pos);
1238 if (pos >= 0)
1239 {
1240 TSubString subStr = logEntry(pos + searchStr.Length(), logEntry.Length());
1241 //printf("%s\n", subStr.String().Data());
1242 TString newStr(subStr.Data());
1243 TObjArray* token = newStr.Tokenize(' ');
1244 if (token)
1245 {
1246 //token->Print();
1247 TObjString* tmpStr = dynamic_cast<TObjString*> (token->First());
1248 if (tmpStr)
1249 {
1250 Int_t testMode = tmpStr->String().Atoi();
1251 if (testMode > 0)
1252 {
c88ad5db 1253 Log("SHUTTLE", Form("Process - Enabling test mode %d", testMode));
9827400b 1254 SetTestMode((TestMode) testMode);
1255 }
1256 }
1257 delete token;
1258 }
1259 }
1260 }
c88ad5db 1261
eba76848 1262 fLogbookEntry->Print("all");
57f50b3c 1263
1264 // Initialization
d477ad88 1265 Bool_t hasError = kFALSE;
5164a766 1266
675f64cd 1267 // Set the CDB and Reference folders according to the year and LHC period
1268 TString lhcPeriod(GetLHCPeriod());
1269 if (lhcPeriod.Length() == 0)
1270 {
7d4cf768 1271 Log("SHUTTLE","Process - LHCPeriod not found in logbook!");
1272 return 0;
675f64cd 1273 }
1274
1265724d 1275 // build cdb paths (repeat each time, LHCperiod might have changed)
1276 fgkMainCDB.Form("alien://folder=%s%d/%s/OCDB?user=alidaq?cacheFold=/tmp/OCDBCache",
b8cf5e0f 1277 fConfig->GetAlienPath(), GetCurrentYear(), lhcPeriod.Data());
675f64cd 1278
1265724d 1279 fgkMainRefStorage.Form("alien://folder=%s%d/%s/Reference?user=alidaq?cacheFold=/tmp/OCDBCache",
b8cf5e0f 1280 fConfig->GetAlienPath(), GetCurrentYear(), lhcPeriod.Data());
675f64cd 1281
57f50b3c 1282 // Loop on detectors in the configuration
b948db8d 1283 TIter iter(fConfig->GetDetectors());
2bb7b766 1284 TObjString* aDetector = 0;
b948db8d 1285
4508ab8b 1286 Bool_t first = kTRUE;
1287
be48e3ea 1288 while ((aDetector = (TObjString*) iter.Next()))
1289 {
7bfb2090 1290 fCurrentDetector = aDetector->String();
5164a766 1291
5e993b6f 1292 if (ContinueProcessing() == kFALSE)
1293 continue;
4508ab8b 1294
1295 if (first)
1296 {
1297 // only read QueryCDB when needed and only once
1298 AliCDBStorage *mainCDBSto = AliCDBManager::Instance()->GetStorage(fgkMainCDB);
1299 if(mainCDBSto) mainCDBSto->QueryCDB(GetCurrentRun());
1300 AliCDBStorage *mainRefSto = AliCDBManager::Instance()->GetStorage(fgkMainRefStorage);
1301 if(mainRefSto) mainRefSto->QueryCDB(GetCurrentRun());
1302 first = kFALSE;
1303 }
9e080f92 1304
c88ad5db 1305 Log("SHUTTLE", Form("\t\t\t****** run %d - %s: START ******",
2bb7b766 1306 GetCurrentRun(), aDetector->GetName()));
1307
9d733021 1308 for(Int_t iSys=0;iSys<3;iSys++) fFXSCalled[iSys]=kFALSE;
1309
c88ad5db 1310 Log(fCurrentDetector.Data(), "Process - Starting processing");
85a80aa9 1311
be48e3ea 1312 Int_t pid = fork();
1313
1314 if (pid < 0)
1315 {
c88ad5db 1316 Log("SHUTTLE", "Process - ERROR: Forking failed");
be48e3ea 1317 }
1318 else if (pid > 0)
1319 {
1320 // parent
c88ad5db 1321 Log("SHUTTLE", Form("Process - In parent process of %d - %s: Starting monitoring",
be48e3ea 1322 GetCurrentRun(), aDetector->GetName()));
1323
1324 Long_t begin = time(0);
1325
1326 int status; // to be used with waitpid, on purpose an int (not Int_t)!
1327 while (waitpid(pid, &status, WNOHANG) == 0)
1328 {
1329 Long_t expiredTime = time(0) - begin;
1330
1331 if (expiredTime > fConfig->GetPPTimeOut())
1332 {
9827400b 1333 TString tmp;
c88ad5db 1334 tmp.Form("Process - Process of %s time out. "
1335 "Run time: %d seconds. Killing...",
1336 fCurrentDetector.Data(), expiredTime);
9827400b 1337 Log("SHUTTLE", tmp);
1338 Log(fCurrentDetector, tmp);
be48e3ea 1339
1340 kill(pid, 9);
1341
26b1866e 1342 UpdateShuttleStatus(AliShuttleStatus::kPPTimeOut);
be48e3ea 1343 hasError = kTRUE;
1344
1345 gSystem->Sleep(1000);
1346 }
1347 else
1348 {
be48e3ea 1349 gSystem->Sleep(1000);
9827400b 1350
1351 TString checkStr;
1352 checkStr.Form("ps -o vsize --pid %d | tail -n 1", pid);
1353 FILE* pipe = gSystem->OpenPipe(checkStr, "r");
1354 if (!pipe)
1355 {
c88ad5db 1356 Log("SHUTTLE", Form("Process - Error: "
1357 "Could not open pipe to %s", checkStr.Data()));
9827400b 1358 continue;
1359 }
1360
1361 char buffer[100];
1362 if (!fgets(buffer, 100, pipe))
1363 {
c88ad5db 1364 Log("SHUTTLE", "Process - Error: ps did not return anything");
9827400b 1365 gSystem->ClosePipe(pipe);
1366 continue;
1367 }
1368 gSystem->ClosePipe(pipe);
1369
1370 //Log("SHUTTLE", Form("ps returned %s", buffer));
1371
1372 Int_t mem = 0;
1373 if ((sscanf(buffer, "%d\n", &mem) != 1) || !mem)
1374 {
c88ad5db 1375 Log("SHUTTLE", "Process - Error: Could not parse output of ps");
9827400b 1376 continue;
1377 }
1378
1379 if (expiredTime % 60 == 0)
ee6f7523 1380 {
c88ad5db 1381 Log("SHUTTLE", Form("Process - %s: Checking process. "
1382 "Run time: %d seconds - Memory consumption: %d KB",
1383 fCurrentDetector.Data(), expiredTime, mem));
ee6f7523 1384 SendAlive();
1385 }
9827400b 1386
1387 if (mem > fConfig->GetPPMaxMem())
1388 {
1389 TString tmp;
c88ad5db 1390 tmp.Form("Process - Process exceeds maximum allowed memory "
1391 "(%d KB > %d KB). Killing...",
9827400b 1392 mem, fConfig->GetPPMaxMem());
1393 Log("SHUTTLE", tmp);
1394 Log(fCurrentDetector, tmp);
1395
1396 kill(pid, 9);
1397
1398 UpdateShuttleStatus(AliShuttleStatus::kPPOutOfMemory);
1399 hasError = kTRUE;
1400
1401 gSystem->Sleep(1000);
1402 }
be48e3ea 1403 }
1404 }
1405
c88ad5db 1406 Log("SHUTTLE", Form("Process - In parent process of %d - %s: Client has terminated.",
be48e3ea 1407 GetCurrentRun(), aDetector->GetName()));
1408
1409 if (WIFEXITED(status))
1410 {
1411 Int_t returnCode = WEXITSTATUS(status);
1412
c88ad5db 1413 Log("SHUTTLE", Form("Process - %s: the return code is %d", fCurrentDetector.Data(),
3301427a 1414 returnCode));
be48e3ea 1415
9827400b 1416 if (returnCode == 0) hasError = kTRUE;
be48e3ea 1417 }
1418 }
1419 else if (pid == 0)
1420 {
26b1866e 1421 // child
1422 Log("SHUTTLE", Form("Process - In child process of %d - %s", GetCurrentRun(),
c88ad5db 1423 aDetector->GetName()));
be48e3ea 1424
c88ad5db 1425 Log("SHUTTLE", Form("Process - Redirecting output to %s log",fCurrentDetector.Data()));
ffa29e93 1426
546242fb 1427 if ((freopen(GetLogFileName(fCurrentDetector), "a", stdout)) == 0)
ffa29e93 1428 {
c88ad5db 1429 Log("SHUTTLE", "Process - Could not freopen stdout");
ffa29e93 1430 }
1431 else
1432 {
1433 fOutputRedirected = kTRUE;
1434 if ((dup2(fileno(stdout), fileno(stderr))) < 0)
c88ad5db 1435 Log("SHUTTLE", "Process - Could not redirect stderr");
ffa29e93 1436
1437 }
1438
5bac2bde 1439 TString wd = gSystem->WorkingDirectory();
675f64cd 1440 TString tmpDir = Form("%s/%s_%d_process", GetShuttleTempDir(),
1441 fCurrentDetector.Data(), GetCurrentRun());
5bac2bde 1442
d524ade6 1443 Int_t result = gSystem->GetPathInfo(tmpDir.Data(), 0, (Long64_t*) 0, 0, 0);
1444 if (!result) // temp dir already exists!
1445 {
1446 Log(fCurrentDetector.Data(),
1447 Form("Process - %s dir already exists! Removing...", tmpDir.Data()));
1448 gSystem->Exec(Form("rm -rf %s",tmpDir.Data()));
675f64cd 1449 }
1450
1451 if (gSystem->mkdir(tmpDir.Data(), 1))
1452 {
1453 Log(fCurrentDetector.Data(), "Process - could not make temp directory!!");
1454 gSystem->Exit(1);
d524ade6 1455 }
1456
1457 if (!gSystem->ChangeDirectory(tmpDir.Data()))
1458 {
1459 Log(fCurrentDetector.Data(), "Process - could not change directory!!");
1460 gSystem->Exit(1);
1461 }
5bac2bde 1462
ffa25f20 1463 Int_t success = ProcessCurrentDetector();
26b1866e 1464
5bac2bde 1465 gSystem->ChangeDirectory(wd.Data());
b1d18693 1466
ffa25f20 1467 if (success == 1) // Preprocessor finished successfully!
9827400b 1468 {
fb40ead4 1469 // remove temporary folder or DCS map
1470 if (!fConfig->KeepTempFolder())
1471 {
1472 gSystem->Exec(Form("rm -rf %s",tmpDir.Data()));
1473 } else if (!fConfig->KeepDCSMap())
1474 {
1475 gSystem->Exec(Form("rm -f %s/DCSMap.root",tmpDir.Data()));
1476 }
b1d18693 1477
3301427a 1478 // Update time_processed field in FXS DB
1479 if (UpdateTable() == kFALSE)
5bac2bde 1480 Log("SHUTTLE", Form("Process - %s: Could not update FXS databases!",
1481 fCurrentDetector.Data()));
3301427a 1482
1483 // Transfer the data from local storage to main storage (Grid)
3301427a 1484 if (StoreOCDB() == kFALSE)
9827400b 1485 success = kFALSE;
5e993b6f 1486 }
ffa25f20 1487 else if (success == 0)
c88ad5db 1488 {
1489 Log("SHUTTLE",
1490 Form("\t\t\t****** run %d - %s: PP ERROR ******",
1491 GetCurrentRun(), aDetector->GetName()));
be48e3ea 1492 }
1493
4b95672b 1494 for (UInt_t iSys=0; iSys<3; iSys++)
1495 {
1496 if (fFXSCalled[iSys]) fFXSlist[iSys].Clear();
1497 }
1498
c88ad5db 1499 Log("SHUTTLE", Form("Process - Client process of %d - %s is exiting now with %d.",
9827400b 1500 GetCurrentRun(), aDetector->GetName(), success));
be48e3ea 1501
1502 // the client exits here
9827400b 1503 gSystem->Exit(success);
be48e3ea 1504
1505 AliError("We should never get here!!!");
1506 }
7bfb2090 1507 }
5164a766 1508
c88ad5db 1509 Log("SHUTTLE", Form("\t\t\t^*^*^*^*^*^*^*^*^*^*^*^* run %d: FINISH ^*^*^*^*^*^*^*^*^*^*^*^*",
2bb7b766 1510 GetCurrentRun()));
1511
1512 //check if shuttle is done for this run, if so update logbook
1513 TObjArray checkEntryArray;
1514 checkEntryArray.SetOwner(1);
9e080f92 1515 TString whereClause = Form("where run=%d", GetCurrentRun());
b0e53b15 1516 if (!QueryShuttleLogbook(whereClause.Data(), checkEntryArray) ||
1517 checkEntryArray.GetEntries() == 0) {
9e080f92 1518 Log("SHUTTLE", Form("Process - Warning: Cannot check status of run %d on Shuttle logbook!",
1519 GetCurrentRun()));
1520 return hasError == kFALSE;
1521 }
b948db8d 1522
9e080f92 1523 AliShuttleLogbookEntry* checkEntry = dynamic_cast<AliShuttleLogbookEntry*>
1524 (checkEntryArray.At(0));
2bb7b766 1525
9e080f92 1526 if (checkEntry)
1527 {
1528 if (checkEntry->IsDone())
be48e3ea 1529 {
9e080f92 1530 Log("SHUTTLE","Process - Shuttle is DONE. Updating logbook");
1531 UpdateShuttleLogbook("shuttle_done");
1532 }
1533 else
1534 {
1535 for (UInt_t iDet=0; iDet<NDetectors(); iDet++)
be48e3ea 1536 {
9e080f92 1537 if (checkEntry->GetDetectorStatus(iDet) == AliShuttleLogbookEntry::kUnprocessed)
be48e3ea 1538 {
9e080f92 1539 AliDebug(2, Form("Run %d: setting %s as \"not first time unprocessed\"",
1540 checkEntry->GetRun(), GetDetName(iDet)));
1541 fFirstUnprocessed[iDet] = kFALSE;
be48e3ea 1542 }
1543 }
339adafa 1544 TMonaLisaText mlStatusPending("SHUTTLE_status", "Pending");
1545 mlList.Clear();
1546 mlList.Add(&mlStatusPending);
1547 fMonaLisa->SendParameters(&mlList, mlID);
2bb7b766 1548 }
1549 }
1550
1551 fLogbookEntry = 0;
85a80aa9 1552
a7160fe9 1553 return hasError == kFALSE;
73abe331 1554}
1555
b948db8d 1556//______________________________________________________________________________________________
ffa25f20 1557Int_t AliShuttle::ProcessCurrentDetector()
73abe331 1558{
1559 //
2bb7b766 1560 // Makes data retrieval just for a specific detector (fCurrentDetector).
73abe331 1561 // Threre should be a configuration for this detector.
73abe331 1562
1d172743 1563 Log("SHUTTLE", Form("ProcessCurrentDetector - Retrieving values for %s, run %d",
1564 fCurrentDetector.Data(), GetCurrentRun()));
73abe331 1565
d524ade6 1566 TString wd = gSystem->WorkingDirectory();
1567
2d9019b4 1568 if (!CleanReferenceStorage(fCurrentDetector.Data()))
ffa25f20 1569 return 0;
d524ade6 1570
1571 gSystem->ChangeDirectory(wd.Data());
1572
3301427a 1573 // call preprocessor
1574 AliPreprocessor* aPreprocessor =
1575 dynamic_cast<AliPreprocessor*> (fPreprocessorMap.GetValue(fCurrentDetector));
1576
8579c6fb 1577 // check if the preprocessor wants to process this run type
1578 if (aPreprocessor->ProcessRunType() == kFALSE)
ffa25f20 1579 {
1580 UpdateShuttleStatus(AliShuttleStatus::kSkipped);
1581 UpdateShuttleLogbook(fCurrentDetector, "DONE");
1582 Log(fCurrentDetector, Form("ProcessCurrentDetector - %s preprocessor is not interested in this run type", fCurrentDetector.Data()));
1583
1584 return 2;
1585 }
1586
1587 TMap* dcsMap = new TMap();
1588
3301427a 1589 aPreprocessor->Initialize(GetCurrentRun(), GetCurrentStartTime(), GetCurrentEndTime());
1590
1591 Bool_t processDCS = aPreprocessor->ProcessDCS();
d477ad88 1592
651fdaab 1593 if (!processDCS)
1594 {
1d172743 1595 Log(fCurrentDetector, "ProcessCurrentDetector -"
1596 " The preprocessor requested to skip the retrieval of DCS values");
651fdaab 1597 }
8b739301 1598 else if (fTestMode & kSkipDCS)
2c15234c 1599 {
1d172743 1600 Log(fCurrentDetector, "ProcessCurrentDetector - In TESTMODE: Skipping DCS processing");
9827400b 1601 }
1602 else if (fTestMode & kErrorDCS)
1603 {
1d172743 1604 Log(fCurrentDetector, "ProcessCurrentDetector - In TESTMODE: Simulating DCS error");
3d8bc902 1605 UpdateShuttleStatus(AliShuttleStatus::kDCSStarted);
9827400b 1606 UpdateShuttleStatus(AliShuttleStatus::kDCSError);
1d172743 1607 delete dcsMap;
ffa25f20 1608 return 0;
2c15234c 1609 } else {
3301427a 1610
1611 UpdateShuttleStatus(AliShuttleStatus::kDCSStarted);
1612
1d172743 1613 // Query DCS archive
1614 Int_t nServers = fConfig->GetNServers(fCurrentDetector);
a038aa70 1615
1d172743 1616 for (int iServ=0; iServ<nServers; iServ++)
2c15234c 1617 {
1d172743 1618
1619 TString host(fConfig->GetDCSHost(fCurrentDetector, iServ));
1620 Int_t port = fConfig->GetDCSPort(fCurrentDetector, iServ);
542b6cc8 1621 Int_t multiSplit = fConfig->GetMultiSplit(fCurrentDetector, iServ);
1622
1790d4b7 1623 Log(fCurrentDetector, Form("ProcessCurrentDetector -"
1624 " Querying DCS Amanda server %s:%d (%d of %d)",
1625 host.Data(), port, iServ+1, nServers));
1d172743 1626
1627 TMap* aliasMap = 0;
1628 TMap* dpMap = 0;
26b1866e 1629
1d172743 1630 if (fConfig->GetDCSAliases(fCurrentDetector, iServ)->GetEntries() > 0)
2c15234c 1631 {
1d172743 1632 aliasMap = GetValueSet(host, port,
542b6cc8 1633 fConfig->GetDCSAliases(fCurrentDetector, iServ),
1634 kAlias, multiSplit);
1d172743 1635 if (!aliasMap)
1636 {
1637 Log(fCurrentDetector,
1638 Form("ProcessCurrentDetector -"
675f64cd 1639 " Error retrieving DCS aliases from server %s."
1640 " Sending mail to DCS experts!", host.Data()));
1d172743 1641 UpdateShuttleStatus(AliShuttleStatus::kDCSError);
675f64cd 1642
339adafa 1643 if (!SendMail(kDCSEMail))
fb40ead4 1644 Log("SHUTTLE", Form("ProcessCurrentDetector - "
95d4aaef 1645 "Could not send mail to DCS experts!"));
675f64cd 1646
a038aa70 1647 delete dcsMap;
ffa25f20 1648 return 0;
1d172743 1649 }
2c15234c 1650 }
a038aa70 1651
1d172743 1652 if (fConfig->GetDCSDataPoints(fCurrentDetector, iServ)->GetEntries() > 0)
a038aa70 1653 {
1d172743 1654 dpMap = GetValueSet(host, port,
542b6cc8 1655 fConfig->GetDCSDataPoints(fCurrentDetector, iServ),
1656 kDP, multiSplit);
1d172743 1657 if (!dpMap)
1658 {
1659 Log(fCurrentDetector,
1660 Form("ProcessCurrentDetector -"
675f64cd 1661 " Error retrieving DCS data points from server %s."
1662 " Sending mail to DCS experts!", host.Data()));
1d172743 1663 UpdateShuttleStatus(AliShuttleStatus::kDCSError);
675f64cd 1664
339adafa 1665 if (!SendMail(kDCSEMail))
fb40ead4 1666 Log("SHUTTLE", Form("ProcessCurrentDetector - "
95d4aaef 1667 "Could not send mail to DCS experts!"));
675f64cd 1668
1d172743 1669 if (aliasMap) delete aliasMap;
1670 delete dcsMap;
ffa25f20 1671 return 0;
1d172743 1672 }
a038aa70 1673 }
1d172743 1674
1675 // merge aliasMap and dpMap into dcsMap
1676 if(aliasMap) {
1677 TIter iter(aliasMap);
a038aa70 1678 TObjString* key = 0;
1679 while ((key = (TObjString*) iter.Next()))
1d172743 1680 dcsMap->Add(key, aliasMap->GetValue(key->String()));
1681
1682 aliasMap->SetOwner(kFALSE);
1683 delete aliasMap;
1684 }
1685
1686 if(dpMap) {
1687 TIter iter(dpMap);
1688 TObjString* key = 0;
1689 while ((key = (TObjString*) iter.Next()))
1690 dcsMap->Add(key, dpMap->GetValue(key->String()));
1691
1692 dpMap->SetOwner(kFALSE);
1693 delete dpMap;
a038aa70 1694 }
73abe331 1695 }
1696 }
dc25836b 1697
b1d18693 1698 // save map into file, to help debugging in case of preprocessor error
fbc112e3 1699 TFile* f = TFile::Open("DCSMap.root","recreate");
b1d18693 1700 f->cd();
1701 dcsMap->Write("DCSMap", TObject::kSingleKey);
1702 f->Close();
fbc112e3 1703 delete f;
b1d18693 1704
2bb7b766 1705 // DCS Archive DB processing successful. Call Preprocessor!
85a80aa9 1706 UpdateShuttleStatus(AliShuttleStatus::kPPStarted);
a7160fe9 1707
339adafa 1708 fFXSError = -1; // this variable is kTRUE after ::Process if an FXS error occured
8da81210 1709
a038aa70 1710 UInt_t returnValue = aPreprocessor->Process(dcsMap);
8da81210 1711
339adafa 1712 if (fFXSError!=-1) {
8da81210 1713 UpdateShuttleStatus(AliShuttleStatus::kFXSError);
339adafa 1714 SendMail(kFXSEMail, fFXSError);
8da81210 1715 dcsMap->DeleteAll();
1716 delete dcsMap;
ffa25f20 1717 return 0;
8da81210 1718 }
b948db8d 1719
3301427a 1720 if (returnValue > 0) // Preprocessor error!
1721 {
c88ad5db 1722 Log(fCurrentDetector, Form("ProcessCurrentDetector - "
1723 "Preprocessor failed. Process returned %d.", returnValue));
cb343cfd 1724 UpdateShuttleStatus(AliShuttleStatus::kPPError);
a038aa70 1725 dcsMap->DeleteAll();
1726 delete dcsMap;
ffa25f20 1727 return 0;
9827400b 1728 }
1729
1730 // preprocessor ok!
1731 UpdateShuttleStatus(AliShuttleStatus::kPPDone);
1732 Log(fCurrentDetector, Form("ProcessCurrentDetector - %s preprocessor returned success",
1733 fCurrentDetector.Data()));
b948db8d 1734
a038aa70 1735 dcsMap->DeleteAll();
1736 delete dcsMap;
b948db8d 1737
ffa25f20 1738 return 1;
2bb7b766 1739}
1740
fbc112e3 1741//______________________________________________________________________________________________
1742void AliShuttle::CountOpenRuns()
1743{
1744 // Query DAQ's Shuttle logbook and sends the number of open runs to ML
1745
1746 // check connection, in case connect
1747 if (!Connect(3))
1748 return;
1749
1750 TString sqlQuery;
1751 sqlQuery = Form("select count(*) from %s where shuttle_done=0", fConfig->GetShuttlelbTable());
1752
1753 TSQLResult* aResult = fServer[3]->Query(sqlQuery);
1754 if (!aResult) {
1755 AliError(Form("Can't execute query <%s>!", sqlQuery.Data()));
1756 return;
1757 }
1758
1759 AliDebug(2,Form("Query = %s", sqlQuery.Data()));
1760
1761 if (aResult->GetRowCount() == 0) {
1762 AliError(Form("No result for query %s received", sqlQuery.Data()));
1763 return;
1764 }
1765
1766 if (aResult->GetFieldCount() != 1) {
1767 AliError(Form("Invalid field count for query %s received", sqlQuery.Data()));
1768 return;
1769 }
1770
1771 TSQLRow* aRow = aResult->Next();
1772 if (!aRow) {
1773 AliError(Form("Could not receive result of query %s", sqlQuery.Data()));
1774 return;
1775 }
1776
1777 TString result(aRow->GetField(0), aRow->GetFieldLength(0));
1778 Int_t count = result.Atoi();
1779
1780 Log("SHUTTLE", Form("%d unprocessed runs", count));
1781
1782 delete aRow;
1783 delete aResult;
1784
1785 TMonaLisaValue mlStatus("SHUTTLE_openruns", count);
1786
1787 TList mlList;
1788 mlList.Add(&mlStatus);
1789
1790 fMonaLisa->SendParameters(&mlList, "__PROCESSINGINFO__");
1791}
1792
2bb7b766 1793//______________________________________________________________________________________________
1794Bool_t AliShuttle::QueryShuttleLogbook(const char* whereClause,
1795 TObjArray& entries)
1796{
9827400b 1797 // Query DAQ's Shuttle logbook and fills detector status object.
1798 // Call QueryRunParameters to query DAQ logbook for run parameters.
1799 //
2bb7b766 1800
fc5a4708 1801 entries.SetOwner(1);
1802
2bb7b766 1803 // check connection, in case connect
fbc112e3 1804 if (!Connect(3)) return kFALSE;
2bb7b766 1805
1806 TString sqlQuery;
441b0e9c 1807 sqlQuery = Form("select * from %s %s order by run", fConfig->GetShuttlelbTable(), whereClause);
2bb7b766 1808
be48e3ea 1809 TSQLResult* aResult = fServer[3]->Query(sqlQuery);
2bb7b766 1810 if (!aResult) {
1811 AliError(Form("Can't execute query <%s>!", sqlQuery.Data()));
1812 return kFALSE;
1813 }
1814
fc5a4708 1815 AliDebug(2,Form("Query = %s", sqlQuery.Data()));
1816
2bb7b766 1817 if(aResult->GetRowCount() == 0) {
c88ad5db 1818 Log("SHUTTLE", "No entries in Shuttle Logbook match request");
9827400b 1819 delete aResult;
1820 return kTRUE;
2bb7b766 1821 }
1822
1823 // TODO Check field count!
db99d43e 1824 const UInt_t nCols = 23;
2bb7b766 1825 if (aResult->GetFieldCount() != (Int_t) nCols) {
c88ad5db 1826 Log("SHUTTLE", "Invalid SQL result field number!");
2bb7b766 1827 delete aResult;
1828 return kFALSE;
1829 }
1830
2bb7b766 1831 TSQLRow* aRow;
1832 while ((aRow = aResult->Next())) {
1833 TString runString(aRow->GetField(0), aRow->GetFieldLength(0));
1834 Int_t run = runString.Atoi();
1835
eba76848 1836 AliShuttleLogbookEntry *entry = QueryRunParameters(run);
1837 if (!entry)
1838 continue;
2bb7b766 1839
1840 // loop on detectors
eba76848 1841 for(UInt_t ii = 0; ii < nCols; ii++)
1842 entry->SetDetectorStatus(aResult->GetFieldName(ii), aRow->GetField(ii));
2bb7b766 1843
eba76848 1844 entries.AddLast(entry);
2bb7b766 1845 delete aRow;
1846 }
1847
2bb7b766 1848 delete aResult;
1849 return kTRUE;
1850}
1851
1852//______________________________________________________________________________________________
eba76848 1853AliShuttleLogbookEntry* AliShuttle::QueryRunParameters(Int_t run)
2bb7b766 1854{
eba76848 1855 //
1856 // Retrieve run parameters written in the DAQ logbook and sets them into AliShuttleLogbookEntry object
1857 //
2bb7b766 1858
1859 // check connection, in case connect
be48e3ea 1860 if (!Connect(3))
eba76848 1861 return 0;
2bb7b766 1862
1863 TString sqlQuery;
2c15234c 1864 sqlQuery.Form("select * from %s where run=%d", fConfig->GetDAQlbTable(), run);
2bb7b766 1865
be48e3ea 1866 TSQLResult* aResult = fServer[3]->Query(sqlQuery);
2bb7b766 1867 if (!aResult) {
c88ad5db 1868 Log("SHUTTLE", Form("Can't execute query <%s>!", sqlQuery.Data()));
eba76848 1869 return 0;
2bb7b766 1870 }
1871
eba76848 1872 if (aResult->GetRowCount() == 0) {
2bb7b766 1873 Log("SHUTTLE", Form("QueryRunParameters - No entry in DAQ Logbook for run %d. Skipping", run));
1874 delete aResult;
eba76848 1875 return 0;
2bb7b766 1876 }
1877
eba76848 1878 if (aResult->GetRowCount() > 1) {
c88ad5db 1879 Log("SHUTTLE", Form("QueryRunParameters - UNEXPECTED: "
1880 "more than one entry in DAQ Logbook for run %d!", run));
2bb7b766 1881 delete aResult;
eba76848 1882 return 0;
2bb7b766 1883 }
1884
eba76848 1885 TSQLRow* aRow = aResult->Next();
1886 if (!aRow)
1887 {
c88ad5db 1888 Log("SHUTTLE", Form("QueryRunParameters - Could not retrieve row for run %d. Skipping", run));
eba76848 1889 delete aResult;
1890 return 0;
1891 }
2bb7b766 1892
eba76848 1893 AliShuttleLogbookEntry* entry = new AliShuttleLogbookEntry(run);
2bb7b766 1894
eba76848 1895 for (Int_t ii = 0; ii < aResult->GetFieldCount(); ii++)
1896 entry->SetRunParameter(aResult->GetFieldName(ii), aRow->GetField(ii));
2bb7b766 1897
c9db68c7 1898 delete aRow;
1899 delete aResult;
1900
eba76848 1901 UInt_t startTime = entry->GetStartTime();
1902 UInt_t endTime = entry->GetEndTime();
89bcf3b4 1903 Bool_t ecsSuccess = entry->GetECSSuccess();
c9db68c7 1904
b0e53b15 1905 TString totEventsStr = entry->GetRunParameter("totalEvents");
1906 Int_t totEvents = totEventsStr.Atoi();
c9db68c7 1907
89bcf3b4 1908 if (startTime != 0 && endTime != 0 && endTime > startTime && totEvents > 0 && ecsSuccess)
1909 return entry;
1910
1911 if (ecsSuccess == kFALSE)
1912 {
1913 Log("SHUTTLE", Form("Skipped run %d due to ECS failure, Reason: %s", run, entry->GetRunParameter("eor_reason")));
1914 }
1915 else if (totEvents < 1)
1916 {
c9db68c7 1917 Log("SHUTTLE", Form("QueryRunParameters - Run %d has 0 events - Skipping!", run));
89bcf3b4 1918 }
1919 else
1920 {
1921 Log("SHUTTLE", Form("QueryRunParameters - Invalid parameters for Run %d: "
1922 "startTime = %d, endTime = %d. Skipping (Shuttle won't be marked as DONE)!",
1923 run, startTime, endTime));
1924 }
c9db68c7 1925
1926 //Log("SHUTTLE", Form("Marking SHUTTLE done for run %d", run));
1927 //fLogbookEntry = entry;
1928 //if (!UpdateShuttleLogbook("shuttle_done"))
1929 //{
1930 // AliError(Form("Could not update logbook for run %d !", run));
1931 //}
1932 //fLogbookEntry = 0;
1933
1934 delete entry;
1935 return 0;
2bb7b766 1936}
1937
a038aa70 1938//______________________________________________________________________________________________
1939TMap* AliShuttle::GetValueSet(const char* host, Int_t port, const TSeqCollection* entries,
542b6cc8 1940 DCSType type, Int_t multiSplit)
a038aa70 1941{
1942 // Retrieve all "entry" data points from the DCS server
1943 // host, port: TSocket connection parameters
1944 // entries: list of name of the alias or data point
1945 // type: kAlias or kDP
1946 // returns TMap of values, 0 when failure
542b6cc8 1947
1948 AliDCSClient client(host, port, fTimeout, fRetries, multiSplit);
b41b252a 1949
a038aa70 1950 TMap* result = 0;
b41b252a 1951 if (type == kAlias)
a038aa70 1952 {
b41b252a 1953 result = client.GetAliasValues(entries, GetCurrentStartTime(),
1954 GetCurrentEndTime());
1955 }
1956 else if (type == kDP)
1957 {
1958 result = client.GetDPValues(entries, GetCurrentStartTime(),
1959 GetCurrentEndTime());
1960 }
a038aa70 1961
b41b252a 1962 if (result == 0)
1963 {
1964 Log(fCurrentDetector.Data(), Form("GetValueSet - Can't get entries! Reason: %s",
1790d4b7 1965 client.GetErrorString(client.GetResultErrorCode())));
1966 if (client.GetResultErrorCode() == AliDCSClient::fgkServerError)
1967 Log(fCurrentDetector.Data(), Form("GetValueSet - Server error code: %s",
1968 client.GetServerError().Data()));
a038aa70 1969
b41b252a 1970 return 0;
a038aa70 1971 }
b41b252a 1972
a038aa70 1973 return result;
1974}
b41b252a 1975
b948db8d 1976//______________________________________________________________________________________________
57f50b3c 1977const char* AliShuttle::GetFile(Int_t system, const char* detector,
1978 const char* id, const char* source)
b948db8d 1979{
9827400b 1980 // Get calibration file from file exchange servers
1981 // First queris the FXS database for the file name, using the run, detector, id and source info
1982 // then calls RetrieveFile(filename) for actual copy to local disk
1983 // run: current run being processed (given by Logbook entry fLogbookEntry)
1984 // detector: the Preprocessor name
1985 // id: provided as a parameter by the Preprocessor
1986 // source: provided by the Preprocessor through GetFileSources function
1987
1988 // check if test mode should simulate a FXS error
1989 if (fTestMode & kErrorFXSFiles)
1990 {
1991 Log(detector, Form("GetFile - In TESTMODE - Simulating error while connecting to %s FXS", GetSystemName(system)));
1992 return 0;
1993 }
1994
57f50b3c 1995 // check connection, in case connect
9d733021 1996 if (!Connect(system))
eba76848 1997 {
9d733021 1998 Log(detector, Form("GetFile - Couldn't connect to %s FXS database", GetSystemName(system)));
339adafa 1999 fFXSError = system;
57f50b3c 2000 return 0;
2001 }
2002
2003 // Query preparation
9d733021 2004 TString sourceName(source);
d386d623 2005 Int_t nFields = 3;
2006 TString sqlQueryStart = Form("select filePath,size,fileChecksum from %s where",
2007 fConfig->GetFXSdbTable(system));
2008 TString whereClause = Form("run=%d and detector=\"%s\" and fileId=\"%s\"",
2009 GetCurrentRun(), detector, id);
2010
9d733021 2011 if (system == kDAQ)
2012 {
d386d623 2013 whereClause += Form(" and DAQsource=\"%s\"", source);
57f50b3c 2014 }
9d733021 2015 else if (system == kDCS)
eba76848 2016 {
9d733021 2017 sourceName="none";
57f50b3c 2018 }
9d733021 2019 else if (system == kHLT)
9e080f92 2020 {
d386d623 2021 whereClause += Form(" and DDLnumbers=\"%s\"", source);
9e080f92 2022 }
2023
9e080f92 2024 TString sqlQuery = Form("%s %s", sqlQueryStart.Data(), whereClause.Data());
2025
2026 AliDebug(2, Form("SQL query: \n%s",sqlQuery.Data()));
2027
2028 // Query execution
2029 TSQLResult* aResult = 0;
9d733021 2030 aResult = dynamic_cast<TSQLResult*> (fServer[system]->Query(sqlQuery));
9e080f92 2031 if (!aResult) {
8da81210 2032 Log(detector, Form("GetFile - Can't execute SQL query to %s database for: id = %s, source = %s",
9d733021 2033 GetSystemName(system), id, sourceName.Data()));
339adafa 2034 fFXSError = system;
9e080f92 2035 return 0;
2036 }
2037
8da81210 2038 if (aResult->GetRowCount() == 0)
9e080f92 2039 {
2040 Log(detector,
8da81210 2041 Form("GetFile - No entry in %s FXS db for: id = %s, source = %s",
9d733021 2042 GetSystemName(system), id, sourceName.Data()));
9e080f92 2043 delete aResult;
2044 return 0;
2045 }
2bb7b766 2046
9e080f92 2047 if (aResult->GetRowCount() > 1) {
2048 Log(detector,
8da81210 2049 Form("GetFile - More than one entry in %s FXS db for: id = %s, source = %s",
9d733021 2050 GetSystemName(system), id, sourceName.Data()));
339adafa 2051 fFXSError = system;
9e080f92 2052 delete aResult;
2053 return 0;
2054 }
2055
9d733021 2056 if (aResult->GetFieldCount() != nFields) {
9e080f92 2057 Log(detector,
9d733021 2058 Form("GetFileName - Wrong field count in %s FXS db for: id = %s, source = %s",
2059 GetSystemName(system), id, sourceName.Data()));
339adafa 2060 fFXSError = system;
9e080f92 2061 delete aResult;
2062 return 0;
2063 }
2064
2065 TSQLRow* aRow = dynamic_cast<TSQLRow*> (aResult->Next());
2066
2067 if (!aRow){
8da81210 2068 Log(detector, Form("GetFile - Empty set result in %s FXS db from query: id = %s, source = %s",
9d733021 2069 GetSystemName(system), id, sourceName.Data()));
339adafa 2070 fFXSError = system;
9e080f92 2071 delete aResult;
2072 return 0;
2073 }
2074
2075 TString filePath(aRow->GetField(0), aRow->GetFieldLength(0));
2076 TString fileSize(aRow->GetField(1), aRow->GetFieldLength(1));
d386d623 2077 TString fileChecksum(aRow->GetField(2), aRow->GetFieldLength(2));
9e080f92 2078
2079 delete aResult;
2080 delete aRow;
2081
d386d623 2082 AliDebug(2, Form("filePath = %s; size = %s, fileChecksum = %s",
2083 filePath.Data(), fileSize.Data(), fileChecksum.Data()));
9e080f92 2084
9e080f92 2085 // retrieved file is renamed to make it unique
675f64cd 2086 TString localFileName = Form("%s/%s_%d_process/%s_%s_%d_%s_%s.shuttle",
2087 GetShuttleTempDir(), detector, GetCurrentRun(),
d524ade6 2088 GetSystemName(system), detector, GetCurrentRun(),
2089 id, sourceName.Data());
9d733021 2090
9e080f92 2091
9d733021 2092 // file retrieval from FXS
4b95672b 2093 UInt_t nRetries = 0;
2094 UInt_t maxRetries = 3;
2095 Bool_t result = kFALSE;
2096
2097 // copy!! if successful TSystem::Exec returns 0
8da81210 2098 while (nRetries++ < maxRetries) {
4b95672b 2099 AliDebug(2, Form("Trying to copy file. Retry # %d", nRetries));
2100 result = RetrieveFile(system, filePath.Data(), localFileName.Data());
8da81210 2101 if (!result)
4b95672b 2102 {
8da81210 2103 Log(detector, Form("GetFile - Copy of file %s from %s FXS failed",
9d733021 2104 filePath.Data(), GetSystemName(system)));
4b95672b 2105 continue;
4f0749a8 2106 }
9e080f92 2107
c98417a1 2108 if (fileSize.Length()>0)
2109 {
2110 // compare filesize of local file with the one stored in the FXS DB
926cbc0e 2111 Long_t size = -1;
2112 Int_t sizeComp = gSystem->GetPathInfo(localFileName.Data(), 0, &size, 0, 0);
c98417a1 2113
926cbc0e 2114 if (sizeComp != 0 || size != fileSize.Atoi())
c98417a1 2115 {
8da81210 2116 Log(detector, Form("GetFile - size of file %s does not match with local copy!",
c98417a1 2117 filePath.Data()));
2118 result = kFALSE;
2119 continue;
2120 }
2121
2122 } else {
2123 Log(fCurrentDetector, Form("GetFile - size of file %s not set in %s database, skipping comparison",
2124 filePath.Data(), GetSystemName(system)));
2125 }
2126
d386d623 2127 if (fileChecksum.Length()>0)
4b95672b 2128 {
2129 // compare md5sum of local file with the one stored in the FXS DB
95d4aaef 2130 if(fileChecksum.Contains(' ')) fileChecksum.Resize(fileChecksum.First(' '));
339adafa 2131 Int_t md5Comp = gSystem->Exec(Form("md5sum %s |grep %s > /dev/null 2> /dev/null",
d524ade6 2132 localFileName.Data(), fileChecksum.Data()));
9e080f92 2133
4b95672b 2134 if (md5Comp != 0)
2135 {
8da81210 2136 Log(detector, Form("GetFile - md5sum of file %s does not match with local copy!",
4b95672b 2137 filePath.Data()));
2138 result = kFALSE;
2139 continue;
2140 }
d386d623 2141 } else {
2142 Log(fCurrentDetector, Form("GetFile - md5sum of file %s not set in %s database, skipping comparison",
2143 filePath.Data(), GetSystemName(system)));
9d733021 2144 }
4b95672b 2145 if (result) break;
9e080f92 2146 }
2147
8da81210 2148 if (!result)
2149 {
339adafa 2150 fFXSError = system;
8da81210 2151 return 0;
2152 }
4b95672b 2153
9d733021 2154 fFXSCalled[system]=kTRUE;
2155 TObjString *fileParams = new TObjString(Form("%s#!?!#%s", id, sourceName.Data()));
2156 fFXSlist[system].Add(fileParams);
9e080f92 2157
675f64cd 2158 static TString staticLocalFileName;
2159 staticLocalFileName.Form("%s", localFileName.Data());
2160
c88ad5db 2161 Log(fCurrentDetector, Form("GetFile - Retrieved file with id %s and "
2162 "source %s from %s to %s", id, source,
d524ade6 2163 GetSystemName(system), localFileName.Data()));
675f64cd 2164
d524ade6 2165 return staticLocalFileName.Data();
2bb7b766 2166}
2167
2168//______________________________________________________________________________________________
9d733021 2169Bool_t AliShuttle::RetrieveFile(UInt_t system, const char* fxsFileName, const char* localFileName)
9e080f92 2170{
9827400b 2171 //
2172 // Copies file from FXS to local Shuttle machine
2173 //
2bb7b766 2174
9e080f92 2175 // check temp directory: trying to cd to temp; if it does not exist, create it
d524ade6 2176 AliDebug(2, Form("Copy file %s from %s FXS into %s",
2177 GetSystemName(system), fxsFileName, localFileName));
2178
2179 TString tmpDir(localFileName);
2180
2181 tmpDir = tmpDir(0,tmpDir.Last('/'));
9e080f92 2182
d524ade6 2183 Int_t noDir = gSystem->GetPathInfo(tmpDir.Data(), 0, (Long64_t*) 0, 0, 0);
2184 if (noDir) // temp dir does not exists!
2185 {
2186 if (gSystem->mkdir(tmpDir.Data(), 1))
2187 {
2188 Log(fCurrentDetector.Data(), "RetrieveFile - could not make temp directory!!");
9e080f92 2189 return kFALSE;
2190 }
9e080f92 2191 }
2192
339adafa 2193 TString command = Form("scp -oPort=%d -2 %s@%s:%s/%s %s",
9d733021 2194 fConfig->GetFXSPort(system),
2195 fConfig->GetFXSUser(system),
2196 fConfig->GetFXSHost(system),
339adafa 2197 fConfig->GetFXSBaseFolder(system),
9d733021 2198 fxsFileName,
9e080f92 2199 localFileName);
2200
2201 AliDebug(2, Form("%s",command.Data()));
2202
4b95672b 2203 Bool_t result = (gSystem->Exec(command.Data()) == 0);
9e080f92 2204
4b95672b 2205 return result;
9e080f92 2206}
2207
2208//______________________________________________________________________________________________
9d733021 2209TList* AliShuttle::GetFileSources(Int_t system, const char* detector, const char* id)
2210{
9827400b 2211 //
2212 // Get sources producing the condition file Id from file exchange servers
4a33bdd9 2213 // if id is NULL all sources are returned (distinct)
9827400b 2214 //
1bcd28db 2215
8884d71e 2216 if (id)
2217 {
2218 Log(detector, Form("GetFileSources - Querying %s FXS for files with id %s produced by %s", GetSystemName(system), id, detector));
2219 } else {
2220 Log(detector, Form("GetFileSources - Querying %s FXS for files produced by %s", GetSystemName(system), detector));
2221 }
9827400b 2222
2223 // check if test mode should simulate a FXS error
2224 if (fTestMode & kErrorFXSSources)
2225 {
2226 Log(detector, Form("GetFileSources - In TESTMODE - Simulating error while connecting to %s FXS", GetSystemName(system)));
2227 return 0;
2228 }
2229
9d733021 2230 if (system == kDCS)
2231 {
c88ad5db 2232 Log(detector, "GetFileSources - WARNING: DCS system has only one source of data!");
6297b37d 2233 TList *list = new TList();
2234 list->SetOwner(1);
2235 list->Add(new TObjString(" "));
2236 return list;
9d733021 2237 }
9e080f92 2238
2239 // check connection, in case connect
9d733021 2240 if (!Connect(system))
2241 {
4a33bdd9 2242 Log(detector, Form("GetFileSources - Couldn't connect to %s FXS database", GetSystemName(system)));
339adafa 2243 fFXSError = system;
9d733021 2244 return NULL;
9e080f92 2245 }
2246
9d733021 2247 TString sourceName = 0;
2248 if (system == kDAQ)
2249 {
2250 sourceName = "DAQsource";
2251 } else if (system == kHLT)
2252 {
2253 sourceName = "DDLnumbers";
2254 }
2255
4a33bdd9 2256 TString sqlQueryStart = Form("select distinct %s from %s where", sourceName.Data(), fConfig->GetFXSdbTable(system));
2257 TString whereClause = Form("run=%d and detector=\"%s\"",
2258 GetCurrentRun(), detector);
2259 if (id)
2260 whereClause += Form(" and fileId=\"%s\"", id);
9e080f92 2261 TString sqlQuery = Form("%s %s", sqlQueryStart.Data(), whereClause.Data());
2262
2263 AliDebug(2, Form("SQL query: \n%s",sqlQuery.Data()));
2264
2265 // Query execution
2266 TSQLResult* aResult;
9d733021 2267 aResult = fServer[system]->Query(sqlQuery);
9e080f92 2268 if (!aResult) {
9d733021 2269 Log(detector, Form("GetFileSources - Can't execute SQL query to %s database for id: %s",
2270 GetSystemName(system), id));
339adafa 2271 fFXSError = system;
9e080f92 2272 return 0;
2273 }
2274
86aa42c3 2275 TList *list = new TList();
2276 list->SetOwner(1);
2277
9d733021 2278 if (aResult->GetRowCount() == 0)
2279 {
9e080f92 2280 Log(detector,
9d733021 2281 Form("GetFileSources - No entry in %s FXS table for id: %s", GetSystemName(system), id));
9e080f92 2282 delete aResult;
86aa42c3 2283 return list;
9e080f92 2284 }
2285
1bcd28db 2286 Log(detector, Form("GetFileSources - Found %d sources", aResult->GetRowCount()));
9e080f92 2287
1bcd28db 2288 TSQLRow* aRow;
9d733021 2289 while ((aRow = aResult->Next()))
2290 {
9e080f92 2291
9d733021 2292 TString source(aRow->GetField(0), aRow->GetFieldLength(0));
2293 AliDebug(2, Form("%s = %s", sourceName.Data(), source.Data()));
2294 list->Add(new TObjString(source));
9e080f92 2295 delete aRow;
2296 }
9d733021 2297
9e080f92 2298 delete aResult;
2299
2300 return list;
2bb7b766 2301}
2302
4a33bdd9 2303//______________________________________________________________________________________________
2304TList* AliShuttle::GetFileIDs(Int_t system, const char* detector, const char* source)
2305{
2306 //
2307 // Get all ids of condition files produced by a given source from file exchange servers
2308 //
2309
1bcd28db 2310 Log(detector, Form("GetFileIDs - Retrieving ids with source %s with %s", source, GetSystemName(system)));
2311
4a33bdd9 2312 // check if test mode should simulate a FXS error
2313 if (fTestMode & kErrorFXSSources)
2314 {
2315 Log(detector, Form("GetFileIDs - In TESTMODE - Simulating error while connecting to %s FXS", GetSystemName(system)));
2316 return 0;
2317 }
2318
2319 // check connection, in case connect
2320 if (!Connect(system))
2321 {
2322 Log(detector, Form("GetFileIDs - Couldn't connect to %s FXS database", GetSystemName(system)));
2323 return NULL;
2324 }
2325
2326 TString sourceName = 0;
2327 if (system == kDAQ)
2328 {
2329 sourceName = "DAQsource";
2330 } else if (system == kHLT)
2331 {
2332 sourceName = "DDLnumbers";
2333 }
2334
2335 TString sqlQueryStart = Form("select fileId from %s where", fConfig->GetFXSdbTable(system));
2336 TString whereClause = Form("run=%d and detector=\"%s\"",
2337 GetCurrentRun(), detector);
2338 if (sourceName.Length() > 0 && source)
2339 whereClause += Form(" and %s=\"%s\"", sourceName.Data(), source);
2340 TString sqlQuery = Form("%s %s", sqlQueryStart.Data(), whereClause.Data());
2341
2342 AliDebug(2, Form("SQL query: \n%s",sqlQuery.Data()));
2343
2344 // Query execution
2345 TSQLResult* aResult;
2346 aResult = fServer[system]->Query(sqlQuery);
2347 if (!aResult) {
2348 Log(detector, Form("GetFileIDs - Can't execute SQL query to %s database for source: %s",
2349 GetSystemName(system), source));
2350 return 0;
2351 }
2352
2353 TList *list = new TList();
2354 list->SetOwner(1);
2355
2356 if (aResult->GetRowCount() == 0)
2357 {
2358 Log(detector,
2359 Form("GetFileIDs - No entry in %s FXS table for source: %s", GetSystemName(system), source));
2360 delete aResult;
2361 return list;
2362 }
2363
1bcd28db 2364 Log(detector, Form("GetFileIDs - Found %d ids", aResult->GetRowCount()));
2365
4a33bdd9 2366 TSQLRow* aRow;
2367
2368 while ((aRow = aResult->Next()))
2369 {
2370
2371 TString id(aRow->GetField(0), aRow->GetFieldLength(0));
2372 AliDebug(2, Form("fileId = %s", id.Data()));
2373 list->Add(new TObjString(id));
2374 delete aRow;
2375 }
2376
2377 delete aResult;
2378
2379 return list;
2380}
2381
2bb7b766 2382//______________________________________________________________________________________________
9d733021 2383Bool_t AliShuttle::Connect(Int_t system)
2bb7b766 2384{
9827400b 2385 // Connect to MySQL Server of the system's FXS MySQL databases
2386 // DAQ Logbook, Shuttle Logbook and DAQ FXS db are on the same host
2387 //
57f50b3c 2388
9d733021 2389 // check connection: if already connected return
2390 if(fServer[system] && fServer[system]->IsConnected()) return kTRUE;
57f50b3c 2391
9d733021 2392 TString dbHost, dbUser, dbPass, dbName;
57f50b3c 2393
9d733021 2394 if (system < 3) // FXS db servers
2395 {
2396 dbHost = Form("mysql://%s:%d", fConfig->GetFXSdbHost(system), fConfig->GetFXSdbPort(system));
2397 dbUser = fConfig->GetFXSdbUser(system);
2398 dbPass = fConfig->GetFXSdbPass(system);
2399 dbName = fConfig->GetFXSdbName(system);
2400 } else { // Run & Shuttle logbook servers
2401 // TODO Will the Shuttle logbook server be the same as the Run logbook server ???
2402 dbHost = Form("mysql://%s:%d", fConfig->GetDAQlbHost(), fConfig->GetDAQlbPort());
2403 dbUser = fConfig->GetDAQlbUser();
2404 dbPass = fConfig->GetDAQlbPass();
2405 dbName = fConfig->GetDAQlbDB();
2406 }
57f50b3c 2407
9d733021 2408 fServer[system] = TSQLServer::Connect(dbHost.Data(), dbUser.Data(), dbPass.Data());
2409 if (!fServer[system] || !fServer[system]->IsConnected()) {
2410 if(system < 3)
2411 {
2412 AliError(Form("Can't establish connection to FXS database for %s",
2413 AliShuttleInterface::GetSystemName(system)));
2414 } else {
2415 AliError("Can't establish connection to Run logbook.");
57f50b3c 2416 }
9d733021 2417 if(fServer[system]) delete fServer[system];
2418 return kFALSE;
2bb7b766 2419 }
57f50b3c 2420
9d733021 2421 // Get tables
2422 TSQLResult* aResult=0;
2423 switch(system){
2424 case kDAQ:
2425 aResult = fServer[kDAQ]->GetTables(dbName.Data());
2426 break;
2427 case kDCS:
2428 aResult = fServer[kDCS]->GetTables(dbName.Data());
2429 break;
2430 case kHLT:
2431 aResult = fServer[kHLT]->GetTables(dbName.Data());
2432 break;
2433 default:
2434 aResult = fServer[3]->GetTables(dbName.Data());
2435 break;
2436 }
2437
2438 delete aResult;
2bb7b766 2439 return kTRUE;
2440}
57f50b3c 2441
9e080f92 2442//______________________________________________________________________________________________
9d733021 2443Bool_t AliShuttle::UpdateTable()
9e080f92 2444{
9827400b 2445 //
2446 // Update FXS table filling time_processed field in all rows corresponding to current run and detector
2447 //
9e080f92 2448
9d733021 2449 Bool_t result = kTRUE;
9e080f92 2450
9d733021 2451 for (UInt_t system=0; system<3; system++)
2452 {
2453 if(!fFXSCalled[system]) continue;
9e080f92 2454
9d733021 2455 // check connection, in case connect
2456 if (!Connect(system))
2457 {
2458 Log(fCurrentDetector, Form("UpdateTable - Couldn't connect to %s FXS database", GetSystemName(system)));
2459 result = kFALSE;
2460 continue;
9e080f92 2461 }
9e080f92 2462
9d733021 2463 TTimeStamp now; // now
2464
2465 // Loop on FXS list entries
2466 TIter iter(&fFXSlist[system]);
2467 TObjString *aFXSentry=0;
2468 while ((aFXSentry = dynamic_cast<TObjString*> (iter.Next())))
2469 {
2470 TString aFXSentrystr = aFXSentry->String();
2471 TObjArray *aFXSarray = aFXSentrystr.Tokenize("#!?!#");
2472 if (!aFXSarray || aFXSarray->GetEntries() != 2 )
2473 {
2474 Log(fCurrentDetector, Form("UpdateTable - error updating %s FXS entry. Check string: <%s>",
2475 GetSystemName(system), aFXSentrystr.Data()));
2476 if(aFXSarray) delete aFXSarray;
2477 result = kFALSE;
2478 continue;
2479 }
2480 const char* fileId = ((TObjString*) aFXSarray->At(0))->GetName();
2481 const char* source = ((TObjString*) aFXSarray->At(1))->GetName();
2482
2483 TString whereClause;
2484 if (system == kDAQ)
2485 {
2486 whereClause = Form("where run=%d and detector=\"%s\" and fileId=\"%s\" and DAQsource=\"%s\";",
2487 GetCurrentRun(), fCurrentDetector.Data(), fileId, source);
2488 }
2489 else if (system == kDCS)
2490 {
2491 whereClause = Form("where run=%d and detector=\"%s\" and fileId=\"%s\";",
2492 GetCurrentRun(), fCurrentDetector.Data(), fileId);
2493 }
2494 else if (system == kHLT)
2495 {
2496 whereClause = Form("where run=%d and detector=\"%s\" and fileId=\"%s\" and DDLnumbers=\"%s\";",
2497 GetCurrentRun(), fCurrentDetector.Data(), fileId, source);
2498 }
2499
2500 delete aFXSarray;
9e080f92 2501
9d733021 2502 TString sqlQuery = Form("update %s set time_processed=%d %s", fConfig->GetFXSdbTable(system),
2503 now.GetSec(), whereClause.Data());
9e080f92 2504
9d733021 2505 AliDebug(2, Form("SQL query: \n%s",sqlQuery.Data()));
9e080f92 2506
9d733021 2507 // Query execution
2508 TSQLResult* aResult;
2509 aResult = dynamic_cast<TSQLResult*> (fServer[system]->Query(sqlQuery));
2510 if (!aResult)
2511 {
2512 Log(fCurrentDetector, Form("UpdateTable - %s db: can't execute SQL query <%s>",
2513 GetSystemName(system), sqlQuery.Data()));
2514 result = kFALSE;
2515 continue;
2516 }
2517 delete aResult;
9e080f92 2518 }
9e080f92 2519 }
2520
9d733021 2521 return result;
9e080f92 2522}
57f50b3c 2523
3301427a 2524//______________________________________________________________________________________________
2525Bool_t AliShuttle::UpdateTableFailCase()
2526{
9827400b 2527 // Update FXS table filling time_processed field in all rows corresponding to current run and detector
2528 // this is called in case the preprocessor is declared failed for the current run, because
2529 // the fields are updated only in case of success
3301427a 2530
2531 Bool_t result = kTRUE;
2532
2533 for (UInt_t system=0; system<3; system++)
2534 {
2535 // check connection, in case connect
2536 if (!Connect(system))
2537 {
2538 Log(fCurrentDetector, Form("UpdateTableFailCase - Couldn't connect to %s FXS database",
2539 GetSystemName(system)));
2540 result = kFALSE;
2541 continue;
2542 }
2543
2544 TTimeStamp now; // now
2545
2546 // Loop on FXS list entries
2547
2548 TString whereClause = Form("where run=%d and detector=\"%s\";",
2549 GetCurrentRun(), fCurrentDetector.Data());
2550
2551
2552 TString sqlQuery = Form("update %s set time_processed=%d %s", fConfig->GetFXSdbTable(system),
2553 now.GetSec(), whereClause.Data());
2554
2555 AliDebug(2, Form("SQL query: \n%s",sqlQuery.Data()));
2556
2557 // Query execution
2558 TSQLResult* aResult;
2559 aResult = dynamic_cast<TSQLResult*> (fServer[system]->Query(sqlQuery));
2560 if (!aResult)
2561 {
2562 Log(fCurrentDetector, Form("UpdateTableFailCase - %s db: can't execute SQL query <%s>",
2563 GetSystemName(system), sqlQuery.Data()));
2564 result = kFALSE;
2565 continue;
2566 }
2567 delete aResult;
2568 }
2569
2570 return result;
2571}
2572
2bb7b766 2573//______________________________________________________________________________________________
2574Bool_t AliShuttle::UpdateShuttleLogbook(const char* detector, const char* status)
2575{
e7f62f16 2576 //
2577 // Update Shuttle logbook filling detector or shuttle_done column
2578 // ex. of usage: UpdateShuttleLogbook("PHOS", "DONE") or UpdateShuttleLogbook("shuttle_done")
2579 //
57f50b3c 2580
2bb7b766 2581 // check connection, in case connect
be48e3ea 2582 if(!Connect(3)){
2bb7b766 2583 Log("SHUTTLE", "UpdateShuttleLogbook - Couldn't connect to DAQ Logbook.");
2584 return kFALSE;
57f50b3c 2585 }
2586
2bb7b766 2587 TString detName(detector);
2588 TString setClause;
ee6f7523 2589 if (detName == "shuttle_done" || detName == "shuttle_ignored")
e7f62f16 2590 {
2bb7b766 2591 setClause = "set shuttle_done=1";
e7f62f16 2592
ee6f7523 2593 if (detName == "shuttle_done")
b0e53b15 2594 {
503ff24f 2595 if (TouchFile()==kTRUE){
2596 //Send the information to ML
2597 TMonaLisaText mlStatus("SHUTTLE_status", "Done");
e7f62f16 2598
503ff24f 2599 TList mlList;
2600 mlList.Add(&mlStatus);
2601
2602 TString mlID;
2603 mlID.Form("%d", GetCurrentRun());
2604 fMonaLisa->SendParameters(&mlList, mlID);
2605 }
2606 else{
2607 return kFALSE;
2608 }
2609
b0e53b15 2610 }
2bb7b766 2611 } else {
2bb7b766 2612 TString statusStr(status);
2613 if(statusStr.Contains("done", TString::kIgnoreCase) ||
2614 statusStr.Contains("failed", TString::kIgnoreCase)){
eba76848 2615 setClause = Form("set %s=\"%s\"", detector, status);
2bb7b766 2616 } else {
2617 Log("SHUTTLE",
2618 Form("UpdateShuttleLogbook - Invalid status <%s> for detector %s",
2619 status, detector));
2620 return kFALSE;
2621 }
2622 }
57f50b3c 2623
2bb7b766 2624 TString whereClause = Form("where run=%d", GetCurrentRun());
2625
441b0e9c 2626 TString sqlQuery = Form("update %s %s %s",
2627 fConfig->GetShuttlelbTable(), setClause.Data(), whereClause.Data());
57f50b3c 2628
2bb7b766 2629 AliDebug(2, Form("SQL query: \n%s",sqlQuery.Data()));
2630
2631 // Query execution
2632 TSQLResult* aResult;
be48e3ea 2633 aResult = dynamic_cast<TSQLResult*> (fServer[3]->Query(sqlQuery));
2bb7b766 2634 if (!aResult) {
2635 Log("SHUTTLE", Form("UpdateShuttleLogbook - Can't execute query <%s>", sqlQuery.Data()));
2636 return kFALSE;
57f50b3c 2637 }
2bb7b766 2638 delete aResult;
57f50b3c 2639
2640 return kTRUE;
2641}
2642
2643//______________________________________________________________________________________________
2bb7b766 2644Int_t AliShuttle::GetCurrentRun() const
2645{
9827400b 2646 //
2647 // Get current run from logbook entry
2648 //
57f50b3c 2649
2bb7b766 2650 return fLogbookEntry ? fLogbookEntry->GetRun() : -1;
57f50b3c 2651}
2652
2653//______________________________________________________________________________________________
2bb7b766 2654UInt_t AliShuttle::GetCurrentStartTime() const
2655{
9827400b 2656 //
2657 // get current start time
2658 //
57f50b3c 2659
2bb7b766 2660 return fLogbookEntry ? fLogbookEntry->GetStartTime() : 0;
57f50b3c 2661}
2662
2663//______________________________________________________________________________________________
2bb7b766 2664UInt_t AliShuttle::GetCurrentEndTime() const
2665{
9827400b 2666 //
2667 // get current end time from logbook entry
2668 //
57f50b3c 2669
2bb7b766 2670 return fLogbookEntry ? fLogbookEntry->GetEndTime() : 0;
57f50b3c 2671}
2672
675f64cd 2673//______________________________________________________________________________________________
2674UInt_t AliShuttle::GetCurrentYear() const
2675{
2676 //
2677 // Get current year from logbook entry
2678 //
2679
2680 if (!fLogbookEntry) return 0;
2681
2682 TTimeStamp startTime(GetCurrentStartTime());
2683 TString year = Form("%d",startTime.GetDate());
2684 year = year(0,4);
2685
2686 return year.Atoi();
2687}
2688
2689//______________________________________________________________________________________________
2690const char* AliShuttle::GetLHCPeriod() const
2691{
2692 //
2693 // Get current LHC period from logbook entry
2694 //
2695
2696 if (!fLogbookEntry) return 0;
2697
2698 return fLogbookEntry->GetRunParameter("LHCperiod");
2699}
2700
b948db8d 2701//______________________________________________________________________________________________
2702void AliShuttle::Log(const char* detector, const char* message)
2703{
9827400b 2704 //
2705 // Fill log string with a message
2706 //
b948db8d 2707
7d4cf768 2708 TString logRunDir = GetShuttleLogDir();
2709 if (GetCurrentRun() >=0)
2710 logRunDir += Form("/%d", GetCurrentRun());
2711
2712 void* dir = gSystem->OpenDirectory(logRunDir.Data());
84090f85 2713 if (dir == NULL) {
7d4cf768 2714 if (gSystem->mkdir(logRunDir.Data(), kTRUE)) {
36c99a6a 2715 AliError(Form("Can't open directory <%s>", GetShuttleLogDir()));
84090f85 2716 return;
2717 }
b948db8d 2718
84090f85 2719 } else {
2720 gSystem->FreeDirectory(dir);
2721 }
b948db8d 2722
0b025a52 2723 TString toLog = Form("%s UTC (%d): %s - ", TTimeStamp(time(0)).AsString("s"), getpid(), detector);
e7f62f16 2724 if (GetCurrentRun() >= 0)
2725 toLog += Form("run %d - ", GetCurrentRun());
2bb7b766 2726 toLog += Form("%s", message);
2727
84090f85 2728 AliInfo(toLog.Data());
ffa29e93 2729
2730 // if we redirect the log output already to the file, leave here
2731 if (fOutputRedirected && strcmp(detector, "SHUTTLE") != 0)
2732 return;
b948db8d 2733
ffa29e93 2734 TString fileName = GetLogFileName(detector);
e7f62f16 2735
84090f85 2736 gSystem->ExpandPathName(fileName);
2737
2738 ofstream logFile;
2739 logFile.open(fileName, ofstream::out | ofstream::app);
2740
2741 if (!logFile.is_open()) {
2742 AliError(Form("Could not open file %s", fileName.Data()));
2743 return;
2744 }
7bfb2090 2745
84090f85 2746 logFile << toLog.Data() << "\n";
b948db8d 2747
84090f85 2748 logFile.close();
b948db8d 2749}
2bb7b766 2750
ffa29e93 2751//______________________________________________________________________________________________
2752TString AliShuttle::GetLogFileName(const char* detector) const
2753{
2754 //
2755 // returns the name of the log file for a given sub detector
2756 //
2757
2758 TString fileName;
2759
2760 if (GetCurrentRun() >= 0)
7d4cf768 2761 {
2762 fileName.Form("%s/%d/%s_%d.log", GetShuttleLogDir(), GetCurrentRun(),
2763 detector, GetCurrentRun());
2764 } else {
ffa29e93 2765 fileName.Form("%s/%s.log", GetShuttleLogDir(), detector);
7d4cf768 2766 }
ffa29e93 2767
2768 return fileName;
2769}
2770
ee6f7523 2771//______________________________________________________________________________________________
2772void AliShuttle::SendAlive()
2773{
2774 // sends alive message to ML
2775
2776 TMonaLisaText mlStatus("SHUTTLE_status", "Alive");
2777
2778 TList mlList;
2779 mlList.Add(&mlStatus);
2780
2781 fMonaLisa->SendParameters(&mlList, "__PROCESSINGINFO__");
2782}
2783
2bb7b766 2784//______________________________________________________________________________________________
2785Bool_t AliShuttle::Collect(Int_t run)
2786{
9827400b 2787 //
2788 // Collects conditions data for all UNPROCESSED run written to DAQ LogBook in case of run = -1 (default)
2789 // If a dedicated run is given this run is processed
2790 //
2791 // In operational mode, this is the Shuttle function triggered by the EOR signal.
2792 //
2bb7b766 2793
eba76848 2794 if (run == -1)
2795 Log("SHUTTLE","Collect - Shuttle called. Collecting conditions data for unprocessed runs");
2796 else
2797 Log("SHUTTLE", Form("Collect - Shuttle called. Collecting conditions data for run %d", run));
cb343cfd 2798
2799 SetLastAction("Starting");
2bb7b766 2800
ee6f7523 2801 // create ML instance
2802 if (!fMonaLisa)
2803 fMonaLisa = new TMonaLisaWriter(fConfig->GetMonitorHost(), fConfig->GetMonitorTable());
2804
ee6f7523 2805 SendAlive();
fbc112e3 2806 CountOpenRuns();
ee6f7523 2807
2bb7b766 2808 TString whereClause("where shuttle_done=0");
eba76848 2809 if (run != -1)
2810 whereClause += Form(" and run=%d", run);
2bb7b766 2811
2812 TObjArray shuttleLogbookEntries;
be48e3ea 2813 if (!QueryShuttleLogbook(whereClause, shuttleLogbookEntries))
2814 {
cb343cfd 2815 Log("SHUTTLE", "Collect - Can't retrieve entries from Shuttle logbook");
2bb7b766 2816 return kFALSE;
2817 }
2818
9e080f92 2819 if (shuttleLogbookEntries.GetEntries() == 0)
2820 {
2821 if (run == -1)
2822 Log("SHUTTLE","Collect - Found no UNPROCESSED runs in Shuttle logbook");
2823 else
2824 Log("SHUTTLE", Form("Collect - Run %d is already DONE "
2825 "or it does not exist in Shuttle logbook", run));
2826 return kTRUE;
2827 }
2828
be48e3ea 2829 for (UInt_t iDet=0; iDet<NDetectors(); iDet++)
2830 fFirstUnprocessed[iDet] = kTRUE;
2831
fc5a4708 2832 if (run != -1)
be48e3ea 2833 {
2834 // query Shuttle logbook for earlier runs, check if some detectors are unprocessed,
2835 // flag them into fFirstUnprocessed array
2836 TString whereClause(Form("where shuttle_done=0 and run < %d", run));
2837 TObjArray tmpLogbookEntries;
2838 if (!QueryShuttleLogbook(whereClause, tmpLogbookEntries))
2839 {
2840 Log("SHUTTLE", "Collect - Can't retrieve entries from Shuttle logbook");
2841 return kFALSE;
2842 }
2843
2844 TIter iter(&tmpLogbookEntries);
2845 AliShuttleLogbookEntry* anEntry = 0;
2846 while ((anEntry = dynamic_cast<AliShuttleLogbookEntry*> (iter.Next())))
2847 {
2848 for (UInt_t iDet=0; iDet<NDetectors(); iDet++)
2849 {
2850 if (anEntry->GetDetectorStatus(iDet) == AliShuttleLogbookEntry::kUnprocessed)
2851 {
2852 AliDebug(2, Form("Run %d: setting %s as \"not first time unprocessed\"",
2853 anEntry->GetRun(), GetDetName(iDet)));
2854 fFirstUnprocessed[iDet] = kFALSE;
2855 }
2856 }
2857
2858 }
2859
2860 }
2861
2862 if (!RetrieveConditionsData(shuttleLogbookEntries))
2863 {
cb343cfd 2864 Log("SHUTTLE", "Collect - Process of at least one run failed");
a3ca69fe 2865 CountOpenRuns();
2bb7b766 2866 return kFALSE;
2867 }
2868
36c99a6a 2869 Log("SHUTTLE", "Collect - Requested run(s) successfully processed");
a3ca69fe 2870 CountOpenRuns();
eba76848 2871 return kTRUE;
2bb7b766 2872}
2873
2bb7b766 2874//______________________________________________________________________________________________
2875Bool_t AliShuttle::RetrieveConditionsData(const TObjArray& dateEntries)
2876{
9827400b 2877 //
2878 // Retrieve conditions data for all runs that aren't processed yet
2879 //
2bb7b766 2880
2881 Bool_t hasError = kFALSE;
2882
2883 TIter iter(&dateEntries);
2884 AliShuttleLogbookEntry* anEntry;
2885
2886 while ((anEntry = (AliShuttleLogbookEntry*) iter.Next())){
2887 if (!Process(anEntry)){
2888 hasError = kTRUE;
2889 }
4b95672b 2890
2891 // clean SHUTTLE temp directory
d524ade6 2892 //TString filename = Form("%s/*.shuttle", GetShuttleTempDir());
2893 //RemoveFile(filename.Data());
2bb7b766 2894 }
2895
2896 return hasError == kFALSE;
2897}
cb343cfd 2898
2899//______________________________________________________________________________________________
2900ULong_t AliShuttle::GetTimeOfLastAction() const
2901{
9827400b 2902 //
2903 // Gets time of last action
2904 //
2905
cb343cfd 2906 ULong_t tmp;
36c99a6a 2907
cb343cfd 2908 fMonitoringMutex->Lock();
be48e3ea 2909
cb343cfd 2910 tmp = fLastActionTime;
36c99a6a 2911
cb343cfd 2912 fMonitoringMutex->UnLock();
36c99a6a 2913
cb343cfd 2914 return tmp;
2915}
2916
2917//______________________________________________________________________________________________
2918const TString AliShuttle::GetLastAction() const
2919{
9827400b 2920 //
cb343cfd 2921 // returns a string description of the last action
9827400b 2922 //
cb343cfd 2923
2924 TString tmp;
36c99a6a 2925
cb343cfd 2926 fMonitoringMutex->Lock();
2927
2928 tmp = fLastAction;
2929
2930 fMonitoringMutex->UnLock();
2931
36c99a6a 2932 return tmp;
cb343cfd 2933}
2934
2935//______________________________________________________________________________________________
2936void AliShuttle::SetLastAction(const char* action)
2937{
9827400b 2938 //
cb343cfd 2939 // updates the monitoring variables
9827400b 2940 //
36c99a6a 2941
cb343cfd 2942 fMonitoringMutex->Lock();
36c99a6a 2943
cb343cfd 2944 fLastAction = action;
2945 fLastActionTime = time(0);
2946
2947 fMonitoringMutex->UnLock();
2948}
eba76848 2949
2950//______________________________________________________________________________________________
2951const char* AliShuttle::GetRunParameter(const char* param)
2952{
9827400b 2953 //
2954 // returns run parameter read from DAQ logbook
2955 //
eba76848 2956
2957 if(!fLogbookEntry) {
2958 AliError("No logbook entry!");
2959 return 0;
2960 }
2961
2962 return fLogbookEntry->GetRunParameter(param);
2963}
57c1a579 2964
d386d623 2965//______________________________________________________________________________________________
9827400b 2966AliCDBEntry* AliShuttle::GetFromOCDB(const char* detector, const AliCDBPath& path)
d386d623 2967{
9827400b 2968 //
2969 // returns object from OCDB valid for current run
2970 //
d386d623 2971
9827400b 2972 if (fTestMode & kErrorOCDB)
2973 {
2974 Log(detector, "GetFromOCDB - In TESTMODE - Simulating error with OCDB");
2975 return 0;
2976 }
2977
d386d623 2978 AliCDBStorage *sto = AliCDBManager::Instance()->GetStorage(fgkMainCDB);
2979 if (!sto)
2980 {
9827400b 2981 Log(detector, "GetFromOCDB - Cannot activate main OCDB for query!");
d386d623 2982 return 0;
2983 }
2984
2985 return dynamic_cast<AliCDBEntry*> (sto->Get(path, GetCurrentRun()));
2986}
2987
57c1a579 2988//______________________________________________________________________________________________
339adafa 2989Bool_t AliShuttle::SendMail(EMailTarget target, Int_t system)
57c1a579 2990{
9827400b 2991 //
2992 // sends a mail to the subdetector expert in case of preprocessor error
2993 //
2994
2995 if (fTestMode != kNone)
2996 return kTRUE;
fb40ead4 2997
926cbc0e 2998 if (!fConfig->SendMail())
2999 return kTRUE;
57c1a579 3000
339adafa 3001 if (target == kDCSEMail || target == kFXSEMail) {
3002 if (!fFirstProcessing)
3003 return kTRUE;
36c99a6a 3004 }
3005
7bdfe1fa 3006 Int_t runMode = (Int_t)fConfig->GetRunMode();
3007 TString tmpStr;
3008 if (runMode == 0) tmpStr = " Nightly Test:";
3009 else tmpStr = " Data Taking:";
fbc112e3 3010 void* dir = gSystem->OpenDirectory(GetShuttleLogDir());
3011 if (dir == NULL)
3012 {
3013 if (gSystem->mkdir(GetShuttleLogDir(), kTRUE))
3014 {
3015 Log("SHUTTLE", Form("SendMail - Can't open directory <%s>", GetShuttleLogDir()));
3016 return kFALSE;
3017 }
3018
3019 } else {
3020 gSystem->FreeDirectory(dir);
3021 }
3022
339adafa 3023 // det experts in to
3024 TString to="";
3835a613 3025 TIter *iterExperts = 0;
339adafa 3026 if (target == kDCSEMail) {
3027 iterExperts = new TIter(fConfig->GetAdmins(AliShuttleConfig::kAmanda));
3028 }
3029 else if (target == kFXSEMail) {
3030 iterExperts = new TIter(fConfig->GetAdmins(system));
3031 }
3835a613 3032 if (iterExperts) {
3033 TObjString *anExpert=0;
3034 while ((anExpert = (TObjString*) iterExperts->Next()))
3035 {
3036 to += Form("%s,", anExpert->GetName());
3037 }
3038 delete iterExperts;
339adafa 3039 }
3835a613 3040
3041 // add subdetector experts
3042 iterExperts = new TIter(fConfig->GetResponsibles(fCurrentDetector));
339adafa 3043 TObjString *anExpert=0;
3044 while ((anExpert = (TObjString*) iterExperts->Next()))
4508ab8b 3045 {
339adafa 3046 to += Form("%s,", anExpert->GetName());
3047 }
3048 delete iterExperts;
3835a613 3049
339adafa 3050 if (to.Length() > 0)
3051 to.Remove(to.Length()-1);
3052 AliDebug(2, Form("to: %s",to.Data()));
4508ab8b 3053
339adafa 3054 if (to.IsNull()) {
3055 Log("SHUTTLE", Form("List of %d responsibles not set!", (Int_t) target));
3056 return kFALSE;
3057 }
3058
3059 // SHUTTLE responsibles in cc
fb40ead4 3060 TString cc="";
3061 TIter iterAdmins(fConfig->GetAdmins(AliShuttleConfig::kGlobal));
3062 TObjString *anAdmin=0;
3063 while ((anAdmin = (TObjString*) iterAdmins.Next()))
3064 {
3065 cc += Form("%s,", anAdmin->GetName());
3066 }
3067 if (cc.Length() > 0)
c19963db 3068 cc.Remove(cc.Length()-1);
fb40ead4 3069 AliDebug(2, Form("cc: %s",to.Data()));
57c1a579 3070
339adafa 3071 // mail body
675f64cd 3072 TString bodyFileName;
3073 bodyFileName.Form("%s/mail.body", GetShuttleLogDir());
3074 gSystem->ExpandPathName(bodyFileName);
3075
3076 ofstream mailBody;
3077 mailBody.open(bodyFileName, ofstream::out);
3078
3079 if (!mailBody.is_open())
3080 {
339adafa 3081 Log("SHUTTLE", Form("Could not open mail body file %s", bodyFileName.Data()));
675f64cd 3082 return kFALSE;
3083 }
3084
675f64cd 3085
339adafa 3086 TString subject;
3087 TString body;
675f64cd 3088
339adafa 3089 if (target == kDCSEMail){
7bdfe1fa 3090 subject = Form("%s Retrieval of data points for %s FAILED in run %d !",
3091 tmpStr.Data(), fCurrentDetector.Data(), GetCurrentRun());
339adafa 3092 AliDebug(2, Form("subject: %s", subject.Data()));
3093
3094 body = Form("Dear DCS experts, \n\n");
3095 body += Form("SHUTTLE couldn\'t retrieve the data points for detector %s "
3096 "in run %d!!\n\n", fCurrentDetector.Data(), GetCurrentRun());
fb40ead4 3097 }
339adafa 3098 else if (target == kFXSEMail){
7bdfe1fa 3099 subject = Form("%s FXS communication for %s FAILED in run %d !",
3100 tmpStr.Data(), fCurrentDetector.Data(), GetCurrentRun());
339adafa 3101 AliDebug(2, Form("subject: %s", subject.Data()));
3102 TString sys;
3103 if (system == kDAQ) sys="DAQ";
3104 else if (system == kDCS) sys="DCS";
3105 else if (system == kHLT) sys="HLT";
3106 else return kFALSE;
9c0e2d64 3107 body = Form("Dear %s FXS experts, \n\n",sys.Data());
339adafa 3108 body += Form("SHUTTLE couldn\'t retrieve data from the FXS for detector %s "
9c0e2d64 3109 "in run %d!!\n\n", fCurrentDetector.Data(), GetCurrentRun());
339adafa 3110 }
3111 else {
7bdfe1fa 3112 subject = Form("%s %s Shuttle preprocessor FAILED in run %d (run type = %s)!",
3113 tmpStr.Data(), fCurrentDetector.Data(), GetCurrentRun(), GetRunType());
339adafa 3114 AliDebug(2, Form("subject: %s", subject.Data()));
3115
3116 body = Form("Dear %s expert(s), \n\n", fCurrentDetector.Data());
3117 body += Form("SHUTTLE just detected that your preprocessor "
3118 "failed processing run %d (run type = %s)!!\n\n",
3119 GetCurrentRun(), GetRunType());
3120 }
675f64cd 3121
7d4cf768 3122 body += Form("Please check %s status on the SHUTTLE monitoring page: \n\n",
3123 fCurrentDetector.Data());
b0e53b15 3124 if (fConfig->GetRunMode() == AliShuttleConfig::kTest)
3125 {
3126 body += Form("\thttp://pcalimonitor.cern.ch:8889/shuttle.jsp?time=168 \n\n");
3127 } else {
339adafa 3128 body += Form("\thttp://pcalimonitor.cern.ch/shuttle.jsp?instance=PROD&time=168 \n\n");
b0e53b15 3129 }
339adafa 3130
3131
7d4cf768 3132 TString logFolder = "logs";
3133 if (fConfig->GetRunMode() == AliShuttleConfig::kProd)
3134 logFolder += "_PROD";
3135
3136
675f64cd 3137 body += Form("Find the %s log for the current run on \n\n"
7d4cf768 3138 "\thttp://pcalishuttle01.cern.ch:8880/%s/%d/%s_%d.log \n\n",
3139 fCurrentDetector.Data(), logFolder.Data(), GetCurrentRun(),
3140 fCurrentDetector.Data(), GetCurrentRun());
3141 body += Form("The last 10 lines of %s log file are following:\n\n", fCurrentDetector.Data());
675f64cd 3142
3143 AliDebug(2, Form("Body begin: %s", body.Data()));
3144
3145 mailBody << body.Data();
3146 mailBody.close();
3147 mailBody.open(bodyFileName, ofstream::out | ofstream::app);
3148
339adafa 3149 TString logFileName = Form("%s/%d/%s_%d.log", GetShuttleLogDir(),
3150 GetCurrentRun(), fCurrentDetector.Data(), GetCurrentRun());
675f64cd 3151 TString tailCommand = Form("tail -n 10 %s >> %s", logFileName.Data(), bodyFileName.Data());
3152 if (gSystem->Exec(tailCommand.Data()))
3153 {
3154 mailBody << Form("%s log file not found ...\n\n", fCurrentDetector.Data());
3155 }
3156
3157 TString endBody = Form("------------------------------------------------------\n\n");
36c99a6a 3158 endBody += Form("In case of problems please contact the SHUTTLE core team.\n\n");
3159 endBody += "Please do not answer this message directly, it is automatically generated.\n\n";
546242fb 3160 endBody += "Greetings,\n\n \t\t\tthe SHUTTLE\n";
57c1a579 3161
909732f7 3162 AliDebug(2, Form("Body end: %s", endBody.Data()));
57c1a579 3163
3164 mailBody << endBody.Data();
3165
3166 mailBody.close();
3167
3168 // send mail!
3169 TString mailCommand = Form("mail -s \"%s\" -c %s %s < %s",
3170 subject.Data(),
3171 cc.Data(),
3172 to.Data(),
3173 bodyFileName.Data());
909732f7 3174 AliDebug(2, Form("mail command: %s", mailCommand.Data()));
57c1a579 3175
3176 Bool_t result = gSystem->Exec(mailCommand.Data());
3177
3178 return result == 0;
3179}
d386d623 3180
441b0e9c 3181//______________________________________________________________________________________________
9827400b 3182const char* AliShuttle::GetRunType()
441b0e9c 3183{
9827400b 3184 //
3185 // returns run type read from "run type" logbook
3186 //
441b0e9c 3187
3188 if(!fLogbookEntry) {
3189 AliError("No logbook entry!");
3190 return 0;
3191 }
3192
9827400b 3193 return fLogbookEntry->GetRunType();
441b0e9c 3194}
3195
4859271b 3196//______________________________________________________________________________________________
3197Bool_t AliShuttle::GetHLTStatus()
3198{
3199 // Return HLT status (ON=1 OFF=0)
af76b20d 3200 // Converts the HLT status from the mode string read in the run logbook (not just a bool)
4859271b 3201
3202 if(!fLogbookEntry) {
3203 AliError("No logbook entry!");
3204 return 0;
3205 }
3206
af76b20d 3207 // TODO implement when HLTMode is inserted in run logbook
3208 TString hltMode = fLogbookEntry->GetRunParameter("HLTMode");
3209 TSubString firstChar = hltMode(0,1);
3210 AliDebug(2,Form("First char = %s ",firstChar.Data()));
3211 if (firstChar == "A") {
3212 return kFALSE;
3213 }
3214 else if ((firstChar == "B") || (firstChar == "C") || (firstChar == "D") || (firstChar == "E")) {
3215 return kTRUE;
3216 }
3217 else {
3218 Log("SHUTTLE","Unexpected HLT mode! Returning 0....");
3219 return kFALSE;
3220 }
4859271b 3221}
ca8ea066 3222
3223//______________________________________________________________________________________________
3224const char* AliShuttle::GetTriggerConfiguration()
3225{
3226 // Receives the trigger configuration from the DAQ logbook for the current run
3227
3228 // check connection, if needed reconnect
3229 if (!Connect(3))
3230 return 0;
3231
3232 TString sqlQuery;
3233 sqlQuery.Form("SELECT configFile FROM logbook_trigger_config WHERE run = %d", GetCurrentRun());
3234 TSQLResult* result = fServer[3]->Query(sqlQuery);
3235 if (!result)
3236 {
3237 Log("SHUTTLE", Form("ERROR: Can't execute query <%s>!", sqlQuery.Data()));
3238 return 0;
3239 }
3240
3241 if (result->GetRowCount() == 0)
3242 {
3243 Log("SHUTTLE", "ERROR: Trigger configuration not found in logbook_trigger_config");
3244 delete result;
3245 return 0;
3246 }
3247
3248 TSQLRow* row = result->Next();
3249 if (!row)
3250 {
3251 Log("SHUTTLE", "ERROR: Could not receive logbook_trigger_config data");
3252 delete result;
3253 return 0;
3254 }
3255
3256 // static, so that pointer remains valid when it is returned to the calling class
3257 static TString triggerConfig(row->GetField(0));
3258
3259 delete row;
3260 row = 0;
3261
3262 delete result;
3263 result = 0;
3264
3265 Log("SHUTTLE", Form("Found trigger configuration: %s", triggerConfig.Data()));
3266
3267 return triggerConfig;
3268}
3269
d386d623 3270//______________________________________________________________________________________________
3271void AliShuttle::SetShuttleTempDir(const char* tmpDir)
3272{
9827400b 3273 //
3274 // sets Shuttle temp directory
3275 //
d386d623 3276
3277 fgkShuttleTempDir = gSystem->ExpandPathName(tmpDir);
3278}
3279
3280//______________________________________________________________________________________________
3281void AliShuttle::SetShuttleLogDir(const char* logDir)
3282{
9827400b 3283 //
3284 // sets Shuttle log directory
3285 //
d386d623 3286
3287 fgkShuttleLogDir = gSystem->ExpandPathName(logDir);
3288}
503ff24f 3289//______________________________________________________________________________________________
3290Bool_t AliShuttle::TouchFile()
3291{
3292 //
3293 // touching a file on the grid if run has been DONE
3294 //
3295
3296 if (!gGrid)
3297 {
3298 Log("SHUTTLE",Form("No TGrid connection estabilished!"));
3299 Log("SHUTTLE",Form("Could not touch file for run %i",GetCurrentRun()));
3300 return kFALSE;
3301 }
3302
b8cf5e0f 3303 TString dir;
3304 dir.Form("%s%d/%s/SHUTTLE_DONE", fConfig->GetAlienPath(), GetCurrentYear(), GetLHCPeriod());
3305 // checking whether directory for touch command exists
3306 TString commandLs;
3307 commandLs.Form("ls %s",dir.Data());
3308 TGridResult *resultLs = dynamic_cast<TGridResult*>(gGrid->Command(commandLs));
3309 if (!resultLs){
3310 Log("SHUTTLE",Form("No result for %s command, returning without touching",commandLs.Data()));
3311 return kFALSE;
3312 }
3313 TMap *mapLs = dynamic_cast<TMap*>(resultLs->At(0));
3314 if (!mapLs){
3315 Log("SHUTTLE",Form("No map for %s command, returning without touching",commandLs.Data()));
3316 return kFALSE;
3317 }
3318 TObjString *valueLsPath = dynamic_cast<TObjString*>(mapLs->GetValue("path"));
3319 if (!valueLsPath || (TString)(valueLsPath->GetString()).CompareTo(dir)!=1){
3320 Log("SHUTTLE",Form("No directory %s found, creating it",dir.Data()));
3321
3322 // creating the directory
3323
3324 Bool_t boolMkdir = gGrid->Mkdir(dir.Data());
3325 if (!boolMkdir) {
3326 Log("SHUTTLE",Form("Impossible to create dir %s in alien catalogue for run %i!",dir.Data(),GetCurrentRun()));
3327 return kFALSE;
3328 }
3329 Log("SHUTTLE",Form("Directory %s successfully created in alien catalogue for run %i",dir.Data(),GetCurrentRun()));
3330 }
3331 else {
3332 Log("SHUTTLE",Form("Directory %s correctly found for run %i",dir.Data(),GetCurrentRun()));
3333 }
3334
503ff24f 3335 TString command;
b8cf5e0f 3336 command.Form("touch %s/%i", dir.Data(), GetCurrentRun());
e660ff60 3337 Log("SHUTTLE", Form("Creating entry in file catalog: %s", command.Data()));
503ff24f 3338 TGridResult *resultTouch = dynamic_cast<TGridResult*>(gGrid->Command(command));
b8cf5e0f 3339 if (!resultTouch){
3340 Log("SHUTTLE",Form("No result for touching command, returning without touching for run %i",GetCurrentRun()));
3341 return kFALSE;
503ff24f 3342 }
b8cf5e0f 3343 TMap *mapTouch = dynamic_cast<TMap*>(resultTouch->At(0));
3344 if (!mapTouch){
3345 Log("SHUTTLE",Form("No map for touching command, returning without touching for run %i",GetCurrentRun()));
3346 return kFALSE;
3347 }
3348 TObjString *valueTouch = dynamic_cast<TObjString*>(mapTouch->GetValue("__result__"));
3349 if (!valueTouch){
3350 Log("SHUTTLE",Form("No value for \"__result__\" key set in the map for touching command, returning without touching for run %i",GetCurrentRun()));
3351 return kFALSE;
503ff24f 3352 }
b8cf5e0f 3353 if (valueTouch->GetString()!="1"){
3354 Log("SHUTTLE",Form("Failing the touching command, returning without touching for run %i",GetCurrentRun()));
3355 return kFALSE;
3356 }
3357 return kTRUE;
503ff24f 3358}
3359
3360