]> git.uio.no Git - u/mrichter/AliRoot.git/blame - SHUTTLE/AliShuttle.cxx
better just warning than error in case of mirrors requesting when putting to local...
[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 !!!!! **********");
efec19bd 99 for(int iSys=0;iSys<5;iSys++) {
57f50b3c 100 fServer[iSys]=0;
efec19bd 101 if (iSys < 4)
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();
efec19bd 120 for(int iSys=0;iSys<5;iSys++)
57f50b3c 121 if(fServer[iSys]) {
122 fServer[iSys]->Close();
123 delete fServer[iSys];
eba76848 124 fServer[iSys] = 0;
57f50b3c 125 }
2bb7b766 126
127 if (fStatusEntry){
128 delete fStatusEntry;
129 fStatusEntry = 0;
130 }
cb343cfd 131
132 if (fMonitoringMutex)
133 {
134 delete fMonitoringMutex;
135 fMonitoringMutex = 0;
136 }
73abe331 137}
138
b948db8d 139//______________________________________________________________________________________________
57f50b3c 140void AliShuttle::RegisterPreprocessor(AliPreprocessor* preprocessor)
58bc3020 141{
73abe331 142 //
b948db8d 143 // Registers new AliPreprocessor.
73abe331 144 // It uses GetName() for indentificator of the pre processor.
145 // The pre processor is registered it there isn't any other
146 // with the same identificator (GetName()).
147 //
148
eba76848 149 const char* detName = preprocessor->GetName();
150 if(GetDetPos(detName) < 0)
151 AliFatal(Form("********** !!!!! Invalid detector name: %s !!!!! **********", detName));
152
153 if (fPreprocessorMap.GetValue(detName)) {
154 AliWarning(Form("AliPreprocessor %s is already registered!", detName));
73abe331 155 return;
156 }
157
eba76848 158 fPreprocessorMap.Add(new TObjString(detName), preprocessor);
73abe331 159}
b948db8d 160//______________________________________________________________________________________________
3301427a 161Bool_t AliShuttle::Store(const AliCDBPath& path, TObject* object,
84090f85 162 AliCDBMetaData* metaData, Int_t validityStart, Bool_t validityInfinite)
73abe331 163{
9827400b 164 // Stores a CDB object in the storage for offline reconstruction. Objects that are not needed for
165 // offline reconstruction, but should be stored anyway (e.g. for debugging) should NOT be stored
166 // using this function. Use StoreReferenceData instead!
167 // It calls StoreLocally function which temporarily stores the data locally; when the preprocessor
168 // finishes the data are transferred to the main storage (Grid).
b948db8d 169
3301427a 170 return StoreLocally(fgkLocalCDB, path, object, metaData, validityStart, validityInfinite);
84090f85 171}
172
173//______________________________________________________________________________________________
3301427a 174Bool_t AliShuttle::StoreReferenceData(const AliCDBPath& path, TObject* object, AliCDBMetaData* metaData)
84090f85 175{
9827400b 176 // Stores a CDB object in the storage for reference data. This objects will not be available during
177 // offline reconstrunction. Use this function for reference data only!
178 // It calls StoreLocally function which temporarily stores the data locally; when the preprocessor
179 // finishes the data are transferred to the main storage (Grid).
85a80aa9 180
3301427a 181 return StoreLocally(fgkLocalRefStorage, path, object, metaData);
85a80aa9 182}
183
184//______________________________________________________________________________________________
3301427a 185Bool_t AliShuttle::StoreLocally(const TString& localUri,
85a80aa9 186 const AliCDBPath& path, TObject* object, AliCDBMetaData* metaData,
187 Int_t validityStart, Bool_t validityInfinite)
188{
9827400b 189 // Store object temporarily in local storage. Parameters are passed by Store and StoreReferenceData functions.
190 // when the preprocessor finishes the data are transferred to the main storage (Grid).
191 // The parameters are:
192 // 1) Uri of the backup storage (Local)
193 // 2) the object's path.
194 // 3) the object to be stored
195 // 4) the metaData to be associated with the object
196 // 5) the validity start run number w.r.t. the current run,
197 // if the data is valid only for this run leave the default 0
198 // 6) specifies if the calibration data is valid for infinity (this means until updated),
199 // typical for calibration runs, the default is kFALSE
200 //
201 // returns 0 if fail, 1 otherwise
84090f85 202
924fb6e9 203
9827400b 204 if (fTestMode & kErrorStorage)
205 {
206 Log(fCurrentDetector, "StoreLocally - In TESTMODE - Simulating error while storing locally");
207 return kFALSE;
208 }
209
3301427a 210 const char* cdbType = (localUri == fgkLocalCDB) ? "CDB" : "Reference";
2bb7b766 211
85a80aa9 212 Int_t firstRun = GetCurrentRun() - validityStart;
84090f85 213 if(firstRun < 0) {
9827400b 214 AliWarning("First valid run happens to be less than 0! Setting it to 0.");
84090f85 215 firstRun=0;
216 }
217
218 Int_t lastRun = -1;
219 if(validityInfinite) {
220 lastRun = AliCDBRunRange::Infinity();
221 } else {
222 lastRun = GetCurrentRun();
223 }
224
3301427a 225 // Version is set to current run, it will be used later to transfer data to Grid
226 AliCDBId id(path, firstRun, lastRun, GetCurrentRun(), -1);
2bb7b766 227
228 if(! dynamic_cast<TObjString*> (metaData->GetProperty("RunUsed(TObjString)"))){
229 TObjString runUsed = Form("%d", GetCurrentRun());
9e080f92 230 metaData->SetProperty("RunUsed(TObjString)", runUsed.Clone());
2bb7b766 231 }
84090f85 232
3301427a 233 Bool_t result = kFALSE;
84090f85 234
3301427a 235 if (!(AliCDBManager::Instance()->GetStorage(localUri))) {
236 Log("SHUTTLE", Form("StoreLocally - Cannot activate local %s storage", cdbType));
84090f85 237 } else {
924fb6e9 238 Int_t logLevel = AliLog::GetGlobalLogLevel();
239 AliLog::SetGlobalLogLevel(AliLog::kError);
3301427a 240 result = AliCDBManager::Instance()->GetStorage(localUri)
84090f85 241 ->Put(object, id, metaData);
924fb6e9 242 AliLog::SetGlobalLogLevel((AliLog::EType_t)logLevel);
84090f85 243 }
244
245 if(!result) {
246
9827400b 247 Log(fCurrentDetector, Form("StoreLocally - Can't store object <%s>!", id.ToString().Data()));
3301427a 248 }
2bb7b766 249
924fb6e9 250
3301427a 251 return result;
252}
84090f85 253
3301427a 254//______________________________________________________________________________________________
255Bool_t AliShuttle::StoreOCDB()
256{
9827400b 257 //
258 // Called when preprocessor ends successfully or when previous storage attempt failed (kStoreError status)
259 // Calls underlying StoreOCDB(const char*) function twice, for OCDB and Reference storage.
260 // Then calls StoreRefFilesToGrid to store reference files.
261 //
262
5e993b6f 263 UpdateShuttleStatus(AliShuttleStatus::kStoreStarted);
56d3e9e4 264
9827400b 265 if (fTestMode & kErrorGrid)
266 {
267 Log("SHUTTLE", "StoreOCDB - In TESTMODE - Simulating error while storing in the Grid");
268 Log(fCurrentDetector, "StoreOCDB - In TESTMODE - Simulating error while storing in the Grid");
269 return kFALSE;
270 }
271
c88ad5db 272 Log("SHUTTLE","StoreOCDB - Storing OCDB data ...");
5e993b6f 273 Int_t resultCDB = StoreOCDB(fgkMainCDB);
86aa42c3 274
c88ad5db 275 Log("SHUTTLE","StoreOCDB - Storing reference data ...");
5e993b6f 276 Int_t resultRef = StoreOCDB(fgkMainRefStorage);
9827400b 277
c88ad5db 278 Log("SHUTTLE","StoreOCDB - Storing reference files ...");
279 Bool_t resultRefFiles = CopyFilesToGrid("reference");
280
281 Bool_t resultMetadata = kTRUE;
282 if(fCurrentDetector == "GRP")
283 {
7802e884 284 Log("SHUTTLE","StoreOCDB - Storing Run Metadata file ...");
c88ad5db 285 resultMetadata = CopyFilesToGrid("metadata");
286 }
9827400b 287
5e993b6f 288 Int_t storeResult = 0;
289
290 if (resultCDB < 0 || resultRef < 0 || resultRefFiles == kFALSE || resultMetadata == kFALSE)
291 storeResult = -1;
292 else if (resultCDB > 0 || resultRef > 0)
293 storeResult = 1;
294
295 if (storeResult < 0)
296 {
297 Log("SHUTTLE",
298 Form("\t\t\t****** run %d - %s: STORAGE ERROR ******",
299 GetCurrentRun(), fCurrentDetector.Data()));
300 UpdateShuttleStatus(AliShuttleStatus::kStoreError);
301 }
302 else if (storeResult > 0)
303 {
304 Log("SHUTTLE",
305 Form("\t\t\t****** run %d - %s: STORAGE DELAYED ******",
306 GetCurrentRun(), fCurrentDetector.Data()));
307 UpdateShuttleStatus(AliShuttleStatus::kStoreDelayed);
308 }
309 else if (storeResult == 0)
310 {
311 Log("SHUTTLE",
312 Form("\t\t\t****** run %d - %s: DONE ******",
313 GetCurrentRun(), fCurrentDetector.Data()));
314 UpdateShuttleStatus(AliShuttleStatus::kDone);
315 UpdateShuttleLogbook(fCurrentDetector, "DONE");
316 }
317
318 return (storeResult == 0);
3301427a 319}
320
321//______________________________________________________________________________________________
5e993b6f 322Int_t AliShuttle::StoreOCDB(const TString& gridURI)
3301427a 323{
324 //
325 // Called by StoreOCDB(), performs actual storage to the main OCDB and reference storages (Grid)
326 //
5e993b6f 327 // Return code:
328 // -2 initialization error
329 // -1 storage error
330 // 0 success
331 // 1 storage delayed (e.g. previous unprocessed runs)
332 //
3301427a 333
334 TObjArray* gridIds=0;
335
336 Bool_t result = kTRUE;
5e993b6f 337 Bool_t delayed = kFALSE;
3301427a 338
339 const char* type = 0;
340 TString localURI;
341 if(gridURI == fgkMainCDB) {
342 type = "OCDB";
343 localURI = fgkLocalCDB;
344 } else if(gridURI == fgkMainRefStorage) {
345 type = "reference";
346 localURI = fgkLocalRefStorage;
347 } else {
348 AliError(Form("Invalid storage URI: %s", gridURI.Data()));
5e993b6f 349 return -2;
3301427a 350 }
351
352 AliCDBManager* man = AliCDBManager::Instance();
353
354 AliCDBStorage *gridSto = man->GetStorage(gridURI);
355 if(!gridSto) {
356 Log("SHUTTLE",
357 Form("StoreOCDB - cannot activate main %s storage", type));
5e993b6f 358 return -2;
3301427a 359 }
360
361 gridIds = gridSto->GetQueryCDBList();
362
363 // get objects previously stored in local CDB
364 AliCDBStorage *localSto = man->GetStorage(localURI);
365 if(!localSto) {
366 Log("SHUTTLE",
367 Form("StoreOCDB - cannot activate local %s storage", type));
5e993b6f 368 return -2;
3301427a 369 }
370 AliCDBPath aPath(GetOfflineDetName(fCurrentDetector.Data()),"*","*");
371 // Local objects were stored with current run as Grid version!
372 TList* localEntries = localSto->GetAll(aPath.GetPath(), GetCurrentRun(), GetCurrentRun());
373 localEntries->SetOwner(1);
374
375 // loop on local stored objects
376 TIter localIter(localEntries);
377 AliCDBEntry *aLocEntry = 0;
378 while((aLocEntry = dynamic_cast<AliCDBEntry*> (localIter.Next()))){
379 aLocEntry->SetOwner(1);
380 AliCDBId aLocId = aLocEntry->GetId();
381 aLocEntry->SetVersion(-1);
382 aLocEntry->SetSubVersion(-1);
6c0c9948 383
384 Log(fCurrentDetector.Data(), Form("Attempting to store %s", aLocId.ToString().Data()));
3301427a 385
386 // If local object is valid up to infinity we store it only if it is
387 // the first unprocessed run!
388 if (aLocId.GetLastRun() == AliCDBRunRange::Infinity() &&
389 !fFirstUnprocessed[GetDetPos(fCurrentDetector)])
390 {
391 Log("SHUTTLE", Form("StoreOCDB - %s: object %s has validity infinite but "
392 "there are previous unprocessed runs!",
393 fCurrentDetector.Data(), aLocId.GetPath().Data()));
5e993b6f 394 Log(fCurrentDetector.Data(), Form("StoreOCDB - %s: object %s has validity infinite but "
395 "there are previous unprocessed runs!",
396 fCurrentDetector.Data(), aLocId.GetPath().Data()));
397 delayed = kTRUE;
3301427a 398 continue;
399 }
400
401 // loop on Grid valid Id's
402 Bool_t store = kTRUE;
403 TIter gridIter(gridIds);
404 AliCDBId* aGridId = 0;
6c0c9948 405 while ((aGridId = dynamic_cast<AliCDBId*> (gridIter.Next()))) {
406 if (aGridId->GetPath() != aLocId.GetPath())
407 continue;
3301427a 408 // skip all objects valid up to infinity
6c0c9948 409 if (aGridId->GetLastRun() == AliCDBRunRange::Infinity())
410 continue;
411
3301427a 412 // if we get here, it means there's already some more recent object stored on Grid!
6c0c9948 413 Log(fCurrentDetector.Data(),
414 Form("StoreOCDB - A more recent object already exists in %s storage: <%s>",
415 type, aGridId->ToString().Data()));
416
3301427a 417 store = kFALSE;
418 break;
419 }
420
6c0c9948 421 Bool_t storeOk = kFALSE;
422 if (store)
423 {
424 Log(fCurrentDetector.Data(), Form("Prechecks succeeded. Ready to store %s", aLocId.ToString().Data()));
425 storeOk = gridSto->Put(aLocEntry);
426 if (storeOk) {
3301427a 427 Log("SHUTTLE",
6c0c9948 428 Form("StoreOCDB - Object <%s> successfully put into %s storage",
429 aLocId.ToString().Data(), type));
2d9019b4 430 Log(fCurrentDetector.Data(),
431 Form("StoreOCDB - Object <%s> successfully put into %s storage",
6c0c9948 432 aLocId.ToString().Data(), type));
433 } else {
434 Log("SHUTTLE",
435 Form("StoreOCDB - Grid %s storage of object <%s> failed",
436 type, aLocId.ToString().Data()));
437 Log(fCurrentDetector.Data(),
438 Form("StoreOCDB - Grid %s storage of object <%s> failed",
439 type, aLocId.ToString().Data()));
440 result = kFALSE;
3301427a 441 }
6c0c9948 442 }
443
444 if (!store || storeOk) {
445 // removing local file...
3301427a 446 TString filename;
447 localSto->IdToFilename(aLocId, filename);
c88ad5db 448 Log("SHUTTLE", Form("StoreOCDB - Removing local file %s", filename.Data()));
3301427a 449 RemoveFile(filename.Data());
b948db8d 450 }
451 }
3301427a 452 localEntries->Clear();
2bb7b766 453
5e993b6f 454 Int_t returnCode = 0;
455
456 if (result == kFALSE)
457 returnCode = -1;
458 else if (delayed != kFALSE)
459 returnCode = 1;
460
461 Log("SHUTTLE", Form("StoreOCDB - Returning with %d (result = %d, delayed = %d)", returnCode, result, delayed));
462 Log(fCurrentDetector.Data(), Form("StoreOCDB - Returning with %d (result = %d, delayed = %d)", returnCode, result, delayed));
463
464 return returnCode;
3301427a 465}
466
546242fb 467//______________________________________________________________________________________________
468Bool_t AliShuttle::CleanReferenceStorage(const char* detector)
469{
2d9019b4 470 // clears the directory used to store reference files of a given subdetector
546242fb 471
472 AliCDBManager* man = AliCDBManager::Instance();
473 AliCDBStorage* sto = man->GetStorage(fgkLocalRefStorage);
2d9019b4 474 TString localBaseFolder = sto->GetBaseFolder();
475
476 TString targetDir = GetRefFilePrefix(localBaseFolder.Data(), detector);
477
d524ade6 478 Log("SHUTTLE", Form("CleanReferenceStorage - Cleaning %s", targetDir.Data()));
2d9019b4 479
480 TString begin;
481 begin.Form("%d_", GetCurrentRun());
482
483 TSystemDirectory* baseDir = new TSystemDirectory("/", targetDir);
484 if (!baseDir)
485 return kTRUE;
486
487 TList* dirList = baseDir->GetListOfFiles();
488 delete baseDir;
489
490 if (!dirList) return kTRUE;
491
efec19bd 492 if (dirList->GetEntries() < 3) // to be changed to 4?
2d9019b4 493 {
494 delete dirList;
495 return kTRUE;
496 }
497
498 Int_t nDirs = 0, nDel = 0;
499 TIter dirIter(dirList);
500 TSystemFile* entry = 0;
546242fb 501
2d9019b4 502 Bool_t success = kTRUE;
546242fb 503
2d9019b4 504 while ((entry = dynamic_cast<TSystemFile*> (dirIter.Next())))
505 {
506 if (entry->IsDirectory())
507 continue;
508
509 TString fileName(entry->GetName());
510 if (!fileName.BeginsWith(begin))
511 continue;
512
513 nDirs++;
514
515 // delete file
516 Int_t result = gSystem->Unlink(fileName.Data());
517
518 if (result)
519 {
d524ade6 520 Log("SHUTTLE", Form("CleanReferenceStorage - Could not delete file %s!", fileName.Data()));
2d9019b4 521 success = kFALSE;
522 } else {
523 nDel++;
524 }
525 }
526
527 if(nDirs > 0)
528 Log("SHUTTLE", Form("CleanReferenceStorage - %d (over %d) reference files in folder %s were deleted.",
529 nDel, nDirs, targetDir.Data()));
530
531
532 delete dirList;
533 return success;
534
535
536
537
538
546242fb 539
540 Int_t result = gSystem->GetPathInfo(targetDir, 0, (Long64_t*) 0, 0, 0);
541 if (result == 0)
542 {
543 // delete directory
d524ade6 544 result = gSystem->Exec(Form("rm -rf %s", targetDir.Data()));
546242fb 545 if (result != 0)
546 {
d524ade6 547 Log("SHUTTLE", Form("CleanReferenceStorage - Could not clean directory %s", targetDir.Data()));
546242fb 548 return kFALSE;
549 }
550 }
551
552 result = gSystem->mkdir(targetDir, kTRUE);
553 if (result != 0)
554 {
c88ad5db 555 Log("SHUTTLE", Form("CleanReferenceStorage - Error creating base directory %s", targetDir.Data()));
546242fb 556 return kFALSE;
557 }
558
559 return kTRUE;
560}
561
9827400b 562//______________________________________________________________________________________________
563Bool_t AliShuttle::StoreReferenceFile(const char* detector, const char* localFile, const char* gridFileName)
564{
565 //
3c2a21c8 566 // Stores reference file directly (without opening it). This function stores the file locally.
9827400b 567 //
3c2a21c8 568 // The file is stored under the following location:
569 // <base folder of local reference storage>/<DET>/<RUN#>_<gridFileName>
570 // where <gridFileName> is the second parameter given to the function
571 //
9827400b 572
573 if (fTestMode & kErrorStorage)
574 {
575 Log(fCurrentDetector, "StoreReferenceFile - In TESTMODE - Simulating error while storing locally");
576 return kFALSE;
577 }
578
579 AliCDBManager* man = AliCDBManager::Instance();
580 AliCDBStorage* sto = man->GetStorage(fgkLocalRefStorage);
581
582 TString localBaseFolder = sto->GetBaseFolder();
583
d524ade6 584 TString target = GetRefFilePrefix(localBaseFolder.Data(), detector);
585 target.Append(Form("/%d_%s", GetCurrentRun(), gridFileName));
9827400b 586
d524ade6 587 return CopyFileLocally(localFile, target);
c88ad5db 588}
589
590//______________________________________________________________________________________________
591Bool_t AliShuttle::StoreRunMetadataFile(const char* localFile, const char* gridFileName)
592{
593 //
594 // Stores Run metadata file to the Grid, in the run folder
595 //
596 // Only GRP can call this function.
597
598 if (fTestMode & kErrorStorage)
599 {
600 Log(fCurrentDetector, "StoreRunMetaDataFile - In TESTMODE - Simulating error while storing locally");
601 return kFALSE;
602 }
603
604 AliCDBManager* man = AliCDBManager::Instance();
605 AliCDBStorage* sto = man->GetStorage(fgkLocalRefStorage);
606
607 TString localBaseFolder = sto->GetBaseFolder();
608
609 // Build Run level folder
955d2abc 610 // folder = /alice/data/year/lhcPeriod/runNb/raw
c88ad5db 611
c88ad5db 612
675f64cd 613 TString lhcPeriod = GetLHCPeriod();
c88ad5db 614 if (lhcPeriod.Length() == 0)
615 {
616 Log("SHUTTLE","StoreRunMetaDataFile - LHCPeriod not found in logbook!");
617 return 0;
618 }
a1c5723b 619
620 // TODO partitions with one detector only write data into LHCperiod_DET
2694f253 621 TString partition = GetRunParameter("detector");
a1c5723b 622
623 if (partition.Length() > 0 && partition != "ALICE")
624 {
625 lhcPeriod.Append(Form("_%s", partition.Data()));
626 Log(fCurrentDetector, Form("Run data tags merged file will be written in %s",
627 lhcPeriod.Data()));
628 }
8884d71e 629
b8cf5e0f 630 TString target = Form("%s/GRP/RunMetadata%s%d/%s/%09d/raw/%s",
631 localBaseFolder.Data(), fConfig->GetAlienPath(), GetCurrentYear(),
d524ade6 632 lhcPeriod.Data(), GetCurrentRun(), gridFileName);
c88ad5db 633
d524ade6 634 return CopyFileLocally(localFile, target);
c88ad5db 635}
636
637//______________________________________________________________________________________________
d524ade6 638Bool_t AliShuttle::CopyFileLocally(const char* localFile, const TString& target)
c88ad5db 639{
640 //
641 // Stores file locally. Called by StoreReferenceFile and StoreRunMetadataFile
d524ade6 642 // Files are temporarily stored in the local reference storage. When the preprocessor
643 // finishes, the Shuttle calls CopyFilesToGrid to transfer the files to AliEn
644 // (in reference or run level folders)
c88ad5db 645 //
646
d524ade6 647 TString targetDir(target(0, target.Last('/')));
648
649 //try to open base dir folder, if it does not exist
2d9019b4 650 void* dir = gSystem->OpenDirectory(targetDir.Data());
651 if (dir == NULL) {
652 if (gSystem->mkdir(targetDir.Data(), kTRUE)) {
a1c5723b 653 Log("SHUTTLE", Form("CopyFileLocally - Can't open directory <%s>", targetDir.Data()));
2d9019b4 654 return kFALSE;
655 }
656
657 } else {
658 gSystem->FreeDirectory(dir);
659 }
9827400b 660
7d43a416 661 Int_t result = 0;
662
663 result = gSystem->GetPathInfo(localFile, 0, (Long64_t*) 0, 0, 0);
9827400b 664 if (result)
665 {
a1c5723b 666 Log("SHUTTLE", Form("CopyFileLocally - %s does not exist", localFile));
546242fb 667 return kFALSE;
9827400b 668 }
546242fb 669
7d43a416 670 result = gSystem->GetPathInfo(target, 0, (Long64_t*) 0, 0, 0);
671 if (!result)
672 {
a1c5723b 673 Log("SHUTTLE", Form("CopyFileLocally - target file %s already exist, removing...", target.Data()));
7d43a416 674 if (gSystem->Unlink(target.Data()))
675 {
a1c5723b 676 Log("SHUTTLE", Form("CopyFileLocally - Could not remove existing target file %s!", target.Data()));
7d43a416 677 return kFALSE;
678 }
679 }
680
9827400b 681 result = gSystem->CopyFile(localFile, target);
682
683 if (result == 0)
684 {
a1c5723b 685 Log("SHUTTLE", Form("CopyFileLocally - File %s stored locally to %s", localFile, target.Data()));
9827400b 686 return kTRUE;
687 }
688 else
689 {
a1c5723b 690 Log("SHUTTLE", Form("CopyFileLocally - Could not store file %s to %s! Error code = %d",
546242fb 691 localFile, target.Data(), result));
9827400b 692 return kFALSE;
693 }
c88ad5db 694
695
696
9827400b 697}
698
699//______________________________________________________________________________________________
c88ad5db 700Bool_t AliShuttle::CopyFilesToGrid(const char* type)
9827400b 701{
702 //
c88ad5db 703 // Transfers local files to the Grid. Local files can be reference files
704 // or run metadata file (from GRP only).
9827400b 705 //
c88ad5db 706 // According to the type (ref, metadata) the files are stored under the following location:
707 // ref --> <base folder of reference storage>/<DET>/<RUN#>_<gridFileName>
708 // metadata --> <run data folder>/<MetadataFileName>
86aa42c3 709 //
c88ad5db 710
9827400b 711 AliCDBManager* man = AliCDBManager::Instance();
712 AliCDBStorage* sto = man->GetStorage(fgkLocalRefStorage);
713 if (!sto)
714 return kFALSE;
715 TString localBaseFolder = sto->GetBaseFolder();
9827400b 716
c88ad5db 717 TString dir;
718 TString alienDir;
9827400b 719 TString begin;
9827400b 720
c88ad5db 721 if (strcmp(type, "reference") == 0)
722 {
723 dir = GetRefFilePrefix(localBaseFolder.Data(), fCurrentDetector.Data());
724 AliCDBStorage* gridSto = man->GetStorage(fgkMainRefStorage);
725 if (!gridSto)
726 return kFALSE;
727 TString gridBaseFolder = gridSto->GetBaseFolder();
728 alienDir = GetRefFilePrefix(gridBaseFolder.Data(), fCurrentDetector.Data());
729 begin = Form("%d_", GetCurrentRun());
730 }
731 else if (strcmp(type, "metadata") == 0)
732 {
c88ad5db 733
675f64cd 734 TString lhcPeriod = GetLHCPeriod();
c88ad5db 735
736 if (lhcPeriod.Length() == 0)
737 {
738 Log("SHUTTLE","CopyFilesToGrid - LHCPeriod not found in logbook!");
739 return 0;
740 }
741
a1c5723b 742 // TODO partitions with one detector only write data into LHCperiod_DET
2694f253 743 TString partition = GetRunParameter("detector");
a1c5723b 744
745 if (partition.Length() > 0 && partition != "ALICE")
746 {
747 lhcPeriod.Append(Form("_%s", partition.Data()));
748 }
749
b8cf5e0f 750 dir = Form("%s/GRP/RunMetadata%s%d/%s/%09d/raw",
751 localBaseFolder.Data(), fConfig->GetAlienPath(), GetCurrentYear(),
c88ad5db 752 lhcPeriod.Data(), GetCurrentRun());
b8cf5e0f 753 alienDir = dir(dir.Index(fConfig->GetAlienPath()), dir.Length());
675f64cd 754
c88ad5db 755 begin = "";
756 }
757 else
758 {
759 Log("SHUTTLE", "CopyFilesToGrid - Unexpected: type label must be reference or metadata!");
760 return kFALSE;
761 }
762
9827400b 763 TSystemDirectory* baseDir = new TSystemDirectory("/", dir);
3d8bc902 764 if (!baseDir)
765 return kTRUE;
766
2d9019b4 767 TList* dirList = baseDir->GetListOfFiles();
768 delete baseDir;
769
770 if (!dirList) return kTRUE;
771
772 if (dirList->GetEntries() < 3)
3d8bc902 773 {
2d9019b4 774 delete dirList;
9827400b 775 return kTRUE;
3d8bc902 776 }
2d9019b4 777
546242fb 778 if (!gGrid)
779 {
c88ad5db 780 Log("SHUTTLE", "CopyFilesToGrid - Connection to Grid failed: Cannot continue!");
2d9019b4 781 delete dirList;
546242fb 782 return kFALSE;
783 }
784
2d9019b4 785 Int_t nDirs = 0, nTransfer = 0;
786 TIter dirIter(dirList);
787 TSystemFile* entry = 0;
788
9827400b 789 Bool_t success = kTRUE;
3d8bc902 790 Bool_t first = kTRUE;
9827400b 791
2d9019b4 792 while ((entry = dynamic_cast<TSystemFile*> (dirIter.Next())))
793 {
9827400b 794 if (entry->IsDirectory())
795 continue;
796
797 TString fileName(entry->GetName());
798 if (!fileName.BeginsWith(begin))
799 continue;
800
2d9019b4 801 nDirs++;
802
3d8bc902 803 if (first)
804 {
805 first = kFALSE;
c88ad5db 806 // check that folder exists, otherwise create it
3d8bc902 807 TGridResult* result = gGrid->Ls(alienDir.Data(), "a");
808
809 if (!result)
2d9019b4 810 {
811 delete dirList;
3d8bc902 812 return kFALSE;
2d9019b4 813 }
3d8bc902 814
546242fb 815 if (!result->GetFileName(1)) // TODO: It looks like element 0 is always 0!!
3d8bc902 816 {
675f64cd 817 // TODO It does not work currently! Bug in TAliEn::Mkdir
818 // TODO Manually fixed in local root v5-16-00
c88ad5db 819 if (!gGrid->Mkdir(alienDir.Data(),"-p",0))
3d8bc902 820 {
c88ad5db 821 Log("SHUTTLE", Form("CopyFilesToGrid - Cannot create directory %s",
3d8bc902 822 alienDir.Data()));
2d9019b4 823 delete dirList;
3d8bc902 824 return kFALSE;
546242fb 825 } else {
c88ad5db 826 Log("SHUTTLE",Form("CopyFilesToGrid - Folder %s created", alienDir.Data()));
3d8bc902 827 }
828
546242fb 829 } else {
c88ad5db 830 Log("SHUTTLE",Form("CopyFilesToGrid - Folder %s found", alienDir.Data()));
3d8bc902 831 }
832 }
833
9827400b 834 TString fullLocalPath;
835 fullLocalPath.Form("%s/%s", dir.Data(), fileName.Data());
836
837 TString fullGridPath;
838 fullGridPath.Form("alien://%s/%s", alienDir.Data(), fileName.Data());
839
a986b218 840 Bool_t result = TFile::Cp(fullLocalPath, fullGridPath);
9827400b 841
842 if (result)
843 {
c88ad5db 844 Log("SHUTTLE", Form("CopyFilesToGrid - Copying local file %s to %s succeeded!",
845 fullLocalPath.Data(), fullGridPath.Data()));
9827400b 846 RemoveFile(fullLocalPath);
2d9019b4 847 nTransfer++;
9827400b 848 }
849 else
850 {
c88ad5db 851 Log("SHUTTLE", Form("CopyFilesToGrid - Copying local file %s to %s FAILED!",
852 fullLocalPath.Data(), fullGridPath.Data()));
9827400b 853 success = kFALSE;
854 }
855 }
2d9019b4 856
c88ad5db 857 Log("SHUTTLE", Form("CopyFilesToGrid - %d (over %d) files in folder %s copied to Grid.",
858 nTransfer, nDirs, dir.Data()));
2d9019b4 859
860
861 delete dirList;
9827400b 862 return success;
863}
864
2d9019b4 865//______________________________________________________________________________________________
866const char* AliShuttle::GetRefFilePrefix(const char* base, const char* detector)
867{
868 //
869 // Get folder name of reference files
870 //
871
872 TString offDetStr(GetOfflineDetName(detector));
3dd835d8 873 static TString dir;
2d9019b4 874 if (offDetStr == "ITS" || offDetStr == "MUON" || offDetStr == "PHOS")
875 {
876 dir.Form("%s/%s/%s", base, offDetStr.Data(), detector);
877 } else {
878 dir.Form("%s/%s", base, offDetStr.Data());
879 }
880
881 return dir.Data();
2d9019b4 882}
c88ad5db 883
3301427a 884//______________________________________________________________________________________________
885void AliShuttle::CleanLocalStorage(const TString& uri)
886{
9827400b 887 //
888 // Called in case the preprocessor is declared failed. Remove remaining objects from the local storages.
889 //
3301427a 890
891 const char* type = 0;
892 if(uri == fgkLocalCDB) {
893 type = "OCDB";
894 } else if(uri == fgkLocalRefStorage) {
546242fb 895 type = "Reference";
3301427a 896 } else {
897 AliError(Form("Invalid storage URI: %s", uri.Data()));
898 return;
899 }
900
901 AliCDBManager* man = AliCDBManager::Instance();
b948db8d 902
3301427a 903 // open local storage
904 AliCDBStorage *localSto = man->GetStorage(uri);
905 if(!localSto) {
906 Log("SHUTTLE",
907 Form("CleanLocalStorage - cannot activate local %s storage", type));
908 return;
909 }
910
911 TString filename(Form("%s/%s/*/Run*_v%d_s*.root",
546242fb 912 localSto->GetBaseFolder().Data(), GetOfflineDetName(fCurrentDetector.Data()), GetCurrentRun()));
3301427a 913
c88ad5db 914 AliDebug(2, Form("filename = %s", filename.Data()));
3301427a 915
c88ad5db 916 Log("SHUTTLE", Form("Removing remaining local files for run %d and detector %s ...",
3301427a 917 GetCurrentRun(), fCurrentDetector.Data()));
918
919 RemoveFile(filename.Data());
920
921}
922
923//______________________________________________________________________________________________
924void AliShuttle::RemoveFile(const char* filename)
925{
9827400b 926 //
927 // removes local file
928 //
3301427a 929
930 TString command(Form("rm -f %s", filename));
931
932 Int_t result = gSystem->Exec(command.Data());
933 if(result != 0)
934 {
935 Log("SHUTTLE", Form("RemoveFile - %s: Cannot remove file %s!",
936 fCurrentDetector.Data(), filename));
937 }
73abe331 938}
939
b948db8d 940//______________________________________________________________________________________________
5164a766 941AliShuttleStatus* AliShuttle::ReadShuttleStatus()
942{
9827400b 943 //
944 // Reads the AliShuttleStatus from the CDB
945 //
5164a766 946
2bb7b766 947 if (fStatusEntry){
948 delete fStatusEntry;
949 fStatusEntry = 0;
950 }
5164a766 951
3ebbf3c4 952 Int_t path1 = GetCurrentRun()/10000;
10a5a932 953 fStatusEntry = AliCDBManager::Instance()->GetStorage(GetLocalCDB())
3ebbf3c4 954 ->Get(Form("/SHUTTLE/%s/%d", fCurrentDetector.Data(), path1), GetCurrentRun());
5164a766 955
2bb7b766 956 if (!fStatusEntry) return 0;
957 fStatusEntry->SetOwner(1);
5164a766 958
2bb7b766 959 AliShuttleStatus* status = dynamic_cast<AliShuttleStatus*> (fStatusEntry->GetObject());
960 if (!status) {
961 AliError("Invalid object stored to CDB!");
962 return 0;
963 }
5164a766 964
2bb7b766 965 return status;
5164a766 966}
967
968//______________________________________________________________________________________________
7bfb2090 969Bool_t AliShuttle::WriteShuttleStatus(AliShuttleStatus* status)
5164a766 970{
9827400b 971 //
972 // writes the status for one subdetector
973 //
2bb7b766 974
975 if (fStatusEntry){
976 delete fStatusEntry;
977 fStatusEntry = 0;
978 }
5164a766 979
2bb7b766 980 Int_t run = GetCurrentRun();
3ebbf3c4 981 Int_t path1 = run/10000;
982 TString path1_string = Form("%d",path1);
5164a766 983
3ebbf3c4 984 AliCDBId id(AliCDBPath("SHUTTLE", fCurrentDetector, path1_string), run, run);
5164a766 985
2bb7b766 986 fStatusEntry = new AliCDBEntry(status, id, new AliCDBMetaData);
987 fStatusEntry->SetOwner(1);
5164a766 988
924fb6e9 989 Int_t logLevel = AliLog::GetGlobalLogLevel();
990 AliLog::SetGlobalLogLevel(AliLog::kError);
991
2bb7b766 992 UInt_t result = AliCDBManager::Instance()->GetStorage(fgkLocalCDB)->Put(fStatusEntry);
7bfb2090 993
924fb6e9 994 AliLog::SetGlobalLogLevel((AliLog::EType_t)logLevel);
995
996 if (!result) {
3301427a 997 Log("SHUTTLE", Form("WriteShuttleStatus - Failed for %s, run %d",
998 fCurrentDetector.Data(), run));
2bb7b766 999 return kFALSE;
1000 }
e7f62f16 1001
8e828d39 1002 SendMLDetInfo();
7bfb2090 1003
2bb7b766 1004 return kTRUE;
5164a766 1005}
1006
1007//______________________________________________________________________________________________
1008void AliShuttle::UpdateShuttleStatus(AliShuttleStatus::Status newStatus, Bool_t increaseCount)
1009{
9827400b 1010 //
1011 // changes the AliShuttleStatus for the given detector and run to the given status
1012 //
5164a766 1013
2bb7b766 1014 if (!fStatusEntry){
1015 AliError("UNEXPECTED: fStatusEntry empty");
1016 return;
1017 }
5164a766 1018
2bb7b766 1019 AliShuttleStatus* status = dynamic_cast<AliShuttleStatus*> (fStatusEntry->GetObject());
5164a766 1020
2bb7b766 1021 if (!status){
c88ad5db 1022 Log("SHUTTLE", "UpdateShuttleStatus - UNEXPECTED: status could not be read from current CDB entry");
2bb7b766 1023 return;
1024 }
5164a766 1025
2c15234c 1026 TString actionStr = Form("UpdateShuttleStatus - %s: Changing state from %s to %s",
eba76848 1027 fCurrentDetector.Data(),
36c99a6a 1028 status->GetStatusName(),
eba76848 1029 status->GetStatusName(newStatus));
cb343cfd 1030 Log("SHUTTLE", actionStr);
1031 SetLastAction(actionStr);
5164a766 1032
2bb7b766 1033 status->SetStatus(newStatus);
1034 if (increaseCount) status->IncreaseCount();
5164a766 1035
924fb6e9 1036 Int_t logLevel = AliLog::GetGlobalLogLevel();
1037 AliLog::SetGlobalLogLevel(AliLog::kError);
1038
2bb7b766 1039 AliCDBManager::Instance()->GetStorage(fgkLocalCDB)->Put(fStatusEntry);
e7f62f16 1040
924fb6e9 1041 AliLog::SetGlobalLogLevel((AliLog::EType_t)logLevel);
1042
8e828d39 1043 SendMLDetInfo();
5164a766 1044}
e7f62f16 1045
1046//______________________________________________________________________________________________
8e828d39 1047void AliShuttle::SendMLDetInfo()
e7f62f16 1048{
1049 //
1050 // sends ML information about the current status of the current detector being processed
1051 //
1052
1053 AliShuttleStatus* status = dynamic_cast<AliShuttleStatus*> (fStatusEntry->GetObject());
1054
1055 if (!status){
8e828d39 1056 Log("SHUTTLE", "SendMLDetInfo - UNEXPECTED: status could not be read from current CDB entry");
e7f62f16 1057 return;
1058 }
1059
1060 TMonaLisaText mlStatus(Form("%s_status", fCurrentDetector.Data()), status->GetStatusName());
1061 TMonaLisaValue mlRetryCount(Form("%s_count", fCurrentDetector.Data()), status->GetCount());
1062
1063 TList mlList;
1064 mlList.Add(&mlStatus);
1065 mlList.Add(&mlRetryCount);
1066
ee6f7523 1067 TString mlID;
1068 mlID.Form("%d", GetCurrentRun());
1069 fMonaLisa->SendParameters(&mlList, mlID);
e7f62f16 1070}
1071
5164a766 1072//______________________________________________________________________________________________
1073Bool_t AliShuttle::ContinueProcessing()
1074{
9827400b 1075 // this function reads the AliShuttleStatus information from CDB and
1076 // checks if the processing should be continued
1077 // if yes it returns kTRUE and updates the AliShuttleStatus with nextStatus
2bb7b766 1078
926cbc0e 1079 if (!fConfig->HostProcessDetector(fCurrentDetector))
1080 return kFALSE;
57c1a579 1081
1082 AliPreprocessor* aPreprocessor =
1083 dynamic_cast<AliPreprocessor*> (fPreprocessorMap.GetValue(fCurrentDetector));
1084 if (!aPreprocessor)
1085 {
82ac3683 1086 Log("SHUTTLE", Form("ContinueProcessing - %s: no preprocessor registered", fCurrentDetector.Data()));
1087 if(fCurrentDetector==TString("HLT")){
1088 Log("SHUTTLE", "ContinueProcessing: Fake processing HLT, to avoid pending runs! => temporary, till an HLT preprocessor is made available.");
1089 UpdateShuttleLogbook("HLT","DONE");
1090 Log(fCurrentDetector, "ContinueProcessing - skipping HLT preprocessor.");
1091
1092 AliShuttleStatus* status = ReadShuttleStatus();
1093 if (!status) {
1094 status = new AliShuttleStatus(AliShuttleStatus::kDone);
1095 WriteShuttleStatus(status);
1096 }
1097 }
1098 return kFALSE;
57c1a579 1099 }
1100
2bb7b766 1101 AliShuttleLogbookEntry::Status entryStatus =
eba76848 1102 fLogbookEntry->GetDetectorStatus(fCurrentDetector);
2bb7b766 1103
5e993b6f 1104 if (entryStatus != AliShuttleLogbookEntry::kUnprocessed) {
c88ad5db 1105 Log("SHUTTLE", Form("ContinueProcessing - %s is %s",
2bb7b766 1106 fCurrentDetector.Data(),
1107 fLogbookEntry->GetDetectorStatusName(entryStatus)));
1108 return kFALSE;
1109 }
1110
1111 // if we get here, according to Shuttle logbook subdetector is in UNPROCESSED state
be48e3ea 1112
1113 // check if current run is first unprocessed run for current detector
1114 if (fConfig->StrictRunOrder(fCurrentDetector) &&
1115 !fFirstUnprocessed[GetDetPos(fCurrentDetector)])
1116 {
86aa42c3 1117 if (fTestMode == kNone)
1118 {
c88ad5db 1119 Log("SHUTTLE", Form("ContinueProcessing - %s requires strict run ordering"
0a0b087a 1120 " but this is not the first unprocessed run!",fCurrentDetector.Data()));
86aa42c3 1121 return kFALSE;
1122 }
1123 else
1124 {
c88ad5db 1125 Log("SHUTTLE", Form("ContinueProcessing - In TESTMODE - "
1126 "Although %s requires strict run ordering "
1127 "and this is not the first unprocessed run, "
0a0b087a 1128 "the SHUTTLE continues",fCurrentDetector.Data()));
86aa42c3 1129 }
be48e3ea 1130 }
1131
926cbc0e 1132 // Is the subdetector processed first time for this run?
1133 fFirstProcessing = kFALSE;
1134
2bb7b766 1135 AliShuttleStatus* status = ReadShuttleStatus();
1136 if (!status) {
1137 // first time
1138 Log("SHUTTLE", Form("ContinueProcessing - %s: Processing first time",
1139 fCurrentDetector.Data()));
1140 status = new AliShuttleStatus(AliShuttleStatus::kStarted);
926cbc0e 1141 fFirstProcessing = kTRUE;
2bb7b766 1142 return WriteShuttleStatus(status);
1143 }
1144
ffa25f20 1145 // The following case shouldn't happen if Shuttle Logbook was correctly updated.
2bb7b766 1146 // If it happens it may mean Logbook updating failed... let's do it now!
1147 if (status->GetStatus() == AliShuttleStatus::kDone ||
ffa25f20 1148 status->GetStatus() == AliShuttleStatus::kFailed ||
1149 status->GetStatus() == AliShuttleStatus::kSkipped) {
2bb7b766 1150 Log("SHUTTLE", Form("ContinueProcessing - %s is already %s. Updating Shuttle Logbook",
1151 fCurrentDetector.Data(),
1152 status->GetStatusName(status->GetStatus())));
ffa25f20 1153
1154 if (status->GetStatus() == AliShuttleStatus::kSkipped)
1155 {
1156 UpdateShuttleLogbook(fCurrentDetector.Data(), "DONE");
1157 }
1158 else
1159 UpdateShuttleLogbook(fCurrentDetector.Data(), status->GetStatusName(status->GetStatus()));
1160
2bb7b766 1161 return kFALSE;
1162 }
1163
5e993b6f 1164 if (status->GetStatus() == AliShuttleStatus::kStoreStarted || status->GetStatus() == AliShuttleStatus::kStoreDelayed ||status->GetStatus() == AliShuttleStatus::kStoreError) {
2bb7b766 1165 Log("SHUTTLE",
c88ad5db 1166 Form("ContinueProcessing - %s: Grid storage of one or more "
1167 "objects failed. Trying again now",
2bb7b766 1168 fCurrentDetector.Data()));
5e993b6f 1169 StoreOCDB();
2bb7b766 1170 return kFALSE;
1171 }
1172
1173 // if we get here, there is a restart
57c1a579 1174 Bool_t cont = kFALSE;
2bb7b766 1175
1176 // abort conditions
cb343cfd 1177 if (status->GetCount() >= fConfig->GetMaxRetries()) {
57c1a579 1178 Log("SHUTTLE", Form("ContinueProcessing - %s failed %d times in status %s - "
1179 "Updating Shuttle Logbook", fCurrentDetector.Data(),
2bb7b766 1180 status->GetCount(), status->GetStatusName()));
1181 UpdateShuttleLogbook(fCurrentDetector.Data(), "FAILED");
e7f62f16 1182 UpdateShuttleStatus(AliShuttleStatus::kFailed);
3301427a 1183
1184 // there may still be objects in local OCDB and reference storage
1185 // and FXS databases may be not updated: do it now!
9827400b 1186
1187 // TODO Currently disabled, we want to keep files in case of failure!
1188 // CleanLocalStorage(fgkLocalCDB);
1189 // CleanLocalStorage(fgkLocalRefStorage);
1190 // UpdateTableFailCase();
1191
1192 // Send mail to detector expert!
c88ad5db 1193 Log("SHUTTLE", Form("ContinueProcessing - Sending mail to %s expert...",
926cbc0e 1194 fCurrentDetector.Data()));
fe6a3257 1195 // det experts in to
1196 TString to="";
1197 TIter *iterExperts = 0;
1198 iterExperts = new TIter(fConfig->GetResponsibles(fCurrentDetector));
1199 TObjString *anExpert=0;
1200 while ((anExpert = (TObjString*) iterExperts->Next()))
1201 {
1202 to += Form("%s, \n", anExpert->GetName());
1203 }
1204 delete iterExperts;
1205
1206 if (to.Length() > 0)
1207 to.Remove(to.Length()-3);
1208 AliDebug(2, Form("to: %s",to.Data()));
1209
1210 if (to.IsNull()) {
1211 Log("SHUTTLE", Form("List of %s responsibles not set!", fCurrentDetector.Data()));
1212 return kFALSE;
1213 }
1214
1215 Log(fCurrentDetector.Data(), Form("ContinueProcessing - Sending mail to %s expert(s):",
1216 fCurrentDetector.Data()));
1217 Log(fCurrentDetector.Data(), Form("\n%s", to.Data()));
339adafa 1218 if (!SendMail(kPPEMail))
9827400b 1219 Log("SHUTTLE", Form("ContinueProcessing - Could not send mail to %s expert",
926cbc0e 1220 fCurrentDetector.Data()));
3301427a 1221
57c1a579 1222 } else {
1223 Log("SHUTTLE", Form("ContinueProcessing - %s: restarting. "
1224 "Aborted before with %s. Retry number %d.", fCurrentDetector.Data(),
1225 status->GetStatusName(), status->GetCount()));
9827400b 1226 Bool_t increaseCount = kTRUE;
c88ad5db 1227 if (status->GetStatus() == AliShuttleStatus::kDCSError ||
8da81210 1228 status->GetStatus() == AliShuttleStatus::kDCSStarted ||
5ca8f352 1229 status->GetStatus() == AliShuttleStatus::kFXSError ||
1230 status->GetStatus() == AliShuttleStatus::kOCDBError)
c88ad5db 1231 increaseCount = kFALSE;
675f64cd 1232
9827400b 1233 UpdateShuttleStatus(AliShuttleStatus::kStarted, increaseCount);
57c1a579 1234 cont = kTRUE;
2bb7b766 1235 }
1236
57c1a579 1237 return cont;
5164a766 1238}
1239
8e828d39 1240//______________________________________________________________________________________________
1241void AliShuttle::SendMLRunInfo(const char* status)
1242{
1243 //
1244 // Send information about this run to ML
1245
1246 TMonaLisaText mlStatus("SHUTTLE_status", status);
1247 TString runType(fLogbookEntry->GetRunType());
1248 if (strlen(fLogbookEntry->GetRunParameter("log")) > 0){
1249
1250 runType += "(";
1251 runType += fLogbookEntry->GetRunParameter("log");
1252 runType += ")";
1253 }
728290a4 1254 if (fLogbookEntry->GetDATestMode()){
1255 runType += " (DATest)";
1256 }
8e828d39 1257 TMonaLisaText mlRunType("SHUTTLE_runtype", runType);
1258
1259 TList mlList;
1260 mlList.Add(&mlStatus);
1261 mlList.Add(&mlRunType);
1262
1263 TString mlID;
1264 mlID.Form("%d", GetCurrentRun());
1265 fMonaLisa->SendParameters(&mlList, mlID);
1266}
1267
ef83773f 1268//______________________________________________________________________________________________
1269Int_t AliShuttle::GetMem(Int_t pid)
1270{
1271 // invokes ps to get the memory consumption of the process <pid>
1272 // returns -1 in case of error
1273
1274 TString checkStr;
1275 checkStr.Form("ps -o vsize --pid %d | tail -n 1", pid);
1276 FILE* pipe = gSystem->OpenPipe(checkStr, "r");
1277 if (!pipe)
1278 {
1279 Log("SHUTTLE", Form("Process - Error: "
1280 "Could not open pipe to %s", checkStr.Data()));
1281 return -1;
1282 }
1283
1284 char buffer[100];
1285 if (!fgets(buffer, 100, pipe))
1286 {
1287 Log("SHUTTLE", "Process - Error: ps did not return anything");
1288 gSystem->ClosePipe(pipe);
1289 return -1;
1290 }
1291 gSystem->ClosePipe(pipe);
1292
1293 //Log("SHUTTLE", Form("ps returned %s", buffer));
1294
1295 Int_t mem = 0;
1296 if ((sscanf(buffer, "%d\n", &mem) != 1) || !mem)
1297 {
1298 Log("SHUTTLE", "Process - Error: Could not parse output of ps");
1299 return -1;
1300 }
1301
1302 return mem;
1303}
1304
5164a766 1305//______________________________________________________________________________________________
2bb7b766 1306Bool_t AliShuttle::Process(AliShuttleLogbookEntry* entry)
58bc3020 1307{
73abe331 1308 //
b948db8d 1309 // Makes data retrieval for all detectors in the configuration.
2bb7b766 1310 // entry: Shuttle logbook entry, contains run paramenters and status of detectors
1311 // (Unprocessed, Inactive, Failed or Done).
d477ad88 1312 // Returns kFALSE in case of error occured and kTRUE otherwise
73abe331 1313 //
1314
9827400b 1315 if (!entry) return kFALSE;
2bb7b766 1316
1317 fLogbookEntry = entry;
1318
c88ad5db 1319 Log("SHUTTLE", Form("\t\t\t^*^*^*^*^*^*^*^*^*^*^*^* run %d: START ^*^*^*^*^*^*^*^*^*^*^*^*",
9827400b 1320 GetCurrentRun()));
2bb7b766 1321
339adafa 1322 CountOpenRuns();
1323
8e828d39 1324 // Send the information to ML
1325 SendMLRunInfo("Processing");
3301427a 1326
9827400b 1327 if (fLogbookEntry->IsDone())
1328 {
1329 Log("SHUTTLE","Process - Shuttle is already DONE. Updating logbook");
1330 UpdateShuttleLogbook("shuttle_done");
1331 fLogbookEntry = 0;
1332 return kTRUE;
1333 }
1334
1335 // read test mode if flag is set
1336 if (fReadTestMode)
1337 {
3d8bc902 1338 fTestMode = kNone;
9827400b 1339 TString logEntry(entry->GetRunParameter("log"));
1340 //printf("log entry = %s\n", logEntry.Data());
1341 TString searchStr("Testmode: ");
1342 Int_t pos = logEntry.Index(searchStr.Data());
1343 //printf("%d\n", pos);
1344 if (pos >= 0)
1345 {
1346 TSubString subStr = logEntry(pos + searchStr.Length(), logEntry.Length());
1347 //printf("%s\n", subStr.String().Data());
1348 TString newStr(subStr.Data());
1349 TObjArray* token = newStr.Tokenize(' ');
1350 if (token)
1351 {
1352 //token->Print();
1353 TObjString* tmpStr = dynamic_cast<TObjString*> (token->First());
1354 if (tmpStr)
1355 {
1356 Int_t testMode = tmpStr->String().Atoi();
1357 if (testMode > 0)
1358 {
c88ad5db 1359 Log("SHUTTLE", Form("Process - Enabling test mode %d", testMode));
9827400b 1360 SetTestMode((TestMode) testMode);
1361 }
1362 }
1363 delete token;
1364 }
1365 }
1366 }
c88ad5db 1367
eba76848 1368 fLogbookEntry->Print("all");
57f50b3c 1369
1370 // Initialization
d477ad88 1371 Bool_t hasError = kFALSE;
5164a766 1372
728290a4 1373 // Set the CDB and Reference folders according to the year
1374
1375 // build cdb paths (repeat each time, run might be a DATest run)
1376 if (!fLogbookEntry->GetDATestMode()){
1377 fgkMainCDB.Form("alien://folder=%s%d/OCDB?user=alidaq?cacheFold=/tmp/OCDBCache",
1378 fConfig->GetAlienPath(), GetCurrentYear());
1379
1380 fgkMainRefStorage.Form("alien://folder=%s%d/Reference?user=alidaq?cacheFold=/tmp/OCDBCache",
1381 fConfig->GetAlienPath(), GetCurrentYear());
1382 }
1383 else {
1384 fgkMainCDB.Form("alien://folder=%s%d/DATest/OCDB?user=alidaq?cacheFold=/tmp/OCDBCache",
1385 fConfig->GetAlienPath(), GetCurrentYear());
1386
1387 fgkMainRefStorage.Form("alien://folder=%s%d/DATest/Reference?user=alidaq?cacheFold=/tmp/OCDBCache",
1388 fConfig->GetAlienPath(), GetCurrentYear());
1389 }
1390
1391 AliDebug(2,Form("Main CDB storage = %s",fgkMainCDB.Data()));
1392 AliDebug(2,Form("Main Reference storage = %s",fgkMainRefStorage.Data()));
1393
57f50b3c 1394 // Loop on detectors in the configuration
b948db8d 1395 TIter iter(fConfig->GetDetectors());
2bb7b766 1396 TObjString* aDetector = 0;
b948db8d 1397
4508ab8b 1398 Bool_t first = kTRUE;
1399
be48e3ea 1400 while ((aDetector = (TObjString*) iter.Next()))
1401 {
7bfb2090 1402 fCurrentDetector = aDetector->String();
5164a766 1403
5e993b6f 1404 if (ContinueProcessing() == kFALSE)
1405 continue;
4508ab8b 1406
1407 if (first)
1408 {
1409 // only read QueryCDB when needed and only once
1410 AliCDBStorage *mainCDBSto = AliCDBManager::Instance()->GetStorage(fgkMainCDB);
1411 if(mainCDBSto) mainCDBSto->QueryCDB(GetCurrentRun());
1412 AliCDBStorage *mainRefSto = AliCDBManager::Instance()->GetStorage(fgkMainRefStorage);
1413 if(mainRefSto) mainRefSto->QueryCDB(GetCurrentRun());
1414 first = kFALSE;
1415 }
9e080f92 1416
c88ad5db 1417 Log("SHUTTLE", Form("\t\t\t****** run %d - %s: START ******",
2bb7b766 1418 GetCurrentRun(), aDetector->GetName()));
1419
6e25f828 1420 for(Int_t iSys=0;iSys<4;iSys++) fFXSCalled[iSys]=kFALSE;
ef83773f 1421
1422 Int_t initialMem = GetMem(getpid());
1423 Log("SHUTTLE", Form("Memory consumption before forking is %d", initialMem));
9d733021 1424
c88ad5db 1425 Log(fCurrentDetector.Data(), "Process - Starting processing");
85a80aa9 1426
be48e3ea 1427 Int_t pid = fork();
1428
1429 if (pid < 0)
1430 {
c88ad5db 1431 Log("SHUTTLE", "Process - ERROR: Forking failed");
be48e3ea 1432 }
1433 else if (pid > 0)
1434 {
1435 // parent
c88ad5db 1436 Log("SHUTTLE", Form("Process - In parent process of %d - %s: Starting monitoring",
be48e3ea 1437 GetCurrentRun(), aDetector->GetName()));
1438
1439 Long_t begin = time(0);
1440
1441 int status; // to be used with waitpid, on purpose an int (not Int_t)!
1442 while (waitpid(pid, &status, WNOHANG) == 0)
1443 {
1444 Long_t expiredTime = time(0) - begin;
1445
1446 if (expiredTime > fConfig->GetPPTimeOut())
1447 {
56d3e9e4 1448 TString logMsg;
1449 AliShuttleStatus *currentStatus = ReadShuttleStatus();
1450 AliShuttleStatus::Status newStatus = AliShuttleStatus::kInvalid;
1451
838d9827 1452 if (currentStatus->GetStatus() == AliShuttleStatus::kDCSStarted)
1453 {
1454 // in case the pp goes in TimeOut while retrieving the DCS DPs
1455 // set status to kDCSError
1456
0a0b087a 1457 logMsg.Form("Process - Process of %s timed out while retrieving the DCS DataPoints. Run time: %ld seconds. Killing... and setting status to DCSError.",
838d9827 1458 fCurrentDetector.Data(), expiredTime);
1459 newStatus = AliShuttleStatus::kDCSError;
1460 }
1461 else if (currentStatus->GetStatus() <= AliShuttleStatus::kPPDone)
56d3e9e4 1462 {
1463 // in case pp not yet done set status to kPPTimeOut
1464
0a0b087a 1465 logMsg.Form("Process - Process of %s timed out. Run time: %ld seconds. Killing...",
56d3e9e4 1466 fCurrentDetector.Data(), expiredTime);
1467 newStatus = AliShuttleStatus::kPPTimeOut;
1468 }
1469 else if (currentStatus->GetStatus() == AliShuttleStatus::kStoreStarted)
1470 {
1471 // in case the pp goes in TimeOut while storing the objects in the OCDB
1472 // set status to kStoreError
1473
0a0b087a 1474 logMsg.Form("Process - Process of %s timed out while storing the OCDB object. Run time: %ld seconds. Killing... and setting status to StoreError.",
56d3e9e4 1475 fCurrentDetector.Data(), expiredTime);
1476 newStatus = AliShuttleStatus::kStoreError;
1477 }
1478 else
1479 {
1480 // in other cases don't change the status
1481
0a0b087a 1482 logMsg.Form("Process - Process of %s timed out in status = %s. Run time: %ld seconds. Killing... without changing the status",
56d3e9e4 1483 fCurrentDetector.Data(), currentStatus->GetStatusName(), expiredTime);
1484 }
1485
1486 Log("SHUTTLE", logMsg);
1487 Log(fCurrentDetector, logMsg);
be48e3ea 1488
1489 kill(pid, 9);
1490
56d3e9e4 1491 if (newStatus != AliShuttleStatus::kInvalid)
1492 UpdateShuttleStatus(newStatus);
be48e3ea 1493 hasError = kTRUE;
1494
1495 gSystem->Sleep(1000);
1496 }
1497 else
1498 {
be48e3ea 1499 gSystem->Sleep(1000);
9827400b 1500
ef83773f 1501 Int_t mem = GetMem(pid);
1502
1503 if (mem < 0)
9827400b 1504 continue;
9827400b 1505
ef83773f 1506 mem -= initialMem;
1507 if (mem < 0)
1508 mem = 0;
9827400b 1509
1510 if (expiredTime % 60 == 0)
ee6f7523 1511 {
c88ad5db 1512 Log("SHUTTLE", Form("Process - %s: Checking process. "
0a0b087a 1513 "Run time: %ld seconds - Memory consumption: %d KB",
c88ad5db 1514 fCurrentDetector.Data(), expiredTime, mem));
ee6f7523 1515 SendAlive();
1516 }
9827400b 1517
1518 if (mem > fConfig->GetPPMaxMem())
1519 {
1520 TString tmp;
c88ad5db 1521 tmp.Form("Process - Process exceeds maximum allowed memory "
1522 "(%d KB > %d KB). Killing...",
9827400b 1523 mem, fConfig->GetPPMaxMem());
1524 Log("SHUTTLE", tmp);
1525 Log(fCurrentDetector, tmp);
1526
1527 kill(pid, 9);
1528
1529 UpdateShuttleStatus(AliShuttleStatus::kPPOutOfMemory);
1530 hasError = kTRUE;
1531
1532 gSystem->Sleep(1000);
1533 }
be48e3ea 1534 }
1535 }
1536
c88ad5db 1537 Log("SHUTTLE", Form("Process - In parent process of %d - %s: Client has terminated.",
be48e3ea 1538 GetCurrentRun(), aDetector->GetName()));
1539
1540 if (WIFEXITED(status))
1541 {
1542 Int_t returnCode = WEXITSTATUS(status);
1543
c88ad5db 1544 Log("SHUTTLE", Form("Process - %s: the return code is %d", fCurrentDetector.Data(),
3301427a 1545 returnCode));
be48e3ea 1546
9827400b 1547 if (returnCode == 0) hasError = kTRUE;
be48e3ea 1548 }
1549 }
1550 else if (pid == 0)
1551 {
26b1866e 1552 // child
1553 Log("SHUTTLE", Form("Process - In child process of %d - %s", GetCurrentRun(),
c88ad5db 1554 aDetector->GetName()));
be48e3ea 1555
c88ad5db 1556 Log("SHUTTLE", Form("Process - Redirecting output to %s log",fCurrentDetector.Data()));
ffa29e93 1557
546242fb 1558 if ((freopen(GetLogFileName(fCurrentDetector), "a", stdout)) == 0)
ffa29e93 1559 {
c88ad5db 1560 Log("SHUTTLE", "Process - Could not freopen stdout");
ffa29e93 1561 }
1562 else
1563 {
1564 fOutputRedirected = kTRUE;
1565 if ((dup2(fileno(stdout), fileno(stderr))) < 0)
c88ad5db 1566 Log("SHUTTLE", "Process - Could not redirect stderr");
ffa29e93 1567
1568 }
51c3744e 1569
1570 Log("SHUTTLE", "Executing TGrid::Connect");
1571 TGrid::Connect("alien://");
ffa29e93 1572
5bac2bde 1573 TString wd = gSystem->WorkingDirectory();
3ebbf3c4 1574 Int_t dir_lev1 = GetCurrentRun()/10000;
1575 TString tmpDir = Form("%s/%d/%d/%s_process", GetShuttleTempDir(),
1576 dir_lev1, GetCurrentRun(), fCurrentDetector.Data());
5bac2bde 1577
d524ade6 1578 Int_t result = gSystem->GetPathInfo(tmpDir.Data(), 0, (Long64_t*) 0, 0, 0);
1579 if (!result) // temp dir already exists!
1580 {
1581 Log(fCurrentDetector.Data(),
1582 Form("Process - %s dir already exists! Removing...", tmpDir.Data()));
1583 gSystem->Exec(Form("rm -rf %s",tmpDir.Data()));
675f64cd 1584 }
1585
1586 if (gSystem->mkdir(tmpDir.Data(), 1))
1587 {
1588 Log(fCurrentDetector.Data(), "Process - could not make temp directory!!");
1589 gSystem->Exit(1);
d524ade6 1590 }
1591
1592 if (!gSystem->ChangeDirectory(tmpDir.Data()))
1593 {
1594 Log(fCurrentDetector.Data(), "Process - could not change directory!!");
1595 gSystem->Exit(1);
1596 }
5bac2bde 1597
ffa25f20 1598 Int_t success = ProcessCurrentDetector();
26b1866e 1599
5bac2bde 1600 gSystem->ChangeDirectory(wd.Data());
b1d18693 1601
ffa25f20 1602 if (success == 1) // Preprocessor finished successfully!
9827400b 1603 {
fb40ead4 1604 // remove temporary folder or DCS map
1605 if (!fConfig->KeepTempFolder())
1606 {
1607 gSystem->Exec(Form("rm -rf %s",tmpDir.Data()));
1608 } else if (!fConfig->KeepDCSMap())
1609 {
1610 gSystem->Exec(Form("rm -f %s/DCSMap.root",tmpDir.Data()));
1611 }
b1d18693 1612
3301427a 1613 // Update time_processed field in FXS DB
1614 if (UpdateTable() == kFALSE)
5bac2bde 1615 Log("SHUTTLE", Form("Process - %s: Could not update FXS databases!",
1616 fCurrentDetector.Data()));
3301427a 1617
1618 // Transfer the data from local storage to main storage (Grid)
3301427a 1619 if (StoreOCDB() == kFALSE)
9827400b 1620 success = kFALSE;
5e993b6f 1621 }
ffa25f20 1622 else if (success == 0)
c88ad5db 1623 {
1624 Log("SHUTTLE",
5ca8f352 1625 Form("\t\t\t****** run %d - %s: ERROR ******",
c88ad5db 1626 GetCurrentRun(), aDetector->GetName()));
be48e3ea 1627 }
1628
6e25f828 1629 for (UInt_t iSys=0; iSys<4; iSys++)
4b95672b 1630 {
1631 if (fFXSCalled[iSys]) fFXSlist[iSys].Clear();
1632 }
1633
c88ad5db 1634 Log("SHUTTLE", Form("Process - Client process of %d - %s is exiting now with %d.",
9827400b 1635 GetCurrentRun(), aDetector->GetName(), success));
be48e3ea 1636
1637 // the client exits here
9827400b 1638 gSystem->Exit(success);
be48e3ea 1639
1640 AliError("We should never get here!!!");
1641 }
7bfb2090 1642 }
5164a766 1643
c88ad5db 1644 Log("SHUTTLE", Form("\t\t\t^*^*^*^*^*^*^*^*^*^*^*^* run %d: FINISH ^*^*^*^*^*^*^*^*^*^*^*^*",
2bb7b766 1645 GetCurrentRun()));
1646
1647 //check if shuttle is done for this run, if so update logbook
1648 TObjArray checkEntryArray;
1649 checkEntryArray.SetOwner(1);
9e080f92 1650 TString whereClause = Form("where run=%d", GetCurrentRun());
b0e53b15 1651 if (!QueryShuttleLogbook(whereClause.Data(), checkEntryArray) ||
1652 checkEntryArray.GetEntries() == 0) {
9e080f92 1653 Log("SHUTTLE", Form("Process - Warning: Cannot check status of run %d on Shuttle logbook!",
1654 GetCurrentRun()));
1655 return hasError == kFALSE;
1656 }
b948db8d 1657
9e080f92 1658 AliShuttleLogbookEntry* checkEntry = dynamic_cast<AliShuttleLogbookEntry*>
1659 (checkEntryArray.At(0));
2bb7b766 1660
9e080f92 1661 if (checkEntry)
1662 {
1663 if (checkEntry->IsDone())
be48e3ea 1664 {
9e080f92 1665 Log("SHUTTLE","Process - Shuttle is DONE. Updating logbook");
1666 UpdateShuttleLogbook("shuttle_done");
1667 }
1668 else
1669 {
1670 for (UInt_t iDet=0; iDet<NDetectors(); iDet++)
be48e3ea 1671 {
9e080f92 1672 if (checkEntry->GetDetectorStatus(iDet) == AliShuttleLogbookEntry::kUnprocessed)
be48e3ea 1673 {
9e080f92 1674 AliDebug(2, Form("Run %d: setting %s as \"not first time unprocessed\"",
1675 checkEntry->GetRun(), GetDetName(iDet)));
1676 fFirstUnprocessed[iDet] = kFALSE;
be48e3ea 1677 }
1678 }
8e828d39 1679 SendMLRunInfo("Pending");
2bb7b766 1680 }
1681 }
1682
1683 fLogbookEntry = 0;
85a80aa9 1684
a7160fe9 1685 return hasError == kFALSE;
73abe331 1686}
1687
b948db8d 1688//______________________________________________________________________________________________
ffa25f20 1689Int_t AliShuttle::ProcessCurrentDetector()
73abe331 1690{
1691 //
2bb7b766 1692 // Makes data retrieval just for a specific detector (fCurrentDetector).
73abe331 1693 // Threre should be a configuration for this detector.
73abe331 1694
1d172743 1695 Log("SHUTTLE", Form("ProcessCurrentDetector - Retrieving values for %s, run %d",
1696 fCurrentDetector.Data(), GetCurrentRun()));
73abe331 1697
d524ade6 1698 TString wd = gSystem->WorkingDirectory();
1699
2d9019b4 1700 if (!CleanReferenceStorage(fCurrentDetector.Data()))
ffa25f20 1701 return 0;
d524ade6 1702
1703 gSystem->ChangeDirectory(wd.Data());
1704
3301427a 1705 // call preprocessor
1706 AliPreprocessor* aPreprocessor =
1707 dynamic_cast<AliPreprocessor*> (fPreprocessorMap.GetValue(fCurrentDetector));
1708
8579c6fb 1709 // check if the preprocessor wants to process this run type
1710 if (aPreprocessor->ProcessRunType() == kFALSE)
ffa25f20 1711 {
1712 UpdateShuttleStatus(AliShuttleStatus::kSkipped);
1713 UpdateShuttleLogbook(fCurrentDetector, "DONE");
bd23d48b 1714 if (!UpdateTableSkippedCase(fCurrentDetector.Data()))
1715 {
1716 AliError(Form("Could not update FXS tables for run %d !", GetCurrentRun()));
1717 }
ffa25f20 1718 Log(fCurrentDetector, Form("ProcessCurrentDetector - %s preprocessor is not interested in this run type", fCurrentDetector.Data()));
1719
1720 return 2;
1721 }
1722
5ca8f352 1723 // checking if OCDB is reachable
1724 AliCDBEntry* testEntry = GetFromOCDB("SHUTTLE","GRP/CTP/DummyConfig");
1725 if (!testEntry){
1726 // OCDB is not accessible, going in OCDBError for current detector
1727 AliError("OCDB Test entry not accessible");
1728 UpdateShuttleStatus(AliShuttleStatus::kOCDBError);
1729 return 0;
1730 }
1731
ffa25f20 1732 TMap* dcsMap = new TMap();
1733
3301427a 1734 aPreprocessor->Initialize(GetCurrentRun(), GetCurrentStartTime(), GetCurrentEndTime());
1735
1736 Bool_t processDCS = aPreprocessor->ProcessDCS();
d477ad88 1737
651fdaab 1738 if (!processDCS)
1739 {
1d172743 1740 Log(fCurrentDetector, "ProcessCurrentDetector -"
1741 " The preprocessor requested to skip the retrieval of DCS values");
651fdaab 1742 }
8b739301 1743 else if (fTestMode & kSkipDCS)
2c15234c 1744 {
1d172743 1745 Log(fCurrentDetector, "ProcessCurrentDetector - In TESTMODE: Skipping DCS processing");
9827400b 1746 }
1747 else if (fTestMode & kErrorDCS)
1748 {
1d172743 1749 Log(fCurrentDetector, "ProcessCurrentDetector - In TESTMODE: Simulating DCS error");
3d8bc902 1750 UpdateShuttleStatus(AliShuttleStatus::kDCSStarted);
9827400b 1751 UpdateShuttleStatus(AliShuttleStatus::kDCSError);
1d172743 1752 delete dcsMap;
ffa25f20 1753 return 0;
2c15234c 1754 } else {
3301427a 1755
1756 UpdateShuttleStatus(AliShuttleStatus::kDCSStarted);
1757
1d172743 1758 // Query DCS archive
1759 Int_t nServers = fConfig->GetNServers(fCurrentDetector);
a038aa70 1760
1d172743 1761 for (int iServ=0; iServ<nServers; iServ++)
2c15234c 1762 {
1d172743 1763
1764 TString host(fConfig->GetDCSHost(fCurrentDetector, iServ));
1765 Int_t port = fConfig->GetDCSPort(fCurrentDetector, iServ);
542b6cc8 1766 Int_t multiSplit = fConfig->GetMultiSplit(fCurrentDetector, iServ);
1767
1790d4b7 1768 Log(fCurrentDetector, Form("ProcessCurrentDetector -"
1769 " Querying DCS Amanda server %s:%d (%d of %d)",
1770 host.Data(), port, iServ+1, nServers));
1d172743 1771
1772 TMap* aliasMap = 0;
1773 TMap* dpMap = 0;
26b1866e 1774
1d172743 1775 if (fConfig->GetDCSAliases(fCurrentDetector, iServ)->GetEntries() > 0)
2c15234c 1776 {
5ed060b1 1777 Log(fCurrentDetector, Form("Querying %d DCS aliases", fConfig->GetDCSAliases(fCurrentDetector, iServ)->GetEntries()));
1d172743 1778 aliasMap = GetValueSet(host, port,
542b6cc8 1779 fConfig->GetDCSAliases(fCurrentDetector, iServ),
1780 kAlias, multiSplit);
1d172743 1781 if (!aliasMap)
1782 {
1783 Log(fCurrentDetector,
1784 Form("ProcessCurrentDetector -"
675f64cd 1785 " Error retrieving DCS aliases from server %s."
1786 " Sending mail to DCS experts!", host.Data()));
1d172743 1787 UpdateShuttleStatus(AliShuttleStatus::kDCSError);
675f64cd 1788
339adafa 1789 if (!SendMail(kDCSEMail))
fb40ead4 1790 Log("SHUTTLE", Form("ProcessCurrentDetector - "
95d4aaef 1791 "Could not send mail to DCS experts!"));
675f64cd 1792
a038aa70 1793 delete dcsMap;
ffa25f20 1794 return 0;
1d172743 1795 }
2c15234c 1796 }
a038aa70 1797
1d172743 1798 if (fConfig->GetDCSDataPoints(fCurrentDetector, iServ)->GetEntries() > 0)
a038aa70 1799 {
5ed060b1 1800 Log(fCurrentDetector, Form("Querying %d DCS data points", fConfig->GetDCSDataPoints(fCurrentDetector, iServ)->GetEntries()));
1d172743 1801 dpMap = GetValueSet(host, port,
542b6cc8 1802 fConfig->GetDCSDataPoints(fCurrentDetector, iServ),
1803 kDP, multiSplit);
1d172743 1804 if (!dpMap)
1805 {
1806 Log(fCurrentDetector,
1807 Form("ProcessCurrentDetector -"
675f64cd 1808 " Error retrieving DCS data points from server %s."
1809 " Sending mail to DCS experts!", host.Data()));
1d172743 1810 UpdateShuttleStatus(AliShuttleStatus::kDCSError);
675f64cd 1811
339adafa 1812 if (!SendMail(kDCSEMail))
fb40ead4 1813 Log("SHUTTLE", Form("ProcessCurrentDetector - "
95d4aaef 1814 "Could not send mail to DCS experts!"));
675f64cd 1815
1d172743 1816 if (aliasMap) delete aliasMap;
1817 delete dcsMap;
ffa25f20 1818 return 0;
1d172743 1819 }
a038aa70 1820 }
1d172743 1821
1822 // merge aliasMap and dpMap into dcsMap
1823 if(aliasMap) {
1824 TIter iter(aliasMap);
a038aa70 1825 TObjString* key = 0;
1826 while ((key = (TObjString*) iter.Next()))
1d172743 1827 dcsMap->Add(key, aliasMap->GetValue(key->String()));
1828
1829 aliasMap->SetOwner(kFALSE);
1830 delete aliasMap;
1831 }
1832
1833 if(dpMap) {
1834 TIter iter(dpMap);
1835 TObjString* key = 0;
1836 while ((key = (TObjString*) iter.Next()))
1837 dcsMap->Add(key, dpMap->GetValue(key->String()));
1838
1839 dpMap->SetOwner(kFALSE);
1840 delete dpMap;
a038aa70 1841 }
73abe331 1842 }
1843 }
dc25836b 1844
b1d18693 1845 // save map into file, to help debugging in case of preprocessor error
fbc112e3 1846 TFile* f = TFile::Open("DCSMap.root","recreate");
b1d18693 1847 f->cd();
1848 dcsMap->Write("DCSMap", TObject::kSingleKey);
1849 f->Close();
fbc112e3 1850 delete f;
b1d18693 1851
2bb7b766 1852 // DCS Archive DB processing successful. Call Preprocessor!
85a80aa9 1853 UpdateShuttleStatus(AliShuttleStatus::kPPStarted);
a7160fe9 1854
339adafa 1855 fFXSError = -1; // this variable is kTRUE after ::Process if an FXS error occured
8da81210 1856
a038aa70 1857 UInt_t returnValue = aPreprocessor->Process(dcsMap);
8da81210 1858
339adafa 1859 if (fFXSError!=-1) {
8da81210 1860 UpdateShuttleStatus(AliShuttleStatus::kFXSError);
339adafa 1861 SendMail(kFXSEMail, fFXSError);
8da81210 1862 dcsMap->DeleteAll();
1863 delete dcsMap;
ffa25f20 1864 return 0;
8da81210 1865 }
b948db8d 1866
3301427a 1867 if (returnValue > 0) // Preprocessor error!
1868 {
c88ad5db 1869 Log(fCurrentDetector, Form("ProcessCurrentDetector - "
1870 "Preprocessor failed. Process returned %d.", returnValue));
cb343cfd 1871 UpdateShuttleStatus(AliShuttleStatus::kPPError);
a038aa70 1872 dcsMap->DeleteAll();
1873 delete dcsMap;
ffa25f20 1874 return 0;
9827400b 1875 }
1876
1877 // preprocessor ok!
1878 UpdateShuttleStatus(AliShuttleStatus::kPPDone);
1879 Log(fCurrentDetector, Form("ProcessCurrentDetector - %s preprocessor returned success",
1880 fCurrentDetector.Data()));
b948db8d 1881
a038aa70 1882 dcsMap->DeleteAll();
1883 delete dcsMap;
b948db8d 1884
ffa25f20 1885 return 1;
2bb7b766 1886}
1887
fbc112e3 1888//______________________________________________________________________________________________
1889void AliShuttle::CountOpenRuns()
1890{
1891 // Query DAQ's Shuttle logbook and sends the number of open runs to ML
1892
982278d6 1893 SendAlive();
1894
fbc112e3 1895 // check connection, in case connect
efec19bd 1896 if (!Connect(4))
fbc112e3 1897 return;
1898
1899 TString sqlQuery;
1900 sqlQuery = Form("select count(*) from %s where shuttle_done=0", fConfig->GetShuttlelbTable());
1901
efec19bd 1902 TSQLResult* aResult = fServer[4]->Query(sqlQuery);
fbc112e3 1903 if (!aResult) {
1904 AliError(Form("Can't execute query <%s>!", sqlQuery.Data()));
1905 return;
1906 }
1907
1908 AliDebug(2,Form("Query = %s", sqlQuery.Data()));
1909
1910 if (aResult->GetRowCount() == 0) {
1911 AliError(Form("No result for query %s received", sqlQuery.Data()));
1912 return;
1913 }
1914
1915 if (aResult->GetFieldCount() != 1) {
1916 AliError(Form("Invalid field count for query %s received", sqlQuery.Data()));
1917 return;
1918 }
1919
1920 TSQLRow* aRow = aResult->Next();
1921 if (!aRow) {
1922 AliError(Form("Could not receive result of query %s", sqlQuery.Data()));
1923 return;
1924 }
1925
1926 TString result(aRow->GetField(0), aRow->GetFieldLength(0));
1927 Int_t count = result.Atoi();
1928
1929 Log("SHUTTLE", Form("%d unprocessed runs", count));
1930
1931 delete aRow;
1932 delete aResult;
1933
1934 TMonaLisaValue mlStatus("SHUTTLE_openruns", count);
1935
1936 TList mlList;
1937 mlList.Add(&mlStatus);
1938
1939 fMonaLisa->SendParameters(&mlList, "__PROCESSINGINFO__");
1940}
1941
2bb7b766 1942//______________________________________________________________________________________________
1943Bool_t AliShuttle::QueryShuttleLogbook(const char* whereClause,
1944 TObjArray& entries)
1945{
9827400b 1946 // Query DAQ's Shuttle logbook and fills detector status object.
1947 // Call QueryRunParameters to query DAQ logbook for run parameters.
1948 //
2bb7b766 1949
fc5a4708 1950 entries.SetOwner(1);
1951
2bb7b766 1952 // check connection, in case connect
efec19bd 1953 if (!Connect(4)) return kFALSE;
2bb7b766 1954
1955 TString sqlQuery;
441b0e9c 1956 sqlQuery = Form("select * from %s %s order by run", fConfig->GetShuttlelbTable(), whereClause);
2bb7b766 1957
efec19bd 1958 TSQLResult* aResult = fServer[4]->Query(sqlQuery);
2bb7b766 1959 if (!aResult) {
1960 AliError(Form("Can't execute query <%s>!", sqlQuery.Data()));
1961 return kFALSE;
1962 }
1963
fc5a4708 1964 AliDebug(2,Form("Query = %s", sqlQuery.Data()));
1965
2bb7b766 1966 if(aResult->GetRowCount() == 0) {
c88ad5db 1967 Log("SHUTTLE", "No entries in Shuttle Logbook match request");
9827400b 1968 delete aResult;
1969 return kTRUE;
2bb7b766 1970 }
1971
1972 // TODO Check field count!
5b23e2da 1973 const UInt_t nCols = 26;
2bb7b766 1974 if (aResult->GetFieldCount() != (Int_t) nCols) {
c88ad5db 1975 Log("SHUTTLE", "Invalid SQL result field number!");
2bb7b766 1976 delete aResult;
1977 return kFALSE;
1978 }
1979
2bb7b766 1980 TSQLRow* aRow;
1981 while ((aRow = aResult->Next())) {
1982 TString runString(aRow->GetField(0), aRow->GetFieldLength(0));
1983 Int_t run = runString.Atoi();
1984
eba76848 1985 AliShuttleLogbookEntry *entry = QueryRunParameters(run);
1986 if (!entry)
1987 continue;
2bb7b766 1988
728290a4 1989 // DA test mode flag
1990 TString daTestModeString(aRow->GetField(2), aRow->GetFieldLength(2)); // field 2 = DA test mode flag
1991 Bool_t daTestMode = (Bool_t)daTestModeString.Atoi();
1992 entry->SetDATestMode(daTestMode);
1993
2bb7b766 1994 // loop on detectors
eba76848 1995 for(UInt_t ii = 0; ii < nCols; ii++)
1996 entry->SetDetectorStatus(aResult->GetFieldName(ii), aRow->GetField(ii));
2bb7b766 1997
eba76848 1998 entries.AddLast(entry);
2bb7b766 1999 delete aRow;
2000 }
2001
2bb7b766 2002 delete aResult;
2003 return kTRUE;
2004}
2005
2006//______________________________________________________________________________________________
eba76848 2007AliShuttleLogbookEntry* AliShuttle::QueryRunParameters(Int_t run)
2bb7b766 2008{
eba76848 2009 //
2010 // Retrieve run parameters written in the DAQ logbook and sets them into AliShuttleLogbookEntry object
2011 //
2bb7b766 2012
2013 // check connection, in case connect
efec19bd 2014 if (!Connect(4))
eba76848 2015 return 0;
2bb7b766 2016
2017 TString sqlQuery;
2c15234c 2018 sqlQuery.Form("select * from %s where run=%d", fConfig->GetDAQlbTable(), run);
2bb7b766 2019
efec19bd 2020 TSQLResult* aResult = fServer[4]->Query(sqlQuery);
2bb7b766 2021 if (!aResult) {
c88ad5db 2022 Log("SHUTTLE", Form("Can't execute query <%s>!", sqlQuery.Data()));
eba76848 2023 return 0;
2bb7b766 2024 }
2025
eba76848 2026 if (aResult->GetRowCount() == 0) {
2bb7b766 2027 Log("SHUTTLE", Form("QueryRunParameters - No entry in DAQ Logbook for run %d. Skipping", run));
2028 delete aResult;
eba76848 2029 return 0;
2bb7b766 2030 }
2031
eba76848 2032 if (aResult->GetRowCount() > 1) {
c88ad5db 2033 Log("SHUTTLE", Form("QueryRunParameters - UNEXPECTED: "
2034 "more than one entry in DAQ Logbook for run %d!", run));
2bb7b766 2035 delete aResult;
eba76848 2036 return 0;
2bb7b766 2037 }
2038
eba76848 2039 TSQLRow* aRow = aResult->Next();
2040 if (!aRow)
2041 {
c88ad5db 2042 Log("SHUTTLE", Form("QueryRunParameters - Could not retrieve row for run %d. Skipping", run));
eba76848 2043 delete aResult;
2044 return 0;
2045 }
2bb7b766 2046
eba76848 2047 AliShuttleLogbookEntry* entry = new AliShuttleLogbookEntry(run);
2bb7b766 2048
eba76848 2049 for (Int_t ii = 0; ii < aResult->GetFieldCount(); ii++)
2050 entry->SetRunParameter(aResult->GetFieldName(ii), aRow->GetField(ii));
2bb7b766 2051
c9db68c7 2052 delete aRow;
2053 delete aResult;
2054
eba76848 2055 UInt_t startTime = entry->GetStartTime();
2056 UInt_t endTime = entry->GetEndTime();
89bcf3b4 2057 Bool_t ecsSuccess = entry->GetECSSuccess();
1415d3f9 2058 TString runType = entry->GetRunType();
2059 TString tmpdaqstartTime = entry->GetRunParameter("DAQ_time_start");
f38b266c 2060 TString recordingFlagString = entry->GetRunParameter("GDCmStreamRecording");
2061 UInt_t recordingFlag = recordingFlagString.Atoi();
1415d3f9 2062 UInt_t daqstartTime = tmpdaqstartTime.Atoi();
c9db68c7 2063
af93e5d7 2064 UInt_t now = time(0);
2eb98462 2065 Int_t dcsDelay = fConfig->GetDCSDelay()+fConfig->GetDCSQueryOffset();
8e828d39 2066
2067 Bool_t skip = kFALSE;
2068
1415d3f9 2069 // runs are processed if
2070 // a) runType is PHYSICS and ecsSuccess is set
2071 // b) runType is not PHYSICS and (ecsSuccess is set or DAQ_time_start is non-0)
2072 // effectively this means that all runs are processed that started properly (ecsSucess behaviour is different for PHYSICS and non-PHYSICS runs (check with ECS!)
1bbcc21b 2073 if (startTime != 0 && endTime != 0) {
2074 if (endTime > startTime) {
2075 if (endTime >= now - dcsDelay) {
2076 Log("SHUTTLE", Form("Skipping run %d for now, because DCS buffer time is not yet expired", run));
2077 } else {
f38b266c 2078 if ((runType == "PHYSICS" || runType == "STANDALONE") && recordingFlag == 0){
2079 Log("SHUTTLE", Form("QueryRunParameters - Run type for run %d is %s but the recording is OFF - Skipping!", run, runType.Data()));
2080 skip = kTRUE;
2081 }
2082 else {
2083 if (runType == "PHYSICS") {
2084 if (ecsSuccess) {
2085 return entry;
2086 } else {
2087 Log("SHUTTLE", Form("QueryRunParameters - Run type for run %d is PHYSICS but ECS success flag not set (Reason = %s) - Skipping!", run, entry->GetRunParameter("eor_reason")));
2088 skip = kTRUE;
2089 }
1bbcc21b 2090 } else {
f38b266c 2091 if (ecsSuccess || daqstartTime > 0) {
2092 if (ecsSuccess == kFALSE)
2093 Log("SHUTTLE", Form("Processing run %d although in status ECS failure (Reason: %s), since run type != PHYSICS and DAQ_time_start != 0", run, entry->GetRunParameter("eor_reason")));
2094 return entry;
2095 } else {
2096 Log("SHUTTLE", Form("QueryRunParameters - Run type for run %d is %s, ECS success flag was not set (Reason = %s) and DAQ_time_start was NULL - Skipping!", run, runType.Data(), entry->GetRunParameter("eor_reason")));
2097 skip = kTRUE;
2098 }
1bbcc21b 2099 }
2100 }
2101 }
2102 } else {
2103 Log("SHUTTLE", Form("QueryRunParameters - Invalid parameters for run %d: startTime equal to endTime: %d %d - Skipping!", run, startTime, endTime));
2104 skip = kTRUE;
2105 }
1415d3f9 2106 } else {
2107 Log("SHUTTLE", Form("QueryRunParameters - Invalid parameters for Run %d: "
2108 "startTime = %d, endTime = %d. Skipping (Shuttle won't be marked as DONE)!",
2109 run, startTime, endTime));
89bcf3b4 2110 }
8e828d39 2111
2112 if (skip)
2113 {
2114 Log("SHUTTLE", Form("Marking SHUTTLE skipped for run %d", run));
2115 fLogbookEntry = entry;
2116 if (!UpdateShuttleLogbook("shuttle_skipped"))
2117 {
2118 AliError(Form("Could not update logbook for run %d !", run));
2119 }
bd23d48b 2120 if (!UpdateTableSkippedCase("ALL"))
2121 {
2122 AliError(Form("Could not update FXS tables for run %d !", run));
2123 }
8e828d39 2124 fLogbookEntry = 0;
2125 }
c9db68c7 2126
2127 delete entry;
2128 return 0;
2bb7b766 2129}
2130
a038aa70 2131//______________________________________________________________________________________________
2132TMap* AliShuttle::GetValueSet(const char* host, Int_t port, const TSeqCollection* entries,
542b6cc8 2133 DCSType type, Int_t multiSplit)
a038aa70 2134{
2135 // Retrieve all "entry" data points from the DCS server
2136 // host, port: TSocket connection parameters
2137 // entries: list of name of the alias or data point
2138 // type: kAlias or kDP
2139 // returns TMap of values, 0 when failure
542b6cc8 2140
2141 AliDCSClient client(host, port, fTimeout, fRetries, multiSplit);
b41b252a 2142
a038aa70 2143 TMap* result = 0;
b41b252a 2144 if (type == kAlias)
a038aa70 2145 {
5f3fe74e 2146 //result = client.GetAliasValues(entries, GetCurrentStartTime()-offset,
2147 // GetCurrentEndTime()+offset);
eff647d5 2148 result = client.GetAliasValues(entries, GetStartTimeDCSQuery(),
2149 GetEndTimeDCSQuery());
b41b252a 2150 }
2151 else if (type == kDP)
2152 {
5f3fe74e 2153 //result = client.GetDPValues(entries, GetCurrentStartTime()-offset,
2154 // GetCurrentEndTime()+offset);
eff647d5 2155 result = client.GetDPValues(entries, GetStartTimeDCSQuery(),
2156 GetEndTimeDCSQuery());
b41b252a 2157 }
a038aa70 2158
b41b252a 2159 if (result == 0)
2160 {
2161 Log(fCurrentDetector.Data(), Form("GetValueSet - Can't get entries! Reason: %s",
1790d4b7 2162 client.GetErrorString(client.GetResultErrorCode())));
2163 if (client.GetResultErrorCode() == AliDCSClient::fgkServerError)
2164 Log(fCurrentDetector.Data(), Form("GetValueSet - Server error code: %s",
2165 client.GetServerError().Data()));
a038aa70 2166
b41b252a 2167 return 0;
a038aa70 2168 }
b41b252a 2169
a038aa70 2170 return result;
2171}
b41b252a 2172
b948db8d 2173//______________________________________________________________________________________________
57f50b3c 2174const char* AliShuttle::GetFile(Int_t system, const char* detector,
2175 const char* id, const char* source)
b948db8d 2176{
9827400b 2177 // Get calibration file from file exchange servers
2178 // First queris the FXS database for the file name, using the run, detector, id and source info
2179 // then calls RetrieveFile(filename) for actual copy to local disk
2180 // run: current run being processed (given by Logbook entry fLogbookEntry)
2181 // detector: the Preprocessor name
2182 // id: provided as a parameter by the Preprocessor
2183 // source: provided by the Preprocessor through GetFileSources function
2184
2185 // check if test mode should simulate a FXS error
2186 if (fTestMode & kErrorFXSFiles)
2187 {
2188 Log(detector, Form("GetFile - In TESTMODE - Simulating error while connecting to %s FXS", GetSystemName(system)));
2189 return 0;
2190 }
2191
57f50b3c 2192 // check connection, in case connect
9d733021 2193 if (!Connect(system))
eba76848 2194 {
9d733021 2195 Log(detector, Form("GetFile - Couldn't connect to %s FXS database", GetSystemName(system)));
339adafa 2196 fFXSError = system;
57f50b3c 2197 return 0;
2198 }
2199
2200 // Query preparation
9d733021 2201 TString sourceName(source);
d386d623 2202 Int_t nFields = 3;
2203 TString sqlQueryStart = Form("select filePath,size,fileChecksum from %s where",
2204 fConfig->GetFXSdbTable(system));
2205 TString whereClause = Form("run=%d and detector=\"%s\" and fileId=\"%s\"",
2206 GetCurrentRun(), detector, id);
2207
efec19bd 2208 if (system == kDAQ || system == kDQM)
9d733021 2209 {
d386d623 2210 whereClause += Form(" and DAQsource=\"%s\"", source);
57f50b3c 2211 }
9d733021 2212 else if (system == kDCS)
eba76848 2213 {
9d733021 2214 sourceName="none";
57f50b3c 2215 }
9d733021 2216 else if (system == kHLT)
9e080f92 2217 {
d386d623 2218 whereClause += Form(" and DDLnumbers=\"%s\"", source);
9e080f92 2219 }
2220
9e080f92 2221 TString sqlQuery = Form("%s %s", sqlQueryStart.Data(), whereClause.Data());
2222
2223 AliDebug(2, Form("SQL query: \n%s",sqlQuery.Data()));
2224
2225 // Query execution
2226 TSQLResult* aResult = 0;
9d733021 2227 aResult = dynamic_cast<TSQLResult*> (fServer[system]->Query(sqlQuery));
9e080f92 2228 if (!aResult) {
8da81210 2229 Log(detector, Form("GetFile - Can't execute SQL query to %s database for: id = %s, source = %s",
9d733021 2230 GetSystemName(system), id, sourceName.Data()));
339adafa 2231 fFXSError = system;
9e080f92 2232 return 0;
2233 }
2234
8da81210 2235 if (aResult->GetRowCount() == 0)
9e080f92 2236 {
2237 Log(detector,
8da81210 2238 Form("GetFile - No entry in %s FXS db for: id = %s, source = %s",
9d733021 2239 GetSystemName(system), id, sourceName.Data()));
9e080f92 2240 delete aResult;
2241 return 0;
2242 }
2bb7b766 2243
9e080f92 2244 if (aResult->GetRowCount() > 1) {
2245 Log(detector,
8da81210 2246 Form("GetFile - More than one entry in %s FXS db for: id = %s, source = %s",
9d733021 2247 GetSystemName(system), id, sourceName.Data()));
339adafa 2248 fFXSError = system;
9e080f92 2249 delete aResult;
2250 return 0;
2251 }
2252
9d733021 2253 if (aResult->GetFieldCount() != nFields) {
9e080f92 2254 Log(detector,
9d733021 2255 Form("GetFileName - Wrong field count in %s FXS db for: id = %s, source = %s",
2256 GetSystemName(system), id, sourceName.Data()));
339adafa 2257 fFXSError = system;
9e080f92 2258 delete aResult;
2259 return 0;
2260 }
2261
2262 TSQLRow* aRow = dynamic_cast<TSQLRow*> (aResult->Next());
2263
2264 if (!aRow){
8da81210 2265 Log(detector, Form("GetFile - Empty set result in %s FXS db from query: id = %s, source = %s",
9d733021 2266 GetSystemName(system), id, sourceName.Data()));
339adafa 2267 fFXSError = system;
9e080f92 2268 delete aResult;
2269 return 0;
2270 }
2271
2272 TString filePath(aRow->GetField(0), aRow->GetFieldLength(0));
2273 TString fileSize(aRow->GetField(1), aRow->GetFieldLength(1));
d386d623 2274 TString fileChecksum(aRow->GetField(2), aRow->GetFieldLength(2));
9e080f92 2275
2276 delete aResult;
2277 delete aRow;
2278
d386d623 2279 AliDebug(2, Form("filePath = %s; size = %s, fileChecksum = %s",
2280 filePath.Data(), fileSize.Data(), fileChecksum.Data()));
9e080f92 2281
9e080f92 2282 // retrieved file is renamed to make it unique
3ebbf3c4 2283 Int_t dir_lev1 = GetCurrentRun()/10000;
2284 TString localFileName = Form("%s/%d/%d/%s_process/%s_%s_%d_%s_%s.shuttle",
2285 GetShuttleTempDir(), dir_lev1, GetCurrentRun(), detector,
d524ade6 2286 GetSystemName(system), detector, GetCurrentRun(),
2287 id, sourceName.Data());
3ebbf3c4 2288 Log("SHUTTLE",Form("file from FXS = %s",localFileName.Data()));
9d733021 2289
9e080f92 2290
9d733021 2291 // file retrieval from FXS
4b95672b 2292 UInt_t nRetries = 0;
2293 UInt_t maxRetries = 3;
2294 Bool_t result = kFALSE;
2295
2296 // copy!! if successful TSystem::Exec returns 0
8da81210 2297 while (nRetries++ < maxRetries) {
4b95672b 2298 AliDebug(2, Form("Trying to copy file. Retry # %d", nRetries));
2299 result = RetrieveFile(system, filePath.Data(), localFileName.Data());
8da81210 2300 if (!result)
4b95672b 2301 {
8da81210 2302 Log(detector, Form("GetFile - Copy of file %s from %s FXS failed",
9d733021 2303 filePath.Data(), GetSystemName(system)));
4b95672b 2304 continue;
4f0749a8 2305 }
9e080f92 2306
c98417a1 2307 if (fileSize.Length()>0)
2308 {
2309 // compare filesize of local file with the one stored in the FXS DB
926cbc0e 2310 Long_t size = -1;
2311 Int_t sizeComp = gSystem->GetPathInfo(localFileName.Data(), 0, &size, 0, 0);
c98417a1 2312
926cbc0e 2313 if (sizeComp != 0 || size != fileSize.Atoi())
c98417a1 2314 {
8da81210 2315 Log(detector, Form("GetFile - size of file %s does not match with local copy!",
c98417a1 2316 filePath.Data()));
2317 result = kFALSE;
2318 continue;
2319 }
2320
2321 } else {
2322 Log(fCurrentDetector, Form("GetFile - size of file %s not set in %s database, skipping comparison",
2323 filePath.Data(), GetSystemName(system)));
2324 }
2325
d386d623 2326 if (fileChecksum.Length()>0)
4b95672b 2327 {
2328 // compare md5sum of local file with the one stored in the FXS DB
95d4aaef 2329 if(fileChecksum.Contains(' ')) fileChecksum.Resize(fileChecksum.First(' '));
339adafa 2330 Int_t md5Comp = gSystem->Exec(Form("md5sum %s |grep %s > /dev/null 2> /dev/null",
d524ade6 2331 localFileName.Data(), fileChecksum.Data()));
9e080f92 2332
4b95672b 2333 if (md5Comp != 0)
2334 {
8da81210 2335 Log(detector, Form("GetFile - md5sum of file %s does not match with local copy!",
4b95672b 2336 filePath.Data()));
2337 result = kFALSE;
2338 continue;
2339 }
d386d623 2340 } else {
2341 Log(fCurrentDetector, Form("GetFile - md5sum of file %s not set in %s database, skipping comparison",
2342 filePath.Data(), GetSystemName(system)));
9d733021 2343 }
4b95672b 2344 if (result) break;
9e080f92 2345 }
2346
8da81210 2347 if (!result)
2348 {
339adafa 2349 fFXSError = system;
8da81210 2350 return 0;
2351 }
4b95672b 2352
9d733021 2353 fFXSCalled[system]=kTRUE;
2354 TObjString *fileParams = new TObjString(Form("%s#!?!#%s", id, sourceName.Data()));
2355 fFXSlist[system].Add(fileParams);
9e080f92 2356
675f64cd 2357 static TString staticLocalFileName;
2358 staticLocalFileName.Form("%s", localFileName.Data());
2359
c88ad5db 2360 Log(fCurrentDetector, Form("GetFile - Retrieved file with id %s and "
2361 "source %s from %s to %s", id, source,
d524ade6 2362 GetSystemName(system), localFileName.Data()));
675f64cd 2363
d524ade6 2364 return staticLocalFileName.Data();
2bb7b766 2365}
2366
2367//______________________________________________________________________________________________
9d733021 2368Bool_t AliShuttle::RetrieveFile(UInt_t system, const char* fxsFileName, const char* localFileName)
9e080f92 2369{
9827400b 2370 //
2371 // Copies file from FXS to local Shuttle machine
2372 //
2bb7b766 2373
9e080f92 2374 // check temp directory: trying to cd to temp; if it does not exist, create it
d524ade6 2375 AliDebug(2, Form("Copy file %s from %s FXS into %s",
2376 GetSystemName(system), fxsFileName, localFileName));
2377
2378 TString tmpDir(localFileName);
2379
2380 tmpDir = tmpDir(0,tmpDir.Last('/'));
9e080f92 2381
d524ade6 2382 Int_t noDir = gSystem->GetPathInfo(tmpDir.Data(), 0, (Long64_t*) 0, 0, 0);
2383 if (noDir) // temp dir does not exists!
2384 {
2385 if (gSystem->mkdir(tmpDir.Data(), 1))
2386 {
2387 Log(fCurrentDetector.Data(), "RetrieveFile - could not make temp directory!!");
9e080f92 2388 return kFALSE;
2389 }
9e080f92 2390 }
2391
339adafa 2392 TString command = Form("scp -oPort=%d -2 %s@%s:%s/%s %s",
9d733021 2393 fConfig->GetFXSPort(system),
2394 fConfig->GetFXSUser(system),
2395 fConfig->GetFXSHost(system),
339adafa 2396 fConfig->GetFXSBaseFolder(system),
9d733021 2397 fxsFileName,
9e080f92 2398 localFileName);
2399
2400 AliDebug(2, Form("%s",command.Data()));
2401
4b95672b 2402 Bool_t result = (gSystem->Exec(command.Data()) == 0);
9e080f92 2403
4b95672b 2404 return result;
9e080f92 2405}
2406
2407//______________________________________________________________________________________________
9d733021 2408TList* AliShuttle::GetFileSources(Int_t system, const char* detector, const char* id)
2409{
9827400b 2410 //
2411 // Get sources producing the condition file Id from file exchange servers
4a33bdd9 2412 // if id is NULL all sources are returned (distinct)
9827400b 2413 //
1bcd28db 2414
8884d71e 2415 if (id)
2416 {
2417 Log(detector, Form("GetFileSources - Querying %s FXS for files with id %s produced by %s", GetSystemName(system), id, detector));
2418 } else {
2419 Log(detector, Form("GetFileSources - Querying %s FXS for files produced by %s", GetSystemName(system), detector));
2420 }
9827400b 2421
2422 // check if test mode should simulate a FXS error
2423 if (fTestMode & kErrorFXSSources)
2424 {
2425 Log(detector, Form("GetFileSources - In TESTMODE - Simulating error while connecting to %s FXS", GetSystemName(system)));
2426 return 0;
2427 }
2428
9d733021 2429 if (system == kDCS)
2430 {
c88ad5db 2431 Log(detector, "GetFileSources - WARNING: DCS system has only one source of data!");
6297b37d 2432 TList *list = new TList();
2433 list->SetOwner(1);
2434 list->Add(new TObjString(" "));
2435 return list;
9d733021 2436 }
9e080f92 2437
2438 // check connection, in case connect
9d733021 2439 if (!Connect(system))
2440 {
4a33bdd9 2441 Log(detector, Form("GetFileSources - Couldn't connect to %s FXS database", GetSystemName(system)));
339adafa 2442 fFXSError = system;
9d733021 2443 return NULL;
9e080f92 2444 }
2445
b827f957 2446 TString sourceName = "";
efec19bd 2447 if (system == kDAQ || system == kDQM)
9d733021 2448 {
2449 sourceName = "DAQsource";
2450 } else if (system == kHLT)
2451 {
2452 sourceName = "DDLnumbers";
2453 }
2454
4a33bdd9 2455 TString sqlQueryStart = Form("select distinct %s from %s where", sourceName.Data(), fConfig->GetFXSdbTable(system));
2456 TString whereClause = Form("run=%d and detector=\"%s\"",
2457 GetCurrentRun(), detector);
2458 if (id)
2459 whereClause += Form(" and fileId=\"%s\"", id);
9e080f92 2460 TString sqlQuery = Form("%s %s", sqlQueryStart.Data(), whereClause.Data());
2461
2462 AliDebug(2, Form("SQL query: \n%s",sqlQuery.Data()));
2463
2464 // Query execution
2465 TSQLResult* aResult;
9d733021 2466 aResult = fServer[system]->Query(sqlQuery);
9e080f92 2467 if (!aResult) {
9d733021 2468 Log(detector, Form("GetFileSources - Can't execute SQL query to %s database for id: %s",
2469 GetSystemName(system), id));
339adafa 2470 fFXSError = system;
9e080f92 2471 return 0;
2472 }
2473
86aa42c3 2474 TList *list = new TList();
2475 list->SetOwner(1);
2476
9d733021 2477 if (aResult->GetRowCount() == 0)
2478 {
9e080f92 2479 Log(detector,
9d733021 2480 Form("GetFileSources - No entry in %s FXS table for id: %s", GetSystemName(system), id));
9e080f92 2481 delete aResult;
86aa42c3 2482 return list;
9e080f92 2483 }
2484
1bcd28db 2485 Log(detector, Form("GetFileSources - Found %d sources", aResult->GetRowCount()));
9e080f92 2486
1bcd28db 2487 TSQLRow* aRow;
9d733021 2488 while ((aRow = aResult->Next()))
2489 {
9e080f92 2490
9d733021 2491 TString source(aRow->GetField(0), aRow->GetFieldLength(0));
2492 AliDebug(2, Form("%s = %s", sourceName.Data(), source.Data()));
2493 list->Add(new TObjString(source));
9e080f92 2494 delete aRow;
2495 }
9d733021 2496
9e080f92 2497 delete aResult;
2498
2499 return list;
2bb7b766 2500}
2501
4a33bdd9 2502//______________________________________________________________________________________________
2503TList* AliShuttle::GetFileIDs(Int_t system, const char* detector, const char* source)
2504{
2505 //
2506 // Get all ids of condition files produced by a given source from file exchange servers
2507 //
2508
1bcd28db 2509 Log(detector, Form("GetFileIDs - Retrieving ids with source %s with %s", source, GetSystemName(system)));
2510
4a33bdd9 2511 // check if test mode should simulate a FXS error
2512 if (fTestMode & kErrorFXSSources)
2513 {
2514 Log(detector, Form("GetFileIDs - In TESTMODE - Simulating error while connecting to %s FXS", GetSystemName(system)));
2515 return 0;
2516 }
2517
2518 // check connection, in case connect
2519 if (!Connect(system))
2520 {
2521 Log(detector, Form("GetFileIDs - Couldn't connect to %s FXS database", GetSystemName(system)));
2522 return NULL;
2523 }
2524
b827f957 2525 TString sourceName = "";
4a33bdd9 2526 if (system == kDAQ)
2527 {
2528 sourceName = "DAQsource";
2529 } else if (system == kHLT)
2530 {
2531 sourceName = "DDLnumbers";
2532 }
2533
2534 TString sqlQueryStart = Form("select fileId from %s where", fConfig->GetFXSdbTable(system));
2535 TString whereClause = Form("run=%d and detector=\"%s\"",
2536 GetCurrentRun(), detector);
2537 if (sourceName.Length() > 0 && source)
2538 whereClause += Form(" and %s=\"%s\"", sourceName.Data(), source);
2539 TString sqlQuery = Form("%s %s", sqlQueryStart.Data(), whereClause.Data());
2540
2541 AliDebug(2, Form("SQL query: \n%s",sqlQuery.Data()));
2542
2543 // Query execution
2544 TSQLResult* aResult;
2545 aResult = fServer[system]->Query(sqlQuery);
2546 if (!aResult) {
2547 Log(detector, Form("GetFileIDs - Can't execute SQL query to %s database for source: %s",
2548 GetSystemName(system), source));
2549 return 0;
2550 }
2551
2552 TList *list = new TList();
2553 list->SetOwner(1);
2554
2555 if (aResult->GetRowCount() == 0)
2556 {
2557 Log(detector,
2558 Form("GetFileIDs - No entry in %s FXS table for source: %s", GetSystemName(system), source));
2559 delete aResult;
2560 return list;
2561 }
2562
1bcd28db 2563 Log(detector, Form("GetFileIDs - Found %d ids", aResult->GetRowCount()));
2564
4a33bdd9 2565 TSQLRow* aRow;
2566
2567 while ((aRow = aResult->Next()))
2568 {
2569
2570 TString id(aRow->GetField(0), aRow->GetFieldLength(0));
2571 AliDebug(2, Form("fileId = %s", id.Data()));
2572 list->Add(new TObjString(id));
2573 delete aRow;
2574 }
2575
2576 delete aResult;
2577
2578 return list;
2579}
2580
2bb7b766 2581//______________________________________________________________________________________________
9d733021 2582Bool_t AliShuttle::Connect(Int_t system)
2bb7b766 2583{
9827400b 2584 // Connect to MySQL Server of the system's FXS MySQL databases
2585 // DAQ Logbook, Shuttle Logbook and DAQ FXS db are on the same host
2586 //
57f50b3c 2587
9d733021 2588 // check connection: if already connected return
dd84c76d 2589
3a4ddf0e 2590 if(fServer[system] && fServer[system]->IsConnected()) {
dd84c76d 2591 // ping the server
2592 if (fServer[system]->PingVerify()==kTRUE){ // connection is still alive
2593 return kTRUE;
2594 }
2595 else{
2596 AliWarning(Form("Connection got lost to FXS database for %s. Closing and reconnecting.",
2597 AliShuttleInterface::GetSystemName(system)));
2598 fServer[system]->Close();
2599 delete fServer[system];
2600 fServer[system] = 0x0;
2601 }
3a4ddf0e 2602 }
57f50b3c 2603
9d733021 2604 TString dbHost, dbUser, dbPass, dbName;
57f50b3c 2605
efec19bd 2606 if (system < 4) // FXS db servers
9d733021 2607 {
2608 dbHost = Form("mysql://%s:%d", fConfig->GetFXSdbHost(system), fConfig->GetFXSdbPort(system));
2609 dbUser = fConfig->GetFXSdbUser(system);
2610 dbPass = fConfig->GetFXSdbPass(system);
2611 dbName = fConfig->GetFXSdbName(system);
2612 } else { // Run & Shuttle logbook servers
2613 // TODO Will the Shuttle logbook server be the same as the Run logbook server ???
2614 dbHost = Form("mysql://%s:%d", fConfig->GetDAQlbHost(), fConfig->GetDAQlbPort());
2615 dbUser = fConfig->GetDAQlbUser();
2616 dbPass = fConfig->GetDAQlbPass();
2617 dbName = fConfig->GetDAQlbDB();
2618 }
57f50b3c 2619
9d733021 2620 fServer[system] = TSQLServer::Connect(dbHost.Data(), dbUser.Data(), dbPass.Data());
efec19bd 2621 if (!fServer[system] || !fServer[system]->IsConnected()) {
2622 if(system < 4)
9d733021 2623 {
2624 AliError(Form("Can't establish connection to FXS database for %s",
2625 AliShuttleInterface::GetSystemName(system)));
2626 } else {
2627 AliError("Can't establish connection to Run logbook.");
57f50b3c 2628 }
9d733021 2629 if(fServer[system]) delete fServer[system];
2630 return kFALSE;
2bb7b766 2631 }
57f50b3c 2632
9d733021 2633 // Get tables
2634 TSQLResult* aResult=0;
2635 switch(system){
2636 case kDAQ:
2637 aResult = fServer[kDAQ]->GetTables(dbName.Data());
2638 break;
2639 case kDCS:
2640 aResult = fServer[kDCS]->GetTables(dbName.Data());
2641 break;
2642 case kHLT:
2643 aResult = fServer[kHLT]->GetTables(dbName.Data());
2644 break;
efec19bd 2645 case kDQM:
2646 aResult = fServer[kDQM]->GetTables(dbName.Data());
2647 break;
9d733021 2648 default:
efec19bd 2649 aResult = fServer[4]->GetTables(dbName.Data());
9d733021 2650 break;
2651 }
2652
2653 delete aResult;
2bb7b766 2654 return kTRUE;
2655}
57f50b3c 2656
9e080f92 2657//______________________________________________________________________________________________
9d733021 2658Bool_t AliShuttle::UpdateTable()
9e080f92 2659{
9827400b 2660 //
2661 // Update FXS table filling time_processed field in all rows corresponding to current run and detector
2662 //
9e080f92 2663
9d733021 2664 Bool_t result = kTRUE;
9e080f92 2665
6e25f828 2666 for (UInt_t system=0; system<4; system++)
9d733021 2667 {
2668 if(!fFXSCalled[system]) continue;
9e080f92 2669
9d733021 2670 // check connection, in case connect
2671 if (!Connect(system))
2672 {
2673 Log(fCurrentDetector, Form("UpdateTable - Couldn't connect to %s FXS database", GetSystemName(system)));
2674 result = kFALSE;
2675 continue;
9e080f92 2676 }
9e080f92 2677
9d733021 2678 TTimeStamp now; // now
2679
2680 // Loop on FXS list entries
2681 TIter iter(&fFXSlist[system]);
2682 TObjString *aFXSentry=0;
2683 while ((aFXSentry = dynamic_cast<TObjString*> (iter.Next())))
2684 {
2685 TString aFXSentrystr = aFXSentry->String();
2686 TObjArray *aFXSarray = aFXSentrystr.Tokenize("#!?!#");
2687 if (!aFXSarray || aFXSarray->GetEntries() != 2 )
2688 {
2689 Log(fCurrentDetector, Form("UpdateTable - error updating %s FXS entry. Check string: <%s>",
2690 GetSystemName(system), aFXSentrystr.Data()));
2691 if(aFXSarray) delete aFXSarray;
2692 result = kFALSE;
2693 continue;
2694 }
2695 const char* fileId = ((TObjString*) aFXSarray->At(0))->GetName();
2696 const char* source = ((TObjString*) aFXSarray->At(1))->GetName();
2697
2698 TString whereClause;
efec19bd 2699 if (system == kDAQ || system == kDQM)
9d733021 2700 {
2701 whereClause = Form("where run=%d and detector=\"%s\" and fileId=\"%s\" and DAQsource=\"%s\";",
2702 GetCurrentRun(), fCurrentDetector.Data(), fileId, source);
2703 }
2704 else if (system == kDCS)
2705 {
2706 whereClause = Form("where run=%d and detector=\"%s\" and fileId=\"%s\";",
2707 GetCurrentRun(), fCurrentDetector.Data(), fileId);
2708 }
2709 else if (system == kHLT)
2710 {
2711 whereClause = Form("where run=%d and detector=\"%s\" and fileId=\"%s\" and DDLnumbers=\"%s\";",
2712 GetCurrentRun(), fCurrentDetector.Data(), fileId, source);
2713 }
2714
2715 delete aFXSarray;
9e080f92 2716
0a0b087a 2717 TString sqlQuery = Form("update %s set time_processed=%ld %s", fConfig->GetFXSdbTable(system),
2718 (ULong_t)now.GetSec(), whereClause.Data());
9e080f92 2719
9d733021 2720 AliDebug(2, Form("SQL query: \n%s",sqlQuery.Data()));
9e080f92 2721
9d733021 2722 // Query execution
2723 TSQLResult* aResult;
2724 aResult = dynamic_cast<TSQLResult*> (fServer[system]->Query(sqlQuery));
2725 if (!aResult)
2726 {
2727 Log(fCurrentDetector, Form("UpdateTable - %s db: can't execute SQL query <%s>",
2728 GetSystemName(system), sqlQuery.Data()));
2729 result = kFALSE;
2730 continue;
2731 }
2732 delete aResult;
9e080f92 2733 }
9e080f92 2734 }
2735
9d733021 2736 return result;
9e080f92 2737}
57f50b3c 2738
bd23d48b 2739//_______________________________________________________________________________
2740Bool_t AliShuttle::UpdateTableSkippedCase(const char* detector)
2741{
2742 //
2743 // Update FXS table filling time_processed field in all rows corresponding to current run and detector
2744 // if detector = "ALL" update all detectors
2745 //
2746
2747 Bool_t result = kTRUE;
2748
7a782903 2749 TString detName(detector);
2750
6e25f828 2751 for (UInt_t system=0; system<4; system++)
bd23d48b 2752 {
2753
2754 // check connection, in case connect
2755 if (!Connect(system))
2756 {
2757 Log(fCurrentDetector, Form("UpdateTableSkippedCase - Couldn't connect to %s FXS database", GetSystemName(system)));
2758 result = kFALSE;
2759 continue;
2760 }
2761
2762 TTimeStamp now; // now
2763
2764 // Loop on FXS list entries
2765 TIter iter(&fFXSlist[system]);
2766
2767 TString whereClause;
7a782903 2768 if (detName == "ALL") whereClause = Form("where run=%d and time_processed IS NULL;",GetCurrentRun());
bd23d48b 2769 else whereClause = Form("where run=%d and detector=\"%s\" and time_processed IS NULL;",GetCurrentRun(), detector);
2770
4be2f262 2771 //Log("SHUTTLE",Form(" whereClause = %s ",whereClause.Data()));
bd23d48b 2772
0a0b087a 2773 TString sqlQuery = Form("update %s set time_processed=%ld %s", fConfig->GetFXSdbTable(system),
2774 (ULong_t)now.GetSec(), whereClause.Data());
bd23d48b 2775
2776 AliDebug(2, Form("SQL query: \n%s",sqlQuery.Data()));
2777
2778 // Query execution
2779 TSQLResult* aResult;
2780 aResult = dynamic_cast<TSQLResult*> (fServer[system]->Query(sqlQuery));
2781 if (!aResult)
2782 {
2783 Log("SHUTTLE", Form("UpdateTableSkippedCase - %s db: can't execute SQL query <%s>",
2784 GetSystemName(system), sqlQuery.Data()));
2785 result = kFALSE;
2786 continue;
2787 }
2788 delete aResult;
2789
2790 }
2791
2792 return result;
2793}
3301427a 2794//______________________________________________________________________________________________
2795Bool_t AliShuttle::UpdateTableFailCase()
2796{
9827400b 2797 // Update FXS table filling time_processed field in all rows corresponding to current run and detector
2798 // this is called in case the preprocessor is declared failed for the current run, because
2799 // the fields are updated only in case of success
3301427a 2800
2801 Bool_t result = kTRUE;
2802
6e25f828 2803 for (UInt_t system=0; system<4; system++)
3301427a 2804 {
2805 // check connection, in case connect
2806 if (!Connect(system))
2807 {
2808 Log(fCurrentDetector, Form("UpdateTableFailCase - Couldn't connect to %s FXS database",
2809 GetSystemName(system)));
2810 result = kFALSE;
2811 continue;
2812 }
2813
2814 TTimeStamp now; // now
2815
2816 // Loop on FXS list entries
2817
2818 TString whereClause = Form("where run=%d and detector=\"%s\";",
2819 GetCurrentRun(), fCurrentDetector.Data());
2820
2821
0a0b087a 2822 TString sqlQuery = Form("update %s set time_processed=%ld %s", fConfig->GetFXSdbTable(system),
2823 (ULong_t)now.GetSec(), whereClause.Data());
3301427a 2824
2825 AliDebug(2, Form("SQL query: \n%s",sqlQuery.Data()));
2826
2827 // Query execution
2828 TSQLResult* aResult;
2829 aResult = dynamic_cast<TSQLResult*> (fServer[system]->Query(sqlQuery));
2830 if (!aResult)
2831 {
2832 Log(fCurrentDetector, Form("UpdateTableFailCase - %s db: can't execute SQL query <%s>",
2833 GetSystemName(system), sqlQuery.Data()));
2834 result = kFALSE;
2835 continue;
2836 }
2837 delete aResult;
2838 }
2839
2840 return result;
2841}
2842
2bb7b766 2843//______________________________________________________________________________________________
2844Bool_t AliShuttle::UpdateShuttleLogbook(const char* detector, const char* status)
2845{
e7f62f16 2846 //
2847 // Update Shuttle logbook filling detector or shuttle_done column
2848 // ex. of usage: UpdateShuttleLogbook("PHOS", "DONE") or UpdateShuttleLogbook("shuttle_done")
2849 //
57f50b3c 2850
2bb7b766 2851 // check connection, in case connect
efec19bd 2852 if(!Connect(4)){
2bb7b766 2853 Log("SHUTTLE", "UpdateShuttleLogbook - Couldn't connect to DAQ Logbook.");
2854 return kFALSE;
57f50b3c 2855 }
2856
2bb7b766 2857 TString detName(detector);
2858 TString setClause;
8e828d39 2859 if (detName == "shuttle_done" || detName == "shuttle_skipped")
e7f62f16 2860 {
2bb7b766 2861 setClause = "set shuttle_done=1";
8e828d39 2862
ee6f7523 2863 if (detName == "shuttle_done")
b0e53b15 2864 {
8e828d39 2865 if (TouchFile() != kTRUE)
2156c755 2866 {
2867 SendMLRunInfo("Pending");
503ff24f 2868 return kFALSE;
2156c755 2869 }
8e828d39 2870
2871 SendMLRunInfo("Done");
b0e53b15 2872 }
8e828d39 2873 else
2874 SendMLRunInfo("Skipped");
2875 }
2876 else {
2bb7b766 2877 TString statusStr(status);
2878 if(statusStr.Contains("done", TString::kIgnoreCase) ||
2879 statusStr.Contains("failed", TString::kIgnoreCase)){
eba76848 2880 setClause = Form("set %s=\"%s\"", detector, status);
2bb7b766 2881 } else {
2882 Log("SHUTTLE",
2883 Form("UpdateShuttleLogbook - Invalid status <%s> for detector %s",
2884 status, detector));
2885 return kFALSE;
2886 }
2887 }
57f50b3c 2888
2bb7b766 2889 TString whereClause = Form("where run=%d", GetCurrentRun());
2890
441b0e9c 2891 TString sqlQuery = Form("update %s %s %s",
2892 fConfig->GetShuttlelbTable(), setClause.Data(), whereClause.Data());
57f50b3c 2893
2bb7b766 2894 AliDebug(2, Form("SQL query: \n%s",sqlQuery.Data()));
2895
2896 // Query execution
2897 TSQLResult* aResult;
efec19bd 2898 aResult = dynamic_cast<TSQLResult*> (fServer[4]->Query(sqlQuery));
2bb7b766 2899 if (!aResult) {
2900 Log("SHUTTLE", Form("UpdateShuttleLogbook - Can't execute query <%s>", sqlQuery.Data()));
2901 return kFALSE;
57f50b3c 2902 }
2bb7b766 2903 delete aResult;
57f50b3c 2904
2905 return kTRUE;
2906}
2907
2908//______________________________________________________________________________________________
2bb7b766 2909Int_t AliShuttle::GetCurrentRun() const
2910{
9827400b 2911 //
2912 // Get current run from logbook entry
2913 //
57f50b3c 2914
2bb7b766 2915 return fLogbookEntry ? fLogbookEntry->GetRun() : -1;
57f50b3c 2916}
2917
2918//______________________________________________________________________________________________
2bb7b766 2919UInt_t AliShuttle::GetCurrentStartTime() const
2920{
9827400b 2921 //
2922 // get current start time
2923 //
57f50b3c 2924
2bb7b766 2925 return fLogbookEntry ? fLogbookEntry->GetStartTime() : 0;
57f50b3c 2926}
2927
2928//______________________________________________________________________________________________
2bb7b766 2929UInt_t AliShuttle::GetCurrentEndTime() const
2930{
9827400b 2931 //
2932 // get current end time from logbook entry
2933 //
57f50b3c 2934
2bb7b766 2935 return fLogbookEntry ? fLogbookEntry->GetEndTime() : 0;
57f50b3c 2936}
675f64cd 2937//______________________________________________________________________________________________
2938UInt_t AliShuttle::GetCurrentYear() const
2939{
2940 //
2941 // Get current year from logbook entry
2942 //
2943
2944 if (!fLogbookEntry) return 0;
2945
2946 TTimeStamp startTime(GetCurrentStartTime());
2947 TString year = Form("%d",startTime.GetDate());
2948 year = year(0,4);
2949
2950 return year.Atoi();
2951}
2952
2953//______________________________________________________________________________________________
2954const char* AliShuttle::GetLHCPeriod() const
2955{
2956 //
2957 // Get current LHC period from logbook entry
2958 //
2959
2960 if (!fLogbookEntry) return 0;
2961
2962 return fLogbookEntry->GetRunParameter("LHCperiod");
2963}
2964
b948db8d 2965//______________________________________________________________________________________________
6a150f83 2966void AliShuttle::Log(const char* detector, const char* message, UInt_t level)
b948db8d 2967{
9827400b 2968 //
2969 // Fill log string with a message
2970 //
7d4cf768 2971
3ebbf3c4 2972 TString logRunDir = GetShuttleLogDir();
2973 if (GetCurrentRun() >=0) {
2974 Int_t logDir_lev1 = GetCurrentRun()/10000;
2975 logRunDir += Form("/%d/%d", logDir_lev1, GetCurrentRun());
2976 }
7d4cf768 2977 void* dir = gSystem->OpenDirectory(logRunDir.Data());
84090f85 2978 if (dir == NULL) {
7d4cf768 2979 if (gSystem->mkdir(logRunDir.Data(), kTRUE)) {
36c99a6a 2980 AliError(Form("Can't open directory <%s>", GetShuttleLogDir()));
84090f85 2981 return;
2982 }
b948db8d 2983
84090f85 2984 } else {
2985 gSystem->FreeDirectory(dir);
2986 }
b948db8d 2987
0b025a52 2988 TString toLog = Form("%s UTC (%d): %s - ", TTimeStamp(time(0)).AsString("s"), getpid(), detector);
e7f62f16 2989 if (GetCurrentRun() >= 0)
2990 toLog += Form("run %d - ", GetCurrentRun());
2bb7b766 2991 toLog += Form("%s", message);
2992
6a150f83 2993 AliLog::Message(level, toLog, MODULENAME(), ClassName(), FUNCTIONNAME(), __FILE__, __LINE__);
ffa29e93 2994
2995 // if we redirect the log output already to the file, leave here
2996 if (fOutputRedirected && strcmp(detector, "SHUTTLE") != 0)
2997 return;
b948db8d 2998
ffa29e93 2999 TString fileName = GetLogFileName(detector);
e7f62f16 3000
84090f85 3001 gSystem->ExpandPathName(fileName);
3002
3003 ofstream logFile;
3004 logFile.open(fileName, ofstream::out | ofstream::app);
3005
3006 if (!logFile.is_open()) {
3007 AliError(Form("Could not open file %s", fileName.Data()));
3008 return;
3009 }
7bfb2090 3010
84090f85 3011 logFile << toLog.Data() << "\n";
b948db8d 3012
84090f85 3013 logFile.close();
b948db8d 3014}
2bb7b766 3015
ffa29e93 3016//______________________________________________________________________________________________
3017TString AliShuttle::GetLogFileName(const char* detector) const
3018{
3019 //
3020 // returns the name of the log file for a given sub detector
3021 //
3022
3023 TString fileName;
3024
3025 if (GetCurrentRun() >= 0)
7d4cf768 3026 {
3ebbf3c4 3027 Int_t logDir_lev1 = GetCurrentRun()/10000;
3028 fileName.Form("%s/%d/%d/%s.log", GetShuttleLogDir(), logDir_lev1, GetCurrentRun(),
3029 detector);
7d4cf768 3030 } else {
ffa29e93 3031 fileName.Form("%s/%s.log", GetShuttleLogDir(), detector);
7d4cf768 3032 }
ffa29e93 3033
3034 return fileName;
3035}
3036
ee6f7523 3037//______________________________________________________________________________________________
3038void AliShuttle::SendAlive()
3039{
3040 // sends alive message to ML
3041
3042 TMonaLisaText mlStatus("SHUTTLE_status", "Alive");
3043
3044 TList mlList;
3045 mlList.Add(&mlStatus);
3046
3047 fMonaLisa->SendParameters(&mlList, "__PROCESSINGINFO__");
3048}
3049
2bb7b766 3050//______________________________________________________________________________________________
3051Bool_t AliShuttle::Collect(Int_t run)
3052{
9827400b 3053 //
3054 // Collects conditions data for all UNPROCESSED run written to DAQ LogBook in case of run = -1 (default)
3055 // If a dedicated run is given this run is processed
3056 //
3057 // In operational mode, this is the Shuttle function triggered by the EOR signal.
3058 //
2bb7b766 3059
eba76848 3060 if (run == -1)
3061 Log("SHUTTLE","Collect - Shuttle called. Collecting conditions data for unprocessed runs");
3062 else
3063 Log("SHUTTLE", Form("Collect - Shuttle called. Collecting conditions data for run %d", run));
cb343cfd 3064
3065 SetLastAction("Starting");
2bb7b766 3066
ee6f7523 3067 // create ML instance
3068 if (!fMonaLisa)
3069 fMonaLisa = new TMonaLisaWriter(fConfig->GetMonitorHost(), fConfig->GetMonitorTable());
3070
fbc112e3 3071 CountOpenRuns();
ee6f7523 3072
2bb7b766 3073 TString whereClause("where shuttle_done=0");
eba76848 3074 if (run != -1)
3075 whereClause += Form(" and run=%d", run);
2bb7b766 3076
3077 TObjArray shuttleLogbookEntries;
be48e3ea 3078 if (!QueryShuttleLogbook(whereClause, shuttleLogbookEntries))
3079 {
cb343cfd 3080 Log("SHUTTLE", "Collect - Can't retrieve entries from Shuttle logbook");
2bb7b766 3081 return kFALSE;
3082 }
3083
9e080f92 3084 if (shuttleLogbookEntries.GetEntries() == 0)
3085 {
3086 if (run == -1)
3087 Log("SHUTTLE","Collect - Found no UNPROCESSED runs in Shuttle logbook");
3088 else
3089 Log("SHUTTLE", Form("Collect - Run %d is already DONE "
3090 "or it does not exist in Shuttle logbook", run));
3091 return kTRUE;
3092 }
3093
be48e3ea 3094 for (UInt_t iDet=0; iDet<NDetectors(); iDet++)
3095 fFirstUnprocessed[iDet] = kTRUE;
3096
fc5a4708 3097 if (run != -1)
be48e3ea 3098 {
3099 // query Shuttle logbook for earlier runs, check if some detectors are unprocessed,
3100 // flag them into fFirstUnprocessed array
e3ae8d33 3101 TString whereClauseBis(Form("where shuttle_done=0 and run < %d", run));
be48e3ea 3102 TObjArray tmpLogbookEntries;
e3ae8d33 3103 if (!QueryShuttleLogbook(whereClauseBis, tmpLogbookEntries))
be48e3ea 3104 {
3105 Log("SHUTTLE", "Collect - Can't retrieve entries from Shuttle logbook");
3106 return kFALSE;
3107 }
3108
3109 TIter iter(&tmpLogbookEntries);
3110 AliShuttleLogbookEntry* anEntry = 0;
3111 while ((anEntry = dynamic_cast<AliShuttleLogbookEntry*> (iter.Next())))
3112 {
3113 for (UInt_t iDet=0; iDet<NDetectors(); iDet++)
3114 {
3115 if (anEntry->GetDetectorStatus(iDet) == AliShuttleLogbookEntry::kUnprocessed)
3116 {
3117 AliDebug(2, Form("Run %d: setting %s as \"not first time unprocessed\"",
3118 anEntry->GetRun(), GetDetName(iDet)));
3119 fFirstUnprocessed[iDet] = kFALSE;
3120 }
3121 }
3122
3123 }
3124
3125 }
3126
3127 if (!RetrieveConditionsData(shuttleLogbookEntries))
3128 {
cb343cfd 3129 Log("SHUTTLE", "Collect - Process of at least one run failed");
a3ca69fe 3130 CountOpenRuns();
2bb7b766 3131 return kFALSE;
3132 }
3133
36c99a6a 3134 Log("SHUTTLE", "Collect - Requested run(s) successfully processed");
a3ca69fe 3135 CountOpenRuns();
eba76848 3136 return kTRUE;
2bb7b766 3137}
3138
2bb7b766 3139//______________________________________________________________________________________________
3140Bool_t AliShuttle::RetrieveConditionsData(const TObjArray& dateEntries)
3141{
9827400b 3142 //
3143 // Retrieve conditions data for all runs that aren't processed yet
3144 //
2bb7b766 3145
3146 Bool_t hasError = kFALSE;
3147
3148 TIter iter(&dateEntries);
3149 AliShuttleLogbookEntry* anEntry;
3150
3151 while ((anEntry = (AliShuttleLogbookEntry*) iter.Next())){
3152 if (!Process(anEntry)){
3153 hasError = kTRUE;
3154 }
4b95672b 3155
3156 // clean SHUTTLE temp directory
d524ade6 3157 //TString filename = Form("%s/*.shuttle", GetShuttleTempDir());
3158 //RemoveFile(filename.Data());
2bb7b766 3159 }
3160
3161 return hasError == kFALSE;
3162}
cb343cfd 3163
3164//______________________________________________________________________________________________
3165ULong_t AliShuttle::GetTimeOfLastAction() const
3166{
9827400b 3167 //
3168 // Gets time of last action
3169 //
3170
cb343cfd 3171 ULong_t tmp;
36c99a6a 3172
cb343cfd 3173 fMonitoringMutex->Lock();
be48e3ea 3174
cb343cfd 3175 tmp = fLastActionTime;
36c99a6a 3176
cb343cfd 3177 fMonitoringMutex->UnLock();
36c99a6a 3178
cb343cfd 3179 return tmp;
3180}
3181
3182//______________________________________________________________________________________________
3183const TString AliShuttle::GetLastAction() const
3184{
9827400b 3185 //
cb343cfd 3186 // returns a string description of the last action
9827400b 3187 //
cb343cfd 3188
3189 TString tmp;
36c99a6a 3190
cb343cfd 3191 fMonitoringMutex->Lock();
3192
3193 tmp = fLastAction;
3194
3195 fMonitoringMutex->UnLock();
3196
36c99a6a 3197 return tmp;
cb343cfd 3198}
3199
3200//______________________________________________________________________________________________
3201void AliShuttle::SetLastAction(const char* action)
3202{
9827400b 3203 //
cb343cfd 3204 // updates the monitoring variables
9827400b 3205 //
36c99a6a 3206
cb343cfd 3207 fMonitoringMutex->Lock();
36c99a6a 3208
cb343cfd 3209 fLastAction = action;
3210 fLastActionTime = time(0);
3211
3212 fMonitoringMutex->UnLock();
3213}
eba76848 3214
3215//______________________________________________________________________________________________
3216const char* AliShuttle::GetRunParameter(const char* param)
3217{
9827400b 3218 //
3219 // returns run parameter read from DAQ logbook
3220 //
eba76848 3221
3222 if(!fLogbookEntry) {
3223 AliError("No logbook entry!");
3224 return 0;
3225 }
3226
3227 return fLogbookEntry->GetRunParameter(param);
3228}
57c1a579 3229
d386d623 3230//______________________________________________________________________________________________
9827400b 3231AliCDBEntry* AliShuttle::GetFromOCDB(const char* detector, const AliCDBPath& path)
d386d623 3232{
9827400b 3233 //
3234 // returns object from OCDB valid for current run
3235 //
d386d623 3236
9827400b 3237 if (fTestMode & kErrorOCDB)
3238 {
3239 Log(detector, "GetFromOCDB - In TESTMODE - Simulating error with OCDB");
3240 return 0;
3241 }
3242
d386d623 3243 AliCDBStorage *sto = AliCDBManager::Instance()->GetStorage(fgkMainCDB);
3244 if (!sto)
3245 {
9827400b 3246 Log(detector, "GetFromOCDB - Cannot activate main OCDB for query!");
d386d623 3247 return 0;
3248 }
3249
3250 return dynamic_cast<AliCDBEntry*> (sto->Get(path, GetCurrentRun()));
3251}
3252
57c1a579 3253//______________________________________________________________________________________________
339adafa 3254Bool_t AliShuttle::SendMail(EMailTarget target, Int_t system)
57c1a579 3255{
9827400b 3256 //
3257 // sends a mail to the subdetector expert in case of preprocessor error
3258 //
3259
3260 if (fTestMode != kNone)
3261 return kTRUE;
fb40ead4 3262
926cbc0e 3263 if (!fConfig->SendMail())
3264 return kTRUE;
57c1a579 3265
339adafa 3266 if (target == kDCSEMail || target == kFXSEMail) {
3267 if (!fFirstProcessing)
3a4ddf0e 3268 return kTRUE;
36c99a6a 3269 }
3270
7bdfe1fa 3271 Int_t runMode = (Int_t)fConfig->GetRunMode();
3272 TString tmpStr;
3273 if (runMode == 0) tmpStr = " Nightly Test:";
3274 else tmpStr = " Data Taking:";
fbc112e3 3275 void* dir = gSystem->OpenDirectory(GetShuttleLogDir());
3276 if (dir == NULL)
3277 {
3278 if (gSystem->mkdir(GetShuttleLogDir(), kTRUE))
3279 {
3280 Log("SHUTTLE", Form("SendMail - Can't open directory <%s>", GetShuttleLogDir()));
3281 return kFALSE;
3282 }
3283
3284 } else {
3285 gSystem->FreeDirectory(dir);
3286 }
3287
339adafa 3288 // det experts in to
3289 TString to="";
3835a613 3290 TIter *iterExperts = 0;
339adafa 3291 if (target == kDCSEMail) {
3292 iterExperts = new TIter(fConfig->GetAdmins(AliShuttleConfig::kAmanda));
3293 }
3294 else if (target == kFXSEMail) {
3295 iterExperts = new TIter(fConfig->GetAdmins(system));
3296 }
3835a613 3297 if (iterExperts) {
3298 TObjString *anExpert=0;
3299 while ((anExpert = (TObjString*) iterExperts->Next()))
3300 {
3301 to += Form("%s,", anExpert->GetName());
3302 }
3303 delete iterExperts;
339adafa 3304 }
3835a613 3305
3306 // add subdetector experts
3307 iterExperts = new TIter(fConfig->GetResponsibles(fCurrentDetector));
339adafa 3308 TObjString *anExpert=0;
3309 while ((anExpert = (TObjString*) iterExperts->Next()))
4508ab8b 3310 {
339adafa 3311 to += Form("%s,", anExpert->GetName());
3312 }
3313 delete iterExperts;
3835a613 3314
339adafa 3315 if (to.Length() > 0)
3316 to.Remove(to.Length()-1);
3317 AliDebug(2, Form("to: %s",to.Data()));
4508ab8b 3318
339adafa 3319 if (to.IsNull()) {
3320 Log("SHUTTLE", Form("List of %d responsibles not set!", (Int_t) target));
3321 return kFALSE;
3322 }
3323
3324 // SHUTTLE responsibles in cc
fb40ead4 3325 TString cc="";
3326 TIter iterAdmins(fConfig->GetAdmins(AliShuttleConfig::kGlobal));
3327 TObjString *anAdmin=0;
3328 while ((anAdmin = (TObjString*) iterAdmins.Next()))
3329 {
3330 cc += Form("%s,", anAdmin->GetName());
3331 }
3332 if (cc.Length() > 0)
c19963db 3333 cc.Remove(cc.Length()-1);
fb40ead4 3334 AliDebug(2, Form("cc: %s",to.Data()));
57c1a579 3335
339adafa 3336 // mail body
675f64cd 3337 TString bodyFileName;
3338 bodyFileName.Form("%s/mail.body", GetShuttleLogDir());
3339 gSystem->ExpandPathName(bodyFileName);
3340
3341 ofstream mailBody;
3342 mailBody.open(bodyFileName, ofstream::out);
3343
3344 if (!mailBody.is_open())
3345 {
339adafa 3346 Log("SHUTTLE", Form("Could not open mail body file %s", bodyFileName.Data()));
675f64cd 3347 return kFALSE;
3348 }
3349
675f64cd 3350
339adafa 3351 TString subject;
3352 TString body;
675f64cd 3353
339adafa 3354 if (target == kDCSEMail){
af93e5d7 3355 subject = Form("%s CRITICAL Retrieval of data points for %s FAILED in run %d !",
7bdfe1fa 3356 tmpStr.Data(), fCurrentDetector.Data(), GetCurrentRun());
339adafa 3357 AliDebug(2, Form("subject: %s", subject.Data()));
3358
3359 body = Form("Dear DCS experts, \n\n");
3360 body += Form("SHUTTLE couldn\'t retrieve the data points for detector %s "
3361 "in run %d!!\n\n", fCurrentDetector.Data(), GetCurrentRun());
fb40ead4 3362 }
339adafa 3363 else if (target == kFXSEMail){
af93e5d7 3364 subject = Form("%s CRITICAL FXS communication for %s FAILED in run %d !",
7bdfe1fa 3365 tmpStr.Data(), fCurrentDetector.Data(), GetCurrentRun());
339adafa 3366 AliDebug(2, Form("subject: %s", subject.Data()));
3367 TString sys;
3368 if (system == kDAQ) sys="DAQ";
3369 else if (system == kDCS) sys="DCS";
3370 else if (system == kHLT) sys="HLT";
3d2b6cdd 3371 else if (system == kDQM) sys="DQM";
339adafa 3372 else return kFALSE;
9c0e2d64 3373 body = Form("Dear %s FXS experts, \n\n",sys.Data());
339adafa 3374 body += Form("SHUTTLE couldn\'t retrieve data from the FXS for detector %s "
9c0e2d64 3375 "in run %d!!\n\n", fCurrentDetector.Data(), GetCurrentRun());
8662f0a8 3376 body += Form("The contacted server was:\nDB: %s\nFXS:%s\n\n", fConfig->GetFXSdbHost(system), fConfig->GetFXSHost(system));
339adafa 3377 }
3378 else {
7bdfe1fa 3379 subject = Form("%s %s Shuttle preprocessor FAILED in run %d (run type = %s)!",
3380 tmpStr.Data(), fCurrentDetector.Data(), GetCurrentRun(), GetRunType());
339adafa 3381 AliDebug(2, Form("subject: %s", subject.Data()));
3382
3383 body = Form("Dear %s expert(s), \n\n", fCurrentDetector.Data());
3384 body += Form("SHUTTLE just detected that your preprocessor "
3385 "failed processing run %d (run type = %s)!!\n\n",
3386 GetCurrentRun(), GetRunType());
3387 }
675f64cd 3388
7d4cf768 3389 body += Form("Please check %s status on the SHUTTLE monitoring page: \n\n",
3390 fCurrentDetector.Data());
b0e53b15 3391 if (fConfig->GetRunMode() == AliShuttleConfig::kTest)
3392 {
8e828d39 3393 body += Form("\thttp://pcalimonitor.cern.ch/shuttle.jsp?time=24 \n\n");
b0e53b15 3394 } else {
8e828d39 3395 body += Form("\thttp://pcalimonitor.cern.ch/shuttle.jsp?instance=PROD&time=24 \n\n");
b0e53b15 3396 }
339adafa 3397
3398
7d4cf768 3399 TString logFolder = "logs";
3400 if (fConfig->GetRunMode() == AliShuttleConfig::kProd)
3401 logFolder += "_PROD";
3402
3403
675f64cd 3404 body += Form("Find the %s log for the current run on \n\n"
29fb87f4 3405 "\thttp://pcalishuttle02.cern.ch/%s/%d/%d/%s.log \n\n",
3ebbf3c4 3406 fCurrentDetector.Data(), logFolder.Data(), GetCurrentRun()/10000,
3407 GetCurrentRun(), fCurrentDetector.Data());
fe6a3257 3408 body += Form("The last 15 lines of %s log file are following:\n\n", fCurrentDetector.Data());
675f64cd 3409
3410 AliDebug(2, Form("Body begin: %s", body.Data()));
3411
3412 mailBody << body.Data();
3413 mailBody.close();
3414 mailBody.open(bodyFileName, ofstream::out | ofstream::app);
3415
3ebbf3c4 3416 TString logFileName = Form("%s/%d/%d/%s.log", GetShuttleLogDir(),
3417 GetCurrentRun()/10000, GetCurrentRun(), fCurrentDetector.Data());
fe6a3257 3418 TString tailCommand = Form("tail -n 15 %s >> %s", logFileName.Data(), bodyFileName.Data());
675f64cd 3419 if (gSystem->Exec(tailCommand.Data()))
3420 {
3421 mailBody << Form("%s log file not found ...\n\n", fCurrentDetector.Data());
3422 }
3423
3424 TString endBody = Form("------------------------------------------------------\n\n");
36c99a6a 3425 endBody += Form("In case of problems please contact the SHUTTLE core team.\n\n");
3426 endBody += "Please do not answer this message directly, it is automatically generated.\n\n";
546242fb 3427 endBody += "Greetings,\n\n \t\t\tthe SHUTTLE\n";
57c1a579 3428
909732f7 3429 AliDebug(2, Form("Body end: %s", endBody.Data()));
57c1a579 3430
3431 mailBody << endBody.Data();
3432
3433 mailBody.close();
3434
3435 // send mail!
3436 TString mailCommand = Form("mail -s \"%s\" -c %s %s < %s",
3437 subject.Data(),
3438 cc.Data(),
3439 to.Data(),
3440 bodyFileName.Data());
909732f7 3441 AliDebug(2, Form("mail command: %s", mailCommand.Data()));
57c1a579 3442
3443 Bool_t result = gSystem->Exec(mailCommand.Data());
3444
3445 return result == 0;
3446}
441b0e9c 3447//______________________________________________________________________________________________
9827400b 3448const char* AliShuttle::GetRunType()
441b0e9c 3449{
9827400b 3450 //
3451 // returns run type read from "run type" logbook
3452 //
441b0e9c 3453
3454 if(!fLogbookEntry) {
3455 AliError("No logbook entry!");
3456 return 0;
3457 }
3458
9827400b 3459 return fLogbookEntry->GetRunType();
441b0e9c 3460}
3461
4859271b 3462//______________________________________________________________________________________________
3463Bool_t AliShuttle::GetHLTStatus()
3464{
3465 // Return HLT status (ON=1 OFF=0)
af76b20d 3466 // Converts the HLT status from the mode string read in the run logbook (not just a bool)
4859271b 3467
3468 if(!fLogbookEntry) {
3469 AliError("No logbook entry!");
3470 return 0;
3471 }
3472
af76b20d 3473 // TODO implement when HLTMode is inserted in run logbook
b4ebc2d5 3474 TString hltMode = fLogbookEntry->GetRunParameter("HLTmode");
af76b20d 3475 TSubString firstChar = hltMode(0,1);
3476 AliDebug(2,Form("First char = %s ",firstChar.Data()));
3477 if (firstChar == "A") {
3478 return kFALSE;
3479 }
3480 else if ((firstChar == "B") || (firstChar == "C") || (firstChar == "D") || (firstChar == "E")) {
3481 return kTRUE;
3482 }
3483 else {
3484 Log("SHUTTLE","Unexpected HLT mode! Returning 0....");
3485 return kFALSE;
3486 }
4859271b 3487}
ca8ea066 3488
3489//______________________________________________________________________________________________
3490const char* AliShuttle::GetTriggerConfiguration()
3491{
3492 // Receives the trigger configuration from the DAQ logbook for the current run
3493
3494 // check connection, if needed reconnect
efec19bd 3495 if (!Connect(4))
ca8ea066 3496 return 0;
3497
3498 TString sqlQuery;
3499 sqlQuery.Form("SELECT configFile FROM logbook_trigger_config WHERE run = %d", GetCurrentRun());
efec19bd 3500 TSQLResult* result = fServer[4]->Query(sqlQuery);
ca8ea066 3501 if (!result)
3502 {
3503 Log("SHUTTLE", Form("ERROR: Can't execute query <%s>!", sqlQuery.Data()));
3504 return 0;
3505 }
3506
3507 if (result->GetRowCount() == 0)
3508 {
ff0191c7 3509 Log("SHUTTLE", "WARNING: Trigger configuration not found in logbook_trigger_config");
ca8ea066 3510 delete result;
3511 return 0;
3512 }
3513
3514 TSQLRow* row = result->Next();
3515 if (!row)
3516 {
3517 Log("SHUTTLE", "ERROR: Could not receive logbook_trigger_config data");
3518 delete result;
3519 return 0;
3520 }
3521
3522 // static, so that pointer remains valid when it is returned to the calling class
3523 static TString triggerConfig(row->GetField(0));
3524
3525 delete row;
3526 row = 0;
3527
3528 delete result;
3529 result = 0;
3530
3531 Log("SHUTTLE", Form("Found trigger configuration: %s", triggerConfig.Data()));
3532
3533 return triggerConfig;
3534}
3535
2c5f9d06 3536//______________________________________________________________________________________________
3537const char* AliShuttle::GetCTPTimeParams()
3538{
3539 // Receives the CTP time parameters from the DAQ logbook for the current run
3540
3541 // check connection, if needed reconnect
efec19bd 3542 if (!Connect(4))
2c5f9d06 3543 return 0;
3544
3545 TString sqlQuery;
3546 sqlQuery.Form("SELECT alignmentFile FROM logbook_trigger_config WHERE run = %d", GetCurrentRun());
efec19bd 3547 TSQLResult* result = fServer[4]->Query(sqlQuery);
2c5f9d06 3548 if (!result)
3549 {
3550 Log("SHUTTLE", Form("ERROR: Can't execute query <%s>!", sqlQuery.Data()));
3551 return 0;
3552 }
3553
3554 if (result->GetRowCount() == 0)
3555 {
ff0191c7 3556 Log("SHUTTLE", "WARNING: CTP time params not found in logbook_trigger_config");
2c5f9d06 3557 delete result;
3558 return 0;
3559 }
3560
3561 TSQLRow* row = result->Next();
3562 if (!row)
3563 {
3564 Log("SHUTTLE", "ERROR: Could not receive logbook_trigger_config data");
3565 delete result;
3566 return 0;
3567 }
3568
3569 // static, so that pointer remains valid when it is returned to the calling class
3570 static TString triggerTimeParams(row->GetField(0));
3571
3572 delete row;
3573 row = 0;
3574
3575 delete result;
3576 result = 0;
3577
a191c286 3578 Log("SHUTTLE", Form("Found trigger time parameters: %s", triggerTimeParams.Data()));
2c5f9d06 3579
3580 return triggerTimeParams;
3581}
3582
d386d623 3583//______________________________________________________________________________________________
b48d2542 3584const char* AliShuttle::GetTriggerDetectorMask()
3585{
3586 // Receives the trigger detector mask from DAQ logbook
3587
3588 // check connection, if needed reconnect
efec19bd 3589 if (!Connect(4))
b48d2542 3590 return 0;
3591
3592 TString sqlQuery;
3593 sqlQuery.Form("SELECT BIN(BIT_OR(inputDetectorMask)) from logbook_trigger_clusters WHERE run = %d;", GetCurrentRun());
efec19bd 3594 TSQLResult* result = fServer[4]->Query(sqlQuery);
b48d2542 3595 if (!result)
3596 {
3597 Log("SHUTTLE", Form("ERROR: Can't execute query <%s>!", sqlQuery.Data()));
3598 return 0;
3599 }
3600
3601 if (result->GetRowCount() == 0)
3602 {
3603 Log("SHUTTLE", "ERROR: Trigger Detector Mask not found in logbook_trigger_clusters");
3604 delete result;
3605 return 0;
3606 }
3607
3608 TSQLRow* row = result->Next();
3609 if (!row)
3610 {
3611 Log("SHUTTLE", "ERROR: Could not receive logbook_trigger_clusters data");
3612 delete result;
3613 return 0;
3614 }
3615
3616 // static, so that pointer remains valid when it is returned to the calling class
3617 static TString triggerDetectorMask(row->GetField(0));
3618
3619 delete row;
3620 row = 0;
3621
3622 delete result;
3623 result = 0;
3624
2e7d9963 3625 Log("SHUTTLE", Form("Found Trigger Detector Mask: %s", triggerDetectorMask.Data()));
b48d2542 3626
3627 return triggerDetectorMask;
3628}
3629
3630//______________________________________________________________________________________________
d386d623 3631void AliShuttle::SetShuttleTempDir(const char* tmpDir)
3632{
9827400b 3633 //
3634 // sets Shuttle temp directory
3635 //
d386d623 3636
3637 fgkShuttleTempDir = gSystem->ExpandPathName(tmpDir);
3638}
3639
3640//______________________________________________________________________________________________
3641void AliShuttle::SetShuttleLogDir(const char* logDir)
3642{
9827400b 3643 //
3644 // sets Shuttle log directory
3645 //
d386d623 3646
3647 fgkShuttleLogDir = gSystem->ExpandPathName(logDir);
3648}
503ff24f 3649//______________________________________________________________________________________________
3650Bool_t AliShuttle::TouchFile()
3651{
3652 //
3653 // touching a file on the grid if run has been DONE
3654 //
3655
3656 if (!gGrid)
3657 {
3658 Log("SHUTTLE",Form("No TGrid connection estabilished!"));
3659 Log("SHUTTLE",Form("Could not touch file for run %i",GetCurrentRun()));
3660 return kFALSE;
3661 }
3662
b8cf5e0f 3663 TString dir;
728290a4 3664 dir.Form("%s%d/SHUTTLE_DONE", fConfig->GetAlienPath(), GetCurrentYear());
b8cf5e0f 3665 // checking whether directory for touch command exists
3666 TString commandLs;
3667 commandLs.Form("ls %s",dir.Data());
3668 TGridResult *resultLs = dynamic_cast<TGridResult*>(gGrid->Command(commandLs));
3669 if (!resultLs){
3670 Log("SHUTTLE",Form("No result for %s command, returning without touching",commandLs.Data()));
3671 return kFALSE;
3672 }
3673 TMap *mapLs = dynamic_cast<TMap*>(resultLs->At(0));
3674 if (!mapLs){
3675 Log("SHUTTLE",Form("No map for %s command, returning without touching",commandLs.Data()));
97d38a5a 3676 delete resultLs;
3677 resultLs = 0x0;
b8cf5e0f 3678 return kFALSE;
3679 }
3680 TObjString *valueLsPath = dynamic_cast<TObjString*>(mapLs->GetValue("path"));
eff13ba0 3681 if (!valueLsPath || (valueLsPath->GetString()).CompareTo(dir)!=1){
b8cf5e0f 3682 Log("SHUTTLE",Form("No directory %s found, creating it",dir.Data()));
3683
3684 // creating the directory
3685
3686 Bool_t boolMkdir = gGrid->Mkdir(dir.Data());
3687 if (!boolMkdir) {
3688 Log("SHUTTLE",Form("Impossible to create dir %s in alien catalogue for run %i!",dir.Data(),GetCurrentRun()));
97d38a5a 3689 delete resultLs;
3690 resultLs = 0x0;
b8cf5e0f 3691 return kFALSE;
3692 }
3693 Log("SHUTTLE",Form("Directory %s successfully created in alien catalogue for run %i",dir.Data(),GetCurrentRun()));
3694 }
3695 else {
3696 Log("SHUTTLE",Form("Directory %s correctly found for run %i",dir.Data(),GetCurrentRun()));
3697 }
3698
97d38a5a 3699 delete resultLs;
3700 resultLs = 0x0;
3701
503ff24f 3702 TString command;
b8cf5e0f 3703 command.Form("touch %s/%i", dir.Data(), GetCurrentRun());
e660ff60 3704 Log("SHUTTLE", Form("Creating entry in file catalog: %s", command.Data()));
503ff24f 3705 TGridResult *resultTouch = dynamic_cast<TGridResult*>(gGrid->Command(command));
b8cf5e0f 3706 if (!resultTouch){
3707 Log("SHUTTLE",Form("No result for touching command, returning without touching for run %i",GetCurrentRun()));
3708 return kFALSE;
503ff24f 3709 }
b8cf5e0f 3710 TMap *mapTouch = dynamic_cast<TMap*>(resultTouch->At(0));
3711 if (!mapTouch){
3712 Log("SHUTTLE",Form("No map for touching command, returning without touching for run %i",GetCurrentRun()));
97d38a5a 3713 delete resultTouch;
3714 resultTouch = 0x0;
b8cf5e0f 3715 return kFALSE;
3716 }
3717 TObjString *valueTouch = dynamic_cast<TObjString*>(mapTouch->GetValue("__result__"));
3718 if (!valueTouch){
3719 Log("SHUTTLE",Form("No value for \"__result__\" key set in the map for touching command, returning without touching for run %i",GetCurrentRun()));
97d38a5a 3720 delete resultTouch;
3721 resultTouch = 0x0;
b8cf5e0f 3722 return kFALSE;
503ff24f 3723 }
b8cf5e0f 3724 if (valueTouch->GetString()!="1"){
3725 Log("SHUTTLE",Form("Failing the touching command, returning without touching for run %i",GetCurrentRun()));
97d38a5a 3726 delete resultTouch;
3727 resultTouch = 0x0;
b8cf5e0f 3728 return kFALSE;
3729 }
97d38a5a 3730 delete resultTouch;
3731 resultTouch = 0x0;
98400bd9 3732 Log("SHUTTLE", "Sucessfully touched the file");
b8cf5e0f 3733 return kTRUE;
503ff24f 3734}
2eb98462 3735//______________________________________________________________________________________________
f100adeb 3736UInt_t AliShuttle::GetStartTimeDCSQuery()
2eb98462 3737{
3738 // Return Start Time for the DCS query
3739 //
3740 // The call is delegated to AliShuttleInterface
3741
3d24ad5a 3742 return GetCurrentStartTime()-fConfig->GetDCSQueryOffset();
2eb98462 3743}
3744//______________________________________________________________________________________________
f100adeb 3745UInt_t AliShuttle::GetEndTimeDCSQuery()
2eb98462 3746{
3747 // Return End Time for the DCS query
3748 //
3749 // The call is delegated to AliShuttleInterface
3750
3751 return GetCurrentEndTime()+fConfig->GetDCSQueryOffset();
3752}
1f0f7c2e 3753//______________________________________________________________________________________________
3754void AliShuttle::SendMLFromDet(const char* value)
3755{
3756 //
3757 // Sending an information coming from the current detector to ML
3758 //
3759
3760 TMonaLisaText mlText(Form("%s_RunCondition", fCurrentDetector.Data()), value);
3761
3762 TList mlList;
3763 mlList.Add(&mlText);
3764
3765 TString mlID;
3766 mlID.Form("%d", GetCurrentRun());
3767 fMonaLisa->SendParameters(&mlList, mlID);
3768
3769 return;
3770}
6ecb073a 3771//______________________________________________________________________________________________
3772TString* AliShuttle::GetLTUConfig(const char* det)
3773{
3774 //
3775 // Getting ltuFineDelay1, ltuFineDelay2, ltuBCDelay for detector det from logbook_detectors table in logbook
3776 //
3777
efec19bd 3778 if (!Connect(4))
6ecb073a 3779 return 0;
3780
3781 TString sqlQuery;
3782 sqlQuery.Form("select LTUFineDelay1, LTUFineDelay2, LTUBCDelayAdd from logbook_detectors WHERE run_number = %d and detector = \"%s\";", GetCurrentRun(),det);
3783
efec19bd 3784 TSQLResult* result = fServer[4]->Query(sqlQuery);
6ecb073a 3785 if (!result){
3786 Log("SHUTTLE","ERROR: No result found for the LTU configuration query");
3787 return 0x0;
3788 }
3789 if (result->GetRowCount() == 0){
3790 Log("SHUTTLE",Form("ERROR: LTU configuration not found in logbook_detectors for detector %s, returning null pointer",det));
3791 delete result;
3792 return 0x0;
3793 }
3794 if (result->GetFieldCount() != 3){
3795 Log("SHUTTLE",Form("ERROR: not all the required fields are there for the LTU configuration for detector %s (only %d found), returning a null pointer",det, result->GetFieldCount()));
3796 delete result;
3797 return 0x0;
3798 }
3799 TSQLRow* row = result->Next();
3800 if (!row){
3801 Printf("ERROR: Could not receive logbook_detectors data, returning null pointer");
3802 delete result;
3803 return 0x0;
3804 }
3805 TString* ltuConfigString = new TString[3];
3806
3807 ltuConfigString[0] = row->GetField(0);
3808 ltuConfigString[1] = row->GetField(1);
3809 ltuConfigString[2] = row->GetField(2);
503ff24f 3810
6ecb073a 3811 return ltuConfigString;
3812
3813}