Method for touching a file on the grid after processing a run in case of shuttle...
[u/mrichter/AliRoot.git] / SHUTTLE / AliShuttle.cxx
CommitLineData
73abe331 1/**************************************************************************
2 * Copyright(c) 1998-1999, ALICE Experiment at CERN, All rights reserved. *
3 * *
4 * Author: The ALICE Off-line Project. *
5 * Contributors are mentioned in the code where appropriate. *
6 * *
7 * Permission to use, copy, modify and distribute this software and its *
8 * documentation strictly for non-commercial purposes is hereby granted *
9 * without fee, provided that the above copyright notice appears in all *
10 * copies and that both the copyright notice and this permission notice *
11 * appear in the supporting documentation. The authors make no claims *
12 * about the suitability of this software for any purpose. It is *
13 * provided "as is" without express or implied warranty. *
14 **************************************************************************/
15
ff3781ad 16/* $Id$ */
73abe331 17
18//
19// This class is the main manager for AliShuttle.
20// It organizes the data retrieval from DCS and call the
b948db8d 21// interface methods of AliPreprocessor.
73abe331 22// For every detector in AliShuttleConfgi (see AliShuttleConfig),
23// data for its set of aliases is retrieved. If there is registered
b948db8d 24// AliPreprocessor for this detector then it will be used
25// accroding to the schema (see AliPreprocessor).
26// If there isn't registered AliPreprocessor than the retrieved
73abe331 27// data is stored automatically to the undelying AliCDBStorage.
28// For detSpec is used the alias name.
29//
30
31#include "AliShuttle.h"
32
33#include "AliCDBManager.h"
34#include "AliCDBStorage.h"
35#include "AliCDBId.h"
84090f85 36#include "AliCDBRunRange.h"
37#include "AliCDBPath.h"
5164a766 38#include "AliCDBEntry.h"
73abe331 39#include "AliShuttleConfig.h"
eba76848 40#include "DCSClient/AliDCSClient.h"
73abe331 41#include "AliLog.h"
b948db8d 42#include "AliPreprocessor.h"
5164a766 43#include "AliShuttleStatus.h"
2bb7b766 44#include "AliShuttleLogbookEntry.h"
73abe331 45
57f50b3c 46#include <TSystem.h>
58bc3020 47#include <TObject.h>
b948db8d 48#include <TString.h>
57f50b3c 49#include <TTimeStamp.h>
73abe331 50#include <TObjString.h>
57f50b3c 51#include <TSQLServer.h>
52#include <TSQLResult.h>
53#include <TSQLRow.h>
cb343cfd 54#include <TMutex.h>
9827400b 55#include <TSystemDirectory.h>
56#include <TSystemFile.h>
a986b218 57#include <TFile.h>
9827400b 58#include <TGrid.h>
59#include <TGridResult.h>
503ff24f 60#include <TMap.h>
73abe331 61
e7f62f16 62#include <TMonaLisaWriter.h>
63
5164a766 64#include <fstream>
65
cb343cfd 66#include <sys/types.h>
67#include <sys/wait.h>
68
1bd183ba 69#include <signal.h>
70
73abe331 71ClassImp(AliShuttle)
72
b948db8d 73//______________________________________________________________________________________________
74AliShuttle::AliShuttle(const AliShuttleConfig* config,
75 UInt_t timeout, Int_t retries):
4f0ab988 76fConfig(config),
77fTimeout(timeout), fRetries(retries),
78fPreprocessorMap(),
2bb7b766 79fLogbookEntry(0),
eba76848 80fCurrentDetector(),
926cbc0e 81fFirstProcessing(0),
339adafa 82fFXSError(-1),
85a80aa9 83fStatusEntry(0),
cb343cfd 84fMonitoringMutex(0),
eba76848 85fLastActionTime(0),
e7f62f16 86fLastAction(),
9827400b 87fMonaLisa(0),
88fTestMode(kNone),
ffa29e93 89fReadTestMode(kFALSE),
90fOutputRedirected(kFALSE)
73abe331 91{
92 //
93 // config: AliShuttleConfig used
73abe331 94 // timeout: timeout used for AliDCSClient connection
95 // retries: the number of retries in case of connection error.
96 //
97
57f50b3c 98 if (!fConfig->IsValid()) AliFatal("********** !!!!! Invalid configuration !!!!! **********");
be48e3ea 99 for(int iSys=0;iSys<4;iSys++) {
57f50b3c 100 fServer[iSys]=0;
be48e3ea 101 if (iSys < 3)
2c15234c 102 fFXSlist[iSys].SetOwner(kTRUE);
57f50b3c 103 }
2bb7b766 104 fPreprocessorMap.SetOwner(kTRUE);
be48e3ea 105
106 for (UInt_t iDet=0; iDet<NDetectors(); iDet++)
107 fFirstUnprocessed[iDet] = kFALSE;
108
cb343cfd 109 fMonitoringMutex = new TMutex();
58bc3020 110}
111
b948db8d 112//______________________________________________________________________________________________
57f50b3c 113AliShuttle::~AliShuttle()
58bc3020 114{
9827400b 115 //
116 // destructor
117 //
58bc3020 118
b948db8d 119 fPreprocessorMap.DeleteAll();
be48e3ea 120 for(int iSys=0;iSys<4;iSys++)
57f50b3c 121 if(fServer[iSys]) {
122 fServer[iSys]->Close();
123 delete fServer[iSys];
eba76848 124 fServer[iSys] = 0;
57f50b3c 125 }
2bb7b766 126
127 if (fStatusEntry){
128 delete fStatusEntry;
129 fStatusEntry = 0;
130 }
cb343cfd 131
132 if (fMonitoringMutex)
133 {
134 delete fMonitoringMutex;
135 fMonitoringMutex = 0;
136 }
73abe331 137}
138
b948db8d 139//______________________________________________________________________________________________
57f50b3c 140void AliShuttle::RegisterPreprocessor(AliPreprocessor* preprocessor)
58bc3020 141{
73abe331 142 //
b948db8d 143 // Registers new AliPreprocessor.
73abe331 144 // It uses GetName() for indentificator of the pre processor.
145 // The pre processor is registered it there isn't any other
146 // with the same identificator (GetName()).
147 //
148
eba76848 149 const char* detName = preprocessor->GetName();
150 if(GetDetPos(detName) < 0)
151 AliFatal(Form("********** !!!!! Invalid detector name: %s !!!!! **********", detName));
152
153 if (fPreprocessorMap.GetValue(detName)) {
154 AliWarning(Form("AliPreprocessor %s is already registered!", detName));
73abe331 155 return;
156 }
157
eba76848 158 fPreprocessorMap.Add(new TObjString(detName), preprocessor);
73abe331 159}
b948db8d 160//______________________________________________________________________________________________
3301427a 161Bool_t AliShuttle::Store(const AliCDBPath& path, TObject* object,
84090f85 162 AliCDBMetaData* metaData, Int_t validityStart, Bool_t validityInfinite)
73abe331 163{
9827400b 164 // Stores a CDB object in the storage for offline reconstruction. Objects that are not needed for
165 // offline reconstruction, but should be stored anyway (e.g. for debugging) should NOT be stored
166 // using this function. Use StoreReferenceData instead!
167 // It calls StoreLocally function which temporarily stores the data locally; when the preprocessor
168 // finishes the data are transferred to the main storage (Grid).
b948db8d 169
3301427a 170 return StoreLocally(fgkLocalCDB, path, object, metaData, validityStart, validityInfinite);
84090f85 171}
172
173//______________________________________________________________________________________________
3301427a 174Bool_t AliShuttle::StoreReferenceData(const AliCDBPath& path, TObject* object, AliCDBMetaData* metaData)
84090f85 175{
9827400b 176 // Stores a CDB object in the storage for reference data. This objects will not be available during
177 // offline reconstrunction. Use this function for reference data only!
178 // It calls StoreLocally function which temporarily stores the data locally; when the preprocessor
179 // finishes the data are transferred to the main storage (Grid).
85a80aa9 180
3301427a 181 return StoreLocally(fgkLocalRefStorage, path, object, metaData);
85a80aa9 182}
183
184//______________________________________________________________________________________________
3301427a 185Bool_t AliShuttle::StoreLocally(const TString& localUri,
85a80aa9 186 const AliCDBPath& path, TObject* object, AliCDBMetaData* metaData,
187 Int_t validityStart, Bool_t validityInfinite)
188{
9827400b 189 // Store object temporarily in local storage. Parameters are passed by Store and StoreReferenceData functions.
190 // when the preprocessor finishes the data are transferred to the main storage (Grid).
191 // The parameters are:
192 // 1) Uri of the backup storage (Local)
193 // 2) the object's path.
194 // 3) the object to be stored
195 // 4) the metaData to be associated with the object
196 // 5) the validity start run number w.r.t. the current run,
197 // if the data is valid only for this run leave the default 0
198 // 6) specifies if the calibration data is valid for infinity (this means until updated),
199 // typical for calibration runs, the default is kFALSE
200 //
201 // returns 0 if fail, 1 otherwise
84090f85 202
9827400b 203 if (fTestMode & kErrorStorage)
204 {
205 Log(fCurrentDetector, "StoreLocally - In TESTMODE - Simulating error while storing locally");
206 return kFALSE;
207 }
208
3301427a 209 const char* cdbType = (localUri == fgkLocalCDB) ? "CDB" : "Reference";
2bb7b766 210
85a80aa9 211 Int_t firstRun = GetCurrentRun() - validityStart;
84090f85 212 if(firstRun < 0) {
9827400b 213 AliWarning("First valid run happens to be less than 0! Setting it to 0.");
84090f85 214 firstRun=0;
215 }
216
217 Int_t lastRun = -1;
218 if(validityInfinite) {
219 lastRun = AliCDBRunRange::Infinity();
220 } else {
221 lastRun = GetCurrentRun();
222 }
223
3301427a 224 // Version is set to current run, it will be used later to transfer data to Grid
225 AliCDBId id(path, firstRun, lastRun, GetCurrentRun(), -1);
2bb7b766 226
227 if(! dynamic_cast<TObjString*> (metaData->GetProperty("RunUsed(TObjString)"))){
228 TObjString runUsed = Form("%d", GetCurrentRun());
9e080f92 229 metaData->SetProperty("RunUsed(TObjString)", runUsed.Clone());
2bb7b766 230 }
84090f85 231
3301427a 232 Bool_t result = kFALSE;
84090f85 233
3301427a 234 if (!(AliCDBManager::Instance()->GetStorage(localUri))) {
235 Log("SHUTTLE", Form("StoreLocally - Cannot activate local %s storage", cdbType));
84090f85 236 } else {
3301427a 237 result = AliCDBManager::Instance()->GetStorage(localUri)
84090f85 238 ->Put(object, id, metaData);
239 }
240
241 if(!result) {
242
9827400b 243 Log(fCurrentDetector, Form("StoreLocally - Can't store object <%s>!", id.ToString().Data()));
3301427a 244 }
2bb7b766 245
3301427a 246 return result;
247}
84090f85 248
3301427a 249//______________________________________________________________________________________________
250Bool_t AliShuttle::StoreOCDB()
251{
9827400b 252 //
253 // Called when preprocessor ends successfully or when previous storage attempt failed (kStoreError status)
254 // Calls underlying StoreOCDB(const char*) function twice, for OCDB and Reference storage.
255 // Then calls StoreRefFilesToGrid to store reference files.
256 //
257
5e993b6f 258 UpdateShuttleStatus(AliShuttleStatus::kStoreStarted);
259
9827400b 260 if (fTestMode & kErrorGrid)
261 {
262 Log("SHUTTLE", "StoreOCDB - In TESTMODE - Simulating error while storing in the Grid");
263 Log(fCurrentDetector, "StoreOCDB - In TESTMODE - Simulating error while storing in the Grid");
264 return kFALSE;
265 }
266
c88ad5db 267 Log("SHUTTLE","StoreOCDB - Storing OCDB data ...");
5e993b6f 268 Int_t resultCDB = StoreOCDB(fgkMainCDB);
86aa42c3 269
c88ad5db 270 Log("SHUTTLE","StoreOCDB - Storing reference data ...");
5e993b6f 271 Int_t resultRef = StoreOCDB(fgkMainRefStorage);
9827400b 272
c88ad5db 273 Log("SHUTTLE","StoreOCDB - Storing reference files ...");
274 Bool_t resultRefFiles = CopyFilesToGrid("reference");
275
276 Bool_t resultMetadata = kTRUE;
277 if(fCurrentDetector == "GRP")
278 {
279 Log("StoreOCDB - SHUTTLE","Storing Run Metadata file ...");
280 resultMetadata = CopyFilesToGrid("metadata");
281 }
9827400b 282
5e993b6f 283 Int_t storeResult = 0;
284
285 if (resultCDB < 0 || resultRef < 0 || resultRefFiles == kFALSE || resultMetadata == kFALSE)
286 storeResult = -1;
287 else if (resultCDB > 0 || resultRef > 0)
288 storeResult = 1;
289
290 if (storeResult < 0)
291 {
292 Log("SHUTTLE",
293 Form("\t\t\t****** run %d - %s: STORAGE ERROR ******",
294 GetCurrentRun(), fCurrentDetector.Data()));
295 UpdateShuttleStatus(AliShuttleStatus::kStoreError);
296 }
297 else if (storeResult > 0)
298 {
299 Log("SHUTTLE",
300 Form("\t\t\t****** run %d - %s: STORAGE DELAYED ******",
301 GetCurrentRun(), fCurrentDetector.Data()));
302 UpdateShuttleStatus(AliShuttleStatus::kStoreDelayed);
303 }
304 else if (storeResult == 0)
305 {
306 Log("SHUTTLE",
307 Form("\t\t\t****** run %d - %s: DONE ******",
308 GetCurrentRun(), fCurrentDetector.Data()));
309 UpdateShuttleStatus(AliShuttleStatus::kDone);
310 UpdateShuttleLogbook(fCurrentDetector, "DONE");
311 }
312
313 return (storeResult == 0);
3301427a 314}
315
316//______________________________________________________________________________________________
5e993b6f 317Int_t AliShuttle::StoreOCDB(const TString& gridURI)
3301427a 318{
319 //
320 // Called by StoreOCDB(), performs actual storage to the main OCDB and reference storages (Grid)
321 //
5e993b6f 322 // Return code:
323 // -2 initialization error
324 // -1 storage error
325 // 0 success
326 // 1 storage delayed (e.g. previous unprocessed runs)
327 //
3301427a 328
329 TObjArray* gridIds=0;
330
331 Bool_t result = kTRUE;
5e993b6f 332 Bool_t delayed = kFALSE;
3301427a 333
334 const char* type = 0;
335 TString localURI;
336 if(gridURI == fgkMainCDB) {
337 type = "OCDB";
338 localURI = fgkLocalCDB;
339 } else if(gridURI == fgkMainRefStorage) {
340 type = "reference";
341 localURI = fgkLocalRefStorage;
342 } else {
343 AliError(Form("Invalid storage URI: %s", gridURI.Data()));
5e993b6f 344 return -2;
3301427a 345 }
346
347 AliCDBManager* man = AliCDBManager::Instance();
348
349 AliCDBStorage *gridSto = man->GetStorage(gridURI);
350 if(!gridSto) {
351 Log("SHUTTLE",
352 Form("StoreOCDB - cannot activate main %s storage", type));
5e993b6f 353 return -2;
3301427a 354 }
355
356 gridIds = gridSto->GetQueryCDBList();
357
358 // get objects previously stored in local CDB
359 AliCDBStorage *localSto = man->GetStorage(localURI);
360 if(!localSto) {
361 Log("SHUTTLE",
362 Form("StoreOCDB - cannot activate local %s storage", type));
5e993b6f 363 return -2;
3301427a 364 }
365 AliCDBPath aPath(GetOfflineDetName(fCurrentDetector.Data()),"*","*");
366 // Local objects were stored with current run as Grid version!
367 TList* localEntries = localSto->GetAll(aPath.GetPath(), GetCurrentRun(), GetCurrentRun());
368 localEntries->SetOwner(1);
369
370 // loop on local stored objects
371 TIter localIter(localEntries);
372 AliCDBEntry *aLocEntry = 0;
373 while((aLocEntry = dynamic_cast<AliCDBEntry*> (localIter.Next()))){
374 aLocEntry->SetOwner(1);
375 AliCDBId aLocId = aLocEntry->GetId();
376 aLocEntry->SetVersion(-1);
377 aLocEntry->SetSubVersion(-1);
378
379 // If local object is valid up to infinity we store it only if it is
380 // the first unprocessed run!
381 if (aLocId.GetLastRun() == AliCDBRunRange::Infinity() &&
382 !fFirstUnprocessed[GetDetPos(fCurrentDetector)])
383 {
384 Log("SHUTTLE", Form("StoreOCDB - %s: object %s has validity infinite but "
385 "there are previous unprocessed runs!",
386 fCurrentDetector.Data(), aLocId.GetPath().Data()));
5e993b6f 387 Log(fCurrentDetector.Data(), Form("StoreOCDB - %s: object %s has validity infinite but "
388 "there are previous unprocessed runs!",
389 fCurrentDetector.Data(), aLocId.GetPath().Data()));
390 delayed = kTRUE;
3301427a 391 continue;
392 }
393
394 // loop on Grid valid Id's
395 Bool_t store = kTRUE;
396 TIter gridIter(gridIds);
397 AliCDBId* aGridId = 0;
398 while((aGridId = dynamic_cast<AliCDBId*> (gridIter.Next()))){
399 if(aGridId->GetPath() != aLocId.GetPath()) continue;
400 // skip all objects valid up to infinity
401 if(aGridId->GetLastRun() == AliCDBRunRange::Infinity()) continue;
402 // if we get here, it means there's already some more recent object stored on Grid!
403 store = kFALSE;
404 break;
405 }
406
407 // If we get here, the file can be stored!
408 Bool_t storeOk = gridSto->Put(aLocEntry);
409 if(!store || storeOk){
410
411 if (!store)
412 {
413 Log(fCurrentDetector.Data(),
414 Form("StoreOCDB - A more recent object already exists in %s storage: <%s>",
415 type, aGridId->ToString().Data()));
416 } else {
417 Log("SHUTTLE",
418 Form("StoreOCDB - Object <%s> successfully put into %s storage",
419 aLocId.ToString().Data(), type));
2d9019b4 420 Log(fCurrentDetector.Data(),
421 Form("StoreOCDB - Object <%s> successfully put into %s storage",
422 aLocId.ToString().Data(), type));
3301427a 423 }
84090f85 424
3301427a 425 // removing local filename...
426 TString filename;
427 localSto->IdToFilename(aLocId, filename);
c88ad5db 428 Log("SHUTTLE", Form("StoreOCDB - Removing local file %s", filename.Data()));
3301427a 429 RemoveFile(filename.Data());
430 continue;
431 } else {
432 Log("SHUTTLE",
433 Form("StoreOCDB - Grid %s storage of object <%s> failed",
434 type, aLocId.ToString().Data()));
2d9019b4 435 Log(fCurrentDetector.Data(),
436 Form("StoreOCDB - Grid %s storage of object <%s> failed",
437 type, aLocId.ToString().Data()));
3301427a 438 result = kFALSE;
b948db8d 439 }
440 }
3301427a 441 localEntries->Clear();
2bb7b766 442
5e993b6f 443 Int_t returnCode = 0;
444
445 if (result == kFALSE)
446 returnCode = -1;
447 else if (delayed != kFALSE)
448 returnCode = 1;
449
450 Log("SHUTTLE", Form("StoreOCDB - Returning with %d (result = %d, delayed = %d)", returnCode, result, delayed));
451 Log(fCurrentDetector.Data(), Form("StoreOCDB - Returning with %d (result = %d, delayed = %d)", returnCode, result, delayed));
452
453 return returnCode;
3301427a 454}
455
546242fb 456//______________________________________________________________________________________________
457Bool_t AliShuttle::CleanReferenceStorage(const char* detector)
458{
2d9019b4 459 // clears the directory used to store reference files of a given subdetector
546242fb 460
461 AliCDBManager* man = AliCDBManager::Instance();
462 AliCDBStorage* sto = man->GetStorage(fgkLocalRefStorage);
2d9019b4 463 TString localBaseFolder = sto->GetBaseFolder();
464
465 TString targetDir = GetRefFilePrefix(localBaseFolder.Data(), detector);
466
d524ade6 467 Log("SHUTTLE", Form("CleanReferenceStorage - Cleaning %s", targetDir.Data()));
2d9019b4 468
469 TString begin;
470 begin.Form("%d_", GetCurrentRun());
471
472 TSystemDirectory* baseDir = new TSystemDirectory("/", targetDir);
473 if (!baseDir)
474 return kTRUE;
475
476 TList* dirList = baseDir->GetListOfFiles();
477 delete baseDir;
478
479 if (!dirList) return kTRUE;
480
481 if (dirList->GetEntries() < 3)
482 {
483 delete dirList;
484 return kTRUE;
485 }
486
487 Int_t nDirs = 0, nDel = 0;
488 TIter dirIter(dirList);
489 TSystemFile* entry = 0;
546242fb 490
2d9019b4 491 Bool_t success = kTRUE;
546242fb 492
2d9019b4 493 while ((entry = dynamic_cast<TSystemFile*> (dirIter.Next())))
494 {
495 if (entry->IsDirectory())
496 continue;
497
498 TString fileName(entry->GetName());
499 if (!fileName.BeginsWith(begin))
500 continue;
501
502 nDirs++;
503
504 // delete file
505 Int_t result = gSystem->Unlink(fileName.Data());
506
507 if (result)
508 {
d524ade6 509 Log("SHUTTLE", Form("CleanReferenceStorage - Could not delete file %s!", fileName.Data()));
2d9019b4 510 success = kFALSE;
511 } else {
512 nDel++;
513 }
514 }
515
516 if(nDirs > 0)
517 Log("SHUTTLE", Form("CleanReferenceStorage - %d (over %d) reference files in folder %s were deleted.",
518 nDel, nDirs, targetDir.Data()));
519
520
521 delete dirList;
522 return success;
523
524
525
526
527
546242fb 528
529 Int_t result = gSystem->GetPathInfo(targetDir, 0, (Long64_t*) 0, 0, 0);
530 if (result == 0)
531 {
532 // delete directory
d524ade6 533 result = gSystem->Exec(Form("rm -rf %s", targetDir.Data()));
546242fb 534 if (result != 0)
535 {
d524ade6 536 Log("SHUTTLE", Form("CleanReferenceStorage - Could not clean directory %s", targetDir.Data()));
546242fb 537 return kFALSE;
538 }
539 }
540
541 result = gSystem->mkdir(targetDir, kTRUE);
542 if (result != 0)
543 {
c88ad5db 544 Log("SHUTTLE", Form("CleanReferenceStorage - Error creating base directory %s", targetDir.Data()));
546242fb 545 return kFALSE;
546 }
547
548 return kTRUE;
549}
550
9827400b 551//______________________________________________________________________________________________
552Bool_t AliShuttle::StoreReferenceFile(const char* detector, const char* localFile, const char* gridFileName)
553{
554 //
3c2a21c8 555 // Stores reference file directly (without opening it). This function stores the file locally.
9827400b 556 //
3c2a21c8 557 // The file is stored under the following location:
558 // <base folder of local reference storage>/<DET>/<RUN#>_<gridFileName>
559 // where <gridFileName> is the second parameter given to the function
560 //
9827400b 561
562 if (fTestMode & kErrorStorage)
563 {
564 Log(fCurrentDetector, "StoreReferenceFile - In TESTMODE - Simulating error while storing locally");
565 return kFALSE;
566 }
567
568 AliCDBManager* man = AliCDBManager::Instance();
569 AliCDBStorage* sto = man->GetStorage(fgkLocalRefStorage);
570
571 TString localBaseFolder = sto->GetBaseFolder();
572
d524ade6 573 TString target = GetRefFilePrefix(localBaseFolder.Data(), detector);
574 target.Append(Form("/%d_%s", GetCurrentRun(), gridFileName));
9827400b 575
d524ade6 576 return CopyFileLocally(localFile, target);
c88ad5db 577}
578
579//______________________________________________________________________________________________
580Bool_t AliShuttle::StoreRunMetadataFile(const char* localFile, const char* gridFileName)
581{
582 //
583 // Stores Run metadata file to the Grid, in the run folder
584 //
585 // Only GRP can call this function.
586
587 if (fTestMode & kErrorStorage)
588 {
589 Log(fCurrentDetector, "StoreRunMetaDataFile - In TESTMODE - Simulating error while storing locally");
590 return kFALSE;
591 }
592
593 AliCDBManager* man = AliCDBManager::Instance();
594 AliCDBStorage* sto = man->GetStorage(fgkLocalRefStorage);
595
596 TString localBaseFolder = sto->GetBaseFolder();
597
598 // Build Run level folder
955d2abc 599 // folder = /alice/data/year/lhcPeriod/runNb/raw
c88ad5db 600
c88ad5db 601
675f64cd 602 TString lhcPeriod = GetLHCPeriod();
c88ad5db 603 if (lhcPeriod.Length() == 0)
604 {
605 Log("SHUTTLE","StoreRunMetaDataFile - LHCPeriod not found in logbook!");
606 return 0;
607 }
a1c5723b 608
609 // TODO partitions with one detector only write data into LHCperiod_DET
2694f253 610 TString partition = GetRunParameter("detector");
a1c5723b 611
612 if (partition.Length() > 0 && partition != "ALICE")
613 {
614 lhcPeriod.Append(Form("_%s", partition.Data()));
615 Log(fCurrentDetector, Form("Run data tags merged file will be written in %s",
616 lhcPeriod.Data()));
617 }
8884d71e 618
955d2abc 619 TString target = Form("%s/GRP/RunMetadata/alice/data/%d/%s/%09d/raw/%s",
675f64cd 620 localBaseFolder.Data(), GetCurrentYear(),
d524ade6 621 lhcPeriod.Data(), GetCurrentRun(), gridFileName);
c88ad5db 622
d524ade6 623 return CopyFileLocally(localFile, target);
c88ad5db 624}
625
626//______________________________________________________________________________________________
d524ade6 627Bool_t AliShuttle::CopyFileLocally(const char* localFile, const TString& target)
c88ad5db 628{
629 //
630 // Stores file locally. Called by StoreReferenceFile and StoreRunMetadataFile
d524ade6 631 // Files are temporarily stored in the local reference storage. When the preprocessor
632 // finishes, the Shuttle calls CopyFilesToGrid to transfer the files to AliEn
633 // (in reference or run level folders)
c88ad5db 634 //
635
d524ade6 636 TString targetDir(target(0, target.Last('/')));
637
638 //try to open base dir folder, if it does not exist
2d9019b4 639 void* dir = gSystem->OpenDirectory(targetDir.Data());
640 if (dir == NULL) {
641 if (gSystem->mkdir(targetDir.Data(), kTRUE)) {
a1c5723b 642 Log("SHUTTLE", Form("CopyFileLocally - Can't open directory <%s>", targetDir.Data()));
2d9019b4 643 return kFALSE;
644 }
645
646 } else {
647 gSystem->FreeDirectory(dir);
648 }
9827400b 649
7d43a416 650 Int_t result = 0;
651
652 result = gSystem->GetPathInfo(localFile, 0, (Long64_t*) 0, 0, 0);
9827400b 653 if (result)
654 {
a1c5723b 655 Log("SHUTTLE", Form("CopyFileLocally - %s does not exist", localFile));
546242fb 656 return kFALSE;
9827400b 657 }
546242fb 658
7d43a416 659 result = gSystem->GetPathInfo(target, 0, (Long64_t*) 0, 0, 0);
660 if (!result)
661 {
a1c5723b 662 Log("SHUTTLE", Form("CopyFileLocally - target file %s already exist, removing...", target.Data()));
7d43a416 663 if (gSystem->Unlink(target.Data()))
664 {
a1c5723b 665 Log("SHUTTLE", Form("CopyFileLocally - Could not remove existing target file %s!", target.Data()));
7d43a416 666 return kFALSE;
667 }
668 }
669
9827400b 670 result = gSystem->CopyFile(localFile, target);
671
672 if (result == 0)
673 {
a1c5723b 674 Log("SHUTTLE", Form("CopyFileLocally - File %s stored locally to %s", localFile, target.Data()));
9827400b 675 return kTRUE;
676 }
677 else
678 {
a1c5723b 679 Log("SHUTTLE", Form("CopyFileLocally - Could not store file %s to %s! Error code = %d",
546242fb 680 localFile, target.Data(), result));
9827400b 681 return kFALSE;
682 }
c88ad5db 683
684
685
9827400b 686}
687
688//______________________________________________________________________________________________
c88ad5db 689Bool_t AliShuttle::CopyFilesToGrid(const char* type)
9827400b 690{
691 //
c88ad5db 692 // Transfers local files to the Grid. Local files can be reference files
693 // or run metadata file (from GRP only).
9827400b 694 //
c88ad5db 695 // According to the type (ref, metadata) the files are stored under the following location:
696 // ref --> <base folder of reference storage>/<DET>/<RUN#>_<gridFileName>
697 // metadata --> <run data folder>/<MetadataFileName>
86aa42c3 698 //
c88ad5db 699
9827400b 700 AliCDBManager* man = AliCDBManager::Instance();
701 AliCDBStorage* sto = man->GetStorage(fgkLocalRefStorage);
702 if (!sto)
703 return kFALSE;
704 TString localBaseFolder = sto->GetBaseFolder();
9827400b 705
c88ad5db 706 TString dir;
707 TString alienDir;
9827400b 708 TString begin;
9827400b 709
c88ad5db 710 if (strcmp(type, "reference") == 0)
711 {
712 dir = GetRefFilePrefix(localBaseFolder.Data(), fCurrentDetector.Data());
713 AliCDBStorage* gridSto = man->GetStorage(fgkMainRefStorage);
714 if (!gridSto)
715 return kFALSE;
716 TString gridBaseFolder = gridSto->GetBaseFolder();
717 alienDir = GetRefFilePrefix(gridBaseFolder.Data(), fCurrentDetector.Data());
718 begin = Form("%d_", GetCurrentRun());
719 }
720 else if (strcmp(type, "metadata") == 0)
721 {
c88ad5db 722
675f64cd 723 TString lhcPeriod = GetLHCPeriod();
c88ad5db 724
725 if (lhcPeriod.Length() == 0)
726 {
727 Log("SHUTTLE","CopyFilesToGrid - LHCPeriod not found in logbook!");
728 return 0;
729 }
730
a1c5723b 731 // TODO partitions with one detector only write data into LHCperiod_DET
2694f253 732 TString partition = GetRunParameter("detector");
a1c5723b 733
734 if (partition.Length() > 0 && partition != "ALICE")
735 {
736 lhcPeriod.Append(Form("_%s", partition.Data()));
737 }
738
955d2abc 739 dir = Form("%s/GRP/RunMetadata/alice/data/%d/%s/%09d/raw",
675f64cd 740 localBaseFolder.Data(), GetCurrentYear(),
c88ad5db 741 lhcPeriod.Data(), GetCurrentRun());
675f64cd 742 alienDir = dir(dir.Index("/alice/data/"), dir.Length());
743
c88ad5db 744 begin = "";
745 }
746 else
747 {
748 Log("SHUTTLE", "CopyFilesToGrid - Unexpected: type label must be reference or metadata!");
749 return kFALSE;
750 }
751
9827400b 752 TSystemDirectory* baseDir = new TSystemDirectory("/", dir);
3d8bc902 753 if (!baseDir)
754 return kTRUE;
755
2d9019b4 756 TList* dirList = baseDir->GetListOfFiles();
757 delete baseDir;
758
759 if (!dirList) return kTRUE;
760
761 if (dirList->GetEntries() < 3)
3d8bc902 762 {
2d9019b4 763 delete dirList;
9827400b 764 return kTRUE;
3d8bc902 765 }
2d9019b4 766
546242fb 767 if (!gGrid)
768 {
c88ad5db 769 Log("SHUTTLE", "CopyFilesToGrid - Connection to Grid failed: Cannot continue!");
2d9019b4 770 delete dirList;
546242fb 771 return kFALSE;
772 }
773
2d9019b4 774 Int_t nDirs = 0, nTransfer = 0;
775 TIter dirIter(dirList);
776 TSystemFile* entry = 0;
777
9827400b 778 Bool_t success = kTRUE;
3d8bc902 779 Bool_t first = kTRUE;
9827400b 780
2d9019b4 781 while ((entry = dynamic_cast<TSystemFile*> (dirIter.Next())))
782 {
9827400b 783 if (entry->IsDirectory())
784 continue;
785
786 TString fileName(entry->GetName());
787 if (!fileName.BeginsWith(begin))
788 continue;
789
2d9019b4 790 nDirs++;
791
3d8bc902 792 if (first)
793 {
794 first = kFALSE;
c88ad5db 795 // check that folder exists, otherwise create it
3d8bc902 796 TGridResult* result = gGrid->Ls(alienDir.Data(), "a");
797
798 if (!result)
2d9019b4 799 {
800 delete dirList;
3d8bc902 801 return kFALSE;
2d9019b4 802 }
3d8bc902 803
546242fb 804 if (!result->GetFileName(1)) // TODO: It looks like element 0 is always 0!!
3d8bc902 805 {
675f64cd 806 // TODO It does not work currently! Bug in TAliEn::Mkdir
807 // TODO Manually fixed in local root v5-16-00
c88ad5db 808 if (!gGrid->Mkdir(alienDir.Data(),"-p",0))
3d8bc902 809 {
c88ad5db 810 Log("SHUTTLE", Form("CopyFilesToGrid - Cannot create directory %s",
3d8bc902 811 alienDir.Data()));
2d9019b4 812 delete dirList;
3d8bc902 813 return kFALSE;
546242fb 814 } else {
c88ad5db 815 Log("SHUTTLE",Form("CopyFilesToGrid - Folder %s created", alienDir.Data()));
3d8bc902 816 }
817
546242fb 818 } else {
c88ad5db 819 Log("SHUTTLE",Form("CopyFilesToGrid - Folder %s found", alienDir.Data()));
3d8bc902 820 }
821 }
822
9827400b 823 TString fullLocalPath;
824 fullLocalPath.Form("%s/%s", dir.Data(), fileName.Data());
825
826 TString fullGridPath;
827 fullGridPath.Form("alien://%s/%s", alienDir.Data(), fileName.Data());
828
a986b218 829 Bool_t result = TFile::Cp(fullLocalPath, fullGridPath);
9827400b 830
831 if (result)
832 {
c88ad5db 833 Log("SHUTTLE", Form("CopyFilesToGrid - Copying local file %s to %s succeeded!",
834 fullLocalPath.Data(), fullGridPath.Data()));
9827400b 835 RemoveFile(fullLocalPath);
2d9019b4 836 nTransfer++;
9827400b 837 }
838 else
839 {
c88ad5db 840 Log("SHUTTLE", Form("CopyFilesToGrid - Copying local file %s to %s FAILED!",
841 fullLocalPath.Data(), fullGridPath.Data()));
9827400b 842 success = kFALSE;
843 }
844 }
2d9019b4 845
c88ad5db 846 Log("SHUTTLE", Form("CopyFilesToGrid - %d (over %d) files in folder %s copied to Grid.",
847 nTransfer, nDirs, dir.Data()));
2d9019b4 848
849
850 delete dirList;
9827400b 851 return success;
852}
853
2d9019b4 854//______________________________________________________________________________________________
855const char* AliShuttle::GetRefFilePrefix(const char* base, const char* detector)
856{
857 //
858 // Get folder name of reference files
859 //
860
861 TString offDetStr(GetOfflineDetName(detector));
862 TString dir;
863 if (offDetStr == "ITS" || offDetStr == "MUON" || offDetStr == "PHOS")
864 {
865 dir.Form("%s/%s/%s", base, offDetStr.Data(), detector);
866 } else {
867 dir.Form("%s/%s", base, offDetStr.Data());
868 }
869
870 return dir.Data();
871
872
873}
c88ad5db 874
3301427a 875//______________________________________________________________________________________________
876void AliShuttle::CleanLocalStorage(const TString& uri)
877{
9827400b 878 //
879 // Called in case the preprocessor is declared failed. Remove remaining objects from the local storages.
880 //
3301427a 881
882 const char* type = 0;
883 if(uri == fgkLocalCDB) {
884 type = "OCDB";
885 } else if(uri == fgkLocalRefStorage) {
546242fb 886 type = "Reference";
3301427a 887 } else {
888 AliError(Form("Invalid storage URI: %s", uri.Data()));
889 return;
890 }
891
892 AliCDBManager* man = AliCDBManager::Instance();
b948db8d 893
3301427a 894 // open local storage
895 AliCDBStorage *localSto = man->GetStorage(uri);
896 if(!localSto) {
897 Log("SHUTTLE",
898 Form("CleanLocalStorage - cannot activate local %s storage", type));
899 return;
900 }
901
902 TString filename(Form("%s/%s/*/Run*_v%d_s*.root",
546242fb 903 localSto->GetBaseFolder().Data(), GetOfflineDetName(fCurrentDetector.Data()), GetCurrentRun()));
3301427a 904
c88ad5db 905 AliDebug(2, Form("filename = %s", filename.Data()));
3301427a 906
c88ad5db 907 Log("SHUTTLE", Form("Removing remaining local files for run %d and detector %s ...",
3301427a 908 GetCurrentRun(), fCurrentDetector.Data()));
909
910 RemoveFile(filename.Data());
911
912}
913
914//______________________________________________________________________________________________
915void AliShuttle::RemoveFile(const char* filename)
916{
9827400b 917 //
918 // removes local file
919 //
3301427a 920
921 TString command(Form("rm -f %s", filename));
922
923 Int_t result = gSystem->Exec(command.Data());
924 if(result != 0)
925 {
926 Log("SHUTTLE", Form("RemoveFile - %s: Cannot remove file %s!",
927 fCurrentDetector.Data(), filename));
928 }
73abe331 929}
930
b948db8d 931//______________________________________________________________________________________________
5164a766 932AliShuttleStatus* AliShuttle::ReadShuttleStatus()
933{
9827400b 934 //
935 // Reads the AliShuttleStatus from the CDB
936 //
5164a766 937
2bb7b766 938 if (fStatusEntry){
939 delete fStatusEntry;
940 fStatusEntry = 0;
941 }
5164a766 942
10a5a932 943 fStatusEntry = AliCDBManager::Instance()->GetStorage(GetLocalCDB())
2bb7b766 944 ->Get(Form("/SHUTTLE/STATUS/%s", fCurrentDetector.Data()), GetCurrentRun());
5164a766 945
2bb7b766 946 if (!fStatusEntry) return 0;
947 fStatusEntry->SetOwner(1);
5164a766 948
2bb7b766 949 AliShuttleStatus* status = dynamic_cast<AliShuttleStatus*> (fStatusEntry->GetObject());
950 if (!status) {
951 AliError("Invalid object stored to CDB!");
952 return 0;
953 }
5164a766 954
2bb7b766 955 return status;
5164a766 956}
957
958//______________________________________________________________________________________________
7bfb2090 959Bool_t AliShuttle::WriteShuttleStatus(AliShuttleStatus* status)
5164a766 960{
9827400b 961 //
962 // writes the status for one subdetector
963 //
2bb7b766 964
965 if (fStatusEntry){
966 delete fStatusEntry;
967 fStatusEntry = 0;
968 }
5164a766 969
2bb7b766 970 Int_t run = GetCurrentRun();
5164a766 971
2bb7b766 972 AliCDBId id(AliCDBPath("SHUTTLE", "STATUS", fCurrentDetector), run, run);
5164a766 973
2bb7b766 974 fStatusEntry = new AliCDBEntry(status, id, new AliCDBMetaData);
975 fStatusEntry->SetOwner(1);
5164a766 976
2bb7b766 977 UInt_t result = AliCDBManager::Instance()->GetStorage(fgkLocalCDB)->Put(fStatusEntry);
7bfb2090 978
2bb7b766 979 if (!result) {
3301427a 980 Log("SHUTTLE", Form("WriteShuttleStatus - Failed for %s, run %d",
981 fCurrentDetector.Data(), run));
2bb7b766 982 return kFALSE;
983 }
e7f62f16 984
985 SendMLInfo();
7bfb2090 986
2bb7b766 987 return kTRUE;
5164a766 988}
989
990//______________________________________________________________________________________________
991void AliShuttle::UpdateShuttleStatus(AliShuttleStatus::Status newStatus, Bool_t increaseCount)
992{
9827400b 993 //
994 // changes the AliShuttleStatus for the given detector and run to the given status
995 //
5164a766 996
2bb7b766 997 if (!fStatusEntry){
998 AliError("UNEXPECTED: fStatusEntry empty");
999 return;
1000 }
5164a766 1001
2bb7b766 1002 AliShuttleStatus* status = dynamic_cast<AliShuttleStatus*> (fStatusEntry->GetObject());
5164a766 1003
2bb7b766 1004 if (!status){
c88ad5db 1005 Log("SHUTTLE", "UpdateShuttleStatus - UNEXPECTED: status could not be read from current CDB entry");
2bb7b766 1006 return;
1007 }
5164a766 1008
2c15234c 1009 TString actionStr = Form("UpdateShuttleStatus - %s: Changing state from %s to %s",
eba76848 1010 fCurrentDetector.Data(),
36c99a6a 1011 status->GetStatusName(),
eba76848 1012 status->GetStatusName(newStatus));
cb343cfd 1013 Log("SHUTTLE", actionStr);
1014 SetLastAction(actionStr);
5164a766 1015
2bb7b766 1016 status->SetStatus(newStatus);
1017 if (increaseCount) status->IncreaseCount();
5164a766 1018
2bb7b766 1019 AliCDBManager::Instance()->GetStorage(fgkLocalCDB)->Put(fStatusEntry);
e7f62f16 1020
1021 SendMLInfo();
5164a766 1022}
e7f62f16 1023
1024//______________________________________________________________________________________________
1025void AliShuttle::SendMLInfo()
1026{
1027 //
1028 // sends ML information about the current status of the current detector being processed
1029 //
1030
1031 AliShuttleStatus* status = dynamic_cast<AliShuttleStatus*> (fStatusEntry->GetObject());
1032
1033 if (!status){
3301427a 1034 Log("SHUTTLE", "SendMLInfo - UNEXPECTED: status could not be read from current CDB entry");
e7f62f16 1035 return;
1036 }
1037
1038 TMonaLisaText mlStatus(Form("%s_status", fCurrentDetector.Data()), status->GetStatusName());
1039 TMonaLisaValue mlRetryCount(Form("%s_count", fCurrentDetector.Data()), status->GetCount());
1040
1041 TList mlList;
1042 mlList.Add(&mlStatus);
1043 mlList.Add(&mlRetryCount);
1044
ee6f7523 1045 TString mlID;
1046 mlID.Form("%d", GetCurrentRun());
1047 fMonaLisa->SendParameters(&mlList, mlID);
e7f62f16 1048}
1049
5164a766 1050//______________________________________________________________________________________________
1051Bool_t AliShuttle::ContinueProcessing()
1052{
9827400b 1053 // this function reads the AliShuttleStatus information from CDB and
1054 // checks if the processing should be continued
1055 // if yes it returns kTRUE and updates the AliShuttleStatus with nextStatus
2bb7b766 1056
926cbc0e 1057 if (!fConfig->HostProcessDetector(fCurrentDetector))
1058 return kFALSE;
57c1a579 1059
1060 AliPreprocessor* aPreprocessor =
1061 dynamic_cast<AliPreprocessor*> (fPreprocessorMap.GetValue(fCurrentDetector));
1062 if (!aPreprocessor)
1063 {
c88ad5db 1064 Log("SHUTTLE", Form("ContinueProcessing - %s: no preprocessor registered", fCurrentDetector.Data()));
57c1a579 1065 return kFALSE;
1066 }
1067
2bb7b766 1068 AliShuttleLogbookEntry::Status entryStatus =
eba76848 1069 fLogbookEntry->GetDetectorStatus(fCurrentDetector);
2bb7b766 1070
5e993b6f 1071 if (entryStatus != AliShuttleLogbookEntry::kUnprocessed) {
c88ad5db 1072 Log("SHUTTLE", Form("ContinueProcessing - %s is %s",
2bb7b766 1073 fCurrentDetector.Data(),
1074 fLogbookEntry->GetDetectorStatusName(entryStatus)));
1075 return kFALSE;
1076 }
1077
1078 // if we get here, according to Shuttle logbook subdetector is in UNPROCESSED state
be48e3ea 1079
1080 // check if current run is first unprocessed run for current detector
1081 if (fConfig->StrictRunOrder(fCurrentDetector) &&
1082 !fFirstUnprocessed[GetDetPos(fCurrentDetector)])
1083 {
86aa42c3 1084 if (fTestMode == kNone)
1085 {
c88ad5db 1086 Log("SHUTTLE", Form("ContinueProcessing - %s requires strict run ordering"
1087 " but this is not the first unprocessed run!"));
86aa42c3 1088 return kFALSE;
1089 }
1090 else
1091 {
c88ad5db 1092 Log("SHUTTLE", Form("ContinueProcessing - In TESTMODE - "
1093 "Although %s requires strict run ordering "
1094 "and this is not the first unprocessed run, "
1095 "the SHUTTLE continues"));
86aa42c3 1096 }
be48e3ea 1097 }
1098
926cbc0e 1099 // Is the subdetector processed first time for this run?
1100 fFirstProcessing = kFALSE;
1101
2bb7b766 1102 AliShuttleStatus* status = ReadShuttleStatus();
1103 if (!status) {
1104 // first time
1105 Log("SHUTTLE", Form("ContinueProcessing - %s: Processing first time",
1106 fCurrentDetector.Data()));
1107 status = new AliShuttleStatus(AliShuttleStatus::kStarted);
926cbc0e 1108 fFirstProcessing = kTRUE;
2bb7b766 1109 return WriteShuttleStatus(status);
1110 }
1111
ffa25f20 1112 // The following case shouldn't happen if Shuttle Logbook was correctly updated.
2bb7b766 1113 // If it happens it may mean Logbook updating failed... let's do it now!
1114 if (status->GetStatus() == AliShuttleStatus::kDone ||
ffa25f20 1115 status->GetStatus() == AliShuttleStatus::kFailed ||
1116 status->GetStatus() == AliShuttleStatus::kSkipped) {
2bb7b766 1117 Log("SHUTTLE", Form("ContinueProcessing - %s is already %s. Updating Shuttle Logbook",
1118 fCurrentDetector.Data(),
1119 status->GetStatusName(status->GetStatus())));
ffa25f20 1120
1121 if (status->GetStatus() == AliShuttleStatus::kSkipped)
1122 {
1123 UpdateShuttleLogbook(fCurrentDetector.Data(), "DONE");
1124 }
1125 else
1126 UpdateShuttleLogbook(fCurrentDetector.Data(), status->GetStatusName(status->GetStatus()));
1127
2bb7b766 1128 return kFALSE;
1129 }
1130
5e993b6f 1131 if (status->GetStatus() == AliShuttleStatus::kStoreStarted || status->GetStatus() == AliShuttleStatus::kStoreDelayed ||status->GetStatus() == AliShuttleStatus::kStoreError) {
2bb7b766 1132 Log("SHUTTLE",
c88ad5db 1133 Form("ContinueProcessing - %s: Grid storage of one or more "
1134 "objects failed. Trying again now",
2bb7b766 1135 fCurrentDetector.Data()));
5e993b6f 1136 StoreOCDB();
2bb7b766 1137 return kFALSE;
1138 }
1139
1140 // if we get here, there is a restart
57c1a579 1141 Bool_t cont = kFALSE;
2bb7b766 1142
1143 // abort conditions
cb343cfd 1144 if (status->GetCount() >= fConfig->GetMaxRetries()) {
57c1a579 1145 Log("SHUTTLE", Form("ContinueProcessing - %s failed %d times in status %s - "
1146 "Updating Shuttle Logbook", fCurrentDetector.Data(),
2bb7b766 1147 status->GetCount(), status->GetStatusName()));
1148 UpdateShuttleLogbook(fCurrentDetector.Data(), "FAILED");
e7f62f16 1149 UpdateShuttleStatus(AliShuttleStatus::kFailed);
3301427a 1150
1151 // there may still be objects in local OCDB and reference storage
1152 // and FXS databases may be not updated: do it now!
9827400b 1153
1154 // TODO Currently disabled, we want to keep files in case of failure!
1155 // CleanLocalStorage(fgkLocalCDB);
1156 // CleanLocalStorage(fgkLocalRefStorage);
1157 // UpdateTableFailCase();
1158
1159 // Send mail to detector expert!
c88ad5db 1160 Log("SHUTTLE", Form("ContinueProcessing - Sending mail to %s expert...",
926cbc0e 1161 fCurrentDetector.Data()));
339adafa 1162 if (!SendMail(kPPEMail))
9827400b 1163 Log("SHUTTLE", Form("ContinueProcessing - Could not send mail to %s expert",
926cbc0e 1164 fCurrentDetector.Data()));
3301427a 1165
57c1a579 1166 } else {
1167 Log("SHUTTLE", Form("ContinueProcessing - %s: restarting. "
1168 "Aborted before with %s. Retry number %d.", fCurrentDetector.Data(),
1169 status->GetStatusName(), status->GetCount()));
9827400b 1170 Bool_t increaseCount = kTRUE;
c88ad5db 1171 if (status->GetStatus() == AliShuttleStatus::kDCSError ||
8da81210 1172 status->GetStatus() == AliShuttleStatus::kDCSStarted ||
1173 status->GetStatus() == AliShuttleStatus::kFXSError)
c88ad5db 1174 increaseCount = kFALSE;
675f64cd 1175
9827400b 1176 UpdateShuttleStatus(AliShuttleStatus::kStarted, increaseCount);
57c1a579 1177 cont = kTRUE;
2bb7b766 1178 }
1179
57c1a579 1180 return cont;
5164a766 1181}
1182
1183//______________________________________________________________________________________________
2bb7b766 1184Bool_t AliShuttle::Process(AliShuttleLogbookEntry* entry)
58bc3020 1185{
73abe331 1186 //
b948db8d 1187 // Makes data retrieval for all detectors in the configuration.
2bb7b766 1188 // entry: Shuttle logbook entry, contains run paramenters and status of detectors
1189 // (Unprocessed, Inactive, Failed or Done).
d477ad88 1190 // Returns kFALSE in case of error occured and kTRUE otherwise
73abe331 1191 //
1192
9827400b 1193 if (!entry) return kFALSE;
2bb7b766 1194
1195 fLogbookEntry = entry;
1196
c88ad5db 1197 Log("SHUTTLE", Form("\t\t\t^*^*^*^*^*^*^*^*^*^*^*^* run %d: START ^*^*^*^*^*^*^*^*^*^*^*^*",
9827400b 1198 GetCurrentRun()));
2bb7b766 1199
e7f62f16 1200 // Send the information to ML
339adafa 1201 CountOpenRuns();
1202
e7f62f16 1203 TMonaLisaText mlStatus("SHUTTLE_status", "Processing");
339adafa 1204 TString runType(entry->GetRunType());
1205 if (strlen(entry->GetRunParameter("log")) > 0){
1206
1207 runType += "(";
1208 runType += entry->GetRunParameter("log");
1209 runType += ")";
1210 }
1211 TMonaLisaText mlRunType("SHUTTLE_runtype", runType);
e7f62f16 1212
1213 TList mlList;
1214 mlList.Add(&mlStatus);
9827400b 1215 mlList.Add(&mlRunType);
e7f62f16 1216
ee6f7523 1217 TString mlID;
1218 mlID.Form("%d", GetCurrentRun());
1219 fMonaLisa->SendParameters(&mlList, mlID);
3301427a 1220
9827400b 1221 if (fLogbookEntry->IsDone())
1222 {
1223 Log("SHUTTLE","Process - Shuttle is already DONE. Updating logbook");
1224 UpdateShuttleLogbook("shuttle_done");
1225 fLogbookEntry = 0;
1226 return kTRUE;
1227 }
1228
1229 // read test mode if flag is set
1230 if (fReadTestMode)
1231 {
3d8bc902 1232 fTestMode = kNone;
9827400b 1233 TString logEntry(entry->GetRunParameter("log"));
1234 //printf("log entry = %s\n", logEntry.Data());
1235 TString searchStr("Testmode: ");
1236 Int_t pos = logEntry.Index(searchStr.Data());
1237 //printf("%d\n", pos);
1238 if (pos >= 0)
1239 {
1240 TSubString subStr = logEntry(pos + searchStr.Length(), logEntry.Length());
1241 //printf("%s\n", subStr.String().Data());
1242 TString newStr(subStr.Data());
1243 TObjArray* token = newStr.Tokenize(' ');
1244 if (token)
1245 {
1246 //token->Print();
1247 TObjString* tmpStr = dynamic_cast<TObjString*> (token->First());
1248 if (tmpStr)
1249 {
1250 Int_t testMode = tmpStr->String().Atoi();
1251 if (testMode > 0)
1252 {
c88ad5db 1253 Log("SHUTTLE", Form("Process - Enabling test mode %d", testMode));
9827400b 1254 SetTestMode((TestMode) testMode);
1255 }
1256 }
1257 delete token;
1258 }
1259 }
1260 }
c88ad5db 1261
eba76848 1262 fLogbookEntry->Print("all");
57f50b3c 1263
1264 // Initialization
d477ad88 1265 Bool_t hasError = kFALSE;
5164a766 1266
675f64cd 1267 // Set the CDB and Reference folders according to the year and LHC period
1268 TString lhcPeriod(GetLHCPeriod());
1269 if (lhcPeriod.Length() == 0)
1270 {
7d4cf768 1271 Log("SHUTTLE","Process - LHCPeriod not found in logbook!");
1272 return 0;
675f64cd 1273 }
1274
1275 if (fgkMainCDB.Length() == 0)
1276 fgkMainCDB = Form("alien://folder=/alice/data/%d/%s/OCDB?user=alidaq?cacheFold=/tmp/OCDBCache",
1277 GetCurrentYear(), lhcPeriod.Data());
1278
1279 if (fgkMainRefStorage.Length() == 0)
1280 fgkMainRefStorage = Form("alien://folder=/alice/data/%d/%s/Reference?user=alidaq?cacheFold=/tmp/OCDBCache",
1281 GetCurrentYear(), lhcPeriod.Data());
1282
57f50b3c 1283 // Loop on detectors in the configuration
b948db8d 1284 TIter iter(fConfig->GetDetectors());
2bb7b766 1285 TObjString* aDetector = 0;
b948db8d 1286
4508ab8b 1287 Bool_t first = kTRUE;
1288
be48e3ea 1289 while ((aDetector = (TObjString*) iter.Next()))
1290 {
7bfb2090 1291 fCurrentDetector = aDetector->String();
5164a766 1292
5e993b6f 1293 if (ContinueProcessing() == kFALSE)
1294 continue;
4508ab8b 1295
1296 if (first)
1297 {
1298 // only read QueryCDB when needed and only once
1299 AliCDBStorage *mainCDBSto = AliCDBManager::Instance()->GetStorage(fgkMainCDB);
1300 if(mainCDBSto) mainCDBSto->QueryCDB(GetCurrentRun());
1301 AliCDBStorage *mainRefSto = AliCDBManager::Instance()->GetStorage(fgkMainRefStorage);
1302 if(mainRefSto) mainRefSto->QueryCDB(GetCurrentRun());
1303 first = kFALSE;
1304 }
9e080f92 1305
c88ad5db 1306 Log("SHUTTLE", Form("\t\t\t****** run %d - %s: START ******",
2bb7b766 1307 GetCurrentRun(), aDetector->GetName()));
1308
9d733021 1309 for(Int_t iSys=0;iSys<3;iSys++) fFXSCalled[iSys]=kFALSE;
1310
c88ad5db 1311 Log(fCurrentDetector.Data(), "Process - Starting processing");
85a80aa9 1312
be48e3ea 1313 Int_t pid = fork();
1314
1315 if (pid < 0)
1316 {
c88ad5db 1317 Log("SHUTTLE", "Process - ERROR: Forking failed");
be48e3ea 1318 }
1319 else if (pid > 0)
1320 {
1321 // parent
c88ad5db 1322 Log("SHUTTLE", Form("Process - In parent process of %d - %s: Starting monitoring",
be48e3ea 1323 GetCurrentRun(), aDetector->GetName()));
1324
1325 Long_t begin = time(0);
1326
1327 int status; // to be used with waitpid, on purpose an int (not Int_t)!
1328 while (waitpid(pid, &status, WNOHANG) == 0)
1329 {
1330 Long_t expiredTime = time(0) - begin;
1331
1332 if (expiredTime > fConfig->GetPPTimeOut())
1333 {
9827400b 1334 TString tmp;
c88ad5db 1335 tmp.Form("Process - Process of %s time out. "
1336 "Run time: %d seconds. Killing...",
1337 fCurrentDetector.Data(), expiredTime);
9827400b 1338 Log("SHUTTLE", tmp);
1339 Log(fCurrentDetector, tmp);
be48e3ea 1340
1341 kill(pid, 9);
1342
3301427a 1343 UpdateShuttleStatus(AliShuttleStatus::kPPTimeOut);
be48e3ea 1344 hasError = kTRUE;
1345
1346 gSystem->Sleep(1000);
1347 }
1348 else
1349 {
be48e3ea 1350 gSystem->Sleep(1000);
9827400b 1351
1352 TString checkStr;
1353 checkStr.Form("ps -o vsize --pid %d | tail -n 1", pid);
1354 FILE* pipe = gSystem->OpenPipe(checkStr, "r");
1355 if (!pipe)
1356 {
c88ad5db 1357 Log("SHUTTLE", Form("Process - Error: "
1358 "Could not open pipe to %s", checkStr.Data()));
9827400b 1359 continue;
1360 }
1361
1362 char buffer[100];
1363 if (!fgets(buffer, 100, pipe))
1364 {
c88ad5db 1365 Log("SHUTTLE", "Process - Error: ps did not return anything");
9827400b 1366 gSystem->ClosePipe(pipe);
1367 continue;
1368 }
1369 gSystem->ClosePipe(pipe);
1370
1371 //Log("SHUTTLE", Form("ps returned %s", buffer));
1372
1373 Int_t mem = 0;
1374 if ((sscanf(buffer, "%d\n", &mem) != 1) || !mem)
1375 {
c88ad5db 1376 Log("SHUTTLE", "Process - Error: Could not parse output of ps");
9827400b 1377 continue;
1378 }
1379
1380 if (expiredTime % 60 == 0)
ee6f7523 1381 {
c88ad5db 1382 Log("SHUTTLE", Form("Process - %s: Checking process. "
1383 "Run time: %d seconds - Memory consumption: %d KB",
1384 fCurrentDetector.Data(), expiredTime, mem));
ee6f7523 1385 SendAlive();
1386 }
9827400b 1387
1388 if (mem > fConfig->GetPPMaxMem())
1389 {
1390 TString tmp;
c88ad5db 1391 tmp.Form("Process - Process exceeds maximum allowed memory "
1392 "(%d KB > %d KB). Killing...",
9827400b 1393 mem, fConfig->GetPPMaxMem());
1394 Log("SHUTTLE", tmp);
1395 Log(fCurrentDetector, tmp);
1396
1397 kill(pid, 9);
1398
1399 UpdateShuttleStatus(AliShuttleStatus::kPPOutOfMemory);
1400 hasError = kTRUE;
1401
1402 gSystem->Sleep(1000);
1403 }
be48e3ea 1404 }
1405 }
1406
c88ad5db 1407 Log("SHUTTLE", Form("Process - In parent process of %d - %s: Client has terminated.",
be48e3ea 1408 GetCurrentRun(), aDetector->GetName()));
1409
1410 if (WIFEXITED(status))
1411 {
1412 Int_t returnCode = WEXITSTATUS(status);
1413
c88ad5db 1414 Log("SHUTTLE", Form("Process - %s: the return code is %d", fCurrentDetector.Data(),
3301427a 1415 returnCode));
be48e3ea 1416
9827400b 1417 if (returnCode == 0) hasError = kTRUE;
be48e3ea 1418 }
1419 }
1420 else if (pid == 0)
1421 {
1422 // client
c88ad5db 1423 Log("SHUTTLE", Form("Process - In client process of %d - %s", GetCurrentRun(),
1424 aDetector->GetName()));
be48e3ea 1425
c88ad5db 1426 Log("SHUTTLE", Form("Process - Redirecting output to %s log",fCurrentDetector.Data()));
ffa29e93 1427
546242fb 1428 if ((freopen(GetLogFileName(fCurrentDetector), "a", stdout)) == 0)
ffa29e93 1429 {
c88ad5db 1430 Log("SHUTTLE", "Process - Could not freopen stdout");
ffa29e93 1431 }
1432 else
1433 {
1434 fOutputRedirected = kTRUE;
1435 if ((dup2(fileno(stdout), fileno(stderr))) < 0)
c88ad5db 1436 Log("SHUTTLE", "Process - Could not redirect stderr");
ffa29e93 1437
1438 }
1439
5bac2bde 1440 TString wd = gSystem->WorkingDirectory();
675f64cd 1441 TString tmpDir = Form("%s/%s_%d_process", GetShuttleTempDir(),
1442 fCurrentDetector.Data(), GetCurrentRun());
5bac2bde 1443
d524ade6 1444 Int_t result = gSystem->GetPathInfo(tmpDir.Data(), 0, (Long64_t*) 0, 0, 0);
1445 if (!result) // temp dir already exists!
1446 {
1447 Log(fCurrentDetector.Data(),
1448 Form("Process - %s dir already exists! Removing...", tmpDir.Data()));
1449 gSystem->Exec(Form("rm -rf %s",tmpDir.Data()));
675f64cd 1450 }
1451
1452 if (gSystem->mkdir(tmpDir.Data(), 1))
1453 {
1454 Log(fCurrentDetector.Data(), "Process - could not make temp directory!!");
1455 gSystem->Exit(1);
d524ade6 1456 }
1457
1458 if (!gSystem->ChangeDirectory(tmpDir.Data()))
1459 {
1460 Log(fCurrentDetector.Data(), "Process - could not change directory!!");
1461 gSystem->Exit(1);
1462 }
5bac2bde 1463
ffa25f20 1464 Int_t success = ProcessCurrentDetector();
5bac2bde 1465
1466 gSystem->ChangeDirectory(wd.Data());
b1d18693 1467
ffa25f20 1468 if (success == 1) // Preprocessor finished successfully!
9827400b 1469 {
fb40ead4 1470 // remove temporary folder or DCS map
1471 if (!fConfig->KeepTempFolder())
1472 {
1473 gSystem->Exec(Form("rm -rf %s",tmpDir.Data()));
1474 } else if (!fConfig->KeepDCSMap())
1475 {
1476 gSystem->Exec(Form("rm -f %s/DCSMap.root",tmpDir.Data()));
1477 }
b1d18693 1478
3301427a 1479 // Update time_processed field in FXS DB
1480 if (UpdateTable() == kFALSE)
5bac2bde 1481 Log("SHUTTLE", Form("Process - %s: Could not update FXS databases!",
1482 fCurrentDetector.Data()));
3301427a 1483
1484 // Transfer the data from local storage to main storage (Grid)
3301427a 1485 if (StoreOCDB() == kFALSE)
9827400b 1486 success = kFALSE;
5e993b6f 1487 }
ffa25f20 1488 else if (success == 0)
c88ad5db 1489 {
1490 Log("SHUTTLE",
1491 Form("\t\t\t****** run %d - %s: PP ERROR ******",
1492 GetCurrentRun(), aDetector->GetName()));
be48e3ea 1493 }
1494
4b95672b 1495 for (UInt_t iSys=0; iSys<3; iSys++)
1496 {
1497 if (fFXSCalled[iSys]) fFXSlist[iSys].Clear();
1498 }
1499
c88ad5db 1500 Log("SHUTTLE", Form("Process - Client process of %d - %s is exiting now with %d.",
9827400b 1501 GetCurrentRun(), aDetector->GetName(), success));
be48e3ea 1502
1503 // the client exits here
9827400b 1504 gSystem->Exit(success);
be48e3ea 1505
1506 AliError("We should never get here!!!");
1507 }
7bfb2090 1508 }
5164a766 1509
c88ad5db 1510 Log("SHUTTLE", Form("\t\t\t^*^*^*^*^*^*^*^*^*^*^*^* run %d: FINISH ^*^*^*^*^*^*^*^*^*^*^*^*",
2bb7b766 1511 GetCurrentRun()));
1512
1513 //check if shuttle is done for this run, if so update logbook
1514 TObjArray checkEntryArray;
1515 checkEntryArray.SetOwner(1);
9e080f92 1516 TString whereClause = Form("where run=%d", GetCurrentRun());
b0e53b15 1517 if (!QueryShuttleLogbook(whereClause.Data(), checkEntryArray) ||
1518 checkEntryArray.GetEntries() == 0) {
9e080f92 1519 Log("SHUTTLE", Form("Process - Warning: Cannot check status of run %d on Shuttle logbook!",
1520 GetCurrentRun()));
1521 return hasError == kFALSE;
1522 }
b948db8d 1523
9e080f92 1524 AliShuttleLogbookEntry* checkEntry = dynamic_cast<AliShuttleLogbookEntry*>
1525 (checkEntryArray.At(0));
2bb7b766 1526
9e080f92 1527 if (checkEntry)
1528 {
1529 if (checkEntry->IsDone())
be48e3ea 1530 {
9e080f92 1531 Log("SHUTTLE","Process - Shuttle is DONE. Updating logbook");
1532 UpdateShuttleLogbook("shuttle_done");
1533 }
1534 else
1535 {
1536 for (UInt_t iDet=0; iDet<NDetectors(); iDet++)
be48e3ea 1537 {
9e080f92 1538 if (checkEntry->GetDetectorStatus(iDet) == AliShuttleLogbookEntry::kUnprocessed)
be48e3ea 1539 {
9e080f92 1540 AliDebug(2, Form("Run %d: setting %s as \"not first time unprocessed\"",
1541 checkEntry->GetRun(), GetDetName(iDet)));
1542 fFirstUnprocessed[iDet] = kFALSE;
be48e3ea 1543 }
1544 }
339adafa 1545 TMonaLisaText mlStatusPending("SHUTTLE_status", "Pending");
1546 mlList.Clear();
1547 mlList.Add(&mlStatusPending);
1548 fMonaLisa->SendParameters(&mlList, mlID);
2bb7b766 1549 }
1550 }
1551
1552 fLogbookEntry = 0;
85a80aa9 1553
a7160fe9 1554 return hasError == kFALSE;
73abe331 1555}
1556
b948db8d 1557//______________________________________________________________________________________________
ffa25f20 1558Int_t AliShuttle::ProcessCurrentDetector()
73abe331 1559{
1560 //
2bb7b766 1561 // Makes data retrieval just for a specific detector (fCurrentDetector).
73abe331 1562 // Threre should be a configuration for this detector.
73abe331 1563
1d172743 1564 Log("SHUTTLE", Form("ProcessCurrentDetector - Retrieving values for %s, run %d",
1565 fCurrentDetector.Data(), GetCurrentRun()));
73abe331 1566
d524ade6 1567 TString wd = gSystem->WorkingDirectory();
1568
2d9019b4 1569 if (!CleanReferenceStorage(fCurrentDetector.Data()))
ffa25f20 1570 return 0;
d524ade6 1571
1572 gSystem->ChangeDirectory(wd.Data());
1573
3301427a 1574 // call preprocessor
1575 AliPreprocessor* aPreprocessor =
1576 dynamic_cast<AliPreprocessor*> (fPreprocessorMap.GetValue(fCurrentDetector));
1577
8579c6fb 1578 // check if the preprocessor wants to process this run type
1579 if (aPreprocessor->ProcessRunType() == kFALSE)
ffa25f20 1580 {
1581 UpdateShuttleStatus(AliShuttleStatus::kSkipped);
1582 UpdateShuttleLogbook(fCurrentDetector, "DONE");
1583 Log(fCurrentDetector, Form("ProcessCurrentDetector - %s preprocessor is not interested in this run type", fCurrentDetector.Data()));
1584
1585 return 2;
1586 }
1587
1588 TMap* dcsMap = new TMap();
1589
3301427a 1590 aPreprocessor->Initialize(GetCurrentRun(), GetCurrentStartTime(), GetCurrentEndTime());
1591
1592 Bool_t processDCS = aPreprocessor->ProcessDCS();
d477ad88 1593
651fdaab 1594 if (!processDCS)
1595 {
1d172743 1596 Log(fCurrentDetector, "ProcessCurrentDetector -"
1597 " The preprocessor requested to skip the retrieval of DCS values");
651fdaab 1598 }
8b739301 1599 else if (fTestMode & kSkipDCS)
2c15234c 1600 {
1d172743 1601 Log(fCurrentDetector, "ProcessCurrentDetector - In TESTMODE: Skipping DCS processing");
9827400b 1602 }
1603 else if (fTestMode & kErrorDCS)
1604 {
1d172743 1605 Log(fCurrentDetector, "ProcessCurrentDetector - In TESTMODE: Simulating DCS error");
3d8bc902 1606 UpdateShuttleStatus(AliShuttleStatus::kDCSStarted);
9827400b 1607 UpdateShuttleStatus(AliShuttleStatus::kDCSError);
1d172743 1608 delete dcsMap;
ffa25f20 1609 return 0;
2c15234c 1610 } else {
3301427a 1611
1612 UpdateShuttleStatus(AliShuttleStatus::kDCSStarted);
1613
1d172743 1614 // Query DCS archive
1615 Int_t nServers = fConfig->GetNServers(fCurrentDetector);
a038aa70 1616
1d172743 1617 for (int iServ=0; iServ<nServers; iServ++)
2c15234c 1618 {
1d172743 1619
1620 TString host(fConfig->GetDCSHost(fCurrentDetector, iServ));
1621 Int_t port = fConfig->GetDCSPort(fCurrentDetector, iServ);
542b6cc8 1622 Int_t multiSplit = fConfig->GetMultiSplit(fCurrentDetector, iServ);
1623
1790d4b7 1624 Log(fCurrentDetector, Form("ProcessCurrentDetector -"
1625 " Querying DCS Amanda server %s:%d (%d of %d)",
1626 host.Data(), port, iServ+1, nServers));
1d172743 1627
1628 TMap* aliasMap = 0;
1629 TMap* dpMap = 0;
1630
1631 if (fConfig->GetDCSAliases(fCurrentDetector, iServ)->GetEntries() > 0)
2c15234c 1632 {
1d172743 1633 aliasMap = GetValueSet(host, port,
542b6cc8 1634 fConfig->GetDCSAliases(fCurrentDetector, iServ),
1635 kAlias, multiSplit);
1d172743 1636 if (!aliasMap)
1637 {
1638 Log(fCurrentDetector,
1639 Form("ProcessCurrentDetector -"
675f64cd 1640 " Error retrieving DCS aliases from server %s."
1641 " Sending mail to DCS experts!", host.Data()));
1d172743 1642 UpdateShuttleStatus(AliShuttleStatus::kDCSError);
675f64cd 1643
339adafa 1644 if (!SendMail(kDCSEMail))
fb40ead4 1645 Log("SHUTTLE", Form("ProcessCurrentDetector - "
95d4aaef 1646 "Could not send mail to DCS experts!"));
675f64cd 1647
a038aa70 1648 delete dcsMap;
ffa25f20 1649 return 0;
1d172743 1650 }
2c15234c 1651 }
a038aa70 1652
1d172743 1653 if (fConfig->GetDCSDataPoints(fCurrentDetector, iServ)->GetEntries() > 0)
a038aa70 1654 {
1d172743 1655 dpMap = GetValueSet(host, port,
542b6cc8 1656 fConfig->GetDCSDataPoints(fCurrentDetector, iServ),
1657 kDP, multiSplit);
1d172743 1658 if (!dpMap)
1659 {
1660 Log(fCurrentDetector,
1661 Form("ProcessCurrentDetector -"
675f64cd 1662 " Error retrieving DCS data points from server %s."
1663 " Sending mail to DCS experts!", host.Data()));
1d172743 1664 UpdateShuttleStatus(AliShuttleStatus::kDCSError);
675f64cd 1665
339adafa 1666 if (!SendMail(kDCSEMail))
fb40ead4 1667 Log("SHUTTLE", Form("ProcessCurrentDetector - "
95d4aaef 1668 "Could not send mail to DCS experts!"));
675f64cd 1669
1d172743 1670 if (aliasMap) delete aliasMap;
1671 delete dcsMap;
ffa25f20 1672 return 0;
1d172743 1673 }
a038aa70 1674 }
1d172743 1675
1676 // merge aliasMap and dpMap into dcsMap
1677 if(aliasMap) {
1678 TIter iter(aliasMap);
a038aa70 1679 TObjString* key = 0;
1680 while ((key = (TObjString*) iter.Next()))
1d172743 1681 dcsMap->Add(key, aliasMap->GetValue(key->String()));
1682
1683 aliasMap->SetOwner(kFALSE);
1684 delete aliasMap;
1685 }
1686
1687 if(dpMap) {
1688 TIter iter(dpMap);
1689 TObjString* key = 0;
1690 while ((key = (TObjString*) iter.Next()))
1691 dcsMap->Add(key, dpMap->GetValue(key->String()));
1692
1693 dpMap->SetOwner(kFALSE);
1694 delete dpMap;
a038aa70 1695 }
73abe331 1696 }
1697 }
dc25836b 1698
b1d18693 1699 // save map into file, to help debugging in case of preprocessor error
fbc112e3 1700 TFile* f = TFile::Open("DCSMap.root","recreate");
b1d18693 1701 f->cd();
1702 dcsMap->Write("DCSMap", TObject::kSingleKey);
1703 f->Close();
fbc112e3 1704 delete f;
b1d18693 1705
2bb7b766 1706 // DCS Archive DB processing successful. Call Preprocessor!
85a80aa9 1707 UpdateShuttleStatus(AliShuttleStatus::kPPStarted);
a7160fe9 1708
339adafa 1709 fFXSError = -1; // this variable is kTRUE after ::Process if an FXS error occured
8da81210 1710
a038aa70 1711 UInt_t returnValue = aPreprocessor->Process(dcsMap);
8da81210 1712
339adafa 1713 if (fFXSError!=-1) {
8da81210 1714 UpdateShuttleStatus(AliShuttleStatus::kFXSError);
339adafa 1715 SendMail(kFXSEMail, fFXSError);
8da81210 1716 dcsMap->DeleteAll();
1717 delete dcsMap;
ffa25f20 1718 return 0;
8da81210 1719 }
b948db8d 1720
3301427a 1721 if (returnValue > 0) // Preprocessor error!
1722 {
c88ad5db 1723 Log(fCurrentDetector, Form("ProcessCurrentDetector - "
1724 "Preprocessor failed. Process returned %d.", returnValue));
cb343cfd 1725 UpdateShuttleStatus(AliShuttleStatus::kPPError);
a038aa70 1726 dcsMap->DeleteAll();
1727 delete dcsMap;
ffa25f20 1728 return 0;
9827400b 1729 }
1730
1731 // preprocessor ok!
1732 UpdateShuttleStatus(AliShuttleStatus::kPPDone);
1733 Log(fCurrentDetector, Form("ProcessCurrentDetector - %s preprocessor returned success",
1734 fCurrentDetector.Data()));
b948db8d 1735
a038aa70 1736 dcsMap->DeleteAll();
1737 delete dcsMap;
b948db8d 1738
ffa25f20 1739 return 1;
2bb7b766 1740}
1741
fbc112e3 1742//______________________________________________________________________________________________
1743void AliShuttle::CountOpenRuns()
1744{
1745 // Query DAQ's Shuttle logbook and sends the number of open runs to ML
1746
1747 // check connection, in case connect
1748 if (!Connect(3))
1749 return;
1750
1751 TString sqlQuery;
1752 sqlQuery = Form("select count(*) from %s where shuttle_done=0", fConfig->GetShuttlelbTable());
1753
1754 TSQLResult* aResult = fServer[3]->Query(sqlQuery);
1755 if (!aResult) {
1756 AliError(Form("Can't execute query <%s>!", sqlQuery.Data()));
1757 return;
1758 }
1759
1760 AliDebug(2,Form("Query = %s", sqlQuery.Data()));
1761
1762 if (aResult->GetRowCount() == 0) {
1763 AliError(Form("No result for query %s received", sqlQuery.Data()));
1764 return;
1765 }
1766
1767 if (aResult->GetFieldCount() != 1) {
1768 AliError(Form("Invalid field count for query %s received", sqlQuery.Data()));
1769 return;
1770 }
1771
1772 TSQLRow* aRow = aResult->Next();
1773 if (!aRow) {
1774 AliError(Form("Could not receive result of query %s", sqlQuery.Data()));
1775 return;
1776 }
1777
1778 TString result(aRow->GetField(0), aRow->GetFieldLength(0));
1779 Int_t count = result.Atoi();
1780
1781 Log("SHUTTLE", Form("%d unprocessed runs", count));
1782
1783 delete aRow;
1784 delete aResult;
1785
1786 TMonaLisaValue mlStatus("SHUTTLE_openruns", count);
1787
1788 TList mlList;
1789 mlList.Add(&mlStatus);
1790
1791 fMonaLisa->SendParameters(&mlList, "__PROCESSINGINFO__");
1792}
1793
2bb7b766 1794//______________________________________________________________________________________________
1795Bool_t AliShuttle::QueryShuttleLogbook(const char* whereClause,
1796 TObjArray& entries)
1797{
9827400b 1798 // Query DAQ's Shuttle logbook and fills detector status object.
1799 // Call QueryRunParameters to query DAQ logbook for run parameters.
1800 //
2bb7b766 1801
fc5a4708 1802 entries.SetOwner(1);
1803
2bb7b766 1804 // check connection, in case connect
fbc112e3 1805 if (!Connect(3)) return kFALSE;
2bb7b766 1806
1807 TString sqlQuery;
441b0e9c 1808 sqlQuery = Form("select * from %s %s order by run", fConfig->GetShuttlelbTable(), whereClause);
2bb7b766 1809
be48e3ea 1810 TSQLResult* aResult = fServer[3]->Query(sqlQuery);
2bb7b766 1811 if (!aResult) {
1812 AliError(Form("Can't execute query <%s>!", sqlQuery.Data()));
1813 return kFALSE;
1814 }
1815
fc5a4708 1816 AliDebug(2,Form("Query = %s", sqlQuery.Data()));
1817
2bb7b766 1818 if(aResult->GetRowCount() == 0) {
c88ad5db 1819 Log("SHUTTLE", "No entries in Shuttle Logbook match request");
9827400b 1820 delete aResult;
1821 return kTRUE;
2bb7b766 1822 }
1823
1824 // TODO Check field count!
db99d43e 1825 const UInt_t nCols = 23;
2bb7b766 1826 if (aResult->GetFieldCount() != (Int_t) nCols) {
c88ad5db 1827 Log("SHUTTLE", "Invalid SQL result field number!");
2bb7b766 1828 delete aResult;
1829 return kFALSE;
1830 }
1831
2bb7b766 1832 TSQLRow* aRow;
1833 while ((aRow = aResult->Next())) {
1834 TString runString(aRow->GetField(0), aRow->GetFieldLength(0));
1835 Int_t run = runString.Atoi();
1836
eba76848 1837 AliShuttleLogbookEntry *entry = QueryRunParameters(run);
1838 if (!entry)
1839 continue;
2bb7b766 1840
1841 // loop on detectors
eba76848 1842 for(UInt_t ii = 0; ii < nCols; ii++)
1843 entry->SetDetectorStatus(aResult->GetFieldName(ii), aRow->GetField(ii));
2bb7b766 1844
eba76848 1845 entries.AddLast(entry);
2bb7b766 1846 delete aRow;
1847 }
1848
2bb7b766 1849 delete aResult;
1850 return kTRUE;
1851}
1852
1853//______________________________________________________________________________________________
eba76848 1854AliShuttleLogbookEntry* AliShuttle::QueryRunParameters(Int_t run)
2bb7b766 1855{
eba76848 1856 //
1857 // Retrieve run parameters written in the DAQ logbook and sets them into AliShuttleLogbookEntry object
1858 //
2bb7b766 1859
1860 // check connection, in case connect
be48e3ea 1861 if (!Connect(3))
eba76848 1862 return 0;
2bb7b766 1863
1864 TString sqlQuery;
2c15234c 1865 sqlQuery.Form("select * from %s where run=%d", fConfig->GetDAQlbTable(), run);
2bb7b766 1866
be48e3ea 1867 TSQLResult* aResult = fServer[3]->Query(sqlQuery);
2bb7b766 1868 if (!aResult) {
c88ad5db 1869 Log("SHUTTLE", Form("Can't execute query <%s>!", sqlQuery.Data()));
eba76848 1870 return 0;
2bb7b766 1871 }
1872
eba76848 1873 if (aResult->GetRowCount() == 0) {
2bb7b766 1874 Log("SHUTTLE", Form("QueryRunParameters - No entry in DAQ Logbook for run %d. Skipping", run));
1875 delete aResult;
eba76848 1876 return 0;
2bb7b766 1877 }
1878
eba76848 1879 if (aResult->GetRowCount() > 1) {
c88ad5db 1880 Log("SHUTTLE", Form("QueryRunParameters - UNEXPECTED: "
1881 "more than one entry in DAQ Logbook for run %d!", run));
2bb7b766 1882 delete aResult;
eba76848 1883 return 0;
2bb7b766 1884 }
1885
eba76848 1886 TSQLRow* aRow = aResult->Next();
1887 if (!aRow)
1888 {
c88ad5db 1889 Log("SHUTTLE", Form("QueryRunParameters - Could not retrieve row for run %d. Skipping", run));
eba76848 1890 delete aResult;
1891 return 0;
1892 }
2bb7b766 1893
eba76848 1894 AliShuttleLogbookEntry* entry = new AliShuttleLogbookEntry(run);
2bb7b766 1895
eba76848 1896 for (Int_t ii = 0; ii < aResult->GetFieldCount(); ii++)
1897 entry->SetRunParameter(aResult->GetFieldName(ii), aRow->GetField(ii));
2bb7b766 1898
c9db68c7 1899 delete aRow;
1900 delete aResult;
1901
eba76848 1902 UInt_t startTime = entry->GetStartTime();
1903 UInt_t endTime = entry->GetEndTime();
c9db68c7 1904
b0e53b15 1905 TString totEventsStr = entry->GetRunParameter("totalEvents");
1906 Int_t totEvents = totEventsStr.Atoi();
c9db68c7 1907
1908 if (startTime != 0 && endTime != 0 && endTime > startTime && totEvents > 0)
1909 return entry;
2bb7b766 1910
c9db68c7 1911 Log("SHUTTLE",
1912 Form("QueryRunParameters - Invalid parameters for Run %d: "
1913 "startTime = %d, endTime = %d. Skipping (Shuttle won't be marked as DONE)!",
1914 run, startTime, endTime));
eba76848 1915
c9db68c7 1916 if (totEvents < 1)
1917 Log("SHUTTLE", Form("QueryRunParameters - Run %d has 0 events - Skipping!", run));
1918
1919 //Log("SHUTTLE", Form("Marking SHUTTLE done for run %d", run));
1920 //fLogbookEntry = entry;
1921 //if (!UpdateShuttleLogbook("shuttle_done"))
1922 //{
1923 // AliError(Form("Could not update logbook for run %d !", run));
1924 //}
1925 //fLogbookEntry = 0;
1926
1927 delete entry;
1928 return 0;
2bb7b766 1929}
1930
a038aa70 1931//______________________________________________________________________________________________
1932TMap* AliShuttle::GetValueSet(const char* host, Int_t port, const TSeqCollection* entries,
542b6cc8 1933 DCSType type, Int_t multiSplit)
a038aa70 1934{
1935 // Retrieve all "entry" data points from the DCS server
1936 // host, port: TSocket connection parameters
1937 // entries: list of name of the alias or data point
1938 // type: kAlias or kDP
1939 // returns TMap of values, 0 when failure
542b6cc8 1940
1941 AliDCSClient client(host, port, fTimeout, fRetries, multiSplit);
b41b252a 1942
a038aa70 1943 TMap* result = 0;
b41b252a 1944 if (type == kAlias)
a038aa70 1945 {
b41b252a 1946 result = client.GetAliasValues(entries, GetCurrentStartTime(),
1947 GetCurrentEndTime());
1948 }
1949 else if (type == kDP)
1950 {
1951 result = client.GetDPValues(entries, GetCurrentStartTime(),
1952 GetCurrentEndTime());
1953 }
a038aa70 1954
b41b252a 1955 if (result == 0)
1956 {
1957 Log(fCurrentDetector.Data(), Form("GetValueSet - Can't get entries! Reason: %s",
1790d4b7 1958 client.GetErrorString(client.GetResultErrorCode())));
1959 if (client.GetResultErrorCode() == AliDCSClient::fgkServerError)
1960 Log(fCurrentDetector.Data(), Form("GetValueSet - Server error code: %s",
1961 client.GetServerError().Data()));
a038aa70 1962
b41b252a 1963 return 0;
a038aa70 1964 }
b41b252a 1965
a038aa70 1966 return result;
1967}
b41b252a 1968
b948db8d 1969//______________________________________________________________________________________________
57f50b3c 1970const char* AliShuttle::GetFile(Int_t system, const char* detector,
1971 const char* id, const char* source)
b948db8d 1972{
9827400b 1973 // Get calibration file from file exchange servers
1974 // First queris the FXS database for the file name, using the run, detector, id and source info
1975 // then calls RetrieveFile(filename) for actual copy to local disk
1976 // run: current run being processed (given by Logbook entry fLogbookEntry)
1977 // detector: the Preprocessor name
1978 // id: provided as a parameter by the Preprocessor
1979 // source: provided by the Preprocessor through GetFileSources function
1980
1981 // check if test mode should simulate a FXS error
1982 if (fTestMode & kErrorFXSFiles)
1983 {
1984 Log(detector, Form("GetFile - In TESTMODE - Simulating error while connecting to %s FXS", GetSystemName(system)));
1985 return 0;
1986 }
1987
57f50b3c 1988 // check connection, in case connect
9d733021 1989 if (!Connect(system))
eba76848 1990 {
9d733021 1991 Log(detector, Form("GetFile - Couldn't connect to %s FXS database", GetSystemName(system)));
339adafa 1992 fFXSError = system;
57f50b3c 1993 return 0;
1994 }
1995
1996 // Query preparation
9d733021 1997 TString sourceName(source);
d386d623 1998 Int_t nFields = 3;
1999 TString sqlQueryStart = Form("select filePath,size,fileChecksum from %s where",
2000 fConfig->GetFXSdbTable(system));
2001 TString whereClause = Form("run=%d and detector=\"%s\" and fileId=\"%s\"",
2002 GetCurrentRun(), detector, id);
2003
9d733021 2004 if (system == kDAQ)
2005 {
d386d623 2006 whereClause += Form(" and DAQsource=\"%s\"", source);
57f50b3c 2007 }
9d733021 2008 else if (system == kDCS)
eba76848 2009 {
9d733021 2010 sourceName="none";
57f50b3c 2011 }
9d733021 2012 else if (system == kHLT)
9e080f92 2013 {
d386d623 2014 whereClause += Form(" and DDLnumbers=\"%s\"", source);
9e080f92 2015 }
2016
9e080f92 2017 TString sqlQuery = Form("%s %s", sqlQueryStart.Data(), whereClause.Data());
2018
2019 AliDebug(2, Form("SQL query: \n%s",sqlQuery.Data()));
2020
2021 // Query execution
2022 TSQLResult* aResult = 0;
9d733021 2023 aResult = dynamic_cast<TSQLResult*> (fServer[system]->Query(sqlQuery));
9e080f92 2024 if (!aResult) {
8da81210 2025 Log(detector, Form("GetFile - Can't execute SQL query to %s database for: id = %s, source = %s",
9d733021 2026 GetSystemName(system), id, sourceName.Data()));
339adafa 2027 fFXSError = system;
9e080f92 2028 return 0;
2029 }
2030
8da81210 2031 if (aResult->GetRowCount() == 0)
9e080f92 2032 {
2033 Log(detector,
8da81210 2034 Form("GetFile - No entry in %s FXS db for: id = %s, source = %s",
9d733021 2035 GetSystemName(system), id, sourceName.Data()));
9e080f92 2036 delete aResult;
2037 return 0;
2038 }
2bb7b766 2039
9e080f92 2040 if (aResult->GetRowCount() > 1) {
2041 Log(detector,
8da81210 2042 Form("GetFile - More than one entry in %s FXS db for: id = %s, source = %s",
9d733021 2043 GetSystemName(system), id, sourceName.Data()));
339adafa 2044 fFXSError = system;
9e080f92 2045 delete aResult;
2046 return 0;
2047 }
2048
9d733021 2049 if (aResult->GetFieldCount() != nFields) {
9e080f92 2050 Log(detector,
9d733021 2051 Form("GetFileName - Wrong field count in %s FXS db for: id = %s, source = %s",
2052 GetSystemName(system), id, sourceName.Data()));
339adafa 2053 fFXSError = system;
9e080f92 2054 delete aResult;
2055 return 0;
2056 }
2057
2058 TSQLRow* aRow = dynamic_cast<TSQLRow*> (aResult->Next());
2059
2060 if (!aRow){
8da81210 2061 Log(detector, Form("GetFile - Empty set result in %s FXS db from query: id = %s, source = %s",
9d733021 2062 GetSystemName(system), id, sourceName.Data()));
339adafa 2063 fFXSError = system;
9e080f92 2064 delete aResult;
2065 return 0;
2066 }
2067
2068 TString filePath(aRow->GetField(0), aRow->GetFieldLength(0));
2069 TString fileSize(aRow->GetField(1), aRow->GetFieldLength(1));
d386d623 2070 TString fileChecksum(aRow->GetField(2), aRow->GetFieldLength(2));
9e080f92 2071
2072 delete aResult;
2073 delete aRow;
2074
d386d623 2075 AliDebug(2, Form("filePath = %s; size = %s, fileChecksum = %s",
2076 filePath.Data(), fileSize.Data(), fileChecksum.Data()));
9e080f92 2077
9e080f92 2078 // retrieved file is renamed to make it unique
675f64cd 2079 TString localFileName = Form("%s/%s_%d_process/%s_%s_%d_%s_%s.shuttle",
2080 GetShuttleTempDir(), detector, GetCurrentRun(),
d524ade6 2081 GetSystemName(system), detector, GetCurrentRun(),
2082 id, sourceName.Data());
9d733021 2083
9e080f92 2084
9d733021 2085 // file retrieval from FXS
4b95672b 2086 UInt_t nRetries = 0;
2087 UInt_t maxRetries = 3;
2088 Bool_t result = kFALSE;
2089
2090 // copy!! if successful TSystem::Exec returns 0
8da81210 2091 while (nRetries++ < maxRetries) {
4b95672b 2092 AliDebug(2, Form("Trying to copy file. Retry # %d", nRetries));
2093 result = RetrieveFile(system, filePath.Data(), localFileName.Data());
8da81210 2094 if (!result)
4b95672b 2095 {
8da81210 2096 Log(detector, Form("GetFile - Copy of file %s from %s FXS failed",
9d733021 2097 filePath.Data(), GetSystemName(system)));
4b95672b 2098 continue;
4f0749a8 2099 }
9e080f92 2100
c98417a1 2101 if (fileSize.Length()>0)
2102 {
2103 // compare filesize of local file with the one stored in the FXS DB
926cbc0e 2104 Long_t size = -1;
2105 Int_t sizeComp = gSystem->GetPathInfo(localFileName.Data(), 0, &size, 0, 0);
c98417a1 2106
926cbc0e 2107 if (sizeComp != 0 || size != fileSize.Atoi())
c98417a1 2108 {
8da81210 2109 Log(detector, Form("GetFile - size of file %s does not match with local copy!",
c98417a1 2110 filePath.Data()));
2111 result = kFALSE;
2112 continue;
2113 }
2114
2115 } else {
2116 Log(fCurrentDetector, Form("GetFile - size of file %s not set in %s database, skipping comparison",
2117 filePath.Data(), GetSystemName(system)));
2118 }
2119
d386d623 2120 if (fileChecksum.Length()>0)
4b95672b 2121 {
2122 // compare md5sum of local file with the one stored in the FXS DB
95d4aaef 2123 if(fileChecksum.Contains(' ')) fileChecksum.Resize(fileChecksum.First(' '));
339adafa 2124 Int_t md5Comp = gSystem->Exec(Form("md5sum %s |grep %s > /dev/null 2> /dev/null",
d524ade6 2125 localFileName.Data(), fileChecksum.Data()));
9e080f92 2126
4b95672b 2127 if (md5Comp != 0)
2128 {
8da81210 2129 Log(detector, Form("GetFile - md5sum of file %s does not match with local copy!",
4b95672b 2130 filePath.Data()));
2131 result = kFALSE;
2132 continue;
2133 }
d386d623 2134 } else {
2135 Log(fCurrentDetector, Form("GetFile - md5sum of file %s not set in %s database, skipping comparison",
2136 filePath.Data(), GetSystemName(system)));
9d733021 2137 }
4b95672b 2138 if (result) break;
9e080f92 2139 }
2140
8da81210 2141 if (!result)
2142 {
339adafa 2143 fFXSError = system;
8da81210 2144 return 0;
2145 }
4b95672b 2146
9d733021 2147 fFXSCalled[system]=kTRUE;
2148 TObjString *fileParams = new TObjString(Form("%s#!?!#%s", id, sourceName.Data()));
2149 fFXSlist[system].Add(fileParams);
9e080f92 2150
675f64cd 2151 static TString staticLocalFileName;
2152 staticLocalFileName.Form("%s", localFileName.Data());
2153
c88ad5db 2154 Log(fCurrentDetector, Form("GetFile - Retrieved file with id %s and "
2155 "source %s from %s to %s", id, source,
d524ade6 2156 GetSystemName(system), localFileName.Data()));
675f64cd 2157
d524ade6 2158 return staticLocalFileName.Data();
2bb7b766 2159}
2160
2161//______________________________________________________________________________________________
9d733021 2162Bool_t AliShuttle::RetrieveFile(UInt_t system, const char* fxsFileName, const char* localFileName)
9e080f92 2163{
9827400b 2164 //
2165 // Copies file from FXS to local Shuttle machine
2166 //
2bb7b766 2167
9e080f92 2168 // check temp directory: trying to cd to temp; if it does not exist, create it
d524ade6 2169 AliDebug(2, Form("Copy file %s from %s FXS into %s",
2170 GetSystemName(system), fxsFileName, localFileName));
2171
2172 TString tmpDir(localFileName);
2173
2174 tmpDir = tmpDir(0,tmpDir.Last('/'));
9e080f92 2175
d524ade6 2176 Int_t noDir = gSystem->GetPathInfo(tmpDir.Data(), 0, (Long64_t*) 0, 0, 0);
2177 if (noDir) // temp dir does not exists!
2178 {
2179 if (gSystem->mkdir(tmpDir.Data(), 1))
2180 {
2181 Log(fCurrentDetector.Data(), "RetrieveFile - could not make temp directory!!");
9e080f92 2182 return kFALSE;
2183 }
9e080f92 2184 }
2185
339adafa 2186 TString command = Form("scp -oPort=%d -2 %s@%s:%s/%s %s",
9d733021 2187 fConfig->GetFXSPort(system),
2188 fConfig->GetFXSUser(system),
2189 fConfig->GetFXSHost(system),
339adafa 2190 fConfig->GetFXSBaseFolder(system),
9d733021 2191 fxsFileName,
9e080f92 2192 localFileName);
2193
2194 AliDebug(2, Form("%s",command.Data()));
2195
4b95672b 2196 Bool_t result = (gSystem->Exec(command.Data()) == 0);
9e080f92 2197
4b95672b 2198 return result;
9e080f92 2199}
2200
2201//______________________________________________________________________________________________
9d733021 2202TList* AliShuttle::GetFileSources(Int_t system, const char* detector, const char* id)
2203{
9827400b 2204 //
2205 // Get sources producing the condition file Id from file exchange servers
4a33bdd9 2206 // if id is NULL all sources are returned (distinct)
9827400b 2207 //
1bcd28db 2208
8884d71e 2209 if (id)
2210 {
2211 Log(detector, Form("GetFileSources - Querying %s FXS for files with id %s produced by %s", GetSystemName(system), id, detector));
2212 } else {
2213 Log(detector, Form("GetFileSources - Querying %s FXS for files produced by %s", GetSystemName(system), detector));
2214 }
9827400b 2215
2216 // check if test mode should simulate a FXS error
2217 if (fTestMode & kErrorFXSSources)
2218 {
2219 Log(detector, Form("GetFileSources - In TESTMODE - Simulating error while connecting to %s FXS", GetSystemName(system)));
2220 return 0;
2221 }
2222
9d733021 2223 if (system == kDCS)
2224 {
c88ad5db 2225 Log(detector, "GetFileSources - WARNING: DCS system has only one source of data!");
6297b37d 2226 TList *list = new TList();
2227 list->SetOwner(1);
2228 list->Add(new TObjString(" "));
2229 return list;
9d733021 2230 }
9e080f92 2231
2232 // check connection, in case connect
9d733021 2233 if (!Connect(system))
2234 {
4a33bdd9 2235 Log(detector, Form("GetFileSources - Couldn't connect to %s FXS database", GetSystemName(system)));
339adafa 2236 fFXSError = system;
9d733021 2237 return NULL;
9e080f92 2238 }
2239
9d733021 2240 TString sourceName = 0;
2241 if (system == kDAQ)
2242 {
2243 sourceName = "DAQsource";
2244 } else if (system == kHLT)
2245 {
2246 sourceName = "DDLnumbers";
2247 }
2248
4a33bdd9 2249 TString sqlQueryStart = Form("select distinct %s from %s where", sourceName.Data(), fConfig->GetFXSdbTable(system));
2250 TString whereClause = Form("run=%d and detector=\"%s\"",
2251 GetCurrentRun(), detector);
2252 if (id)
2253 whereClause += Form(" and fileId=\"%s\"", id);
9e080f92 2254 TString sqlQuery = Form("%s %s", sqlQueryStart.Data(), whereClause.Data());
2255
2256 AliDebug(2, Form("SQL query: \n%s",sqlQuery.Data()));
2257
2258 // Query execution
2259 TSQLResult* aResult;
9d733021 2260 aResult = fServer[system]->Query(sqlQuery);
9e080f92 2261 if (!aResult) {
9d733021 2262 Log(detector, Form("GetFileSources - Can't execute SQL query to %s database for id: %s",
2263 GetSystemName(system), id));
339adafa 2264 fFXSError = system;
9e080f92 2265 return 0;
2266 }
2267
86aa42c3 2268 TList *list = new TList();
2269 list->SetOwner(1);
2270
9d733021 2271 if (aResult->GetRowCount() == 0)
2272 {
9e080f92 2273 Log(detector,
9d733021 2274 Form("GetFileSources - No entry in %s FXS table for id: %s", GetSystemName(system), id));
9e080f92 2275 delete aResult;
86aa42c3 2276 return list;
9e080f92 2277 }
2278
1bcd28db 2279 Log(detector, Form("GetFileSources - Found %d sources", aResult->GetRowCount()));
9e080f92 2280
1bcd28db 2281 TSQLRow* aRow;
9d733021 2282 while ((aRow = aResult->Next()))
2283 {
9e080f92 2284
9d733021 2285 TString source(aRow->GetField(0), aRow->GetFieldLength(0));
2286 AliDebug(2, Form("%s = %s", sourceName.Data(), source.Data()));
2287 list->Add(new TObjString(source));
9e080f92 2288 delete aRow;
2289 }
9d733021 2290
9e080f92 2291 delete aResult;
2292
2293 return list;
2bb7b766 2294}
2295
4a33bdd9 2296//______________________________________________________________________________________________
2297TList* AliShuttle::GetFileIDs(Int_t system, const char* detector, const char* source)
2298{
2299 //
2300 // Get all ids of condition files produced by a given source from file exchange servers
2301 //
2302
1bcd28db 2303 Log(detector, Form("GetFileIDs - Retrieving ids with source %s with %s", source, GetSystemName(system)));
2304
4a33bdd9 2305 // check if test mode should simulate a FXS error
2306 if (fTestMode & kErrorFXSSources)
2307 {
2308 Log(detector, Form("GetFileIDs - In TESTMODE - Simulating error while connecting to %s FXS", GetSystemName(system)));
2309 return 0;
2310 }
2311
2312 // check connection, in case connect
2313 if (!Connect(system))
2314 {
2315 Log(detector, Form("GetFileIDs - Couldn't connect to %s FXS database", GetSystemName(system)));
2316 return NULL;
2317 }
2318
2319 TString sourceName = 0;
2320 if (system == kDAQ)
2321 {
2322 sourceName = "DAQsource";
2323 } else if (system == kHLT)
2324 {
2325 sourceName = "DDLnumbers";
2326 }
2327
2328 TString sqlQueryStart = Form("select fileId from %s where", fConfig->GetFXSdbTable(system));
2329 TString whereClause = Form("run=%d and detector=\"%s\"",
2330 GetCurrentRun(), detector);
2331 if (sourceName.Length() > 0 && source)
2332 whereClause += Form(" and %s=\"%s\"", sourceName.Data(), source);
2333 TString sqlQuery = Form("%s %s", sqlQueryStart.Data(), whereClause.Data());
2334
2335 AliDebug(2, Form("SQL query: \n%s",sqlQuery.Data()));
2336
2337 // Query execution
2338 TSQLResult* aResult;
2339 aResult = fServer[system]->Query(sqlQuery);
2340 if (!aResult) {
2341 Log(detector, Form("GetFileIDs - Can't execute SQL query to %s database for source: %s",
2342 GetSystemName(system), source));
2343 return 0;
2344 }
2345
2346 TList *list = new TList();
2347 list->SetOwner(1);
2348
2349 if (aResult->GetRowCount() == 0)
2350 {
2351 Log(detector,
2352 Form("GetFileIDs - No entry in %s FXS table for source: %s", GetSystemName(system), source));
2353 delete aResult;
2354 return list;
2355 }
2356
1bcd28db 2357 Log(detector, Form("GetFileIDs - Found %d ids", aResult->GetRowCount()));
2358
4a33bdd9 2359 TSQLRow* aRow;
2360
2361 while ((aRow = aResult->Next()))
2362 {
2363
2364 TString id(aRow->GetField(0), aRow->GetFieldLength(0));
2365 AliDebug(2, Form("fileId = %s", id.Data()));
2366 list->Add(new TObjString(id));
2367 delete aRow;
2368 }
2369
2370 delete aResult;
2371
2372 return list;
2373}
2374
2bb7b766 2375//______________________________________________________________________________________________
9d733021 2376Bool_t AliShuttle::Connect(Int_t system)
2bb7b766 2377{
9827400b 2378 // Connect to MySQL Server of the system's FXS MySQL databases
2379 // DAQ Logbook, Shuttle Logbook and DAQ FXS db are on the same host
2380 //
57f50b3c 2381
9d733021 2382 // check connection: if already connected return
2383 if(fServer[system] && fServer[system]->IsConnected()) return kTRUE;
57f50b3c 2384
9d733021 2385 TString dbHost, dbUser, dbPass, dbName;
57f50b3c 2386
9d733021 2387 if (system < 3) // FXS db servers
2388 {
2389 dbHost = Form("mysql://%s:%d", fConfig->GetFXSdbHost(system), fConfig->GetFXSdbPort(system));
2390 dbUser = fConfig->GetFXSdbUser(system);
2391 dbPass = fConfig->GetFXSdbPass(system);
2392 dbName = fConfig->GetFXSdbName(system);
2393 } else { // Run & Shuttle logbook servers
2394 // TODO Will the Shuttle logbook server be the same as the Run logbook server ???
2395 dbHost = Form("mysql://%s:%d", fConfig->GetDAQlbHost(), fConfig->GetDAQlbPort());
2396 dbUser = fConfig->GetDAQlbUser();
2397 dbPass = fConfig->GetDAQlbPass();
2398 dbName = fConfig->GetDAQlbDB();
2399 }
57f50b3c 2400
9d733021 2401 fServer[system] = TSQLServer::Connect(dbHost.Data(), dbUser.Data(), dbPass.Data());
2402 if (!fServer[system] || !fServer[system]->IsConnected()) {
2403 if(system < 3)
2404 {
2405 AliError(Form("Can't establish connection to FXS database for %s",
2406 AliShuttleInterface::GetSystemName(system)));
2407 } else {
2408 AliError("Can't establish connection to Run logbook.");
57f50b3c 2409 }
9d733021 2410 if(fServer[system]) delete fServer[system];
2411 return kFALSE;
2bb7b766 2412 }
57f50b3c 2413
9d733021 2414 // Get tables
2415 TSQLResult* aResult=0;
2416 switch(system){
2417 case kDAQ:
2418 aResult = fServer[kDAQ]->GetTables(dbName.Data());
2419 break;
2420 case kDCS:
2421 aResult = fServer[kDCS]->GetTables(dbName.Data());
2422 break;
2423 case kHLT:
2424 aResult = fServer[kHLT]->GetTables(dbName.Data());
2425 break;
2426 default:
2427 aResult = fServer[3]->GetTables(dbName.Data());
2428 break;
2429 }
2430
2431 delete aResult;
2bb7b766 2432 return kTRUE;
2433}
57f50b3c 2434
9e080f92 2435//______________________________________________________________________________________________
9d733021 2436Bool_t AliShuttle::UpdateTable()
9e080f92 2437{
9827400b 2438 //
2439 // Update FXS table filling time_processed field in all rows corresponding to current run and detector
2440 //
9e080f92 2441
9d733021 2442 Bool_t result = kTRUE;
9e080f92 2443
9d733021 2444 for (UInt_t system=0; system<3; system++)
2445 {
2446 if(!fFXSCalled[system]) continue;
9e080f92 2447
9d733021 2448 // check connection, in case connect
2449 if (!Connect(system))
2450 {
2451 Log(fCurrentDetector, Form("UpdateTable - Couldn't connect to %s FXS database", GetSystemName(system)));
2452 result = kFALSE;
2453 continue;
9e080f92 2454 }
9e080f92 2455
9d733021 2456 TTimeStamp now; // now
2457
2458 // Loop on FXS list entries
2459 TIter iter(&fFXSlist[system]);
2460 TObjString *aFXSentry=0;
2461 while ((aFXSentry = dynamic_cast<TObjString*> (iter.Next())))
2462 {
2463 TString aFXSentrystr = aFXSentry->String();
2464 TObjArray *aFXSarray = aFXSentrystr.Tokenize("#!?!#");
2465 if (!aFXSarray || aFXSarray->GetEntries() != 2 )
2466 {
2467 Log(fCurrentDetector, Form("UpdateTable - error updating %s FXS entry. Check string: <%s>",
2468 GetSystemName(system), aFXSentrystr.Data()));
2469 if(aFXSarray) delete aFXSarray;
2470 result = kFALSE;
2471 continue;
2472 }
2473 const char* fileId = ((TObjString*) aFXSarray->At(0))->GetName();
2474 const char* source = ((TObjString*) aFXSarray->At(1))->GetName();
2475
2476 TString whereClause;
2477 if (system == kDAQ)
2478 {
2479 whereClause = Form("where run=%d and detector=\"%s\" and fileId=\"%s\" and DAQsource=\"%s\";",
2480 GetCurrentRun(), fCurrentDetector.Data(), fileId, source);
2481 }
2482 else if (system == kDCS)
2483 {
2484 whereClause = Form("where run=%d and detector=\"%s\" and fileId=\"%s\";",
2485 GetCurrentRun(), fCurrentDetector.Data(), fileId);
2486 }
2487 else if (system == kHLT)
2488 {
2489 whereClause = Form("where run=%d and detector=\"%s\" and fileId=\"%s\" and DDLnumbers=\"%s\";",
2490 GetCurrentRun(), fCurrentDetector.Data(), fileId, source);
2491 }
2492
2493 delete aFXSarray;
9e080f92 2494
9d733021 2495 TString sqlQuery = Form("update %s set time_processed=%d %s", fConfig->GetFXSdbTable(system),
2496 now.GetSec(), whereClause.Data());
9e080f92 2497
9d733021 2498 AliDebug(2, Form("SQL query: \n%s",sqlQuery.Data()));
9e080f92 2499
9d733021 2500 // Query execution
2501 TSQLResult* aResult;
2502 aResult = dynamic_cast<TSQLResult*> (fServer[system]->Query(sqlQuery));
2503 if (!aResult)
2504 {
2505 Log(fCurrentDetector, Form("UpdateTable - %s db: can't execute SQL query <%s>",
2506 GetSystemName(system), sqlQuery.Data()));
2507 result = kFALSE;
2508 continue;
2509 }
2510 delete aResult;
9e080f92 2511 }
9e080f92 2512 }
2513
9d733021 2514 return result;
9e080f92 2515}
57f50b3c 2516
3301427a 2517//______________________________________________________________________________________________
2518Bool_t AliShuttle::UpdateTableFailCase()
2519{
9827400b 2520 // Update FXS table filling time_processed field in all rows corresponding to current run and detector
2521 // this is called in case the preprocessor is declared failed for the current run, because
2522 // the fields are updated only in case of success
3301427a 2523
2524 Bool_t result = kTRUE;
2525
2526 for (UInt_t system=0; system<3; system++)
2527 {
2528 // check connection, in case connect
2529 if (!Connect(system))
2530 {
2531 Log(fCurrentDetector, Form("UpdateTableFailCase - Couldn't connect to %s FXS database",
2532 GetSystemName(system)));
2533 result = kFALSE;
2534 continue;
2535 }
2536
2537 TTimeStamp now; // now
2538
2539 // Loop on FXS list entries
2540
2541 TString whereClause = Form("where run=%d and detector=\"%s\";",
2542 GetCurrentRun(), fCurrentDetector.Data());
2543
2544
2545 TString sqlQuery = Form("update %s set time_processed=%d %s", fConfig->GetFXSdbTable(system),
2546 now.GetSec(), whereClause.Data());
2547
2548 AliDebug(2, Form("SQL query: \n%s",sqlQuery.Data()));
2549
2550 // Query execution
2551 TSQLResult* aResult;
2552 aResult = dynamic_cast<TSQLResult*> (fServer[system]->Query(sqlQuery));
2553 if (!aResult)
2554 {
2555 Log(fCurrentDetector, Form("UpdateTableFailCase - %s db: can't execute SQL query <%s>",
2556 GetSystemName(system), sqlQuery.Data()));
2557 result = kFALSE;
2558 continue;
2559 }
2560 delete aResult;
2561 }
2562
2563 return result;
2564}
2565
2bb7b766 2566//______________________________________________________________________________________________
2567Bool_t AliShuttle::UpdateShuttleLogbook(const char* detector, const char* status)
2568{
e7f62f16 2569 //
2570 // Update Shuttle logbook filling detector or shuttle_done column
2571 // ex. of usage: UpdateShuttleLogbook("PHOS", "DONE") or UpdateShuttleLogbook("shuttle_done")
2572 //
57f50b3c 2573
2bb7b766 2574 // check connection, in case connect
be48e3ea 2575 if(!Connect(3)){
2bb7b766 2576 Log("SHUTTLE", "UpdateShuttleLogbook - Couldn't connect to DAQ Logbook.");
2577 return kFALSE;
57f50b3c 2578 }
2579
2bb7b766 2580 TString detName(detector);
2581 TString setClause;
ee6f7523 2582 if (detName == "shuttle_done" || detName == "shuttle_ignored")
e7f62f16 2583 {
2bb7b766 2584 setClause = "set shuttle_done=1";
e7f62f16 2585
ee6f7523 2586 if (detName == "shuttle_done")
b0e53b15 2587 {
503ff24f 2588 if (TouchFile()==kTRUE){
2589 //Send the information to ML
2590 TMonaLisaText mlStatus("SHUTTLE_status", "Done");
e7f62f16 2591
503ff24f 2592 TList mlList;
2593 mlList.Add(&mlStatus);
2594
2595 TString mlID;
2596 mlID.Form("%d", GetCurrentRun());
2597 fMonaLisa->SendParameters(&mlList, mlID);
2598 }
2599 else{
2600 return kFALSE;
2601 }
2602
b0e53b15 2603 }
2bb7b766 2604 } else {
2bb7b766 2605 TString statusStr(status);
2606 if(statusStr.Contains("done", TString::kIgnoreCase) ||
2607 statusStr.Contains("failed", TString::kIgnoreCase)){
eba76848 2608 setClause = Form("set %s=\"%s\"", detector, status);
2bb7b766 2609 } else {
2610 Log("SHUTTLE",
2611 Form("UpdateShuttleLogbook - Invalid status <%s> for detector %s",
2612 status, detector));
2613 return kFALSE;
2614 }
2615 }
57f50b3c 2616
2bb7b766 2617 TString whereClause = Form("where run=%d", GetCurrentRun());
2618
441b0e9c 2619 TString sqlQuery = Form("update %s %s %s",
2620 fConfig->GetShuttlelbTable(), setClause.Data(), whereClause.Data());
57f50b3c 2621
2bb7b766 2622 AliDebug(2, Form("SQL query: \n%s",sqlQuery.Data()));
2623
2624 // Query execution
2625 TSQLResult* aResult;
be48e3ea 2626 aResult = dynamic_cast<TSQLResult*> (fServer[3]->Query(sqlQuery));
2bb7b766 2627 if (!aResult) {
2628 Log("SHUTTLE", Form("UpdateShuttleLogbook - Can't execute query <%s>", sqlQuery.Data()));
2629 return kFALSE;
57f50b3c 2630 }
2bb7b766 2631 delete aResult;
57f50b3c 2632
2633 return kTRUE;
2634}
2635
2636//______________________________________________________________________________________________
2bb7b766 2637Int_t AliShuttle::GetCurrentRun() const
2638{
9827400b 2639 //
2640 // Get current run from logbook entry
2641 //
57f50b3c 2642
2bb7b766 2643 return fLogbookEntry ? fLogbookEntry->GetRun() : -1;
57f50b3c 2644}
2645
2646//______________________________________________________________________________________________
2bb7b766 2647UInt_t AliShuttle::GetCurrentStartTime() const
2648{
9827400b 2649 //
2650 // get current start time
2651 //
57f50b3c 2652
2bb7b766 2653 return fLogbookEntry ? fLogbookEntry->GetStartTime() : 0;
57f50b3c 2654}
2655
2656//______________________________________________________________________________________________
2bb7b766 2657UInt_t AliShuttle::GetCurrentEndTime() const
2658{
9827400b 2659 //
2660 // get current end time from logbook entry
2661 //
57f50b3c 2662
2bb7b766 2663 return fLogbookEntry ? fLogbookEntry->GetEndTime() : 0;
57f50b3c 2664}
2665
675f64cd 2666//______________________________________________________________________________________________
2667UInt_t AliShuttle::GetCurrentYear() const
2668{
2669 //
2670 // Get current year from logbook entry
2671 //
2672
2673 if (!fLogbookEntry) return 0;
2674
2675 TTimeStamp startTime(GetCurrentStartTime());
2676 TString year = Form("%d",startTime.GetDate());
2677 year = year(0,4);
2678
2679 return year.Atoi();
2680}
2681
2682//______________________________________________________________________________________________
2683const char* AliShuttle::GetLHCPeriod() const
2684{
2685 //
2686 // Get current LHC period from logbook entry
2687 //
2688
2689 if (!fLogbookEntry) return 0;
2690
2691 return fLogbookEntry->GetRunParameter("LHCperiod");
2692}
2693
b948db8d 2694//______________________________________________________________________________________________
2695void AliShuttle::Log(const char* detector, const char* message)
2696{
9827400b 2697 //
2698 // Fill log string with a message
2699 //
b948db8d 2700
7d4cf768 2701 TString logRunDir = GetShuttleLogDir();
2702 if (GetCurrentRun() >=0)
2703 logRunDir += Form("/%d", GetCurrentRun());
2704
2705 void* dir = gSystem->OpenDirectory(logRunDir.Data());
84090f85 2706 if (dir == NULL) {
7d4cf768 2707 if (gSystem->mkdir(logRunDir.Data(), kTRUE)) {
36c99a6a 2708 AliError(Form("Can't open directory <%s>", GetShuttleLogDir()));
84090f85 2709 return;
2710 }
b948db8d 2711
84090f85 2712 } else {
2713 gSystem->FreeDirectory(dir);
2714 }
b948db8d 2715
cb343cfd 2716 TString toLog = Form("%s (%d): %s - ", TTimeStamp(time(0)).AsString("s"), getpid(), detector);
e7f62f16 2717 if (GetCurrentRun() >= 0)
2718 toLog += Form("run %d - ", GetCurrentRun());
2bb7b766 2719 toLog += Form("%s", message);
2720
84090f85 2721 AliInfo(toLog.Data());
ffa29e93 2722
2723 // if we redirect the log output already to the file, leave here
2724 if (fOutputRedirected && strcmp(detector, "SHUTTLE") != 0)
2725 return;
b948db8d 2726
ffa29e93 2727 TString fileName = GetLogFileName(detector);
e7f62f16 2728
84090f85 2729 gSystem->ExpandPathName(fileName);
2730
2731 ofstream logFile;
2732 logFile.open(fileName, ofstream::out | ofstream::app);
2733
2734 if (!logFile.is_open()) {
2735 AliError(Form("Could not open file %s", fileName.Data()));
2736 return;
2737 }
7bfb2090 2738
84090f85 2739 logFile << toLog.Data() << "\n";
b948db8d 2740
84090f85 2741 logFile.close();
b948db8d 2742}
2bb7b766 2743
ffa29e93 2744//______________________________________________________________________________________________
2745TString AliShuttle::GetLogFileName(const char* detector) const
2746{
2747 //
2748 // returns the name of the log file for a given sub detector
2749 //
2750
2751 TString fileName;
2752
2753 if (GetCurrentRun() >= 0)
7d4cf768 2754 {
2755 fileName.Form("%s/%d/%s_%d.log", GetShuttleLogDir(), GetCurrentRun(),
2756 detector, GetCurrentRun());
2757 } else {
ffa29e93 2758 fileName.Form("%s/%s.log", GetShuttleLogDir(), detector);
7d4cf768 2759 }
ffa29e93 2760
2761 return fileName;
2762}
2763
ee6f7523 2764//______________________________________________________________________________________________
2765void AliShuttle::SendAlive()
2766{
2767 // sends alive message to ML
2768
2769 TMonaLisaText mlStatus("SHUTTLE_status", "Alive");
2770
2771 TList mlList;
2772 mlList.Add(&mlStatus);
2773
2774 fMonaLisa->SendParameters(&mlList, "__PROCESSINGINFO__");
2775}
2776
2bb7b766 2777//______________________________________________________________________________________________
2778Bool_t AliShuttle::Collect(Int_t run)
2779{
9827400b 2780 //
2781 // Collects conditions data for all UNPROCESSED run written to DAQ LogBook in case of run = -1 (default)
2782 // If a dedicated run is given this run is processed
2783 //
2784 // In operational mode, this is the Shuttle function triggered by the EOR signal.
2785 //
2bb7b766 2786
eba76848 2787 if (run == -1)
2788 Log("SHUTTLE","Collect - Shuttle called. Collecting conditions data for unprocessed runs");
2789 else
2790 Log("SHUTTLE", Form("Collect - Shuttle called. Collecting conditions data for run %d", run));
cb343cfd 2791
2792 SetLastAction("Starting");
2bb7b766 2793
ee6f7523 2794 // create ML instance
2795 if (!fMonaLisa)
2796 fMonaLisa = new TMonaLisaWriter(fConfig->GetMonitorHost(), fConfig->GetMonitorTable());
2797
ee6f7523 2798 SendAlive();
fbc112e3 2799 CountOpenRuns();
ee6f7523 2800
2bb7b766 2801 TString whereClause("where shuttle_done=0");
eba76848 2802 if (run != -1)
2803 whereClause += Form(" and run=%d", run);
2bb7b766 2804
2805 TObjArray shuttleLogbookEntries;
be48e3ea 2806 if (!QueryShuttleLogbook(whereClause, shuttleLogbookEntries))
2807 {
cb343cfd 2808 Log("SHUTTLE", "Collect - Can't retrieve entries from Shuttle logbook");
2bb7b766 2809 return kFALSE;
2810 }
2811
9e080f92 2812 if (shuttleLogbookEntries.GetEntries() == 0)
2813 {
2814 if (run == -1)
2815 Log("SHUTTLE","Collect - Found no UNPROCESSED runs in Shuttle logbook");
2816 else
2817 Log("SHUTTLE", Form("Collect - Run %d is already DONE "
2818 "or it does not exist in Shuttle logbook", run));
2819 return kTRUE;
2820 }
2821
be48e3ea 2822 for (UInt_t iDet=0; iDet<NDetectors(); iDet++)
2823 fFirstUnprocessed[iDet] = kTRUE;
2824
fc5a4708 2825 if (run != -1)
be48e3ea 2826 {
2827 // query Shuttle logbook for earlier runs, check if some detectors are unprocessed,
2828 // flag them into fFirstUnprocessed array
2829 TString whereClause(Form("where shuttle_done=0 and run < %d", run));
2830 TObjArray tmpLogbookEntries;
2831 if (!QueryShuttleLogbook(whereClause, tmpLogbookEntries))
2832 {
2833 Log("SHUTTLE", "Collect - Can't retrieve entries from Shuttle logbook");
2834 return kFALSE;
2835 }
2836
2837 TIter iter(&tmpLogbookEntries);
2838 AliShuttleLogbookEntry* anEntry = 0;
2839 while ((anEntry = dynamic_cast<AliShuttleLogbookEntry*> (iter.Next())))
2840 {
2841 for (UInt_t iDet=0; iDet<NDetectors(); iDet++)
2842 {
2843 if (anEntry->GetDetectorStatus(iDet) == AliShuttleLogbookEntry::kUnprocessed)
2844 {
2845 AliDebug(2, Form("Run %d: setting %s as \"not first time unprocessed\"",
2846 anEntry->GetRun(), GetDetName(iDet)));
2847 fFirstUnprocessed[iDet] = kFALSE;
2848 }
2849 }
2850
2851 }
2852
2853 }
2854
2855 if (!RetrieveConditionsData(shuttleLogbookEntries))
2856 {
cb343cfd 2857 Log("SHUTTLE", "Collect - Process of at least one run failed");
a3ca69fe 2858 CountOpenRuns();
2bb7b766 2859 return kFALSE;
2860 }
2861
36c99a6a 2862 Log("SHUTTLE", "Collect - Requested run(s) successfully processed");
a3ca69fe 2863 CountOpenRuns();
eba76848 2864 return kTRUE;
2bb7b766 2865}
2866
2bb7b766 2867//______________________________________________________________________________________________
2868Bool_t AliShuttle::RetrieveConditionsData(const TObjArray& dateEntries)
2869{
9827400b 2870 //
2871 // Retrieve conditions data for all runs that aren't processed yet
2872 //
2bb7b766 2873
2874 Bool_t hasError = kFALSE;
2875
2876 TIter iter(&dateEntries);
2877 AliShuttleLogbookEntry* anEntry;
2878
2879 while ((anEntry = (AliShuttleLogbookEntry*) iter.Next())){
2880 if (!Process(anEntry)){
2881 hasError = kTRUE;
2882 }
4b95672b 2883
2884 // clean SHUTTLE temp directory
d524ade6 2885 //TString filename = Form("%s/*.shuttle", GetShuttleTempDir());
2886 //RemoveFile(filename.Data());
2bb7b766 2887 }
2888
2889 return hasError == kFALSE;
2890}
cb343cfd 2891
2892//______________________________________________________________________________________________
2893ULong_t AliShuttle::GetTimeOfLastAction() const
2894{
9827400b 2895 //
2896 // Gets time of last action
2897 //
2898
cb343cfd 2899 ULong_t tmp;
36c99a6a 2900
cb343cfd 2901 fMonitoringMutex->Lock();
be48e3ea 2902
cb343cfd 2903 tmp = fLastActionTime;
36c99a6a 2904
cb343cfd 2905 fMonitoringMutex->UnLock();
36c99a6a 2906
cb343cfd 2907 return tmp;
2908}
2909
2910//______________________________________________________________________________________________
2911const TString AliShuttle::GetLastAction() const
2912{
9827400b 2913 //
cb343cfd 2914 // returns a string description of the last action
9827400b 2915 //
cb343cfd 2916
2917 TString tmp;
36c99a6a 2918
cb343cfd 2919 fMonitoringMutex->Lock();
2920
2921 tmp = fLastAction;
2922
2923 fMonitoringMutex->UnLock();
2924
36c99a6a 2925 return tmp;
cb343cfd 2926}
2927
2928//______________________________________________________________________________________________
2929void AliShuttle::SetLastAction(const char* action)
2930{
9827400b 2931 //
cb343cfd 2932 // updates the monitoring variables
9827400b 2933 //
36c99a6a 2934
cb343cfd 2935 fMonitoringMutex->Lock();
36c99a6a 2936
cb343cfd 2937 fLastAction = action;
2938 fLastActionTime = time(0);
2939
2940 fMonitoringMutex->UnLock();
2941}
eba76848 2942
2943//______________________________________________________________________________________________
2944const char* AliShuttle::GetRunParameter(const char* param)
2945{
9827400b 2946 //
2947 // returns run parameter read from DAQ logbook
2948 //
eba76848 2949
2950 if(!fLogbookEntry) {
2951 AliError("No logbook entry!");
2952 return 0;
2953 }
2954
2955 return fLogbookEntry->GetRunParameter(param);
2956}
57c1a579 2957
d386d623 2958//______________________________________________________________________________________________
9827400b 2959AliCDBEntry* AliShuttle::GetFromOCDB(const char* detector, const AliCDBPath& path)
d386d623 2960{
9827400b 2961 //
2962 // returns object from OCDB valid for current run
2963 //
d386d623 2964
9827400b 2965 if (fTestMode & kErrorOCDB)
2966 {
2967 Log(detector, "GetFromOCDB - In TESTMODE - Simulating error with OCDB");
2968 return 0;
2969 }
2970
d386d623 2971 AliCDBStorage *sto = AliCDBManager::Instance()->GetStorage(fgkMainCDB);
2972 if (!sto)
2973 {
9827400b 2974 Log(detector, "GetFromOCDB - Cannot activate main OCDB for query!");
d386d623 2975 return 0;
2976 }
2977
2978 return dynamic_cast<AliCDBEntry*> (sto->Get(path, GetCurrentRun()));
2979}
2980
57c1a579 2981//______________________________________________________________________________________________
339adafa 2982Bool_t AliShuttle::SendMail(EMailTarget target, Int_t system)
57c1a579 2983{
9827400b 2984 //
2985 // sends a mail to the subdetector expert in case of preprocessor error
2986 //
2987
2988 if (fTestMode != kNone)
2989 return kTRUE;
fb40ead4 2990
926cbc0e 2991 if (!fConfig->SendMail())
2992 return kTRUE;
57c1a579 2993
339adafa 2994 if (target == kDCSEMail || target == kFXSEMail) {
2995 if (!fFirstProcessing)
2996 return kTRUE;
36c99a6a 2997 }
2998
fbc112e3 2999 void* dir = gSystem->OpenDirectory(GetShuttleLogDir());
3000 if (dir == NULL)
3001 {
3002 if (gSystem->mkdir(GetShuttleLogDir(), kTRUE))
3003 {
3004 Log("SHUTTLE", Form("SendMail - Can't open directory <%s>", GetShuttleLogDir()));
3005 return kFALSE;
3006 }
3007
3008 } else {
3009 gSystem->FreeDirectory(dir);
3010 }
3011
339adafa 3012 // det experts in to
3013 TString to="";
3014 TIter *iterExperts;
3015 if (target == kDCSEMail) {
3016 iterExperts = new TIter(fConfig->GetAdmins(AliShuttleConfig::kAmanda));
3017 }
3018 else if (target == kFXSEMail) {
3019 iterExperts = new TIter(fConfig->GetAdmins(system));
3020 }
3021 else {
3022 iterExperts = new TIter(fConfig->GetResponsibles(fCurrentDetector));
3023 }
3024 TObjString *anExpert=0;
3025 while ((anExpert = (TObjString*) iterExperts->Next()))
4508ab8b 3026 {
339adafa 3027 to += Form("%s,", anExpert->GetName());
3028 }
3029 delete iterExperts;
3030 if (to.Length() > 0)
3031 to.Remove(to.Length()-1);
3032 AliDebug(2, Form("to: %s",to.Data()));
4508ab8b 3033
339adafa 3034 if (to.IsNull()) {
3035 Log("SHUTTLE", Form("List of %d responsibles not set!", (Int_t) target));
3036 return kFALSE;
3037 }
3038
3039 // SHUTTLE responsibles in cc
fb40ead4 3040 TString cc="";
3041 TIter iterAdmins(fConfig->GetAdmins(AliShuttleConfig::kGlobal));
3042 TObjString *anAdmin=0;
3043 while ((anAdmin = (TObjString*) iterAdmins.Next()))
3044 {
3045 cc += Form("%s,", anAdmin->GetName());
3046 }
3047 if (cc.Length() > 0)
c19963db 3048 cc.Remove(cc.Length()-1);
fb40ead4 3049 AliDebug(2, Form("cc: %s",to.Data()));
57c1a579 3050
339adafa 3051 // mail body
675f64cd 3052 TString bodyFileName;
3053 bodyFileName.Form("%s/mail.body", GetShuttleLogDir());
3054 gSystem->ExpandPathName(bodyFileName);
3055
3056 ofstream mailBody;
3057 mailBody.open(bodyFileName, ofstream::out);
3058
3059 if (!mailBody.is_open())
3060 {
339adafa 3061 Log("SHUTTLE", Form("Could not open mail body file %s", bodyFileName.Data()));
675f64cd 3062 return kFALSE;
3063 }
3064
675f64cd 3065
339adafa 3066 TString subject;
3067 TString body;
675f64cd 3068
339adafa 3069 if (target == kDCSEMail){
3070 subject = Form("Retrieval of data points for %s FAILED in run %d !",
3071 fCurrentDetector.Data(), GetCurrentRun());
3072 AliDebug(2, Form("subject: %s", subject.Data()));
3073
3074 body = Form("Dear DCS experts, \n\n");
3075 body += Form("SHUTTLE couldn\'t retrieve the data points for detector %s "
3076 "in run %d!!\n\n", fCurrentDetector.Data(), GetCurrentRun());
fb40ead4 3077 }
339adafa 3078 else if (target == kFXSEMail){
3079 subject = Form("FXS communication for %s FAILED in run %d !",
675f64cd 3080 fCurrentDetector.Data(), GetCurrentRun());
339adafa 3081 AliDebug(2, Form("subject: %s", subject.Data()));
3082 TString sys;
3083 if (system == kDAQ) sys="DAQ";
3084 else if (system == kDCS) sys="DCS";
3085 else if (system == kHLT) sys="HLT";
3086 else return kFALSE;
9c0e2d64 3087 body = Form("Dear %s FXS experts, \n\n",sys.Data());
339adafa 3088 body += Form("SHUTTLE couldn\'t retrieve data from the FXS for detector %s "
9c0e2d64 3089 "in run %d!!\n\n", fCurrentDetector.Data(), GetCurrentRun());
339adafa 3090 }
3091 else {
3092 subject = Form("%s Shuttle preprocessor FAILED in run %d (run type = %s)!",
3093 fCurrentDetector.Data(), GetCurrentRun(), GetRunType());
3094 AliDebug(2, Form("subject: %s", subject.Data()));
3095
3096 body = Form("Dear %s expert(s), \n\n", fCurrentDetector.Data());
3097 body += Form("SHUTTLE just detected that your preprocessor "
3098 "failed processing run %d (run type = %s)!!\n\n",
3099 GetCurrentRun(), GetRunType());
3100 }
675f64cd 3101
7d4cf768 3102 body += Form("Please check %s status on the SHUTTLE monitoring page: \n\n",
3103 fCurrentDetector.Data());
b0e53b15 3104 if (fConfig->GetRunMode() == AliShuttleConfig::kTest)
3105 {
3106 body += Form("\thttp://pcalimonitor.cern.ch:8889/shuttle.jsp?time=168 \n\n");
3107 } else {
339adafa 3108 body += Form("\thttp://pcalimonitor.cern.ch/shuttle.jsp?instance=PROD&time=168 \n\n");
b0e53b15 3109 }
339adafa 3110
3111
7d4cf768 3112 TString logFolder = &