]> git.uio.no Git - u/mrichter/AliRoot.git/blame - SHUTTLE/AliShuttle.cxx
git://git.uio.no / u / mrichter / AliRoot.git / blame
? search:
summary | shortlog | log | commit | commitdiff | tree
blob | blame (incremental) | history | HEAD
Renamed output file to Vertex.Performance.root
[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
924fb6e9 203
9827400b 204 if (fTestMode & kErrorStorage)
205 {
206 Log(fCurrentDetector, "StoreLocally - In TESTMODE - Simulating error while storing locally");
207 return kFALSE;
208 }
209
3301427a 210 const char* cdbType = (localUri == fgkLocalCDB) ? "CDB" : "Reference";
2bb7b766 211
85a80aa9 212 Int_t firstRun = GetCurrentRun() - validityStart;
84090f85 213 if(firstRun < 0) {
9827400b 214 AliWarning("First valid run happens to be less than 0! Setting it to 0.");
84090f85 215 firstRun=0;
216 }
217
218 Int_t lastRun = -1;
219 if(validityInfinite) {
220 lastRun = AliCDBRunRange::Infinity();
221 } else {
222 lastRun = GetCurrentRun();
223 }
224
3301427a 225 // Version is set to current run, it will be used later to transfer data to Grid
226 AliCDBId id(path, firstRun, lastRun, GetCurrentRun(), -1);
2bb7b766 227
228 if(! dynamic_cast<TObjString*> (metaData->GetProperty("RunUsed(TObjString)"))){
229 TObjString runUsed = Form("%d", GetCurrentRun());
9e080f92 230 metaData->SetProperty("RunUsed(TObjString)", runUsed.Clone());
2bb7b766 231 }
84090f85 232
3301427a 233 Bool_t result = kFALSE;
84090f85 234
3301427a 235 if (!(AliCDBManager::Instance()->GetStorage(localUri))) {
236 Log("SHUTTLE", Form("StoreLocally - Cannot activate local %s storage", cdbType));
84090f85 237 } else {
924fb6e9 238 Int_t logLevel = AliLog::GetGlobalLogLevel();
239 AliLog::SetGlobalLogLevel(AliLog::kError);
3301427a 240 result = AliCDBManager::Instance()->GetStorage(localUri)
84090f85 241 ->Put(object, id, metaData);
924fb6e9 242 AliLog::SetGlobalLogLevel((AliLog::EType_t)logLevel);
84090f85 243 }
244
245 if(!result) {
246
9827400b 247 Log(fCurrentDetector, Form("StoreLocally - Can't store object <%s>!", id.ToString().Data()));
3301427a 248 }
2bb7b766 249
924fb6e9 250
3301427a 251 return result;
252}
84090f85 253
3301427a 254//______________________________________________________________________________________________
255Bool_t AliShuttle::StoreOCDB()
256{
9827400b 257 //
258 // Called when preprocessor ends successfully or when previous storage attempt failed (kStoreError status)
259 // Calls underlying StoreOCDB(const char*) function twice, for OCDB and Reference storage.
260 // Then calls StoreRefFilesToGrid to store reference files.
261 //
262
5e993b6f 263 UpdateShuttleStatus(AliShuttleStatus::kStoreStarted);
56d3e9e4 264
9827400b 265 if (fTestMode & kErrorGrid)
266 {
267 Log("SHUTTLE", "StoreOCDB - In TESTMODE - Simulating error while storing in the Grid");
268 Log(fCurrentDetector, "StoreOCDB - In TESTMODE - Simulating error while storing in the Grid");
269 return kFALSE;
270 }
271
c88ad5db 272 Log("SHUTTLE","StoreOCDB - Storing OCDB data ...");
5e993b6f 273 Int_t resultCDB = StoreOCDB(fgkMainCDB);
86aa42c3 274
c88ad5db 275 Log("SHUTTLE","StoreOCDB - Storing reference data ...");
5e993b6f 276 Int_t resultRef = StoreOCDB(fgkMainRefStorage);
9827400b 277
c88ad5db 278 Log("SHUTTLE","StoreOCDB - Storing reference files ...");
279 Bool_t resultRefFiles = CopyFilesToGrid("reference");
280
281 Bool_t resultMetadata = kTRUE;
282 if(fCurrentDetector == "GRP")
283 {
7802e884 284 Log("SHUTTLE","StoreOCDB - Storing Run Metadata file ...");
c88ad5db 285 resultMetadata = CopyFilesToGrid("metadata");
286 }
9827400b 287
5e993b6f 288 Int_t storeResult = 0;
289
290 if (resultCDB < 0 || resultRef < 0 || resultRefFiles == kFALSE || resultMetadata == kFALSE)
291 storeResult = -1;
292 else if (resultCDB > 0 || resultRef > 0)
293 storeResult = 1;
294
295 if (storeResult < 0)
296 {
297 Log("SHUTTLE",
298 Form("\t\t\t****** run %d - %s: STORAGE ERROR ******",
299 GetCurrentRun(), fCurrentDetector.Data()));
300 UpdateShuttleStatus(AliShuttleStatus::kStoreError);
301 }
302 else if (storeResult > 0)
303 {
304 Log("SHUTTLE",
305 Form("\t\t\t****** run %d - %s: STORAGE DELAYED ******",
306 GetCurrentRun(), fCurrentDetector.Data()));
307 UpdateShuttleStatus(AliShuttleStatus::kStoreDelayed);
308 }
309 else if (storeResult == 0)
310 {
311 Log("SHUTTLE",
312 Form("\t\t\t****** run %d - %s: DONE ******",
313 GetCurrentRun(), fCurrentDetector.Data()));
314 UpdateShuttleStatus(AliShuttleStatus::kDone);
315 UpdateShuttleLogbook(fCurrentDetector, "DONE");
316 }
317
318 return (storeResult == 0);
3301427a 319}
320
321//______________________________________________________________________________________________
5e993b6f 322Int_t AliShuttle::StoreOCDB(const TString& gridURI)
3301427a 323{
324 //
325 // Called by StoreOCDB(), performs actual storage to the main OCDB and reference storages (Grid)
326 //
5e993b6f 327 // Return code:
328 // -2 initialization error
329 // -1 storage error
330 // 0 success
331 // 1 storage delayed (e.g. previous unprocessed runs)
332 //
3301427a 333
334 TObjArray* gridIds=0;
335
336 Bool_t result = kTRUE;
5e993b6f 337 Bool_t delayed = kFALSE;
3301427a 338
339 const char* type = 0;
340 TString localURI;
341 if(gridURI == fgkMainCDB) {
342 type = "OCDB";
343 localURI = fgkLocalCDB;
344 } else if(gridURI == fgkMainRefStorage) {
345 type = "reference";
346 localURI = fgkLocalRefStorage;
347 } else {
348 AliError(Form("Invalid storage URI: %s", gridURI.Data()));
5e993b6f 349 return -2;
3301427a 350 }
351
352 AliCDBManager* man = AliCDBManager::Instance();
353
354 AliCDBStorage *gridSto = man->GetStorage(gridURI);
355 if(!gridSto) {
356 Log("SHUTTLE",
357 Form("StoreOCDB - cannot activate main %s storage", type));
5e993b6f 358 return -2;
3301427a 359 }
360
361 gridIds = gridSto->GetQueryCDBList();
362
363 // get objects previously stored in local CDB
364 AliCDBStorage *localSto = man->GetStorage(localURI);
365 if(!localSto) {
366 Log("SHUTTLE",
367 Form("StoreOCDB - cannot activate local %s storage", type));
5e993b6f 368 return -2;
3301427a 369 }
370 AliCDBPath aPath(GetOfflineDetName(fCurrentDetector.Data()),"*","*");
371 // Local objects were stored with current run as Grid version!
372 TList* localEntries = localSto->GetAll(aPath.GetPath(), GetCurrentRun(), GetCurrentRun());
373 localEntries->SetOwner(1);
374
375 // loop on local stored objects
376 TIter localIter(localEntries);
377 AliCDBEntry *aLocEntry = 0;
378 while((aLocEntry = dynamic_cast<AliCDBEntry*> (localIter.Next()))){
379 aLocEntry->SetOwner(1);
380 AliCDBId aLocId = aLocEntry->GetId();
381 aLocEntry->SetVersion(-1);
382 aLocEntry->SetSubVersion(-1);
6c0c9948 383
384 Log(fCurrentDetector.Data(), Form("Attempting to store %s", aLocId.ToString().Data()));
3301427a 385
386 // If local object is valid up to infinity we store it only if it is
387 // the first unprocessed run!
388 if (aLocId.GetLastRun() == AliCDBRunRange::Infinity() &&
389 !fFirstUnprocessed[GetDetPos(fCurrentDetector)])
390 {
391 Log("SHUTTLE", Form("StoreOCDB - %s: object %s has validity infinite but "
392 "there are previous unprocessed runs!",
393 fCurrentDetector.Data(), aLocId.GetPath().Data()));
5e993b6f 394 Log(fCurrentDetector.Data(), Form("StoreOCDB - %s: object %s has validity infinite but "
395 "there are previous unprocessed runs!",
396 fCurrentDetector.Data(), aLocId.GetPath().Data()));
397 delayed = kTRUE;
3301427a 398 continue;
399 }
400
401 // loop on Grid valid Id's
402 Bool_t store = kTRUE;
403 TIter gridIter(gridIds);
404 AliCDBId* aGridId = 0;
6c0c9948 405 while ((aGridId = dynamic_cast<AliCDBId*> (gridIter.Next()))) {
406 if (aGridId->GetPath() != aLocId.GetPath())
407 continue;
3301427a 408 // skip all objects valid up to infinity
6c0c9948 409 if (aGridId->GetLastRun() == AliCDBRunRange::Infinity())
410 continue;
411
3301427a 412 // if we get here, it means there's already some more recent object stored on Grid!
6c0c9948 413 Log(fCurrentDetector.Data(),
414 Form("StoreOCDB - A more recent object already exists in %s storage: <%s>",
415 type, aGridId->ToString().Data()));
416
3301427a 417 store = kFALSE;
418 break;
419 }
420
6c0c9948 421 Bool_t storeOk = kFALSE;
422 if (store)
423 {
424 Log(fCurrentDetector.Data(), Form("Prechecks succeeded. Ready to store %s", aLocId.ToString().Data()));
425 storeOk = gridSto->Put(aLocEntry);
426 if (storeOk) {
3301427a 427 Log("SHUTTLE",
6c0c9948 428 Form("StoreOCDB - Object <%s> successfully put into %s storage",
429 aLocId.ToString().Data(), type));
2d9019b4 430 Log(fCurrentDetector.Data(),
431 Form("StoreOCDB - Object <%s> successfully put into %s storage",
6c0c9948 432 aLocId.ToString().Data(), type));
433 } else {
434 Log("SHUTTLE",
435 Form("StoreOCDB - Grid %s storage of object <%s> failed",
436 type, aLocId.ToString().Data()));
437 Log(fCurrentDetector.Data(),
438 Form("StoreOCDB - Grid %s storage of object <%s> failed",
439 type, aLocId.ToString().Data()));
440 result = kFALSE;
3301427a 441 }
6c0c9948 442 }
443
444 if (!store || storeOk) {
445 // removing local file...
3301427a 446 TString filename;
447 localSto->IdToFilename(aLocId, filename);
c88ad5db 448 Log("SHUTTLE", Form("StoreOCDB - Removing local file %s", filename.Data()));
3301427a 449 RemoveFile(filename.Data());
b948db8d 450 }
451 }
3301427a 452 localEntries->Clear();
2bb7b766 453
5e993b6f 454 Int_t returnCode = 0;
455
456 if (result == kFALSE)
457 returnCode = -1;
458 else if (delayed != kFALSE)
459 returnCode = 1;
460
461 Log("SHUTTLE", Form("StoreOCDB - Returning with %d (result = %d, delayed = %d)", returnCode, result, delayed));
462 Log(fCurrentDetector.Data(), Form("StoreOCDB - Returning with %d (result = %d, delayed = %d)", returnCode, result, delayed));
463
464 return returnCode;
3301427a 465}
466
546242fb 467//______________________________________________________________________________________________
468Bool_t AliShuttle::CleanReferenceStorage(const char* detector)
469{
2d9019b4 470 // clears the directory used to store reference files of a given subdetector
546242fb 471
472 AliCDBManager* man = AliCDBManager::Instance();
473 AliCDBStorage* sto = man->GetStorage(fgkLocalRefStorage);
2d9019b4 474 TString localBaseFolder = sto->GetBaseFolder();
475
476 TString targetDir = GetRefFilePrefix(localBaseFolder.Data(), detector);
477
d524ade6 478 Log("SHUTTLE", Form("CleanReferenceStorage - Cleaning %s", targetDir.Data()));
2d9019b4 479
480 TString begin;
481 begin.Form("%d_", GetCurrentRun());
482
483 TSystemDirectory* baseDir = new TSystemDirectory("/", targetDir);
484 if (!baseDir)
485 return kTRUE;
486
487 TList* dirList = baseDir->GetListOfFiles();
488 delete baseDir;
489
490 if (!dirList) return kTRUE;
491
492 if (dirList->GetEntries() < 3)
493 {
494 delete dirList;
495 return kTRUE;
496 }
497
498 Int_t nDirs = 0, nDel = 0;
499 TIter dirIter(dirList);
500 TSystemFile* entry = 0;
546242fb 501
2d9019b4 502 Bool_t success = kTRUE;
546242fb 503
2d9019b4 504 while ((entry = dynamic_cast<TSystemFile*> (dirIter.Next())))
505 {
506 if (entry->IsDirectory())
507 continue;
508
509 TString fileName(entry->GetName());
510 if (!fileName.BeginsWith(begin))
511 continue;
512
513 nDirs++;
514
515 // delete file
516 Int_t result = gSystem->Unlink(fileName.Data());
517
518 if (result)
519 {
d524ade6 520 Log("SHUTTLE", Form("CleanReferenceStorage - Could not delete file %s!", fileName.Data()));
2d9019b4 521 success = kFALSE;
522 } else {
523 nDel++;
524 }
525 }
526
527 if(nDirs > 0)
528 Log("SHUTTLE", Form("CleanReferenceStorage - %d (over %d) reference files in folder %s were deleted.",
529 nDel, nDirs, targetDir.Data()));
530
531
532 delete dirList;
533 return success;
534
535
536
537
538
546242fb 539
540 Int_t result = gSystem->GetPathInfo(targetDir, 0, (Long64_t*) 0, 0, 0);
541 if (result == 0)
542 {
543 // delete directory
d524ade6 544 result = gSystem->Exec(Form("rm -rf %s", targetDir.Data()));
546242fb 545 if (result != 0)
546 {
d524ade6 547 Log("SHUTTLE", Form("CleanReferenceStorage - Could not clean directory %s", targetDir.Data()));
546242fb 548 return kFALSE;
549 }
550 }
551
552 result = gSystem->mkdir(targetDir, kTRUE);
553 if (result != 0)
554 {
c88ad5db 555 Log("SHUTTLE", Form("CleanReferenceStorage - Error creating base directory %s", targetDir.Data()));
546242fb 556 return kFALSE;
557 }
558
559 return kTRUE;
560}
561
9827400b 562//______________________________________________________________________________________________
563Bool_t AliShuttle::StoreReferenceFile(const char* detector, const char* localFile, const char* gridFileName)
564{
565 //
3c2a21c8 566 // Stores reference file directly (without opening it). This function stores the file locally.
9827400b 567 //
3c2a21c8 568 // The file is stored under the following location:
569 // <base folder of local reference storage>/<DET>/<RUN#>_<gridFileName>
570 // where <gridFileName> is the second parameter given to the function
571 //
9827400b 572
573 if (fTestMode & kErrorStorage)
574 {
575 Log(fCurrentDetector, "StoreReferenceFile - In TESTMODE - Simulating error while storing locally");
576 return kFALSE;
577 }
578
579 AliCDBManager* man = AliCDBManager::Instance();
580 AliCDBStorage* sto = man->GetStorage(fgkLocalRefStorage);
581
582 TString localBaseFolder = sto->GetBaseFolder();
583
d524ade6 584 TString target = GetRefFilePrefix(localBaseFolder.Data(), detector);
585 target.Append(Form("/%d_%s", GetCurrentRun(), gridFileName));
9827400b 586
d524ade6 587 return CopyFileLocally(localFile, target);
c88ad5db 588}
589
590//______________________________________________________________________________________________
591Bool_t AliShuttle::StoreRunMetadataFile(const char* localFile, const char* gridFileName)
592{
593 //
594 // Stores Run metadata file to the Grid, in the run folder
595 //
596 // Only GRP can call this function.
597
598 if (fTestMode & kErrorStorage)
599 {
600 Log(fCurrentDetector, "StoreRunMetaDataFile - In TESTMODE - Simulating error while storing locally");
601 return kFALSE;
602 }
603
604 AliCDBManager* man = AliCDBManager::Instance();
605 AliCDBStorage* sto = man->GetStorage(fgkLocalRefStorage);
606
607 TString localBaseFolder = sto->GetBaseFolder();
608
609 // Build Run level folder
955d2abc 610 // folder = /alice/data/year/lhcPeriod/runNb/raw
c88ad5db 611
c88ad5db 612
675f64cd 613 TString lhcPeriod = GetLHCPeriod();
c88ad5db 614 if (lhcPeriod.Length() == 0)
615 {
616 Log("SHUTTLE","StoreRunMetaDataFile - LHCPeriod not found in logbook!");
617 return 0;
618 }
a1c5723b 619
620 // TODO partitions with one detector only write data into LHCperiod_DET
2694f253 621 TString partition = GetRunParameter("detector");
a1c5723b 622
623 if (partition.Length() > 0 && partition != "ALICE")
624 {
625 lhcPeriod.Append(Form("_%s", partition.Data()));
626 Log(fCurrentDetector, Form("Run data tags merged file will be written in %s",
627 lhcPeriod.Data()));
628 }
8884d71e 629
b8cf5e0f 630 TString target = Form("%s/GRP/RunMetadata%s%d/%s/%09d/raw/%s",
631 localBaseFolder.Data(), fConfig->GetAlienPath(), GetCurrentYear(),
d524ade6 632 lhcPeriod.Data(), GetCurrentRun(), gridFileName);
c88ad5db 633
d524ade6 634 return CopyFileLocally(localFile, target);
c88ad5db 635}
636
637//______________________________________________________________________________________________
d524ade6 638Bool_t AliShuttle::CopyFileLocally(const char* localFile, const TString& target)
c88ad5db 639{
640 //
641 // Stores file locally. Called by StoreReferenceFile and StoreRunMetadataFile
d524ade6 642 // Files are temporarily stored in the local reference storage. When the preprocessor
643 // finishes, the Shuttle calls CopyFilesToGrid to transfer the files to AliEn
644 // (in reference or run level folders)
c88ad5db 645 //
646
d524ade6 647 TString targetDir(target(0, target.Last('/')));
648
649 //try to open base dir folder, if it does not exist
2d9019b4 650 void* dir = gSystem->OpenDirectory(targetDir.Data());
651 if (dir == NULL) {
652 if (gSystem->mkdir(targetDir.Data(), kTRUE)) {
a1c5723b 653 Log("SHUTTLE", Form("CopyFileLocally - Can't open directory <%s>", targetDir.Data()));
2d9019b4 654 return kFALSE;
655 }
656
657 } else {
658 gSystem->FreeDirectory(dir);
659 }
9827400b 660
7d43a416 661 Int_t result = 0;
662
663 result = gSystem->GetPathInfo(localFile, 0, (Long64_t*) 0, 0, 0);
9827400b 664 if (result)
665 {
a1c5723b 666 Log("SHUTTLE", Form("CopyFileLocally - %s does not exist", localFile));
546242fb 667 return kFALSE;
9827400b 668 }
546242fb 669
7d43a416 670 result = gSystem->GetPathInfo(target, 0, (Long64_t*) 0, 0, 0);
671 if (!result)
672 {
a1c5723b 673 Log("SHUTTLE", Form("CopyFileLocally - target file %s already exist, removing...", target.Data()));
7d43a416 674 if (gSystem->Unlink(target.Data()))
675 {
a1c5723b 676 Log("SHUTTLE", Form("CopyFileLocally - Could not remove existing target file %s!", target.Data()));
7d43a416 677 return kFALSE;
678 }
679 }
680
9827400b 681 result = gSystem->CopyFile(localFile, target);
682
683 if (result == 0)
684 {
a1c5723b 685 Log("SHUTTLE", Form("CopyFileLocally - File %s stored locally to %s", localFile, target.Data()));
9827400b 686 return kTRUE;
687 }
688 else
689 {
a1c5723b 690 Log("SHUTTLE", Form("CopyFileLocally - Could not store file %s to %s! Error code = %d",
546242fb 691 localFile, target.Data(), result));
9827400b 692 return kFALSE;
693 }
c88ad5db 694
695
696
9827400b 697}
698
699//______________________________________________________________________________________________
c88ad5db 700Bool_t AliShuttle::CopyFilesToGrid(const char* type)
9827400b 701{
702 //
c88ad5db 703 // Transfers local files to the Grid. Local files can be reference files
704 // or run metadata file (from GRP only).
9827400b 705 //
c88ad5db 706 // According to the type (ref, metadata) the files are stored under the following location:
707 // ref --> <base folder of reference storage>/<DET>/<RUN#>_<gridFileName>
708 // metadata --> <run data folder>/<MetadataFileName>
86aa42c3 709 //
c88ad5db 710
9827400b 711 AliCDBManager* man = AliCDBManager::Instance();
712 AliCDBStorage* sto = man->GetStorage(fgkLocalRefStorage);
713 if (!sto)
714 return kFALSE;
715 TString localBaseFolder = sto->GetBaseFolder();
9827400b 716
c88ad5db 717 TString dir;
718 TString alienDir;
9827400b 719 TString begin;
9827400b 720
c88ad5db 721 if (strcmp(type, "reference") == 0)
722 {
723 dir = GetRefFilePrefix(localBaseFolder.Data(), fCurrentDetector.Data());
724 AliCDBStorage* gridSto = man->GetStorage(fgkMainRefStorage);
725 if (!gridSto)
726 return kFALSE;
727 TString gridBaseFolder = gridSto->GetBaseFolder();
728 alienDir = GetRefFilePrefix(gridBaseFolder.Data(), fCurrentDetector.Data());
729 begin = Form("%d_", GetCurrentRun());
730 }
731 else if (strcmp(type, "metadata") == 0)
732 {
c88ad5db 733
675f64cd 734 TString lhcPeriod = GetLHCPeriod();
c88ad5db 735
736 if (lhcPeriod.Length() == 0)
737 {
738 Log("SHUTTLE","CopyFilesToGrid - LHCPeriod not found in logbook!");
739 return 0;
740 }
741
a1c5723b 742 // TODO partitions with one detector only write data into LHCperiod_DET
2694f253 743 TString partition = GetRunParameter("detector");
a1c5723b 744
745 if (partition.Length() > 0 && partition != "ALICE")
746 {
747 lhcPeriod.Append(Form("_%s", partition.Data()));
748 }
749
b8cf5e0f 750 dir = Form("%s/GRP/RunMetadata%s%d/%s/%09d/raw",
751 localBaseFolder.Data(), fConfig->GetAlienPath(), GetCurrentYear(),
c88ad5db 752 lhcPeriod.Data(), GetCurrentRun());
b8cf5e0f 753 alienDir = dir(dir.Index(fConfig->GetAlienPath()), dir.Length());
675f64cd 754
c88ad5db 755 begin = "";
756 }
757 else
758 {
759 Log("SHUTTLE", "CopyFilesToGrid - Unexpected: type label must be reference or metadata!");
760 return kFALSE;
761 }
762
9827400b 763 TSystemDirectory* baseDir = new TSystemDirectory("/", dir);
3d8bc902 764 if (!baseDir)
765 return kTRUE;
766
2d9019b4 767 TList* dirList = baseDir->GetListOfFiles();
768 delete baseDir;
769
770 if (!dirList) return kTRUE;
771
772 if (dirList->GetEntries() < 3)
3d8bc902 773 {
2d9019b4 774 delete dirList;
9827400b 775 return kTRUE;
3d8bc902 776 }
2d9019b4 777
546242fb 778 if (!gGrid)
779 {
c88ad5db 780 Log("SHUTTLE", "CopyFilesToGrid - Connection to Grid failed: Cannot continue!");
2d9019b4 781 delete dirList;
546242fb 782 return kFALSE;
783 }
784
2d9019b4 785 Int_t nDirs = 0, nTransfer = 0;
786 TIter dirIter(dirList);
787 TSystemFile* entry = 0;
788
9827400b 789 Bool_t success = kTRUE;
3d8bc902 790 Bool_t first = kTRUE;
9827400b 791
2d9019b4 792 while ((entry = dynamic_cast<TSystemFile*> (dirIter.Next())))
793 {
9827400b 794 if (entry->IsDirectory())
795 continue;
796
797 TString fileName(entry->GetName());
798 if (!fileName.BeginsWith(begin))
799 continue;
800
2d9019b4 801 nDirs++;
802
3d8bc902 803 if (first)
804 {
805 first = kFALSE;
c88ad5db 806 // check that folder exists, otherwise create it
3d8bc902 807 TGridResult* result = gGrid->Ls(alienDir.Data(), "a");
808
809 if (!result)
2d9019b4 810 {
811 delete dirList;
3d8bc902 812 return kFALSE;
2d9019b4 813 }
3d8bc902 814
546242fb 815 if (!result->GetFileName(1)) // TODO: It looks like element 0 is always 0!!
3d8bc902 816 {
675f64cd 817 // TODO It does not work currently! Bug in TAliEn::Mkdir
818 // TODO Manually fixed in local root v5-16-00
c88ad5db 819 if (!gGrid->Mkdir(alienDir.Data(),"-p",0))
3d8bc902 820 {
c88ad5db 821 Log("SHUTTLE", Form("CopyFilesToGrid - Cannot create directory %s",
3d8bc902 822 alienDir.Data()));
2d9019b4 823 delete dirList;
3d8bc902 824 return kFALSE;
546242fb 825 } else {
c88ad5db 826 Log("SHUTTLE",Form("CopyFilesToGrid - Folder %s created", alienDir.Data()));
3d8bc902 827 }
828
546242fb 829 } else {
c88ad5db 830 Log("SHUTTLE",Form("CopyFilesToGrid - Folder %s found", alienDir.Data()));
3d8bc902 831 }
832 }
833
9827400b 834 TString fullLocalPath;
835 fullLocalPath.Form("%s/%s", dir.Data(), fileName.Data());
836
837 TString fullGridPath;
838 fullGridPath.Form("alien://%s/%s", alienDir.Data(), fileName.Data());
839
a986b218 840 Bool_t result = TFile::Cp(fullLocalPath, fullGridPath);
9827400b 841
842 if (result)
843 {
c88ad5db 844 Log("SHUTTLE", Form("CopyFilesToGrid - Copying local file %s to %s succeeded!",
845 fullLocalPath.Data(), fullGridPath.Data()));
9827400b 846 RemoveFile(fullLocalPath);
2d9019b4 847 nTransfer++;
9827400b 848 }
849 else
850 {
c88ad5db 851 Log("SHUTTLE", Form("CopyFilesToGrid - Copying local file %s to %s FAILED!",
852 fullLocalPath.Data(), fullGridPath.Data()));
9827400b 853 success = kFALSE;
854 }
855 }
2d9019b4 856
c88ad5db 857 Log("SHUTTLE", Form("CopyFilesToGrid - %d (over %d) files in folder %s copied to Grid.",
858 nTransfer, nDirs, dir.Data()));
2d9019b4 859
860
861 delete dirList;
9827400b 862 return success;
863}
864
2d9019b4 865//______________________________________________________________________________________________
866const char* AliShuttle::GetRefFilePrefix(const char* base, const char* detector)
867{
868 //
869 // Get folder name of reference files
870 //
871
872 TString offDetStr(GetOfflineDetName(detector));
3dd835d8 873 static TString dir;
2d9019b4 874 if (offDetStr == "ITS" || offDetStr == "MUON" || offDetStr == "PHOS")
875 {
876 dir.Form("%s/%s/%s", base, offDetStr.Data(), detector);
877 } else {
878 dir.Form("%s/%s", base, offDetStr.Data());
879 }
880
881 return dir.Data();
2d9019b4 882}
c88ad5db 883
3301427a 884//______________________________________________________________________________________________
885void AliShuttle::CleanLocalStorage(const TString& uri)
886{
9827400b 887 //
888 // Called in case the preprocessor is declared failed. Remove remaining objects from the local storages.
889 //
3301427a 890
891 const char* type = 0;
892 if(uri == fgkLocalCDB) {
893 type = "OCDB";
894 } else if(uri == fgkLocalRefStorage) {
546242fb 895 type = "Reference";
3301427a 896 } else {
897 AliError(Form("Invalid storage URI: %s", uri.Data()));
898 return;
899 }
900
901 AliCDBManager* man = AliCDBManager::Instance();
b948db8d 902
3301427a 903 // open local storage
904 AliCDBStorage *localSto = man->GetStorage(uri);
905 if(!localSto) {
906 Log("SHUTTLE",
907 Form("CleanLocalStorage - cannot activate local %s storage", type));
908 return;
909 }
910
911 TString filename(Form("%s/%s/*/Run*_v%d_s*.root",
546242fb 912 localSto->GetBaseFolder().Data(), GetOfflineDetName(fCurrentDetector.Data()), GetCurrentRun()));
3301427a 913
c88ad5db 914 AliDebug(2, Form("filename = %s", filename.Data()));
3301427a 915
c88ad5db 916 Log("SHUTTLE", Form("Removing remaining local files for run %d and detector %s ...",
3301427a 917 GetCurrentRun(), fCurrentDetector.Data()));
918
919 RemoveFile(filename.Data());
920
921}
922
923//______________________________________________________________________________________________
924void AliShuttle::RemoveFile(const char* filename)
925{
9827400b 926 //
927 // removes local file
928 //
3301427a 929
930 TString command(Form("rm -f %s", filename));
931
932 Int_t result = gSystem->Exec(command.Data());
933 if(result != 0)
934 {
935 Log("SHUTTLE", Form("RemoveFile - %s: Cannot remove file %s!",
936 fCurrentDetector.Data(), filename));
937 }
73abe331 938}
939
b948db8d 940//______________________________________________________________________________________________
5164a766 941AliShuttleStatus* AliShuttle::ReadShuttleStatus()
942{
9827400b 943 //
944 // Reads the AliShuttleStatus from the CDB
945 //
5164a766 946
2bb7b766 947 if (fStatusEntry){
948 delete fStatusEntry;
949 fStatusEntry = 0;
950 }
5164a766 951
3ebbf3c4 952 Int_t path1 = GetCurrentRun()/10000;
10a5a932 953 fStatusEntry = AliCDBManager::Instance()->GetStorage(GetLocalCDB())
3ebbf3c4 954 ->Get(Form("/SHUTTLE/%s/%d", fCurrentDetector.Data(), path1), GetCurrentRun());
5164a766 955
2bb7b766 956 if (!fStatusEntry) return 0;
957 fStatusEntry->SetOwner(1);
5164a766 958
2bb7b766 959 AliShuttleStatus* status = dynamic_cast<AliShuttleStatus*> (fStatusEntry->GetObject());
960 if (!status) {
961 AliError("Invalid object stored to CDB!");
962 return 0;
963 }
5164a766 964
2bb7b766 965 return status;
5164a766 966}
967
968//______________________________________________________________________________________________
7bfb2090 969Bool_t AliShuttle::WriteShuttleStatus(AliShuttleStatus* status)
5164a766 970{
9827400b 971 //
972 // writes the status for one subdetector
973 //
2bb7b766 974
975 if (fStatusEntry){
976 delete fStatusEntry;
977 fStatusEntry = 0;
978 }
5164a766 979
2bb7b766 980 Int_t run = GetCurrentRun();
3ebbf3c4 981 Int_t path1 = run/10000;
982 TString path1_string = Form("%d",path1);
5164a766 983
3ebbf3c4 984 AliCDBId id(AliCDBPath("SHUTTLE", fCurrentDetector, path1_string), run, run);
5164a766 985
2bb7b766 986 fStatusEntry = new AliCDBEntry(status, id, new AliCDBMetaData);
987 fStatusEntry->SetOwner(1);
5164a766 988
924fb6e9 989 Int_t logLevel = AliLog::GetGlobalLogLevel();
990 AliLog::SetGlobalLogLevel(AliLog::kError);
991
2bb7b766 992 UInt_t result = AliCDBManager::Instance()->GetStorage(fgkLocalCDB)->Put(fStatusEntry);
7bfb2090 993
924fb6e9 994 AliLog::SetGlobalLogLevel((AliLog::EType_t)logLevel);
995
996 if (!result) {
3301427a 997 Log("SHUTTLE", Form("WriteShuttleStatus - Failed for %s, run %d",
998 fCurrentDetector.Data(), run));
2bb7b766 999 return kFALSE;
1000 }
e7f62f16 1001
8e828d39 1002 SendMLDetInfo();
7bfb2090 1003
2bb7b766 1004 return kTRUE;
5164a766 1005}
1006
1007//______________________________________________________________________________________________
1008void AliShuttle::UpdateShuttleStatus(AliShuttleStatus::Status newStatus, Bool_t increaseCount)
1009{
9827400b 1010 //
1011 // changes the AliShuttleStatus for the given detector and run to the given status
1012 //
5164a766 1013
2bb7b766 1014 if (!fStatusEntry){
1015 AliError("UNEXPECTED: fStatusEntry empty");
1016 return;
1017 }
5164a766 1018
2bb7b766 1019 AliShuttleStatus* status = dynamic_cast<AliShuttleStatus*> (fStatusEntry->GetObject());
5164a766 1020
2bb7b766 1021 if (!status){
c88ad5db 1022 Log("SHUTTLE", "UpdateShuttleStatus - UNEXPECTED: status could not be read from current CDB entry");
2bb7b766 1023 return;
1024 }
5164a766 1025
2c15234c 1026 TString actionStr = Form("UpdateShuttleStatus - %s: Changing state from %s to %s",
eba76848 1027 fCurrentDetector.Data(),
36c99a6a 1028 status->GetStatusName(),
eba76848 1029 status->GetStatusName(newStatus));
cb343cfd 1030 Log("SHUTTLE", actionStr);
1031 SetLastAction(actionStr);
5164a766 1032
2bb7b766 1033 status->SetStatus(newStatus);
1034 if (increaseCount) status->IncreaseCount();
5164a766 1035
924fb6e9 1036 Int_t logLevel = AliLog::GetGlobalLogLevel();
1037 AliLog::SetGlobalLogLevel(AliLog::kError);
1038
2bb7b766 1039 AliCDBManager::Instance()->GetStorage(fgkLocalCDB)->Put(fStatusEntry);
e7f62f16 1040
924fb6e9 1041 AliLog::SetGlobalLogLevel((AliLog::EType_t)logLevel);
1042
8e828d39 1043 SendMLDetInfo();
5164a766 1044}
e7f62f16 1045
1046//______________________________________________________________________________________________
8e828d39 1047void AliShuttle::SendMLDetInfo()
e7f62f16 1048{
1049 //
1050 // sends ML information about the current status of the current detector being processed
1051 //
1052
1053 AliShuttleStatus* status = dynamic_cast<AliShuttleStatus*> (fStatusEntry->GetObject());
1054
1055 if (!status){
8e828d39 1056 Log("SHUTTLE", "SendMLDetInfo - UNEXPECTED: status could not be read from current CDB entry");
e7f62f16 1057 return;
1058 }
1059
1060 TMonaLisaText mlStatus(Form("%s_status", fCurrentDetector.Data()), status->GetStatusName());
1061 TMonaLisaValue mlRetryCount(Form("%s_count", fCurrentDetector.Data()), status->GetCount());
1062
1063 TList mlList;
1064 mlList.Add(&mlStatus);
1065 mlList.Add(&mlRetryCount);
1066
ee6f7523 1067 TString mlID;
1068 mlID.Form("%d", GetCurrentRun());
1069 fMonaLisa->SendParameters(&mlList, mlID);
e7f62f16 1070}
1071
5164a766 1072//______________________________________________________________________________________________
1073Bool_t AliShuttle::ContinueProcessing()
1074{
9827400b 1075 // this function reads the AliShuttleStatus information from CDB and
1076 // checks if the processing should be continued
1077 // if yes it returns kTRUE and updates the AliShuttleStatus with nextStatus
2bb7b766 1078
926cbc0e 1079 if (!fConfig->HostProcessDetector(fCurrentDetector))
1080 return kFALSE;
57c1a579 1081
1082 AliPreprocessor* aPreprocessor =
1083 dynamic_cast<AliPreprocessor*> (fPreprocessorMap.GetValue(fCurrentDetector));
1084 if (!aPreprocessor)
1085 {
c88ad5db 1086 Log("SHUTTLE", Form("ContinueProcessing - %s: no preprocessor registered", fCurrentDetector.Data()));
57c1a579 1087 return kFALSE;
1088 }
1089
2bb7b766 1090 AliShuttleLogbookEntry::Status entryStatus =
eba76848 1091 fLogbookEntry->GetDetectorStatus(fCurrentDetector);
2bb7b766 1092
5e993b6f 1093 if (entryStatus != AliShuttleLogbookEntry::kUnprocessed) {
c88ad5db 1094 Log("SHUTTLE", Form("ContinueProcessing - %s is %s",
2bb7b766 1095 fCurrentDetector.Data(),
1096 fLogbookEntry->GetDetectorStatusName(entryStatus)));
1097 return kFALSE;
1098 }
1099
1100 // if we get here, according to Shuttle logbook subdetector is in UNPROCESSED state
be48e3ea 1101
1102 // check if current run is first unprocessed run for current detector
1103 if (fConfig->StrictRunOrder(fCurrentDetector) &&
1104 !fFirstUnprocessed[GetDetPos(fCurrentDetector)])
1105 {
86aa42c3 1106 if (fTestMode == kNone)
1107 {
c88ad5db 1108 Log("SHUTTLE", Form("ContinueProcessing - %s requires strict run ordering"
1109 " but this is not the first unprocessed run!"));
86aa42c3 1110 return kFALSE;
1111 }
1112 else
1113 {
c88ad5db 1114 Log("SHUTTLE", Form("ContinueProcessing - In TESTMODE - "
1115 "Although %s requires strict run ordering "
1116 "and this is not the first unprocessed run, "
1117 "the SHUTTLE continues"));
86aa42c3 1118 }
be48e3ea 1119 }
1120
926cbc0e 1121 // Is the subdetector processed first time for this run?
1122 fFirstProcessing = kFALSE;
1123
2bb7b766 1124 AliShuttleStatus* status = ReadShuttleStatus();
1125 if (!status) {
1126 // first time
1127 Log("SHUTTLE", Form("ContinueProcessing - %s: Processing first time",
1128 fCurrentDetector.Data()));
1129 status = new AliShuttleStatus(AliShuttleStatus::kStarted);
926cbc0e 1130 fFirstProcessing = kTRUE;
2bb7b766 1131 return WriteShuttleStatus(status);
1132 }
1133
ffa25f20 1134 // The following case shouldn't happen if Shuttle Logbook was correctly updated.
2bb7b766 1135 // If it happens it may mean Logbook updating failed... let's do it now!
1136 if (status->GetStatus() == AliShuttleStatus::kDone ||
ffa25f20 1137 status->GetStatus() == AliShuttleStatus::kFailed ||
1138 status->GetStatus() == AliShuttleStatus::kSkipped) {
2bb7b766 1139 Log("SHUTTLE", Form("ContinueProcessing - %s is already %s. Updating Shuttle Logbook",
1140 fCurrentDetector.Data(),
1141 status->GetStatusName(status->GetStatus())));
ffa25f20 1142
1143 if (status->GetStatus() == AliShuttleStatus::kSkipped)
1144 {
1145 UpdateShuttleLogbook(fCurrentDetector.Data(), "DONE");
1146 }
1147 else
1148 UpdateShuttleLogbook(fCurrentDetector.Data(), status->GetStatusName(status->GetStatus()));
1149
2bb7b766 1150 return kFALSE;
1151 }
1152
5e993b6f 1153 if (status->GetStatus() == AliShuttleStatus::kStoreStarted || status->GetStatus() == AliShuttleStatus::kStoreDelayed ||status->GetStatus() == AliShuttleStatus::kStoreError) {
2bb7b766 1154 Log("SHUTTLE",
c88ad5db 1155 Form("ContinueProcessing - %s: Grid storage of one or more "
1156 "objects failed. Trying again now",
2bb7b766 1157 fCurrentDetector.Data()));
5e993b6f 1158 StoreOCDB();
2bb7b766 1159 return kFALSE;
1160 }
1161
1162 // if we get here, there is a restart
57c1a579 1163 Bool_t cont = kFALSE;
2bb7b766 1164
1165 // abort conditions
cb343cfd 1166 if (status->GetCount() >= fConfig->GetMaxRetries()) {
57c1a579 1167 Log("SHUTTLE", Form("ContinueProcessing - %s failed %d times in status %s - "
1168 "Updating Shuttle Logbook", fCurrentDetector.Data(),
2bb7b766 1169 status->GetCount(), status->GetStatusName()));
1170 UpdateShuttleLogbook(fCurrentDetector.Data(), "FAILED");
e7f62f16 1171 UpdateShuttleStatus(AliShuttleStatus::kFailed);
3301427a 1172
1173 // there may still be objects in local OCDB and reference storage
1174 // and FXS databases may be not updated: do it now!
9827400b 1175
1176 // TODO Currently disabled, we want to keep files in case of failure!
1177 // CleanLocalStorage(fgkLocalCDB);
1178 // CleanLocalStorage(fgkLocalRefStorage);
1179 // UpdateTableFailCase();
1180
1181 // Send mail to detector expert!
c88ad5db 1182 Log("SHUTTLE", Form("ContinueProcessing - Sending mail to %s expert...",
926cbc0e 1183 fCurrentDetector.Data()));
fe6a3257 1184 // det experts in to
1185 TString to="";
1186 TIter *iterExperts = 0;
1187 iterExperts = new TIter(fConfig->GetResponsibles(fCurrentDetector));
1188 TObjString *anExpert=0;
1189 while ((anExpert = (TObjString*) iterExperts->Next()))
1190 {
1191 to += Form("%s, \n", anExpert->GetName());
1192 }
1193 delete iterExperts;
1194
1195 if (to.Length() > 0)
1196 to.Remove(to.Length()-3);
1197 AliDebug(2, Form("to: %s",to.Data()));
1198
1199 if (to.IsNull()) {
1200 Log("SHUTTLE", Form("List of %s responsibles not set!", fCurrentDetector.Data()));
1201 return kFALSE;
1202 }
1203
1204 Log(fCurrentDetector.Data(), Form("ContinueProcessing - Sending mail to %s expert(s):",
1205 fCurrentDetector.Data()));
1206 Log(fCurrentDetector.Data(), Form("\n%s", to.Data()));
339adafa 1207 if (!SendMail(kPPEMail))
9827400b 1208 Log("SHUTTLE", Form("ContinueProcessing - Could not send mail to %s expert",
926cbc0e 1209 fCurrentDetector.Data()));
3301427a 1210
57c1a579 1211 } else {
1212 Log("SHUTTLE", Form("ContinueProcessing - %s: restarting. "
1213 "Aborted before with %s. Retry number %d.", fCurrentDetector.Data(),
1214 status->GetStatusName(), status->GetCount()));
9827400b 1215 Bool_t increaseCount = kTRUE;
c88ad5db 1216 if (status->GetStatus() == AliShuttleStatus::kDCSError ||
8da81210 1217 status->GetStatus() == AliShuttleStatus::kDCSStarted ||
1218 status->GetStatus() == AliShuttleStatus::kFXSError)
c88ad5db 1219 increaseCount = kFALSE;
675f64cd 1220
9827400b 1221 UpdateShuttleStatus(AliShuttleStatus::kStarted, increaseCount);
57c1a579 1222 cont = kTRUE;
2bb7b766 1223 }
1224
57c1a579 1225 return cont;
5164a766 1226}
1227
8e828d39 1228//______________________________________________________________________________________________
1229void AliShuttle::SendMLRunInfo(const char* status)
1230{
1231 //
1232 // Send information about this run to ML
1233
1234 TMonaLisaText mlStatus("SHUTTLE_status", status);
1235 TString runType(fLogbookEntry->GetRunType());
1236 if (strlen(fLogbookEntry->GetRunParameter("log")) > 0){
1237
1238 runType += "(";
1239 runType += fLogbookEntry->GetRunParameter("log");
1240 runType += ")";
1241 }
728290a4 1242 if (fLogbookEntry->GetDATestMode()){
1243 runType += " (DATest)";
1244 }
8e828d39 1245 TMonaLisaText mlRunType("SHUTTLE_runtype", runType);
1246
1247 TList mlList;
1248 mlList.Add(&mlStatus);
1249 mlList.Add(&mlRunType);
1250
1251 TString mlID;
1252 mlID.Form("%d", GetCurrentRun());
1253 fMonaLisa->SendParameters(&mlList, mlID);
1254}
1255
ef83773f 1256//______________________________________________________________________________________________
1257Int_t AliShuttle::GetMem(Int_t pid)
1258{
1259 // invokes ps to get the memory consumption of the process <pid>
1260 // returns -1 in case of error
1261
1262 TString checkStr;
1263 checkStr.Form("ps -o vsize --pid %d | tail -n 1", pid);
1264 FILE* pipe = gSystem->OpenPipe(checkStr, "r");
1265 if (!pipe)
1266 {
1267 Log("SHUTTLE", Form("Process - Error: "
1268 "Could not open pipe to %s", checkStr.Data()));
1269 return -1;
1270 }
1271
1272 char buffer[100];
1273 if (!fgets(buffer, 100, pipe))
1274 {
1275 Log("SHUTTLE", "Process - Error: ps did not return anything");
1276 gSystem->ClosePipe(pipe);
1277 return -1;
1278 }
1279 gSystem->ClosePipe(pipe);
1280
1281 //Log("SHUTTLE", Form("ps returned %s", buffer));
1282
1283 Int_t mem = 0;
1284 if ((sscanf(buffer, "%d\n", &mem) != 1) || !mem)
1285 {
1286 Log("SHUTTLE", "Process - Error: Could not parse output of ps");
1287 return -1;
1288 }
1289
1290 return mem;
1291}
1292
5164a766 1293//______________________________________________________________________________________________
2bb7b766 1294Bool_t AliShuttle::Process(AliShuttleLogbookEntry* entry)
58bc3020 1295{
73abe331 1296 //
b948db8d 1297 // Makes data retrieval for all detectors in the configuration.
2bb7b766 1298 // entry: Shuttle logbook entry, contains run paramenters and status of detectors
1299 // (Unprocessed, Inactive, Failed or Done).
d477ad88 1300 // Returns kFALSE in case of error occured and kTRUE otherwise
73abe331 1301 //
1302
9827400b 1303 if (!entry) return kFALSE;
2bb7b766 1304
1305 fLogbookEntry = entry;
1306
c88ad5db 1307 Log("SHUTTLE", Form("\t\t\t^*^*^*^*^*^*^*^*^*^*^*^* run %d: START ^*^*^*^*^*^*^*^*^*^*^*^*",
9827400b 1308 GetCurrentRun()));
2bb7b766 1309
339adafa 1310 CountOpenRuns();
1311
8e828d39 1312 // Send the information to ML
1313 SendMLRunInfo("Processing");
3301427a 1314
9827400b 1315 if (fLogbookEntry->IsDone())
1316 {
1317 Log("SHUTTLE","Process - Shuttle is already DONE. Updating logbook");
1318 UpdateShuttleLogbook("shuttle_done");
1319 fLogbookEntry = 0;
1320 return kTRUE;
1321 }
1322
1323 // read test mode if flag is set
1324 if (fReadTestMode)
1325 {
3d8bc902 1326 fTestMode = kNone;
9827400b 1327 TString logEntry(entry->GetRunParameter("log"));
1328 //printf("log entry = %s\n", logEntry.Data());
1329 TString searchStr("Testmode: ");
1330 Int_t pos = logEntry.Index(searchStr.Data());
1331 //printf("%d\n", pos);
1332 if (pos >= 0)
1333 {
1334 TSubString subStr = logEntry(pos + searchStr.Length(), logEntry.Length());
1335 //printf("%s\n", subStr.String().Data());
1336 TString newStr(subStr.Data());
1337 TObjArray* token = newStr.Tokenize(' ');
1338 if (token)
1339 {
1340 //token->Print();
1341 TObjString* tmpStr = dynamic_cast<TObjString*> (token->First());
1342 if (tmpStr)
1343 {
1344 Int_t testMode = tmpStr->String().Atoi();
1345 if (testMode > 0)
1346 {
c88ad5db 1347 Log("SHUTTLE", Form("Process - Enabling test mode %d", testMode));
9827400b 1348 SetTestMode((TestMode) testMode);
1349 }
1350 }
1351 delete token;
1352 }
1353 }
1354 }
c88ad5db 1355
eba76848 1356 fLogbookEntry->Print("all");
57f50b3c 1357
1358 // Initialization
d477ad88 1359 Bool_t hasError = kFALSE;
5164a766 1360
728290a4 1361 // Set the CDB and Reference folders according to the year
1362
1363 // build cdb paths (repeat each time, run might be a DATest run)
1364 if (!fLogbookEntry->GetDATestMode()){
1365 fgkMainCDB.Form("alien://folder=%s%d/OCDB?user=alidaq?cacheFold=/tmp/OCDBCache",
1366 fConfig->GetAlienPath(), GetCurrentYear());
1367
1368 fgkMainRefStorage.Form("alien://folder=%s%d/Reference?user=alidaq?cacheFold=/tmp/OCDBCache",
1369 fConfig->GetAlienPath(), GetCurrentYear());
1370 }
1371 else {
1372 fgkMainCDB.Form("alien://folder=%s%d/DATest/OCDB?user=alidaq?cacheFold=/tmp/OCDBCache",
1373 fConfig->GetAlienPath(), GetCurrentYear());
1374
1375 fgkMainRefStorage.Form("alien://folder=%s%d/DATest/Reference?user=alidaq?cacheFold=/tmp/OCDBCache",
1376 fConfig->GetAlienPath(), GetCurrentYear());
1377 }
1378
1379 AliDebug(2,Form("Main CDB storage = %s",fgkMainCDB.Data()));
1380 AliDebug(2,Form("Main Reference storage = %s",fgkMainRefStorage.Data()));
1381
57f50b3c 1382 // Loop on detectors in the configuration
b948db8d 1383 TIter iter(fConfig->GetDetectors());
2bb7b766 1384 TObjString* aDetector = 0;
b948db8d 1385
4508ab8b 1386 Bool_t first = kTRUE;
1387
be48e3ea 1388 while ((aDetector = (TObjString*) iter.Next()))
1389 {
7bfb2090 1390 fCurrentDetector = aDetector->String();
5164a766 1391
5e993b6f 1392 if (ContinueProcessing() == kFALSE)
1393 continue;
4508ab8b 1394
1395 if (first)
1396 {
1397 // only read QueryCDB when needed and only once
1398 AliCDBStorage *mainCDBSto = AliCDBManager::Instance()->GetStorage(fgkMainCDB);
1399 if(mainCDBSto) mainCDBSto->QueryCDB(GetCurrentRun());
1400 AliCDBStorage *mainRefSto = AliCDBManager::Instance()->GetStorage(fgkMainRefStorage);
1401 if(mainRefSto) mainRefSto->QueryCDB(GetCurrentRun());
1402 first = kFALSE;
1403 }
9e080f92 1404
c88ad5db 1405 Log("SHUTTLE", Form("\t\t\t****** run %d - %s: START ******",
2bb7b766 1406 GetCurrentRun(), aDetector->GetName()));
1407
9d733021 1408 for(Int_t iSys=0;iSys<3;iSys++) fFXSCalled[iSys]=kFALSE;
ef83773f 1409
1410 Int_t initialMem = GetMem(getpid());
1411 Log("SHUTTLE", Form("Memory consumption before forking is %d", initialMem));
9d733021 1412
c88ad5db 1413 Log(fCurrentDetector.Data(), "Process - Starting processing");
85a80aa9 1414
be48e3ea 1415 Int_t pid = fork();
1416
1417 if (pid < 0)
1418 {
c88ad5db 1419 Log("SHUTTLE", "Process - ERROR: Forking failed");
be48e3ea 1420 }
1421 else if (pid > 0)
1422 {
1423 // parent
c88ad5db 1424 Log("SHUTTLE", Form("Process - In parent process of %d - %s: Starting monitoring",
be48e3ea 1425 GetCurrentRun(), aDetector->GetName()));
1426
1427 Long_t begin = time(0);
1428
1429 int status; // to be used with waitpid, on purpose an int (not Int_t)!
1430 while (waitpid(pid, &status, WNOHANG) == 0)
1431 {
1432 Long_t expiredTime = time(0) - begin;
1433
1434 if (expiredTime > fConfig->GetPPTimeOut())
1435 {
56d3e9e4 1436 TString logMsg;
1437 AliShuttleStatus *currentStatus = ReadShuttleStatus();
1438 AliShuttleStatus::Status newStatus = AliShuttleStatus::kInvalid;
1439
838d9827 1440 if (currentStatus->GetStatus() == AliShuttleStatus::kDCSStarted)
1441 {
1442 // in case the pp goes in TimeOut while retrieving the DCS DPs
1443 // set status to kDCSError
1444
1445 logMsg.Form("Process - Process of %s timed out while retrieving the DCS DataPoints. Run time: %d seconds. Killing... and setting status to DCSError.",
1446 fCurrentDetector.Data(), expiredTime);
1447 newStatus = AliShuttleStatus::kDCSError;
1448 }
1449 else if (currentStatus->GetStatus() <= AliShuttleStatus::kPPDone)
56d3e9e4 1450 {
1451 // in case pp not yet done set status to kPPTimeOut
1452
1453 logMsg.Form("Process - Process of %s timed out. Run time: %d seconds. Killing...",
1454 fCurrentDetector.Data(), expiredTime);
1455 newStatus = AliShuttleStatus::kPPTimeOut;
1456 }
1457 else if (currentStatus->GetStatus() == AliShuttleStatus::kStoreStarted)
1458 {
1459 // in case the pp goes in TimeOut while storing the objects in the OCDB
1460 // set status to kStoreError
1461
1462 logMsg.Form("Process - Process of %s timed out while storing the OCDB object. Run time: %d seconds. Killing... and setting status to StoreError.",
1463 fCurrentDetector.Data(), expiredTime);
1464 newStatus = AliShuttleStatus::kStoreError;
1465 }
1466 else
1467 {
1468 // in other cases don't change the status
1469
1470 logMsg.Form("Process - Process of %s timed out in status = %s. Run time: %d seconds. Killing... without changing the status",
1471 fCurrentDetector.Data(), currentStatus->GetStatusName(), expiredTime);
1472 }
1473
1474 Log("SHUTTLE", logMsg);
1475 Log(fCurrentDetector, logMsg);
be48e3ea 1476
1477 kill(pid, 9);
1478
56d3e9e4 1479 if (newStatus != AliShuttleStatus::kInvalid)
1480 UpdateShuttleStatus(newStatus);
be48e3ea 1481 hasError = kTRUE;
1482
1483 gSystem->Sleep(1000);
1484 }
1485 else
1486 {
be48e3ea 1487 gSystem->Sleep(1000);
9827400b 1488
ef83773f 1489 Int_t mem = GetMem(pid);
1490
1491 if (mem < 0)
9827400b 1492 continue;
9827400b 1493
ef83773f 1494 mem -= initialMem;
1495 if (mem < 0)
1496 mem = 0;
9827400b 1497
1498 if (expiredTime % 60 == 0)
ee6f7523 1499 {
c88ad5db 1500 Log("SHUTTLE", Form("Process - %s: Checking process. "
1501 "Run time: %d seconds - Memory consumption: %d KB",
1502 fCurrentDetector.Data(), expiredTime, mem));
ee6f7523 1503 SendAlive();
1504 }
9827400b 1505
1506 if (mem > fConfig->GetPPMaxMem())
1507 {
1508 TString tmp;
c88ad5db 1509 tmp.Form("Process - Process exceeds maximum allowed memory "
1510 "(%d KB > %d KB). Killing...",
9827400b 1511 mem, fConfig->GetPPMaxMem());
1512 Log("SHUTTLE", tmp);
1513 Log(fCurrentDetector, tmp);
1514
1515 kill(pid, 9);
1516
1517 UpdateShuttleStatus(AliShuttleStatus::kPPOutOfMemory);
1518 hasError = kTRUE;
1519
1520 gSystem->Sleep(1000);
1521 }
be48e3ea 1522 }
1523 }
1524
c88ad5db 1525 Log("SHUTTLE", Form("Process - In parent process of %d - %s: Client has terminated.",
be48e3ea 1526 GetCurrentRun(), aDetector->GetName()));
1527
1528 if (WIFEXITED(status))
1529 {
1530 Int_t returnCode = WEXITSTATUS(status);
1531
c88ad5db 1532 Log("SHUTTLE", Form("Process - %s: the return code is %d", fCurrentDetector.Data(),
3301427a 1533 returnCode));
be48e3ea 1534
9827400b 1535 if (returnCode == 0) hasError = kTRUE;
be48e3ea 1536 }
1537 }
1538 else if (pid == 0)
1539 {
26b1866e 1540 // child
1541 Log("SHUTTLE", Form("Process - In child process of %d - %s", GetCurrentRun(),
c88ad5db 1542 aDetector->GetName()));
be48e3ea 1543
c88ad5db 1544 Log("SHUTTLE", Form("Process - Redirecting output to %s log",fCurrentDetector.Data()));
ffa29e93 1545
546242fb 1546 if ((freopen(GetLogFileName(fCurrentDetector), "a", stdout)) == 0)
ffa29e93 1547 {
c88ad5db 1548 Log("SHUTTLE", "Process - Could not freopen stdout");
ffa29e93 1549 }
1550 else
1551 {
1552 fOutputRedirected = kTRUE;
1553 if ((dup2(fileno(stdout), fileno(stderr))) < 0)
c88ad5db 1554 Log("SHUTTLE", "Process - Could not redirect stderr");
ffa29e93 1555
1556 }
1557
5bac2bde 1558 TString wd = gSystem->WorkingDirectory();
3ebbf3c4 1559 Int_t dir_lev1 = GetCurrentRun()/10000;
1560 TString tmpDir = Form("%s/%d/%d/%s_process", GetShuttleTempDir(),
1561 dir_lev1, GetCurrentRun(), fCurrentDetector.Data());
5bac2bde 1562
d524ade6 1563 Int_t result = gSystem->GetPathInfo(tmpDir.Data(), 0, (Long64_t*) 0, 0, 0);
1564 if (!result) // temp dir already exists!
1565 {
1566 Log(fCurrentDetector.Data(),
1567 Form("Process - %s dir already exists! Removing...", tmpDir.Data()));
1568 gSystem->Exec(Form("rm -rf %s",tmpDir.Data()));
675f64cd 1569 }
1570
1571 if (gSystem->mkdir(tmpDir.Data(), 1))
1572 {
1573 Log(fCurrentDetector.Data(), "Process - could not make temp directory!!");
1574 gSystem->Exit(1);
d524ade6 1575 }
1576
1577 if (!gSystem->ChangeDirectory(tmpDir.Data()))
1578 {
1579 Log(fCurrentDetector.Data(), "Process - could not change directory!!");
1580 gSystem->Exit(1);
1581 }
5bac2bde 1582
ffa25f20 1583 Int_t success = ProcessCurrentDetector();
26b1866e 1584
5bac2bde 1585 gSystem->ChangeDirectory(wd.Data());
b1d18693 1586
ffa25f20 1587 if (success == 1) // Preprocessor finished successfully!
9827400b 1588 {
fb40ead4 1589 // remove temporary folder or DCS map
1590 if (!fConfig->KeepTempFolder())
1591 {
1592 gSystem->Exec(Form("rm -rf %s",tmpDir.Data()));
1593 } else if (!fConfig->KeepDCSMap())
1594 {
1595 gSystem->Exec(Form("rm -f %s/DCSMap.root",tmpDir.Data()));
1596 }
b1d18693 1597
3301427a 1598 // Update time_processed field in FXS DB
1599 if (UpdateTable() == kFALSE)
5bac2bde 1600 Log("SHUTTLE", Form("Process - %s: Could not update FXS databases!",
1601 fCurrentDetector.Data()));
3301427a 1602
1603 // Transfer the data from local storage to main storage (Grid)
3301427a 1604 if (StoreOCDB() == kFALSE)
9827400b 1605 success = kFALSE;
5e993b6f 1606 }
ffa25f20 1607 else if (success == 0)
c88ad5db 1608 {
1609 Log("SHUTTLE",
1610 Form("\t\t\t****** run %d - %s: PP ERROR ******",
1611 GetCurrentRun(), aDetector->GetName()));
be48e3ea 1612 }
1613
4b95672b 1614 for (UInt_t iSys=0; iSys<3; iSys++)
1615 {
1616 if (fFXSCalled[iSys]) fFXSlist[iSys].Clear();
1617 }
1618
c88ad5db 1619 Log("SHUTTLE", Form("Process - Client process of %d - %s is exiting now with %d.",
9827400b 1620 GetCurrentRun(), aDetector->GetName(), success));
be48e3ea 1621
1622 // the client exits here
9827400b 1623 gSystem->Exit(success);
be48e3ea 1624
1625 AliError("We should never get here!!!");
1626 }
7bfb2090 1627 }
5164a766 1628
c88ad5db 1629 Log("SHUTTLE", Form("\t\t\t^*^*^*^*^*^*^*^*^*^*^*^* run %d: FINISH ^*^*^*^*^*^*^*^*^*^*^*^*",
2bb7b766 1630 GetCurrentRun()));
1631
1632 //check if shuttle is done for this run, if so update logbook
1633 TObjArray checkEntryArray;
1634 checkEntryArray.SetOwner(1);
9e080f92 1635 TString whereClause = Form("where run=%d", GetCurrentRun());
b0e53b15 1636 if (!QueryShuttleLogbook(whereClause.Data(), checkEntryArray) ||
1637 checkEntryArray.GetEntries() == 0) {
9e080f92 1638 Log("SHUTTLE", Form("Process - Warning: Cannot check status of run %d on Shuttle logbook!",
1639 GetCurrentRun()));
1640 return hasError == kFALSE;
1641 }
b948db8d 1642
9e080f92 1643 AliShuttleLogbookEntry* checkEntry = dynamic_cast<AliShuttleLogbookEntry*>
1644 (checkEntryArray.At(0));
2bb7b766 1645
9e080f92 1646 if (checkEntry)
1647 {
1648 if (checkEntry->IsDone())
be48e3ea 1649 {
9e080f92 1650 Log("SHUTTLE","Process - Shuttle is DONE. Updating logbook");
1651 UpdateShuttleLogbook("shuttle_done");
1652 }
1653 else
1654 {
1655 for (UInt_t iDet=0; iDet<NDetectors(); iDet++)
be48e3ea 1656 {
9e080f92 1657 if (checkEntry->GetDetectorStatus(iDet) == AliShuttleLogbookEntry::kUnprocessed)
be48e3ea 1658 {
9e080f92 1659 AliDebug(2, Form("Run %d: setting %s as \"not first time unprocessed\"",
1660 checkEntry->GetRun(), GetDetName(iDet)));
1661 fFirstUnprocessed[iDet] = kFALSE;
be48e3ea 1662 }
1663 }
8e828d39 1664 SendMLRunInfo("Pending");
2bb7b766 1665 }
1666 }
1667
1668 fLogbookEntry = 0;
85a80aa9 1669
a7160fe9 1670 return hasError == kFALSE;
73abe331 1671}
1672
b948db8d 1673//______________________________________________________________________________________________
ffa25f20 1674Int_t AliShuttle::ProcessCurrentDetector()
73abe331 1675{
1676 //
2bb7b766 1677 // Makes data retrieval just for a specific detector (fCurrentDetector).
73abe331 1678 // Threre should be a configuration for this detector.
73abe331 1679
1d172743 1680 Log("SHUTTLE", Form("ProcessCurrentDetector - Retrieving values for %s, run %d",
1681 fCurrentDetector.Data(), GetCurrentRun()));
73abe331 1682
d524ade6 1683 TString wd = gSystem->WorkingDirectory();
1684
2d9019b4 1685 if (!CleanReferenceStorage(fCurrentDetector.Data()))
ffa25f20 1686 return 0;
d524ade6 1687
1688 gSystem->ChangeDirectory(wd.Data());
1689
3301427a 1690 // call preprocessor
1691 AliPreprocessor* aPreprocessor =
1692 dynamic_cast<AliPreprocessor*> (fPreprocessorMap.GetValue(fCurrentDetector));
1693
8579c6fb 1694 // check if the preprocessor wants to process this run type
1695 if (aPreprocessor->ProcessRunType() == kFALSE)
ffa25f20 1696 {
1697 UpdateShuttleStatus(AliShuttleStatus::kSkipped);
1698 UpdateShuttleLogbook(fCurrentDetector, "DONE");
bd23d48b 1699 if (!UpdateTableSkippedCase(fCurrentDetector.Data()))
1700 {
1701 AliError(Form("Could not update FXS tables for run %d !", GetCurrentRun()));
1702 }
ffa25f20 1703 Log(fCurrentDetector, Form("ProcessCurrentDetector - %s preprocessor is not interested in this run type", fCurrentDetector.Data()));
1704
1705 return 2;
1706 }
1707
1708 TMap* dcsMap = new TMap();
1709
3301427a 1710 aPreprocessor->Initialize(GetCurrentRun(), GetCurrentStartTime(), GetCurrentEndTime());
1711
1712 Bool_t processDCS = aPreprocessor->ProcessDCS();
d477ad88 1713
651fdaab 1714 if (!processDCS)
1715 {
1d172743 1716 Log(fCurrentDetector, "ProcessCurrentDetector -"
1717 " The preprocessor requested to skip the retrieval of DCS values");
651fdaab 1718 }
8b739301 1719 else if (fTestMode & kSkipDCS)
2c15234c 1720 {
1d172743 1721 Log(fCurrentDetector, "ProcessCurrentDetector - In TESTMODE: Skipping DCS processing");
9827400b 1722 }
1723 else if (fTestMode & kErrorDCS)
1724 {
1d172743 1725 Log(fCurrentDetector, "ProcessCurrentDetector - In TESTMODE: Simulating DCS error");
3d8bc902 1726 UpdateShuttleStatus(AliShuttleStatus::kDCSStarted);
9827400b 1727 UpdateShuttleStatus(AliShuttleStatus::kDCSError);
1d172743 1728 delete dcsMap;
ffa25f20 1729 return 0;
2c15234c 1730 } else {
3301427a 1731
1732 UpdateShuttleStatus(AliShuttleStatus::kDCSStarted);
1733
1d172743 1734 // Query DCS archive
1735 Int_t nServers = fConfig->GetNServers(fCurrentDetector);
a038aa70 1736
1d172743 1737 for (int iServ=0; iServ<nServers; iServ++)
2c15234c 1738 {
1d172743 1739
1740 TString host(fConfig->GetDCSHost(fCurrentDetector, iServ));
1741 Int_t port = fConfig->GetDCSPort(fCurrentDetector, iServ);
542b6cc8 1742 Int_t multiSplit = fConfig->GetMultiSplit(fCurrentDetector, iServ);
1743
1790d4b7 1744 Log(fCurrentDetector, Form("ProcessCurrentDetector -"
1745 " Querying DCS Amanda server %s:%d (%d of %d)",
1746 host.Data(), port, iServ+1, nServers));
1d172743 1747
1748 TMap* aliasMap = 0;
1749 TMap* dpMap = 0;
26b1866e 1750
1d172743 1751 if (fConfig->GetDCSAliases(fCurrentDetector, iServ)->GetEntries() > 0)
2c15234c 1752 {
5ed060b1 1753 Log(fCurrentDetector, Form("Querying %d DCS aliases", fConfig->GetDCSAliases(fCurrentDetector, iServ)->GetEntries()));
1d172743 1754 aliasMap = GetValueSet(host, port,
542b6cc8 1755 fConfig->GetDCSAliases(fCurrentDetector, iServ),
1756 kAlias, multiSplit);
1d172743 1757 if (!aliasMap)
1758 {
1759 Log(fCurrentDetector,
1760 Form("ProcessCurrentDetector -"
675f64cd 1761 " Error retrieving DCS aliases from server %s."
1762 " Sending mail to DCS experts!", host.Data()));
1d172743 1763 UpdateShuttleStatus(AliShuttleStatus::kDCSError);
675f64cd 1764
339adafa 1765 if (!SendMail(kDCSEMail))
fb40ead4 1766 Log("SHUTTLE", Form("ProcessCurrentDetector - "
95d4aaef 1767 "Could not send mail to DCS experts!"));
675f64cd 1768
a038aa70 1769 delete dcsMap;
ffa25f20 1770 return 0;
1d172743 1771 }
2c15234c 1772 }
a038aa70 1773
1d172743 1774 if (fConfig->GetDCSDataPoints(fCurrentDetector, iServ)->GetEntries() > 0)
a038aa70 1775 {
5ed060b1 1776 Log(fCurrentDetector, Form("Querying %d DCS data points", fConfig->GetDCSDataPoints(fCurrentDetector, iServ)->GetEntries()));
1d172743 1777 dpMap = GetValueSet(host, port,
542b6cc8 1778 fConfig->GetDCSDataPoints(fCurrentDetector, iServ),
1779 kDP, multiSplit);
1d172743 1780 if (!dpMap)
1781 {
1782 Log(fCurrentDetector,
1783 Form("ProcessCurrentDetector -"
675f64cd 1784 " Error retrieving DCS data points from server %s."
1785 " Sending mail to DCS experts!", host.Data()));
1d172743 1786 UpdateShuttleStatus(AliShuttleStatus::kDCSError);
675f64cd 1787
339adafa 1788 if (!SendMail(kDCSEMail))
fb40ead4 1789 Log("SHUTTLE", Form("ProcessCurrentDetector - "
95d4aaef 1790 "Could not send mail to DCS experts!"));
675f64cd 1791
1d172743 1792 if (aliasMap) delete aliasMap;
1793 delete dcsMap;
ffa25f20 1794 return 0;
1d172743 1795 }
a038aa70 1796 }
1d172743 1797
1798 // merge aliasMap and dpMap into dcsMap
1799 if(aliasMap) {
1800 TIter iter(aliasMap);
a038aa70 1801 TObjString* key = 0;
1802 while ((key = (TObjString*) iter.Next()))
1d172743 1803 dcsMap->Add(key, aliasMap->GetValue(key->String()));
1804
1805 aliasMap->SetOwner(kFALSE);
1806 delete aliasMap;
1807 }
1808
1809 if(dpMap) {
1810 TIter iter(dpMap);
1811 TObjString* key = 0;
1812 while ((key = (TObjString*) iter.Next()))
1813 dcsMap->Add(key, dpMap->GetValue(key->String()));
1814
1815 dpMap->SetOwner(kFALSE);
1816 delete dpMap;
a038aa70 1817 }
73abe331 1818 }
1819 }
dc25836b 1820
b1d18693 1821 // save map into file, to help debugging in case of preprocessor error
fbc112e3 1822 TFile* f = TFile::Open("DCSMap.root","recreate");
b1d18693 1823 f->cd();
1824 dcsMap->Write("DCSMap", TObject::kSingleKey);
1825 f->Close();
fbc112e3 1826 delete f;
b1d18693 1827
2bb7b766 1828 // DCS Archive DB processing successful. Call Preprocessor!
85a80aa9 1829 UpdateShuttleStatus(AliShuttleStatus::kPPStarted);
a7160fe9 1830
339adafa 1831 fFXSError = -1; // this variable is kTRUE after ::Process if an FXS error occured
8da81210 1832
a038aa70 1833 UInt_t returnValue = aPreprocessor->Process(dcsMap);
8da81210 1834
339adafa 1835 if (fFXSError!=-1) {
8da81210 1836 UpdateShuttleStatus(AliShuttleStatus::kFXSError);
339adafa 1837 SendMail(kFXSEMail, fFXSError);
8da81210 1838 dcsMap->DeleteAll();
1839 delete dcsMap;
ffa25f20 1840 return 0;
8da81210 1841 }
b948db8d 1842
3301427a 1843 if (returnValue > 0) // Preprocessor error!
1844 {
c88ad5db 1845 Log(fCurrentDetector, Form("ProcessCurrentDetector - "
1846 "Preprocessor failed. Process returned %d.", returnValue));
cb343cfd 1847 UpdateShuttleStatus(AliShuttleStatus::kPPError);
a038aa70 1848 dcsMap->DeleteAll();
1849 delete dcsMap;
ffa25f20 1850 return 0;
9827400b 1851 }
1852
1853 // preprocessor ok!
1854 UpdateShuttleStatus(AliShuttleStatus::kPPDone);
1855 Log(fCurrentDetector, Form("ProcessCurrentDetector - %s preprocessor returned success",
1856 fCurrentDetector.Data()));
b948db8d 1857
a038aa70 1858 dcsMap->DeleteAll();
1859 delete dcsMap;
b948db8d 1860
ffa25f20 1861 return 1;
2bb7b766 1862}
1863
fbc112e3 1864//______________________________________________________________________________________________
1865void AliShuttle::CountOpenRuns()
1866{
1867 // Query DAQ's Shuttle logbook and sends the number of open runs to ML
1868
982278d6 1869 SendAlive();
1870
fbc112e3 1871 // check connection, in case connect
1872 if (!Connect(3))
1873 return;
1874
1875 TString sqlQuery;
1876 sqlQuery = Form("select count(*) from %s where shuttle_done=0", fConfig->GetShuttlelbTable());
1877
1878 TSQLResult* aResult = fServer[3]->Query(sqlQuery);
1879 if (!aResult) {
1880 AliError(Form("Can't execute query <%s>!", sqlQuery.Data()));
1881 return;
1882 }
1883
1884 AliDebug(2,Form("Query = %s", sqlQuery.Data()));
1885
1886 if (aResult->GetRowCount() == 0) {
1887 AliError(Form("No result for query %s received", sqlQuery.Data()));
1888 return;
1889 }
1890
1891 if (aResult->GetFieldCount() != 1) {
1892 AliError(Form("Invalid field count for query %s received", sqlQuery.Data()));
1893 return;
1894 }
1895
1896 TSQLRow* aRow = aResult->Next();
1897 if (!aRow) {
1898 AliError(Form("Could not receive result of query %s", sqlQuery.Data()));
1899 return;
1900 }
1901
1902 TString result(aRow->GetField(0), aRow->GetFieldLength(0));
1903 Int_t count = result.Atoi();
1904
1905 Log("SHUTTLE", Form("%d unprocessed runs", count));
1906
1907 delete aRow;
1908 delete aResult;
1909
1910 TMonaLisaValue mlStatus("SHUTTLE_openruns", count);
1911
1912 TList mlList;
1913 mlList.Add(&mlStatus);
1914
1915 fMonaLisa->SendParameters(&mlList, "__PROCESSINGINFO__");
1916}
1917
2bb7b766 1918//______________________________________________________________________________________________
1919Bool_t AliShuttle::QueryShuttleLogbook(const char* whereClause,
1920 TObjArray& entries)
1921{
9827400b 1922 // Query DAQ's Shuttle logbook and fills detector status object.
1923 // Call QueryRunParameters to query DAQ logbook for run parameters.
1924 //
2bb7b766 1925
fc5a4708 1926 entries.SetOwner(1);
1927
2bb7b766 1928 // check connection, in case connect
fbc112e3 1929 if (!Connect(3)) return kFALSE;
2bb7b766 1930
1931 TString sqlQuery;
441b0e9c 1932 sqlQuery = Form("select * from %s %s order by run", fConfig->GetShuttlelbTable(), whereClause);
2bb7b766 1933
be48e3ea 1934 TSQLResult* aResult = fServer[3]->Query(sqlQuery);
2bb7b766 1935 if (!aResult) {
1936 AliError(Form("Can't execute query <%s>!", sqlQuery.Data()));
1937 return kFALSE;
1938 }
1939
fc5a4708 1940 AliDebug(2,Form("Query = %s", sqlQuery.Data()));
1941
2bb7b766 1942 if(aResult->GetRowCount() == 0) {
c88ad5db 1943 Log("SHUTTLE", "No entries in Shuttle Logbook match request");
9827400b 1944 delete aResult;
1945 return kTRUE;
2bb7b766 1946 }
1947
1948 // TODO Check field count!
5b23e2da 1949 const UInt_t nCols = 26;
2bb7b766 1950 if (aResult->GetFieldCount() != (Int_t) nCols) {
c88ad5db 1951 Log("SHUTTLE", "Invalid SQL result field number!");
2bb7b766 1952 delete aResult;
1953 return kFALSE;
1954 }
1955
2bb7b766 1956 TSQLRow* aRow;
1957 while ((aRow = aResult->Next())) {
1958 TString runString(aRow->GetField(0), aRow->GetFieldLength(0));
1959 Int_t run = runString.Atoi();
1960
eba76848 1961 AliShuttleLogbookEntry *entry = QueryRunParameters(run);
1962 if (!entry)
1963 continue;
2bb7b766 1964
728290a4 1965 // DA test mode flag
1966 TString daTestModeString(aRow->GetField(2), aRow->GetFieldLength(2)); // field 2 = DA test mode flag
1967 Bool_t daTestMode = (Bool_t)daTestModeString.Atoi();
1968 entry->SetDATestMode(daTestMode);
1969
2bb7b766 1970 // loop on detectors
eba76848 1971 for(UInt_t ii = 0; ii < nCols; ii++)
1972 entry->SetDetectorStatus(aResult->GetFieldName(ii), aRow->GetField(ii));
2bb7b766 1973
eba76848 1974 entries.AddLast(entry);
2bb7b766 1975 delete aRow;
1976 }
1977
2bb7b766 1978 delete aResult;
1979 return kTRUE;
1980}
1981
1982//______________________________________________________________________________________________
eba76848 1983AliShuttleLogbookEntry* AliShuttle::QueryRunParameters(Int_t run)
2bb7b766 1984{
eba76848 1985 //
1986 // Retrieve run parameters written in the DAQ logbook and sets them into AliShuttleLogbookEntry object
1987 //
2bb7b766 1988
1989 // check connection, in case connect
be48e3ea 1990 if (!Connect(3))
eba76848 1991 return 0;
2bb7b766 1992
1993 TString sqlQuery;
2c15234c 1994 sqlQuery.Form("select * from %s where run=%d", fConfig->GetDAQlbTable(), run);
2bb7b766 1995
be48e3ea 1996 TSQLResult* aResult = fServer[3]->Query(sqlQuery);
2bb7b766 1997 if (!aResult) {
c88ad5db 1998 Log("SHUTTLE", Form("Can't execute query <%s>!", sqlQuery.Data()));
eba76848 1999 return 0;
2bb7b766 2000 }
2001
eba76848 2002 if (aResult->GetRowCount() == 0) {
2bb7b766 2003 Log("SHUTTLE", Form("QueryRunParameters - No entry in DAQ Logbook for run %d. Skipping", run));
2004 delete aResult;
eba76848 2005 return 0;
2bb7b766 2006 }
2007
eba76848 2008 if (aResult->GetRowCount() > 1) {
c88ad5db 2009 Log("SHUTTLE", Form("QueryRunParameters - UNEXPECTED: "
2010 "more than one entry in DAQ Logbook for run %d!", run));
2bb7b766 2011 delete aResult;
eba76848 2012 return 0;
2bb7b766 2013 }
2014
eba76848 2015 TSQLRow* aRow = aResult->Next();
2016 if (!aRow)
2017 {
c88ad5db 2018 Log("SHUTTLE", Form("QueryRunParameters - Could not retrieve row for run %d. Skipping", run));
eba76848 2019 delete aResult;
2020 return 0;
2021 }
2bb7b766 2022
eba76848 2023 AliShuttleLogbookEntry* entry = new AliShuttleLogbookEntry(run);
2bb7b766 2024
eba76848 2025 for (Int_t ii = 0; ii < aResult->GetFieldCount(); ii++)
2026 entry->SetRunParameter(aResult->GetFieldName(ii), aRow->GetField(ii));
2bb7b766 2027
c9db68c7 2028 delete aRow;
2029 delete aResult;
2030
eba76848 2031 UInt_t startTime = entry->GetStartTime();
2032 UInt_t endTime = entry->GetEndTime();
89bcf3b4 2033 Bool_t ecsSuccess = entry->GetECSSuccess();
1415d3f9 2034 TString runType = entry->GetRunType();
2035 TString tmpdaqstartTime = entry->GetRunParameter("DAQ_time_start");
2036 UInt_t daqstartTime = tmpdaqstartTime.Atoi();
c9db68c7 2037
af93e5d7 2038 UInt_t now = time(0);
2eb98462 2039 Int_t dcsDelay = fConfig->GetDCSDelay()+fConfig->GetDCSQueryOffset();
8e828d39 2040
2041 Bool_t skip = kFALSE;
2042
1415d3f9 2043 // runs are processed if
2044 // a) runType is PHYSICS and ecsSuccess is set
2045 // b) runType is not PHYSICS and (ecsSuccess is set or DAQ_time_start is non-0)
2046 // effectively this means that all runs are processed that started properly (ecsSucess behaviour is different for PHYSICS and non-PHYSICS runs (check with ECS!)
1bbcc21b 2047 if (startTime != 0 && endTime != 0) {
2048 if (endTime > startTime) {
2049 if (endTime >= now - dcsDelay) {
2050 Log("SHUTTLE", Form("Skipping run %d for now, because DCS buffer time is not yet expired", run));
2051 } else {
2052 if (runType == "PHYSICS") {
2053 if (ecsSuccess) {
2054 return entry;
2055 } else {
2056 Log("SHUTTLE", Form("QueryRunParameters - Run type for run %d is PHYSICS but ECS success flag not set (Reason = %s) - Skipping!", run, entry->GetRunParameter("eor_reason")));
2057 skip = kTRUE;
2058 }
2059 } else {
2060 if (ecsSuccess || daqstartTime > 0) {
2061 if (ecsSuccess == kFALSE)
2062 Log("SHUTTLE", Form("Processing run %d although in status ECS failure (Reason: %s), since run type != PHYSICS and DAQ_time_start != 0", run, entry->GetRunParameter("eor_reason")));
2063 return entry;
2064 } else {
2065 Log("SHUTTLE", Form("QueryRunParameters - Run type for run %d is %s, ECS success flag was not set (Reason = %s) and DAQ_time_start was NULL - Skipping!", run, runType.Data(), entry->GetRunParameter("eor_reason")));
2066 skip = kTRUE;
2067 }
2068 }
2069 }
2070 } else {
2071 Log("SHUTTLE", Form("QueryRunParameters - Invalid parameters for run %d: startTime equal to endTime: %d %d - Skipping!", run, startTime, endTime));
2072 skip = kTRUE;
2073 }
1415d3f9 2074 } else {
2075 Log("SHUTTLE", Form("QueryRunParameters - Invalid parameters for Run %d: "
2076 "startTime = %d, endTime = %d. Skipping (Shuttle won't be marked as DONE)!",
2077 run, startTime, endTime));
89bcf3b4 2078 }
8e828d39 2079
2080 if (skip)
2081 {
2082 Log("SHUTTLE", Form("Marking SHUTTLE skipped for run %d", run));
2083 fLogbookEntry = entry;
2084 if (!UpdateShuttleLogbook("shuttle_skipped"))
2085 {
2086 AliError(Form("Could not update logbook for run %d !", run));
2087 }
bd23d48b 2088 if (!UpdateTableSkippedCase("ALL"))
2089 {
2090 AliError(Form("Could not update FXS tables for run %d !", run));
2091 }
8e828d39 2092 fLogbookEntry = 0;
2093 }
c9db68c7 2094
2095 delete entry;
2096 return 0;
2bb7b766 2097}
2098
a038aa70 2099//______________________________________________________________________________________________
2100TMap* AliShuttle::GetValueSet(const char* host, Int_t port, const TSeqCollection* entries,
542b6cc8 2101 DCSType type, Int_t multiSplit)
a038aa70 2102{
2103 // Retrieve all "entry" data points from the DCS server
2104 // host, port: TSocket connection parameters
2105 // entries: list of name of the alias or data point
2106 // type: kAlias or kDP
2107 // returns TMap of values, 0 when failure
542b6cc8 2108
2109 AliDCSClient client(host, port, fTimeout, fRetries, multiSplit);
b41b252a 2110
a038aa70 2111 TMap* result = 0;
b41b252a 2112 if (type == kAlias)
a038aa70 2113 {
5f3fe74e 2114 //result = client.GetAliasValues(entries, GetCurrentStartTime()-offset,
2115 // GetCurrentEndTime()+offset);
eff647d5 2116 result = client.GetAliasValues(entries, GetStartTimeDCSQuery(),
2117 GetEndTimeDCSQuery());
b41b252a 2118 }
2119 else if (type == kDP)
2120 {
5f3fe74e 2121 //result = client.GetDPValues(entries, GetCurrentStartTime()-offset,
2122 // GetCurrentEndTime()+offset);
eff647d5 2123 result = client.GetDPValues(entries, GetStartTimeDCSQuery(),
2124 GetEndTimeDCSQuery());
b41b252a 2125 }
a038aa70 2126
b41b252a 2127 if (result == 0)
2128 {
2129 Log(fCurrentDetector.Data(), Form("GetValueSet - Can't get entries! Reason: %s",
1790d4b7 2130 client.GetErrorString(client.GetResultErrorCode())));
2131 if (client.GetResultErrorCode() == AliDCSClient::fgkServerError)
2132 Log(fCurrentDetector.Data(), Form("GetValueSet - Server error code: %s",
2133 client.GetServerError().Data()));
a038aa70 2134
b41b252a 2135 return 0;
a038aa70 2136 }
b41b252a 2137
a038aa70 2138 return result;
2139}
b41b252a 2140
b948db8d 2141//______________________________________________________________________________________________
57f50b3c 2142const char* AliShuttle::GetFile(Int_t system, const char* detector,
2143 const char* id, const char* source)
b948db8d 2144{
9827400b 2145 // Get calibration file from file exchange servers
2146 // First queris the FXS database for the file name, using the run, detector, id and source info
2147 // then calls RetrieveFile(filename) for actual copy to local disk
2148 // run: current run being processed (given by Logbook entry fLogbookEntry)
2149 // detector: the Preprocessor name
2150 // id: provided as a parameter by the Preprocessor
2151 // source: provided by the Preprocessor through GetFileSources function
2152
2153 // check if test mode should simulate a FXS error
2154 if (fTestMode & kErrorFXSFiles)
2155 {
2156 Log(detector, Form("GetFile - In TESTMODE - Simulating error while connecting to %s FXS", GetSystemName(system)));
2157 return 0;
2158 }
2159
57f50b3c 2160 // check connection, in case connect
9d733021 2161 if (!Connect(system))
eba76848 2162 {
9d733021 2163 Log(detector, Form("GetFile - Couldn't connect to %s FXS database", GetSystemName(system)));
339adafa 2164 fFXSError = system;
57f50b3c 2165 return 0;
2166 }
2167
2168 // Query preparation
9d733021 2169 TString sourceName(source);
d386d623 2170 Int_t nFields = 3;
2171 TString sqlQueryStart = Form("select filePath,size,fileChecksum from %s where",
2172 fConfig->GetFXSdbTable(system));
2173 TString whereClause = Form("run=%d and detector=\"%s\" and fileId=\"%s\"",
2174 GetCurrentRun(), detector, id);
2175
9d733021 2176 if (system == kDAQ)
2177 {
d386d623 2178 whereClause += Form(" and DAQsource=\"%s\"", source);
57f50b3c 2179 }
9d733021 2180 else if (system == kDCS)
eba76848 2181 {
9d733021 2182 sourceName="none";
57f50b3c 2183 }
9d733021 2184 else if (system == kHLT)
9e080f92 2185 {
d386d623 2186 whereClause += Form(" and DDLnumbers=\"%s\"", source);
9e080f92 2187 }
2188
9e080f92 2189 TString sqlQuery = Form("%s %s", sqlQueryStart.Data(), whereClause.Data());
2190
2191 AliDebug(2, Form("SQL query: \n%s",sqlQuery.Data()));
2192
2193 // Query execution
2194 TSQLResult* aResult = 0;
9d733021 2195 aResult = dynamic_cast<TSQLResult*> (fServer[system]->Query(sqlQuery));
9e080f92 2196 if (!aResult) {
8da81210 2197 Log(detector, Form("GetFile - Can't execute SQL query to %s database for: id = %s, source = %s",
9d733021 2198 GetSystemName(system), id, sourceName.Data()));
339adafa 2199 fFXSError = system;
9e080f92 2200 return 0;
2201 }
2202
8da81210 2203 if (aResult->GetRowCount() == 0)
9e080f92 2204 {
2205 Log(detector,
8da81210 2206 Form("GetFile - No entry in %s FXS db for: id = %s, source = %s",
9d733021 2207 GetSystemName(system), id, sourceName.Data()));
9e080f92 2208 delete aResult;
2209 return 0;
2210 }
2bb7b766 2211
9e080f92 2212 if (aResult->GetRowCount() > 1) {
2213 Log(detector,
8da81210 2214 Form("GetFile - More than one entry in %s FXS db for: id = %s, source = %s",
9d733021 2215 GetSystemName(system), id, sourceName.Data()));
339adafa 2216 fFXSError = system;
9e080f92 2217 delete aResult;
2218 return 0;
2219 }
2220
9d733021 2221 if (aResult->GetFieldCount() != nFields) {
9e080f92 2222 Log(detector,
9d733021 2223 Form("GetFileName - Wrong field count in %s FXS db for: id = %s, source = %s",
2224 GetSystemName(system), id, sourceName.Data()));
339adafa 2225 fFXSError = system;
9e080f92 2226 delete aResult;
2227 return 0;
2228 }
2229
2230 TSQLRow* aRow = dynamic_cast<TSQLRow*> (aResult->Next());
2231
2232 if (!aRow){
8da81210 2233 Log(detector, Form("GetFile - Empty set result in %s FXS db from query: id = %s, source = %s",
9d733021 2234 GetSystemName(system), id, sourceName.Data()));
339adafa 2235 fFXSError = system;
9e080f92 2236 delete aResult;
2237 return 0;
2238 }
2239
2240 TString filePath(aRow->GetField(0), aRow->GetFieldLength(0));
2241 TString fileSize(aRow->GetField(1), aRow->GetFieldLength(1));
d386d623 2242 TString fileChecksum(aRow->GetField(2), aRow->GetFieldLength(2));
9e080f92 2243
2244 delete aResult;
2245 delete aRow;
2246
d386d623 2247 AliDebug(2, Form("filePath = %s; size = %s, fileChecksum = %s",
2248 filePath.Data(), fileSize.Data(), fileChecksum.Data()));
9e080f92 2249
9e080f92 2250 // retrieved file is renamed to make it unique
3ebbf3c4 2251 Int_t dir_lev1 = GetCurrentRun()/10000;
2252 TString localFileName = Form("%s/%d/%d/%s_process/%s_%s_%d_%s_%s.shuttle",
2253 GetShuttleTempDir(), dir_lev1, GetCurrentRun(), detector,
d524ade6 2254 GetSystemName(system), detector, GetCurrentRun(),
2255 id, sourceName.Data());
3ebbf3c4 2256 Log("SHUTTLE",Form("file from FXS = %s",localFileName.Data()));
9d733021 2257
9e080f92 2258
9d733021 2259 // file retrieval from FXS
4b95672b 2260 UInt_t nRetries = 0;
2261 UInt_t maxRetries = 3;
2262 Bool_t result = kFALSE;
2263
2264 // copy!! if successful TSystem::Exec returns 0
8da81210 2265 while (nRetries++ < maxRetries) {
4b95672b 2266 AliDebug(2, Form("Trying to copy file. Retry # %d", nRetries));
2267 result = RetrieveFile(system, filePath.Data(), localFileName.Data());
8da81210 2268 if (!result)
4b95672b 2269 {
8da81210 2270 Log(detector, Form("GetFile - Copy of file %s from %s FXS failed",
9d733021 2271 filePath.Data(), GetSystemName(system)));
4b95672b 2272 continue;
4f0749a8 2273 }
9e080f92 2274
c98417a1 2275 if (fileSize.Length()>0)
2276 {
2277 // compare filesize of local file with the one stored in the FXS DB
926cbc0e 2278 Long_t size = -1;
2279 Int_t sizeComp = gSystem->GetPathInfo(localFileName.Data(), 0, &size, 0, 0);
c98417a1 2280
926cbc0e 2281 if (sizeComp != 0 || size != fileSize.Atoi())
c98417a1 2282 {
8da81210 2283 Log(detector, Form("GetFile - size of file %s does not match with local copy!",
c98417a1 2284 filePath.Data()));
2285 result = kFALSE;
2286 continue;
2287 }
2288
2289 } else {
2290 Log(fCurrentDetector, Form("GetFile - size of file %s not set in %s database, skipping comparison",
2291 filePath.Data(), GetSystemName(system)));
2292 }
2293
d386d623 2294 if (fileChecksum.Length()>0)
4b95672b 2295 {
2296 // compare md5sum of local file with the one stored in the FXS DB
95d4aaef 2297 if(fileChecksum.Contains(' ')) fileChecksum.Resize(fileChecksum.First(' '));
339adafa 2298 Int_t md5Comp = gSystem->Exec(Form("md5sum %s |grep %s > /dev/null 2> /dev/null",
d524ade6 2299 localFileName.Data(), fileChecksum.Data()));
9e080f92 2300
4b95672b 2301 if (md5Comp != 0)
2302 {
8da81210 2303 Log(detector, Form("GetFile - md5sum of file %s does not match with local copy!",
4b95672b 2304 filePath.Data()));
2305 result = kFALSE;
2306 continue;
2307 }
d386d623 2308 } else {
2309 Log(fCurrentDetector, Form("GetFile - md5sum of file %s not set in %s database, skipping comparison",
2310 filePath.Data(), GetSystemName(system)));
9d733021 2311 }
4b95672b 2312 if (result) break;
9e080f92 2313 }
2314
8da81210 2315 if (!result)
2316 {
339adafa 2317 fFXSError = system;
8da81210 2318 return 0;
2319 }
4b95672b 2320
9d733021 2321 fFXSCalled[system]=kTRUE;
2322 TObjString *fileParams = new TObjString(Form("%s#!?!#%s", id, sourceName.Data()));
2323 fFXSlist[system].Add(fileParams);
9e080f92 2324
675f64cd 2325 static TString staticLocalFileName;
2326 staticLocalFileName.Form("%s", localFileName.Data());
2327
c88ad5db 2328 Log(fCurrentDetector, Form("GetFile - Retrieved file with id %s and "
2329 "source %s from %s to %s", id, source,
d524ade6 2330 GetSystemName(system), localFileName.Data()));
675f64cd 2331
d524ade6 2332 return staticLocalFileName.Data();
2bb7b766 2333}
2334
2335//______________________________________________________________________________________________
9d733021 2336Bool_t AliShuttle::RetrieveFile(UInt_t system, const char* fxsFileName, const char* localFileName)
9e080f92 2337{
9827400b 2338 //
2339 // Copies file from FXS to local Shuttle machine
2340 //
2bb7b766 2341
9e080f92 2342 // check temp directory: trying to cd to temp; if it does not exist, create it
d524ade6 2343 AliDebug(2, Form("Copy file %s from %s FXS into %s",
2344 GetSystemName(system), fxsFileName, localFileName));
2345
2346 TString tmpDir(localFileName);
2347
2348 tmpDir = tmpDir(0,tmpDir.Last('/'));
9e080f92 2349
d524ade6 2350 Int_t noDir = gSystem->GetPathInfo(tmpDir.Data(), 0, (Long64_t*) 0, 0, 0);
2351 if (noDir) // temp dir does not exists!
2352 {
2353 if (gSystem->mkdir(tmpDir.Data(), 1))
2354 {
2355 Log(fCurrentDetector.Data(), "RetrieveFile - could not make temp directory!!");
9e080f92 2356 return kFALSE;
2357 }
9e080f92 2358 }
2359
339adafa 2360 TString command = Form("scp -oPort=%d -2 %s@%s:%s/%s %s",
9d733021 2361 fConfig->GetFXSPort(system),
2362 fConfig->GetFXSUser(system),
2363 fConfig->GetFXSHost(system),
339adafa 2364 fConfig->GetFXSBaseFolder(system),
9d733021 2365 fxsFileName,
9e080f92 2366 localFileName);
2367
2368 AliDebug(2, Form("%s",command.Data()));
2369
4b95672b 2370 Bool_t result = (gSystem->Exec(command.Data()) == 0);
9e080f92 2371
4b95672b 2372 return result;
9e080f92 2373}
2374
2375//______________________________________________________________________________________________
9d733021 2376TList* AliShuttle::GetFileSources(Int_t system, const char* detector, const char* id)
2377{
9827400b 2378 //
2379 // Get sources producing the condition file Id from file exchange servers
4a33bdd9 2380 // if id is NULL all sources are returned (distinct)
9827400b 2381 //
1bcd28db 2382
8884d71e 2383 if (id)
2384 {
2385 Log(detector, Form("GetFileSources - Querying %s FXS for files with id %s produced by %s", GetSystemName(system), id, detector));
2386 } else {
2387 Log(detector, Form("GetFileSources - Querying %s FXS for files produced by %s", GetSystemName(system), detector));
2388 }
9827400b 2389
2390 // check if test mode should simulate a FXS error
2391 if (fTestMode & kErrorFXSSources)
2392 {
2393 Log(detector, Form("GetFileSources - In TESTMODE - Simulating error while connecting to %s FXS", GetSystemName(system)));
2394 return 0;
2395 }
2396
9d733021 2397 if (system == kDCS)
2398 {
c88ad5db 2399 Log(detector, "GetFileSources - WARNING: DCS system has only one source of data!");
6297b37d 2400 TList *list = new TList();
2401 list->SetOwner(1);
2402 list->Add(new TObjString(" "));
2403 return list;
9d733021 2404 }
9e080f92 2405
2406 // check connection, in case connect
9d733021 2407 if (!Connect(system))
2408 {
4a33bdd9 2409 Log(detector, Form("GetFileSources - Couldn't connect to %s FXS database", GetSystemName(system)));
339adafa 2410 fFXSError = system;
9d733021 2411 return NULL;
9e080f92 2412 }
2413
b827f957 2414 TString sourceName = "";
9d733021 2415 if (system == kDAQ)
2416 {
2417 sourceName = "DAQsource";
2418 } else if (system == kHLT)
2419 {
2420 sourceName = "DDLnumbers";
2421 }
2422
4a33bdd9 2423 TString sqlQueryStart = Form("select distinct %s from %s where", sourceName.Data(), fConfig->GetFXSdbTable(system));
2424 TString whereClause = Form("run=%d and detector=\"%s\"",
2425 GetCurrentRun(), detector);
2426 if (id)
2427 whereClause += Form(" and fileId=\"%s\"", id);
9e080f92 2428 TString sqlQuery = Form("%s %s", sqlQueryStart.Data(), whereClause.Data());
2429
2430 AliDebug(2, Form("SQL query: \n%s",sqlQuery.Data()));
2431
2432 // Query execution
2433 TSQLResult* aResult;
9d733021 2434 aResult = fServer[system]->Query(sqlQuery);
9e080f92 2435 if (!aResult) {
9d733021 2436 Log(detector, Form("GetFileSources - Can't execute SQL query to %s database for id: %s",
2437 GetSystemName(system), id));
339adafa 2438 fFXSError = system;
9e080f92 2439 return 0;
2440 }
2441
86aa42c3 2442 TList *list = new TList();
2443 list->SetOwner(1);
2444
9d733021 2445 if (aResult->GetRowCount() == 0)
2446 {
9e080f92 2447 Log(detector,
9d733021 2448 Form("GetFileSources - No entry in %s FXS table for id: %s", GetSystemName(system), id));
9e080f92 2449 delete aResult;
86aa42c3 2450 return list;
9e080f92 2451 }
2452
1bcd28db 2453 Log(detector, Form("GetFileSources - Found %d sources", aResult->GetRowCount()));
9e080f92 2454
1bcd28db 2455 TSQLRow* aRow;
9d733021 2456 while ((aRow = aResult->Next()))
2457 {
9e080f92 2458
9d733021 2459 TString source(aRow->GetField(0), aRow->GetFieldLength(0));
2460 AliDebug(2, Form("%s = %s", sourceName.Data(), source.Data()));
2461 list->Add(new TObjString(source));
9e080f92 2462 delete aRow;
2463 }
9d733021 2464
9e080f92 2465 delete aResult;
2466
2467 return list;
2bb7b766 2468}
2469
4a33bdd9 2470//______________________________________________________________________________________________
2471TList* AliShuttle::GetFileIDs(Int_t system, const char* detector, const char* source)
2472{
2473 //
2474 // Get all ids of condition files produced by a given source from file exchange servers
2475 //
2476
1bcd28db 2477 Log(detector, Form("GetFileIDs - Retrieving ids with source %s with %s", source, GetSystemName(system)));
2478
4a33bdd9 2479 // check if test mode should simulate a FXS error
2480 if (fTestMode & kErrorFXSSources)
2481 {
2482 Log(detector, Form("GetFileIDs - In TESTMODE - Simulating error while connecting to %s FXS", GetSystemName(system)));
2483 return 0;
2484 }
2485
2486 // check connection, in case connect
2487 if (!Connect(system))
2488 {
2489 Log(detector, Form("GetFileIDs - Couldn't connect to %s FXS database", GetSystemName(system)));
2490 return NULL;
2491 }
2492
b827f957 2493 TString sourceName = "";
4a33bdd9 2494 if (system == kDAQ)
2495 {
2496 sourceName = "DAQsource";
2497 } else if (system == kHLT)
2498 {
2499 sourceName = "DDLnumbers";
2500 }
2501
2502 TString sqlQueryStart = Form("select fileId from %s where", fConfig->GetFXSdbTable(system));
2503 TString whereClause = Form("run=%d and detector=\"%s\"",
2504 GetCurrentRun(), detector);
2505 if (sourceName.Length() > 0 && source)
2506 whereClause += Form(" and %s=\"%s\"", sourceName.Data(), source);
2507 TString sqlQuery = Form("%s %s", sqlQueryStart.Data(), whereClause.Data());
2508
2509 AliDebug(2, Form("SQL query: \n%s",sqlQuery.Data()));
2510
2511 // Query execution
2512 TSQLResult* aResult;
2513 aResult = fServer[system]->Query(sqlQuery);
2514 if (!aResult) {
2515 Log(detector, Form("GetFileIDs - Can't execute SQL query to %s database for source: %s",
2516 GetSystemName(system), source));
2517 return 0;
2518 }
2519
2520 TList *list = new TList();
2521 list->SetOwner(1);
2522
2523 if (aResult->GetRowCount() == 0)
2524 {
2525 Log(detector,
2526 Form("GetFileIDs - No entry in %s FXS table for source: %s", GetSystemName(system), source));
2527 delete aResult;
2528 return list;
2529 }
2530
1bcd28db 2531 Log(detector, Form("GetFileIDs - Found %d ids", aResult->GetRowCount()));
2532
4a33bdd9 2533 TSQLRow* aRow;
2534
2535 while ((aRow = aResult->Next()))
2536 {
2537
2538 TString id(aRow->GetField(0), aRow->GetFieldLength(0));
2539 AliDebug(2, Form("fileId = %s", id.Data()));
2540 list->Add(new TObjString(id));
2541 delete aRow;
2542 }
2543
2544 delete aResult;
2545
2546 return list;
2547}
2548
2bb7b766 2549//______________________________________________________________________________________________
9d733021 2550Bool_t AliShuttle::Connect(Int_t system)
2bb7b766 2551{
9827400b 2552 // Connect to MySQL Server of the system's FXS MySQL databases
2553 // DAQ Logbook, Shuttle Logbook and DAQ FXS db are on the same host
2554 //
57f50b3c 2555
9d733021 2556 // check connection: if already connected return
dd84c76d 2557
3a4ddf0e 2558 if(fServer[system] && fServer[system]->IsConnected()) {
dd84c76d 2559 // ping the server
2560 if (fServer[system]->PingVerify()==kTRUE){ // connection is still alive
2561 return kTRUE;
2562 }
2563 else{
2564 AliWarning(Form("Connection got lost to FXS database for %s. Closing and reconnecting.",
2565 AliShuttleInterface::GetSystemName(system)));
2566 fServer[system]->Close();
2567 delete fServer[system];
2568 fServer[system] = 0x0;
2569 }
3a4ddf0e 2570 }
57f50b3c 2571
9d733021 2572 TString dbHost, dbUser, dbPass, dbName;
57f50b3c 2573
9d733021 2574 if (system < 3) // FXS db servers
2575 {
2576 dbHost = Form("mysql://%s:%d", fConfig->GetFXSdbHost(system), fConfig->GetFXSdbPort(system));
2577 dbUser = fConfig->GetFXSdbUser(system);
2578 dbPass = fConfig->GetFXSdbPass(system);
2579 dbName = fConfig->GetFXSdbName(system);
2580 } else { // Run & Shuttle logbook servers
2581 // TODO Will the Shuttle logbook server be the same as the Run logbook server ???
2582 dbHost = Form("mysql://%s:%d", fConfig->GetDAQlbHost(), fConfig->GetDAQlbPort());
2583 dbUser = fConfig->GetDAQlbUser();
2584 dbPass = fConfig->GetDAQlbPass();
2585 dbName = fConfig->GetDAQlbDB();
2586 }
57f50b3c 2587
9d733021 2588 fServer[system] = TSQLServer::Connect(dbHost.Data(), dbUser.Data(), dbPass.Data());
2589 if (!fServer[system] || !fServer[system]->IsConnected()) {
2590 if(system < 3)
2591 {
2592 AliError(Form("Can't establish connection to FXS database for %s",
2593 AliShuttleInterface::GetSystemName(system)));
2594 } else {
2595 AliError("Can't establish connection to Run logbook.");
57f50b3c 2596 }
9d733021 2597 if(fServer[system]) delete fServer[system];
2598 return kFALSE;
2bb7b766 2599 }
57f50b3c 2600
9d733021 2601 // Get tables
2602 TSQLResult* aResult=0;
2603 switch(system){
2604 case kDAQ:
2605 aResult = fServer[kDAQ]->GetTables(dbName.Data());
2606 break;
2607 case kDCS:
2608 aResult = fServer[kDCS]->GetTables(dbName.Data());
2609 break;
2610 case kHLT:
2611 aResult = fServer[kHLT]->GetTables(dbName.Data());
2612 break;
2613 default:
2614 aResult = fServer[3]->GetTables(dbName.Data());
2615 break;
2616 }
2617
2618 delete aResult;
2bb7b766 2619 return kTRUE;
2620}
57f50b3c 2621
9e080f92 2622//______________________________________________________________________________________________
9d733021 2623Bool_t AliShuttle::UpdateTable()
9e080f92 2624{
9827400b 2625 //
2626 // Update FXS table filling time_processed field in all rows corresponding to current run and detector
2627 //
9e080f92 2628
9d733021 2629 Bool_t result = kTRUE;
9e080f92 2630
9d733021 2631 for (UInt_t system=0; system<3; system++)
2632 {
2633 if(!fFXSCalled[system]) continue;
9e080f92 2634
9d733021 2635 // check connection, in case connect
2636 if (!Connect(system))
2637 {
2638 Log(fCurrentDetector, Form("UpdateTable - Couldn't connect to %s FXS database", GetSystemName(system)));
2639 result = kFALSE;
2640 continue;
9e080f92 2641 }
9e080f92 2642
9d733021 2643 TTimeStamp now; // now
2644
2645 // Loop on FXS list entries
2646 TIter iter(&fFXSlist[system]);
2647 TObjString *aFXSentry=0;
2648 while ((aFXSentry = dynamic_cast<TObjString*> (iter.Next())))
2649 {
2650 TString aFXSentrystr = aFXSentry->String();
2651 TObjArray *aFXSarray = aFXSentrystr.Tokenize("#!?!#");
2652 if (!aFXSarray || aFXSarray->GetEntries() != 2 )
2653 {
2654 Log(fCurrentDetector, Form("UpdateTable - error updating %s FXS entry. Check string: <%s>",
2655 GetSystemName(system), aFXSentrystr.Data()));
2656 if(aFXSarray) delete aFXSarray;
2657 result = kFALSE;
2658 continue;
2659 }
2660 const char* fileId = ((TObjString*) aFXSarray->At(0))->GetName();
2661 const char* source = ((TObjString*) aFXSarray->At(1))->GetName();
2662
2663 TString whereClause;
2664 if (system == kDAQ)
2665 {
2666 whereClause = Form("where run=%d and detector=\"%s\" and fileId=\"%s\" and DAQsource=\"%s\";",
2667 GetCurrentRun(), fCurrentDetector.Data(), fileId, source);
2668 }
2669 else if (system == kDCS)
2670 {
2671 whereClause = Form("where run=%d and detector=\"%s\" and fileId=\"%s\";",
2672 GetCurrentRun(), fCurrentDetector.Data(), fileId);
2673 }
2674 else if (system == kHLT)
2675 {
2676 whereClause = Form("where run=%d and detector=\"%s\" and fileId=\"%s\" and DDLnumbers=\"%s\";",
2677 GetCurrentRun(), fCurrentDetector.Data(), fileId, source);
2678 }
2679
2680 delete aFXSarray;
9e080f92 2681
9d733021 2682 TString sqlQuery = Form("update %s set time_processed=%d %s", fConfig->GetFXSdbTable(system),
2683 now.GetSec(), whereClause.Data());
9e080f92 2684
9d733021 2685 AliDebug(2, Form("SQL query: \n%s",sqlQuery.Data()));
9e080f92 2686
9d733021 2687 // Query execution
2688 TSQLResult* aResult;
2689 aResult = dynamic_cast<TSQLResult*> (fServer[system]->Query(sqlQuery));
2690 if (!aResult)
2691 {
2692 Log(fCurrentDetector, Form("UpdateTable - %s db: can't execute SQL query <%s>",
2693 GetSystemName(system), sqlQuery.Data()));
2694 result = kFALSE;
2695 continue;
2696 }
2697 delete aResult;
9e080f92 2698 }
9e080f92 2699 }
2700
9d733021 2701 return result;
9e080f92 2702}
57f50b3c 2703
bd23d48b 2704//_______________________________________________________________________________
2705Bool_t AliShuttle::UpdateTableSkippedCase(const char* detector)
2706{
2707 //
2708 // Update FXS table filling time_processed field in all rows corresponding to current run and detector
2709 // if detector = "ALL" update all detectors
2710 //
2711
2712 Bool_t result = kTRUE;
2713
7a782903 2714 TString detName(detector);
2715
bd23d48b 2716 for (UInt_t system=0; system<3; system++)
2717 {
2718
2719 // check connection, in case connect
2720 if (!Connect(system))
2721 {
2722 Log(fCurrentDetector, Form("UpdateTableSkippedCase - Couldn't connect to %s FXS database", GetSystemName(system)));
2723 result = kFALSE;
2724 continue;
2725 }
2726
2727 TTimeStamp now; // now
2728
2729 // Loop on FXS list entries
2730 TIter iter(&fFXSlist[system]);
2731
2732 TString whereClause;
7a782903 2733 if (detName == "ALL") whereClause = Form("where run=%d and time_processed IS NULL;",GetCurrentRun());
bd23d48b 2734 else whereClause = Form("where run=%d and detector=\"%s\" and time_processed IS NULL;",GetCurrentRun(), detector);
2735
4be2f262 2736 //Log("SHUTTLE",Form(" whereClause = %s ",whereClause.Data()));
bd23d48b 2737
2738 TString sqlQuery = Form("update %s set time_processed=%d %s", fConfig->GetFXSdbTable(system),
2739 now.GetSec(), whereClause.Data());
2740
2741 AliDebug(2, Form("SQL query: \n%s",sqlQuery.Data()));
2742
2743 // Query execution
2744 TSQLResult* aResult;
2745 aResult = dynamic_cast<TSQLResult*> (fServer[system]->Query(sqlQuery));
2746 if (!aResult)
2747 {
2748 Log("SHUTTLE", Form("UpdateTableSkippedCase - %s db: can't execute SQL query <%s>",
2749 GetSystemName(system), sqlQuery.Data()));
2750 result = kFALSE;
2751 continue;
2752 }
2753 delete aResult;
2754
2755 }
2756
2757 return result;
2758}
3301427a 2759//______________________________________________________________________________________________
2760Bool_t AliShuttle::UpdateTableFailCase()
2761{
9827400b 2762 // Update FXS table filling time_processed field in all rows corresponding to current run and detector
2763 // this is called in case the preprocessor is declared failed for the current run, because
2764 // the fields are updated only in case of success
3301427a 2765
2766 Bool_t result = kTRUE;
2767
2768 for (UInt_t system=0; system<3; system++)
2769 {
2770 // check connection, in case connect
2771 if (!Connect(system))
2772 {
2773 Log(fCurrentDetector, Form("UpdateTableFailCase - Couldn't connect to %s FXS database",
2774 GetSystemName(system)));
2775 result = kFALSE;
2776 continue;
2777 }
2778
2779 TTimeStamp now; // now
2780
2781 // Loop on FXS list entries
2782
2783 TString whereClause = Form("where run=%d and detector=\"%s\";",
2784 GetCurrentRun(), fCurrentDetector.Data());
2785
2786
2787 TString sqlQuery = Form("update %s set time_processed=%d %s", fConfig->GetFXSdbTable(system),
2788 now.GetSec(), whereClause.Data());
2789
2790 AliDebug(2, Form("SQL query: \n%s",sqlQuery.Data()));
2791
2792 // Query execution
2793 TSQLResult* aResult;
2794 aResult = dynamic_cast<TSQLResult*> (fServer[system]->Query(sqlQuery));
2795 if (!aResult)
2796 {
2797 Log(fCurrentDetector, Form("UpdateTableFailCase - %s db: can't execute SQL query <%s>",
2798 GetSystemName(system), sqlQuery.Data()));
2799 result = kFALSE;
2800 continue;
2801 }
2802 delete aResult;
2803 }
2804
2805 return result;
2806}
2807
2bb7b766 2808//______________________________________________________________________________________________
2809Bool_t AliShuttle::UpdateShuttleLogbook(const char* detector, const char* status)
2810{
e7f62f16 2811 //
2812 // Update Shuttle logbook filling detector or shuttle_done column
2813 // ex. of usage: UpdateShuttleLogbook("PHOS", "DONE") or UpdateShuttleLogbook("shuttle_done")
2814 //
57f50b3c 2815
2bb7b766 2816 // check connection, in case connect
be48e3ea 2817 if(!Connect(3)){
2bb7b766 2818 Log("SHUTTLE", "UpdateShuttleLogbook - Couldn't connect to DAQ Logbook.");
2819 return kFALSE;
57f50b3c 2820 }
2821
2bb7b766 2822 TString detName(detector);
2823 TString setClause;
8e828d39 2824 if (detName == "shuttle_done" || detName == "shuttle_skipped")
e7f62f16 2825 {
2bb7b766 2826 setClause = "set shuttle_done=1";
8e828d39 2827
ee6f7523 2828 if (detName == "shuttle_done")
b0e53b15 2829 {
8e828d39 2830 if (TouchFile() != kTRUE)
2156c755 2831 {
2832 SendMLRunInfo("Pending");
503ff24f 2833 return kFALSE;
2156c755 2834 }
8e828d39 2835
2836 SendMLRunInfo("Done");
b0e53b15 2837 }
8e828d39 2838 else
2839 SendMLRunInfo("Skipped");
2840 }
2841 else {
2bb7b766 2842 TString statusStr(status);
2843 if(statusStr.Contains("done", TString::kIgnoreCase) ||
2844 statusStr.Contains("failed", TString::kIgnoreCase)){
eba76848 2845 setClause = Form("set %s=\"%s\"", detector, status);
2bb7b766 2846 } else {
2847 Log("SHUTTLE",
2848 Form("UpdateShuttleLogbook - Invalid status <%s> for detector %s",
2849 status, detector));
2850 return kFALSE;
2851 }
2852 }
57f50b3c 2853
2bb7b766 2854 TString whereClause = Form("where run=%d", GetCurrentRun());
2855
441b0e9c 2856 TString sqlQuery = Form("update %s %s %s",
2857 fConfig->GetShuttlelbTable(), setClause.Data(), whereClause.Data());
57f50b3c 2858
2bb7b766 2859 AliDebug(2, Form("SQL query: \n%s",sqlQuery.Data()));
2860
2861 // Query execution
2862 TSQLResult* aResult;
be48e3ea 2863 aResult = dynamic_cast<TSQLResult*> (fServer[3]->Query(sqlQuery));
2bb7b766 2864 if (!aResult) {
2865 Log("SHUTTLE", Form("UpdateShuttleLogbook - Can't execute query <%s>", sqlQuery.Data()));
2866 return kFALSE;
57f50b3c 2867 }
2bb7b766 2868 delete aResult;
57f50b3c 2869
2870 return kTRUE;
2871}
2872
2873//______________________________________________________________________________________________
2bb7b766 2874Int_t AliShuttle::GetCurrentRun() const
2875{
9827400b 2876 //
2877 // Get current run from logbook entry
2878 //
57f50b3c 2879
2bb7b766 2880 return fLogbookEntry ? fLogbookEntry->GetRun() : -1;
57f50b3c 2881}
2882
2883//______________________________________________________________________________________________
2bb7b766 2884UInt_t AliShuttle::GetCurrentStartTime() const
2885{
9827400b 2886 //
2887 // get current start time
2888 //
57f50b3c 2889
2bb7b766 2890 return fLogbookEntry ? fLogbookEntry->GetStartTime() : 0;
57f50b3c 2891}
2892
2893//______________________________________________________________________________________________
2bb7b766 2894UInt_t AliShuttle::GetCurrentEndTime() const
2895{
9827400b 2896 //
2897 // get current end time from logbook entry
2898 //
57f50b3c 2899
2bb7b766 2900 return fLogbookEntry ? fLogbookEntry->GetEndTime() : 0;
57f50b3c 2901}
675f64cd 2902//______________________________________________________________________________________________
2903UInt_t AliShuttle::GetCurrentYear() const
2904{
2905 //
2906 // Get current year from logbook entry
2907 //
2908
2909 if (!fLogbookEntry) return 0;
2910
2911 TTimeStamp startTime(GetCurrentStartTime());
2912 TString year = Form("%d",startTime.GetDate());
2913 year = year(0,4);
2914
2915 return year.Atoi();
2916}
2917
2918//______________________________________________________________________________________________
2919const char* AliShuttle::GetLHCPeriod() const
2920{
2921 //
2922 // Get current LHC period from logbook entry
2923 //
2924
2925 if (!fLogbookEntry) return 0;
2926
2927 return fLogbookEntry->GetRunParameter("LHCperiod");
2928}
2929
b948db8d 2930//______________________________________________________________________________________________
2931void AliShuttle::Log(const char* detector, const char* message)
2932{
9827400b 2933 //
2934 // Fill log string with a message
2935 //
7d4cf768 2936
3ebbf3c4 2937 TString logRunDir = GetShuttleLogDir();
2938 if (GetCurrentRun() >=0) {
2939 Int_t logDir_lev1 = GetCurrentRun()/10000;
2940 logRunDir += Form("/%d/%d", logDir_lev1, GetCurrentRun());
2941 }
7d4cf768 2942 void* dir = gSystem->OpenDirectory(logRunDir.Data());
84090f85 2943 if (dir == NULL) {
7d4cf768 2944 if (gSystem->mkdir(logRunDir.Data(), kTRUE)) {
36c99a6a 2945 AliError(Form("Can't open directory <%s>", GetShuttleLogDir()));
84090f85 2946 return;
2947 }
b948db8d 2948
84090f85 2949 } else {
2950 gSystem->FreeDirectory(dir);
2951 }
b948db8d 2952
0b025a52 2953 TString toLog = Form("%s UTC (%d): %s - ", TTimeStamp(time(0)).AsString("s"), getpid(), detector);
e7f62f16 2954 if (GetCurrentRun() >= 0)
2955 toLog += Form("run %d - ", GetCurrentRun());
2bb7b766 2956 toLog += Form("%s", message);
2957
84090f85 2958 AliInfo(toLog.Data());
ffa29e93 2959
2960 // if we redirect the log output already to the file, leave here
2961 if (fOutputRedirected && strcmp(detector, "SHUTTLE") != 0)
2962 return;
b948db8d 2963
ffa29e93 2964 TString fileName = GetLogFileName(detector);
e7f62f16 2965
84090f85 2966 gSystem->ExpandPathName(fileName);
2967
2968 ofstream logFile;
2969 logFile.open(fileName, ofstream::out | ofstream::app);
2970
2971 if (!logFile.is_open()) {
2972 AliError(Form("Could not open file %s", fileName.Data()));
2973 return;
2974 }
7bfb2090 2975
84090f85 2976 logFile << toLog.Data() << "\n";
b948db8d 2977
84090f85 2978 logFile.close();
b948db8d 2979}
2bb7b766 2980
ffa29e93 2981//______________________________________________________________________________________________
2982TString AliShuttle::GetLogFileName(const char* detector) const
2983{
2984 //
2985 // returns the name of the log file for a given sub detector
2986 //
2987
2988 TString fileName;
2989
2990 if (GetCurrentRun() >= 0)
7d4cf768 2991 {
3ebbf3c4 2992 Int_t logDir_lev1 = GetCurrentRun()/10000;
2993 fileName.Form("%s/%d/%d/%s.log", GetShuttleLogDir(), logDir_lev1, GetCurrentRun(),
2994 detector);
7d4cf768 2995 } else {
ffa29e93 2996 fileName.Form("%s/%s.log", GetShuttleLogDir(), detector);
7d4cf768 2997 }
ffa29e93 2998
2999 return fileName;
3000}
3001
ee6f7523 3002//______________________________________________________________________________________________
3003void AliShuttle::SendAlive()
3004{
3005 // sends alive message to ML
3006
3007 TMonaLisaText mlStatus("SHUTTLE_status", "Alive");
3008
3009 TList mlList;
3010 mlList.Add(&mlStatus);
3011
3012 fMonaLisa->SendParameters(&mlList, "__PROCESSINGINFO__");
3013}
3014
2bb7b766 3015//______________________________________________________________________________________________
3016Bool_t AliShuttle::Collect(Int_t run)
3017{
9827400b 3018 //
3019 // Collects conditions data for all UNPROCESSED run written to DAQ LogBook in case of run = -1 (default)
3020 // If a dedicated run is given this run is processed
3021 //
3022 // In operational mode, this is the Shuttle function triggered by the EOR signal.
3023 //
2bb7b766 3024
eba76848 3025 if (run == -1)
3026 Log("SHUTTLE","Collect - Shuttle called. Collecting conditions data for unprocessed runs");
3027 else
3028 Log("SHUTTLE", Form("Collect - Shuttle called. Collecting conditions data for run %d", run));
cb343cfd 3029
3030 SetLastAction("Starting");
2bb7b766 3031
ee6f7523 3032 // create ML instance
3033 if (!fMonaLisa)
3034 fMonaLisa = new TMonaLisaWriter(fConfig->GetMonitorHost(), fConfig->GetMonitorTable());
3035
fbc112e3 3036 CountOpenRuns();
ee6f7523 3037
2bb7b766 3038 TString whereClause("where shuttle_done=0");
eba76848 3039 if (run != -1)
3040 whereClause += Form(" and run=%d", run);
2bb7b766 3041
3042 TObjArray shuttleLogbookEntries;
be48e3ea 3043 if (!QueryShuttleLogbook(whereClause, shuttleLogbookEntries))
3044 {
cb343cfd 3045 Log("SHUTTLE", "Collect - Can't retrieve entries from Shuttle logbook");
2bb7b766 3046 return kFALSE;
3047 }
3048
9e080f92 3049 if (shuttleLogbookEntries.GetEntries() == 0)
3050 {
3051 if (run == -1)
3052 Log("SHUTTLE","Collect - Found no UNPROCESSED runs in Shuttle logbook");
3053 else
3054 Log("SHUTTLE", Form("Collect - Run %d is already DONE "
3055 "or it does not exist in Shuttle logbook", run));
3056 return kTRUE;
3057 }
3058
be48e3ea 3059 for (UInt_t iDet=0; iDet<NDetectors(); iDet++)
3060 fFirstUnprocessed[iDet] = kTRUE;
3061
fc5a4708 3062 if (run != -1)
be48e3ea 3063 {
3064 // query Shuttle logbook for earlier runs, check if some detectors are unprocessed,
3065 // flag them into fFirstUnprocessed array
e3ae8d33 3066 TString whereClauseBis(Form("where shuttle_done=0 and run < %d", run));
be48e3ea 3067 TObjArray tmpLogbookEntries;
e3ae8d33 3068 if (!QueryShuttleLogbook(whereClauseBis, tmpLogbookEntries))
be48e3ea 3069 {
3070 Log("SHUTTLE", "Collect - Can't retrieve entries from Shuttle logbook");
3071 return kFALSE;
3072 }
3073
3074 TIter iter(&tmpLogbookEntries);
3075 AliShuttleLogbookEntry* anEntry = 0;
3076 while ((anEntry = dynamic_cast<AliShuttleLogbookEntry*> (iter.Next())))
3077 {
3078 for (UInt_t iDet=0; iDet<NDetectors(); iDet++)
3079 {
3080 if (anEntry->GetDetectorStatus(iDet) == AliShuttleLogbookEntry::kUnprocessed)
3081 {
3082 AliDebug(2, Form("Run %d: setting %s as \"not first time unprocessed\"",
3083 anEntry->GetRun(), GetDetName(iDet)));
3084 fFirstUnprocessed[iDet] = kFALSE;
3085 }
3086 }
3087
3088 }
3089
3090 }
3091
3092 if (!RetrieveConditionsData(shuttleLogbookEntries))
3093 {
cb343cfd 3094 Log("SHUTTLE", "Collect - Process of at least one run failed");
a3ca69fe 3095 CountOpenRuns();
2bb7b766 3096 return kFALSE;
3097 }
3098
36c99a6a 3099 Log("SHUTTLE", "Collect - Requested run(s) successfully processed");
a3ca69fe 3100 CountOpenRuns();
eba76848 3101 return kTRUE;
2bb7b766 3102}
3103
2bb7b766 3104//______________________________________________________________________________________________
3105Bool_t AliShuttle::RetrieveConditionsData(const TObjArray& dateEntries)
3106{
9827400b 3107 //
3108 // Retrieve conditions data for all runs that aren't processed yet
3109 //
2bb7b766 3110
3111 Bool_t hasError = kFALSE;
3112
3113 TIter iter(&dateEntries);
3114 AliShuttleLogbookEntry* anEntry;
3115
3116 while ((anEntry = (AliShuttleLogbookEntry*) iter.Next())){
3117 if (!Process(anEntry)){
3118 hasError = kTRUE;
3119 }
4b95672b 3120
3121 // clean SHUTTLE temp directory
d524ade6 3122 //TString filename = Form("%s/*.shuttle", GetShuttleTempDir());
3123 //RemoveFile(filename.Data());
2bb7b766 3124 }
3125
3126 return hasError == kFALSE;
3127}
cb343cfd 3128
3129//______________________________________________________________________________________________
3130ULong_t AliShuttle::GetTimeOfLastAction() const
3131{
9827400b 3132 //
3133 // Gets time of last action
3134 //
3135
cb343cfd 3136 ULong_t tmp;
36c99a6a 3137
cb343cfd 3138 fMonitoringMutex->Lock();
be48e3ea 3139
cb343cfd 3140 tmp = fLastActionTime;
36c99a6a 3141
cb343cfd 3142 fMonitoringMutex->UnLock();
36c99a6a 3143
cb343cfd 3144 return tmp;
3145}
3146
3147//______________________________________________________________________________________________
3148const TString AliShuttle::GetLastAction() const
3149{
9827400b 3150 //
cb343cfd 3151 // returns a string description of the last action
9827400b 3152 //
cb343cfd 3153
3154 TString tmp;
36c99a6a 3155
cb343cfd 3156 fMonitoringMutex->Lock();
3157
3158 tmp = fLastAction;
3159
3160 fMonitoringMutex->UnLock();
3161
36c99a6a 3162 return tmp;
cb343cfd 3163}
3164
3165//______________________________________________________________________________________________
3166void AliShuttle::SetLastAction(const char* action)
3167{
9827400b 3168 //
cb343cfd 3169 // updates the monitoring variables
9827400b 3170 //
36c99a6a 3171
cb343cfd 3172 fMonitoringMutex->Lock();
36c99a6a 3173
cb343cfd 3174 fLastAction = action;
3175 fLastActionTime = time(0);
3176
3177 fMonitoringMutex->UnLock();
3178}
eba76848 3179
3180//______________________________________________________________________________________________
3181const char* AliShuttle::GetRunParameter(const char* param)
3182{
9827400b 3183 //
3184 // returns run parameter read from DAQ logbook
3185 //
eba76848 3186
3187 if(!fLogbookEntry) {
3188 AliError("No logbook entry!");
3189 return 0;
3190 }
3191
3192 return fLogbookEntry->GetRunParameter(param);
3193}
57c1a579 3194
d386d623 3195//______________________________________________________________________________________________
9827400b 3196AliCDBEntry* AliShuttle::GetFromOCDB(const char* detector, const AliCDBPath& path)
d386d623 3197{
9827400b 3198 //
3199 // returns object from OCDB valid for current run
3200 //
d386d623 3201
9827400b 3202 if (fTestMode & kErrorOCDB)
3203 {
3204 Log(detector, "GetFromOCDB - In TESTMODE - Simulating error with OCDB");
3205 return 0;
3206 }
3207
d386d623 3208 AliCDBStorage *sto = AliCDBManager::Instance()->GetStorage(fgkMainCDB);
3209 if (!sto)
3210 {
9827400b 3211 Log(detector, "GetFromOCDB - Cannot activate main OCDB for query!");
d386d623 3212 return 0;
3213 }
3214
3215 return dynamic_cast<AliCDBEntry*> (sto->Get(path, GetCurrentRun()));
3216}
3217
57c1a579 3218//______________________________________________________________________________________________
339adafa 3219Bool_t AliShuttle::SendMail(EMailTarget target, Int_t system)
57c1a579 3220{
9827400b 3221 //
3222 // sends a mail to the subdetector expert in case of preprocessor error
3223 //
3224
3225 if (fTestMode != kNone)
3226 return kTRUE;
fb40ead4 3227
926cbc0e 3228 if (!fConfig->SendMail())
3229 return kTRUE;
57c1a579 3230
339adafa 3231 if (target == kDCSEMail || target == kFXSEMail) {
3232 if (!fFirstProcessing)
3a4ddf0e 3233 return kTRUE;
36c99a6a 3234 }
3235
7bdfe1fa 3236 Int_t runMode = (Int_t)fConfig->GetRunMode();
3237 TString tmpStr;
3238 if (runMode == 0) tmpStr = " Nightly Test:";
3239 else tmpStr = " Data Taking:";
fbc112e3 3240 void* dir = gSystem->OpenDirectory(GetShuttleLogDir());
3241 if (dir == NULL)
3242 {
3243 if (gSystem->mkdir(GetShuttleLogDir(), kTRUE))
3244 {
3245 Log("SHUTTLE", Form("SendMail - Can't open directory <%s>", GetShuttleLogDir()));
3246 return kFALSE;
3247 }
3248
3249 } else {
3250 gSystem->FreeDirectory(dir);
3251 }
3252
339adafa 3253 // det experts in to
3254 TString to="";
3835a613 3255 TIter *iterExperts = 0;
339adafa 3256 if (target == kDCSEMail) {
3257 iterExperts = new TIter(fConfig->GetAdmins(AliShuttleConfig::kAmanda));
3258 }
3259 else if (target == kFXSEMail) {
3260 iterExperts = new TIter(fConfig->GetAdmins(system));
3261 }
3835a613 3262 if (iterExperts) {
3263 TObjString *anExpert=0;
3264 while ((anExpert = (TObjString*) iterExperts->Next()))
3265 {
3266 to += Form("%s,", anExpert->GetName());
3267 }
3268 delete iterExperts;
339adafa 3269 }
3835a613 3270
3271 // add subdetector experts
3272 iterExperts = new TIter(fConfig->GetResponsibles(fCurrentDetector));
339adafa 3273 TObjString *anExpert=0;
3274 while ((anExpert = (TObjString*) iterExperts->Next()))
4508ab8b 3275 {
339adafa 3276 to += Form("%s,", anExpert->GetName());
3277 }
3278 delete iterExperts;
3835a613 3279
339adafa 3280 if (to.Length() > 0)
3281 to.Remove(to.Length()-1);
3282 AliDebug(2, Form("to: %s",to.Data()));
4508ab8b 3283
339adafa 3284 if (to.IsNull()) {
3285 Log("SHUTTLE", Form("List of %d responsibles not set!", (Int_t) target));
3286 return kFALSE;
3287 }
3288
3289 // SHUTTLE responsibles in cc
fb40ead4 3290 TString cc="";
3291 TIter iterAdmins(fConfig->GetAdmins(AliShuttleConfig::kGlobal));
3292 TObjString *anAdmin=0;
3293 while ((anAdmin = (TObjString*) iterAdmins.Next()))
3294 {
3295 cc += Form("%s,", anAdmin->GetName());
3296 }
3297 if (cc.Length() > 0)
c19963db 3298 cc.Remove(cc.Length()-1);
fb40ead4 3299 AliDebug(2, Form("cc: %s",to.Data()));
57c1a579 3300
339adafa 3301 // mail body
675f64cd 3302 TString bodyFileName;
3303 bodyFileName.Form("%s/mail.body", GetShuttleLogDir());
3304 gSystem->ExpandPathName(bodyFileName);
3305
3306 ofstream mailBody;
3307 mailBody.open(bodyFileName, ofstream::out);
3308
3309 if (!mailBody.is_open())
3310 {
339adafa 3311 Log("SHUTTLE", Form("Could not open mail body file %s", bodyFileName.Data()));
675f64cd 3312 return kFALSE;
3313 }
3314
675f64cd 3315
339adafa 3316 TString subject;
3317 TString body;
675f64cd 3318
339adafa 3319 if (target == kDCSEMail){
af93e5d7 3320 subject = Form("%s CRITICAL Retrieval of data points for %s FAILED in run %d !",
7bdfe1fa 3321 tmpStr.Data(), fCurrentDetector.Data(), GetCurrentRun());
339adafa 3322 AliDebug(2, Form("subject: %s", subject.Data()));
3323
3324 body = Form("Dear DCS experts, \n\n");
3325 body += Form("SHUTTLE couldn\'t retrieve the data points for detector %s "
3326 "in run %d!!\n\n", fCurrentDetector.Data(), GetCurrentRun());
fb40ead4 3327 }
339adafa 3328 else if (target == kFXSEMail){
af93e5d7 3329 subject = Form("%s CRITICAL FXS communication for %s FAILED in run %d !",
7bdfe1fa 3330 tmpStr.Data(), fCurrentDetector.Data(), GetCurrentRun());
339adafa 3331 AliDebug(2, Form("subject: %s", subject.Data()));
3332 TString sys;
3333 if (system == kDAQ) sys="DAQ";
3334 else if (system == kDCS) sys="DCS";
3335 else if (system == kHLT) sys="HLT";
3336 else return kFALSE;
9c0e2d64 3337 body = Form("Dear %s FXS experts, \n\n",sys.Data());
339adafa 3338 body += Form("SHUTTLE couldn\'t retrieve data from the FXS for detector %s "
9c0e2d64 3339 "in run %d!!\n\n", fCurrentDetector.Data(), GetCurrentRun());
8662f0a8 3340 body += Form("The contacted server was:\nDB: %s\nFXS:%s\n\n", fConfig->GetFXSdbHost(system), fConfig->GetFXSHost(system));
339adafa 3341 }
3342 else {
7bdfe1fa 3343 subject = Form("%s %s Shuttle preprocessor FAILED in run %d (run type = %s)!",
3344 tmpStr.Data(), fCurrentDetector.Data(), GetCurrentRun(), GetRunType());
339adafa 3345 AliDebug(2, Form("subject: %s", subject.Data()));
3346
3347 body = Form("Dear %s expert(s), \n\n", fCurrentDetector.Data());
3348 body += Form("SHUTTLE just detected that your preprocessor "
3349 "failed processing run %d (run type = %s)!!\n\n",
3350 GetCurrentRun(), GetRunType());
3351 }
675f64cd 3352
7d4cf768 3353 body += Form("Please check %s status on the SHUTTLE monitoring page: \n\n",
3354 fCurrentDetector.Data());
b0e53b15 3355 if (fConfig->GetRunMode() == AliShuttleConfig::kTest)
3356 {
8e828d39 3357 body += Form("\thttp://pcalimonitor.cern.ch/shuttle.jsp?time=24 \n\n");
b0e53b15 3358 } else {
8e828d39 3359 body += Form("\thttp://pcalimonitor.cern.ch/shuttle.jsp?instance=PROD&time=24 \n\n");
b0e53b15 3360 }
339adafa 3361
3362
7d4cf768 3363 TString logFolder = "logs";
3364 if (fConfig->GetRunMode() == AliShuttleConfig::kProd)
3365 logFolder += "_PROD";
3366
3367
675f64cd 3368 body += Form("Find the %s log for the current run on \n\n"
29fb87f4 3369 "\thttp://pcalishuttle02.cern.ch/%s/%d/%d/%s.log \n\n",
3ebbf3c4 3370 fCurrentDetector.Data(), logFolder.Data(), GetCurrentRun()/10000,
3371 GetCurrentRun(), fCurrentDetector.Data());
fe6a3257 3372 body += Form("The last 15 lines of %s log file are following:\n\n", fCurrentDetector.Data());
675f64cd 3373
3374 AliDebug(2, Form("Body begin: %s", body.Data()));
3375
3376 mailBody << body.Data();
3377 mailBody.close();
3378 mailBody.open(bodyFileName, ofstream::out | ofstream::app);
3379
3ebbf3c4 3380 TString logFileName = Form("%s/%d/%d/%s.log", GetShuttleLogDir(),
3381 GetCurrentRun()/10000, GetCurrentRun(), fCurrentDetector.Data());
fe6a3257 3382 TString tailCommand = Form("tail -n 15 %s >> %s", logFileName.Data(), bodyFileName.Data());
675f64cd 3383 if (gSystem->Exec(tailCommand.Data()))
3384 {
3385 mailBody << Form("%s log file not found ...\n\n", fCurrentDetector.Data());
3386 }
3387
3388 TString endBody = Form("------------------------------------------------------\n\n");
36c99a6a 3389 endBody += Form("In case of problems please contact the SHUTTLE core team.\n\n");
3390 endBody += "Please do not answer this message directly, it is automatically generated.\n\n";
546242fb 3391 endBody += "Greetings,\n\n \t\t\tthe SHUTTLE\n";
57c1a579 3392
909732f7 3393 AliDebug(2, Form("Body end: %s", endBody.Data()));
57c1a579 3394
3395 mailBody << endBody.Data();
3396
3397 mailBody.close();
3398
3399 // send mail!
3400 TString mailCommand = Form("mail -s \"%s\" -c %s %s < %s",
3401 subject.Data(),
3402 cc.Data(),
3403 to.Data(),
3404 bodyFileName.Data());
909732f7 3405 AliDebug(2, Form("mail command: %s", mailCommand.Data()));
57c1a579 3406
3407 Bool_t result = gSystem->Exec(mailCommand.Data());
3408
3409 return result == 0;
3410}
441b0e9c 3411//______________________________________________________________________________________________
9827400b 3412const char* AliShuttle::GetRunType()
441b0e9c 3413{
9827400b 3414 //
3415 // returns run type read from "run type" logbook
3416 //
441b0e9c 3417
3418 if(!fLogbookEntry) {
3419 AliError("No logbook entry!");
3420 return 0;
3421 }
3422
9827400b 3423 return fLogbookEntry->GetRunType();
441b0e9c 3424}
3425
4859271b 3426//______________________________________________________________________________________________
3427Bool_t AliShuttle::GetHLTStatus()
3428{
3429 // Return HLT status (ON=1 OFF=0)
af76b20d 3430 // Converts the HLT status from the mode string read in the run logbook (not just a bool)
4859271b 3431
3432 if(!fLogbookEntry) {
3433 AliError("No logbook entry!");
3434 return 0;
3435 }
3436
af76b20d 3437 // TODO implement when HLTMode is inserted in run logbook
b4ebc2d5 3438 TString hltMode = fLogbookEntry->GetRunParameter("HLTmode");
af76b20d 3439 TSubString firstChar = hltMode(0,1);
3440 AliDebug(2,Form("First char = %s ",firstChar.Data()));
3441 if (firstChar == "A") {
3442 return kFALSE;
3443 }
3444 else if ((firstChar == "B") || (firstChar == "C") || (firstChar == "D") || (firstChar == "E")) {
3445 return kTRUE;
3446 }
3447 else {
3448 Log("SHUTTLE","Unexpected HLT mode! Returning 0....");
3449 return kFALSE;
3450 }
4859271b 3451}
ca8ea066 3452
3453//______________________________________________________________________________________________
3454const char* AliShuttle::GetTriggerConfiguration()
3455{
3456 // Receives the trigger configuration from the DAQ logbook for the current run
3457
3458 // check connection, if needed reconnect
3459 if (!Connect(3))
3460 return 0;
3461
3462 TString sqlQuery;
3463 sqlQuery.Form("SELECT configFile FROM logbook_trigger_config WHERE run = %d", GetCurrentRun());
3464 TSQLResult* result = fServer[3]->Query(sqlQuery);
3465 if (!result)
3466 {
3467 Log("SHUTTLE", Form("ERROR: Can't execute query <%s>!", sqlQuery.Data()));
3468 return 0;
3469 }
3470
3471 if (result->GetRowCount() == 0)
3472 {
ff0191c7 3473 Log("SHUTTLE", "WARNING: Trigger configuration not found in logbook_trigger_config");
ca8ea066 3474 delete result;
3475 return 0;
3476 }
3477
3478 TSQLRow* row = result->Next();
3479 if (!row)
3480 {
3481 Log("SHUTTLE", "ERROR: Could not receive logbook_trigger_config data");
3482 delete result;
3483 return 0;
3484 }
3485
3486 // static, so that pointer remains valid when it is returned to the calling class
3487 static TString triggerConfig(row->GetField(0));
3488
3489 delete row;
3490 row = 0;
3491
3492 delete result;
3493 result = 0;
3494
3495 Log("SHUTTLE", Form("Found trigger configuration: %s", triggerConfig.Data()));
3496
3497 return triggerConfig;
3498}
3499
2c5f9d06 3500//______________________________________________________________________________________________
3501const char* AliShuttle::GetCTPTimeParams()
3502{
3503 // Receives the CTP time parameters from the DAQ logbook for the current run
3504
3505 // check connection, if needed reconnect
3506 if (!Connect(3))
3507 return 0;
3508
3509 TString sqlQuery;
3510 sqlQuery.Form("SELECT alignmentFile FROM logbook_trigger_config WHERE run = %d", GetCurrentRun());
3511 TSQLResult* result = fServer[3]->Query(sqlQuery);
3512 if (!result)
3513 {
3514 Log("SHUTTLE", Form("ERROR: Can't execute query <%s>!", sqlQuery.Data()));
3515 return 0;
3516 }
3517
3518 if (result->GetRowCount() == 0)
3519 {
ff0191c7 3520 Log("SHUTTLE", "WARNING: CTP time params not found in logbook_trigger_config");
2c5f9d06 3521 delete result;
3522 return 0;
3523 }
3524
3525 TSQLRow* row = result->Next();
3526 if (!row)
3527 {
3528 Log("SHUTTLE", "ERROR: Could not receive logbook_trigger_config data");
3529 delete result;
3530 return 0;
3531 }
3532
3533 // static, so that pointer remains valid when it is returned to the calling class
3534 static TString triggerTimeParams(row->GetField(0));
3535
3536 delete row;
3537 row = 0;
3538
3539 delete result;
3540 result = 0;
3541
a191c286 3542 Log("SHUTTLE", Form("Found trigger time parameters: %s", triggerTimeParams.Data()));
2c5f9d06 3543
3544 return triggerTimeParams;
3545}
3546
d386d623 3547//______________________________________________________________________________________________
b48d2542 3548const char* AliShuttle::GetTriggerDetectorMask()
3549{
3550 // Receives the trigger detector mask from DAQ logbook
3551
3552 // check connection, if needed reconnect
3553 if (!Connect(3))
3554 return 0;
3555
3556 TString sqlQuery;
3557 sqlQuery.Form("SELECT BIN(BIT_OR(inputDetectorMask)) from logbook_trigger_clusters WHERE run = %d;", GetCurrentRun());
3558 TSQLResult* result = fServer[3]->Query(sqlQuery);
3559 if (!result)
3560 {
3561 Log("SHUTTLE", Form("ERROR: Can't execute query <%s>!", sqlQuery.Data()));
3562 return 0;
3563 }
3564
3565 if (result->GetRowCount() == 0)
3566 {
3567 Log("SHUTTLE", "ERROR: Trigger Detector Mask not found in logbook_trigger_clusters");
3568 delete result;
3569 return 0;
3570 }
3571
3572 TSQLRow* row = result->Next();
3573 if (!row)
3574 {
3575 Log("SHUTTLE", "ERROR: Could not receive logbook_trigger_clusters data");
3576 delete result;
3577 return 0;
3578 }
3579
3580 // static, so that pointer remains valid when it is returned to the calling class
3581 static TString triggerDetectorMask(row->GetField(0));
3582
3583 delete row;
3584 row = 0;
3585
3586 delete result;
3587 result = 0;
3588
2e7d9963 3589 Log("SHUTTLE", Form("Found Trigger Detector Mask: %s", triggerDetectorMask.Data()));
b48d2542 3590
3591 return triggerDetectorMask;
3592}
3593
3594//______________________________________________________________________________________________
d386d623 3595void AliShuttle::SetShuttleTempDir(const char* tmpDir)
3596{
9827400b 3597 //
3598 // sets Shuttle temp directory
3599 //
d386d623 3600
3601 fgkShuttleTempDir = gSystem->ExpandPathName(tmpDir);
3602}
3603
3604//______________________________________________________________________________________________
3605void AliShuttle::SetShuttleLogDir(const char* logDir)
3606{
9827400b 3607 //
3608 // sets Shuttle log directory
3609 //
d386d623 3610
3611 fgkShuttleLogDir = gSystem->ExpandPathName(logDir);
3612}
503ff24f 3613//______________________________________________________________________________________________
3614Bool_t AliShuttle::TouchFile()
3615{
3616 //
3617 // touching a file on the grid if run has been DONE
3618 //
3619
3620 if (!gGrid)
3621 {
3622 Log("SHUTTLE",Form("No TGrid connection estabilished!"));
3623 Log("SHUTTLE",Form("Could not touch file for run %i",GetCurrentRun()));
3624 return kFALSE;
3625 }
3626
b8cf5e0f 3627 TString dir;
728290a4 3628 dir.Form("%s%d/SHUTTLE_DONE", fConfig->GetAlienPath(), GetCurrentYear());
b8cf5e0f 3629 // checking whether directory for touch command exists
3630 TString commandLs;
3631 commandLs.Form("ls %s",dir.Data());
3632 TGridResult *resultLs = dynamic_cast<TGridResult*>(gGrid->Command(commandLs));
3633 if (!resultLs){
3634 Log("SHUTTLE",Form("No result for %s command, returning without touching",commandLs.Data()));
3635 return kFALSE;
3636 }
3637 TMap *mapLs = dynamic_cast<TMap*>(resultLs->At(0));
3638 if (!mapLs){
3639 Log("SHUTTLE",Form("No map for %s command, returning without touching",commandLs.Data()));
97d38a5a 3640 delete resultLs;
3641 resultLs = 0x0;
b8cf5e0f 3642 return kFALSE;
3643 }
3644 TObjString *valueLsPath = dynamic_cast<TObjString*>(mapLs->GetValue("path"));
eff13ba0 3645 if (!valueLsPath || (valueLsPath->GetString()).CompareTo(dir)!=1){
b8cf5e0f 3646 Log("SHUTTLE",Form("No directory %s found, creating it",dir.Data()));
3647
3648 // creating the directory
3649
3650 Bool_t boolMkdir = gGrid->Mkdir(dir.Data());
3651 if (!boolMkdir) {
3652 Log("SHUTTLE",Form("Impossible to create dir %s in alien catalogue for run %i!",dir.Data(),GetCurrentRun()));
97d38a5a 3653 delete resultLs;
3654 resultLs = 0x0;
b8cf5e0f 3655 return kFALSE;
3656 }
3657 Log("SHUTTLE",Form("Directory %s successfully created in alien catalogue for run %i",dir.Data(),GetCurrentRun()));
3658 }
3659 else {
3660 Log("SHUTTLE",Form("Directory %s correctly found for run %i",dir.Data(),GetCurrentRun()));
3661 }
3662
97d38a5a 3663 delete resultLs;
3664 resultLs = 0x0;
3665
503ff24f 3666 TString command;
b8cf5e0f 3667 command.Form("touch %s/%i", dir.Data(), GetCurrentRun());
e660ff60 3668 Log("SHUTTLE", Form("Creating entry in file catalog: %s", command.Data()));
503ff24f 3669 TGridResult *resultTouch = dynamic_cast<TGridResult*>(gGrid->Command(command));
b8cf5e0f 3670 if (!resultTouch){
3671 Log("SHUTTLE",Form("No result for touching command, returning without touching for run %i",GetCurrentRun()));
3672 return kFALSE;
503ff24f 3673 }
b8cf5e0f 3674 TMap *mapTouch = dynamic_cast<TMap*>(resultTouch->At(0));
3675 if (!mapTouch){
3676 Log("SHUTTLE",Form("No map for touching command, returning without touching for run %i",GetCurrentRun()));
97d38a5a 3677 delete resultTouch;
3678 resultTouch = 0x0;
b8cf5e0f 3679 return kFALSE;
3680 }
3681 TObjString *valueTouch = dynamic_cast<TObjString*>(mapTouch->GetValue("__result__"));
3682 if (!valueTouch){
3683 Log("SHUTTLE",Form("No value for \"__result__\" key set in the map for touching command, returning without touching for run %i",GetCurrentRun()));
97d38a5a 3684 delete resultTouch;
3685 resultTouch = 0x0;
b8cf5e0f 3686 return kFALSE;
503ff24f 3687 }
b8cf5e0f 3688 if (valueTouch->GetString()!="1"){
3689 Log("SHUTTLE",Form("Failing the touching command, returning without touching for run %i",GetCurrentRun()));
97d38a5a 3690 delete resultTouch;
3691 resultTouch = 0x0;
b8cf5e0f 3692 return kFALSE;
3693 }
97d38a5a 3694 delete resultTouch;
3695 resultTouch = 0x0;
98400bd9 3696 Log("SHUTTLE", "Sucessfully touched the file");
b8cf5e0f 3697 return kTRUE;
503ff24f 3698}
2eb98462 3699//______________________________________________________________________________________________
f100adeb 3700UInt_t AliShuttle::GetStartTimeDCSQuery()
2eb98462 3701{
3702 // Return Start Time for the DCS query
3703 //
3704 // The call is delegated to AliShuttleInterface
3705
3d24ad5a 3706 return GetCurrentStartTime()-fConfig->GetDCSQueryOffset();
2eb98462 3707}
3708//______________________________________________________________________________________________
f100adeb 3709UInt_t AliShuttle::GetEndTimeDCSQuery()
2eb98462 3710{
3711 // Return End Time for the DCS query
3712 //
3713 // The call is delegated to AliShuttleInterface
3714
3715 return GetCurrentEndTime()+fConfig->GetDCSQueryOffset();
3716}
503ff24f 3717