Changing printout in case trigger configuration/time paramters are not found in
[u/mrichter/AliRoot.git] / SHUTTLE / AliShuttle.cxx
CommitLineData
73abe331 1/**************************************************************************
2 * Copyright(c) 1998-1999, ALICE Experiment at CERN, All rights reserved. *
3 * *
4 * Author: The ALICE Off-line Project. *
5 * Contributors are mentioned in the code where appropriate. *
6 * *
7 * Permission to use, copy, modify and distribute this software and its *
8 * documentation strictly for non-commercial purposes is hereby granted *
9 * without fee, provided that the above copyright notice appears in all *
10 * copies and that both the copyright notice and this permission notice *
11 * appear in the supporting documentation. The authors make no claims *
12 * about the suitability of this software for any purpose. It is *
13 * provided "as is" without express or implied warranty. *
14 **************************************************************************/
15
ff3781ad 16/* $Id$ */
73abe331 17
18//
19// This class is the main manager for AliShuttle.
20// It organizes the data retrieval from DCS and call the
b948db8d 21// interface methods of AliPreprocessor.
73abe331 22// For every detector in AliShuttleConfgi (see AliShuttleConfig),
23// data for its set of aliases is retrieved. If there is registered
b948db8d 24// AliPreprocessor for this detector then it will be used
25// accroding to the schema (see AliPreprocessor).
26// If there isn't registered AliPreprocessor than the retrieved
73abe331 27// data is stored automatically to the undelying AliCDBStorage.
28// For detSpec is used the alias name.
29//
30
31#include "AliShuttle.h"
32
33#include "AliCDBManager.h"
34#include "AliCDBStorage.h"
35#include "AliCDBId.h"
84090f85 36#include "AliCDBRunRange.h"
37#include "AliCDBPath.h"
5164a766 38#include "AliCDBEntry.h"
73abe331 39#include "AliShuttleConfig.h"
eba76848 40#include "DCSClient/AliDCSClient.h"
73abe331 41#include "AliLog.h"
b948db8d 42#include "AliPreprocessor.h"
5164a766 43#include "AliShuttleStatus.h"
2bb7b766 44#include "AliShuttleLogbookEntry.h"
73abe331 45
57f50b3c 46#include <TSystem.h>
58bc3020 47#include <TObject.h>
b948db8d 48#include <TString.h>
57f50b3c 49#include <TTimeStamp.h>
73abe331 50#include <TObjString.h>
57f50b3c 51#include <TSQLServer.h>
52#include <TSQLResult.h>
53#include <TSQLRow.h>
cb343cfd 54#include <TMutex.h>
9827400b 55#include <TSystemDirectory.h>
56#include <TSystemFile.h>
a986b218 57#include <TFile.h>
9827400b 58#include <TGrid.h>
59#include <TGridResult.h>
503ff24f 60#include <TMap.h>
73abe331 61
e7f62f16 62#include <TMonaLisaWriter.h>
63
5164a766 64#include <fstream>
65
cb343cfd 66#include <sys/types.h>
67#include <sys/wait.h>
68
1bd183ba 69#include <signal.h>
70
73abe331 71ClassImp(AliShuttle)
72
b948db8d 73//______________________________________________________________________________________________
74AliShuttle::AliShuttle(const AliShuttleConfig* config,
75 UInt_t timeout, Int_t retries):
4f0ab988 76fConfig(config),
77fTimeout(timeout), fRetries(retries),
78fPreprocessorMap(),
2bb7b766 79fLogbookEntry(0),
eba76848 80fCurrentDetector(),
926cbc0e 81fFirstProcessing(0),
339adafa 82fFXSError(-1),
85a80aa9 83fStatusEntry(0),
cb343cfd 84fMonitoringMutex(0),
eba76848 85fLastActionTime(0),
e7f62f16 86fLastAction(),
9827400b 87fMonaLisa(0),
88fTestMode(kNone),
ffa29e93 89fReadTestMode(kFALSE),
90fOutputRedirected(kFALSE)
73abe331 91{
92 //
93 // config: AliShuttleConfig used
73abe331 94 // timeout: timeout used for AliDCSClient connection
95 // retries: the number of retries in case of connection error.
96 //
97
57f50b3c 98 if (!fConfig->IsValid()) AliFatal("********** !!!!! Invalid configuration !!!!! **********");
be48e3ea 99 for(int iSys=0;iSys<4;iSys++) {
57f50b3c 100 fServer[iSys]=0;
be48e3ea 101 if (iSys < 3)
2c15234c 102 fFXSlist[iSys].SetOwner(kTRUE);
57f50b3c 103 }
2bb7b766 104 fPreprocessorMap.SetOwner(kTRUE);
be48e3ea 105
106 for (UInt_t iDet=0; iDet<NDetectors(); iDet++)
107 fFirstUnprocessed[iDet] = kFALSE;
108
cb343cfd 109 fMonitoringMutex = new TMutex();
58bc3020 110}
111
b948db8d 112//______________________________________________________________________________________________
57f50b3c 113AliShuttle::~AliShuttle()
58bc3020 114{
9827400b 115 //
116 // destructor
117 //
58bc3020 118
b948db8d 119 fPreprocessorMap.DeleteAll();
be48e3ea 120 for(int iSys=0;iSys<4;iSys++)
57f50b3c 121 if(fServer[iSys]) {
122 fServer[iSys]->Close();
123 delete fServer[iSys];
eba76848 124 fServer[iSys] = 0;
57f50b3c 125 }
2bb7b766 126
127 if (fStatusEntry){
128 delete fStatusEntry;
129 fStatusEntry = 0;
130 }
cb343cfd 131
132 if (fMonitoringMutex)
133 {
134 delete fMonitoringMutex;
135 fMonitoringMutex = 0;
136 }
73abe331 137}
138
b948db8d 139//______________________________________________________________________________________________
57f50b3c 140void AliShuttle::RegisterPreprocessor(AliPreprocessor* preprocessor)
58bc3020 141{
73abe331 142 //
b948db8d 143 // Registers new AliPreprocessor.
73abe331 144 // It uses GetName() for indentificator of the pre processor.
145 // The pre processor is registered it there isn't any other
146 // with the same identificator (GetName()).
147 //
148
eba76848 149 const char* detName = preprocessor->GetName();
150 if(GetDetPos(detName) < 0)
151 AliFatal(Form("********** !!!!! Invalid detector name: %s !!!!! **********", detName));
152
153 if (fPreprocessorMap.GetValue(detName)) {
154 AliWarning(Form("AliPreprocessor %s is already registered!", detName));
73abe331 155 return;
156 }
157
eba76848 158 fPreprocessorMap.Add(new TObjString(detName), preprocessor);
73abe331 159}
b948db8d 160//______________________________________________________________________________________________
3301427a 161Bool_t AliShuttle::Store(const AliCDBPath& path, TObject* object,
84090f85 162 AliCDBMetaData* metaData, Int_t validityStart, Bool_t validityInfinite)
73abe331 163{
9827400b 164 // Stores a CDB object in the storage for offline reconstruction. Objects that are not needed for
165 // offline reconstruction, but should be stored anyway (e.g. for debugging) should NOT be stored
166 // using this function. Use StoreReferenceData instead!
167 // It calls StoreLocally function which temporarily stores the data locally; when the preprocessor
168 // finishes the data are transferred to the main storage (Grid).
b948db8d 169
3301427a 170 return StoreLocally(fgkLocalCDB, path, object, metaData, validityStart, validityInfinite);
84090f85 171}
172
173//______________________________________________________________________________________________
3301427a 174Bool_t AliShuttle::StoreReferenceData(const AliCDBPath& path, TObject* object, AliCDBMetaData* metaData)
84090f85 175{
9827400b 176 // Stores a CDB object in the storage for reference data. This objects will not be available during
177 // offline reconstrunction. Use this function for reference data only!
178 // It calls StoreLocally function which temporarily stores the data locally; when the preprocessor
179 // finishes the data are transferred to the main storage (Grid).
85a80aa9 180
3301427a 181 return StoreLocally(fgkLocalRefStorage, path, object, metaData);
85a80aa9 182}
183
184//______________________________________________________________________________________________
3301427a 185Bool_t AliShuttle::StoreLocally(const TString& localUri,
85a80aa9 186 const AliCDBPath& path, TObject* object, AliCDBMetaData* metaData,
187 Int_t validityStart, Bool_t validityInfinite)
188{
9827400b 189 // Store object temporarily in local storage. Parameters are passed by Store and StoreReferenceData functions.
190 // when the preprocessor finishes the data are transferred to the main storage (Grid).
191 // The parameters are:
192 // 1) Uri of the backup storage (Local)
193 // 2) the object's path.
194 // 3) the object to be stored
195 // 4) the metaData to be associated with the object
196 // 5) the validity start run number w.r.t. the current run,
197 // if the data is valid only for this run leave the default 0
198 // 6) specifies if the calibration data is valid for infinity (this means until updated),
199 // typical for calibration runs, the default is kFALSE
200 //
201 // returns 0 if fail, 1 otherwise
84090f85 202
924fb6e9 203
9827400b 204 if (fTestMode & kErrorStorage)
205 {
206 Log(fCurrentDetector, "StoreLocally - In TESTMODE - Simulating error while storing locally");
207 return kFALSE;
208 }
209
3301427a 210 const char* cdbType = (localUri == fgkLocalCDB) ? "CDB" : "Reference";
2bb7b766 211
85a80aa9 212 Int_t firstRun = GetCurrentRun() - validityStart;
84090f85 213 if(firstRun < 0) {
9827400b 214 AliWarning("First valid run happens to be less than 0! Setting it to 0.");
84090f85 215 firstRun=0;
216 }
217
218 Int_t lastRun = -1;
219 if(validityInfinite) {
220 lastRun = AliCDBRunRange::Infinity();
221 } else {
222 lastRun = GetCurrentRun();
223 }
224
3301427a 225 // Version is set to current run, it will be used later to transfer data to Grid
226 AliCDBId id(path, firstRun, lastRun, GetCurrentRun(), -1);
2bb7b766 227
228 if(! dynamic_cast<TObjString*> (metaData->GetProperty("RunUsed(TObjString)"))){
229 TObjString runUsed = Form("%d", GetCurrentRun());
9e080f92 230 metaData->SetProperty("RunUsed(TObjString)", runUsed.Clone());
2bb7b766 231 }
84090f85 232
3301427a 233 Bool_t result = kFALSE;
84090f85 234
3301427a 235 if (!(AliCDBManager::Instance()->GetStorage(localUri))) {
236 Log("SHUTTLE", Form("StoreLocally - Cannot activate local %s storage", cdbType));
84090f85 237 } else {
924fb6e9 238 Int_t logLevel = AliLog::GetGlobalLogLevel();
239 AliLog::SetGlobalLogLevel(AliLog::kError);
3301427a 240 result = AliCDBManager::Instance()->GetStorage(localUri)
84090f85 241 ->Put(object, id, metaData);
924fb6e9 242 AliLog::SetGlobalLogLevel((AliLog::EType_t)logLevel);
84090f85 243 }
244
245 if(!result) {
246
9827400b 247 Log(fCurrentDetector, Form("StoreLocally - Can't store object <%s>!", id.ToString().Data()));
3301427a 248 }
2bb7b766 249
924fb6e9 250
3301427a 251 return result;
252}
84090f85 253
3301427a 254//______________________________________________________________________________________________
255Bool_t AliShuttle::StoreOCDB()
256{
9827400b 257 //
258 // Called when preprocessor ends successfully or when previous storage attempt failed (kStoreError status)
259 // Calls underlying StoreOCDB(const char*) function twice, for OCDB and Reference storage.
260 // Then calls StoreRefFilesToGrid to store reference files.
261 //
262
5e993b6f 263 UpdateShuttleStatus(AliShuttleStatus::kStoreStarted);
56d3e9e4 264
9827400b 265 if (fTestMode & kErrorGrid)
266 {
267 Log("SHUTTLE", "StoreOCDB - In TESTMODE - Simulating error while storing in the Grid");
268 Log(fCurrentDetector, "StoreOCDB - In TESTMODE - Simulating error while storing in the Grid");
269 return kFALSE;
270 }
271
c88ad5db 272 Log("SHUTTLE","StoreOCDB - Storing OCDB data ...");
5e993b6f 273 Int_t resultCDB = StoreOCDB(fgkMainCDB);
86aa42c3 274
c88ad5db 275 Log("SHUTTLE","StoreOCDB - Storing reference data ...");
5e993b6f 276 Int_t resultRef = StoreOCDB(fgkMainRefStorage);
9827400b 277
c88ad5db 278 Log("SHUTTLE","StoreOCDB - Storing reference files ...");
279 Bool_t resultRefFiles = CopyFilesToGrid("reference");
280
281 Bool_t resultMetadata = kTRUE;
282 if(fCurrentDetector == "GRP")
283 {
7802e884 284 Log("SHUTTLE","StoreOCDB - Storing Run Metadata file ...");
c88ad5db 285 resultMetadata = CopyFilesToGrid("metadata");
286 }
9827400b 287
5e993b6f 288 Int_t storeResult = 0;
289
290 if (resultCDB < 0 || resultRef < 0 || resultRefFiles == kFALSE || resultMetadata == kFALSE)
291 storeResult = -1;
292 else if (resultCDB > 0 || resultRef > 0)
293 storeResult = 1;
294
295 if (storeResult < 0)
296 {
297 Log("SHUTTLE",
298 Form("\t\t\t****** run %d - %s: STORAGE ERROR ******",
299 GetCurrentRun(), fCurrentDetector.Data()));
300 UpdateShuttleStatus(AliShuttleStatus::kStoreError);
301 }
302 else if (storeResult > 0)
303 {
304 Log("SHUTTLE",
305 Form("\t\t\t****** run %d - %s: STORAGE DELAYED ******",
306 GetCurrentRun(), fCurrentDetector.Data()));
307 UpdateShuttleStatus(AliShuttleStatus::kStoreDelayed);
308 }
309 else if (storeResult == 0)
310 {
311 Log("SHUTTLE",
312 Form("\t\t\t****** run %d - %s: DONE ******",
313 GetCurrentRun(), fCurrentDetector.Data()));
314 UpdateShuttleStatus(AliShuttleStatus::kDone);
315 UpdateShuttleLogbook(fCurrentDetector, "DONE");
316 }
317
318 return (storeResult == 0);
3301427a 319}
320
321//______________________________________________________________________________________________
5e993b6f 322Int_t AliShuttle::StoreOCDB(const TString& gridURI)
3301427a 323{
324 //
325 // Called by StoreOCDB(), performs actual storage to the main OCDB and reference storages (Grid)
326 //
5e993b6f 327 // Return code:
328 // -2 initialization error
329 // -1 storage error
330 // 0 success
331 // 1 storage delayed (e.g. previous unprocessed runs)
332 //
3301427a 333
334 TObjArray* gridIds=0;
335
336 Bool_t result = kTRUE;
5e993b6f 337 Bool_t delayed = kFALSE;
3301427a 338
339 const char* type = 0;
340 TString localURI;
341 if(gridURI == fgkMainCDB) {
342 type = "OCDB";
343 localURI = fgkLocalCDB;
344 } else if(gridURI == fgkMainRefStorage) {
345 type = "reference";
346 localURI = fgkLocalRefStorage;
347 } else {
348 AliError(Form("Invalid storage URI: %s", gridURI.Data()));
5e993b6f 349 return -2;
3301427a 350 }
351
352 AliCDBManager* man = AliCDBManager::Instance();
353
354 AliCDBStorage *gridSto = man->GetStorage(gridURI);
355 if(!gridSto) {
356 Log("SHUTTLE",
357 Form("StoreOCDB - cannot activate main %s storage", type));
5e993b6f 358 return -2;
3301427a 359 }
360
361 gridIds = gridSto->GetQueryCDBList();
362
363 // get objects previously stored in local CDB
364 AliCDBStorage *localSto = man->GetStorage(localURI);
365 if(!localSto) {
366 Log("SHUTTLE",
367 Form("StoreOCDB - cannot activate local %s storage", type));
5e993b6f 368 return -2;
3301427a 369 }
370 AliCDBPath aPath(GetOfflineDetName(fCurrentDetector.Data()),"*","*");
371 // Local objects were stored with current run as Grid version!
372 TList* localEntries = localSto->GetAll(aPath.GetPath(), GetCurrentRun(), GetCurrentRun());
373 localEntries->SetOwner(1);
374
375 // loop on local stored objects
376 TIter localIter(localEntries);
377 AliCDBEntry *aLocEntry = 0;
378 while((aLocEntry = dynamic_cast<AliCDBEntry*> (localIter.Next()))){
379 aLocEntry->SetOwner(1);
380 AliCDBId aLocId = aLocEntry->GetId();
381 aLocEntry->SetVersion(-1);
382 aLocEntry->SetSubVersion(-1);
6c0c9948 383
384 Log(fCurrentDetector.Data(), Form("Attempting to store %s", aLocId.ToString().Data()));
3301427a 385
386 // If local object is valid up to infinity we store it only if it is
387 // the first unprocessed run!
388 if (aLocId.GetLastRun() == AliCDBRunRange::Infinity() &&
389 !fFirstUnprocessed[GetDetPos(fCurrentDetector)])
390 {
391 Log("SHUTTLE", Form("StoreOCDB - %s: object %s has validity infinite but "
392 "there are previous unprocessed runs!",
393 fCurrentDetector.Data(), aLocId.GetPath().Data()));
5e993b6f 394 Log(fCurrentDetector.Data(), Form("StoreOCDB - %s: object %s has validity infinite but "
395 "there are previous unprocessed runs!",
396 fCurrentDetector.Data(), aLocId.GetPath().Data()));
397 delayed = kTRUE;
3301427a 398 continue;
399 }
400
401 // loop on Grid valid Id's
402 Bool_t store = kTRUE;
403 TIter gridIter(gridIds);
404 AliCDBId* aGridId = 0;
6c0c9948 405 while ((aGridId = dynamic_cast<AliCDBId*> (gridIter.Next()))) {
406 if (aGridId->GetPath() != aLocId.GetPath())
407 continue;
3301427a 408 // skip all objects valid up to infinity
6c0c9948 409 if (aGridId->GetLastRun() == AliCDBRunRange::Infinity())
410 continue;
411
3301427a 412 // if we get here, it means there's already some more recent object stored on Grid!
6c0c9948 413 Log(fCurrentDetector.Data(),
414 Form("StoreOCDB - A more recent object already exists in %s storage: <%s>",
415 type, aGridId->ToString().Data()));
416
3301427a 417 store = kFALSE;
418 break;
419 }
420
6c0c9948 421 Bool_t storeOk = kFALSE;
422 if (store)
423 {
424 Log(fCurrentDetector.Data(), Form("Prechecks succeeded. Ready to store %s", aLocId.ToString().Data()));
425 storeOk = gridSto->Put(aLocEntry);
426 if (storeOk) {
3301427a 427 Log("SHUTTLE",
6c0c9948 428 Form("StoreOCDB - Object <%s> successfully put into %s storage",
429 aLocId.ToString().Data(), type));
2d9019b4 430 Log(fCurrentDetector.Data(),
431 Form("StoreOCDB - Object <%s> successfully put into %s storage",
6c0c9948 432 aLocId.ToString().Data(), type));
433 } else {
434 Log("SHUTTLE",
435 Form("StoreOCDB - Grid %s storage of object <%s> failed",
436 type, aLocId.ToString().Data()));
437 Log(fCurrentDetector.Data(),
438 Form("StoreOCDB - Grid %s storage of object <%s> failed",
439 type, aLocId.ToString().Data()));
440 result = kFALSE;
3301427a 441 }
6c0c9948 442 }
443
444 if (!store || storeOk) {
445 // removing local file...
3301427a 446 TString filename;
447 localSto->IdToFilename(aLocId, filename);
c88ad5db 448 Log("SHUTTLE", Form("StoreOCDB - Removing local file %s", filename.Data()));
3301427a 449 RemoveFile(filename.Data());
b948db8d 450 }
451 }
3301427a 452 localEntries->Clear();
2bb7b766 453
5e993b6f 454 Int_t returnCode = 0;
455
456 if (result == kFALSE)
457 returnCode = -1;
458 else if (delayed != kFALSE)
459 returnCode = 1;
460
461 Log("SHUTTLE", Form("StoreOCDB - Returning with %d (result = %d, delayed = %d)", returnCode, result, delayed));
462 Log(fCurrentDetector.Data(), Form("StoreOCDB - Returning with %d (result = %d, delayed = %d)", returnCode, result, delayed));
463
464 return returnCode;
3301427a 465}
466
467//______________________________________________________________________________________________
546242fb 468Bool_t AliShuttle::CleanReferenceStorage(const char* detector)
469{
2d9019b4 470 // clears the directory used to store reference files of a given subdetector
546242fb 471
472 AliCDBManager* man = AliCDBManager::Instance();
473 AliCDBStorage* sto = man->GetStorage(fgkLocalRefStorage);
2d9019b4 474 TString localBaseFolder = sto->GetBaseFolder();
475
476 TString targetDir = GetRefFilePrefix(localBaseFolder.Data(), detector);
477
d524ade6 478 Log("SHUTTLE", Form("CleanReferenceStorage - Cleaning %s", targetDir.Data()));
2d9019b4 479
480 TString begin;
481 begin.Form("%d_", GetCurrentRun());
482
483 TSystemDirectory* baseDir = new TSystemDirectory("/", targetDir);
484 if (!baseDir)
485 return kTRUE;
486
487 TList* dirList = baseDir->GetListOfFiles();
488 delete baseDir;
489
490 if (!dirList) return kTRUE;
491
492 if (dirList->GetEntries() < 3)
493 {
494 delete dirList;
495 return kTRUE;
496 }
497
498 Int_t nDirs = 0, nDel = 0;
499 TIter dirIter(dirList);
500 TSystemFile* entry = 0;
546242fb 501
2d9019b4 502 Bool_t success = kTRUE;
546242fb 503
2d9019b4 504 while ((entry = dynamic_cast<TSystemFile*> (dirIter.Next())))
505 {
506 if (entry->IsDirectory())
507 continue;
508
509 TString fileName(entry->GetName());
510 if (!fileName.BeginsWith(begin))
511 continue;
512
513 nDirs++;
514
515 // delete file
516 Int_t result = gSystem->Unlink(fileName.Data());
517
518 if (result)
519 {
d524ade6 520 Log("SHUTTLE", Form("CleanReferenceStorage - Could not delete file %s!", fileName.Data()));
2d9019b4 521 success = kFALSE;
522 } else {
523 nDel++;
524 }
525 }
526
527 if(nDirs > 0)
528 Log("SHUTTLE", Form("CleanReferenceStorage - %d (over %d) reference files in folder %s were deleted.",
529 nDel, nDirs, targetDir.Data()));
530
531
532 delete dirList;
533 return success;
534
535
536
537
538
546242fb 539
540 Int_t result = gSystem->GetPathInfo(targetDir, 0, (Long64_t*) 0, 0, 0);
541 if (result == 0)
542 {
543 // delete directory
d524ade6 544 result = gSystem->Exec(Form("rm -rf %s", targetDir.Data()));
546242fb 545 if (result != 0)
546 {
d524ade6 547 Log("SHUTTLE", Form("CleanReferenceStorage - Could not clean directory %s", targetDir.Data()));
546242fb 548 return kFALSE;
549 }
550 }
551
552 result = gSystem->mkdir(targetDir, kTRUE);
553 if (result != 0)
554 {
c88ad5db 555 Log("SHUTTLE", Form("CleanReferenceStorage - Error creating base directory %s", targetDir.Data()));
546242fb 556 return kFALSE;
557 }
558
559 return kTRUE;
560}
561
562//______________________________________________________________________________________________
9827400b 563Bool_t AliShuttle::StoreReferenceFile(const char* detector, const char* localFile, const char* gridFileName)
564{
565 //
3c2a21c8 566 // Stores reference file directly (without opening it). This function stores the file locally.
9827400b 567 //
3c2a21c8 568 // The file is stored under the following location:
569 // <base folder of local reference storage>/<DET>/<RUN#>_<gridFileName>
570 // where <gridFileName> is the second parameter given to the function
571 //
9827400b 572
573 if (fTestMode & kErrorStorage)
574 {
575 Log(fCurrentDetector, "StoreReferenceFile - In TESTMODE - Simulating error while storing locally");
576 return kFALSE;
577 }
578
579 AliCDBManager* man = AliCDBManager::Instance();
580 AliCDBStorage* sto = man->GetStorage(fgkLocalRefStorage);
581
582 TString localBaseFolder = sto->GetBaseFolder();
583
d524ade6 584 TString target = GetRefFilePrefix(localBaseFolder.Data(), detector);
585 target.Append(Form("/%d_%s", GetCurrentRun(), gridFileName));
9827400b 586
d524ade6 587 return CopyFileLocally(localFile, target);
c88ad5db 588}
589
590//______________________________________________________________________________________________
591Bool_t AliShuttle::StoreRunMetadataFile(const char* localFile, const char* gridFileName)
592{
593 //
594 // Stores Run metadata file to the Grid, in the run folder
595 //
596 // Only GRP can call this function.
597
598 if (fTestMode & kErrorStorage)
599 {
600 Log(fCurrentDetector, "StoreRunMetaDataFile - In TESTMODE - Simulating error while storing locally");
601 return kFALSE;
602 }
603
604 AliCDBManager* man = AliCDBManager::Instance();
605 AliCDBStorage* sto = man->GetStorage(fgkLocalRefStorage);
606
607 TString localBaseFolder = sto->GetBaseFolder();
608
609 // Build Run level folder
955d2abc 610 // folder = /alice/data/year/lhcPeriod/runNb/raw
c88ad5db 611
c88ad5db 612
675f64cd 613 TString lhcPeriod = GetLHCPeriod();
c88ad5db 614 if (lhcPeriod.Length() == 0)
615 {
616 Log("SHUTTLE","StoreRunMetaDataFile - LHCPeriod not found in logbook!");
617 return 0;
618 }
a1c5723b 619
620 // TODO partitions with one detector only write data into LHCperiod_DET
2694f253 621 TString partition = GetRunParameter("detector");
a1c5723b 622
623 if (partition.Length() > 0 && partition != "ALICE")
624 {
625 lhcPeriod.Append(Form("_%s", partition.Data()));
626 Log(fCurrentDetector, Form("Run data tags merged file will be written in %s",
627 lhcPeriod.Data()));
628 }
8884d71e 629
b8cf5e0f 630 TString target = Form("%s/GRP/RunMetadata%s%d/%s/%09d/raw/%s",
631 localBaseFolder.Data(), fConfig->GetAlienPath(), GetCurrentYear(),
d524ade6 632 lhcPeriod.Data(), GetCurrentRun(), gridFileName);
c88ad5db 633
d524ade6 634 return CopyFileLocally(localFile, target);
c88ad5db 635}
636
637//______________________________________________________________________________________________
d524ade6 638Bool_t AliShuttle::CopyFileLocally(const char* localFile, const TString& target)
c88ad5db 639{
640 //
641 // Stores file locally. Called by StoreReferenceFile and StoreRunMetadataFile
d524ade6 642 // Files are temporarily stored in the local reference storage. When the preprocessor
643 // finishes, the Shuttle calls CopyFilesToGrid to transfer the files to AliEn
644 // (in reference or run level folders)
c88ad5db 645 //
646
d524ade6 647 TString targetDir(target(0, target.Last('/')));
648
649 //try to open base dir folder, if it does not exist
2d9019b4 650 void* dir = gSystem->OpenDirectory(targetDir.Data());
651 if (dir == NULL) {
652 if (gSystem->mkdir(targetDir.Data(), kTRUE)) {
a1c5723b 653 Log("SHUTTLE", Form("CopyFileLocally - Can't open directory <%s>", targetDir.Data()));
2d9019b4 654 return kFALSE;
655 }
656
657 } else {
658 gSystem->FreeDirectory(dir);
659 }
9827400b 660
7d43a416 661 Int_t result = 0;
662
663 result = gSystem->GetPathInfo(localFile, 0, (Long64_t*) 0, 0, 0);
9827400b 664 if (result)
665 {
a1c5723b 666 Log("SHUTTLE", Form("CopyFileLocally - %s does not exist", localFile));
546242fb 667 return kFALSE;
9827400b 668 }
546242fb 669
7d43a416 670 result = gSystem->GetPathInfo(target, 0, (Long64_t*) 0, 0, 0);
671 if (!result)
672 {
a1c5723b 673 Log("SHUTTLE", Form("CopyFileLocally - target file %s already exist, removing...", target.Data()));
7d43a416 674 if (gSystem->Unlink(target.Data()))
675 {
a1c5723b 676 Log("SHUTTLE", Form("CopyFileLocally - Could not remove existing target file %s!", target.Data()));
7d43a416 677 return kFALSE;
678 }
679 }
680
9827400b 681 result = gSystem->CopyFile(localFile, target);
682
683 if (result == 0)
684 {
a1c5723b 685 Log("SHUTTLE", Form("CopyFileLocally - File %s stored locally to %s", localFile, target.Data()));
9827400b 686 return kTRUE;
687 }
688 else
689 {
a1c5723b 690 Log("SHUTTLE", Form("CopyFileLocally - Could not store file %s to %s! Error code = %d",
546242fb 691 localFile, target.Data(), result));
9827400b 692 return kFALSE;
693 }
c88ad5db 694
695
696
9827400b 697}
698
699//______________________________________________________________________________________________
c88ad5db 700Bool_t AliShuttle::CopyFilesToGrid(const char* type)
9827400b 701{
702 //
c88ad5db 703 // Transfers local files to the Grid. Local files can be reference files
704 // or run metadata file (from GRP only).
9827400b 705 //
c88ad5db 706 // According to the type (ref, metadata) the files are stored under the following location:
707 // ref --> <base folder of reference storage>/<DET>/<RUN#>_<gridFileName>
708 // metadata --> <run data folder>/<MetadataFileName>
86aa42c3 709 //
c88ad5db 710
9827400b 711 AliCDBManager* man = AliCDBManager::Instance();
712 AliCDBStorage* sto = man->GetStorage(fgkLocalRefStorage);
713 if (!sto)
714 return kFALSE;
715 TString localBaseFolder = sto->GetBaseFolder();
9827400b 716
c88ad5db 717 TString dir;
718 TString alienDir;
9827400b 719 TString begin;
9827400b 720
c88ad5db 721 if (strcmp(type, "reference") == 0)
722 {
723 dir = GetRefFilePrefix(localBaseFolder.Data(), fCurrentDetector.Data());
724 AliCDBStorage* gridSto = man->GetStorage(fgkMainRefStorage);
725 if (!gridSto)
726 return kFALSE;
727 TString gridBaseFolder = gridSto->GetBaseFolder();
728 alienDir = GetRefFilePrefix(gridBaseFolder.Data(), fCurrentDetector.Data());
729 begin = Form("%d_", GetCurrentRun());
730 }
731 else if (strcmp(type, "metadata") == 0)
732 {
c88ad5db 733
675f64cd 734 TString lhcPeriod = GetLHCPeriod();
c88ad5db 735
736 if (lhcPeriod.Length() == 0)
737 {
738 Log("SHUTTLE","CopyFilesToGrid - LHCPeriod not found in logbook!");
739 return 0;
740 }
741
a1c5723b 742 // TODO partitions with one detector only write data into LHCperiod_DET
2694f253 743 TString partition = GetRunParameter("detector");
a1c5723b 744
745 if (partition.Length() > 0 && partition != "ALICE")
746 {
747 lhcPeriod.Append(Form("_%s", partition.Data()));
748 }
749
b8cf5e0f 750 dir = Form("%s/GRP/RunMetadata%s%d/%s/%09d/raw",
751 localBaseFolder.Data(), fConfig->GetAlienPath(), GetCurrentYear(),
c88ad5db 752 lhcPeriod.Data(), GetCurrentRun());
b8cf5e0f 753 alienDir = dir(dir.Index(fConfig->GetAlienPath()), dir.Length());
675f64cd 754
c88ad5db 755 begin = "";
756 }
757 else
758 {
759 Log("SHUTTLE", "CopyFilesToGrid - Unexpected: type label must be reference or metadata!");
760 return kFALSE;
761 }
762
9827400b 763 TSystemDirectory* baseDir = new TSystemDirectory("/", dir);
3d8bc902 764 if (!baseDir)
765 return kTRUE;
766
2d9019b4 767 TList* dirList = baseDir->GetListOfFiles();
768 delete baseDir;
769
770 if (!dirList) return kTRUE;
771
772 if (dirList->GetEntries() < 3)
3d8bc902 773 {
2d9019b4 774 delete dirList;
9827400b 775 return kTRUE;
3d8bc902 776 }
2d9019b4 777
546242fb 778 if (!gGrid)
779 {
c88ad5db 780 Log("SHUTTLE", "CopyFilesToGrid - Connection to Grid failed: Cannot continue!");
2d9019b4 781 delete dirList;
546242fb 782 return kFALSE;
783 }
784
2d9019b4 785 Int_t nDirs = 0, nTransfer = 0;
786 TIter dirIter(dirList);
787 TSystemFile* entry = 0;
788
9827400b 789 Bool_t success = kTRUE;
3d8bc902 790 Bool_t first = kTRUE;
9827400b 791
2d9019b4 792 while ((entry = dynamic_cast<TSystemFile*> (dirIter.Next())))
793 {
9827400b 794 if (entry->IsDirectory())
795 continue;
796
797 TString fileName(entry->GetName());
798 if (!fileName.BeginsWith(begin))
799 continue;
800
2d9019b4 801 nDirs++;
802
3d8bc902 803 if (first)
804 {
805 first = kFALSE;
c88ad5db 806 // check that folder exists, otherwise create it
3d8bc902 807 TGridResult* result = gGrid->Ls(alienDir.Data(), "a");
808
809 if (!result)
2d9019b4 810 {
811 delete dirList;
3d8bc902 812 return kFALSE;
2d9019b4 813 }
3d8bc902 814
546242fb 815 if (!result->GetFileName(1)) // TODO: It looks like element 0 is always 0!!
3d8bc902 816 {
675f64cd 817 // TODO It does not work currently! Bug in TAliEn::Mkdir
818 // TODO Manually fixed in local root v5-16-00
c88ad5db 819 if (!gGrid->Mkdir(alienDir.Data(),"-p",0))
3d8bc902 820 {
c88ad5db 821 Log("SHUTTLE", Form("CopyFilesToGrid - Cannot create directory %s",
3d8bc902 822 alienDir.Data()));
2d9019b4 823 delete dirList;
3d8bc902 824 return kFALSE;
546242fb 825 } else {
c88ad5db 826 Log("SHUTTLE",Form("CopyFilesToGrid - Folder %s created", alienDir.Data()));
3d8bc902 827 }
828
546242fb 829 } else {
c88ad5db 830 Log("SHUTTLE",Form("CopyFilesToGrid - Folder %s found", alienDir.Data()));
3d8bc902 831 }
832 }
833
9827400b 834 TString fullLocalPath;
835 fullLocalPath.Form("%s/%s", dir.Data(), fileName.Data());
836
837 TString fullGridPath;
838 fullGridPath.Form("alien://%s/%s", alienDir.Data(), fileName.Data());
839
a986b218 840 Bool_t result = TFile::Cp(fullLocalPath, fullGridPath);
9827400b 841
842 if (result)
843 {
c88ad5db 844 Log("SHUTTLE", Form("CopyFilesToGrid - Copying local file %s to %s succeeded!",
845 fullLocalPath.Data(), fullGridPath.Data()));
9827400b 846 RemoveFile(fullLocalPath);
2d9019b4 847 nTransfer++;
9827400b 848 }
849 else
850 {
c88ad5db 851 Log("SHUTTLE", Form("CopyFilesToGrid - Copying local file %s to %s FAILED!",
852 fullLocalPath.Data(), fullGridPath.Data()));
9827400b 853 success = kFALSE;
854 }
855 }
2d9019b4 856
c88ad5db 857 Log("SHUTTLE", Form("CopyFilesToGrid - %d (over %d) files in folder %s copied to Grid.",
858 nTransfer, nDirs, dir.Data()));
2d9019b4 859
860
861 delete dirList;
9827400b 862 return success;
863}
864
865//______________________________________________________________________________________________
2d9019b4 866const char* AliShuttle::GetRefFilePrefix(const char* base, const char* detector)
867{
868 //
869 // Get folder name of reference files
870 //
871
872 TString offDetStr(GetOfflineDetName(detector));
3dd835d8 873 static TString dir;
2d9019b4 874 if (offDetStr == "ITS" || offDetStr == "MUON" || offDetStr == "PHOS")
875 {
876 dir.Form("%s/%s/%s", base, offDetStr.Data(), detector);
877 } else {
878 dir.Form("%s/%s", base, offDetStr.Data());
879 }
880
881 return dir.Data();
2d9019b4 882}
c88ad5db 883
2d9019b4 884//______________________________________________________________________________________________
3301427a 885void AliShuttle::CleanLocalStorage(const TString& uri)
886{
9827400b 887 //
888 // Called in case the preprocessor is declared failed. Remove remaining objects from the local storages.
889 //
3301427a 890
891 const char* type = 0;
892 if(uri == fgkLocalCDB) {
893 type = "OCDB";
894 } else if(uri == fgkLocalRefStorage) {
546242fb 895 type = "Reference";
3301427a 896 } else {
897 AliError(Form("Invalid storage URI: %s", uri.Data()));
898 return;
899 }
900
901 AliCDBManager* man = AliCDBManager::Instance();
b948db8d 902
3301427a 903 // open local storage
904 AliCDBStorage *localSto = man->GetStorage(uri);
905 if(!localSto) {
906 Log("SHUTTLE",
907 Form("CleanLocalStorage - cannot activate local %s storage", type));
908 return;
909 }
910
911 TString filename(Form("%s/%s/*/Run*_v%d_s*.root",
546242fb 912 localSto->GetBaseFolder().Data(), GetOfflineDetName(fCurrentDetector.Data()), GetCurrentRun()));
3301427a 913
c88ad5db 914 AliDebug(2, Form("filename = %s", filename.Data()));
3301427a 915
c88ad5db 916 Log("SHUTTLE", Form("Removing remaining local files for run %d and detector %s ...",
3301427a 917 GetCurrentRun(), fCurrentDetector.Data()));
918
919 RemoveFile(filename.Data());
920
921}
922
923//______________________________________________________________________________________________
924void AliShuttle::RemoveFile(const char* filename)
925{
9827400b 926 //
927 // removes local file
928 //
3301427a 929
930 TString command(Form("rm -f %s", filename));
931
932 Int_t result = gSystem->Exec(command.Data());
933 if(result != 0)
934 {
935 Log("SHUTTLE", Form("RemoveFile - %s: Cannot remove file %s!",
936 fCurrentDetector.Data(), filename));
937 }
73abe331 938}
939
b948db8d 940//______________________________________________________________________________________________
5164a766 941AliShuttleStatus* AliShuttle::ReadShuttleStatus()
942{
9827400b 943 //
944 // Reads the AliShuttleStatus from the CDB
945 //
5164a766 946
2bb7b766 947 if (fStatusEntry){
948 delete fStatusEntry;
949 fStatusEntry = 0;
950 }
5164a766 951
3ebbf3c4 952 Int_t path1 = GetCurrentRun()/10000;
10a5a932 953 fStatusEntry = AliCDBManager::Instance()->GetStorage(GetLocalCDB())
3ebbf3c4 954 ->Get(Form("/SHUTTLE/%s/%d", fCurrentDetector.Data(), path1), GetCurrentRun());
5164a766 955
2bb7b766 956 if (!fStatusEntry) return 0;
957 fStatusEntry->SetOwner(1);
5164a766 958
2bb7b766 959 AliShuttleStatus* status = dynamic_cast<AliShuttleStatus*> (fStatusEntry->GetObject());
960 if (!status) {
961 AliError("Invalid object stored to CDB!");
962 return 0;
963 }
5164a766 964
2bb7b766 965 return status;
5164a766 966}
967
968//______________________________________________________________________________________________
7bfb2090 969Bool_t AliShuttle::WriteShuttleStatus(AliShuttleStatus* status)
5164a766 970{
9827400b 971 //
972 // writes the status for one subdetector
973 //
2bb7b766 974
975 if (fStatusEntry){
976 delete fStatusEntry;
977 fStatusEntry = 0;
978 }
5164a766 979
2bb7b766 980 Int_t run = GetCurrentRun();
3ebbf3c4 981 Int_t path1 = run/10000;
982 TString path1_string = Form("%d",path1);
5164a766 983
3ebbf3c4 984 AliCDBId id(AliCDBPath("SHUTTLE", fCurrentDetector, path1_string), run, run);
5164a766 985
2bb7b766 986 fStatusEntry = new AliCDBEntry(status, id, new AliCDBMetaData);
987 fStatusEntry->SetOwner(1);
5164a766 988
924fb6e9 989 Int_t logLevel = AliLog::GetGlobalLogLevel();
990 AliLog::SetGlobalLogLevel(AliLog::kError);
991
2bb7b766 992 UInt_t result = AliCDBManager::Instance()->GetStorage(fgkLocalCDB)->Put(fStatusEntry);
7bfb2090 993
924fb6e9 994 AliLog::SetGlobalLogLevel((AliLog::EType_t)logLevel);
995
996 if (!result) {
3301427a 997 Log("SHUTTLE", Form("WriteShuttleStatus - Failed for %s, run %d",
998 fCurrentDetector.Data(), run));
2bb7b766 999 return kFALSE;
1000 }
e7f62f16 1001
8e828d39 1002 SendMLDetInfo();
7bfb2090 1003
2bb7b766 1004 return kTRUE;
5164a766 1005}
1006
1007//______________________________________________________________________________________________
1008void AliShuttle::UpdateShuttleStatus(AliShuttleStatus::Status newStatus, Bool_t increaseCount)
1009{
9827400b 1010 //
1011 // changes the AliShuttleStatus for the given detector and run to the given status
1012 //
5164a766 1013
2bb7b766 1014 if (!fStatusEntry){
1015 AliError("UNEXPECTED: fStatusEntry empty");
1016 return;
1017 }
5164a766 1018
2bb7b766 1019 AliShuttleStatus* status = dynamic_cast<AliShuttleStatus*> (fStatusEntry->GetObject());
5164a766 1020
2bb7b766 1021 if (!status){
c88ad5db 1022 Log("SHUTTLE", "UpdateShuttleStatus - UNEXPECTED: status could not be read from current CDB entry");
2bb7b766 1023 return;
1024 }
5164a766 1025
2c15234c 1026 TString actionStr = Form("UpdateShuttleStatus - %s: Changing state from %s to %s",
eba76848 1027 fCurrentDetector.Data(),
36c99a6a 1028 status->GetStatusName(),
eba76848 1029 status->GetStatusName(newStatus));
cb343cfd 1030 Log("SHUTTLE", actionStr);
1031 SetLastAction(actionStr);
5164a766 1032
2bb7b766 1033 status->SetStatus(newStatus);
1034 if (increaseCount) status->IncreaseCount();
5164a766 1035
924fb6e9 1036 Int_t logLevel = AliLog::GetGlobalLogLevel();
1037 AliLog::SetGlobalLogLevel(AliLog::kError);
1038
2bb7b766 1039 AliCDBManager::Instance()->GetStorage(fgkLocalCDB)->Put(fStatusEntry);
e7f62f16 1040
924fb6e9 1041 AliLog::SetGlobalLogLevel((AliLog::EType_t)logLevel);
1042
8e828d39 1043 SendMLDetInfo();
5164a766 1044}
e7f62f16 1045
1046//______________________________________________________________________________________________
8e828d39 1047void AliShuttle::SendMLDetInfo()
e7f62f16 1048{
1049 //
1050 // sends ML information about the current status of the current detector being processed
1051 //
1052
1053 AliShuttleStatus* status = dynamic_cast<AliShuttleStatus*> (fStatusEntry->GetObject());
1054
1055 if (!status){
8e828d39 1056 Log("SHUTTLE", "SendMLDetInfo - UNEXPECTED: status could not be read from current CDB entry");
e7f62f16 1057 return;
1058 }
1059
1060 TMonaLisaText mlStatus(Form("%s_status", fCurrentDetector.Data()), status->GetStatusName());
1061 TMonaLisaValue mlRetryCount(Form("%s_count", fCurrentDetector.Data()), status->GetCount());
1062
1063 TList mlList;
1064 mlList.Add(&mlStatus);
1065 mlList.Add(&mlRetryCount);
1066
ee6f7523 1067 TString mlID;
1068 mlID.Form("%d", GetCurrentRun());
1069 fMonaLisa->SendParameters(&mlList, mlID);
e7f62f16 1070}
1071
5164a766 1072//______________________________________________________________________________________________
1073Bool_t AliShuttle::ContinueProcessing()
1074{
9827400b 1075 // this function reads the AliShuttleStatus information from CDB and
1076 // checks if the processing should be continued
1077 // if yes it returns kTRUE and updates the AliShuttleStatus with nextStatus
2bb7b766 1078
926cbc0e 1079 if (!fConfig->HostProcessDetector(fCurrentDetector))
1080 return kFALSE;
57c1a579 1081
1082 AliPreprocessor* aPreprocessor =
1083 dynamic_cast<AliPreprocessor*> (fPreprocessorMap.GetValue(fCurrentDetector));
1084 if (!aPreprocessor)
1085 {
c88ad5db 1086 Log("SHUTTLE", Form("ContinueProcessing - %s: no preprocessor registered", fCurrentDetector.Data()));
57c1a579 1087 return kFALSE;
1088 }
1089
2bb7b766 1090 AliShuttleLogbookEntry::Status entryStatus =
eba76848 1091 fLogbookEntry->GetDetectorStatus(fCurrentDetector);
2bb7b766 1092
5e993b6f 1093 if (entryStatus != AliShuttleLogbookEntry::kUnprocessed) {
c88ad5db 1094 Log("SHUTTLE", Form("ContinueProcessing - %s is %s",
2bb7b766 1095 fCurrentDetector.Data(),
1096 fLogbookEntry->GetDetectorStatusName(entryStatus)));
1097 return kFALSE;
1098 }
1099
1100 // if we get here, according to Shuttle logbook subdetector is in UNPROCESSED state
be48e3ea 1101
1102 // check if current run is first unprocessed run for current detector
1103 if (fConfig->StrictRunOrder(fCurrentDetector) &&
1104 !fFirstUnprocessed[GetDetPos(fCurrentDetector)])
1105 {
86aa42c3 1106 if (fTestMode == kNone)
1107 {
c88ad5db 1108 Log("SHUTTLE", Form("ContinueProcessing - %s requires strict run ordering"
1109 " but this is not the first unprocessed run!"));
86aa42c3 1110 return kFALSE;
1111 }
1112 else
1113 {
c88ad5db 1114 Log("SHUTTLE", Form("ContinueProcessing - In TESTMODE - "
1115 "Although %s requires strict run ordering "
1116 "and this is not the first unprocessed run, "
1117 "the SHUTTLE continues"));
86aa42c3 1118 }
be48e3ea 1119 }
1120
926cbc0e 1121 // Is the subdetector processed first time for this run?
1122 fFirstProcessing = kFALSE;
1123
2bb7b766 1124 AliShuttleStatus* status = ReadShuttleStatus();
1125 if (!status) {
1126 // first time
1127 Log("SHUTTLE", Form("ContinueProcessing - %s: Processing first time",
1128 fCurrentDetector.Data()));
1129 status = new AliShuttleStatus(AliShuttleStatus::kStarted);
926cbc0e 1130 fFirstProcessing = kTRUE;
2bb7b766 1131 return WriteShuttleStatus(status);
1132 }
1133
ffa25f20 1134 // The following case shouldn't happen if Shuttle Logbook was correctly updated.
2bb7b766 1135 // If it happens it may mean Logbook updating failed... let's do it now!
1136 if (status->GetStatus() == AliShuttleStatus::kDone ||
ffa25f20 1137 status->GetStatus() == AliShuttleStatus::kFailed ||
1138 status->GetStatus() == AliShuttleStatus::kSkipped) {
2bb7b766 1139 Log("SHUTTLE", Form("ContinueProcessing - %s is already %s. Updating Shuttle Logbook",
1140 fCurrentDetector.Data(),
1141 status->GetStatusName(status->GetStatus())));
ffa25f20 1142
1143 if (status->GetStatus() == AliShuttleStatus::kSkipped)
1144 {
1145 UpdateShuttleLogbook(fCurrentDetector.Data(), "DONE");
1146 }
1147 else
1148 UpdateShuttleLogbook(fCurrentDetector.Data(), status->GetStatusName(status->GetStatus()));
1149
2bb7b766 1150 return kFALSE;
1151 }
1152
5e993b6f 1153 if (status->GetStatus() == AliShuttleStatus::kStoreStarted || status->GetStatus() == AliShuttleStatus::kStoreDelayed ||status->GetStatus() == AliShuttleStatus::kStoreError) {
2bb7b766 1154 Log("SHUTTLE",
c88ad5db 1155 Form("ContinueProcessing - %s: Grid storage of one or more "
1156 "objects failed. Trying again now",
2bb7b766 1157 fCurrentDetector.Data()));
5e993b6f 1158 StoreOCDB();
2bb7b766 1159 return kFALSE;
1160 }
1161
1162 // if we get here, there is a restart
57c1a579 1163 Bool_t cont = kFALSE;
2bb7b766 1164
1165 // abort conditions
cb343cfd 1166 if (status->GetCount() >= fConfig->GetMaxRetries()) {
57c1a579 1167 Log("SHUTTLE", Form("ContinueProcessing - %s failed %d times in status %s - "
1168 "Updating Shuttle Logbook", fCurrentDetector.Data(),
2bb7b766 1169 status->GetCount(), status->GetStatusName()));
1170 UpdateShuttleLogbook(fCurrentDetector.Data(), "FAILED");
e7f62f16 1171 UpdateShuttleStatus(AliShuttleStatus::kFailed);
3301427a 1172
1173 // there may still be objects in local OCDB and reference storage
1174 // and FXS databases may be not updated: do it now!
9827400b 1175
1176 // TODO Currently disabled, we want to keep files in case of failure!
1177 // CleanLocalStorage(fgkLocalCDB);
1178 // CleanLocalStorage(fgkLocalRefStorage);
1179 // UpdateTableFailCase();
1180
1181 // Send mail to detector expert!
c88ad5db 1182 Log("SHUTTLE", Form("ContinueProcessing - Sending mail to %s expert...",
926cbc0e 1183 fCurrentDetector.Data()));
fe6a3257 1184 // det experts in to
1185 TString to="";
1186 TIter *iterExperts = 0;
1187 iterExperts = new TIter(fConfig->GetResponsibles(fCurrentDetector));
1188 TObjString *anExpert=0;
1189 while ((anExpert = (TObjString*) iterExperts->Next()))
1190 {
1191 to += Form("%s, \n", anExpert->GetName());
1192 }
1193 delete iterExperts;
1194
1195 if (to.Length() > 0)
1196 to.Remove(to.Length()-3);
1197 AliDebug(2, Form("to: %s",to.Data()));
1198
1199 if (to.IsNull()) {
1200 Log("SHUTTLE", Form("List of %s responsibles not set!", fCurrentDetector.Data()));
1201 return kFALSE;
1202 }
1203
1204 Log(fCurrentDetector.Data(), Form("ContinueProcessing - Sending mail to %s expert(s):",
1205 fCurrentDetector.Data()));
1206 Log(fCurrentDetector.Data(), Form("\n%s", to.Data()));
339adafa 1207 if (!SendMail(kPPEMail))
9827400b 1208 Log("SHUTTLE", Form("ContinueProcessing - Could not send mail to %s expert",
926cbc0e 1209 fCurrentDetector.Data()));
3301427a 1210
57c1a579 1211 } else {
1212 Log("SHUTTLE", Form("ContinueProcessing - %s: restarting. "
1213 "Aborted before with %s. Retry number %d.", fCurrentDetector.Data(),
1214 status->GetStatusName(), status->GetCount()));
9827400b 1215 Bool_t increaseCount = kTRUE;
c88ad5db 1216 if (status->GetStatus() == AliShuttleStatus::kDCSError ||
8da81210 1217 status->GetStatus() == AliShuttleStatus::kDCSStarted ||
1218 status->GetStatus() == AliShuttleStatus::kFXSError)
c88ad5db 1219 increaseCount = kFALSE;
675f64cd 1220
9827400b 1221 UpdateShuttleStatus(AliShuttleStatus::kStarted, increaseCount);
57c1a579 1222 cont = kTRUE;
2bb7b766 1223 }
1224
57c1a579 1225 return cont;
5164a766 1226}
1227
1228//______________________________________________________________________________________________
8e828d39 1229void AliShuttle::SendMLRunInfo(const char* status)
1230{
1231 //
1232 // Send information about this run to ML
1233
1234 TMonaLisaText mlStatus("SHUTTLE_status", status);
1235 TString runType(fLogbookEntry->GetRunType());
1236 if (strlen(fLogbookEntry->GetRunParameter("log")) > 0){
1237
1238 runType += "(";
1239 runType += fLogbookEntry->GetRunParameter("log");
1240 runType += ")";
1241 }
728290a4 1242 if (fLogbookEntry->GetDATestMode()){
1243 runType += " (DATest)";
1244 }
8e828d39 1245 TMonaLisaText mlRunType("SHUTTLE_runtype", runType);
1246
1247 TList mlList;
1248 mlList.Add(&mlStatus);
1249 mlList.Add(&mlRunType);
1250
1251 TString mlID;
1252 mlID.Form("%d", GetCurrentRun());
1253 fMonaLisa->SendParameters(&mlList, mlID);
1254}
1255
1256//______________________________________________________________________________________________
ef83773f 1257Int_t AliShuttle::GetMem(Int_t pid)
1258{
1259 // invokes ps to get the memory consumption of the process <pid>
1260 // returns -1 in case of error
1261
1262 TString checkStr;
1263 checkStr.Form("ps -o vsize --pid %d | tail -n 1", pid);
1264 FILE* pipe = gSystem->OpenPipe(checkStr, "r");
1265 if (!pipe)
1266 {
1267 Log("SHUTTLE", Form("Process - Error: "
1268 "Could not open pipe to %s", checkStr.Data()));
1269 return -1;
1270 }
1271
1272 char buffer[100];
1273 if (!fgets(buffer, 100, pipe))
1274 {
1275 Log("SHUTTLE", "Process - Error: ps did not return anything");
1276 gSystem->ClosePipe(pipe);
1277 return -1;
1278 }
1279 gSystem->ClosePipe(pipe);
1280
1281 //Log("SHUTTLE", Form("ps returned %s", buffer));
1282
1283 Int_t mem = 0;
1284 if ((sscanf(buffer, "%d\n", &mem) != 1) || !mem)
1285 {
1286 Log("SHUTTLE", "Process - Error: Could not parse output of ps");
1287 return -1;
1288 }
1289
1290 return mem;
1291}
1292
1293//______________________________________________________________________________________________
2bb7b766 1294Bool_t AliShuttle::Process(AliShuttleLogbookEntry* entry)
58bc3020 1295{
73abe331 1296 //
b948db8d 1297 // Makes data retrieval for all detectors in the configuration.
2bb7b766 1298 // entry: Shuttle logbook entry, contains run paramenters and status of detectors
1299 // (Unprocessed, Inactive, Failed or Done).
d477ad88 1300 // Returns kFALSE in case of error occured and kTRUE otherwise
73abe331 1301 //
1302
9827400b 1303 if (!entry) return kFALSE;
2bb7b766 1304
1305 fLogbookEntry = entry;
1306
c88ad5db 1307 Log("SHUTTLE", Form("\t\t\t^*^*^*^*^*^*^*^*^*^*^*^* run %d: START ^*^*^*^*^*^*^*^*^*^*^*^*",
9827400b 1308 GetCurrentRun()));
2bb7b766 1309
339adafa 1310 CountOpenRuns();
1311
8e828d39 1312 // Send the information to ML
1313 SendMLRunInfo("Processing");
3301427a 1314
9827400b 1315 if (fLogbookEntry->IsDone())
1316 {
1317 Log("SHUTTLE","Process - Shuttle is already DONE. Updating logbook");
1318 UpdateShuttleLogbook("shuttle_done");
1319 fLogbookEntry = 0;
1320 return kTRUE;
1321 }
1322
1323 // read test mode if flag is set
1324 if (fReadTestMode)
1325 {
3d8bc902 1326 fTestMode = kNone;
9827400b 1327 TString logEntry(entry->GetRunParameter("log"));
1328 //printf("log entry = %s\n", logEntry.Data());
1329 TString searchStr("Testmode: ");
1330 Int_t pos = logEntry.Index(searchStr.Data());
1331 //printf("%d\n", pos);
1332 if (pos >= 0)
1333 {
1334 TSubString subStr = logEntry(pos + searchStr.Length(), logEntry.Length());
1335 //printf("%s\n", subStr.String().Data());
1336 TString newStr(subStr.Data());
1337 TObjArray* token = newStr.Tokenize(' ');
1338 if (token)
1339 {
1340 //token->Print();
1341 TObjString* tmpStr = dynamic_cast<TObjString*> (token->First());
1342 if (tmpStr)
1343 {
1344 Int_t testMode = tmpStr->String().Atoi();
1345 if (testMode > 0)
1346 {
c88ad5db 1347 Log("SHUTTLE", Form("Process - Enabling test mode %d", testMode));
9827400b 1348 SetTestMode((TestMode) testMode);
1349 }
1350 }
1351 delete token;
1352 }
1353 }
1354 }
c88ad5db 1355
eba76848 1356 fLogbookEntry->Print("all");
57f50b3c 1357
1358 // Initialization
d477ad88 1359 Bool_t hasError = kFALSE;
5164a766 1360
728290a4 1361 // Set the CDB and Reference folders according to the year
1362
1363 // build cdb paths (repeat each time, run might be a DATest run)
1364 if (!fLogbookEntry->GetDATestMode()){
1365 fgkMainCDB.Form("alien://folder=%s%d/OCDB?user=alidaq?cacheFold=/tmp/OCDBCache",
1366 fConfig->GetAlienPath(), GetCurrentYear());
1367
1368 fgkMainRefStorage.Form("alien://folder=%s%d/Reference?user=alidaq?cacheFold=/tmp/OCDBCache",
1369 fConfig->GetAlienPath(), GetCurrentYear());
1370 }
1371 else {
1372 fgkMainCDB.Form("alien://folder=%s%d/DATest/OCDB?user=alidaq?cacheFold=/tmp/OCDBCache",
1373 fConfig->GetAlienPath(), GetCurrentYear());
1374
1375 fgkMainRefStorage.Form("alien://folder=%s%d/DATest/Reference?user=alidaq?cacheFold=/tmp/OCDBCache",
1376 fConfig->GetAlienPath(), GetCurrentYear());
1377 }
1378
1379 AliDebug(2,Form("Main CDB storage = %s",fgkMainCDB.Data()));
1380 AliDebug(2,Form("Main Reference storage = %s",fgkMainRefStorage.Data()));
1381
57f50b3c 1382 // Loop on detectors in the configuration
b948db8d 1383 TIter iter(fConfig->GetDetectors());
2bb7b766 1384 TObjString* aDetector = 0;
b948db8d 1385
4508ab8b 1386 Bool_t first = kTRUE;
1387
be48e3ea 1388 while ((aDetector = (TObjString*) iter.Next()))
1389 {
7bfb2090 1390 fCurrentDetector = aDetector->String();
5164a766 1391
5e993b6f 1392 if (ContinueProcessing() == kFALSE)
1393 continue;
4508ab8b 1394
1395 if (first)
1396 {
1397 // only read QueryCDB when needed and only once
1398 AliCDBStorage *mainCDBSto = AliCDBManager::Instance()->GetStorage(fgkMainCDB);
1399 if(mainCDBSto) mainCDBSto->QueryCDB(GetCurrentRun());
1400 AliCDBStorage *mainRefSto = AliCDBManager::Instance()->GetStorage(fgkMainRefStorage);
1401 if(mainRefSto) mainRefSto->QueryCDB(GetCurrentRun());
1402 first = kFALSE;
1403 }
9e080f92 1404
c88ad5db 1405 Log("SHUTTLE", Form("\t\t\t****** run %d - %s: START ******",
2bb7b766 1406 GetCurrentRun(), aDetector->GetName()));
1407
9d733021 1408 for(Int_t iSys=0;iSys<3;iSys++) fFXSCalled[iSys]=kFALSE;
ef83773f 1409
1410 Int_t initialMem = GetMem(getpid());
1411 Log("SHUTTLE", Form("Memory consumption before forking is %d", initialMem));
9d733021 1412
c88ad5db 1413 Log(fCurrentDetector.Data(), "Process - Starting processing");
85a80aa9 1414
be48e3ea 1415 Int_t pid = fork();
1416
1417 if (pid < 0)
1418 {
c88ad5db 1419 Log("SHUTTLE", "Process - ERROR: Forking failed");
be48e3ea 1420 }
1421 else if (pid > 0)
1422 {
1423 // parent
c88ad5db 1424 Log("SHUTTLE", Form("Process - In parent process of %d - %s: Starting monitoring",
be48e3ea 1425 GetCurrentRun(), aDetector->GetName()));
1426
1427 Long_t begin = time(0);
1428
1429 int status; // to be used with waitpid, on purpose an int (not Int_t)!
1430 while (waitpid(pid, &status, WNOHANG) == 0)
1431 {
1432 Long_t expiredTime = time(0) - begin;
1433
1434 if (expiredTime > fConfig->GetPPTimeOut())
1435 {
56d3e9e4 1436 TString logMsg;
1437 AliShuttleStatus *currentStatus = ReadShuttleStatus();
1438 AliShuttleStatus::Status newStatus = AliShuttleStatus::kInvalid;
1439
1440 if (currentStatus->GetStatus() <= AliShuttleStatus::kPPDone)
1441 {
1442 // in case pp not yet done set status to kPPTimeOut
1443
1444 logMsg.Form("Process - Process of %s timed out. Run time: %d seconds. Killing...",
1445 fCurrentDetector.Data(), expiredTime);
1446 newStatus = AliShuttleStatus::kPPTimeOut;
1447 }
1448 else if (currentStatus->GetStatus() == AliShuttleStatus::kStoreStarted)
1449 {
1450 // in case the pp goes in TimeOut while storing the objects in the OCDB
1451 // set status to kStoreError
1452
1453 logMsg.Form("Process - Process of %s timed out while storing the OCDB object. Run time: %d seconds. Killing... and setting status to StoreError.",
1454 fCurrentDetector.Data(), expiredTime);
1455 newStatus = AliShuttleStatus::kStoreError;
1456 }
1457 else
1458 {
1459 // in other cases don't change the status
1460
1461 logMsg.Form("Process - Process of %s timed out in status = %s. Run time: %d seconds. Killing... without changing the status",
1462 fCurrentDetector.Data(), currentStatus->GetStatusName(), expiredTime);
1463 }
1464
1465 Log("SHUTTLE", logMsg);
1466 Log(fCurrentDetector, logMsg);
be48e3ea 1467
1468 kill(pid, 9);
1469
56d3e9e4 1470 if (newStatus != AliShuttleStatus::kInvalid)
1471 UpdateShuttleStatus(newStatus);
be48e3ea 1472 hasError = kTRUE;
1473
1474 gSystem->Sleep(1000);
1475 }
1476 else
1477 {
be48e3ea 1478 gSystem->Sleep(1000);
9827400b 1479
ef83773f 1480 Int_t mem = GetMem(pid);
1481
1482 if (mem < 0)
9827400b 1483 continue;
9827400b 1484
ef83773f 1485 mem -= initialMem;
1486 if (mem < 0)
1487 mem = 0;
9827400b 1488
1489 if (expiredTime % 60 == 0)
ee6f7523 1490 {
c88ad5db 1491 Log("SHUTTLE", Form("Process - %s: Checking process. "
1492 "Run time: %d seconds - Memory consumption: %d KB",
1493 fCurrentDetector.Data(), expiredTime, mem));
ee6f7523 1494 SendAlive();
1495 }
9827400b 1496
1497 if (mem > fConfig->GetPPMaxMem())
1498 {
1499 TString tmp;
c88ad5db 1500 tmp.Form("Process - Process exceeds maximum allowed memory "
1501 "(%d KB > %d KB). Killing...",
9827400b 1502 mem, fConfig->GetPPMaxMem());
1503 Log("SHUTTLE", tmp);
1504 Log(fCurrentDetector, tmp);
1505
1506 kill(pid, 9);
1507
1508 UpdateShuttleStatus(AliShuttleStatus::kPPOutOfMemory);
1509 hasError = kTRUE;
1510
1511 gSystem->Sleep(1000);
1512 }
be48e3ea 1513 }
1514 }
1515
c88ad5db 1516 Log("SHUTTLE", Form("Process - In parent process of %d - %s: Client has terminated.",
be48e3ea 1517 GetCurrentRun(), aDetector->GetName()));
1518
1519 if (WIFEXITED(status))
1520 {
1521 Int_t returnCode = WEXITSTATUS(status);
1522
c88ad5db 1523 Log("SHUTTLE", Form("Process - %s: the return code is %d", fCurrentDetector.Data(),
3301427a 1524 returnCode));
be48e3ea 1525
9827400b 1526 if (returnCode == 0) hasError = kTRUE;
be48e3ea 1527 }
1528 }
1529 else if (pid == 0)
1530 {
26b1866e 1531 // child
1532 Log("SHUTTLE", Form("Process - In child process of %d - %s", GetCurrentRun(),
c88ad5db 1533 aDetector->GetName()));
be48e3ea 1534
c88ad5db 1535 Log("SHUTTLE", Form("Process - Redirecting output to %s log",fCurrentDetector.Data()));
ffa29e93 1536
546242fb 1537 if ((freopen(GetLogFileName(fCurrentDetector), "a", stdout)) == 0)
ffa29e93 1538 {
c88ad5db 1539 Log("SHUTTLE", "Process - Could not freopen stdout");
ffa29e93 1540 }
1541 else
1542 {
1543 fOutputRedirected = kTRUE;
1544 if ((dup2(fileno(stdout), fileno(stderr))) < 0)
c88ad5db 1545 Log("SHUTTLE", "Process - Could not redirect stderr");
ffa29e93 1546
1547 }
1548
5bac2bde 1549 TString wd = gSystem->WorkingDirectory();
3ebbf3c4 1550 Int_t dir_lev1 = GetCurrentRun()/10000;
1551 TString tmpDir = Form("%s/%d/%d/%s_process", GetShuttleTempDir(),
1552 dir_lev1, GetCurrentRun(), fCurrentDetector.Data());
5bac2bde 1553
d524ade6 1554 Int_t result = gSystem->GetPathInfo(tmpDir.Data(), 0, (Long64_t*) 0, 0, 0);
1555 if (!result) // temp dir already exists!
1556 {
1557 Log(fCurrentDetector.Data(),
1558 Form("Process - %s dir already exists! Removing...", tmpDir.Data()));
1559 gSystem->Exec(Form("rm -rf %s",tmpDir.Data()));
675f64cd 1560 }
1561
1562 if (gSystem->mkdir(tmpDir.Data(), 1))
1563 {
1564 Log(fCurrentDetector.Data(), "Process - could not make temp directory!!");
1565 gSystem->Exit(1);
d524ade6 1566 }
1567
1568 if (!gSystem->ChangeDirectory(tmpDir.Data()))
1569 {
1570 Log(fCurrentDetector.Data(), "Process - could not change directory!!");
1571 gSystem->Exit(1);
1572 }
5bac2bde 1573
ffa25f20 1574 Int_t success = ProcessCurrentDetector();
26b1866e 1575
5bac2bde 1576 gSystem->ChangeDirectory(wd.Data());
b1d18693 1577
ffa25f20 1578 if (success == 1) // Preprocessor finished successfully!
9827400b 1579 {
fb40ead4 1580 // remove temporary folder or DCS map
1581 if (!fConfig->KeepTempFolder())
1582 {
1583 gSystem->Exec(Form("rm -rf %s",tmpDir.Data()));
1584 } else if (!fConfig->KeepDCSMap())
1585 {
1586 gSystem->Exec(Form("rm -f %s/DCSMap.root",tmpDir.Data()));
1587 }
b1d18693 1588
3301427a 1589 // Update time_processed field in FXS DB
1590 if (UpdateTable() == kFALSE)
5bac2bde 1591 Log("SHUTTLE", Form("Process - %s: Could not update FXS databases!",
1592 fCurrentDetector.Data()));
3301427a 1593
1594 // Transfer the data from local storage to main storage (Grid)
3301427a 1595 if (StoreOCDB() == kFALSE)
9827400b 1596 success = kFALSE;
5e993b6f 1597 }
ffa25f20 1598 else if (success == 0)
c88ad5db 1599 {
1600 Log("SHUTTLE",
1601 Form("\t\t\t****** run %d - %s: PP ERROR ******",
1602 GetCurrentRun(), aDetector->GetName()));
be48e3ea 1603 }
1604
4b95672b 1605 for (UInt_t iSys=0; iSys<3; iSys++)
1606 {
1607 if (fFXSCalled[iSys]) fFXSlist[iSys].Clear();
1608 }
1609
c88ad5db 1610 Log("SHUTTLE", Form("Process - Client process of %d - %s is exiting now with %d.",
9827400b 1611 GetCurrentRun(), aDetector->GetName(), success));
be48e3ea 1612
1613 // the client exits here
9827400b 1614 gSystem->Exit(success);
be48e3ea 1615
1616 AliError("We should never get here!!!");
1617 }
7bfb2090 1618 }
5164a766 1619
c88ad5db 1620 Log("SHUTTLE", Form("\t\t\t^*^*^*^*^*^*^*^*^*^*^*^* run %d: FINISH ^*^*^*^*^*^*^*^*^*^*^*^*",
2bb7b766 1621 GetCurrentRun()));
1622
1623 //check if shuttle is done for this run, if so update logbook
1624 TObjArray checkEntryArray;
1625 checkEntryArray.SetOwner(1);
9e080f92 1626 TString whereClause = Form("where run=%d", GetCurrentRun());
b0e53b15 1627 if (!QueryShuttleLogbook(whereClause.Data(), checkEntryArray) ||
1628 checkEntryArray.GetEntries() == 0) {
9e080f92 1629 Log("SHUTTLE", Form("Process - Warning: Cannot check status of run %d on Shuttle logbook!",
1630 GetCurrentRun()));
1631 return hasError == kFALSE;
1632 }
b948db8d 1633
9e080f92 1634 AliShuttleLogbookEntry* checkEntry = dynamic_cast<AliShuttleLogbookEntry*>
1635 (checkEntryArray.At(0));
2bb7b766 1636
9e080f92 1637 if (checkEntry)
1638 {
1639 if (checkEntry->IsDone())
be48e3ea 1640 {
9e080f92 1641 Log("SHUTTLE","Process - Shuttle is DONE. Updating logbook");
1642 UpdateShuttleLogbook("shuttle_done");
1643 }
1644 else
1645 {
1646 for (UInt_t iDet=0; iDet<NDetectors(); iDet++)
be48e3ea 1647 {
9e080f92 1648 if (checkEntry->GetDetectorStatus(iDet) == AliShuttleLogbookEntry::kUnprocessed)
be48e3ea 1649 {
9e080f92 1650 AliDebug(2, Form("Run %d: setting %s as \"not first time unprocessed\"",
1651 checkEntry->GetRun(), GetDetName(iDet)));
1652 fFirstUnprocessed[iDet] = kFALSE;
be48e3ea 1653 }
1654 }
8e828d39 1655 SendMLRunInfo("Pending");
2bb7b766 1656 }
1657 }
1658
1659 fLogbookEntry = 0;
85a80aa9 1660
a7160fe9 1661 return hasError == kFALSE;
73abe331 1662}
1663
b948db8d 1664//______________________________________________________________________________________________
ffa25f20 1665Int_t AliShuttle::ProcessCurrentDetector()
73abe331 1666{
1667 //
2bb7b766 1668 // Makes data retrieval just for a specific detector (fCurrentDetector).
73abe331 1669 // Threre should be a configuration for this detector.
73abe331 1670
1d172743 1671 Log("SHUTTLE", Form("ProcessCurrentDetector - Retrieving values for %s, run %d",
1672 fCurrentDetector.Data(), GetCurrentRun()));
73abe331 1673
d524ade6 1674 TString wd = gSystem->WorkingDirectory();
1675
2d9019b4 1676 if (!CleanReferenceStorage(fCurrentDetector.Data()))
ffa25f20 1677 return 0;
d524ade6 1678
1679 gSystem->ChangeDirectory(wd.Data());
1680
3301427a 1681 // call preprocessor
1682 AliPreprocessor* aPreprocessor =
1683 dynamic_cast<AliPreprocessor*> (fPreprocessorMap.GetValue(fCurrentDetector));
1684
8579c6fb 1685 // check if the preprocessor wants to process this run type
1686 if (aPreprocessor->ProcessRunType() == kFALSE)
ffa25f20 1687 {
1688 UpdateShuttleStatus(AliShuttleStatus::kSkipped);
1689 UpdateShuttleLogbook(fCurrentDetector, "DONE");
bd23d48b 1690 if (!UpdateTableSkippedCase(fCurrentDetector.Data()))
1691 {
1692 AliError(Form("Could not update FXS tables for run %d !", GetCurrentRun()));
1693 }
ffa25f20 1694 Log(fCurrentDetector, Form("ProcessCurrentDetector - %s preprocessor is not interested in this run type", fCurrentDetector.Data()));
1695
1696 return 2;
1697 }
1698
1699 TMap* dcsMap = new TMap();
1700
3301427a 1701 aPreprocessor->Initialize(GetCurrentRun(), GetCurrentStartTime(), GetCurrentEndTime());
1702
1703 Bool_t processDCS = aPreprocessor->ProcessDCS();
d477ad88 1704
651fdaab 1705 if (!processDCS)
1706 {
1d172743 1707 Log(fCurrentDetector, "ProcessCurrentDetector -"
1708 " The preprocessor requested to skip the retrieval of DCS values");
651fdaab 1709 }
8b739301 1710 else if (fTestMode & kSkipDCS)
2c15234c 1711 {
1d172743 1712 Log(fCurrentDetector, "ProcessCurrentDetector - In TESTMODE: Skipping DCS processing");
9827400b 1713 }
1714 else if (fTestMode & kErrorDCS)
1715 {
1d172743 1716 Log(fCurrentDetector, "ProcessCurrentDetector - In TESTMODE: Simulating DCS error");
3d8bc902 1717 UpdateShuttleStatus(AliShuttleStatus::kDCSStarted);
9827400b 1718 UpdateShuttleStatus(AliShuttleStatus::kDCSError);
1d172743 1719 delete dcsMap;
ffa25f20 1720 return 0;
2c15234c 1721 } else {
3301427a 1722
1723 UpdateShuttleStatus(AliShuttleStatus::kDCSStarted);
1724
1d172743 1725 // Query DCS archive
1726 Int_t nServers = fConfig->GetNServers(fCurrentDetector);
a038aa70 1727
1d172743 1728 for (int iServ=0; iServ<nServers; iServ++)
2c15234c 1729 {
1d172743 1730
1731 TString host(fConfig->GetDCSHost(fCurrentDetector, iServ));
1732 Int_t port = fConfig->GetDCSPort(fCurrentDetector, iServ);
542b6cc8 1733 Int_t multiSplit = fConfig->GetMultiSplit(fCurrentDetector, iServ);
1734
1790d4b7 1735 Log(fCurrentDetector, Form("ProcessCurrentDetector -"
1736 " Querying DCS Amanda server %s:%d (%d of %d)",
1737 host.Data(), port, iServ+1, nServers));
1d172743 1738
1739 TMap* aliasMap = 0;
1740 TMap* dpMap = 0;
26b1866e 1741
1d172743 1742 if (fConfig->GetDCSAliases(fCurrentDetector, iServ)->GetEntries() > 0)
2c15234c 1743 {
5ed060b1 1744 Log(fCurrentDetector, Form("Querying %d DCS aliases", fConfig->GetDCSAliases(fCurrentDetector, iServ)->GetEntries()));
1d172743 1745 aliasMap = GetValueSet(host, port,
542b6cc8 1746 fConfig->GetDCSAliases(fCurrentDetector, iServ),
1747 kAlias, multiSplit);
1d172743 1748 if (!aliasMap)
1749 {
1750 Log(fCurrentDetector,
1751 Form("ProcessCurrentDetector -"
675f64cd 1752 " Error retrieving DCS aliases from server %s."
1753 " Sending mail to DCS experts!", host.Data()));
1d172743 1754 UpdateShuttleStatus(AliShuttleStatus::kDCSError);
675f64cd 1755
339adafa 1756 if (!SendMail(kDCSEMail))
fb40ead4 1757 Log("SHUTTLE", Form("ProcessCurrentDetector - "
95d4aaef 1758 "Could not send mail to DCS experts!"));
675f64cd 1759
a038aa70 1760 delete dcsMap;
ffa25f20 1761 return 0;
1d172743 1762 }
2c15234c 1763 }
a038aa70 1764
1d172743 1765 if (fConfig->GetDCSDataPoints(fCurrentDetector, iServ)->GetEntries() > 0)
a038aa70 1766 {
5ed060b1 1767 Log(fCurrentDetector, Form("Querying %d DCS data points", fConfig->GetDCSDataPoints(fCurrentDetector, iServ)->GetEntries()));
1d172743 1768 dpMap = GetValueSet(host, port,
542b6cc8 1769 fConfig->GetDCSDataPoints(fCurrentDetector, iServ),
1770 kDP, multiSplit);
1d172743 1771 if (!dpMap)
1772 {
1773 Log(fCurrentDetector,
1774 Form("ProcessCurrentDetector -"
675f64cd 1775 " Error retrieving DCS data points from server %s."
1776 " Sending mail to DCS experts!", host.Data()));
1d172743 1777 UpdateShuttleStatus(AliShuttleStatus::kDCSError);
675f64cd 1778
339adafa 1779 if (!SendMail(kDCSEMail))
fb40ead4 1780 Log("SHUTTLE", Form("ProcessCurrentDetector - "
95d4aaef 1781 "Could not send mail to DCS experts!"));
675f64cd 1782
1d172743 1783 if (aliasMap) delete aliasMap;
1784 delete dcsMap;
ffa25f20 1785 return 0;
1d172743 1786 }
a038aa70 1787 }
1d172743 1788
1789 // merge aliasMap and dpMap into dcsMap
1790 if(aliasMap) {
1791 TIter iter(aliasMap);
a038aa70 1792 TObjString* key = 0;
1793 while ((key = (TObjString*) iter.Next()))
1d172743 1794 dcsMap->Add(key, aliasMap->GetValue(key->String()));
1795
1796 aliasMap->SetOwner(kFALSE);
1797 delete aliasMap;
1798 }
1799
1800 if(dpMap) {
1801 TIter iter(dpMap);
1802 TObjString* key = 0;
1803 while ((key = (TObjString*) iter.Next()))
1804 dcsMap->Add(key, dpMap->GetValue(key->String()));
1805
1806 dpMap->SetOwner(kFALSE);
1807 delete dpMap;
a038aa70 1808 }
73abe331 1809 }
1810 }
dc25836b 1811
b1d18693 1812 // save map into file, to help debugging in case of preprocessor error
fbc112e3 1813 TFile* f = TFile::Open("DCSMap.root","recreate");
b1d18693 1814 f->cd();
1815 dcsMap->Write("DCSMap", TObject::kSingleKey);
1816 f->Close();
fbc112e3 1817 delete f;
b1d18693 1818
2bb7b766 1819 // DCS Archive DB processing successful. Call Preprocessor!
85a80aa9 1820 UpdateShuttleStatus(AliShuttleStatus::kPPStarted);
a7160fe9 1821
339adafa 1822 fFXSError = -1; // this variable is kTRUE after ::Process if an FXS error occured
8da81210 1823
a038aa70 1824 UInt_t returnValue = aPreprocessor->Process(dcsMap);
8da81210 1825
339adafa 1826 if (fFXSError!=-1) {
8da81210 1827 UpdateShuttleStatus(AliShuttleStatus::kFXSError);
339adafa 1828 SendMail(kFXSEMail, fFXSError);
8da81210 1829 dcsMap->DeleteAll();
1830 delete dcsMap;
ffa25f20 1831 return 0;
8da81210 1832 }
b948db8d 1833
3301427a 1834 if (returnValue > 0) // Preprocessor error!
1835 {
c88ad5db 1836 Log(fCurrentDetector, Form("ProcessCurrentDetector - "
1837 "Preprocessor failed. Process returned %d.", returnValue));
cb343cfd 1838 UpdateShuttleStatus(AliShuttleStatus::kPPError);
a038aa70 1839 dcsMap->DeleteAll();
1840 delete dcsMap;
ffa25f20 1841 return 0;
9827400b 1842 }
1843
1844 // preprocessor ok!
1845 UpdateShuttleStatus(AliShuttleStatus::kPPDone);
1846 Log(fCurrentDetector, Form("ProcessCurrentDetector - %s preprocessor returned success",
1847 fCurrentDetector.Data()));
b948db8d 1848
a038aa70 1849 dcsMap->DeleteAll();
1850 delete dcsMap;
b948db8d 1851
ffa25f20 1852 return 1;
2bb7b766 1853}
1854
1855//______________________________________________________________________________________________
fbc112e3 1856void AliShuttle::CountOpenRuns()
1857{
1858 // Query DAQ's Shuttle logbook and sends the number of open runs to ML
1859
982278d6 1860 SendAlive();
1861
fbc112e3 1862 // check connection, in case connect
1863 if (!Connect(3))
1864 return;
1865
1866 TString sqlQuery;
1867 sqlQuery = Form("select count(*) from %s where shuttle_done=0", fConfig->GetShuttlelbTable());
1868
1869 TSQLResult* aResult = fServer[3]->Query(sqlQuery);
1870 if (!aResult) {
1871 AliError(Form("Can't execute query <%s>!", sqlQuery.Data()));
1872 return;
1873 }
1874
1875 AliDebug(2,Form("Query = %s", sqlQuery.Data()));
1876
1877 if (aResult->GetRowCount() == 0) {
1878 AliError(Form("No result for query %s received", sqlQuery.Data()));
1879 return;
1880 }
1881
1882 if (aResult->GetFieldCount() != 1) {
1883 AliError(Form("Invalid field count for query %s received", sqlQuery.Data()));
1884 return;
1885 }
1886
1887 TSQLRow* aRow = aResult->Next();
1888 if (!aRow) {
1889 AliError(Form("Could not receive result of query %s", sqlQuery.Data()));
1890 return;
1891 }
1892
1893 TString result(aRow->GetField(0), aRow->GetFieldLength(0));
1894 Int_t count = result.Atoi();
1895
1896 Log("SHUTTLE", Form("%d unprocessed runs", count));
1897
1898 delete aRow;
1899 delete aResult;
1900
1901 TMonaLisaValue mlStatus("SHUTTLE_openruns", count);
1902
1903 TList mlList;
1904 mlList.Add(&mlStatus);
1905
1906 fMonaLisa->SendParameters(&mlList, "__PROCESSINGINFO__");
1907}
1908
1909//______________________________________________________________________________________________
2bb7b766 1910Bool_t AliShuttle::QueryShuttleLogbook(const char* whereClause,
1911 TObjArray& entries)
1912{
9827400b 1913 // Query DAQ's Shuttle logbook and fills detector status object.
1914 // Call QueryRunParameters to query DAQ logbook for run parameters.
1915 //
2bb7b766 1916
fc5a4708 1917 entries.SetOwner(1);
1918
2bb7b766 1919 // check connection, in case connect
fbc112e3 1920 if (!Connect(3)) return kFALSE;
2bb7b766 1921
1922 TString sqlQuery;
441b0e9c 1923 sqlQuery = Form("select * from %s %s order by run", fConfig->GetShuttlelbTable(), whereClause);
2bb7b766 1924
be48e3ea 1925 TSQLResult* aResult = fServer[3]->Query(sqlQuery);
2bb7b766 1926 if (!aResult) {
1927 AliError(Form("Can't execute query <%s>!", sqlQuery.Data()));
1928 return kFALSE;
1929 }
1930
fc5a4708 1931 AliDebug(2,Form("Query = %s", sqlQuery.Data()));
1932
2bb7b766 1933 if(aResult->GetRowCount() == 0) {
c88ad5db 1934 Log("SHUTTLE", "No entries in Shuttle Logbook match request");
9827400b 1935 delete aResult;
1936 return kTRUE;
2bb7b766 1937 }
1938
1939 // TODO Check field count!
5b23e2da 1940 const UInt_t nCols = 26;
2bb7b766 1941 if (aResult->GetFieldCount() != (Int_t) nCols) {
c88ad5db 1942 Log("SHUTTLE", "Invalid SQL result field number!");
2bb7b766 1943 delete aResult;
1944 return kFALSE;
1945 }
1946
2bb7b766 1947 TSQLRow* aRow;
1948 while ((aRow = aResult->Next())) {
1949 TString runString(aRow->GetField(0), aRow->GetFieldLength(0));
1950 Int_t run = runString.Atoi();
1951
eba76848 1952 AliShuttleLogbookEntry *entry = QueryRunParameters(run);
1953 if (!entry)
1954 continue;
2bb7b766 1955
728290a4 1956 // DA test mode flag
1957 TString daTestModeString(aRow->GetField(2), aRow->GetFieldLength(2)); // field 2 = DA test mode flag
1958 Bool_t daTestMode = (Bool_t)daTestModeString.Atoi();
1959 entry->SetDATestMode(daTestMode);
1960
2bb7b766 1961 // loop on detectors
eba76848 1962 for(UInt_t ii = 0; ii < nCols; ii++)
1963 entry->SetDetectorStatus(aResult->GetFieldName(ii), aRow->GetField(ii));
2bb7b766 1964
eba76848 1965 entries.AddLast(entry);
2bb7b766 1966 delete aRow;
1967 }
1968
2bb7b766 1969 delete aResult;
1970 return kTRUE;
1971}
1972
1973//______________________________________________________________________________________________
eba76848 1974AliShuttleLogbookEntry* AliShuttle::QueryRunParameters(Int_t run)
2bb7b766 1975{
eba76848 1976 //
1977 // Retrieve run parameters written in the DAQ logbook and sets them into AliShuttleLogbookEntry object
1978 //
2bb7b766 1979
1980 // check connection, in case connect
be48e3ea 1981 if (!Connect(3))
eba76848 1982 return 0;
2bb7b766 1983
1984 TString sqlQuery;
2c15234c 1985 sqlQuery.Form("select * from %s where run=%d", fConfig->GetDAQlbTable(), run);
2bb7b766 1986
be48e3ea 1987 TSQLResult* aResult = fServer[3]->Query(sqlQuery);
2bb7b766 1988 if (!aResult) {
c88ad5db 1989 Log("SHUTTLE", Form("Can't execute query <%s>!", sqlQuery.Data()));
eba76848 1990 return 0;
2bb7b766 1991 }
1992
eba76848 1993 if (aResult->GetRowCount() == 0) {
2bb7b766 1994 Log("SHUTTLE", Form("QueryRunParameters - No entry in DAQ Logbook for run %d. Skipping", run));
1995 delete aResult;
eba76848 1996 return 0;
2bb7b766 1997 }
1998
eba76848 1999 if (aResult->GetRowCount() > 1) {
c88ad5db 2000 Log("SHUTTLE", Form("QueryRunParameters - UNEXPECTED: "
2001 "more than one entry in DAQ Logbook for run %d!", run));
2bb7b766 2002 delete aResult;
eba76848 2003 return 0;
2bb7b766 2004 }
2005
eba76848 2006 TSQLRow* aRow = aResult->Next();
2007 if (!aRow)
2008 {
c88ad5db 2009 Log("SHUTTLE", Form("QueryRunParameters - Could not retrieve row for run %d. Skipping", run));
eba76848 2010 delete aResult;
2011 return 0;
2012 }
2bb7b766 2013
eba76848 2014 AliShuttleLogbookEntry* entry = new AliShuttleLogbookEntry(run);
2bb7b766 2015
eba76848 2016 for (Int_t ii = 0; ii < aResult->GetFieldCount(); ii++)
2017 entry->SetRunParameter(aResult->GetFieldName(ii), aRow->GetField(ii));
2bb7b766 2018
c9db68c7 2019 delete aRow;
2020 delete aResult;
2021
eba76848 2022 UInt_t startTime = entry->GetStartTime();
2023 UInt_t endTime = entry->GetEndTime();
89bcf3b4 2024 Bool_t ecsSuccess = entry->GetECSSuccess();
c9db68c7 2025
b0e53b15 2026 TString totEventsStr = entry->GetRunParameter("totalEvents");
2027 Int_t totEvents = totEventsStr.Atoi();
c9db68c7 2028
af93e5d7 2029 UInt_t now = time(0);
2030 // TODO make this a configuration parameter
2eb98462 2031 Int_t dcsDelay = fConfig->GetDCSDelay()+fConfig->GetDCSQueryOffset();
af93e5d7 2032
8e828d39 2033 // runs are accepted if they have ecsSuccess set or more than 1 event
af93e5d7 2034 if (startTime != 0 && endTime != 0 && endTime > startTime && (totEvents > 1 || ecsSuccess) && (endTime < now - dcsDelay))
89bcf3b4 2035 {
8e828d39 2036 if (ecsSuccess == kFALSE)
2037 Log("SHUTTLE", Form("Processing run %d although in status ECS failure, Reason: %s", run, entry->GetRunParameter("eor_reason")));
2038 return entry;
89bcf3b4 2039 }
8e828d39 2040
2041 Bool_t skip = kFALSE;
2042
f8523731 2043 if (endTime != 0 && endTime >= now - dcsDelay)
89bcf3b4 2044 {
f8523731 2045 Log("SHUTTLE", Form("Skipping run %d for now, because DCS buffer time is not yet expired", run));
af93e5d7 2046 }
f8523731 2047 else if (totEvents <= 1)
af93e5d7 2048 {
f8523731 2049 Log("SHUTTLE", Form("QueryRunParameters - Run %d has 1 event or less - Skipping!", run));
2050 skip = kTRUE;
af93e5d7 2051 }
89bcf3b4 2052 else
2053 {
2054 Log("SHUTTLE", Form("QueryRunParameters - Invalid parameters for Run %d: "
2055 "startTime = %d, endTime = %d. Skipping (Shuttle won't be marked as DONE)!",
2056 run, startTime, endTime));
2057 }
8e828d39 2058
2059 if (skip)
2060 {
2061 Log("SHUTTLE", Form("Marking SHUTTLE skipped for run %d", run));
2062 fLogbookEntry = entry;
2063 if (!UpdateShuttleLogbook("shuttle_skipped"))
2064 {
2065 AliError(Form("Could not update logbook for run %d !", run));
2066 }
bd23d48b 2067 if (!UpdateTableSkippedCase("ALL"))
2068 {
2069 AliError(Form("Could not update FXS tables for run %d !", run));
2070 }
8e828d39 2071 fLogbookEntry = 0;
2072 }
c9db68c7 2073
2074 delete entry;
2075 return 0;
2bb7b766 2076}
2077
2078//______________________________________________________________________________________________
a038aa70 2079TMap* AliShuttle::GetValueSet(const char* host, Int_t port, const TSeqCollection* entries,
542b6cc8 2080 DCSType type, Int_t multiSplit)
a038aa70 2081{
2082 // Retrieve all "entry" data points from the DCS server
2083 // host, port: TSocket connection parameters
2084 // entries: list of name of the alias or data point
2085 // type: kAlias or kDP
2086 // returns TMap of values, 0 when failure
542b6cc8 2087
2088 AliDCSClient client(host, port, fTimeout, fRetries, multiSplit);
b41b252a 2089
a038aa70 2090 TMap* result = 0;
b41b252a 2091 if (type == kAlias)
a038aa70 2092 {
5f3fe74e 2093 //result = client.GetAliasValues(entries, GetCurrentStartTime()-offset,
2094 // GetCurrentEndTime()+offset);
eff647d5 2095 result = client.GetAliasValues(entries, GetStartTimeDCSQuery(),
2096 GetEndTimeDCSQuery());
b41b252a 2097 }
2098 else if (type == kDP)
2099 {
5f3fe74e 2100 //result = client.GetDPValues(entries, GetCurrentStartTime()-offset,
2101 // GetCurrentEndTime()+offset);
eff647d5 2102 result = client.GetDPValues(entries, GetStartTimeDCSQuery(),
2103 GetEndTimeDCSQuery());
b41b252a 2104 }
a038aa70 2105
b41b252a 2106 if (result == 0)
2107 {
2108 Log(fCurrentDetector.Data(), Form("GetValueSet - Can't get entries! Reason: %s",
1790d4b7 2109 client.GetErrorString(client.GetResultErrorCode())));
2110 if (client.GetResultErrorCode() == AliDCSClient::fgkServerError)
2111 Log(fCurrentDetector.Data(), Form("GetValueSet - Server error code: %s",
2112 client.GetServerError().Data()));
a038aa70 2113
b41b252a 2114 return 0;
a038aa70 2115 }
b41b252a 2116
a038aa70 2117 return result;
2118}
b41b252a 2119
a038aa70 2120//______________________________________________________________________________________________
57f50b3c 2121const char* AliShuttle::GetFile(Int_t system, const char* detector,
2122 const char* id, const char* source)
b948db8d 2123{
9827400b 2124 // Get calibration file from file exchange servers
2125 // First queris the FXS database for the file name, using the run, detector, id and source info
2126 // then calls RetrieveFile(filename) for actual copy to local disk
2127 // run: current run being processed (given by Logbook entry fLogbookEntry)
2128 // detector: the Preprocessor name
2129 // id: provided as a parameter by the Preprocessor
2130 // source: provided by the Preprocessor through GetFileSources function
2131
2132 // check if test mode should simulate a FXS error
2133 if (fTestMode & kErrorFXSFiles)
2134 {
2135 Log(detector, Form("GetFile - In TESTMODE - Simulating error while connecting to %s FXS", GetSystemName(system)));
2136 return 0;
2137 }
2138
57f50b3c 2139 // check connection, in case connect
9d733021 2140 if (!Connect(system))
eba76848 2141 {
9d733021 2142 Log(detector, Form("GetFile - Couldn't connect to %s FXS database", GetSystemName(system)));
339adafa 2143 fFXSError = system;
57f50b3c 2144 return 0;
2145 }
2146
2147 // Query preparation
9d733021 2148 TString sourceName(source);
d386d623 2149 Int_t nFields = 3;
2150 TString sqlQueryStart = Form("select filePath,size,fileChecksum from %s where",
2151 fConfig->GetFXSdbTable(system));
2152 TString whereClause = Form("run=%d and detector=\"%s\" and fileId=\"%s\"",
2153 GetCurrentRun(), detector, id);
2154
9d733021 2155 if (system == kDAQ)
2156 {
d386d623 2157 whereClause += Form(" and DAQsource=\"%s\"", source);
57f50b3c 2158 }
9d733021 2159 else if (system == kDCS)
eba76848 2160 {
9d733021 2161 sourceName="none";
57f50b3c 2162 }
9d733021 2163 else if (system == kHLT)
9e080f92 2164 {
d386d623 2165 whereClause += Form(" and DDLnumbers=\"%s\"", source);
9e080f92 2166 }
2167
9e080f92 2168 TString sqlQuery = Form("%s %s", sqlQueryStart.Data(), whereClause.Data());
2169
2170 AliDebug(2, Form("SQL query: \n%s",sqlQuery.Data()));
2171
2172 // Query execution
2173 TSQLResult* aResult = 0;
9d733021 2174 aResult = dynamic_cast<TSQLResult*> (fServer[system]->Query(sqlQuery));
9e080f92 2175 if (!aResult) {
8da81210 2176 Log(detector, Form("GetFile - Can't execute SQL query to %s database for: id = %s, source = %s",
9d733021 2177 GetSystemName(system), id, sourceName.Data()));
339adafa 2178 fFXSError = system;
9e080f92 2179 return 0;
2180 }
2181
8da81210 2182 if (aResult->GetRowCount() == 0)
9e080f92 2183 {
2184 Log(detector,
8da81210 2185 Form("GetFile - No entry in %s FXS db for: id = %s, source = %s",
9d733021 2186 GetSystemName(system), id, sourceName.Data()));
9e080f92 2187 delete aResult;
2188 return 0;
2189 }
2bb7b766 2190
9e080f92 2191 if (aResult->GetRowCount() > 1) {
2192 Log(detector,
8da81210 2193 Form("GetFile - More than one entry in %s FXS db for: id = %s, source = %s",
9d733021 2194 GetSystemName(system), id, sourceName.Data()));
339adafa 2195 fFXSError = system;
9e080f92 2196 delete aResult;
2197 return 0;
2198 }
2199
9d733021 2200 if (aResult->GetFieldCount() != nFields) {
9e080f92 2201 Log(detector,
9d733021 2202 Form("GetFileName - Wrong field count in %s FXS db for: id = %s, source = %s",
2203 GetSystemName(system), id, sourceName.Data()));
339adafa 2204 fFXSError = system;
9e080f92 2205 delete aResult;
2206 return 0;
2207 }
2208
2209 TSQLRow* aRow = dynamic_cast<TSQLRow*> (aResult->Next());
2210
2211 if (!aRow){
8da81210 2212 Log(detector, Form("GetFile - Empty set result in %s FXS db from query: id = %s, source = %s",
9d733021 2213 GetSystemName(system), id, sourceName.Data()));
339adafa 2214 fFXSError = system;
9e080f92 2215 delete aResult;
2216 return 0;
2217 }
2218
2219 TString filePath(aRow->GetField(0), aRow->GetFieldLength(0));
2220 TString fileSize(aRow->GetField(1), aRow->GetFieldLength(1));
d386d623 2221 TString fileChecksum(aRow->GetField(2), aRow->GetFieldLength(2));
9e080f92 2222
2223 delete aResult;
2224 delete aRow;
2225
d386d623 2226 AliDebug(2, Form("filePath = %s; size = %s, fileChecksum = %s",
2227 filePath.Data(), fileSize.Data(), fileChecksum.Data()));
9e080f92 2228
9e080f92 2229 // retrieved file is renamed to make it unique
3ebbf3c4 2230 Int_t dir_lev1 = GetCurrentRun()/10000;
2231 TString localFileName = Form("%s/%d/%d/%s_process/%s_%s_%d_%s_%s.shuttle",
2232 GetShuttleTempDir(), dir_lev1, GetCurrentRun(), detector,
d524ade6 2233 GetSystemName(system), detector, GetCurrentRun(),
2234 id, sourceName.Data());
3ebbf3c4 2235 Log("SHUTTLE",Form("file from FXS = %s",localFileName.Data()));
9d733021 2236
9e080f92 2237
9d733021 2238 // file retrieval from FXS
4b95672b 2239 UInt_t nRetries = 0;
2240 UInt_t maxRetries = 3;
2241 Bool_t result = kFALSE;
2242
2243 // copy!! if successful TSystem::Exec returns 0
8da81210 2244 while (nRetries++ < maxRetries) {
4b95672b 2245 AliDebug(2, Form("Trying to copy file. Retry # %d", nRetries));
2246 result = RetrieveFile(system, filePath.Data(), localFileName.Data());
8da81210 2247 if (!result)
4b95672b 2248 {
8da81210 2249 Log(detector, Form("GetFile - Copy of file %s from %s FXS failed",
9d733021 2250 filePath.Data(), GetSystemName(system)));
4b95672b 2251 continue;
4f0749a8 2252 }
9e080f92 2253
c98417a1 2254 if (fileSize.Length()>0)
2255 {
2256 // compare filesize of local file with the one stored in the FXS DB
926cbc0e 2257 Long_t size = -1;
2258 Int_t sizeComp = gSystem->GetPathInfo(localFileName.Data(), 0, &size, 0, 0);
c98417a1 2259
926cbc0e 2260 if (sizeComp != 0 || size != fileSize.Atoi())
c98417a1 2261 {
8da81210 2262 Log(detector, Form("GetFile - size of file %s does not match with local copy!",
c98417a1 2263 filePath.Data()));
2264 result = kFALSE;
2265 continue;
2266 }
2267
2268 } else {
2269 Log(fCurrentDetector, Form("GetFile - size of file %s not set in %s database, skipping comparison",
2270 filePath.Data(), GetSystemName(system)));
2271 }
2272
d386d623 2273 if (fileChecksum.Length()>0)
4b95672b 2274 {
2275 // compare md5sum of local file with the one stored in the FXS DB
95d4aaef 2276 if(fileChecksum.Contains(' ')) fileChecksum.Resize(fileChecksum.First(' '));
339adafa 2277 Int_t md5Comp = gSystem->Exec(Form("md5sum %s |grep %s > /dev/null 2> /dev/null",
d524ade6 2278 localFileName.Data(), fileChecksum.Data()));
9e080f92 2279
4b95672b 2280 if (md5Comp != 0)
2281 {
8da81210 2282 Log(detector, Form("GetFile - md5sum of file %s does not match with local copy!",
4b95672b 2283 filePath.Data()));
2284 result = kFALSE;
2285 continue;
2286 }
d386d623 2287 } else {
2288 Log(fCurrentDetector, Form("GetFile - md5sum of file %s not set in %s database, skipping comparison",
2289 filePath.Data(), GetSystemName(system)));
9d733021 2290 }
4b95672b 2291 if (result) break;
9e080f92 2292 }
2293
8da81210 2294 if (!result)
2295 {
339adafa 2296 fFXSError = system;
8da81210 2297 return 0;
2298 }
4b95672b 2299
9d733021 2300 fFXSCalled[system]=kTRUE;
2301 TObjString *fileParams = new TObjString(Form("%s#!?!#%s", id, sourceName.Data()));
2302 fFXSlist[system].Add(fileParams);
9e080f92 2303
675f64cd 2304 static TString staticLocalFileName;
2305 staticLocalFileName.Form("%s", localFileName.Data());
2306
c88ad5db 2307 Log(fCurrentDetector, Form("GetFile - Retrieved file with id %s and "
2308 "source %s from %s to %s", id, source,
d524ade6 2309 GetSystemName(system), localFileName.Data()));
675f64cd 2310
d524ade6 2311 return staticLocalFileName.Data();
2bb7b766 2312}
2313
2314//______________________________________________________________________________________________
9d733021 2315Bool_t AliShuttle::RetrieveFile(UInt_t system, const char* fxsFileName, const char* localFileName)
9e080f92 2316{
9827400b 2317 //
2318 // Copies file from FXS to local Shuttle machine
2319 //
2bb7b766 2320
9e080f92 2321 // check temp directory: trying to cd to temp; if it does not exist, create it
d524ade6 2322 AliDebug(2, Form("Copy file %s from %s FXS into %s",
2323 GetSystemName(system), fxsFileName, localFileName));
2324
2325 TString tmpDir(localFileName);
2326
2327 tmpDir = tmpDir(0,tmpDir.Last('/'));
9e080f92 2328
d524ade6 2329 Int_t noDir = gSystem->GetPathInfo(tmpDir.Data(), 0, (Long64_t*) 0, 0, 0);
2330 if (noDir) // temp dir does not exists!
2331 {
2332 if (gSystem->mkdir(tmpDir.Data(), 1))
2333 {
2334 Log(fCurrentDetector.Data(), "RetrieveFile - could not make temp directory!!");
9e080f92 2335 return kFALSE;
2336 }
9e080f92 2337 }
2338
339adafa 2339 TString command = Form("scp -oPort=%d -2 %s@%s:%s/%s %s",
9d733021 2340 fConfig->GetFXSPort(system),
2341 fConfig->GetFXSUser(system),
2342 fConfig->GetFXSHost(system),
339adafa 2343 fConfig->GetFXSBaseFolder(system),
9d733021 2344 fxsFileName,
9e080f92 2345 localFileName);
2346
2347 AliDebug(2, Form("%s",command.Data()));
2348
4b95672b 2349 Bool_t result = (gSystem->Exec(command.Data()) == 0);
9e080f92 2350
4b95672b 2351 return result;
9e080f92 2352}
2353
2354//______________________________________________________________________________________________
9d733021 2355TList* AliShuttle::GetFileSources(Int_t system, const char* detector, const char* id)
2356{
9827400b 2357 //
2358 // Get sources producing the condition file Id from file exchange servers
4a33bdd9 2359 // if id is NULL all sources are returned (distinct)
9827400b 2360 //
1bcd28db 2361
8884d71e 2362 if (id)
2363 {
2364 Log(detector, Form("GetFileSources - Querying %s FXS for files with id %s produced by %s", GetSystemName(system), id, detector));
2365 } else {
2366 Log(detector, Form("GetFileSources - Querying %s FXS for files produced by %s", GetSystemName(system), detector));
2367 }
9827400b 2368
2369 // check if test mode should simulate a FXS error
2370 if (fTestMode & kErrorFXSSources)
2371 {
2372 Log(detector, Form("GetFileSources - In TESTMODE - Simulating error while connecting to %s FXS", GetSystemName(system)));
2373 return 0;
2374 }
2375
9d733021 2376 if (system == kDCS)
2377 {
c88ad5db 2378 Log(detector, "GetFileSources - WARNING: DCS system has only one source of data!");
6297b37d 2379 TList *list = new TList();
2380 list->SetOwner(1);
2381 list->Add(new TObjString(" "));
2382 return list;
9d733021 2383 }
9e080f92 2384
2385 // check connection, in case connect
9d733021 2386 if (!Connect(system))
2387 {
4a33bdd9 2388 Log(detector, Form("GetFileSources - Couldn't connect to %s FXS database", GetSystemName(system)));
339adafa 2389 fFXSError = system;
9d733021 2390 return NULL;
9e080f92 2391 }
2392
b827f957 2393 TString sourceName = "";
9d733021 2394 if (system == kDAQ)
2395 {
2396 sourceName = "DAQsource";
2397 } else if (system == kHLT)
2398 {
2399 sourceName = "DDLnumbers";
2400 }
2401
4a33bdd9 2402 TString sqlQueryStart = Form("select distinct %s from %s where", sourceName.Data(), fConfig->GetFXSdbTable(system));
2403 TString whereClause = Form("run=%d and detector=\"%s\"",
2404 GetCurrentRun(), detector);
2405 if (id)
2406 whereClause += Form(" and fileId=\"%s\"", id);
9e080f92 2407 TString sqlQuery = Form("%s %s", sqlQueryStart.Data(), whereClause.Data());
2408
2409 AliDebug(2, Form("SQL query: \n%s",sqlQuery.Data()));
2410
2411 // Query execution
2412 TSQLResult* aResult;
9d733021 2413 aResult = fServer[system]->Query(sqlQuery);
9e080f92 2414 if (!aResult) {
9d733021 2415 Log(detector, Form("GetFileSources - Can't execute SQL query to %s database for id: %s",
2416 GetSystemName(system), id));
339adafa 2417 fFXSError = system;
9e080f92 2418 return 0;
2419 }
2420
86aa42c3 2421 TList *list = new TList();
2422 list->SetOwner(1);
2423
9d733021 2424 if (aResult->GetRowCount() == 0)
2425 {
9e080f92 2426 Log(detector,
9d733021 2427 Form("GetFileSources - No entry in %s FXS table for id: %s", GetSystemName(system), id));
9e080f92 2428 delete aResult;
86aa42c3 2429 return list;
9e080f92 2430 }
2431
1bcd28db 2432 Log(detector, Form("GetFileSources - Found %d sources", aResult->GetRowCount()));
9e080f92 2433
1bcd28db 2434 TSQLRow* aRow;
9d733021 2435 while ((aRow = aResult->Next()))
2436 {
9e080f92 2437
9d733021 2438 TString source(aRow->GetField(0), aRow->GetFieldLength(0));
2439 AliDebug(2, Form("%s = %s", sourceName.Data(), source.Data()));
2440 list->Add(new TObjString(source));
9e080f92 2441 delete aRow;
2442 }
9d733021 2443
9e080f92 2444 delete aResult;
2445
2446 return list;
2bb7b766 2447}
2448
2449//______________________________________________________________________________________________
4a33bdd9 2450TList* AliShuttle::GetFileIDs(Int_t system, const char* detector, const char* source)
2451{
2452 //
2453 // Get all ids of condition files produced by a given source from file exchange servers
2454 //
2455
1bcd28db 2456 Log(detector, Form("GetFileIDs - Retrieving ids with source %s with %s", source, GetSystemName(system)));
2457
4a33bdd9 2458 // check if test mode should simulate a FXS error
2459 if (fTestMode & kErrorFXSSources)
2460 {
2461 Log(detector, Form("GetFileIDs - In TESTMODE - Simulating error while connecting to %s FXS", GetSystemName(system)));
2462 return 0;
2463 }
2464
2465 // check connection, in case connect
2466 if (!Connect(system))
2467 {
2468 Log(detector, Form("GetFileIDs - Couldn't connect to %s FXS database", GetSystemName(system)));
2469 return NULL;
2470 }
2471
b827f957 2472 TString sourceName = "";
4a33bdd9 2473 if (system == kDAQ)
2474 {
2475 sourceName = "DAQsource";
2476 } else if (system == kHLT)
2477 {
2478 sourceName = "DDLnumbers";
2479 }
2480
2481 TString sqlQueryStart = Form("select fileId from %s where", fConfig->GetFXSdbTable(system));
2482 TString whereClause = Form("run=%d and detector=\"%s\"",
2483 GetCurrentRun(), detector);
2484 if (sourceName.Length() > 0 && source)
2485 whereClause += Form(" and %s=\"%s\"", sourceName.Data(), source);
2486 TString sqlQuery = Form("%s %s", sqlQueryStart.Data(), whereClause.Data());
2487
2488 AliDebug(2, Form("SQL query: \n%s",sqlQuery.Data()));
2489
2490 // Query execution
2491 TSQLResult* aResult;
2492 aResult = fServer[system]->Query(sqlQuery);
2493 if (!aResult) {
2494 Log(detector, Form("GetFileIDs - Can't execute SQL query to %s database for source: %s",
2495 GetSystemName(system), source));
2496 return 0;
2497 }
2498
2499 TList *list = new TList();
2500 list->SetOwner(1);
2501
2502 if (aResult->GetRowCount() == 0)
2503 {
2504 Log(detector,
2505 Form("GetFileIDs - No entry in %s FXS table for source: %s", GetSystemName(system), source));
2506 delete aResult;
2507 return list;
2508 }
2509
1bcd28db 2510 Log(detector, Form("GetFileIDs - Found %d ids", aResult->GetRowCount()));
2511
4a33bdd9 2512 TSQLRow* aRow;
2513
2514 while ((aRow = aResult->Next()))
2515 {
2516
2517 TString id(aRow->GetField(0), aRow->GetFieldLength(0));
2518 AliDebug(2, Form("fileId = %s", id.Data()));
2519 list->Add(new TObjString(id));
2520 delete aRow;
2521 }
2522
2523 delete aResult;
2524
2525 return list;
2526}
2527
2528//______________________________________________________________________________________________
9d733021 2529Bool_t AliShuttle::Connect(Int_t system)
2bb7b766 2530{
9827400b 2531 // Connect to MySQL Server of the system's FXS MySQL databases
2532 // DAQ Logbook, Shuttle Logbook and DAQ FXS db are on the same host
2533 //
57f50b3c 2534
9d733021 2535 // check connection: if already connected return
dd84c76d 2536
3a4ddf0e 2537 if(fServer[system] && fServer[system]->IsConnected()) {
dd84c76d 2538 // ping the server
2539 if (fServer[system]->PingVerify()==kTRUE){ // connection is still alive
2540 return kTRUE;
2541 }
2542 else{
2543 AliWarning(Form("Connection got lost to FXS database for %s. Closing and reconnecting.",
2544 AliShuttleInterface::GetSystemName(system)));
2545 fServer[system]->Close();
2546 delete fServer[system];
2547 fServer[system] = 0x0;
2548 }
3a4ddf0e 2549 }
57f50b3c 2550
9d733021 2551 TString dbHost, dbUser, dbPass, dbName;
57f50b3c 2552
9d733021 2553 if (system < 3) // FXS db servers
2554 {
2555 dbHost = Form("mysql://%s:%d", fConfig->GetFXSdbHost(system), fConfig->GetFXSdbPort(system));
2556 dbUser = fConfig->GetFXSdbUser(system);
2557 dbPass = fConfig->GetFXSdbPass(system);
2558 dbName = fConfig->GetFXSdbName(system);
2559 } else { // Run & Shuttle logbook servers
2560 // TODO Will the Shuttle logbook server be the same as the Run logbook server ???
2561 dbHost = Form("mysql://%s:%d", fConfig->GetDAQlbHost(), fConfig->GetDAQlbPort());
2562 dbUser = fConfig->GetDAQlbUser();
2563 dbPass = fConfig->GetDAQlbPass();
2564 dbName = fConfig->GetDAQlbDB();
2565 }
57f50b3c 2566
9d733021 2567 fServer[system] = TSQLServer::Connect(dbHost.Data(), dbUser.Data(), dbPass.Data());
2568 if (!fServer[system] || !fServer[system]->IsConnected()) {
2569 if(system < 3)
2570 {
2571 AliError(Form("Can't establish connection to FXS database for %s",
2572 AliShuttleInterface::GetSystemName(system)));
2573 } else {
2574 AliError("Can't establish connection to Run logbook.");
57f50b3c 2575 }
9d733021 2576 if(fServer[system]) delete fServer[system];
2577 return kFALSE;
2bb7b766 2578 }
57f50b3c 2579
9d733021 2580 // Get tables
2581 TSQLResult* aResult=0;
2582 switch(system){
2583 case kDAQ:
2584 aResult = fServer[kDAQ]->GetTables(dbName.Data());
2585 break;
2586 case kDCS:
2587 aResult = fServer[kDCS]->GetTables(dbName.Data());
2588 break;
2589 case kHLT:
2590 aResult = fServer[kHLT]->GetTables(dbName.Data());
2591 break;
2592 default:
2593 aResult = fServer[3]->GetTables(dbName.Data());
2594 break;
2595 }
2596
2597 delete aResult;
2bb7b766 2598 return kTRUE;
2599}
57f50b3c 2600
9e080f92 2601//______________________________________________________________________________________________
9d733021 2602Bool_t AliShuttle::UpdateTable()
9e080f92 2603{
9827400b 2604 //
2605 // Update FXS table filling time_processed field in all rows corresponding to current run and detector
2606 //
9e080f92 2607
9d733021 2608 Bool_t result = kTRUE;
9e080f92 2609
9d733021 2610 for (UInt_t system=0; system<3; system++)
2611 {
2612 if(!fFXSCalled[system]) continue;
9e080f92 2613
9d733021 2614 // check connection, in case connect
2615 if (!Connect(system))
2616 {
2617 Log(fCurrentDetector, Form("UpdateTable - Couldn't connect to %s FXS database", GetSystemName(system)));
2618 result = kFALSE;
2619 continue;
9e080f92 2620 }
9e080f92 2621
9d733021 2622 TTimeStamp now; // now
2623
2624 // Loop on FXS list entries
2625 TIter iter(&fFXSlist[system]);
2626 TObjString *aFXSentry=0;
2627 while ((aFXSentry = dynamic_cast<TObjString*> (iter.Next())))
2628 {
2629 TString aFXSentrystr = aFXSentry->String();
2630 TObjArray *aFXSarray = aFXSentrystr.Tokenize("#!?!#");
2631 if (!aFXSarray || aFXSarray->GetEntries() != 2 )
2632 {
2633 Log(fCurrentDetector, Form("UpdateTable - error updating %s FXS entry. Check string: <%s>",
2634 GetSystemName(system), aFXSentrystr.Data()));
2635 if(aFXSarray) delete aFXSarray;
2636 result = kFALSE;
2637 continue;
2638 }
2639 const char* fileId = ((TObjString*) aFXSarray->At(0))->GetName();
2640 const char* source = ((TObjString*) aFXSarray->At(1))->GetName();
2641
2642 TString whereClause;
2643 if (system == kDAQ)
2644 {
2645 whereClause = Form("where run=%d and detector=\"%s\" and fileId=\"%s\" and DAQsource=\"%s\";",
2646 GetCurrentRun(), fCurrentDetector.Data(), fileId, source);
2647 }
2648 else if (system == kDCS)
2649 {
2650 whereClause = Form("where run=%d and detector=\"%s\" and fileId=\"%s\";",
2651 GetCurrentRun(), fCurrentDetector.Data(), fileId);
2652 }
2653 else if (system == kHLT)
2654 {
2655 whereClause = Form("where run=%d and detector=\"%s\" and fileId=\"%s\" and DDLnumbers=\"%s\";",
2656 GetCurrentRun(), fCurrentDetector.Data(), fileId, source);
2657 }
2658
2659 delete aFXSarray;
9e080f92 2660
9d733021 2661 TString sqlQuery = Form("update %s set time_processed=%d %s", fConfig->GetFXSdbTable(system),
2662 now.GetSec(), whereClause.Data());
9e080f92 2663
9d733021 2664 AliDebug(2, Form("SQL query: \n%s",sqlQuery.Data()));
9e080f92 2665
9d733021 2666 // Query execution
2667 TSQLResult* aResult;
2668 aResult = dynamic_cast<TSQLResult*> (fServer[system]->Query(sqlQuery));
2669 if (!aResult)
2670 {
2671 Log(fCurrentDetector, Form("UpdateTable - %s db: can't execute SQL query <%s>",
2672 GetSystemName(system), sqlQuery.Data()));
2673 result = kFALSE;
2674 continue;
2675 }
2676 delete aResult;
9e080f92 2677 }
9e080f92 2678 }
2679
9d733021 2680 return result;
9e080f92 2681}
57f50b3c 2682
bd23d48b 2683//_______________________________________________________________________________
2684Bool_t AliShuttle::UpdateTableSkippedCase(const char* detector)
2685{
2686 //
2687 // Update FXS table filling time_processed field in all rows corresponding to current run and detector
2688 // if detector = "ALL" update all detectors
2689 //
2690
2691 Bool_t result = kTRUE;
2692
7a782903 2693 TString detName(detector);
2694
bd23d48b 2695 for (UInt_t system=0; system<3; system++)
2696 {
2697
2698 // check connection, in case connect
2699 if (!Connect(system))
2700 {
2701 Log(fCurrentDetector, Form("UpdateTableSkippedCase - Couldn't connect to %s FXS database", GetSystemName(system)));
2702 result = kFALSE;
2703 continue;
2704 }
2705
2706 TTimeStamp now; // now
2707
2708 // Loop on FXS list entries
2709 TIter iter(&fFXSlist[system]);
2710
2711 TString whereClause;
7a782903 2712 if (detName == "ALL") whereClause = Form("where run=%d and time_processed IS NULL;",GetCurrentRun());
bd23d48b 2713 else whereClause = Form("where run=%d and detector=\"%s\" and time_processed IS NULL;",GetCurrentRun(), detector);
2714
4be2f262 2715 //Log("SHUTTLE",Form(" whereClause = %s ",whereClause.Data()));
bd23d48b 2716
2717 TString sqlQuery = Form("update %s set time_processed=%d %s", fConfig->GetFXSdbTable(system),
2718 now.GetSec(), whereClause.Data());
2719
2720 AliDebug(2, Form("SQL query: \n%s",sqlQuery.Data()));
2721
2722 // Query execution
2723 TSQLResult* aResult;
2724 aResult = dynamic_cast<TSQLResult*> (fServer[system]->Query(sqlQuery));
2725 if (!aResult)
2726 {
2727 Log("SHUTTLE", Form("UpdateTableSkippedCase - %s db: can't execute SQL query <%s>",
2728 GetSystemName(system), sqlQuery.Data()));
2729 result = kFALSE;
2730 continue;
2731 }
2732 delete aResult;
2733
2734 }
2735
2736 return result;
2737}
2bb7b766 2738//______________________________________________________________________________________________
3301427a 2739Bool_t AliShuttle::UpdateTableFailCase()
2740{
9827400b 2741 // Update FXS table filling time_processed field in all rows corresponding to current run and detector
2742 // this is called in case the preprocessor is declared failed for the current run, because
2743 // the fields are updated only in case of success
3301427a 2744
2745 Bool_t result = kTRUE;
2746
2747 for (UInt_t system=0; system<3; system++)
2748 {
2749 // check connection, in case connect
2750 if (!Connect(system))
2751 {
2752 Log(fCurrentDetector, Form("UpdateTableFailCase - Couldn't connect to %s FXS database",
2753 GetSystemName(system)));
2754 result = kFALSE;
2755 continue;
2756 }
2757
2758 TTimeStamp now; // now
2759
2760 // Loop on FXS list entries
2761
2762 TString whereClause = Form("where run=%d and detector=\"%s\";",
2763 GetCurrentRun(), fCurrentDetector.Data());
2764
2765
2766 TString sqlQuery = Form("update %s set time_processed=%d %s", fConfig->GetFXSdbTable(system),
2767 now.GetSec(), whereClause.Data());
2768
2769 AliDebug(2, Form("SQL query: \n%s",sqlQuery.Data()));
2770
2771 // Query execution
2772 TSQLResult* aResult;
2773 aResult = dynamic_cast<TSQLResult*> (fServer[system]->Query(sqlQuery));
2774 if (!aResult)
2775 {
2776 Log(fCurrentDetector, Form("UpdateTableFailCase - %s db: can't execute SQL query <%s>",
2777 GetSystemName(system), sqlQuery.Data()));
2778 result = kFALSE;
2779 continue;
2780 }
2781 delete aResult;
2782 }
2783
2784 return result;
2785}
2786
2787//______________________________________________________________________________________________
2bb7b766 2788Bool_t AliShuttle::UpdateShuttleLogbook(const char* detector, const char* status)
2789{
e7f62f16 2790 //
2791 // Update Shuttle logbook filling detector or shuttle_done column
2792 // ex. of usage: UpdateShuttleLogbook("PHOS", "DONE") or UpdateShuttleLogbook("shuttle_done")
2793 //
57f50b3c 2794
2bb7b766 2795 // check connection, in case connect
be48e3ea 2796 if(!Connect(3)){
2bb7b766 2797 Log("SHUTTLE", "UpdateShuttleLogbook - Couldn't connect to DAQ Logbook.");
2798 return kFALSE;
57f50b3c 2799 }
2800
2bb7b766 2801 TString detName(detector);
2802 TString setClause;
8e828d39 2803 if (detName == "shuttle_done" || detName == "shuttle_skipped")
e7f62f16 2804 {
2bb7b766 2805 setClause = "set shuttle_done=1";
8e828d39 2806
ee6f7523 2807 if (detName == "shuttle_done")
b0e53b15 2808 {
8e828d39 2809 if (TouchFile() != kTRUE)
2156c755 2810 {
2811 SendMLRunInfo("Pending");
503ff24f 2812 return kFALSE;
2156c755 2813 }
8e828d39 2814
2815 SendMLRunInfo("Done");
b0e53b15 2816 }
8e828d39 2817 else
2818 SendMLRunInfo("Skipped");
2819 }
2820 else {
2bb7b766 2821 TString statusStr(status);
2822 if(statusStr.Contains("done", TString::kIgnoreCase) ||
2823 statusStr.Contains("failed", TString::kIgnoreCase)){
eba76848 2824 setClause = Form("set %s=\"%s\"", detector, status);
2bb7b766 2825 } else {
2826 Log("SHUTTLE",
2827 Form("UpdateShuttleLogbook - Invalid status <%s> for detector %s",
2828 status, detector));
2829 return kFALSE;
2830 }
2831 }
57f50b3c 2832
2bb7b766 2833 TString whereClause = Form("where run=%d", GetCurrentRun());
2834
441b0e9c 2835 TString sqlQuery = Form("update %s %s %s",
2836 fConfig->GetShuttlelbTable(), setClause.Data(), whereClause.Data());
57f50b3c 2837
2bb7b766 2838 AliDebug(2, Form("SQL query: \n%s",sqlQuery.Data()));
2839
2840 // Query execution
2841 TSQLResult* aResult;
be48e3ea 2842 aResult = dynamic_cast<TSQLResult*> (fServer[3]->Query(sqlQuery));
2bb7b766 2843 if (!aResult) {
2844 Log("SHUTTLE", Form("UpdateShuttleLogbook - Can't execute query <%s>", sqlQuery.Data()));
2845 return kFALSE;
57f50b3c 2846 }
2bb7b766 2847 delete aResult;
57f50b3c 2848
2849 return kTRUE;
2850}
2851
2852//______________________________________________________________________________________________
2bb7b766 2853Int_t AliShuttle::GetCurrentRun() const
2854{
9827400b 2855 //
2856 // Get current run from logbook entry
2857 //
57f50b3c 2858
2bb7b766 2859 return fLogbookEntry ? fLogbookEntry->GetRun() : -1;
57f50b3c 2860}
2861
2862//______________________________________________________________________________________________
2bb7b766 2863UInt_t AliShuttle::GetCurrentStartTime() const
2864{
9827400b 2865 //
2866 // get current start time
2867 //
57f50b3c 2868
2bb7b766 2869 return fLogbookEntry ? fLogbookEntry->GetStartTime() : 0;
57f50b3c 2870}
2871
2872//______________________________________________________________________________________________
2bb7b766 2873UInt_t AliShuttle::GetCurrentEndTime() const
2874{
9827400b 2875 //
2876 // get current end time from logbook entry
2877 //
57f50b3c 2878
2bb7b766 2879 return fLogbookEntry ? fLogbookEntry->GetEndTime() : 0;
57f50b3c 2880}
5f3fe74e 2881//______________________________________________________________________________________________
675f64cd 2882UInt_t AliShuttle::GetCurrentYear() const
2883{
2884 //
2885 // Get current year from logbook entry
2886 //
2887
2888 if (!fLogbookEntry) return 0;
2889
2890 TTimeStamp startTime(GetCurrentStartTime());
2891 TString year = Form("%d",startTime.GetDate());
2892 year = year(0,4);
2893
2894 return year.Atoi();
2895}
2896
2897//______________________________________________________________________________________________
2898const char* AliShuttle::GetLHCPeriod() const
2899{
2900 //
2901 // Get current LHC period from logbook entry
2902 //
2903
2904 if (!fLogbookEntry) return 0;
2905
2906 return fLogbookEntry->GetRunParameter("LHCperiod");
2907}
2908
2909//______________________________________________________________________________________________
b948db8d 2910void AliShuttle::Log(const char* detector, const char* message)
2911{
9827400b 2912 //
2913 // Fill log string with a message
2914 //
7d4cf768 2915
3ebbf3c4 2916 TString logRunDir = GetShuttleLogDir();
2917 if (GetCurrentRun() >=0) {
2918 Int_t logDir_lev1 = GetCurrentRun()/10000;
2919 logRunDir += Form("/%d/%d", logDir_lev1, GetCurrentRun());
2920 }
7d4cf768 2921 void* dir = gSystem->OpenDirectory(logRunDir.Data());
84090f85 2922 if (dir == NULL) {
7d4cf768 2923 if (gSystem->mkdir(logRunDir.Data(), kTRUE)) {
36c99a6a 2924 AliError(Form("Can't open directory <%s>", GetShuttleLogDir()));
84090f85 2925 return;
2926 }
b948db8d 2927
84090f85 2928 } else {
2929 gSystem->FreeDirectory(dir);
2930 }
b948db8d 2931
0b025a52 2932 TString toLog = Form("%s UTC (%d): %s - ", TTimeStamp(time(0)).AsString("s"), getpid(), detector);
e7f62f16 2933 if (GetCurrentRun() >= 0)
2934 toLog += Form("run %d - ", GetCurrentRun());
2bb7b766 2935 toLog += Form("%s", message);
2936
84090f85 2937 AliInfo(toLog.Data());
ffa29e93 2938
2939 // if we redirect the log output already to the file, leave here
2940 if (fOutputRedirected && strcmp(detector, "SHUTTLE") != 0)
2941 return;
b948db8d 2942
ffa29e93 2943 TString fileName = GetLogFileName(detector);
e7f62f16 2944
84090f85 2945 gSystem->ExpandPathName(fileName);
2946
2947 ofstream logFile;
2948 logFile.open(fileName, ofstream::out | ofstream::app);
2949
2950 if (!logFile.is_open()) {
2951 AliError(Form("Could not open file %s", fileName.Data()));
2952 return;
2953 }
7bfb2090 2954
84090f85 2955 logFile << toLog.Data() << "\n";
b948db8d 2956
84090f85 2957 logFile.close();
b948db8d 2958}
2bb7b766 2959
2bb7b766 2960//______________________________________________________________________________________________
ffa29e93 2961TString AliShuttle::GetLogFileName(const char* detector) const
2962{
2963 //
2964 // returns the name of the log file for a given sub detector
2965 //
2966
2967 TString fileName;
2968
2969 if (GetCurrentRun() >= 0)
7d4cf768 2970 {
3ebbf3c4 2971 Int_t logDir_lev1 = GetCurrentRun()/10000;
2972 fileName.Form("%s/%d/%d/%s.log", GetShuttleLogDir(), logDir_lev1, GetCurrentRun(),
2973 detector);
7d4cf768 2974 } else {
ffa29e93 2975 fileName.Form("%s/%s.log", GetShuttleLogDir(), detector);
7d4cf768 2976 }
ffa29e93 2977
2978 return fileName;
2979}
2980
2981//______________________________________________________________________________________________
ee6f7523 2982void AliShuttle::SendAlive()
2983{
2984 // sends alive message to ML
2985
2986 TMonaLisaText mlStatus("SHUTTLE_status", "Alive");
2987
2988 TList mlList;
2989 mlList.Add(&mlStatus);
2990
2991 fMonaLisa->SendParameters(&mlList, "__PROCESSINGINFO__");
2992}
2993
2994//______________________________________________________________________________________________
2bb7b766 2995Bool_t AliShuttle::Collect(Int_t run)
2996{
9827400b 2997 //
2998 // Collects conditions data for all UNPROCESSED run written to DAQ LogBook in case of run = -1 (default)
2999 // If a dedicated run is given this run is processed
3000 //
3001 // In operational mode, this is the Shuttle function triggered by the EOR signal.
3002 //
2bb7b766 3003
eba76848 3004 if (run == -1)
3005 Log("SHUTTLE","Collect - Shuttle called. Collecting conditions data for unprocessed runs");
3006 else
3007 Log("SHUTTLE", Form("Collect - Shuttle called. Collecting conditions data for run %d", run));
cb343cfd 3008
3009 SetLastAction("Starting");
2bb7b766 3010
ee6f7523 3011 // create ML instance
3012 if (!fMonaLisa)
3013 fMonaLisa = new TMonaLisaWriter(fConfig->GetMonitorHost(), fConfig->GetMonitorTable());
3014
fbc112e3 3015 CountOpenRuns();
ee6f7523 3016
2bb7b766 3017 TString whereClause("where shuttle_done=0");
eba76848 3018 if (run != -1)
3019 whereClause += Form(" and run=%d", run);
2bb7b766 3020
3021 TObjArray shuttleLogbookEntries;
be48e3ea 3022 if (!QueryShuttleLogbook(whereClause, shuttleLogbookEntries))
3023 {
cb343cfd 3024 Log("SHUTTLE", "Collect - Can't retrieve entries from Shuttle logbook");
2bb7b766 3025 return kFALSE;
3026 }
3027
9e080f92 3028 if (shuttleLogbookEntries.GetEntries() == 0)
3029 {
3030 if (run == -1)
3031 Log("SHUTTLE","Collect - Found no UNPROCESSED runs in Shuttle logbook");
3032 else
3033 Log("SHUTTLE", Form("Collect - Run %d is already DONE "
3034 "or it does not exist in Shuttle logbook", run));
3035 return kTRUE;
3036 }
3037
be48e3ea 3038 for (UInt_t iDet=0; iDet<NDetectors(); iDet++)
3039 fFirstUnprocessed[iDet] = kTRUE;
3040
fc5a4708 3041 if (run != -1)
be48e3ea 3042 {
3043 // query Shuttle logbook for earlier runs, check if some detectors are unprocessed,
3044 // flag them into fFirstUnprocessed array
e3ae8d33 3045 TString whereClauseBis(Form("where shuttle_done=0 and run < %d", run));
be48e3ea 3046 TObjArray tmpLogbookEntries;
e3ae8d33 3047 if (!QueryShuttleLogbook(whereClauseBis, tmpLogbookEntries))
be48e3ea 3048 {
3049 Log("SHUTTLE", "Collect - Can't retrieve entries from Shuttle logbook");
3050 return kFALSE;
3051 }
3052
3053 TIter iter(&tmpLogbookEntries);
3054 AliShuttleLogbookEntry* anEntry = 0;
3055 while ((anEntry = dynamic_cast<AliShuttleLogbookEntry*> (iter.Next())))
3056 {
3057 for (UInt_t iDet=0; iDet<NDetectors(); iDet++)
3058 {
3059 if (anEntry->GetDetectorStatus(iDet) == AliShuttleLogbookEntry::kUnprocessed)
3060 {
3061 AliDebug(2, Form("Run %d: setting %s as \"not first time unprocessed\"",
3062 anEntry->GetRun(), GetDetName(iDet)));
3063 fFirstUnprocessed[iDet] = kFALSE;
3064 }
3065 }
3066
3067 }
3068
3069 }
3070
3071 if (!RetrieveConditionsData(shuttleLogbookEntries))
3072 {
cb343cfd 3073 Log("SHUTTLE", "Collect - Process of at least one run failed");
a3ca69fe 3074 CountOpenRuns();
2bb7b766 3075 return kFALSE;
3076 }
3077
36c99a6a 3078 Log("SHUTTLE", "Collect - Requested run(s) successfully processed");
a3ca69fe 3079 CountOpenRuns();
eba76848 3080 return kTRUE;
2bb7b766 3081}
3082
2bb7b766 3083//______________________________________________________________________________________________
3084Bool_t AliShuttle::RetrieveConditionsData(const TObjArray& dateEntries)
3085{
9827400b 3086 //
3087 // Retrieve conditions data for all runs that aren't processed yet
3088 //
2bb7b766 3089
3090 Bool_t hasError = kFALSE;
3091
3092 TIter iter(&dateEntries);
3093 AliShuttleLogbookEntry* anEntry;
3094
3095 while ((anEntry = (AliShuttleLogbookEntry*) iter.Next())){
3096 if (!Process(anEntry)){
3097 hasError = kTRUE;
3098 }
4b95672b 3099
3100 // clean SHUTTLE temp directory
d524ade6 3101 //TString filename = Form("%s/*.shu