]> 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
persistent CommonHist (still need some improvements)
[u/mrichter/AliRoot.git] / SHUTTLE / AliShuttle.cxx
CommitLineData
73abe331 1/**************************************************************************
2 * Copyright(c) 1998-1999, ALICE Experiment at CERN, All rights reserved. *
3 * *
4 * Author: The ALICE Off-line Project. *
5 * Contributors are mentioned in the code where appropriate. *
6 * *
7 * Permission to use, copy, modify and distribute this software and its *
8 * documentation strictly for non-commercial purposes is hereby granted *
9 * without fee, provided that the above copyright notice appears in all *
10 * copies and that both the copyright notice and this permission notice *
11 * appear in the supporting documentation. The authors make no claims *
12 * about the suitability of this software for any purpose. It is *
13 * provided "as is" without express or implied warranty. *
14 **************************************************************************/
15
ff3781ad 16/* $Id$ */
73abe331 17
18//
19// This class is the main manager for AliShuttle.
20// It organizes the data retrieval from DCS and call the
b948db8d 21// interface methods of AliPreprocessor.
73abe331 22// For every detector in AliShuttleConfgi (see AliShuttleConfig),
23// data for its set of aliases is retrieved. If there is registered
b948db8d 24// AliPreprocessor for this detector then it will be used
25// accroding to the schema (see AliPreprocessor).
26// If there isn't registered AliPreprocessor than the retrieved
73abe331 27// data is stored automatically to the undelying AliCDBStorage.
28// For detSpec is used the alias name.
29//
30
31#include "AliShuttle.h"
32
33#include "AliCDBManager.h"
34#include "AliCDBStorage.h"
35#include "AliCDBId.h"
84090f85 36#include "AliCDBRunRange.h"
37#include "AliCDBPath.h"
5164a766 38#include "AliCDBEntry.h"
73abe331 39#include "AliShuttleConfig.h"
eba76848 40#include "DCSClient/AliDCSClient.h"
73abe331 41#include "AliLog.h"
b948db8d 42#include "AliPreprocessor.h"
5164a766 43#include "AliShuttleStatus.h"
2bb7b766 44#include "AliShuttleLogbookEntry.h"
73abe331 45
57f50b3c 46#include <TSystem.h>
58bc3020 47#include <TObject.h>
b948db8d 48#include <TString.h>
57f50b3c 49#include <TTimeStamp.h>
73abe331 50#include <TObjString.h>
57f50b3c 51#include <TSQLServer.h>
52#include <TSQLResult.h>
53#include <TSQLRow.h>
cb343cfd 54#include <TMutex.h>
9827400b 55#include <TSystemDirectory.h>
56#include <TSystemFile.h>
a986b218 57#include <TFile.h>
9827400b 58#include <TGrid.h>
59#include <TGridResult.h>
503ff24f 60#include <TMap.h>
73abe331 61
e7f62f16 62#include <TMonaLisaWriter.h>
63
5164a766 64#include <fstream>
65
cb343cfd 66#include <sys/types.h>
67#include <sys/wait.h>
68
1bd183ba 69#include <signal.h>
70
73abe331 71ClassImp(AliShuttle)
72
b948db8d 73//______________________________________________________________________________________________
74AliShuttle::AliShuttle(const AliShuttleConfig* config,
75 UInt_t timeout, Int_t retries):
4f0ab988 76fConfig(config),
77fTimeout(timeout), fRetries(retries),
78fPreprocessorMap(),
2bb7b766 79fLogbookEntry(0),
eba76848 80fCurrentDetector(),
926cbc0e 81fFirstProcessing(0),
339adafa 82fFXSError(-1),
85a80aa9 83fStatusEntry(0),
cb343cfd 84fMonitoringMutex(0),
eba76848 85fLastActionTime(0),
e7f62f16 86fLastAction(),
9827400b 87fMonaLisa(0),
88fTestMode(kNone),
ffa29e93 89fReadTestMode(kFALSE),
90fOutputRedirected(kFALSE)
73abe331 91{
92 //
93 // config: AliShuttleConfig used
73abe331 94 // timeout: timeout used for AliDCSClient connection
95 // retries: the number of retries in case of connection error.
96 //
97
57f50b3c 98 if (!fConfig->IsValid()) AliFatal("********** !!!!! Invalid configuration !!!!! **********");
be48e3ea 99 for(int iSys=0;iSys<4;iSys++) {
57f50b3c 100 fServer[iSys]=0;
be48e3ea 101 if (iSys < 3)
2c15234c 102 fFXSlist[iSys].SetOwner(kTRUE);
57f50b3c 103 }
2bb7b766 104 fPreprocessorMap.SetOwner(kTRUE);
be48e3ea 105
106 for (UInt_t iDet=0; iDet<NDetectors(); iDet++)
107 fFirstUnprocessed[iDet] = kFALSE;
108
cb343cfd 109 fMonitoringMutex = new TMutex();
58bc3020 110}
111
b948db8d 112//______________________________________________________________________________________________
57f50b3c 113AliShuttle::~AliShuttle()
58bc3020 114{
9827400b 115 //
116 // destructor
117 //
58bc3020 118
b948db8d 119 fPreprocessorMap.DeleteAll();
be48e3ea 120 for(int iSys=0;iSys<4;iSys++)
57f50b3c 121 if(fServer[iSys]) {
122 fServer[iSys]->Close();
123 delete fServer[iSys];
eba76848 124 fServer[iSys] = 0;
57f50b3c 125 }
2bb7b766 126
127 if (fStatusEntry){
128 delete fStatusEntry;
129 fStatusEntry = 0;
130 }
cb343cfd 131
132 if (fMonitoringMutex)
133 {
134 delete fMonitoringMutex;
135 fMonitoringMutex = 0;
136 }
73abe331 137}
138
b948db8d 139//______________________________________________________________________________________________
57f50b3c 140void AliShuttle::RegisterPreprocessor(AliPreprocessor* preprocessor)
58bc3020 141{
73abe331 142 //
b948db8d 143 // Registers new AliPreprocessor.
73abe331 144 // It uses GetName() for indentificator of the pre processor.
145 // The pre processor is registered it there isn't any other
146 // with the same identificator (GetName()).
147 //
148
eba76848 149 const char* detName = preprocessor->GetName();
150 if(GetDetPos(detName) < 0)
151 AliFatal(Form("********** !!!!! Invalid detector name: %s !!!!! **********", detName));
152
153 if (fPreprocessorMap.GetValue(detName)) {
154 AliWarning(Form("AliPreprocessor %s is already registered!", detName));
73abe331 155 return;
156 }
157
eba76848 158 fPreprocessorMap.Add(new TObjString(detName), preprocessor);
73abe331 159}
b948db8d 160//______________________________________________________________________________________________
3301427a 161Bool_t AliShuttle::Store(const AliCDBPath& path, TObject* object,
84090f85 162 AliCDBMetaData* metaData, Int_t validityStart, Bool_t validityInfinite)
73abe331 163{
9827400b 164 // Stores a CDB object in the storage for offline reconstruction. Objects that are not needed for
165 // offline reconstruction, but should be stored anyway (e.g. for debugging) should NOT be stored
166 // using this function. Use StoreReferenceData instead!
167 // It calls StoreLocally function which temporarily stores the data locally; when the preprocessor
168 // finishes the data are transferred to the main storage (Grid).
b948db8d 169
3301427a 170 return StoreLocally(fgkLocalCDB, path, object, metaData, validityStart, validityInfinite);
84090f85 171}
172
173//______________________________________________________________________________________________
3301427a 174Bool_t AliShuttle::StoreReferenceData(const AliCDBPath& path, TObject* object, AliCDBMetaData* metaData)
84090f85 175{
9827400b 176 // Stores a CDB object in the storage for reference data. This objects will not be available during
177 // offline reconstrunction. Use this function for reference data only!
178 // It calls StoreLocally function which temporarily stores the data locally; when the preprocessor
179 // finishes the data are transferred to the main storage (Grid).
85a80aa9 180
3301427a 181 return StoreLocally(fgkLocalRefStorage, path, object, metaData);
85a80aa9 182}
183
184//______________________________________________________________________________________________
3301427a 185Bool_t AliShuttle::StoreLocally(const TString& localUri,
85a80aa9 186 const AliCDBPath& path, TObject* object, AliCDBMetaData* metaData,
187 Int_t validityStart, Bool_t validityInfinite)
188{
9827400b 189 // Store object temporarily in local storage. Parameters are passed by Store and StoreReferenceData functions.
190 // when the preprocessor finishes the data are transferred to the main storage (Grid).
191 // The parameters are:
192 // 1) Uri of the backup storage (Local)
193 // 2) the object's path.
194 // 3) the object to be stored
195 // 4) the metaData to be associated with the object
196 // 5) the validity start run number w.r.t. the current run,
197 // if the data is valid only for this run leave the default 0
198 // 6) specifies if the calibration data is valid for infinity (this means until updated),
199 // typical for calibration runs, the default is kFALSE
200 //
201 // returns 0 if fail, 1 otherwise
84090f85 202
9827400b 203 if (fTestMode & kErrorStorage)
204 {
205 Log(fCurrentDetector, "StoreLocally - In TESTMODE - Simulating error while storing locally");
206 return kFALSE;
207 }
208
3301427a 209 const char* cdbType = (localUri == fgkLocalCDB) ? "CDB" : "Reference";
2bb7b766 210
85a80aa9 211 Int_t firstRun = GetCurrentRun() - validityStart;
84090f85 212 if(firstRun < 0) {
9827400b 213 AliWarning("First valid run happens to be less than 0! Setting it to 0.");
84090f85 214 firstRun=0;
215 }
216
217 Int_t lastRun = -1;
218 if(validityInfinite) {
219 lastRun = AliCDBRunRange::Infinity();
220 } else {
221 lastRun = GetCurrentRun();
222 }
223
3301427a 224 // Version is set to current run, it will be used later to transfer data to Grid
225 AliCDBId id(path, firstRun, lastRun, GetCurrentRun(), -1);
2bb7b766 226
227 if(! dynamic_cast<TObjString*> (metaData->GetProperty("RunUsed(TObjString)"))){
228 TObjString runUsed = Form("%d", GetCurrentRun());
9e080f92 229 metaData->SetProperty("RunUsed(TObjString)", runUsed.Clone());
2bb7b766 230 }
84090f85 231
3301427a 232 Bool_t result = kFALSE;
84090f85 233
3301427a 234 if (!(AliCDBManager::Instance()->GetStorage(localUri))) {
235 Log("SHUTTLE", Form("StoreLocally - Cannot activate local %s storage", cdbType));
84090f85 236 } else {
3301427a 237 result = AliCDBManager::Instance()->GetStorage(localUri)
84090f85 238 ->Put(object, id, metaData);
239 }
240
241 if(!result) {
242
9827400b 243 Log(fCurrentDetector, Form("StoreLocally - Can't store object <%s>!", id.ToString().Data()));
3301427a 244 }
2bb7b766 245
3301427a 246 return result;
247}
84090f85 248
3301427a 249//______________________________________________________________________________________________
250Bool_t AliShuttle::StoreOCDB()
251{
9827400b 252 //
253 // Called when preprocessor ends successfully or when previous storage attempt failed (kStoreError status)
254 // Calls underlying StoreOCDB(const char*) function twice, for OCDB and Reference storage.
255 // Then calls StoreRefFilesToGrid to store reference files.
256 //
257
5e993b6f 258 UpdateShuttleStatus(AliShuttleStatus::kStoreStarted);
56d3e9e4 259
9827400b 260 if (fTestMode & kErrorGrid)
261 {
262 Log("SHUTTLE", "StoreOCDB - In TESTMODE - Simulating error while storing in the Grid");
263 Log(fCurrentDetector, "StoreOCDB - In TESTMODE - Simulating error while storing in the Grid");
264 return kFALSE;
265 }
266
c88ad5db 267 Log("SHUTTLE","StoreOCDB - Storing OCDB data ...");
5e993b6f 268 Int_t resultCDB = StoreOCDB(fgkMainCDB);
86aa42c3 269
c88ad5db 270 Log("SHUTTLE","StoreOCDB - Storing reference data ...");
5e993b6f 271 Int_t resultRef = StoreOCDB(fgkMainRefStorage);
9827400b 272
c88ad5db 273 Log("SHUTTLE","StoreOCDB - Storing reference files ...");
274 Bool_t resultRefFiles = CopyFilesToGrid("reference");
275
276 Bool_t resultMetadata = kTRUE;
277 if(fCurrentDetector == "GRP")
278 {
279 Log("StoreOCDB - SHUTTLE","Storing Run Metadata file ...");
280 resultMetadata = CopyFilesToGrid("metadata");
281 }
9827400b 282
5e993b6f 283 Int_t storeResult = 0;
284
285 if (resultCDB < 0 || resultRef < 0 || resultRefFiles == kFALSE || resultMetadata == kFALSE)
286 storeResult = -1;
287 else if (resultCDB > 0 || resultRef > 0)
288 storeResult = 1;
289
290 if (storeResult < 0)
291 {
292 Log("SHUTTLE",
293 Form("\t\t\t****** run %d - %s: STORAGE ERROR ******",
294 GetCurrentRun(), fCurrentDetector.Data()));
295 UpdateShuttleStatus(AliShuttleStatus::kStoreError);
296 }
297 else if (storeResult > 0)
298 {
299 Log("SHUTTLE",
300 Form("\t\t\t****** run %d - %s: STORAGE DELAYED ******",
301 GetCurrentRun(), fCurrentDetector.Data()));
302 UpdateShuttleStatus(AliShuttleStatus::kStoreDelayed);
303 }
304 else if (storeResult == 0)
305 {
306 Log("SHUTTLE",
307 Form("\t\t\t****** run %d - %s: DONE ******",
308 GetCurrentRun(), fCurrentDetector.Data()));
309 UpdateShuttleStatus(AliShuttleStatus::kDone);
310 UpdateShuttleLogbook(fCurrentDetector, "DONE");
311 }
312
313 return (storeResult == 0);
3301427a 314}
315
316//______________________________________________________________________________________________
5e993b6f 317Int_t AliShuttle::StoreOCDB(const TString& gridURI)
3301427a 318{
319 //
320 // Called by StoreOCDB(), performs actual storage to the main OCDB and reference storages (Grid)
321 //
5e993b6f 322 // Return code:
323 // -2 initialization error
324 // -1 storage error
325 // 0 success
326 // 1 storage delayed (e.g. previous unprocessed runs)
327 //
3301427a 328
329 TObjArray* gridIds=0;
330
331 Bool_t result = kTRUE;
5e993b6f 332 Bool_t delayed = kFALSE;
3301427a 333
334 const char* type = 0;
335 TString localURI;
336 if(gridURI == fgkMainCDB) {
337 type = "OCDB";
338 localURI = fgkLocalCDB;
339 } else if(gridURI == fgkMainRefStorage) {
340 type = "reference";
341 localURI = fgkLocalRefStorage;
342 } else {
343 AliError(Form("Invalid storage URI: %s", gridURI.Data()));
5e993b6f 344 return -2;
3301427a 345 }
346
347 AliCDBManager* man = AliCDBManager::Instance();
348
349 AliCDBStorage *gridSto = man->GetStorage(gridURI);
350 if(!gridSto) {
351 Log("SHUTTLE",
352 Form("StoreOCDB - cannot activate main %s storage", type));
5e993b6f 353 return -2;
3301427a 354 }
355
356 gridIds = gridSto->GetQueryCDBList();
357
358 // get objects previously stored in local CDB
359 AliCDBStorage *localSto = man->GetStorage(localURI);
360 if(!localSto) {
361 Log("SHUTTLE",
362 Form("StoreOCDB - cannot activate local %s storage", type));
5e993b6f 363 return -2;
3301427a 364 }
365 AliCDBPath aPath(GetOfflineDetName(fCurrentDetector.Data()),"*","*");
366 // Local objects were stored with current run as Grid version!
367 TList* localEntries = localSto->GetAll(aPath.GetPath(), GetCurrentRun(), GetCurrentRun());
368 localEntries->SetOwner(1);
369
370 // loop on local stored objects
371 TIter localIter(localEntries);
372 AliCDBEntry *aLocEntry = 0;
373 while((aLocEntry = dynamic_cast<AliCDBEntry*> (localIter.Next()))){
374 aLocEntry->SetOwner(1);
375 AliCDBId aLocId = aLocEntry->GetId();
376 aLocEntry->SetVersion(-1);
377 aLocEntry->SetSubVersion(-1);
6c0c9948 378
379 Log(fCurrentDetector.Data(), Form("Attempting to store %s", aLocId.ToString().Data()));
3301427a 380
381 // If local object is valid up to infinity we store it only if it is
382 // the first unprocessed run!
383 if (aLocId.GetLastRun() == AliCDBRunRange::Infinity() &&
384 !fFirstUnprocessed[GetDetPos(fCurrentDetector)])
385 {
386 Log("SHUTTLE", Form("StoreOCDB - %s: object %s has validity infinite but "
387 "there are previous unprocessed runs!",
388 fCurrentDetector.Data(), aLocId.GetPath().Data()));
5e993b6f 389 Log(fCurrentDetector.Data(), Form("StoreOCDB - %s: object %s has validity infinite but "
390 "there are previous unprocessed runs!",
391 fCurrentDetector.Data(), aLocId.GetPath().Data()));
392 delayed = kTRUE;
3301427a 393 continue;
394 }
395
396 // loop on Grid valid Id's
397 Bool_t store = kTRUE;
398 TIter gridIter(gridIds);
399 AliCDBId* aGridId = 0;
6c0c9948 400 while ((aGridId = dynamic_cast<AliCDBId*> (gridIter.Next()))) {
401 if (aGridId->GetPath() != aLocId.GetPath())
402 continue;
3301427a 403 // skip all objects valid up to infinity
6c0c9948 404 if (aGridId->GetLastRun() == AliCDBRunRange::Infinity())
405 continue;
406
3301427a 407 // if we get here, it means there's already some more recent object stored on Grid!
6c0c9948 408 Log(fCurrentDetector.Data(),
409 Form("StoreOCDB - A more recent object already exists in %s storage: <%s>",
410 type, aGridId->ToString().Data()));
411
3301427a 412 store = kFALSE;
413 break;
414 }
415
6c0c9948 416 Bool_t storeOk = kFALSE;
417 if (store)
418 {
419 Log(fCurrentDetector.Data(), Form("Prechecks succeeded. Ready to store %s", aLocId.ToString().Data()));
420 storeOk = gridSto->Put(aLocEntry);
421 if (storeOk) {
3301427a 422 Log("SHUTTLE",
6c0c9948 423 Form("StoreOCDB - Object <%s> successfully put into %s storage",
424 aLocId.ToString().Data(), type));
2d9019b4 425 Log(fCurrentDetector.Data(),
426 Form("StoreOCDB - Object <%s> successfully put into %s storage",
6c0c9948 427 aLocId.ToString().Data(), type));
428 } else {
429 Log("SHUTTLE",
430 Form("StoreOCDB - Grid %s storage of object <%s> failed",
431 type, aLocId.ToString().Data()));
432 Log(fCurrentDetector.Data(),
433 Form("StoreOCDB - Grid %s storage of object <%s> failed",
434 type, aLocId.ToString().Data()));
435 result = kFALSE;
3301427a 436 }
6c0c9948 437 }
438
439 if (!store || storeOk) {
440 // removing local file...
3301427a 441 TString filename;
442 localSto->IdToFilename(aLocId, filename);
c88ad5db 443 Log("SHUTTLE", Form("StoreOCDB - Removing local file %s", filename.Data()));
3301427a 444 RemoveFile(filename.Data());
b948db8d 445 }
446 }
3301427a 447 localEntries->Clear();
2bb7b766 448
5e993b6f 449 Int_t returnCode = 0;
450
451 if (result == kFALSE)
452 returnCode = -1;
453 else if (delayed != kFALSE)
454 returnCode = 1;
455
456 Log("SHUTTLE", Form("StoreOCDB - Returning with %d (result = %d, delayed = %d)", returnCode, result, delayed));
457 Log(fCurrentDetector.Data(), Form("StoreOCDB - Returning with %d (result = %d, delayed = %d)", returnCode, result, delayed));
458
459 return returnCode;
3301427a 460}
461
546242fb 462//______________________________________________________________________________________________
463Bool_t AliShuttle::CleanReferenceStorage(const char* detector)
464{
2d9019b4 465 // clears the directory used to store reference files of a given subdetector
546242fb 466
467 AliCDBManager* man = AliCDBManager::Instance();
468 AliCDBStorage* sto = man->GetStorage(fgkLocalRefStorage);
2d9019b4 469 TString localBaseFolder = sto->GetBaseFolder();
470
471 TString targetDir = GetRefFilePrefix(localBaseFolder.Data(), detector);
472
d524ade6 473 Log("SHUTTLE", Form("CleanReferenceStorage - Cleaning %s", targetDir.Data()));
2d9019b4 474
475 TString begin;
476 begin.Form("%d_", GetCurrentRun());
477
478 TSystemDirectory* baseDir = new TSystemDirectory("/", targetDir);
479 if (!baseDir)
480 return kTRUE;
481
482 TList* dirList = baseDir->GetListOfFiles();
483 delete baseDir;
484
485 if (!dirList) return kTRUE;
486
487 if (dirList->GetEntries() < 3)
488 {
489 delete dirList;
490 return kTRUE;
491 }
492
493 Int_t nDirs = 0, nDel = 0;
494 TIter dirIter(dirList);
495 TSystemFile* entry = 0;
546242fb 496
2d9019b4 497 Bool_t success = kTRUE;
546242fb 498
2d9019b4 499 while ((entry = dynamic_cast<TSystemFile*> (dirIter.Next())))
500 {
501 if (entry->IsDirectory())
502 continue;
503
504 TString fileName(entry->GetName());
505 if (!fileName.BeginsWith(begin))
506 continue;
507
508 nDirs++;
509
510 // delete file
511 Int_t result = gSystem->Unlink(fileName.Data());
512
513 if (result)
514 {
d524ade6 515 Log("SHUTTLE", Form("CleanReferenceStorage - Could not delete file %s!", fileName.Data()));
2d9019b4 516 success = kFALSE;
517 } else {
518 nDel++;
519 }
520 }
521
522 if(nDirs > 0)
523 Log("SHUTTLE", Form("CleanReferenceStorage - %d (over %d) reference files in folder %s were deleted.",
524 nDel, nDirs, targetDir.Data()));
525
526
527 delete dirList;
528 return success;
529
530
531
532
533
546242fb 534
535 Int_t result = gSystem->GetPathInfo(targetDir, 0, (Long64_t*) 0, 0, 0);
536 if (result == 0)
537 {
538 // delete directory
d524ade6 539 result = gSystem->Exec(Form("rm -rf %s", targetDir.Data()));
546242fb 540 if (result != 0)
541 {
d524ade6 542 Log("SHUTTLE", Form("CleanReferenceStorage - Could not clean directory %s", targetDir.Data()));
546242fb 543 return kFALSE;
544 }
545 }
546
547 result = gSystem->mkdir(targetDir, kTRUE);
548 if (result != 0)
549 {
c88ad5db 550 Log("SHUTTLE", Form("CleanReferenceStorage - Error creating base directory %s", targetDir.Data()));
546242fb 551 return kFALSE;
552 }
553
554 return kTRUE;
555}
556
9827400b 557//______________________________________________________________________________________________
558Bool_t AliShuttle::StoreReferenceFile(const char* detector, const char* localFile, const char* gridFileName)
559{
560 //
3c2a21c8 561 // Stores reference file directly (without opening it). This function stores the file locally.
9827400b 562 //
3c2a21c8 563 // The file is stored under the following location:
564 // <base folder of local reference storage>/<DET>/<RUN#>_<gridFileName>
565 // where <gridFileName> is the second parameter given to the function
566 //
9827400b 567
568 if (fTestMode & kErrorStorage)
569 {
570 Log(fCurrentDetector, "StoreReferenceFile - In TESTMODE - Simulating error while storing locally");
571 return kFALSE;
572 }
573
574 AliCDBManager* man = AliCDBManager::Instance();
575 AliCDBStorage* sto = man->GetStorage(fgkLocalRefStorage);
576
577 TString localBaseFolder = sto->GetBaseFolder();
578
d524ade6 579 TString target = GetRefFilePrefix(localBaseFolder.Data(), detector);
580 target.Append(Form("/%d_%s", GetCurrentRun(), gridFileName));
9827400b 581
d524ade6 582 return CopyFileLocally(localFile, target);
c88ad5db 583}
584
585//______________________________________________________________________________________________
586Bool_t AliShuttle::StoreRunMetadataFile(const char* localFile, const char* gridFileName)
587{
588 //
589 // Stores Run metadata file to the Grid, in the run folder
590 //
591 // Only GRP can call this function.
592
593 if (fTestMode & kErrorStorage)
594 {
595 Log(fCurrentDetector, "StoreRunMetaDataFile - In TESTMODE - Simulating error while storing locally");
596 return kFALSE;
597 }
598
599 AliCDBManager* man = AliCDBManager::Instance();
600 AliCDBStorage* sto = man->GetStorage(fgkLocalRefStorage);
601
602 TString localBaseFolder = sto->GetBaseFolder();
603
604 // Build Run level folder
955d2abc 605 // folder = /alice/data/year/lhcPeriod/runNb/raw
c88ad5db 606
c88ad5db 607
675f64cd 608 TString lhcPeriod = GetLHCPeriod();
c88ad5db 609 if (lhcPeriod.Length() == 0)
610 {
611 Log("SHUTTLE","StoreRunMetaDataFile - LHCPeriod not found in logbook!");
612 return 0;
613 }
a1c5723b 614
615 // TODO partitions with one detector only write data into LHCperiod_DET
2694f253 616 TString partition = GetRunParameter("detector");
a1c5723b 617
618 if (partition.Length() > 0 && partition != "ALICE")
619 {
620 lhcPeriod.Append(Form("_%s", partition.Data()));
621 Log(fCurrentDetector, Form("Run data tags merged file will be written in %s",
622 lhcPeriod.Data()));
623 }
8884d71e 624
b8cf5e0f 625 TString target = Form("%s/GRP/RunMetadata%s%d/%s/%09d/raw/%s",
626 localBaseFolder.Data(), fConfig->GetAlienPath(), GetCurrentYear(),
d524ade6 627 lhcPeriod.Data(), GetCurrentRun(), gridFileName);
c88ad5db 628
d524ade6 629 return CopyFileLocally(localFile, target);
c88ad5db 630}
631
632//______________________________________________________________________________________________
d524ade6 633Bool_t AliShuttle::CopyFileLocally(const char* localFile, const TString& target)
c88ad5db 634{
635 //
636 // Stores file locally. Called by StoreReferenceFile and StoreRunMetadataFile
d524ade6 637 // Files are temporarily stored in the local reference storage. When the preprocessor
638 // finishes, the Shuttle calls CopyFilesToGrid to transfer the files to AliEn
639 // (in reference or run level folders)
c88ad5db 640 //
641
d524ade6 642 TString targetDir(target(0, target.Last('/')));
643
644 //try to open base dir folder, if it does not exist
2d9019b4 645 void* dir = gSystem->OpenDirectory(targetDir.Data());
646 if (dir == NULL) {
647 if (gSystem->mkdir(targetDir.Data(), kTRUE)) {
a1c5723b 648 Log("SHUTTLE", Form("CopyFileLocally - Can't open directory <%s>", targetDir.Data()));
2d9019b4 649 return kFALSE;
650 }
651
652 } else {
653 gSystem->FreeDirectory(dir);
654 }
9827400b 655
7d43a416 656 Int_t result = 0;
657
658 result = gSystem->GetPathInfo(localFile, 0, (Long64_t*) 0, 0, 0);
9827400b 659 if (result)
660 {
a1c5723b 661 Log("SHUTTLE", Form("CopyFileLocally - %s does not exist", localFile));
546242fb 662 return kFALSE;
9827400b 663 }
546242fb 664
7d43a416 665 result = gSystem->GetPathInfo(target, 0, (Long64_t*) 0, 0, 0);
666 if (!result)
667 {
a1c5723b 668 Log("SHUTTLE", Form("CopyFileLocally - target file %s already exist, removing...", target.Data()));
7d43a416 669 if (gSystem->Unlink(target.Data()))
670 {
a1c5723b 671 Log("SHUTTLE", Form("CopyFileLocally - Could not remove existing target file %s!", target.Data()));
7d43a416 672 return kFALSE;
673 }
674 }
675
9827400b 676 result = gSystem->CopyFile(localFile, target);
677
678 if (result == 0)
679 {
a1c5723b 680 Log("SHUTTLE", Form("CopyFileLocally - File %s stored locally to %s", localFile, target.Data()));
9827400b 681 return kTRUE;
682 }
683 else
684 {
a1c5723b 685 Log("SHUTTLE", Form("CopyFileLocally - Could not store file %s to %s! Error code = %d",
546242fb 686 localFile, target.Data(), result));
9827400b 687 return kFALSE;
688 }
c88ad5db 689
690
691
9827400b 692}
693
694//______________________________________________________________________________________________
c88ad5db 695Bool_t AliShuttle::CopyFilesToGrid(const char* type)
9827400b 696{
697 //
c88ad5db 698 // Transfers local files to the Grid. Local files can be reference files
699 // or run metadata file (from GRP only).
9827400b 700 //
c88ad5db 701 // According to the type (ref, metadata) the files are stored under the following location:
702 // ref --> <base folder of reference storage>/<DET>/<RUN#>_<gridFileName>
703 // metadata --> <run data folder>/<MetadataFileName>
86aa42c3 704 //
c88ad5db 705
9827400b 706 AliCDBManager* man = AliCDBManager::Instance();
707 AliCDBStorage* sto = man->GetStorage(fgkLocalRefStorage);
708 if (!sto)
709 return kFALSE;
710 TString localBaseFolder = sto->GetBaseFolder();
9827400b 711
c88ad5db 712 TString dir;
713 TString alienDir;
9827400b 714 TString begin;
9827400b 715
c88ad5db 716 if (strcmp(type, "reference") == 0)
717 {
718 dir = GetRefFilePrefix(localBaseFolder.Data(), fCurrentDetector.Data());
719 AliCDBStorage* gridSto = man->GetStorage(fgkMainRefStorage);
720 if (!gridSto)
721 return kFALSE;
722 TString gridBaseFolder = gridSto->GetBaseFolder();
723 alienDir = GetRefFilePrefix(gridBaseFolder.Data(), fCurrentDetector.Data());
724 begin = Form("%d_", GetCurrentRun());
725 }
726 else if (strcmp(type, "metadata") == 0)
727 {
c88ad5db 728
675f64cd 729 TString lhcPeriod = GetLHCPeriod();
c88ad5db 730
731 if (lhcPeriod.Length() == 0)
732 {
733 Log("SHUTTLE","CopyFilesToGrid - LHCPeriod not found in logbook!");
734 return 0;
735 }
736
a1c5723b 737 // TODO partitions with one detector only write data into LHCperiod_DET
2694f253 738 TString partition = GetRunParameter("detector");
a1c5723b 739
740 if (partition.Length() > 0 && partition != "ALICE")
741 {
742 lhcPeriod.Append(Form("_%s", partition.Data()));
743 }
744
b8cf5e0f 745 dir = Form("%s/GRP/RunMetadata%s%d/%s/%09d/raw",
746 localBaseFolder.Data(), fConfig->GetAlienPath(), GetCurrentYear(),
c88ad5db 747 lhcPeriod.Data(), GetCurrentRun());
b8cf5e0f 748 alienDir = dir(dir.Index(fConfig->GetAlienPath()), dir.Length());
675f64cd 749
c88ad5db 750 begin = "";
751 }
752 else
753 {
754 Log("SHUTTLE", "CopyFilesToGrid - Unexpected: type label must be reference or metadata!");
755 return kFALSE;
756 }
757
9827400b 758 TSystemDirectory* baseDir = new TSystemDirectory("/", dir);
3d8bc902 759 if (!baseDir)
760 return kTRUE;
761
2d9019b4 762 TList* dirList = baseDir->GetListOfFiles();
763 delete baseDir;
764
765 if (!dirList) return kTRUE;
766
767 if (dirList->GetEntries() < 3)
3d8bc902 768 {
2d9019b4 769 delete dirList;
9827400b 770 return kTRUE;
3d8bc902 771 }
2d9019b4 772
546242fb 773 if (!gGrid)
774 {
c88ad5db 775 Log("SHUTTLE", "CopyFilesToGrid - Connection to Grid failed: Cannot continue!");
2d9019b4 776 delete dirList;
546242fb 777 return kFALSE;
778 }
779
2d9019b4 780 Int_t nDirs = 0, nTransfer = 0;
781 TIter dirIter(dirList);
782 TSystemFile* entry = 0;
783
9827400b 784 Bool_t success = kTRUE;
3d8bc902 785 Bool_t first = kTRUE;
9827400b 786
2d9019b4 787 while ((entry = dynamic_cast<TSystemFile*> (dirIter.Next())))
788 {
9827400b 789 if (entry->IsDirectory())
790 continue;
791
792 TString fileName(entry->GetName());
793 if (!fileName.BeginsWith(begin))
794 continue;
795
2d9019b4 796 nDirs++;
797
3d8bc902 798 if (first)
799 {
800 first = kFALSE;
c88ad5db 801 // check that folder exists, otherwise create it
3d8bc902 802 TGridResult* result = gGrid->Ls(alienDir.Data(), "a");
803
804 if (!result)
2d9019b4 805 {
806 delete dirList;
3d8bc902 807 return kFALSE;
2d9019b4 808 }
3d8bc902 809
546242fb 810 if (!result->GetFileName(1)) // TODO: It looks like element 0 is always 0!!
3d8bc902 811 {
675f64cd 812 // TODO It does not work currently! Bug in TAliEn::Mkdir
813 // TODO Manually fixed in local root v5-16-00
c88ad5db 814 if (!gGrid->Mkdir(alienDir.Data(),"-p",0))
3d8bc902 815 {
c88ad5db 816 Log("SHUTTLE", Form("CopyFilesToGrid - Cannot create directory %s",
3d8bc902 817 alienDir.Data()));
2d9019b4 818 delete dirList;
3d8bc902 819 return kFALSE;
546242fb 820 } else {
c88ad5db 821 Log("SHUTTLE",Form("CopyFilesToGrid - Folder %s created", alienDir.Data()));
3d8bc902 822 }
823
546242fb 824 } else {
c88ad5db 825 Log("SHUTTLE",Form("CopyFilesToGrid - Folder %s found", alienDir.Data()));
3d8bc902 826 }
827 }
828
9827400b 829 TString fullLocalPath;
830 fullLocalPath.Form("%s/%s", dir.Data(), fileName.Data());
831
832 TString fullGridPath;
833 fullGridPath.Form("alien://%s/%s", alienDir.Data(), fileName.Data());
834
a986b218 835 Bool_t result = TFile::Cp(fullLocalPath, fullGridPath);
9827400b 836
837 if (result)
838 {
c88ad5db 839 Log("SHUTTLE", Form("CopyFilesToGrid - Copying local file %s to %s succeeded!",
840 fullLocalPath.Data(), fullGridPath.Data()));
9827400b 841 RemoveFile(fullLocalPath);
2d9019b4 842 nTransfer++;
9827400b 843 }
844 else
845 {
c88ad5db 846 Log("SHUTTLE", Form("CopyFilesToGrid - Copying local file %s to %s FAILED!",
847 fullLocalPath.Data(), fullGridPath.Data()));
9827400b 848 success = kFALSE;
849 }
850 }
2d9019b4 851
c88ad5db 852 Log("SHUTTLE", Form("CopyFilesToGrid - %d (over %d) files in folder %s copied to Grid.",
853 nTransfer, nDirs, dir.Data()));
2d9019b4 854
855
856 delete dirList;
9827400b 857 return success;
858}
859
2d9019b4 860//______________________________________________________________________________________________
861const char* AliShuttle::GetRefFilePrefix(const char* base, const char* detector)
862{
863 //
864 // Get folder name of reference files
865 //
866
867 TString offDetStr(GetOfflineDetName(detector));
868 TString dir;
869 if (offDetStr == "ITS" || offDetStr == "MUON" || offDetStr == "PHOS")
870 {
871 dir.Form("%s/%s/%s", base, offDetStr.Data(), detector);
872 } else {
873 dir.Form("%s/%s", base, offDetStr.Data());
874 }
875
876 return dir.Data();
877
878
879}
c88ad5db 880
3301427a 881//______________________________________________________________________________________________
882void AliShuttle::CleanLocalStorage(const TString& uri)
883{
9827400b 884 //
885 // Called in case the preprocessor is declared failed. Remove remaining objects from the local storages.
886 //
3301427a 887
888 const char* type = 0;
889 if(uri == fgkLocalCDB) {
890 type = "OCDB";
891 } else if(uri == fgkLocalRefStorage) {
546242fb 892 type = "Reference";
3301427a 893 } else {
894 AliError(Form("Invalid storage URI: %s", uri.Data()));
895 return;
896 }
897
898 AliCDBManager* man = AliCDBManager::Instance();
b948db8d 899
3301427a 900 // open local storage
901 AliCDBStorage *localSto = man->GetStorage(uri);
902 if(!localSto) {
903 Log("SHUTTLE",
904 Form("CleanLocalStorage - cannot activate local %s storage", type));
905 return;
906 }
907
908 TString filename(Form("%s/%s/*/Run*_v%d_s*.root",
546242fb 909 localSto->GetBaseFolder().Data(), GetOfflineDetName(fCurrentDetector.Data()), GetCurrentRun()));
3301427a 910
c88ad5db 911 AliDebug(2, Form("filename = %s", filename.Data()));
3301427a 912
c88ad5db 913 Log("SHUTTLE", Form("Removing remaining local files for run %d and detector %s ...",
3301427a 914 GetCurrentRun(), fCurrentDetector.Data()));
915
916 RemoveFile(filename.Data());
917
918}
919
920//______________________________________________________________________________________________
921void AliShuttle::RemoveFile(const char* filename)
922{
9827400b 923 //
924 // removes local file
925 //
3301427a 926
927 TString command(Form("rm -f %s", filename));
928
929 Int_t result = gSystem->Exec(command.Data());
930 if(result != 0)
931 {
932 Log("SHUTTLE", Form("RemoveFile - %s: Cannot remove file %s!",
933 fCurrentDetector.Data(), filename));
934 }
73abe331 935}
936
b948db8d 937//______________________________________________________________________________________________
5164a766 938AliShuttleStatus* AliShuttle::ReadShuttleStatus()
939{
9827400b 940 //
941 // Reads the AliShuttleStatus from the CDB
942 //
5164a766 943
2bb7b766 944 if (fStatusEntry){
945 delete fStatusEntry;
946 fStatusEntry = 0;
947 }
5164a766 948
10a5a932 949 fStatusEntry = AliCDBManager::Instance()->GetStorage(GetLocalCDB())
2bb7b766 950 ->Get(Form("/SHUTTLE/STATUS/%s", fCurrentDetector.Data()), GetCurrentRun());
5164a766 951
2bb7b766 952 if (!fStatusEntry) return 0;
953 fStatusEntry->SetOwner(1);
5164a766 954
2bb7b766 955 AliShuttleStatus* status = dynamic_cast<AliShuttleStatus*> (fStatusEntry->GetObject());
956 if (!status) {
957 AliError("Invalid object stored to CDB!");
958 return 0;
959 }
5164a766 960
2bb7b766 961 return status;
5164a766 962}
963
964//______________________________________________________________________________________________
7bfb2090 965Bool_t AliShuttle::WriteShuttleStatus(AliShuttleStatus* status)
5164a766 966{
9827400b 967 //
968 // writes the status for one subdetector
969 //
2bb7b766 970
971 if (fStatusEntry){
972 delete fStatusEntry;
973 fStatusEntry = 0;
974 }
5164a766 975
2bb7b766 976 Int_t run = GetCurrentRun();
5164a766 977
2bb7b766 978 AliCDBId id(AliCDBPath("SHUTTLE", "STATUS", fCurrentDetector), run, run);
5164a766 979
2bb7b766 980 fStatusEntry = new AliCDBEntry(status, id, new AliCDBMetaData);
981 fStatusEntry->SetOwner(1);
5164a766 982
2bb7b766 983 UInt_t result = AliCDBManager::Instance()->GetStorage(fgkLocalCDB)->Put(fStatusEntry);
7bfb2090 984
2bb7b766 985 if (!result) {
3301427a 986 Log("SHUTTLE", Form("WriteShuttleStatus - Failed for %s, run %d",
987 fCurrentDetector.Data(), run));
2bb7b766 988 return kFALSE;
989 }
e7f62f16 990
8e828d39 991 SendMLDetInfo();
7bfb2090 992
2bb7b766 993 return kTRUE;
5164a766 994}
995
996//______________________________________________________________________________________________
997void AliShuttle::UpdateShuttleStatus(AliShuttleStatus::Status newStatus, Bool_t increaseCount)
998{
9827400b 999 //
1000 // changes the AliShuttleStatus for the given detector and run to the given status
1001 //
5164a766 1002
2bb7b766 1003 if (!fStatusEntry){
1004 AliError("UNEXPECTED: fStatusEntry empty");
1005 return;
1006 }
5164a766 1007
2bb7b766 1008 AliShuttleStatus* status = dynamic_cast<AliShuttleStatus*> (fStatusEntry->GetObject());
5164a766 1009
2bb7b766 1010 if (!status){
c88ad5db 1011 Log("SHUTTLE", "UpdateShuttleStatus - UNEXPECTED: status could not be read from current CDB entry");
2bb7b766 1012 return;
1013 }
5164a766 1014
2c15234c 1015 TString actionStr = Form("UpdateShuttleStatus - %s: Changing state from %s to %s",
eba76848 1016 fCurrentDetector.Data(),
36c99a6a 1017 status->GetStatusName(),
eba76848 1018 status->GetStatusName(newStatus));
cb343cfd 1019 Log("SHUTTLE", actionStr);
1020 SetLastAction(actionStr);
5164a766 1021
2bb7b766 1022 status->SetStatus(newStatus);
1023 if (increaseCount) status->IncreaseCount();
5164a766 1024
2bb7b766 1025 AliCDBManager::Instance()->GetStorage(fgkLocalCDB)->Put(fStatusEntry);
e7f62f16 1026
8e828d39 1027 SendMLDetInfo();
5164a766 1028}
e7f62f16 1029
1030//______________________________________________________________________________________________
8e828d39 1031void AliShuttle::SendMLDetInfo()
e7f62f16 1032{
1033 //
1034 // sends ML information about the current status of the current detector being processed
1035 //
1036
1037 AliShuttleStatus* status = dynamic_cast<AliShuttleStatus*> (fStatusEntry->GetObject());
1038
1039 if (!status){
8e828d39 1040 Log("SHUTTLE", "SendMLDetInfo - UNEXPECTED: status could not be read from current CDB entry");
e7f62f16 1041 return;
1042 }
1043
1044 TMonaLisaText mlStatus(Form("%s_status", fCurrentDetector.Data()), status->GetStatusName());
1045 TMonaLisaValue mlRetryCount(Form("%s_count", fCurrentDetector.Data()), status->GetCount());
1046
1047 TList mlList;
1048 mlList.Add(&mlStatus);
1049 mlList.Add(&mlRetryCount);
1050
ee6f7523 1051 TString mlID;
1052 mlID.Form("%d", GetCurrentRun());
1053 fMonaLisa->SendParameters(&mlList, mlID);
e7f62f16 1054}
1055
5164a766 1056//______________________________________________________________________________________________
1057Bool_t AliShuttle::ContinueProcessing()
1058{
9827400b 1059 // this function reads the AliShuttleStatus information from CDB and
1060 // checks if the processing should be continued
1061 // if yes it returns kTRUE and updates the AliShuttleStatus with nextStatus
2bb7b766 1062
926cbc0e 1063 if (!fConfig->HostProcessDetector(fCurrentDetector))
1064 return kFALSE;
57c1a579 1065
1066 AliPreprocessor* aPreprocessor =
1067 dynamic_cast<AliPreprocessor*> (fPreprocessorMap.GetValue(fCurrentDetector));
1068 if (!aPreprocessor)
1069 {
c88ad5db 1070 Log("SHUTTLE", Form("ContinueProcessing - %s: no preprocessor registered", fCurrentDetector.Data()));
57c1a579 1071 return kFALSE;
1072 }
1073
2bb7b766 1074 AliShuttleLogbookEntry::Status entryStatus =
eba76848 1075 fLogbookEntry->GetDetectorStatus(fCurrentDetector);
2bb7b766 1076
5e993b6f 1077 if (entryStatus != AliShuttleLogbookEntry::kUnprocessed) {
c88ad5db 1078 Log("SHUTTLE", Form("ContinueProcessing - %s is %s",
2bb7b766 1079 fCurrentDetector.Data(),
1080 fLogbookEntry->GetDetectorStatusName(entryStatus)));
1081 return kFALSE;
1082 }
1083
1084 // if we get here, according to Shuttle logbook subdetector is in UNPROCESSED state
be48e3ea 1085
1086 // check if current run is first unprocessed run for current detector
1087 if (fConfig->StrictRunOrder(fCurrentDetector) &&
1088 !fFirstUnprocessed[GetDetPos(fCurrentDetector)])
1089 {
86aa42c3 1090 if (fTestMode == kNone)
1091 {
c88ad5db 1092 Log("SHUTTLE", Form("ContinueProcessing - %s requires strict run ordering"
1093 " but this is not the first unprocessed run!"));
86aa42c3 1094 return kFALSE;
1095 }
1096 else
1097 {
c88ad5db 1098 Log("SHUTTLE", Form("ContinueProcessing - In TESTMODE - "
1099 "Although %s requires strict run ordering "
1100 "and this is not the first unprocessed run, "
1101 "the SHUTTLE continues"));
86aa42c3 1102 }
be48e3ea 1103 }
1104
926cbc0e 1105 // Is the subdetector processed first time for this run?
1106 fFirstProcessing = kFALSE;
1107
2bb7b766 1108 AliShuttleStatus* status = ReadShuttleStatus();
1109 if (!status) {
1110 // first time
1111 Log("SHUTTLE", Form("ContinueProcessing - %s: Processing first time",
1112 fCurrentDetector.Data()));
1113 status = new AliShuttleStatus(AliShuttleStatus::kStarted);
926cbc0e 1114 fFirstProcessing = kTRUE;
2bb7b766 1115 return WriteShuttleStatus(status);
1116 }
1117
ffa25f20 1118 // The following case shouldn't happen if Shuttle Logbook was correctly updated.
2bb7b766 1119 // If it happens it may mean Logbook updating failed... let's do it now!
1120 if (status->GetStatus() == AliShuttleStatus::kDone ||
ffa25f20 1121 status->GetStatus() == AliShuttleStatus::kFailed ||
1122 status->GetStatus() == AliShuttleStatus::kSkipped) {
2bb7b766 1123 Log("SHUTTLE", Form("ContinueProcessing - %s is already %s. Updating Shuttle Logbook",
1124 fCurrentDetector.Data(),
1125 status->GetStatusName(status->GetStatus())));
ffa25f20 1126
1127 if (status->GetStatus() == AliShuttleStatus::kSkipped)
1128 {
1129 UpdateShuttleLogbook(fCurrentDetector.Data(), "DONE");
1130 }
1131 else
1132 UpdateShuttleLogbook(fCurrentDetector.Data(), status->GetStatusName(status->GetStatus()));
1133
2bb7b766 1134 return kFALSE;
1135 }
1136
5e993b6f 1137 if (status->GetStatus() == AliShuttleStatus::kStoreStarted || status->GetStatus() == AliShuttleStatus::kStoreDelayed ||status->GetStatus() == AliShuttleStatus::kStoreError) {
2bb7b766 1138 Log("SHUTTLE",
c88ad5db 1139 Form("ContinueProcessing - %s: Grid storage of one or more "
1140 "objects failed. Trying again now",
2bb7b766 1141 fCurrentDetector.Data()));
5e993b6f 1142 StoreOCDB();
2bb7b766 1143 return kFALSE;
1144 }
1145
1146 // if we get here, there is a restart
57c1a579 1147 Bool_t cont = kFALSE;
2bb7b766 1148
1149 // abort conditions
cb343cfd 1150 if (status->GetCount() >= fConfig->GetMaxRetries()) {
57c1a579 1151 Log("SHUTTLE", Form("ContinueProcessing - %s failed %d times in status %s - "
1152 "Updating Shuttle Logbook", fCurrentDetector.Data(),
2bb7b766 1153 status->GetCount(), status->GetStatusName()));
1154 UpdateShuttleLogbook(fCurrentDetector.Data(), "FAILED");
e7f62f16 1155 UpdateShuttleStatus(AliShuttleStatus::kFailed);
3301427a 1156
1157 // there may still be objects in local OCDB and reference storage
1158 // and FXS databases may be not updated: do it now!
9827400b 1159
1160 // TODO Currently disabled, we want to keep files in case of failure!
1161 // CleanLocalStorage(fgkLocalCDB);
1162 // CleanLocalStorage(fgkLocalRefStorage);
1163 // UpdateTableFailCase();
1164
1165 // Send mail to detector expert!
c88ad5db 1166 Log("SHUTTLE", Form("ContinueProcessing - Sending mail to %s expert...",
926cbc0e 1167 fCurrentDetector.Data()));
339adafa 1168 if (!SendMail(kPPEMail))
9827400b 1169 Log("SHUTTLE", Form("ContinueProcessing - Could not send mail to %s expert",
926cbc0e 1170 fCurrentDetector.Data()));
3301427a 1171
57c1a579 1172 } else {
1173 Log("SHUTTLE", Form("ContinueProcessing - %s: restarting. "
1174 "Aborted before with %s. Retry number %d.", fCurrentDetector.Data(),
1175 status->GetStatusName(), status->GetCount()));
9827400b 1176 Bool_t increaseCount = kTRUE;
c88ad5db 1177 if (status->GetStatus() == AliShuttleStatus::kDCSError ||
8da81210 1178 status->GetStatus() == AliShuttleStatus::kDCSStarted ||
1179 status->GetStatus() == AliShuttleStatus::kFXSError)
c88ad5db 1180 increaseCount = kFALSE;
675f64cd 1181
9827400b 1182 UpdateShuttleStatus(AliShuttleStatus::kStarted, increaseCount);
57c1a579 1183 cont = kTRUE;
2bb7b766 1184 }
1185
57c1a579 1186 return cont;
5164a766 1187}
1188
8e828d39 1189//______________________________________________________________________________________________
1190void AliShuttle::SendMLRunInfo(const char* status)
1191{
1192 //
1193 // Send information about this run to ML
1194
1195 TMonaLisaText mlStatus("SHUTTLE_status", status);
1196 TString runType(fLogbookEntry->GetRunType());
1197 if (strlen(fLogbookEntry->GetRunParameter("log")) > 0){
1198
1199 runType += "(";
1200 runType += fLogbookEntry->GetRunParameter("log");
1201 runType += ")";
1202 }
1203 TMonaLisaText mlRunType("SHUTTLE_runtype", runType);
1204
1205 TList mlList;
1206 mlList.Add(&mlStatus);
1207 mlList.Add(&mlRunType);
1208
1209 TString mlID;
1210 mlID.Form("%d", GetCurrentRun());
1211 fMonaLisa->SendParameters(&mlList, mlID);
1212}
1213
ef83773f 1214//______________________________________________________________________________________________
1215Int_t AliShuttle::GetMem(Int_t pid)
1216{
1217 // invokes ps to get the memory consumption of the process <pid>
1218 // returns -1 in case of error
1219
1220 TString checkStr;
1221 checkStr.Form("ps -o vsize --pid %d | tail -n 1", pid);
1222 FILE* pipe = gSystem->OpenPipe(checkStr, "r");
1223 if (!pipe)
1224 {
1225 Log("SHUTTLE", Form("Process - Error: "
1226 "Could not open pipe to %s", checkStr.Data()));
1227 return -1;
1228 }
1229
1230 char buffer[100];
1231 if (!fgets(buffer, 100, pipe))
1232 {
1233 Log("SHUTTLE", "Process - Error: ps did not return anything");
1234 gSystem->ClosePipe(pipe);
1235 return -1;
1236 }
1237 gSystem->ClosePipe(pipe);
1238
1239 //Log("SHUTTLE", Form("ps returned %s", buffer));
1240
1241 Int_t mem = 0;
1242 if ((sscanf(buffer, "%d\n", &mem) != 1) || !mem)
1243 {
1244 Log("SHUTTLE", "Process - Error: Could not parse output of ps");
1245 return -1;
1246 }
1247
1248 return mem;
1249}
1250
5164a766 1251//______________________________________________________________________________________________
2bb7b766 1252Bool_t AliShuttle::Process(AliShuttleLogbookEntry* entry)
58bc3020 1253{
73abe331 1254 //
b948db8d 1255 // Makes data retrieval for all detectors in the configuration.
2bb7b766 1256 // entry: Shuttle logbook entry, contains run paramenters and status of detectors
1257 // (Unprocessed, Inactive, Failed or Done).
d477ad88 1258 // Returns kFALSE in case of error occured and kTRUE otherwise
73abe331 1259 //
1260
9827400b 1261 if (!entry) return kFALSE;
2bb7b766 1262
1263 fLogbookEntry = entry;
1264
c88ad5db 1265 Log("SHUTTLE", Form("\t\t\t^*^*^*^*^*^*^*^*^*^*^*^* run %d: START ^*^*^*^*^*^*^*^*^*^*^*^*",
9827400b 1266 GetCurrentRun()));
2bb7b766 1267
339adafa 1268 CountOpenRuns();
1269
8e828d39 1270 // Send the information to ML
1271 SendMLRunInfo("Processing");
3301427a 1272
9827400b 1273 if (fLogbookEntry->IsDone())
1274 {
1275 Log("SHUTTLE","Process - Shuttle is already DONE. Updating logbook");
1276 UpdateShuttleLogbook("shuttle_done");
1277 fLogbookEntry = 0;
1278 return kTRUE;
1279 }
1280
1281 // read test mode if flag is set
1282 if (fReadTestMode)
1283 {
3d8bc902 1284 fTestMode = kNone;
9827400b 1285 TString logEntry(entry->GetRunParameter("log"));
1286 //printf("log entry = %s\n", logEntry.Data());
1287 TString searchStr("Testmode: ");
1288 Int_t pos = logEntry.Index(searchStr.Data());
1289 //printf("%d\n", pos);
1290 if (pos >= 0)
1291 {
1292 TSubString subStr = logEntry(pos + searchStr.Length(), logEntry.Length());
1293 //printf("%s\n", subStr.String().Data());
1294 TString newStr(subStr.Data());
1295 TObjArray* token = newStr.Tokenize(' ');
1296 if (token)
1297 {
1298 //token->Print();
1299 TObjString* tmpStr = dynamic_cast<TObjString*> (token->First());
1300 if (tmpStr)
1301 {
1302 Int_t testMode = tmpStr->String().Atoi();
1303 if (testMode > 0)
1304 {
c88ad5db 1305 Log("SHUTTLE", Form("Process - Enabling test mode %d", testMode));
9827400b 1306 SetTestMode((TestMode) testMode);
1307 }
1308 }
1309 delete token;
1310 }
1311 }
1312 }
c88ad5db 1313
eba76848 1314 fLogbookEntry->Print("all");
57f50b3c 1315
1316 // Initialization
d477ad88 1317 Bool_t hasError = kFALSE;
5164a766 1318
675f64cd 1319 // Set the CDB and Reference folders according to the year and LHC period
1320 TString lhcPeriod(GetLHCPeriod());
1321 if (lhcPeriod.Length() == 0)
1322 {
7d4cf768 1323 Log("SHUTTLE","Process - LHCPeriod not found in logbook!");
1324 return 0;
675f64cd 1325 }
1326
1265724d 1327 // build cdb paths (repeat each time, LHCperiod might have changed)
1328 fgkMainCDB.Form("alien://folder=%s%d/%s/OCDB?user=alidaq?cacheFold=/tmp/OCDBCache",
b8cf5e0f 1329 fConfig->GetAlienPath(), GetCurrentYear(), lhcPeriod.Data());
675f64cd 1330
1265724d 1331 fgkMainRefStorage.Form("alien://folder=%s%d/%s/Reference?user=alidaq?cacheFold=/tmp/OCDBCache",
b8cf5e0f 1332 fConfig->GetAlienPath(), GetCurrentYear(), lhcPeriod.Data());
675f64cd 1333
57f50b3c 1334 // Loop on detectors in the configuration
b948db8d 1335 TIter iter(fConfig->GetDetectors());
2bb7b766 1336 TObjString* aDetector = 0;
b948db8d 1337
4508ab8b 1338 Bool_t first = kTRUE;
1339
be48e3ea 1340 while ((aDetector = (TObjString*) iter.Next()))
1341 {
7bfb2090 1342 fCurrentDetector = aDetector->String();
5164a766 1343
5e993b6f 1344 if (ContinueProcessing() == kFALSE)
1345 continue;
4508ab8b 1346
1347 if (first)
1348 {
1349 // only read QueryCDB when needed and only once
1350 AliCDBStorage *mainCDBSto = AliCDBManager::Instance()->GetStorage(fgkMainCDB);
1351 if(mainCDBSto) mainCDBSto->QueryCDB(GetCurrentRun());
1352 AliCDBStorage *mainRefSto = AliCDBManager::Instance()->GetStorage(fgkMainRefStorage);
1353 if(mainRefSto) mainRefSto->QueryCDB(GetCurrentRun());
1354 first = kFALSE;
1355 }
9e080f92 1356
c88ad5db 1357 Log("SHUTTLE", Form("\t\t\t****** run %d - %s: START ******",
2bb7b766 1358 GetCurrentRun(), aDetector->GetName()));
1359
9d733021 1360 for(Int_t iSys=0;iSys<3;iSys++) fFXSCalled[iSys]=kFALSE;
ef83773f 1361
1362 Int_t initialMem = GetMem(getpid());
1363 Log("SHUTTLE", Form("Memory consumption before forking is %d", initialMem));
9d733021 1364
c88ad5db 1365 Log(fCurrentDetector.Data(), "Process - Starting processing");
85a80aa9 1366
be48e3ea 1367 Int_t pid = fork();
1368
1369 if (pid < 0)
1370 {
c88ad5db 1371 Log("SHUTTLE", "Process - ERROR: Forking failed");
be48e3ea 1372 }
1373 else if (pid > 0)
1374 {
1375 // parent
c88ad5db 1376 Log("SHUTTLE", Form("Process - In parent process of %d - %s: Starting monitoring",
be48e3ea 1377 GetCurrentRun(), aDetector->GetName()));
1378
1379 Long_t begin = time(0);
1380
1381 int status; // to be used with waitpid, on purpose an int (not Int_t)!
1382 while (waitpid(pid, &status, WNOHANG) == 0)
1383 {
1384 Long_t expiredTime = time(0) - begin;
1385
1386 if (expiredTime > fConfig->GetPPTimeOut())
1387 {
56d3e9e4 1388 TString logMsg;
1389 AliShuttleStatus *currentStatus = ReadShuttleStatus();
1390 AliShuttleStatus::Status newStatus = AliShuttleStatus::kInvalid;
1391
1392 if (currentStatus->GetStatus() <= AliShuttleStatus::kPPDone)
1393 {
1394 // in case pp not yet done set status to kPPTimeOut
1395
1396 logMsg.Form("Process - Process of %s timed out. Run time: %d seconds. Killing...",
1397 fCurrentDetector.Data(), expiredTime);
1398 newStatus = AliShuttleStatus::kPPTimeOut;
1399 }
1400 else if (currentStatus->GetStatus() == AliShuttleStatus::kStoreStarted)
1401 {
1402 // in case the pp goes in TimeOut while storing the objects in the OCDB
1403 // set status to kStoreError
1404
1405 logMsg.Form("Process - Process of %s timed out while storing the OCDB object. Run time: %d seconds. Killing... and setting status to StoreError.",
1406 fCurrentDetector.Data(), expiredTime);
1407 newStatus = AliShuttleStatus::kStoreError;
1408 }
1409 else
1410 {
1411 // in other cases don't change the status
1412
1413 logMsg.Form("Process - Process of %s timed out in status = %s. Run time: %d seconds. Killing... without changing the status",
1414 fCurrentDetector.Data(), currentStatus->GetStatusName(), expiredTime);
1415 }
1416
1417 Log("SHUTTLE", logMsg);
1418 Log(fCurrentDetector, logMsg);
be48e3ea 1419
1420 kill(pid, 9);
1421
56d3e9e4 1422 if (newStatus != AliShuttleStatus::kInvalid)
1423 UpdateShuttleStatus(newStatus);
be48e3ea 1424 hasError = kTRUE;
1425
1426 gSystem->Sleep(1000);
1427 }
1428 else
1429 {
be48e3ea 1430 gSystem->Sleep(1000);
9827400b 1431
ef83773f 1432 Int_t mem = GetMem(pid);
1433
1434 if (mem < 0)
9827400b 1435 continue;
9827400b 1436
ef83773f 1437 mem -= initialMem;
1438 if (mem < 0)
1439 mem = 0;
9827400b 1440
1441 if (expiredTime % 60 == 0)
ee6f7523 1442 {
c88ad5db 1443 Log("SHUTTLE", Form("Process - %s: Checking process. "
1444 "Run time: %d seconds - Memory consumption: %d KB",
1445 fCurrentDetector.Data(), expiredTime, mem));
ee6f7523 1446 SendAlive();
1447 }
9827400b 1448
1449 if (mem > fConfig->GetPPMaxMem())
1450 {
1451 TString tmp;
c88ad5db 1452 tmp.Form("Process - Process exceeds maximum allowed memory "
1453 "(%d KB > %d KB). Killing...",
9827400b 1454 mem, fConfig->GetPPMaxMem());
1455 Log("SHUTTLE", tmp);
1456 Log(fCurrentDetector, tmp);
1457
1458 kill(pid, 9);
1459
1460 UpdateShuttleStatus(AliShuttleStatus::kPPOutOfMemory);
1461 hasError = kTRUE;
1462
1463 gSystem->Sleep(1000);
1464 }
be48e3ea 1465 }
1466 }
1467
c88ad5db 1468 Log("SHUTTLE", Form("Process - In parent process of %d - %s: Client has terminated.",
be48e3ea 1469 GetCurrentRun(), aDetector->GetName()));
1470
1471 if (WIFEXITED(status))
1472 {
1473 Int_t returnCode = WEXITSTATUS(status);
1474
c88ad5db 1475 Log("SHUTTLE", Form("Process - %s: the return code is %d", fCurrentDetector.Data(),
3301427a 1476 returnCode));
be48e3ea 1477
9827400b 1478 if (returnCode == 0) hasError = kTRUE;
be48e3ea 1479 }
1480 }
1481 else if (pid == 0)
1482 {
26b1866e 1483 // child
1484 Log("SHUTTLE", Form("Process - In child process of %d - %s", GetCurrentRun(),
c88ad5db 1485 aDetector->GetName()));
be48e3ea 1486
c88ad5db 1487 Log("SHUTTLE", Form("Process - Redirecting output to %s log",fCurrentDetector.Data()));
ffa29e93 1488
546242fb 1489 if ((freopen(GetLogFileName(fCurrentDetector), "a", stdout)) == 0)
ffa29e93 1490 {
c88ad5db 1491 Log("SHUTTLE", "Process - Could not freopen stdout");
ffa29e93 1492 }
1493 else
1494 {
1495 fOutputRedirected = kTRUE;
1496 if ((dup2(fileno(stdout), fileno(stderr))) < 0)
c88ad5db 1497 Log("SHUTTLE", "Process - Could not redirect stderr");
ffa29e93 1498
1499 }
1500
5bac2bde 1501 TString wd = gSystem->WorkingDirectory();
675f64cd 1502 TString tmpDir = Form("%s/%s_%d_process", GetShuttleTempDir(),
1503 fCurrentDetector.Data(), GetCurrentRun());
5bac2bde 1504
d524ade6 1505 Int_t result = gSystem->GetPathInfo(tmpDir.Data(), 0, (Long64_t*) 0, 0, 0);
1506 if (!result) // temp dir already exists!
1507 {
1508 Log(fCurrentDetector.Data(),
1509 Form("Process - %s dir already exists! Removing...", tmpDir.Data()));
1510 gSystem->Exec(Form("rm -rf %s",tmpDir.Data()));
675f64cd 1511 }
1512
1513 if (gSystem->mkdir(tmpDir.Data(), 1))
1514 {
1515 Log(fCurrentDetector.Data(), "Process - could not make temp directory!!");
1516 gSystem->Exit(1);
d524ade6 1517 }
1518
1519 if (!gSystem->ChangeDirectory(tmpDir.Data()))
1520 {
1521 Log(fCurrentDetector.Data(), "Process - could not change directory!!");
1522 gSystem->Exit(1);
1523 }
5bac2bde 1524
ffa25f20 1525 Int_t success = ProcessCurrentDetector();
26b1866e 1526
5bac2bde 1527 gSystem->ChangeDirectory(wd.Data());
b1d18693 1528
ffa25f20 1529 if (success == 1) // Preprocessor finished successfully!
9827400b 1530 {
fb40ead4 1531 // remove temporary folder or DCS map
1532 if (!fConfig->KeepTempFolder())
1533 {
1534 gSystem->Exec(Form("rm -rf %s",tmpDir.Data()));
1535 } else if (!fConfig->KeepDCSMap())
1536 {
1537 gSystem->Exec(Form("rm -f %s/DCSMap.root",tmpDir.Data()));
1538 }
b1d18693 1539
3301427a 1540 // Update time_processed field in FXS DB
1541 if (UpdateTable() == kFALSE)
5bac2bde 1542 Log("SHUTTLE", Form("Process - %s: Could not update FXS databases!",
1543 fCurrentDetector.Data()));
3301427a 1544
1545 // Transfer the data from local storage to main storage (Grid)
3301427a 1546 if (StoreOCDB() == kFALSE)
9827400b 1547 success = kFALSE;
5e993b6f 1548 }
ffa25f20 1549 else if (success == 0)
c88ad5db 1550 {
1551 Log("SHUTTLE",
1552 Form("\t\t\t****** run %d - %s: PP ERROR ******",
1553 GetCurrentRun(), aDetector->GetName()));
be48e3ea 1554 }
1555
4b95672b 1556 for (UInt_t iSys=0; iSys<3; iSys++)
1557 {
1558 if (fFXSCalled[iSys]) fFXSlist[iSys].Clear();
1559 }
1560
c88ad5db 1561 Log("SHUTTLE", Form("Process - Client process of %d - %s is exiting now with %d.",
9827400b 1562 GetCurrentRun(), aDetector->GetName(), success));
be48e3ea 1563
1564 // the client exits here
9827400b 1565 gSystem->Exit(success);
be48e3ea 1566
1567 AliError("We should never get here!!!");
1568 }
7bfb2090 1569 }
5164a766 1570
c88ad5db 1571 Log("SHUTTLE", Form("\t\t\t^*^*^*^*^*^*^*^*^*^*^*^* run %d: FINISH ^*^*^*^*^*^*^*^*^*^*^*^*",
2bb7b766 1572 GetCurrentRun()));
1573
1574 //check if shuttle is done for this run, if so update logbook
1575 TObjArray checkEntryArray;
1576 checkEntryArray.SetOwner(1);
9e080f92 1577 TString whereClause = Form("where run=%d", GetCurrentRun());
b0e53b15 1578 if (!QueryShuttleLogbook(whereClause.Data(), checkEntryArray) ||
1579 checkEntryArray.GetEntries() == 0) {
9e080f92 1580 Log("SHUTTLE", Form("Process - Warning: Cannot check status of run %d on Shuttle logbook!",
1581 GetCurrentRun()));
1582 return hasError == kFALSE;
1583 }
b948db8d 1584
9e080f92 1585 AliShuttleLogbookEntry* checkEntry = dynamic_cast<AliShuttleLogbookEntry*>
1586 (checkEntryArray.At(0));
2bb7b766 1587
9e080f92 1588 if (checkEntry)
1589 {
1590 if (checkEntry->IsDone())
be48e3ea 1591 {
9e080f92 1592 Log("SHUTTLE","Process - Shuttle is DONE. Updating logbook");
1593 UpdateShuttleLogbook("shuttle_done");
1594 }
1595 else
1596 {
1597 for (UInt_t iDet=0; iDet<NDetectors(); iDet++)
be48e3ea 1598 {
9e080f92 1599 if (checkEntry->GetDetectorStatus(iDet) == AliShuttleLogbookEntry::kUnprocessed)
be48e3ea 1600 {
9e080f92 1601 AliDebug(2, Form("Run %d: setting %s as \"not first time unprocessed\"",
1602 checkEntry->GetRun(), GetDetName(iDet)));
1603 fFirstUnprocessed[iDet] = kFALSE;
be48e3ea 1604 }
1605 }
8e828d39 1606 SendMLRunInfo("Pending");
2bb7b766 1607 }
1608 }
1609
1610 fLogbookEntry = 0;
85a80aa9 1611
a7160fe9 1612 return hasError == kFALSE;
73abe331 1613}
1614
b948db8d 1615//______________________________________________________________________________________________
ffa25f20 1616Int_t AliShuttle::ProcessCurrentDetector()
73abe331 1617{
1618 //
2bb7b766 1619 // Makes data retrieval just for a specific detector (fCurrentDetector).
73abe331 1620 // Threre should be a configuration for this detector.
73abe331 1621
1d172743 1622 Log("SHUTTLE", Form("ProcessCurrentDetector - Retrieving values for %s, run %d",
1623 fCurrentDetector.Data(), GetCurrentRun()));
73abe331 1624
d524ade6 1625 TString wd = gSystem->WorkingDirectory();
1626
2d9019b4 1627 if (!CleanReferenceStorage(fCurrentDetector.Data()))
ffa25f20 1628 return 0;
d524ade6 1629
1630 gSystem->ChangeDirectory(wd.Data());
1631
3301427a 1632 // call preprocessor
1633 AliPreprocessor* aPreprocessor =
1634 dynamic_cast<AliPreprocessor*> (fPreprocessorMap.GetValue(fCurrentDetector));
1635
8579c6fb 1636 // check if the preprocessor wants to process this run type
1637 if (aPreprocessor->ProcessRunType() == kFALSE)
ffa25f20 1638 {
1639 UpdateShuttleStatus(AliShuttleStatus::kSkipped);
1640 UpdateShuttleLogbook(fCurrentDetector, "DONE");
1641 Log(fCurrentDetector, Form("ProcessCurrentDetector - %s preprocessor is not interested in this run type", fCurrentDetector.Data()));
1642
1643 return 2;
1644 }
1645
1646 TMap* dcsMap = new TMap();
1647
3301427a 1648 aPreprocessor->Initialize(GetCurrentRun(), GetCurrentStartTime(), GetCurrentEndTime());
1649
1650 Bool_t processDCS = aPreprocessor->ProcessDCS();
d477ad88 1651
651fdaab 1652 if (!processDCS)
1653 {
1d172743 1654 Log(fCurrentDetector, "ProcessCurrentDetector -"
1655 " The preprocessor requested to skip the retrieval of DCS values");
651fdaab 1656 }
8b739301 1657 else if (fTestMode & kSkipDCS)
2c15234c 1658 {
1d172743 1659 Log(fCurrentDetector, "ProcessCurrentDetector - In TESTMODE: Skipping DCS processing");
9827400b 1660 }
1661 else if (fTestMode & kErrorDCS)
1662 {
1d172743 1663 Log(fCurrentDetector, "ProcessCurrentDetector - In TESTMODE: Simulating DCS error");
3d8bc902 1664 UpdateShuttleStatus(AliShuttleStatus::kDCSStarted);
9827400b 1665 UpdateShuttleStatus(AliShuttleStatus::kDCSError);
1d172743 1666 delete dcsMap;
ffa25f20 1667 return 0;
2c15234c 1668 } else {
3301427a 1669
1670 UpdateShuttleStatus(AliShuttleStatus::kDCSStarted);
1671
1d172743 1672 // Query DCS archive
1673 Int_t nServers = fConfig->GetNServers(fCurrentDetector);
a038aa70 1674
1d172743 1675 for (int iServ=0; iServ<nServers; iServ++)
2c15234c 1676 {
1d172743 1677
1678 TString host(fConfig->GetDCSHost(fCurrentDetector, iServ));
1679 Int_t port = fConfig->GetDCSPort(fCurrentDetector, iServ);
542b6cc8 1680 Int_t multiSplit = fConfig->GetMultiSplit(fCurrentDetector, iServ);
1681
1790d4b7 1682 Log(fCurrentDetector, Form("ProcessCurrentDetector -"
1683 " Querying DCS Amanda server %s:%d (%d of %d)",
1684 host.Data(), port, iServ+1, nServers));
1d172743 1685
1686 TMap* aliasMap = 0;
1687 TMap* dpMap = 0;
26b1866e 1688
1d172743 1689 if (fConfig->GetDCSAliases(fCurrentDetector, iServ)->GetEntries() > 0)
2c15234c 1690 {
1d172743 1691 aliasMap = GetValueSet(host, port,
542b6cc8 1692 fConfig->GetDCSAliases(fCurrentDetector, iServ),
1693 kAlias, multiSplit);
1d172743 1694 if (!aliasMap)
1695 {
1696 Log(fCurrentDetector,
1697 Form("ProcessCurrentDetector -"
675f64cd 1698 " Error retrieving DCS aliases from server %s."
1699 " Sending mail to DCS experts!", host.Data()));
1d172743 1700 UpdateShuttleStatus(AliShuttleStatus::kDCSError);
675f64cd 1701
339adafa 1702 if (!SendMail(kDCSEMail))
fb40ead4 1703 Log("SHUTTLE", Form("ProcessCurrentDetector - "
95d4aaef 1704 "Could not send mail to DCS experts!"));
675f64cd 1705
a038aa70 1706 delete dcsMap;
ffa25f20 1707 return 0;
1d172743 1708 }
2c15234c 1709 }
a038aa70 1710
1d172743 1711 if (fConfig->GetDCSDataPoints(fCurrentDetector, iServ)->GetEntries() > 0)
a038aa70 1712 {
1d172743 1713 dpMap = GetValueSet(host, port,
542b6cc8 1714 fConfig->GetDCSDataPoints(fCurrentDetector, iServ),
1715 kDP, multiSplit);
1d172743 1716 if (!dpMap)
1717 {
1718 Log(fCurrentDetector,
1719 Form("ProcessCurrentDetector -"
675f64cd 1720 " Error retrieving DCS data points from server %s."
1721 " Sending mail to DCS experts!", host.Data()));
1d172743 1722 UpdateShuttleStatus(AliShuttleStatus::kDCSError);
675f64cd 1723
339adafa 1724 if (!SendMail(kDCSEMail))
fb40ead4 1725 Log("SHUTTLE", Form("ProcessCurrentDetector - "
95d4aaef 1726 "Could not send mail to DCS experts!"));
675f64cd 1727
1d172743 1728 if (aliasMap) delete aliasMap;
1729 delete dcsMap;
ffa25f20 1730 return 0;
1d172743 1731 }
a038aa70 1732 }
1d172743 1733
1734 // merge aliasMap and dpMap into dcsMap
1735 if(aliasMap) {
1736 TIter iter(aliasMap);
a038aa70 1737 TObjString* key = 0;
1738 while ((key = (TObjString*) iter.Next()))
1d172743 1739 dcsMap->Add(key, aliasMap->GetValue(key->String()));
1740
1741 aliasMap->SetOwner(kFALSE);
1742 delete aliasMap;
1743 }
1744
1745 if(dpMap) {
1746 TIter iter(dpMap);
1747 TObjString* key = 0;
1748 while ((key = (TObjString*) iter.Next()))
1749 dcsMap->Add(key, dpMap->GetValue(key->String()));
1750
1751 dpMap->SetOwner(kFALSE);
1752 delete dpMap;
a038aa70 1753 }
73abe331 1754 }
1755 }
dc25836b 1756
b1d18693 1757 // save map into file, to help debugging in case of preprocessor error
fbc112e3 1758 TFile* f = TFile::Open("DCSMap.root","recreate");
b1d18693 1759 f->cd();
1760 dcsMap->Write("DCSMap", TObject::kSingleKey);
1761 f->Close();
fbc112e3 1762 delete f;
b1d18693 1763
2bb7b766 1764 // DCS Archive DB processing successful. Call Preprocessor!
85a80aa9 1765 UpdateShuttleStatus(AliShuttleStatus::kPPStarted);
a7160fe9 1766
339adafa 1767 fFXSError = -1; // this variable is kTRUE after ::Process if an FXS error occured
8da81210 1768
a038aa70 1769 UInt_t returnValue = aPreprocessor->Process(dcsMap);
8da81210 1770
339adafa 1771 if (fFXSError!=-1) {
8da81210 1772 UpdateShuttleStatus(AliShuttleStatus::kFXSError);
339adafa 1773 SendMail(kFXSEMail, fFXSError);
8da81210 1774 dcsMap->DeleteAll();
1775 delete dcsMap;
ffa25f20 1776 return 0;
8da81210 1777 }
b948db8d 1778
3301427a 1779 if (returnValue > 0) // Preprocessor error!
1780 {
c88ad5db 1781 Log(fCurrentDetector, Form("ProcessCurrentDetector - "
1782 "Preprocessor failed. Process returned %d.", returnValue));
cb343cfd 1783 UpdateShuttleStatus(AliShuttleStatus::kPPError);
a038aa70 1784 dcsMap->DeleteAll();
1785 delete dcsMap;
ffa25f20 1786 return 0;
9827400b 1787 }
1788
1789 // preprocessor ok!
1790 UpdateShuttleStatus(AliShuttleStatus::kPPDone);
1791 Log(fCurrentDetector, Form("ProcessCurrentDetector - %s preprocessor returned success",
1792 fCurrentDetector.Data()));
b948db8d 1793
a038aa70 1794 dcsMap->DeleteAll();
1795 delete dcsMap;
b948db8d 1796
ffa25f20 1797 return 1;
2bb7b766 1798}
1799
fbc112e3 1800//______________________________________________________________________________________________
1801void AliShuttle::CountOpenRuns()
1802{
1803 // Query DAQ's Shuttle logbook and sends the number of open runs to ML
1804
1805 // check connection, in case connect
1806 if (!Connect(3))
1807 return;
1808
1809 TString sqlQuery;
1810 sqlQuery = Form("select count(*) from %s where shuttle_done=0", fConfig->GetShuttlelbTable());
1811
1812 TSQLResult* aResult = fServer[3]->Query(sqlQuery);
1813 if (!aResult) {
1814 AliError(Form("Can't execute query <%s>!", sqlQuery.Data()));
1815 return;
1816 }
1817
1818 AliDebug(2,Form("Query = %s", sqlQuery.Data()));
1819
1820 if (aResult->GetRowCount() == 0) {
1821 AliError(Form("No result for query %s received", sqlQuery.Data()));
1822 return;
1823 }
1824
1825 if (aResult->GetFieldCount() != 1) {
1826 AliError(Form("Invalid field count for query %s received", sqlQuery.Data()));
1827 return;
1828 }
1829
1830 TSQLRow* aRow = aResult->Next();
1831 if (!aRow) {
1832 AliError(Form("Could not receive result of query %s", sqlQuery.Data()));
1833 return;
1834 }
1835
1836 TString result(aRow->GetField(0), aRow->GetFieldLength(0));
1837 Int_t count = result.Atoi();
1838
1839 Log("SHUTTLE", Form("%d unprocessed runs", count));
1840
1841 delete aRow;
1842 delete aResult;
1843
1844 TMonaLisaValue mlStatus("SHUTTLE_openruns", count);
1845
1846 TList mlList;
1847 mlList.Add(&mlStatus);
1848
1849 fMonaLisa->SendParameters(&mlList, "__PROCESSINGINFO__");
1850}
1851
2bb7b766 1852//______________________________________________________________________________________________
1853Bool_t AliShuttle::QueryShuttleLogbook(const char* whereClause,
1854 TObjArray& entries)
1855{
9827400b 1856 // Query DAQ's Shuttle logbook and fills detector status object.
1857 // Call QueryRunParameters to query DAQ logbook for run parameters.
1858 //
2bb7b766 1859
fc5a4708 1860 entries.SetOwner(1);
1861
2bb7b766 1862 // check connection, in case connect
fbc112e3 1863 if (!Connect(3)) return kFALSE;
2bb7b766 1864
1865 TString sqlQuery;
441b0e9c 1866 sqlQuery = Form("select * from %s %s order by run", fConfig->GetShuttlelbTable(), whereClause);
2bb7b766 1867
be48e3ea 1868 TSQLResult* aResult = fServer[3]->Query(sqlQuery);
2bb7b766 1869 if (!aResult) {
1870 AliError(Form("Can't execute query <%s>!", sqlQuery.Data()));
1871 return kFALSE;
1872 }
1873
fc5a4708 1874 AliDebug(2,Form("Query = %s", sqlQuery.Data()));
1875
2bb7b766 1876 if(aResult->GetRowCount() == 0) {
c88ad5db 1877 Log("SHUTTLE", "No entries in Shuttle Logbook match request");
9827400b 1878 delete aResult;
1879 return kTRUE;
2bb7b766 1880 }
1881
1882 // TODO Check field count!
db99d43e 1883 const UInt_t nCols = 23;
2bb7b766 1884 if (aResult->GetFieldCount() != (Int_t) nCols) {
c88ad5db 1885 Log("SHUTTLE", "Invalid SQL result field number!");
2bb7b766 1886 delete aResult;
1887 return kFALSE;
1888 }
1889
2bb7b766 1890 TSQLRow* aRow;
1891 while ((aRow = aResult->Next())) {
1892 TString runString(aRow->GetField(0), aRow->GetFieldLength(0));
1893 Int_t run = runString.Atoi();
1894
eba76848 1895 AliShuttleLogbookEntry *entry = QueryRunParameters(run);
1896 if (!entry)
1897 continue;
2bb7b766 1898
1899 // loop on detectors
eba76848 1900 for(UInt_t ii = 0; ii < nCols; ii++)
1901 entry->SetDetectorStatus(aResult->GetFieldName(ii), aRow->GetField(ii));
2bb7b766 1902
eba76848 1903 entries.AddLast(entry);
2bb7b766 1904 delete aRow;
1905 }
1906
2bb7b766 1907 delete aResult;
1908 return kTRUE;
1909}
1910
1911//______________________________________________________________________________________________
eba76848 1912AliShuttleLogbookEntry* AliShuttle::QueryRunParameters(Int_t run)
2bb7b766 1913{
eba76848 1914 //
1915 // Retrieve run parameters written in the DAQ logbook and sets them into AliShuttleLogbookEntry object
1916 //
2bb7b766 1917
1918 // check connection, in case connect
be48e3ea 1919 if (!Connect(3))
eba76848 1920 return 0;
2bb7b766 1921
1922 TString sqlQuery;
2c15234c 1923 sqlQuery.Form("select * from %s where run=%d", fConfig->GetDAQlbTable(), run);
2bb7b766 1924
be48e3ea 1925 TSQLResult* aResult = fServer[3]->Query(sqlQuery);
2bb7b766 1926 if (!aResult) {
c88ad5db 1927 Log("SHUTTLE", Form("Can't execute query <%s>!", sqlQuery.Data()));
eba76848 1928 return 0;
2bb7b766 1929 }
1930
eba76848 1931 if (aResult->GetRowCount() == 0) {
2bb7b766 1932 Log("SHUTTLE", Form("QueryRunParameters - No entry in DAQ Logbook for run %d. Skipping", run));
1933 delete aResult;
eba76848 1934 return 0;
2bb7b766 1935 }
1936
eba76848 1937 if (aResult->GetRowCount() > 1) {
c88ad5db 1938 Log("SHUTTLE", Form("QueryRunParameters - UNEXPECTED: "
1939 "more than one entry in DAQ Logbook for run %d!", run));
2bb7b766 1940 delete aResult;
eba76848 1941 return 0;
2bb7b766 1942 }
1943
eba76848 1944 TSQLRow* aRow = aResult->Next();
1945 if (!aRow)
1946 {
c88ad5db 1947 Log("SHUTTLE", Form("QueryRunParameters - Could not retrieve row for run %d. Skipping", run));
eba76848 1948 delete aResult;
1949 return 0;
1950 }
2bb7b766 1951
eba76848 1952 AliShuttleLogbookEntry* entry = new AliShuttleLogbookEntry(run);
2bb7b766 1953
eba76848 1954 for (Int_t ii = 0; ii < aResult->GetFieldCount(); ii++)
1955 entry->SetRunParameter(aResult->GetFieldName(ii), aRow->GetField(ii));
2bb7b766 1956
c9db68c7 1957 delete aRow;
1958 delete aResult;
1959
eba76848 1960 UInt_t startTime = entry->GetStartTime();
1961 UInt_t endTime = entry->GetEndTime();
89bcf3b4 1962 Bool_t ecsSuccess = entry->GetECSSuccess();
c9db68c7 1963
b0e53b15 1964 TString totEventsStr = entry->GetRunParameter("totalEvents");
1965 Int_t totEvents = totEventsStr.Atoi();
c9db68c7 1966
af93e5d7 1967 UInt_t now = time(0);
1968 // TODO make this a configuration parameter
1969 Int_t dcsDelay = 120;
1970
8e828d39 1971 // runs are accepted if they have ecsSuccess set or more than 1 event
af93e5d7 1972 if (startTime != 0 && endTime != 0 && endTime > startTime && (totEvents > 1 || ecsSuccess) && (endTime < now - dcsDelay))
89bcf3b4 1973 {
8e828d39 1974 if (ecsSuccess == kFALSE)
1975 Log("SHUTTLE", Form("Processing run %d although in status ECS failure, Reason: %s", run, entry->GetRunParameter("eor_reason")));
1976 return entry;
89bcf3b4 1977 }
8e828d39 1978
1979 Bool_t skip = kFALSE;
1980
1981 if (totEvents <= 1)
89bcf3b4 1982 {
8e828d39 1983 Log("SHUTTLE", Form("QueryRunParameters - Run %d has 1 event or less - Skipping!", run));
1984 skip = kTRUE;
af93e5d7 1985 }
1986 else if (endTime != 0 && endTime >= now - dcsDelay)
1987 {
1988 Log("SHUTTLE", Form("Skipping run %d for now, because DCS buffer time is not yet expired", run));
1989 }
89bcf3b4 1990 else
1991 {
1992 Log("SHUTTLE", Form("QueryRunParameters - Invalid parameters for Run %d: "
1993 "startTime = %d, endTime = %d. Skipping (Shuttle won't be marked as DONE)!",
1994 run, startTime, endTime));
1995 }
8e828d39 1996
1997 if (skip)
1998 {
1999 Log("SHUTTLE", Form("Marking SHUTTLE skipped for run %d", run));
2000 fLogbookEntry = entry;
2001 if (!UpdateShuttleLogbook("shuttle_skipped"))
2002 {
2003 AliError(Form("Could not update logbook for run %d !", run));
2004 }
2005 fLogbookEntry = 0;
2006 }
c9db68c7 2007
2008 delete entry;
2009 return 0;
2bb7b766 2010}
2011
a038aa70 2012//______________________________________________________________________________________________
2013TMap* AliShuttle::GetValueSet(const char* host, Int_t port, const TSeqCollection* entries,
542b6cc8 2014 DCSType type, Int_t multiSplit)
a038aa70 2015{
2016 // Retrieve all "entry" data points from the DCS server
2017 // host, port: TSocket connection parameters
2018 // entries: list of name of the alias or data point
2019 // type: kAlias or kDP
2020 // returns TMap of values, 0 when failure
542b6cc8 2021
2022 AliDCSClient client(host, port, fTimeout, fRetries, multiSplit);
b41b252a 2023
a038aa70 2024 TMap* result = 0;
b41b252a 2025 if (type == kAlias)
a038aa70 2026 {
b41b252a 2027 result = client.GetAliasValues(entries, GetCurrentStartTime(),
2028 GetCurrentEndTime());
2029 }
2030 else if (type == kDP)
2031 {
2032 result = client.GetDPValues(entries, GetCurrentStartTime(),
2033 GetCurrentEndTime());
2034 }
a038aa70 2035
b41b252a 2036 if (result == 0)
2037 {
2038 Log(fCurrentDetector.Data(), Form("GetValueSet - Can't get entries! Reason: %s",
1790d4b7 2039 client.GetErrorString(client.GetResultErrorCode())));
2040 if (client.GetResultErrorCode() == AliDCSClient::fgkServerError)
2041 Log(fCurrentDetector.Data(), Form("GetValueSet - Server error code: %s",
2042 client.GetServerError().Data()));
a038aa70 2043
b41b252a 2044 return 0;
a038aa70 2045 }
b41b252a 2046
a038aa70 2047 return result;
2048}
b41b252a 2049
b948db8d 2050//______________________________________________________________________________________________
57f50b3c 2051const char* AliShuttle::GetFile(Int_t system, const char* detector,
2052 const char* id, const char* source)
b948db8d 2053{
9827400b 2054 // Get calibration file from file exchange servers
2055 // First queris the FXS database for the file name, using the run, detector, id and source info
2056 // then calls RetrieveFile(filename) for actual copy to local disk
2057 // run: current run being processed (given by Logbook entry fLogbookEntry)
2058 // detector: the Preprocessor name
2059 // id: provided as a parameter by the Preprocessor
2060 // source: provided by the Preprocessor through GetFileSources function
2061
2062 // check if test mode should simulate a FXS error
2063 if (fTestMode & kErrorFXSFiles)
2064 {
2065 Log(detector, Form("GetFile - In TESTMODE - Simulating error while connecting to %s FXS", GetSystemName(system)));
2066 return 0;
2067 }
2068
57f50b3c 2069 // check connection, in case connect
9d733021 2070 if (!Connect(system))
eba76848 2071 {
9d733021 2072 Log(detector, Form("GetFile - Couldn't connect to %s FXS database", GetSystemName(system)));
339adafa 2073 fFXSError = system;
57f50b3c 2074 return 0;
2075 }
2076
2077 // Query preparation
9d733021 2078 TString sourceName(source);
d386d623 2079 Int_t nFields = 3;
2080 TString sqlQueryStart = Form("select filePath,size,fileChecksum from %s where",
2081 fConfig->GetFXSdbTable(system));
2082 TString whereClause = Form("run=%d and detector=\"%s\" and fileId=\"%s\"",
2083 GetCurrentRun(), detector, id);
2084
9d733021 2085 if (system == kDAQ)
2086 {
d386d623 2087 whereClause += Form(" and DAQsource=\"%s\"", source);
57f50b3c 2088 }
9d733021 2089 else if (system == kDCS)
eba76848 2090 {
9d733021 2091 sourceName="none";
57f50b3c 2092 }
9d733021 2093 else if (system == kHLT)
9e080f92 2094 {
d386d623 2095 whereClause += Form(" and DDLnumbers=\"%s\"", source);
9e080f92 2096 }
2097
9e080f92 2098 TString sqlQuery = Form("%s %s", sqlQueryStart.Data(), whereClause.Data());
2099
2100 AliDebug(2, Form("SQL query: \n%s",sqlQuery.Data()));
2101
2102 // Query execution
2103 TSQLResult* aResult = 0;
9d733021 2104 aResult = dynamic_cast<TSQLResult*> (fServer[system]->Query(sqlQuery));
9e080f92 2105 if (!aResult) {
8da81210 2106 Log(detector, Form("GetFile - Can't execute SQL query to %s database for: id = %s, source = %s",
9d733021 2107 GetSystemName(system), id, sourceName.Data()));
339adafa 2108 fFXSError = system;
9e080f92 2109 return 0;
2110 }
2111
8da81210 2112 if (aResult->GetRowCount() == 0)
9e080f92 2113 {
2114 Log(detector,
8da81210 2115 Form("GetFile - No entry in %s FXS db for: id = %s, source = %s",
9d733021 2116 GetSystemName(system), id, sourceName.Data()));
9e080f92 2117 delete aResult;
2118 return 0;
2119 }
2bb7b766 2120
9e080f92 2121 if (aResult->GetRowCount() > 1) {
2122 Log(detector,
8da81210 2123 Form("GetFile - More than one entry in %s FXS db for: id = %s, source = %s",
9d733021 2124 GetSystemName(system), id, sourceName.Data()));
339adafa 2125 fFXSError = system;
9e080f92 2126 delete aResult;
2127 return 0;
2128 }
2129
9d733021 2130 if (aResult->GetFieldCount() != nFields) {
9e080f92 2131 Log(detector,
9d733021 2132 Form("GetFileName - Wrong field count in %s FXS db for: id = %s, source = %s",
2133 GetSystemName(system), id, sourceName.Data()));
339adafa 2134 fFXSError = system;
9e080f92 2135 delete aResult;
2136 return 0;
2137 }
2138
2139 TSQLRow* aRow = dynamic_cast<TSQLRow*> (aResult->Next());
2140
2141 if (!aRow){
8da81210 2142 Log(detector, Form("GetFile - Empty set result in %s FXS db from query: id = %s, source = %s",
9d733021 2143 GetSystemName(system), id, sourceName.Data()));
339adafa 2144 fFXSError = system;
9e080f92 2145 delete aResult;
2146 return 0;
2147 }
2148
2149 TString filePath(aRow->GetField(0), aRow->GetFieldLength(0));
2150 TString fileSize(aRow->GetField(1), aRow->GetFieldLength(1));
d386d623 2151 TString fileChecksum(aRow->GetField(2), aRow->GetFieldLength(2));
9e080f92 2152
2153 delete aResult;
2154 delete aRow;
2155
d386d623 2156 AliDebug(2, Form("filePath = %s; size = %s, fileChecksum = %s",
2157 filePath.Data(), fileSize.Data(), fileChecksum.Data()));
9e080f92 2158
9e080f92 2159 // retrieved file is renamed to make it unique
675f64cd 2160 TString localFileName = Form("%s/%s_%d_process/%s_%s_%d_%s_%s.shuttle",
2161 GetShuttleTempDir(), detector, GetCurrentRun(),
d524ade6 2162 GetSystemName(system), detector, GetCurrentRun(),
2163 id, sourceName.Data());
9d733021 2164
9e080f92 2165
9d733021 2166 // file retrieval from FXS
4b95672b 2167 UInt_t nRetries = 0;
2168 UInt_t maxRetries = 3;
2169 Bool_t result = kFALSE;
2170
2171 // copy!! if successful TSystem::Exec returns 0
8da81210 2172 while (nRetries++ < maxRetries) {
4b95672b 2173 AliDebug(2, Form("Trying to copy file. Retry # %d", nRetries));
2174 result = RetrieveFile(system, filePath.Data(), localFileName.Data());
8da81210 2175 if (!result)
4b95672b 2176 {
8da81210 2177 Log(detector, Form("GetFile - Copy of file %s from %s FXS failed",
9d733021 2178 filePath.Data(), GetSystemName(system)));
4b95672b 2179 continue;
4f0749a8 2180 }
9e080f92 2181
c98417a1 2182 if (fileSize.Length()>0)
2183 {
2184 // compare filesize of local file with the one stored in the FXS DB
926cbc0e 2185 Long_t size = -1;
2186 Int_t sizeComp = gSystem->GetPathInfo(localFileName.Data(), 0, &size, 0, 0);
c98417a1 2187
926cbc0e 2188 if (sizeComp != 0 || size != fileSize.Atoi())
c98417a1 2189 {
8da81210 2190 Log(detector, Form("GetFile - size of file %s does not match with local copy!",
c98417a1 2191 filePath.Data()));
2192 result = kFALSE;
2193 continue;
2194 }
2195
2196 } else {
2197 Log(fCurrentDetector, Form("GetFile - size of file %s not set in %s database, skipping comparison",
2198 filePath.Data(), GetSystemName(system)));
2199 }
2200
d386d623 2201 if (fileChecksum.Length()>0)
4b95672b 2202 {
2203 // compare md5sum of local file with the one stored in the FXS DB
95d4aaef 2204 if(fileChecksum.Contains(' ')) fileChecksum.Resize(fileChecksum.First(' '));
339adafa 2205 Int_t md5Comp = gSystem->Exec(Form("md5sum %s |grep %s > /dev/null 2> /dev/null",
d524ade6 2206 localFileName.Data(), fileChecksum.Data()));
9e080f92 2207
4b95672b 2208 if (md5Comp != 0)
2209 {
8da81210 2210 Log(detector, Form("GetFile - md5sum of file %s does not match with local copy!",
4b95672b 2211 filePath.Data()));
2212 result = kFALSE;
2213 continue;
2214 }
d386d623 2215 } else {
2216 Log(fCurrentDetector, Form("GetFile - md5sum of file %s not set in %s database, skipping comparison",
2217 filePath.Data(), GetSystemName(system)));
9d733021 2218 }
4b95672b 2219 if (result) break;
9e080f92 2220 }
2221
8da81210 2222 if (!result)
2223 {
339adafa 2224 fFXSError = system;
8da81210 2225 return 0;
2226 }
4b95672b 2227
9d733021 2228 fFXSCalled[system]=kTRUE;
2229 TObjString *fileParams = new TObjString(Form("%s#!?!#%s", id, sourceName.Data()));
2230 fFXSlist[system].Add(fileParams);
9e080f92 2231
675f64cd 2232 static TString staticLocalFileName;
2233 staticLocalFileName.Form("%s", localFileName.Data());
2234
c88ad5db 2235 Log(fCurrentDetector, Form("GetFile - Retrieved file with id %s and "
2236 "source %s from %s to %s", id, source,
d524ade6 2237 GetSystemName(system), localFileName.Data()));
675f64cd 2238
d524ade6 2239 return staticLocalFileName.Data();
2bb7b766 2240}
2241
2242//______________________________________________________________________________________________
9d733021 2243Bool_t AliShuttle::RetrieveFile(UInt_t system, const char* fxsFileName, const char* localFileName)
9e080f92 2244{
9827400b 2245 //
2246 // Copies file from FXS to local Shuttle machine
2247 //
2bb7b766 2248
9e080f92 2249 // check temp directory: trying to cd to temp; if it does not exist, create it
d524ade6 2250 AliDebug(2, Form("Copy file %s from %s FXS into %s",
2251 GetSystemName(system), fxsFileName, localFileName));
2252
2253 TString tmpDir(localFileName);
2254
2255 tmpDir = tmpDir(0,tmpDir.Last('/'));
9e080f92 2256
d524ade6 2257 Int_t noDir = gSystem->GetPathInfo(tmpDir.Data(), 0, (Long64_t*) 0, 0, 0);
2258 if (noDir) // temp dir does not exists!
2259 {
2260 if (gSystem->mkdir(tmpDir.Data(), 1))
2261 {
2262 Log(fCurrentDetector.Data(), "RetrieveFile - could not make temp directory!!");
9e080f92 2263 return kFALSE;
2264 }
9e080f92 2265 }
2266
339adafa 2267 TString command = Form("scp -oPort=%d -2 %s@%s:%s/%s %s",
9d733021 2268 fConfig->GetFXSPort(system),
2269 fConfig->GetFXSUser(system),
2270 fConfig->GetFXSHost(system),
339adafa 2271 fConfig->GetFXSBaseFolder(system),
9d733021 2272 fxsFileName,
9e080f92 2273 localFileName);
2274
2275 AliDebug(2, Form("%s",command.Data()));
2276
4b95672b 2277 Bool_t result = (gSystem->Exec(command.Data()) == 0);
9e080f92 2278
4b95672b 2279 return result;
9e080f92 2280}
2281
2282//______________________________________________________________________________________________
9d733021 2283TList* AliShuttle::GetFileSources(Int_t system, const char* detector, const char* id)
2284{
9827400b 2285 //
2286 // Get sources producing the condition file Id from file exchange servers
4a33bdd9 2287 // if id is NULL all sources are returned (distinct)
9827400b 2288 //
1bcd28db 2289
8884d71e 2290 if (id)
2291 {
2292 Log(detector, Form("GetFileSources - Querying %s FXS for files with id %s produced by %s", GetSystemName(system), id, detector));
2293 } else {
2294 Log(detector, Form("GetFileSources - Querying %s FXS for files produced by %s", GetSystemName(system), detector));
2295 }
9827400b 2296
2297 // check if test mode should simulate a FXS error
2298 if (fTestMode & kErrorFXSSources)
2299 {
2300 Log(detector, Form("GetFileSources - In TESTMODE - Simulating error while connecting to %s FXS", GetSystemName(system)));
2301 return 0;
2302 }
2303
9d733021 2304 if (system == kDCS)
2305 {
c88ad5db 2306 Log(detector, "GetFileSources - WARNING: DCS system has only one source of data!");
6297b37d 2307 TList *list = new TList();
2308 list->SetOwner(1);
2309 list->Add(new TObjString(" "));
2310 return list;
9d733021 2311 }
9e080f92 2312
2313 // check connection, in case connect
9d733021 2314 if (!Connect(system))
2315 {
4a33bdd9 2316 Log(detector, Form("GetFileSources - Couldn't connect to %s FXS database", GetSystemName(system)));
339adafa 2317 fFXSError = system;
9d733021 2318 return NULL;
9e080f92 2319 }
2320
9d733021 2321 TString sourceName = 0;
2322 if (system == kDAQ)
2323 {
2324 sourceName = "DAQsource";
2325 } else if (system == kHLT)
2326 {
2327 sourceName = "DDLnumbers";
2328 }
2329
4a33bdd9 2330 TString sqlQueryStart = Form("select distinct %s from %s where", sourceName.Data(), fConfig->GetFXSdbTable(system));
2331 TString whereClause = Form("run=%d and detector=\"%s\"",
2332 GetCurrentRun(), detector);
2333 if (id)
2334 whereClause += Form(" and fileId=\"%s\"", id);
9e080f92 2335 TString sqlQuery = Form("%s %s", sqlQueryStart.Data(), whereClause.Data());
2336
2337 AliDebug(2, Form("SQL query: \n%s",sqlQuery.Data()));
2338
2339 // Query execution
2340 TSQLResult* aResult;
9d733021 2341 aResult = fServer[system]->Query(sqlQuery);
9e080f92 2342 if (!aResult) {
9d733021 2343 Log(detector, Form("GetFileSources - Can't execute SQL query to %s database for id: %s",
2344 GetSystemName(system), id));
339adafa 2345 fFXSError = system;
9e080f92 2346 return 0;
2347 }
2348
86aa42c3 2349 TList *list = new TList();
2350 list->SetOwner(1);
2351
9d733021 2352 if (aResult->GetRowCount() == 0)
2353 {
9e080f92 2354 Log(detector,
9d733021 2355 Form("GetFileSources - No entry in %s FXS table for id: %s", GetSystemName(system), id));
9e080f92 2356 delete aResult;
86aa42c3 2357 return list;
9e080f92 2358 }
2359
1bcd28db 2360 Log(detector, Form("GetFileSources - Found %d sources", aResult->GetRowCount()));
9e080f92 2361
1bcd28db 2362 TSQLRow* aRow;
9d733021 2363 while ((aRow = aResult->Next()))
2364 {
9e080f92 2365
9d733021 2366 TString source(aRow->GetField(0), aRow->GetFieldLength(0));
2367 AliDebug(2, Form("%s = %s", sourceName.Data(), source.Data()));
2368 list->Add(new TObjString(source));
9e080f92 2369 delete aRow;
2370 }
9d733021 2371
9e080f92 2372 delete aResult;
2373
2374 return list;
2bb7b766 2375}
2376
4a33bdd9 2377//______________________________________________________________________________________________
2378TList* AliShuttle::GetFileIDs(Int_t system, const char* detector, const char* source)
2379{
2380 //
2381 // Get all ids of condition files produced by a given source from file exchange servers
2382 //
2383
1bcd28db 2384 Log(detector, Form("GetFileIDs - Retrieving ids with source %s with %s", source, GetSystemName(system)));
2385
4a33bdd9 2386 // check if test mode should simulate a FXS error
2387 if (fTestMode & kErrorFXSSources)
2388 {
2389 Log(detector, Form("GetFileIDs - In TESTMODE - Simulating error while connecting to %s FXS", GetSystemName(system)));
2390 return 0;
2391 }
2392
2393 // check connection, in case connect
2394 if (!Connect(system))
2395 {
2396 Log(detector, Form("GetFileIDs - Couldn't connect to %s FXS database", GetSystemName(system)));
2397 return NULL;
2398 }
2399
2400 TString sourceName = 0;
2401 if (system == kDAQ)
2402 {
2403 sourceName = "DAQsource";
2404 } else if (system == kHLT)
2405 {
2406 sourceName = "DDLnumbers";
2407 }
2408
2409 TString sqlQueryStart = Form("select fileId from %s where", fConfig->GetFXSdbTable(system));
2410 TString whereClause = Form("run=%d and detector=\"%s\"",
2411 GetCurrentRun(), detector);
2412 if (sourceName.Length() > 0 && source)
2413 whereClause += Form(" and %s=\"%s\"", sourceName.Data(), source);
2414 TString sqlQuery = Form("%s %s", sqlQueryStart.Data(), whereClause.Data());
2415
2416 AliDebug(2, Form("SQL query: \n%s",sqlQuery.Data()));
2417
2418 // Query execution
2419 TSQLResult* aResult;
2420 aResult = fServer[system]->Query(sqlQuery);
2421 if (!aResult) {
2422 Log(detector, Form("GetFileIDs - Can't execute SQL query to %s database for source: %s",
2423 GetSystemName(system), source));
2424 return 0;
2425 }
2426
2427 TList *list = new TList();
2428 list->SetOwner(1);
2429
2430 if (aResult->GetRowCount() == 0)
2431 {
2432 Log(detector,
2433 Form("GetFileIDs - No entry in %s FXS table for source: %s", GetSystemName(system), source));
2434 delete aResult;
2435 return list;
2436 }
2437
1bcd28db 2438 Log(detector, Form("GetFileIDs - Found %d ids", aResult->GetRowCount()));
2439
4a33bdd9 2440 TSQLRow* aRow;
2441
2442 while ((aRow = aResult->Next()))
2443 {
2444
2445 TString id(aRow->GetField(0), aRow->GetFieldLength(0));
2446 AliDebug(2, Form("fileId = %s", id.Data()));
2447 list->Add(new TObjString(id));
2448 delete aRow;
2449 }
2450
2451 delete aResult;
2452
2453 return list;
2454}
2455
2bb7b766 2456//______________________________________________________________________________________________
9d733021 2457Bool_t AliShuttle::Connect(Int_t system)
2bb7b766 2458{
9827400b 2459 // Connect to MySQL Server of the system's FXS MySQL databases
2460 // DAQ Logbook, Shuttle Logbook and DAQ FXS db are on the same host
2461 //
57f50b3c 2462
9d733021 2463 // check connection: if already connected return
2464 if(fServer[system] && fServer[system]->IsConnected()) return kTRUE;
57f50b3c 2465
9d733021 2466 TString dbHost, dbUser, dbPass, dbName;
57f50b3c 2467
9d733021 2468 if (system < 3) // FXS db servers
2469 {
2470 dbHost = Form("mysql://%s:%d", fConfig->GetFXSdbHost(system), fConfig->GetFXSdbPort(system));
2471 dbUser = fConfig->GetFXSdbUser(system);
2472 dbPass = fConfig->GetFXSdbPass(system);
2473 dbName = fConfig->GetFXSdbName(system);
2474 } else { // Run & Shuttle logbook servers
2475 // TODO Will the Shuttle logbook server be the same as the Run logbook server ???
2476 dbHost = Form("mysql://%s:%d", fConfig->GetDAQlbHost(), fConfig->GetDAQlbPort());
2477 dbUser = fConfig->GetDAQlbUser();
2478 dbPass = fConfig->GetDAQlbPass();
2479 dbName = fConfig->GetDAQlbDB();
2480 }
57f50b3c 2481
9d733021 2482 fServer[system] = TSQLServer::Connect(dbHost.Data(), dbUser.Data(), dbPass.Data());
2483 if (!fServer[system] || !fServer[system]->IsConnected()) {
2484 if(system < 3)
2485 {
2486 AliError(Form("Can't establish connection to FXS database for %s",
2487 AliShuttleInterface::GetSystemName(system)));
2488 } else {
2489 AliError("Can't establish connection to Run logbook.");
57f50b3c 2490 }
9d733021 2491 if(fServer[system]) delete fServer[system];
2492 return kFALSE;
2bb7b766 2493 }
57f50b3c 2494
9d733021 2495 // Get tables
2496 TSQLResult* aResult=0;
2497 switch(system){
2498 case kDAQ:
2499 aResult = fServer[kDAQ]->GetTables(dbName.Data());
2500 break;
2501 case kDCS:
2502 aResult = fServer[kDCS]->GetTables(dbName.Data());
2503 break;
2504 case kHLT:
2505 aResult = fServer[kHLT]->GetTables(dbName.Data());
2506 break;
2507 default:
2508 aResult = fServer[3]->GetTables(dbName.Data());
2509 break;
2510 }
2511
2512 delete aResult;
2bb7b766 2513 return kTRUE;
2514}
57f50b3c 2515
9e080f92 2516//______________________________________________________________________________________________
9d733021 2517Bool_t AliShuttle::UpdateTable()
9e080f92 2518{
9827400b 2519 //
2520 // Update FXS table filling time_processed field in all rows corresponding to current run and detector
2521 //
9e080f92 2522
9d733021 2523 Bool_t result = kTRUE;
9e080f92 2524
9d733021 2525 for (UInt_t system=0; system<3; system++)
2526 {
2527 if(!fFXSCalled[system]) continue;
9e080f92 2528
9d733021 2529 // check connection, in case connect
2530 if (!Connect(system))
2531 {
2532 Log(fCurrentDetector, Form("UpdateTable - Couldn't connect to %s FXS database", GetSystemName(system)));
2533 result = kFALSE;
2534 continue;
9e080f92 2535 }
9e080f92 2536
9d733021 2537 TTimeStamp now; // now
2538
2539 // Loop on FXS list entries
2540 TIter iter(&fFXSlist[system]);
2541 TObjString *aFXSentry=0;
2542 while ((aFXSentry = dynamic_cast<TObjString*> (iter.Next())))
2543 {
2544 TString aFXSentrystr = aFXSentry->String();
2545 TObjArray *aFXSarray = aFXSentrystr.Tokenize("#!?!#");
2546 if (!aFXSarray || aFXSarray->GetEntries() != 2 )
2547 {
2548 Log(fCurrentDetector, Form("UpdateTable - error updating %s FXS entry. Check string: <%s>",
2549 GetSystemName(system), aFXSentrystr.Data()));
2550 if(aFXSarray) delete aFXSarray;
2551 result = kFALSE;
2552 continue;
2553 }
2554 const char* fileId = ((TObjString*) aFXSarray->At(0))->GetName();
2555 const char* source = ((TObjString*) aFXSarray->At(1))->GetName();
2556
2557 TString whereClause;
2558 if (system == kDAQ)
2559 {
2560 whereClause = Form("where run=%d and detector=\"%s\" and fileId=\"%s\" and DAQsource=\"%s\";",
2561 GetCurrentRun(), fCurrentDetector.Data(), fileId, source);
2562 }
2563 else if (system == kDCS)
2564 {
2565 whereClause = Form("where run=%d and detector=\"%s\" and fileId=\"%s\";",
2566 GetCurrentRun(), fCurrentDetector.Data(), fileId);
2567 }
2568 else if (system == kHLT)
2569 {
2570 whereClause = Form("where run=%d and detector=\"%s\" and fileId=\"%s\" and DDLnumbers=\"%s\";",
2571 GetCurrentRun(), fCurrentDetector.Data(), fileId, source);
2572 }
2573
2574 delete aFXSarray;
9e080f92 2575
9d733021 2576 TString sqlQuery = Form("update %s set time_processed=%d %s", fConfig->GetFXSdbTable(system),
2577 now.GetSec(), whereClause.Data());
9e080f92 2578
9d733021 2579 AliDebug(2, Form("SQL query: \n%s",sqlQuery.Data()));
9e080f92 2580
9d733021 2581 // Query execution
2582 TSQLResult* aResult;
2583 aResult = dynamic_cast<TSQLResult*> (fServer[system]->Query(sqlQuery));
2584 if (!aResult)
2585 {
2586 Log(fCurrentDetector, Form("UpdateTable - %s db: can't execute SQL query <%s>",
2587 GetSystemName(system), sqlQuery.Data()));
2588 result = kFALSE;
2589 continue;
2590 }
2591 delete aResult;
9e080f92 2592 }
9e080f92 2593 }
2594
9d733021 2595 return result;
9e080f92 2596}
57f50b3c 2597
3301427a 2598//______________________________________________________________________________________________
2599Bool_t AliShuttle::UpdateTableFailCase()
2600{
9827400b 2601 // Update FXS table filling time_processed field in all rows corresponding to current run and detector
2602 // this is called in case the preprocessor is declared failed for the current run, because
2603 // the fields are updated only in case of success
3301427a 2604
2605 Bool_t result = kTRUE;
2606
2607 for (UInt_t system=0; system<3; system++)
2608 {
2609 // check connection, in case connect
2610 if (!Connect(system))
2611 {
2612 Log(fCurrentDetector, Form("UpdateTableFailCase - Couldn't connect to %s FXS database",
2613 GetSystemName(system)));
2614 result = kFALSE;
2615 continue;
2616 }
2617
2618 TTimeStamp now; // now
2619
2620 // Loop on FXS list entries
2621
2622 TString whereClause = Form("where run=%d and detector=\"%s\";",
2623 GetCurrentRun(), fCurrentDetector.Data());
2624
2625
2626 TString sqlQuery = Form("update %s set time_processed=%d %s", fConfig->GetFXSdbTable(system),
2627 now.GetSec(), whereClause.Data());
2628
2629 AliDebug(2, Form("SQL query: \n%s",sqlQuery.Data()));
2630
2631 // Query execution
2632 TSQLResult* aResult;
2633 aResult = dynamic_cast<TSQLResult*> (fServer[system]->Query(sqlQuery));
2634 if (!aResult)
2635 {
2636 Log(fCurrentDetector, Form("UpdateTableFailCase - %s db: can't execute SQL query <%s>",
2637 GetSystemName(system), sqlQuery.Data()));
2638 result = kFALSE;
2639 continue;
2640 }
2641 delete aResult;
2642 }
2643
2644 return result;
2645}
2646
2bb7b766 2647//______________________________________________________________________________________________
2648Bool_t AliShuttle::UpdateShuttleLogbook(const char* detector, const char* status)
2649{
e7f62f16 2650 //
2651 // Update Shuttle logbook filling detector or shuttle_done column
2652 // ex. of usage: UpdateShuttleLogbook("PHOS", "DONE") or UpdateShuttleLogbook("shuttle_done")
2653 //
57f50b3c 2654
2bb7b766 2655 // check connection, in case connect
be48e3ea 2656 if(!Connect(3)){
2bb7b766 2657 Log("SHUTTLE", "UpdateShuttleLogbook - Couldn't connect to DAQ Logbook.");
2658 return kFALSE;
57f50b3c 2659 }
2660
2bb7b766 2661 TString detName(detector);
2662 TString setClause;
8e828d39 2663 if (detName == "shuttle_done" || detName == "shuttle_skipped")
e7f62f16 2664 {
2bb7b766 2665 setClause = "set shuttle_done=1";
8e828d39 2666
ee6f7523 2667 if (detName == "shuttle_done")
b0e53b15 2668 {
8e828d39 2669 if (TouchFile() != kTRUE)
2156c755 2670 {
2671 SendMLRunInfo("Pending");
503ff24f 2672 return kFALSE;
2156c755 2673 }
8e828d39 2674
2675 SendMLRunInfo("Done");
b0e53b15 2676 }
8e828d39 2677 else
2678 SendMLRunInfo("Skipped");
2679 }
2680 else {
2bb7b766 2681 TString statusStr(status);
2682 if(statusStr.Contains("done", TString::kIgnoreCase) ||
2683 statusStr.Contains("failed", TString::kIgnoreCase)){
eba76848 2684 setClause = Form("set %s=\"%s\"", detector, status);
2bb7b766 2685 } else {
2686 Log("SHUTTLE",
2687 Form("UpdateShuttleLogbook - Invalid status <%s> for detector %s",
2688 status, detector));
2689 return kFALSE;
2690 }
2691 }
57f50b3c 2692
2bb7b766 2693 TString whereClause = Form("where run=%d", GetCurrentRun());
2694
441b0e9c 2695 TString sqlQuery = Form("update %s %s %s",
2696 fConfig->GetShuttlelbTable(), setClause.Data(), whereClause.Data());
57f50b3c 2697
2bb7b766 2698 AliDebug(2, Form("SQL query: \n%s",sqlQuery.Data()));
2699
2700 // Query execution
2701 TSQLResult* aResult;
be48e3ea 2702 aResult = dynamic_cast<TSQLResult*> (fServer[3]->Query(sqlQuery));
2bb7b766 2703 if (!aResult) {
2704 Log("SHUTTLE", Form("UpdateShuttleLogbook - Can't execute query <%s>", sqlQuery.Data()));
2705 return kFALSE;
57f50b3c 2706 }
2bb7b766 2707 delete aResult;
57f50b3c 2708
2709 return kTRUE;
2710}
2711
2712//______________________________________________________________________________________________
2bb7b766 2713Int_t AliShuttle::GetCurrentRun() const
2714{
9827400b 2715 //
2716 // Get current run from logbook entry
2717 //
57f50b3c 2718
2bb7b766 2719 return fLogbookEntry ? fLogbookEntry->GetRun() : -1;
57f50b3c 2720}
2721
2722//______________________________________________________________________________________________
2bb7b766 2723UInt_t AliShuttle::GetCurrentStartTime() const
2724{
9827400b 2725 //
2726 // get current start time
2727 //
57f50b3c 2728
2bb7b766 2729 return fLogbookEntry ? fLogbookEntry->GetStartTime() : 0;
57f50b3c 2730}
2731
2732//______________________________________________________________________________________________
2bb7b766 2733UInt_t AliShuttle::GetCurrentEndTime() const
2734{
9827400b 2735 //
2736 // get current end time from logbook entry
2737 //
57f50b3c 2738
2bb7b766 2739 return fLogbookEntry ? fLogbookEntry->GetEndTime() : 0;
57f50b3c 2740}
2741
675f64cd 2742//______________________________________________________________________________________________
2743UInt_t AliShuttle::GetCurrentYear() const
2744{
2745 //
2746 // Get current year from logbook entry
2747 //
2748
2749 if (!fLogbookEntry) return 0;
2750
2751 TTimeStamp startTime(GetCurrentStartTime());
2752 TString year = Form("%d",startTime.GetDate());
2753 year = year(0,4);
2754
2755 return year.Atoi();
2756}
2757
2758//______________________________________________________________________________________________
2759const char* AliShuttle::GetLHCPeriod() const
2760{
2761 //
2762 // Get current LHC period from logbook entry
2763 //
2764
2765 if (!fLogbookEntry) return 0;
2766
2767 return fLogbookEntry->GetRunParameter("LHCperiod");
2768}
2769
b948db8d 2770//______________________________________________________________________________________________
2771void AliShuttle::Log(const char* detector, const char* message)
2772{
9827400b 2773 //
2774 // Fill log string with a message
2775 //
b948db8d 2776
7d4cf768 2777 TString logRunDir = GetShuttleLogDir();
2778 if (GetCurrentRun() >=0)
2779 logRunDir += Form("/%d", GetCurrentRun());
2780
2781 void* dir = gSystem->OpenDirectory(logRunDir.Data());
84090f85 2782 if (dir == NULL) {
7d4cf768 2783 if (gSystem->mkdir(logRunDir.Data(), kTRUE)) {
36c99a6a 2784 AliError(Form("Can't open directory <%s>", GetShuttleLogDir()));
84090f85 2785 return;
2786 }
b948db8d 2787
84090f85 2788 } else {
2789 gSystem->FreeDirectory(dir);
2790 }
b948db8d 2791
0b025a52 2792 TString toLog = Form("%s UTC (%d): %s - ", TTimeStamp(time(0)).AsString("s"), getpid(), detector);
e7f62f16 2793 if (GetCurrentRun() >= 0)
2794 toLog += Form("run %d - ", GetCurrentRun());
2bb7b766 2795 toLog += Form("%s", message);
2796
84090f85 2797 AliInfo(toLog.Data());
ffa29e93 2798
2799 // if we redirect the log output already to the file, leave here
2800 if (fOutputRedirected && strcmp(detector, "SHUTTLE") != 0)
2801 return;
b948db8d 2802
ffa29e93 2803 TString fileName = GetLogFileName(detector);
e7f62f16 2804
84090f85 2805 gSystem->ExpandPathName(fileName);
2806
2807 ofstream logFile;
2808 logFile.open(fileName, ofstream::out | ofstream::app);
2809
2810 if (!logFile.is_open()) {
2811 AliError(Form("Could not open file %s", fileName.Data()));
2812 return;
2813 }
7bfb2090 2814
84090f85 2815 logFile << toLog.Data() << "\n";
b948db8d 2816
84090f85 2817 logFile.close();
b948db8d 2818}
2bb7b766 2819
ffa29e93 2820//______________________________________________________________________________________________
2821TString AliShuttle::GetLogFileName(const char* detector) const
2822{
2823 //
2824 // returns the name of the log file for a given sub detector
2825 //
2826
2827 TString fileName;
2828
2829 if (GetCurrentRun() >= 0)
7d4cf768 2830 {
2831 fileName.Form("%s/%d/%s_%d.log", GetShuttleLogDir(), GetCurrentRun(),
2832 detector, GetCurrentRun());
2833 } else {
ffa29e93 2834 fileName.Form("%s/%s.log", GetShuttleLogDir(), detector);
7d4cf768 2835 }
ffa29e93 2836
2837 return fileName;
2838}
2839
ee6f7523 2840//______________________________________________________________________________________________
2841void AliShuttle::SendAlive()
2842{
2843 // sends alive message to ML
2844
2845 TMonaLisaText mlStatus("SHUTTLE_status", "Alive");
2846
2847 TList mlList;
2848 mlList.Add(&mlStatus);
2849
2850 fMonaLisa->SendParameters(&mlList, "__PROCESSINGINFO__");
2851}
2852
2bb7b766 2853//______________________________________________________________________________________________
2854Bool_t AliShuttle::Collect(Int_t run)
2855{
9827400b 2856 //
2857 // Collects conditions data for all UNPROCESSED run written to DAQ LogBook in case of run = -1 (default)
2858 // If a dedicated run is given this run is processed
2859 //
2860 // In operational mode, this is the Shuttle function triggered by the EOR signal.
2861 //
2bb7b766 2862
eba76848 2863 if (run == -1)
2864 Log("SHUTTLE","Collect - Shuttle called. Collecting conditions data for unprocessed runs");
2865 else
2866 Log("SHUTTLE", Form("Collect - Shuttle called. Collecting conditions data for run %d", run));
cb343cfd 2867
2868 SetLastAction("Starting");
2bb7b766 2869
ee6f7523 2870 // create ML instance
2871 if (!fMonaLisa)
2872 fMonaLisa = new TMonaLisaWriter(fConfig->GetMonitorHost(), fConfig->GetMonitorTable());
2873
ee6f7523 2874 SendAlive();
fbc112e3 2875 CountOpenRuns();
ee6f7523 2876
2bb7b766 2877 TString whereClause("where shuttle_done=0");
eba76848 2878 if (run != -1)
2879 whereClause += Form(" and run=%d", run);
2bb7b766 2880
2881 TObjArray shuttleLogbookEntries;
be48e3ea 2882 if (!QueryShuttleLogbook(whereClause, shuttleLogbookEntries))
2883 {
cb343cfd 2884 Log("SHUTTLE", "Collect - Can't retrieve entries from Shuttle logbook");
2bb7b766 2885 return kFALSE;
2886 }
2887
9e080f92 2888 if (shuttleLogbookEntries.GetEntries() == 0)
2889 {
2890 if (run == -1)
2891 Log("SHUTTLE","Collect - Found no UNPROCESSED runs in Shuttle logbook");
2892 else
2893 Log("SHUTTLE", Form("Collect - Run %d is already DONE "
2894 "or it does not exist in Shuttle logbook", run));
2895 return kTRUE;
2896 }
2897
be48e3ea 2898 for (UInt_t iDet=0; iDet<NDetectors(); iDet++)
2899 fFirstUnprocessed[iDet] = kTRUE;
2900
fc5a4708 2901 if (run != -1)
be48e3ea 2902 {
2903 // query Shuttle logbook for earlier runs, check if some detectors are unprocessed,
2904 // flag them into fFirstUnprocessed array
e3ae8d33 2905 TString whereClauseBis(Form("where shuttle_done=0 and run < %d", run));
be48e3ea 2906 TObjArray tmpLogbookEntries;
e3ae8d33 2907 if (!QueryShuttleLogbook(whereClauseBis, tmpLogbookEntries))
be48e3ea 2908 {
2909 Log("SHUTTLE", "Collect - Can't retrieve entries from Shuttle logbook");
2910 return kFALSE;
2911 }
2912
2913 TIter iter(&tmpLogbookEntries);
2914 AliShuttleLogbookEntry* anEntry = 0;
2915 while ((anEntry = dynamic_cast<AliShuttleLogbookEntry*> (iter.Next())))
2916 {
2917 for (UInt_t iDet=0; iDet<NDetectors(); iDet++)
2918 {
2919 if (anEntry->GetDetectorStatus(iDet) == AliShuttleLogbookEntry::kUnprocessed)
2920 {
2921 AliDebug(2, Form("Run %d: setting %s as \"not first time unprocessed\"",
2922 anEntry->GetRun(), GetDetName(iDet)));
2923 fFirstUnprocessed[iDet] = kFALSE;
2924 }
2925 }
2926
2927 }
2928
2929 }
2930
2931 if (!RetrieveConditionsData(shuttleLogbookEntries))
2932 {
cb343cfd 2933 Log("SHUTTLE", "Collect - Process of at least one run failed");
a3ca69fe 2934 CountOpenRuns();
2bb7b766 2935 return kFALSE;
2936 }
2937
36c99a6a 2938 Log("SHUTTLE", "Collect - Requested run(s) successfully processed");
a3ca69fe 2939 CountOpenRuns();
eba76848 2940 return kTRUE;
2bb7b766 2941}
2942
2bb7b766 2943//______________________________________________________________________________________________
2944Bool_t AliShuttle::RetrieveConditionsData(const TObjArray& dateEntries)
2945{
9827400b 2946 //
2947 // Retrieve conditions data for all runs that aren't processed yet
2948 //
2bb7b766 2949
2950 Bool_t hasError = kFALSE;
2951
2952 TIter iter(&dateEntries);
2953 AliShuttleLogbookEntry* anEntry;
2954
2955 while ((anEntry = (AliShuttleLogbookEntry*) iter.Next())){
2956 if (!Process(anEntry)){
2957 hasError = kTRUE;
2958 }
4b95672b 2959
2960 // clean SHUTTLE temp directory
d524ade6 2961 //TString filename = Form("%s/*.shuttle", GetShuttleTempDir());
2962 //RemoveFile(filename.Data());
2bb7b766 2963 }
2964
2965 return hasError == kFALSE;
2966}
cb343cfd 2967
2968//______________________________________________________________________________________________
2969ULong_t AliShuttle::GetTimeOfLastAction() const
2970{
9827400b 2971 //
2972 // Gets time of last action
2973 //
2974
cb343cfd 2975 ULong_t tmp;
36c99a6a 2976
cb343cfd 2977 fMonitoringMutex->Lock();
be48e3ea 2978
cb343cfd 2979 tmp = fLastActionTime;
36c99a6a 2980
cb343cfd 2981 fMonitoringMutex->UnLock();
36c99a6a 2982
cb343cfd 2983 return tmp;
2984}
2985
2986//______________________________________________________________________________________________
2987const TString AliShuttle::GetLastAction() const
2988{
9827400b 2989 //
cb343cfd 2990 // returns a string description of the last action
9827400b 2991 //
cb343cfd 2992
2993 TString tmp;
36c99a6a 2994
cb343cfd 2995 fMonitoringMutex->Lock();
2996
2997 tmp = fLastAction;
2998
2999 fMonitoringMutex->UnLock();
3000
36c99a6a 3001 return tmp;
cb343cfd 3002}
3003
3004//______________________________________________________________________________________________
3005void AliShuttle::SetLastAction(const char* action)
3006{
9827400b 3007 //
cb343cfd 3008 // updates the monitoring variables
9827400b 3009 //
36c99a6a 3010
cb343cfd 3011 fMonitoringMutex->Lock();
36c99a6a 3012
cb343cfd 3013 fLastAction = action;
3014 fLastActionTime = time(0);
3015
3016 fMonitoringMutex->UnLock();
3017}
eba76848 3018
3019//______________________________________________________________________________________________
3020const char* AliShuttle::GetRunParameter(const char* param)
3021{
9827400b 3022 //
3023 // returns run parameter read from DAQ logbook
3024 //
eba76848 3025
3026 if(!fLogbookEntry) {
3027 AliError("No logbook entry!");
3028 return 0;
3029 }
3030
3031 return fLogbookEntry->GetRunParameter(param);
3032}
57c1a579 3033
d386d623 3034//______________________________________________________________________________________________
9827400b 3035AliCDBEntry* AliShuttle::GetFromOCDB(const char* detector, const AliCDBPath& path)
d386d623 3036{
9827400b 3037 //
3038 // returns object from OCDB valid for current run
3039 //
d386d623 3040
9827400b 3041 if (fTestMode & kErrorOCDB)
3042 {
3043 Log(detector, "GetFromOCDB - In TESTMODE - Simulating error with OCDB");
3044 return 0;
3045 }
3046
d386d623 3047 AliCDBStorage *sto = AliCDBManager::Instance()->GetStorage(fgkMainCDB);
3048 if (!sto)
3049 {
9827400b 3050 Log(detector, "GetFromOCDB - Cannot activate main OCDB for query!");
d386d623 3051 return 0;
3052 }
3053
3054 return dynamic_cast<AliCDBEntry*> (sto->Get(path, GetCurrentRun()));
3055}
3056
57c1a579 3057//______________________________________________________________________________________________
339adafa 3058Bool_t AliShuttle::SendMail(EMailTarget target, Int_t system)
57c1a579 3059{
9827400b 3060 //
3061 // sends a mail to the subdetector expert in case of preprocessor error
3062 //
3063
3064 if (fTestMode != kNone)
3065 return kTRUE;
fb40ead4 3066
926cbc0e 3067 if (!fConfig->SendMail())
3068 return kTRUE;
57c1a579 3069
339adafa 3070 if (target == kDCSEMail || target == kFXSEMail) {
3071 if (!fFirstProcessing)
3072 return kTRUE;
36c99a6a 3073 }
3074
7bdfe1fa 3075 Int_t runMode = (Int_t)fConfig->GetRunMode();
3076 TString tmpStr;
3077 if (runMode == 0) tmpStr = " Nightly Test:";
3078 else tmpStr = " Data Taking:";
fbc112e3 3079 void* dir = gSystem->OpenDirectory(GetShuttleLogDir());
3080 if (dir == NULL)
3081 {
3082 if (gSystem->mkdir(GetShuttleLogDir(), kTRUE))
3083 {
3084 Log("SHUTTLE", Form("SendMail - Can't open directory <%s>", GetShuttleLogDir()));
3085 return kFALSE;
3086 }
3087
3088 } else {
3089 gSystem->FreeDirectory(dir);
3090 }
3091
339adafa 3092 // det experts in to
3093 TString to="";
3835a613 3094 TIter *iterExperts = 0;
339adafa 3095 if (target == kDCSEMail) {
3096 iterExperts = new TIter(fConfig->GetAdmins(AliShuttleConfig::kAmanda));
3097 }
3098 else if (target == kFXSEMail) {
3099 iterExperts = new TIter(fConfig->GetAdmins(system));
3100 }
3835a613 3101 if (iterExperts) {
3102 TObjString *anExpert=0;
3103 while ((anExpert = (TObjString*) iterExperts->Next()))
3104 {
3105 to += Form("%s,", anExpert->GetName());
3106 }
3107 delete iterExperts;
339adafa 3108 }
3835a613 3109
3110 // add subdetector experts
3111 iterExperts = new TIter(fConfig->GetResponsibles(fCurrentDetector));
339adafa 3112 TObjString *anExpert=0;
3113 while ((anExpert = (TObjString*) iterExperts->Next()))
4508ab8b 3114 {
339adafa 3115 to += Form("%s,", anExpert->GetName());
3116 }
3117 delete iterExperts;
3835a613 3118
339adafa 3119 if (to.Length() > 0)
3120 to.Remove(to.Length()-1);
3121 AliDebug(2, Form("to: %s",to.Data()));
4508ab8b 3122
339adafa 3123 if (to.IsNull()) {
3124 Log("SHUTTLE", Form("List of %d responsibles not set!", (Int_t) target));
3125 return kFALSE;
3126 }
3127
3128 // SHUTTLE responsibles in cc
fb40ead4 3129 TString cc="";
3130 TIter iterAdmins(fConfig->GetAdmins(AliShuttleConfig::kGlobal));
3131 TObjString *anAdmin=0;
3132 while ((anAdmin = (TObjString*) iterAdmins.Next()))
3133 {
3134 cc += Form("%s,", anAdmin->GetName());
3135 }
3136 if (cc.Length() > 0)
c19963db 3137 cc.Remove(cc.Length()-1);
fb40ead4 3138 AliDebug(2, Form("cc: %s",to.Data()));
57c1a579 3139
339adafa 3140 // mail body
675f64cd 3141 TString bodyFileName;
3142 bodyFileName.Form("%s/mail.body", GetShuttleLogDir());
3143 gSystem->ExpandPathName(bodyFileName);
3144
3145 ofstream mailBody;
3146 mailBody.open(bodyFileName, ofstream::out);
3147
3148 if (!mailBody.is_open())
3149 {
339adafa 3150 Log("SHUTTLE", Form("Could not open mail body file %s", bodyFileName.Data()));
675f64cd 3151 return kFALSE;
3152 }
3153
675f64cd 3154
339adafa 3155 TString subject;
3156 TString body;
675f64cd 3157
339adafa 3158 if (target == kDCSEMail){
af93e5d7 3159 subject = Form("%s CRITICAL Retrieval of data points for %s FAILED in run %d !",
7bdfe1fa 3160 tmpStr.Data(), fCurrentDetector.Data(), GetCurrentRun());
339adafa 3161 AliDebug(2, Form("subject: %s", subject.Data()));
3162
3163 body = Form("Dear DCS experts, \n\n");
3164 body += Form("SHUTTLE couldn\'t retrieve the data points for detector %s "
3165 "in run %d!!\n\n", fCurrentDetector.Data(), GetCurrentRun());
fb40ead4 3166 }
339adafa 3167 else if (target == kFXSEMail){
af93e5d7 3168 subject = Form("%s CRITICAL FXS communication for %s FAILED in run %d !",
7bdfe1fa 3169 tmpStr.Data(), fCurrentDetector.Data(), GetCurrentRun());
339adafa 3170 AliDebug(2, Form("subject: %s", subject.Data()));
3171 TString sys;
3172 if (system == kDAQ) sys="DAQ";
3173 else if (system == kDCS) sys="DCS";
3174 else if (system == kHLT) sys="HLT";
3175 else return kFALSE;
9c0e2d64 3176 body = Form("Dear %s FXS experts, \n\n",sys.Data());
339adafa 3177 body += Form("SHUTTLE couldn\'t retrieve data from the FXS for detector %s "
9c0e2d64 3178 "in run %d!!\n\n", fCurrentDetector.Data(), GetCurrentRun());
339adafa 3179 }
3180 else {
7bdfe1fa 3181 subject = Form("%s %s Shuttle preprocessor FAILED in run %d (run type = %s)!",
3182 tmpStr.Data(), fCurrentDetector.Data(), GetCurrentRun(), GetRunType());
339adafa 3183 AliDebug(2, Form("subject: %s", subject.Data()));
3184
3185 body = Form("Dear %s expert(s), \n\n", fCurrentDetector.Data());
3186 body += Form("SHUTTLE just detected that your preprocessor "
3187 "failed processing run %d (run type = %s)!!\n\n",
3188 GetCurrentRun(), GetRunType());
3189 }
675f64cd 3190
7d4cf768 3191 body += Form("Please check %s status on the SHUTTLE monitoring page: \n\n",
3192 fCurrentDetector.Data());
b0e53b15 3193 if (fConfig->GetRunMode() == AliShuttleConfig::kTest)
3194 {
8e828d39 3195 body += Form("\thttp://pcalimonitor.cern.ch/shuttle.jsp?time=24 \n\n");
b0e53b15 3196 } else {
8e828d39 3197 body += Form("\thttp://pcalimonitor.cern.ch/shuttle.jsp?instance=PROD&time=24 \n\n");
b0e53b15 3198 }
339adafa 3199
3200
7d4cf768 3201 TString logFolder = "logs";
3202 if (fConfig->GetRunMode() == AliShuttleConfig::kProd)
3203 logFolder += "_PROD";
3204
3205
675f64cd 3206 body += Form("Find the %s log for the current run on \n\n"
7d4cf768 3207 "\thttp://pcalishuttle01.cern.ch:8880/%s/%d/%s_%d.log \n\n",
3208 fCurrentDetector.Data(), logFolder.Data(), GetCurrentRun(),
3209 fCurrentDetector.Data(), GetCurrentRun());
3210 body += Form("The last 10 lines of %s log file are following:\n\n", fCurrentDetector.Data());
675f64cd 3211
3212 AliDebug(2, Form("Body begin: %s", body.Data()));
3213
3214 mailBody << body.Data();
3215 mailBody.close();
3216 mailBody.open(bodyFileName, ofstream::out | ofstream::app);
3217
339adafa 3218 TString logFileName = Form("%s/%d/%s_%d.log", GetShuttleLogDir(),
3219 GetCurrentRun(), fCurrentDetector.Data(), GetCurrentRun());
675f64cd 3220 TString tailCommand = Form("tail -n 10 %s >> %s", logFileName.Data(), bodyFileName.Data());
3221 if (gSystem->Exec(tailCommand.Data()))
3222 {
3223 mailBody << Form("%s log file not found ...\n\n", fCurrentDetector.Data());
3224 }
3225
3226 TString endBody = Form("------------------------------------------------------\n\n");
36c99a6a 3227 endBody += Form("In case of problems please contact the SHUTTLE core team.\n\n");
3228 endBody += "Please do not answer this message directly, it is automatically generated.\n\n";
546242fb 3229 endBody += "Greetings,\n\n \t\t\tthe SHUTTLE\n";
57c1a579 3230
909732f7 3231 AliDebug(2, Form("Body end: %s", endBody.Data()));
57c1a579 3232
3233 mailBody << endBody.Data();
3234
3235 mailBody.close();
3236
3237 // send mail!
3238 TString mailCommand = Form("mail -s \"%s\" -c %s %s < %s",
3239 subject.Data(),
3240 cc.Data(),
3241 to.Data(),
3242 bodyFileName.Data());
909732f7 3243 AliDebug(2, Form("mail command: %s", mailCommand.Data()));
57c1a579 3244
3245 Bool_t result = gSystem->Exec(mailCommand.Data());
3246
3247 return result == 0;
3248}
d386d623 3249
441b0e9c 3250//______________________________________________________________________________________________
9827400b 3251const char* AliShuttle::GetRunType()
441b0e9c 3252{
9827400b 3253 //
3254 // returns run type read from "run type" logbook
3255 //
441b0e9c 3256
3257 if(!fLogbookEntry) {
3258 AliError("No logbook entry!");
3259 return 0;
3260 }
3261
9827400b 3262 return fLogbookEntry->GetRunType();
441b0e9c 3263}
3264
4859271b 3265//______________________________________________________________________________________________
3266Bool_t AliShuttle::GetHLTStatus()
3267{
3268 // Return HLT status (ON=1 OFF=0)
af76b20d 3269 // Converts the HLT status from the mode string read in the run logbook (not just a bool)
4859271b 3270
3271 if(!fLogbookEntry) {
3272 AliError("No logbook entry!");
3273 return 0;
3274 }
3275
af76b20d 3276 // TODO implement when HLTMode is inserted in run logbook
b4ebc2d5 3277 TString hltMode = fLogbookEntry->GetRunParameter("HLTmode");
af76b20d 3278 TSubString firstChar = hltMode(0,1);
3279 AliDebug(2,Form("First char = %s ",firstChar.Data()));
3280 if (firstChar == "A") {
3281 return kFALSE;
3282 }
3283 else if ((firstChar == "B") || (firstChar == "C") || (firstChar == "D") || (firstChar == "E")) {
3284 return kTRUE;
3285 }
3286 else {
3287 Log("SHUTTLE","Unexpected HLT mode! Returning 0....");
3288 return kFALSE;
3289 }
4859271b 3290}
ca8ea066 3291
3292//______________________________________________________________________________________________
3293const char* AliShuttle::GetTriggerConfiguration()
3294{
3295 // Receives the trigger configuration from the DAQ logbook for the current run
3296
3297 // check connection, if needed reconnect
3298 if (!Connect(3))
3299 return 0;
3300
3301 TString sqlQuery;
3302 sqlQuery.Form("SELECT configFile FROM logbook_trigger_config WHERE run = %d", GetCurrentRun());
3303 TSQLResult* result = fServer[3]->Query(sqlQuery);
3304 if (!result)
3305 {
3306 Log("SHUTTLE", Form("ERROR: Can't execute query <%s>!", sqlQuery.Data()));
3307 return 0;
3308 }
3309
3310 if (result->GetRowCount() == 0)
3311 {
3312 Log("SHUTTLE", "ERROR: Trigger configuration not found in logbook_trigger_config");
3313 delete result;
3314 return 0;
3315 }
3316
3317 TSQLRow* row = result->Next();
3318 if (!row)
3319 {
3320 Log("SHUTTLE", "ERROR: Could not receive logbook_trigger_config data");
3321 delete result;
3322 return 0;
3323 }
3324
3325 // static, so that pointer remains valid when it is returned to the calling class
3326 static TString triggerConfig(row->GetField(0));
3327
3328 delete row;
3329 row = 0;
3330
3331 delete result;
3332 result = 0;
3333
3334 Log("SHUTTLE", Form("Found trigger configuration: %s", triggerConfig.Data()));
3335
3336 return triggerConfig;
3337}
3338
d386d623 3339//______________________________________________________________________________________________
3340void AliShuttle::SetShuttleTempDir(const char* tmpDir)
3341{
9827400b 3342 //
3343 // sets Shuttle temp directory
3344 //
d386d623 3345
3346 fgkShuttleTempDir = gSystem->ExpandPathName(tmpDir);
3347}
3348
3349//______________________________________________________________________________________________
3350void AliShuttle::SetShuttleLogDir(const char* logDir)
3351{
9827400b 3352 //
3353 // sets Shuttle log directory
3354 //
d386d623 3355
3356 fgkShuttleLogDir = gSystem->ExpandPathName(logDir);
3357}
503ff24f 3358//______________________________________________________________________________________________
3359Bool_t AliShuttle::TouchFile()
3360{
3361 //
3362 // touching a file on the grid if run has been DONE
3363 //
3364
3365 if (!gGrid)
3366 {
3367 Log("SHUTTLE",Form("No TGrid connection estabilished!"));
3368 Log("SHUTTLE",Form("Could not touch file for run %i",GetCurrentRun()));
3369 return kFALSE;
3370 }
3371
b8cf5e0f 3372 TString dir;
3373 dir.Form("%s%d/%s/SHUTTLE_DONE", fConfig->GetAlienPath(), GetCurrentYear(), GetLHCPeriod());
3374 // checking whether directory for touch command exists
3375 TString commandLs;
3376 commandLs.Form("ls %s",dir.Data());
3377 TGridResult *resultLs = dynamic_cast<TGridResult*>(gGrid->Command(commandLs));
3378 if (!resultLs){
3379 Log("SHUTTLE",Form("No result for %s command, returning without touching",commandLs.Data()));
3380 return kFALSE;
3381 }
3382 TMap *mapLs = dynamic_cast<TMap*>(resultLs->At(0));
3383 if (!mapLs){
3384 Log("SHUTTLE",Form("No map for %s command, returning without touching",commandLs.Data()));
97d38a5a 3385 delete resultLs;
3386 resultLs = 0x0;
b8cf5e0f 3387 return kFALSE;
3388 }
3389 TObjString *valueLsPath = dynamic_cast<TObjString*>(mapLs->GetValue("path"));
3390 if (!valueLsPath || (TString)(valueLsPath->GetString()).CompareTo(dir)!=1){
3391 Log("SHUTTLE",Form("No directory %s found, creating it",dir.Data()));
3392
3393 // creating the directory
3394
3395 Bool_t boolMkdir = gGrid->Mkdir(dir.Data());
3396 if (!boolMkdir) {
3397 Log("SHUTTLE",Form("Impossible to create dir %s in alien catalogue for run %i!",dir.Data(),GetCurrentRun()));
97d38a5a 3398 delete resultLs;
3399 resultLs = 0x0;
b8cf5e0f 3400 return kFALSE;
3401 }
3402 Log("SHUTTLE",Form("Directory %s successfully created in alien catalogue for run %i",dir.Data(),GetCurrentRun()));
3403 }
3404 else {
3405 Log("SHUTTLE",Form("Directory %s correctly found for run %i",dir.Data(),GetCurrentRun()));
3406 }
3407
97d38a5a 3408 delete resultLs;
3409 resultLs = 0x0;
3410
503ff24f 3411 TString command;
b8cf5e0f 3412 command.Form("touch %s/%i", dir.Data(), GetCurrentRun());
e660ff60 3413 Log("SHUTTLE", Form("Creating entry in file catalog: %s", command.Data()));
503ff24f 3414 TGridResult *resultTouch = dynamic_cast<TGridResult*>(gGrid->Command(command));
b8cf5e0f 3415 if (!resultTouch){
3416 Log("SHUTTLE",Form("No result for touching command, returning without touching for run %i",GetCurrentRun()));
3417 return kFALSE;
503ff24f 3418 }
b8cf5e0f 3419 TMap *mapTouch = dynamic_cast<TMap*>(resultTouch->At(0));
3420 if (!mapTouch){
3421 Log("SHUTTLE",Form("No map for touching command, returning without touching for run %i",GetCurrentRun()));
97d38a5a 3422 delete resultTouch;
3423 resultTouch = 0x0;
b8cf5e0f 3424 return kFALSE;
3425 }
3426 TObjString *valueTouch = dynamic_cast<TObjString*>(mapTouch->GetValue("__result__"));
3427 if (!valueTouch){
3428 Log("SHUTTLE",Form("No value for \"__result__\" key set in the map for touching command, returning without touching for run %i",GetCurrentRun()));
97d38a5a 3429 delete resultTouch;
3430 resultTouch = 0x0;
b8cf5e0f 3431 return kFALSE;
503ff24f 3432 }
b8cf5e0f 3433 if (valueTouch->GetString()!="1"){
3434 Log("SHUTTLE",Form("Failing the touching command, returning without touching for run %i",GetCurrentRun()));
97d38a5a 3435 delete resultTouch;
3436 resultTouch = 0x0;
b8cf5e0f 3437 return kFALSE;
3438 }
97d38a5a 3439 delete resultTouch;
3440 resultTouch = 0x0;
98400bd9 3441 Log("SHUTTLE", "Sucessfully touched the file");
b8cf5e0f 3442 return kTRUE;
503ff24f 3443}
3444
3445