]>
Commit | Line | Data |
---|---|---|
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> | |
73abe331 | 60 | |
e7f62f16 | 61 | #include <TMonaLisaWriter.h> |
62 | ||
5164a766 | 63 | #include <fstream> |
64 | ||
cb343cfd | 65 | #include <sys/types.h> |
66 | #include <sys/wait.h> | |
67 | ||
1bd183ba | 68 | #include <signal.h> |
69 | ||
73abe331 | 70 | ClassImp(AliShuttle) |
71 | ||
b948db8d | 72 | //______________________________________________________________________________________________ |
73 | AliShuttle::AliShuttle(const AliShuttleConfig* config, | |
74 | UInt_t timeout, Int_t retries): | |
4f0ab988 | 75 | fConfig(config), |
76 | fTimeout(timeout), fRetries(retries), | |
77 | fPreprocessorMap(), | |
2bb7b766 | 78 | fLogbookEntry(0), |
eba76848 | 79 | fCurrentDetector(), |
926cbc0e | 80 | fFirstProcessing(0), |
85a80aa9 | 81 | fStatusEntry(0), |
cb343cfd | 82 | fMonitoringMutex(0), |
eba76848 | 83 | fLastActionTime(0), |
e7f62f16 | 84 | fLastAction(), |
9827400b | 85 | fMonaLisa(0), |
86 | fTestMode(kNone), | |
ffa29e93 | 87 | fReadTestMode(kFALSE), |
88 | fOutputRedirected(kFALSE) | |
73abe331 | 89 | { |
90 | // | |
91 | // config: AliShuttleConfig used | |
73abe331 | 92 | // timeout: timeout used for AliDCSClient connection |
93 | // retries: the number of retries in case of connection error. | |
94 | // | |
95 | ||
57f50b3c | 96 | if (!fConfig->IsValid()) AliFatal("********** !!!!! Invalid configuration !!!!! **********"); |
be48e3ea | 97 | for(int iSys=0;iSys<4;iSys++) { |
57f50b3c | 98 | fServer[iSys]=0; |
be48e3ea | 99 | if (iSys < 3) |
2c15234c | 100 | fFXSlist[iSys].SetOwner(kTRUE); |
57f50b3c | 101 | } |
2bb7b766 | 102 | fPreprocessorMap.SetOwner(kTRUE); |
be48e3ea | 103 | |
104 | for (UInt_t iDet=0; iDet<NDetectors(); iDet++) | |
105 | fFirstUnprocessed[iDet] = kFALSE; | |
106 | ||
cb343cfd | 107 | fMonitoringMutex = new TMutex(); |
58bc3020 | 108 | } |
109 | ||
b948db8d | 110 | //______________________________________________________________________________________________ |
57f50b3c | 111 | AliShuttle::~AliShuttle() |
58bc3020 | 112 | { |
9827400b | 113 | // |
114 | // destructor | |
115 | // | |
58bc3020 | 116 | |
b948db8d | 117 | fPreprocessorMap.DeleteAll(); |
be48e3ea | 118 | for(int iSys=0;iSys<4;iSys++) |
57f50b3c | 119 | if(fServer[iSys]) { |
120 | fServer[iSys]->Close(); | |
121 | delete fServer[iSys]; | |
eba76848 | 122 | fServer[iSys] = 0; |
57f50b3c | 123 | } |
2bb7b766 | 124 | |
125 | if (fStatusEntry){ | |
126 | delete fStatusEntry; | |
127 | fStatusEntry = 0; | |
128 | } | |
cb343cfd | 129 | |
130 | if (fMonitoringMutex) | |
131 | { | |
132 | delete fMonitoringMutex; | |
133 | fMonitoringMutex = 0; | |
134 | } | |
73abe331 | 135 | } |
136 | ||
b948db8d | 137 | //______________________________________________________________________________________________ |
57f50b3c | 138 | void AliShuttle::RegisterPreprocessor(AliPreprocessor* preprocessor) |
58bc3020 | 139 | { |
73abe331 | 140 | // |
b948db8d | 141 | // Registers new AliPreprocessor. |
73abe331 | 142 | // It uses GetName() for indentificator of the pre processor. |
143 | // The pre processor is registered it there isn't any other | |
144 | // with the same identificator (GetName()). | |
145 | // | |
146 | ||
eba76848 | 147 | const char* detName = preprocessor->GetName(); |
148 | if(GetDetPos(detName) < 0) | |
149 | AliFatal(Form("********** !!!!! Invalid detector name: %s !!!!! **********", detName)); | |
150 | ||
151 | if (fPreprocessorMap.GetValue(detName)) { | |
152 | AliWarning(Form("AliPreprocessor %s is already registered!", detName)); | |
73abe331 | 153 | return; |
154 | } | |
155 | ||
eba76848 | 156 | fPreprocessorMap.Add(new TObjString(detName), preprocessor); |
73abe331 | 157 | } |
b948db8d | 158 | //______________________________________________________________________________________________ |
3301427a | 159 | Bool_t AliShuttle::Store(const AliCDBPath& path, TObject* object, |
84090f85 | 160 | AliCDBMetaData* metaData, Int_t validityStart, Bool_t validityInfinite) |
73abe331 | 161 | { |
9827400b | 162 | // Stores a CDB object in the storage for offline reconstruction. Objects that are not needed for |
163 | // offline reconstruction, but should be stored anyway (e.g. for debugging) should NOT be stored | |
164 | // using this function. Use StoreReferenceData instead! | |
165 | // It calls StoreLocally function which temporarily stores the data locally; when the preprocessor | |
166 | // finishes the data are transferred to the main storage (Grid). | |
b948db8d | 167 | |
3301427a | 168 | return StoreLocally(fgkLocalCDB, path, object, metaData, validityStart, validityInfinite); |
84090f85 | 169 | } |
170 | ||
171 | //______________________________________________________________________________________________ | |
3301427a | 172 | Bool_t AliShuttle::StoreReferenceData(const AliCDBPath& path, TObject* object, AliCDBMetaData* metaData) |
84090f85 | 173 | { |
9827400b | 174 | // Stores a CDB object in the storage for reference data. This objects will not be available during |
175 | // offline reconstrunction. Use this function for reference data only! | |
176 | // It calls StoreLocally function which temporarily stores the data locally; when the preprocessor | |
177 | // finishes the data are transferred to the main storage (Grid). | |
85a80aa9 | 178 | |
3301427a | 179 | return StoreLocally(fgkLocalRefStorage, path, object, metaData); |
85a80aa9 | 180 | } |
181 | ||
182 | //______________________________________________________________________________________________ | |
3301427a | 183 | Bool_t AliShuttle::StoreLocally(const TString& localUri, |
85a80aa9 | 184 | const AliCDBPath& path, TObject* object, AliCDBMetaData* metaData, |
185 | Int_t validityStart, Bool_t validityInfinite) | |
186 | { | |
9827400b | 187 | // Store object temporarily in local storage. Parameters are passed by Store and StoreReferenceData functions. |
188 | // when the preprocessor finishes the data are transferred to the main storage (Grid). | |
189 | // The parameters are: | |
190 | // 1) Uri of the backup storage (Local) | |
191 | // 2) the object's path. | |
192 | // 3) the object to be stored | |
193 | // 4) the metaData to be associated with the object | |
194 | // 5) the validity start run number w.r.t. the current run, | |
195 | // if the data is valid only for this run leave the default 0 | |
196 | // 6) specifies if the calibration data is valid for infinity (this means until updated), | |
197 | // typical for calibration runs, the default is kFALSE | |
198 | // | |
199 | // returns 0 if fail, 1 otherwise | |
84090f85 | 200 | |
9827400b | 201 | if (fTestMode & kErrorStorage) |
202 | { | |
203 | Log(fCurrentDetector, "StoreLocally - In TESTMODE - Simulating error while storing locally"); | |
204 | return kFALSE; | |
205 | } | |
206 | ||
3301427a | 207 | const char* cdbType = (localUri == fgkLocalCDB) ? "CDB" : "Reference"; |
2bb7b766 | 208 | |
85a80aa9 | 209 | Int_t firstRun = GetCurrentRun() - validityStart; |
84090f85 | 210 | if(firstRun < 0) { |
9827400b | 211 | AliWarning("First valid run happens to be less than 0! Setting it to 0."); |
84090f85 | 212 | firstRun=0; |
213 | } | |
214 | ||
215 | Int_t lastRun = -1; | |
216 | if(validityInfinite) { | |
217 | lastRun = AliCDBRunRange::Infinity(); | |
218 | } else { | |
219 | lastRun = GetCurrentRun(); | |
220 | } | |
221 | ||
3301427a | 222 | // Version is set to current run, it will be used later to transfer data to Grid |
223 | AliCDBId id(path, firstRun, lastRun, GetCurrentRun(), -1); | |
2bb7b766 | 224 | |
225 | if(! dynamic_cast<TObjString*> (metaData->GetProperty("RunUsed(TObjString)"))){ | |
226 | TObjString runUsed = Form("%d", GetCurrentRun()); | |
9e080f92 | 227 | metaData->SetProperty("RunUsed(TObjString)", runUsed.Clone()); |
2bb7b766 | 228 | } |
84090f85 | 229 | |
3301427a | 230 | Bool_t result = kFALSE; |
84090f85 | 231 | |
3301427a | 232 | if (!(AliCDBManager::Instance()->GetStorage(localUri))) { |
233 | Log("SHUTTLE", Form("StoreLocally - Cannot activate local %s storage", cdbType)); | |
84090f85 | 234 | } else { |
3301427a | 235 | result = AliCDBManager::Instance()->GetStorage(localUri) |
84090f85 | 236 | ->Put(object, id, metaData); |
237 | } | |
238 | ||
239 | if(!result) { | |
240 | ||
9827400b | 241 | Log(fCurrentDetector, Form("StoreLocally - Can't store object <%s>!", id.ToString().Data())); |
3301427a | 242 | } |
2bb7b766 | 243 | |
3301427a | 244 | return result; |
245 | } | |
84090f85 | 246 | |
3301427a | 247 | //______________________________________________________________________________________________ |
248 | Bool_t AliShuttle::StoreOCDB() | |
249 | { | |
9827400b | 250 | // |
251 | // Called when preprocessor ends successfully or when previous storage attempt failed (kStoreError status) | |
252 | // Calls underlying StoreOCDB(const char*) function twice, for OCDB and Reference storage. | |
253 | // Then calls StoreRefFilesToGrid to store reference files. | |
254 | // | |
255 | ||
5e993b6f | 256 | UpdateShuttleStatus(AliShuttleStatus::kStoreStarted); |
257 | ||
9827400b | 258 | if (fTestMode & kErrorGrid) |
259 | { | |
260 | Log("SHUTTLE", "StoreOCDB - In TESTMODE - Simulating error while storing in the Grid"); | |
261 | Log(fCurrentDetector, "StoreOCDB - In TESTMODE - Simulating error while storing in the Grid"); | |
262 | return kFALSE; | |
263 | } | |
264 | ||
c88ad5db | 265 | Log("SHUTTLE","StoreOCDB - Storing OCDB data ..."); |
5e993b6f | 266 | Int_t resultCDB = StoreOCDB(fgkMainCDB); |
86aa42c3 | 267 | |
c88ad5db | 268 | Log("SHUTTLE","StoreOCDB - Storing reference data ..."); |
5e993b6f | 269 | Int_t resultRef = StoreOCDB(fgkMainRefStorage); |
9827400b | 270 | |
c88ad5db | 271 | Log("SHUTTLE","StoreOCDB - Storing reference files ..."); |
272 | Bool_t resultRefFiles = CopyFilesToGrid("reference"); | |
273 | ||
274 | Bool_t resultMetadata = kTRUE; | |
275 | if(fCurrentDetector == "GRP") | |
276 | { | |
277 | Log("StoreOCDB - SHUTTLE","Storing Run Metadata file ..."); | |
278 | resultMetadata = CopyFilesToGrid("metadata"); | |
279 | } | |
9827400b | 280 | |
5e993b6f | 281 | Int_t storeResult = 0; |
282 | ||
283 | if (resultCDB < 0 || resultRef < 0 || resultRefFiles == kFALSE || resultMetadata == kFALSE) | |
284 | storeResult = -1; | |
285 | else if (resultCDB > 0 || resultRef > 0) | |
286 | storeResult = 1; | |
287 | ||
288 | if (storeResult < 0) | |
289 | { | |
290 | Log("SHUTTLE", | |
291 | Form("\t\t\t****** run %d - %s: STORAGE ERROR ******", | |
292 | GetCurrentRun(), fCurrentDetector.Data())); | |
293 | UpdateShuttleStatus(AliShuttleStatus::kStoreError); | |
294 | } | |
295 | else if (storeResult > 0) | |
296 | { | |
297 | Log("SHUTTLE", | |
298 | Form("\t\t\t****** run %d - %s: STORAGE DELAYED ******", | |
299 | GetCurrentRun(), fCurrentDetector.Data())); | |
300 | UpdateShuttleStatus(AliShuttleStatus::kStoreDelayed); | |
301 | } | |
302 | else if (storeResult == 0) | |
303 | { | |
304 | Log("SHUTTLE", | |
305 | Form("\t\t\t****** run %d - %s: DONE ******", | |
306 | GetCurrentRun(), fCurrentDetector.Data())); | |
307 | UpdateShuttleStatus(AliShuttleStatus::kDone); | |
308 | UpdateShuttleLogbook(fCurrentDetector, "DONE"); | |
309 | } | |
310 | ||
311 | return (storeResult == 0); | |
3301427a | 312 | } |
313 | ||
314 | //______________________________________________________________________________________________ | |
5e993b6f | 315 | Int_t AliShuttle::StoreOCDB(const TString& gridURI) |
3301427a | 316 | { |
317 | // | |
318 | // Called by StoreOCDB(), performs actual storage to the main OCDB and reference storages (Grid) | |
319 | // | |
5e993b6f | 320 | // Return code: |
321 | // -2 initialization error | |
322 | // -1 storage error | |
323 | // 0 success | |
324 | // 1 storage delayed (e.g. previous unprocessed runs) | |
325 | // | |
3301427a | 326 | |
327 | TObjArray* gridIds=0; | |
328 | ||
329 | Bool_t result = kTRUE; | |
5e993b6f | 330 | Bool_t delayed = kFALSE; |
3301427a | 331 | |
332 | const char* type = 0; | |
333 | TString localURI; | |
334 | if(gridURI == fgkMainCDB) { | |
335 | type = "OCDB"; | |
336 | localURI = fgkLocalCDB; | |
337 | } else if(gridURI == fgkMainRefStorage) { | |
338 | type = "reference"; | |
339 | localURI = fgkLocalRefStorage; | |
340 | } else { | |
341 | AliError(Form("Invalid storage URI: %s", gridURI.Data())); | |
5e993b6f | 342 | return -2; |
3301427a | 343 | } |
344 | ||
345 | AliCDBManager* man = AliCDBManager::Instance(); | |
346 | ||
347 | AliCDBStorage *gridSto = man->GetStorage(gridURI); | |
348 | if(!gridSto) { | |
349 | Log("SHUTTLE", | |
350 | Form("StoreOCDB - cannot activate main %s storage", type)); | |
5e993b6f | 351 | return -2; |
3301427a | 352 | } |
353 | ||
354 | gridIds = gridSto->GetQueryCDBList(); | |
355 | ||
356 | // get objects previously stored in local CDB | |
357 | AliCDBStorage *localSto = man->GetStorage(localURI); | |
358 | if(!localSto) { | |
359 | Log("SHUTTLE", | |
360 | Form("StoreOCDB - cannot activate local %s storage", type)); | |
5e993b6f | 361 | return -2; |
3301427a | 362 | } |
363 | AliCDBPath aPath(GetOfflineDetName(fCurrentDetector.Data()),"*","*"); | |
364 | // Local objects were stored with current run as Grid version! | |
365 | TList* localEntries = localSto->GetAll(aPath.GetPath(), GetCurrentRun(), GetCurrentRun()); | |
366 | localEntries->SetOwner(1); | |
367 | ||
368 | // loop on local stored objects | |
369 | TIter localIter(localEntries); | |
370 | AliCDBEntry *aLocEntry = 0; | |
371 | while((aLocEntry = dynamic_cast<AliCDBEntry*> (localIter.Next()))){ | |
372 | aLocEntry->SetOwner(1); | |
373 | AliCDBId aLocId = aLocEntry->GetId(); | |
374 | aLocEntry->SetVersion(-1); | |
375 | aLocEntry->SetSubVersion(-1); | |
376 | ||
377 | // If local object is valid up to infinity we store it only if it is | |
378 | // the first unprocessed run! | |
379 | if (aLocId.GetLastRun() == AliCDBRunRange::Infinity() && | |
380 | !fFirstUnprocessed[GetDetPos(fCurrentDetector)]) | |
381 | { | |
382 | Log("SHUTTLE", Form("StoreOCDB - %s: object %s has validity infinite but " | |
383 | "there are previous unprocessed runs!", | |
384 | fCurrentDetector.Data(), aLocId.GetPath().Data())); | |
5e993b6f | 385 | Log(fCurrentDetector.Data(), Form("StoreOCDB - %s: object %s has validity infinite but " |
386 | "there are previous unprocessed runs!", | |
387 | fCurrentDetector.Data(), aLocId.GetPath().Data())); | |
388 | delayed = kTRUE; | |
3301427a | 389 | continue; |
390 | } | |
391 | ||
392 | // loop on Grid valid Id's | |
393 | Bool_t store = kTRUE; | |
394 | TIter gridIter(gridIds); | |
395 | AliCDBId* aGridId = 0; | |
396 | while((aGridId = dynamic_cast<AliCDBId*> (gridIter.Next()))){ | |
397 | if(aGridId->GetPath() != aLocId.GetPath()) continue; | |
398 | // skip all objects valid up to infinity | |
399 | if(aGridId->GetLastRun() == AliCDBRunRange::Infinity()) continue; | |
400 | // if we get here, it means there's already some more recent object stored on Grid! | |
401 | store = kFALSE; | |
402 | break; | |
403 | } | |
404 | ||
405 | // If we get here, the file can be stored! | |
406 | Bool_t storeOk = gridSto->Put(aLocEntry); | |
407 | if(!store || storeOk){ | |
408 | ||
409 | if (!store) | |
410 | { | |
411 | Log(fCurrentDetector.Data(), | |
412 | Form("StoreOCDB - A more recent object already exists in %s storage: <%s>", | |
413 | type, aGridId->ToString().Data())); | |
414 | } else { | |
415 | Log("SHUTTLE", | |
416 | Form("StoreOCDB - Object <%s> successfully put into %s storage", | |
417 | aLocId.ToString().Data(), type)); | |
2d9019b4 | 418 | Log(fCurrentDetector.Data(), |
419 | Form("StoreOCDB - Object <%s> successfully put into %s storage", | |
420 | aLocId.ToString().Data(), type)); | |
3301427a | 421 | } |
84090f85 | 422 | |
3301427a | 423 | // removing local filename... |
424 | TString filename; | |
425 | localSto->IdToFilename(aLocId, filename); | |
c88ad5db | 426 | Log("SHUTTLE", Form("StoreOCDB - Removing local file %s", filename.Data())); |
3301427a | 427 | RemoveFile(filename.Data()); |
428 | continue; | |
429 | } else { | |
430 | Log("SHUTTLE", | |
431 | Form("StoreOCDB - Grid %s storage of object <%s> failed", | |
432 | type, aLocId.ToString().Data())); | |
2d9019b4 | 433 | Log(fCurrentDetector.Data(), |
434 | Form("StoreOCDB - Grid %s storage of object <%s> failed", | |
435 | type, aLocId.ToString().Data())); | |
3301427a | 436 | result = kFALSE; |
b948db8d | 437 | } |
438 | } | |
3301427a | 439 | localEntries->Clear(); |
2bb7b766 | 440 | |
5e993b6f | 441 | Int_t returnCode = 0; |
442 | ||
443 | if (result == kFALSE) | |
444 | returnCode = -1; | |
445 | else if (delayed != kFALSE) | |
446 | returnCode = 1; | |
447 | ||
448 | Log("SHUTTLE", Form("StoreOCDB - Returning with %d (result = %d, delayed = %d)", returnCode, result, delayed)); | |
449 | Log(fCurrentDetector.Data(), Form("StoreOCDB - Returning with %d (result = %d, delayed = %d)", returnCode, result, delayed)); | |
450 | ||
451 | return returnCode; | |
3301427a | 452 | } |
453 | ||
546242fb | 454 | //______________________________________________________________________________________________ |
455 | Bool_t AliShuttle::CleanReferenceStorage(const char* detector) | |
456 | { | |
2d9019b4 | 457 | // clears the directory used to store reference files of a given subdetector |
546242fb | 458 | |
459 | AliCDBManager* man = AliCDBManager::Instance(); | |
460 | AliCDBStorage* sto = man->GetStorage(fgkLocalRefStorage); | |
2d9019b4 | 461 | TString localBaseFolder = sto->GetBaseFolder(); |
462 | ||
463 | TString targetDir = GetRefFilePrefix(localBaseFolder.Data(), detector); | |
464 | ||
d524ade6 | 465 | Log("SHUTTLE", Form("CleanReferenceStorage - Cleaning %s", targetDir.Data())); |
2d9019b4 | 466 | |
467 | TString begin; | |
468 | begin.Form("%d_", GetCurrentRun()); | |
469 | ||
470 | TSystemDirectory* baseDir = new TSystemDirectory("/", targetDir); | |
471 | if (!baseDir) | |
472 | return kTRUE; | |
473 | ||
474 | TList* dirList = baseDir->GetListOfFiles(); | |
475 | delete baseDir; | |
476 | ||
477 | if (!dirList) return kTRUE; | |
478 | ||
479 | if (dirList->GetEntries() < 3) | |
480 | { | |
481 | delete dirList; | |
482 | return kTRUE; | |
483 | } | |
484 | ||
485 | Int_t nDirs = 0, nDel = 0; | |
486 | TIter dirIter(dirList); | |
487 | TSystemFile* entry = 0; | |
546242fb | 488 | |
2d9019b4 | 489 | Bool_t success = kTRUE; |
546242fb | 490 | |
2d9019b4 | 491 | while ((entry = dynamic_cast<TSystemFile*> (dirIter.Next()))) |
492 | { | |
493 | if (entry->IsDirectory()) | |
494 | continue; | |
495 | ||
496 | TString fileName(entry->GetName()); | |
497 | if (!fileName.BeginsWith(begin)) | |
498 | continue; | |
499 | ||
500 | nDirs++; | |
501 | ||
502 | // delete file | |
503 | Int_t result = gSystem->Unlink(fileName.Data()); | |
504 | ||
505 | if (result) | |
506 | { | |
d524ade6 | 507 | Log("SHUTTLE", Form("CleanReferenceStorage - Could not delete file %s!", fileName.Data())); |
2d9019b4 | 508 | success = kFALSE; |
509 | } else { | |
510 | nDel++; | |
511 | } | |
512 | } | |
513 | ||
514 | if(nDirs > 0) | |
515 | Log("SHUTTLE", Form("CleanReferenceStorage - %d (over %d) reference files in folder %s were deleted.", | |
516 | nDel, nDirs, targetDir.Data())); | |
517 | ||
518 | ||
519 | delete dirList; | |
520 | return success; | |
521 | ||
522 | ||
523 | ||
524 | ||
525 | ||
546242fb | 526 | |
527 | Int_t result = gSystem->GetPathInfo(targetDir, 0, (Long64_t*) 0, 0, 0); | |
528 | if (result == 0) | |
529 | { | |
530 | // delete directory | |
d524ade6 | 531 | result = gSystem->Exec(Form("rm -rf %s", targetDir.Data())); |
546242fb | 532 | if (result != 0) |
533 | { | |
d524ade6 | 534 | Log("SHUTTLE", Form("CleanReferenceStorage - Could not clean directory %s", targetDir.Data())); |
546242fb | 535 | return kFALSE; |
536 | } | |
537 | } | |
538 | ||
539 | result = gSystem->mkdir(targetDir, kTRUE); | |
540 | if (result != 0) | |
541 | { | |
c88ad5db | 542 | Log("SHUTTLE", Form("CleanReferenceStorage - Error creating base directory %s", targetDir.Data())); |
546242fb | 543 | return kFALSE; |
544 | } | |
545 | ||
546 | return kTRUE; | |
547 | } | |
548 | ||
9827400b | 549 | //______________________________________________________________________________________________ |
550 | Bool_t AliShuttle::StoreReferenceFile(const char* detector, const char* localFile, const char* gridFileName) | |
551 | { | |
552 | // | |
3c2a21c8 | 553 | // Stores reference file directly (without opening it). This function stores the file locally. |
9827400b | 554 | // |
3c2a21c8 | 555 | // The file is stored under the following location: |
556 | // <base folder of local reference storage>/<DET>/<RUN#>_<gridFileName> | |
557 | // where <gridFileName> is the second parameter given to the function | |
558 | // | |
9827400b | 559 | |
560 | if (fTestMode & kErrorStorage) | |
561 | { | |
562 | Log(fCurrentDetector, "StoreReferenceFile - In TESTMODE - Simulating error while storing locally"); | |
563 | return kFALSE; | |
564 | } | |
565 | ||
566 | AliCDBManager* man = AliCDBManager::Instance(); | |
567 | AliCDBStorage* sto = man->GetStorage(fgkLocalRefStorage); | |
568 | ||
569 | TString localBaseFolder = sto->GetBaseFolder(); | |
570 | ||
d524ade6 | 571 | TString target = GetRefFilePrefix(localBaseFolder.Data(), detector); |
572 | target.Append(Form("/%d_%s", GetCurrentRun(), gridFileName)); | |
9827400b | 573 | |
d524ade6 | 574 | return CopyFileLocally(localFile, target); |
c88ad5db | 575 | } |
576 | ||
577 | //______________________________________________________________________________________________ | |
578 | Bool_t AliShuttle::StoreRunMetadataFile(const char* localFile, const char* gridFileName) | |
579 | { | |
580 | // | |
581 | // Stores Run metadata file to the Grid, in the run folder | |
582 | // | |
583 | // Only GRP can call this function. | |
584 | ||
585 | if (fTestMode & kErrorStorage) | |
586 | { | |
587 | Log(fCurrentDetector, "StoreRunMetaDataFile - In TESTMODE - Simulating error while storing locally"); | |
588 | return kFALSE; | |
589 | } | |
590 | ||
591 | AliCDBManager* man = AliCDBManager::Instance(); | |
592 | AliCDBStorage* sto = man->GetStorage(fgkLocalRefStorage); | |
593 | ||
594 | TString localBaseFolder = sto->GetBaseFolder(); | |
595 | ||
596 | // Build Run level folder | |
955d2abc | 597 | // folder = /alice/data/year/lhcPeriod/runNb/raw |
c88ad5db | 598 | |
c88ad5db | 599 | |
675f64cd | 600 | TString lhcPeriod = GetLHCPeriod(); |
c88ad5db | 601 | if (lhcPeriod.Length() == 0) |
602 | { | |
603 | Log("SHUTTLE","StoreRunMetaDataFile - LHCPeriod not found in logbook!"); | |
604 | return 0; | |
605 | } | |
a1c5723b | 606 | |
607 | // TODO partitions with one detector only write data into LHCperiod_DET | |
2694f253 | 608 | TString partition = GetRunParameter("detector"); |
a1c5723b | 609 | |
610 | if (partition.Length() > 0 && partition != "ALICE") | |
611 | { | |
612 | lhcPeriod.Append(Form("_%s", partition.Data())); | |
613 | Log(fCurrentDetector, Form("Run data tags merged file will be written in %s", | |
614 | lhcPeriod.Data())); | |
615 | } | |
8884d71e | 616 | |
955d2abc | 617 | TString target = Form("%s/GRP/RunMetadata/alice/data/%d/%s/%09d/raw/%s", |
675f64cd | 618 | localBaseFolder.Data(), GetCurrentYear(), |
d524ade6 | 619 | lhcPeriod.Data(), GetCurrentRun(), gridFileName); |
c88ad5db | 620 | |
d524ade6 | 621 | return CopyFileLocally(localFile, target); |
c88ad5db | 622 | } |
623 | ||
624 | //______________________________________________________________________________________________ | |
d524ade6 | 625 | Bool_t AliShuttle::CopyFileLocally(const char* localFile, const TString& target) |
c88ad5db | 626 | { |
627 | // | |
628 | // Stores file locally. Called by StoreReferenceFile and StoreRunMetadataFile | |
d524ade6 | 629 | // Files are temporarily stored in the local reference storage. When the preprocessor |
630 | // finishes, the Shuttle calls CopyFilesToGrid to transfer the files to AliEn | |
631 | // (in reference or run level folders) | |
c88ad5db | 632 | // |
633 | ||
d524ade6 | 634 | TString targetDir(target(0, target.Last('/'))); |
635 | ||
636 | //try to open base dir folder, if it does not exist | |
2d9019b4 | 637 | void* dir = gSystem->OpenDirectory(targetDir.Data()); |
638 | if (dir == NULL) { | |
639 | if (gSystem->mkdir(targetDir.Data(), kTRUE)) { | |
a1c5723b | 640 | Log("SHUTTLE", Form("CopyFileLocally - Can't open directory <%s>", targetDir.Data())); |
2d9019b4 | 641 | return kFALSE; |
642 | } | |
643 | ||
644 | } else { | |
645 | gSystem->FreeDirectory(dir); | |
646 | } | |
9827400b | 647 | |
7d43a416 | 648 | Int_t result = 0; |
649 | ||
650 | result = gSystem->GetPathInfo(localFile, 0, (Long64_t*) 0, 0, 0); | |
9827400b | 651 | if (result) |
652 | { | |
a1c5723b | 653 | Log("SHUTTLE", Form("CopyFileLocally - %s does not exist", localFile)); |
546242fb | 654 | return kFALSE; |
9827400b | 655 | } |
546242fb | 656 | |
7d43a416 | 657 | result = gSystem->GetPathInfo(target, 0, (Long64_t*) 0, 0, 0); |
658 | if (!result) | |
659 | { | |
a1c5723b | 660 | Log("SHUTTLE", Form("CopyFileLocally - target file %s already exist, removing...", target.Data())); |
7d43a416 | 661 | if (gSystem->Unlink(target.Data())) |
662 | { | |
a1c5723b | 663 | Log("SHUTTLE", Form("CopyFileLocally - Could not remove existing target file %s!", target.Data())); |
7d43a416 | 664 | return kFALSE; |
665 | } | |
666 | } | |
667 | ||
9827400b | 668 | result = gSystem->CopyFile(localFile, target); |
669 | ||
670 | if (result == 0) | |
671 | { | |
a1c5723b | 672 | Log("SHUTTLE", Form("CopyFileLocally - File %s stored locally to %s", localFile, target.Data())); |
9827400b | 673 | return kTRUE; |
674 | } | |
675 | else | |
676 | { | |
a1c5723b | 677 | Log("SHUTTLE", Form("CopyFileLocally - Could not store file %s to %s! Error code = %d", |
546242fb | 678 | localFile, target.Data(), result)); |
9827400b | 679 | return kFALSE; |
680 | } | |
c88ad5db | 681 | |
682 | ||
683 | ||
9827400b | 684 | } |
685 | ||
686 | //______________________________________________________________________________________________ | |
c88ad5db | 687 | Bool_t AliShuttle::CopyFilesToGrid(const char* type) |
9827400b | 688 | { |
689 | // | |
c88ad5db | 690 | // Transfers local files to the Grid. Local files can be reference files |
691 | // or run metadata file (from GRP only). | |
9827400b | 692 | // |
c88ad5db | 693 | // According to the type (ref, metadata) the files are stored under the following location: |
694 | // ref --> <base folder of reference storage>/<DET>/<RUN#>_<gridFileName> | |
695 | // metadata --> <run data folder>/<MetadataFileName> | |
86aa42c3 | 696 | // |
c88ad5db | 697 | |
9827400b | 698 | AliCDBManager* man = AliCDBManager::Instance(); |
699 | AliCDBStorage* sto = man->GetStorage(fgkLocalRefStorage); | |
700 | if (!sto) | |
701 | return kFALSE; | |
702 | TString localBaseFolder = sto->GetBaseFolder(); | |
9827400b | 703 | |
c88ad5db | 704 | TString dir; |
705 | TString alienDir; | |
9827400b | 706 | TString begin; |
9827400b | 707 | |
c88ad5db | 708 | if (strcmp(type, "reference") == 0) |
709 | { | |
710 | dir = GetRefFilePrefix(localBaseFolder.Data(), fCurrentDetector.Data()); | |
711 | AliCDBStorage* gridSto = man->GetStorage(fgkMainRefStorage); | |
712 | if (!gridSto) | |
713 | return kFALSE; | |
714 | TString gridBaseFolder = gridSto->GetBaseFolder(); | |
715 | alienDir = GetRefFilePrefix(gridBaseFolder.Data(), fCurrentDetector.Data()); | |
716 | begin = Form("%d_", GetCurrentRun()); | |
717 | } | |
718 | else if (strcmp(type, "metadata") == 0) | |
719 | { | |
c88ad5db | 720 | |
675f64cd | 721 | TString lhcPeriod = GetLHCPeriod(); |
c88ad5db | 722 | |
723 | if (lhcPeriod.Length() == 0) | |
724 | { | |
725 | Log("SHUTTLE","CopyFilesToGrid - LHCPeriod not found in logbook!"); | |
726 | return 0; | |
727 | } | |
728 | ||
a1c5723b | 729 | // TODO partitions with one detector only write data into LHCperiod_DET |
2694f253 | 730 | TString partition = GetRunParameter("detector"); |
a1c5723b | 731 | |
732 | if (partition.Length() > 0 && partition != "ALICE") | |
733 | { | |
734 | lhcPeriod.Append(Form("_%s", partition.Data())); | |
735 | } | |
736 | ||
955d2abc | 737 | dir = Form("%s/GRP/RunMetadata/alice/data/%d/%s/%09d/raw", |
675f64cd | 738 | localBaseFolder.Data(), GetCurrentYear(), |
c88ad5db | 739 | lhcPeriod.Data(), GetCurrentRun()); |
675f64cd | 740 | alienDir = dir(dir.Index("/alice/data/"), dir.Length()); |
741 | ||
c88ad5db | 742 | begin = ""; |
743 | } | |
744 | else | |
745 | { | |
746 | Log("SHUTTLE", "CopyFilesToGrid - Unexpected: type label must be reference or metadata!"); | |
747 | return kFALSE; | |
748 | } | |
749 | ||
9827400b | 750 | TSystemDirectory* baseDir = new TSystemDirectory("/", dir); |
3d8bc902 | 751 | if (!baseDir) |
752 | return kTRUE; | |
753 | ||
2d9019b4 | 754 | TList* dirList = baseDir->GetListOfFiles(); |
755 | delete baseDir; | |
756 | ||
757 | if (!dirList) return kTRUE; | |
758 | ||
759 | if (dirList->GetEntries() < 3) | |
3d8bc902 | 760 | { |
2d9019b4 | 761 | delete dirList; |
9827400b | 762 | return kTRUE; |
3d8bc902 | 763 | } |
2d9019b4 | 764 | |
546242fb | 765 | if (!gGrid) |
766 | { | |
c88ad5db | 767 | Log("SHUTTLE", "CopyFilesToGrid - Connection to Grid failed: Cannot continue!"); |
2d9019b4 | 768 | delete dirList; |
546242fb | 769 | return kFALSE; |
770 | } | |
771 | ||
2d9019b4 | 772 | Int_t nDirs = 0, nTransfer = 0; |
773 | TIter dirIter(dirList); | |
774 | TSystemFile* entry = 0; | |
775 | ||
9827400b | 776 | Bool_t success = kTRUE; |
3d8bc902 | 777 | Bool_t first = kTRUE; |
9827400b | 778 | |
2d9019b4 | 779 | while ((entry = dynamic_cast<TSystemFile*> (dirIter.Next()))) |
780 | { | |
9827400b | 781 | if (entry->IsDirectory()) |
782 | continue; | |
783 | ||
784 | TString fileName(entry->GetName()); | |
785 | if (!fileName.BeginsWith(begin)) | |
786 | continue; | |
787 | ||
2d9019b4 | 788 | nDirs++; |
789 | ||
3d8bc902 | 790 | if (first) |
791 | { | |
792 | first = kFALSE; | |
c88ad5db | 793 | // check that folder exists, otherwise create it |
3d8bc902 | 794 | TGridResult* result = gGrid->Ls(alienDir.Data(), "a"); |
795 | ||
796 | if (!result) | |
2d9019b4 | 797 | { |
798 | delete dirList; | |
3d8bc902 | 799 | return kFALSE; |
2d9019b4 | 800 | } |
3d8bc902 | 801 | |
546242fb | 802 | if (!result->GetFileName(1)) // TODO: It looks like element 0 is always 0!! |
3d8bc902 | 803 | { |
675f64cd | 804 | // TODO It does not work currently! Bug in TAliEn::Mkdir |
805 | // TODO Manually fixed in local root v5-16-00 | |
c88ad5db | 806 | if (!gGrid->Mkdir(alienDir.Data(),"-p",0)) |
3d8bc902 | 807 | { |
c88ad5db | 808 | Log("SHUTTLE", Form("CopyFilesToGrid - Cannot create directory %s", |
3d8bc902 | 809 | alienDir.Data())); |
2d9019b4 | 810 | delete dirList; |
3d8bc902 | 811 | return kFALSE; |
546242fb | 812 | } else { |
c88ad5db | 813 | Log("SHUTTLE",Form("CopyFilesToGrid - Folder %s created", alienDir.Data())); |
3d8bc902 | 814 | } |
815 | ||
546242fb | 816 | } else { |
c88ad5db | 817 | Log("SHUTTLE",Form("CopyFilesToGrid - Folder %s found", alienDir.Data())); |
3d8bc902 | 818 | } |
819 | } | |
820 | ||
9827400b | 821 | TString fullLocalPath; |
822 | fullLocalPath.Form("%s/%s", dir.Data(), fileName.Data()); | |
823 | ||
824 | TString fullGridPath; | |
825 | fullGridPath.Form("alien://%s/%s", alienDir.Data(), fileName.Data()); | |
826 | ||
a986b218 | 827 | Bool_t result = TFile::Cp(fullLocalPath, fullGridPath); |
9827400b | 828 | |
829 | if (result) | |
830 | { | |
c88ad5db | 831 | Log("SHUTTLE", Form("CopyFilesToGrid - Copying local file %s to %s succeeded!", |
832 | fullLocalPath.Data(), fullGridPath.Data())); | |
9827400b | 833 | RemoveFile(fullLocalPath); |
2d9019b4 | 834 | nTransfer++; |
9827400b | 835 | } |
836 | else | |
837 | { | |
c88ad5db | 838 | Log("SHUTTLE", Form("CopyFilesToGrid - Copying local file %s to %s FAILED!", |
839 | fullLocalPath.Data(), fullGridPath.Data())); | |
9827400b | 840 | success = kFALSE; |
841 | } | |
842 | } | |
2d9019b4 | 843 | |
c88ad5db | 844 | Log("SHUTTLE", Form("CopyFilesToGrid - %d (over %d) files in folder %s copied to Grid.", |
845 | nTransfer, nDirs, dir.Data())); | |
2d9019b4 | 846 | |
847 | ||
848 | delete dirList; | |
9827400b | 849 | return success; |
850 | } | |
851 | ||
2d9019b4 | 852 | //______________________________________________________________________________________________ |
853 | const char* AliShuttle::GetRefFilePrefix(const char* base, const char* detector) | |
854 | { | |
855 | // | |
856 | // Get folder name of reference files | |
857 | // | |
858 | ||
859 | TString offDetStr(GetOfflineDetName(detector)); | |
860 | TString dir; | |
861 | if (offDetStr == "ITS" || offDetStr == "MUON" || offDetStr == "PHOS") | |
862 | { | |
863 | dir.Form("%s/%s/%s", base, offDetStr.Data(), detector); | |
864 | } else { | |
865 | dir.Form("%s/%s", base, offDetStr.Data()); | |
866 | } | |
867 | ||
868 | return dir.Data(); | |
869 | ||
870 | ||
871 | } | |
c88ad5db | 872 | |
3301427a | 873 | //______________________________________________________________________________________________ |
874 | void AliShuttle::CleanLocalStorage(const TString& uri) | |
875 | { | |
9827400b | 876 | // |
877 | // Called in case the preprocessor is declared failed. Remove remaining objects from the local storages. | |
878 | // | |
3301427a | 879 | |
880 | const char* type = 0; | |
881 | if(uri == fgkLocalCDB) { | |
882 | type = "OCDB"; | |
883 | } else if(uri == fgkLocalRefStorage) { | |
546242fb | 884 | type = "Reference"; |
3301427a | 885 | } else { |
886 | AliError(Form("Invalid storage URI: %s", uri.Data())); | |
887 | return; | |
888 | } | |
889 | ||
890 | AliCDBManager* man = AliCDBManager::Instance(); | |
b948db8d | 891 | |
3301427a | 892 | // open local storage |
893 | AliCDBStorage *localSto = man->GetStorage(uri); | |
894 | if(!localSto) { | |
895 | Log("SHUTTLE", | |
896 | Form("CleanLocalStorage - cannot activate local %s storage", type)); | |
897 | return; | |
898 | } | |
899 | ||
900 | TString filename(Form("%s/%s/*/Run*_v%d_s*.root", | |
546242fb | 901 | localSto->GetBaseFolder().Data(), GetOfflineDetName(fCurrentDetector.Data()), GetCurrentRun())); |
3301427a | 902 | |
c88ad5db | 903 | AliDebug(2, Form("filename = %s", filename.Data())); |
3301427a | 904 | |
c88ad5db | 905 | Log("SHUTTLE", Form("Removing remaining local files for run %d and detector %s ...", |
3301427a | 906 | GetCurrentRun(), fCurrentDetector.Data())); |
907 | ||
908 | RemoveFile(filename.Data()); | |
909 | ||
910 | } | |
911 | ||
912 | //______________________________________________________________________________________________ | |
913 | void AliShuttle::RemoveFile(const char* filename) | |
914 | { | |
9827400b | 915 | // |
916 | // removes local file | |
917 | // | |
3301427a | 918 | |
919 | TString command(Form("rm -f %s", filename)); | |
920 | ||
921 | Int_t result = gSystem->Exec(command.Data()); | |
922 | if(result != 0) | |
923 | { | |
924 | Log("SHUTTLE", Form("RemoveFile - %s: Cannot remove file %s!", | |
925 | fCurrentDetector.Data(), filename)); | |
926 | } | |
73abe331 | 927 | } |
928 | ||
b948db8d | 929 | //______________________________________________________________________________________________ |
5164a766 | 930 | AliShuttleStatus* AliShuttle::ReadShuttleStatus() |
931 | { | |
9827400b | 932 | // |
933 | // Reads the AliShuttleStatus from the CDB | |
934 | // | |
5164a766 | 935 | |
2bb7b766 | 936 | if (fStatusEntry){ |
937 | delete fStatusEntry; | |
938 | fStatusEntry = 0; | |
939 | } | |
5164a766 | 940 | |
10a5a932 | 941 | fStatusEntry = AliCDBManager::Instance()->GetStorage(GetLocalCDB()) |
2bb7b766 | 942 | ->Get(Form("/SHUTTLE/STATUS/%s", fCurrentDetector.Data()), GetCurrentRun()); |
5164a766 | 943 | |
2bb7b766 | 944 | if (!fStatusEntry) return 0; |
945 | fStatusEntry->SetOwner(1); | |
5164a766 | 946 | |
2bb7b766 | 947 | AliShuttleStatus* status = dynamic_cast<AliShuttleStatus*> (fStatusEntry->GetObject()); |
948 | if (!status) { | |
949 | AliError("Invalid object stored to CDB!"); | |
950 | return 0; | |
951 | } | |
5164a766 | 952 | |
2bb7b766 | 953 | return status; |
5164a766 | 954 | } |
955 | ||
956 | //______________________________________________________________________________________________ | |
7bfb2090 | 957 | Bool_t AliShuttle::WriteShuttleStatus(AliShuttleStatus* status) |
5164a766 | 958 | { |
9827400b | 959 | // |
960 | // writes the status for one subdetector | |
961 | // | |
2bb7b766 | 962 | |
963 | if (fStatusEntry){ | |
964 | delete fStatusEntry; | |
965 | fStatusEntry = 0; | |
966 | } | |
5164a766 | 967 | |
2bb7b766 | 968 | Int_t run = GetCurrentRun(); |
5164a766 | 969 | |
2bb7b766 | 970 | AliCDBId id(AliCDBPath("SHUTTLE", "STATUS", fCurrentDetector), run, run); |
5164a766 | 971 | |
2bb7b766 | 972 | fStatusEntry = new AliCDBEntry(status, id, new AliCDBMetaData); |
973 | fStatusEntry->SetOwner(1); | |
5164a766 | 974 | |
2bb7b766 | 975 | UInt_t result = AliCDBManager::Instance()->GetStorage(fgkLocalCDB)->Put(fStatusEntry); |
7bfb2090 | 976 | |
2bb7b766 | 977 | if (!result) { |
3301427a | 978 | Log("SHUTTLE", Form("WriteShuttleStatus - Failed for %s, run %d", |
979 | fCurrentDetector.Data(), run)); | |
2bb7b766 | 980 | return kFALSE; |
981 | } | |
e7f62f16 | 982 | |
983 | SendMLInfo(); | |
7bfb2090 | 984 | |
2bb7b766 | 985 | return kTRUE; |
5164a766 | 986 | } |
987 | ||
988 | //______________________________________________________________________________________________ | |
989 | void AliShuttle::UpdateShuttleStatus(AliShuttleStatus::Status newStatus, Bool_t increaseCount) | |
990 | { | |
9827400b | 991 | // |
992 | // changes the AliShuttleStatus for the given detector and run to the given status | |
993 | // | |
5164a766 | 994 | |
2bb7b766 | 995 | if (!fStatusEntry){ |
996 | AliError("UNEXPECTED: fStatusEntry empty"); | |
997 | return; | |
998 | } | |
5164a766 | 999 | |
2bb7b766 | 1000 | AliShuttleStatus* status = dynamic_cast<AliShuttleStatus*> (fStatusEntry->GetObject()); |
5164a766 | 1001 | |
2bb7b766 | 1002 | if (!status){ |
c88ad5db | 1003 | Log("SHUTTLE", "UpdateShuttleStatus - UNEXPECTED: status could not be read from current CDB entry"); |
2bb7b766 | 1004 | return; |
1005 | } | |
5164a766 | 1006 | |
2c15234c | 1007 | TString actionStr = Form("UpdateShuttleStatus - %s: Changing state from %s to %s", |
eba76848 | 1008 | fCurrentDetector.Data(), |
36c99a6a | 1009 | status->GetStatusName(), |
eba76848 | 1010 | status->GetStatusName(newStatus)); |
cb343cfd | 1011 | Log("SHUTTLE", actionStr); |
1012 | SetLastAction(actionStr); | |
5164a766 | 1013 | |
2bb7b766 | 1014 | status->SetStatus(newStatus); |
1015 | if (increaseCount) status->IncreaseCount(); | |
5164a766 | 1016 | |
2bb7b766 | 1017 | AliCDBManager::Instance()->GetStorage(fgkLocalCDB)->Put(fStatusEntry); |
e7f62f16 | 1018 | |
1019 | SendMLInfo(); | |
5164a766 | 1020 | } |
e7f62f16 | 1021 | |
1022 | //______________________________________________________________________________________________ | |
1023 | void AliShuttle::SendMLInfo() | |
1024 | { | |
1025 | // | |
1026 | // sends ML information about the current status of the current detector being processed | |
1027 | // | |
1028 | ||
1029 | AliShuttleStatus* status = dynamic_cast<AliShuttleStatus*> (fStatusEntry->GetObject()); | |
1030 | ||
1031 | if (!status){ | |
3301427a | 1032 | Log("SHUTTLE", "SendMLInfo - UNEXPECTED: status could not be read from current CDB entry"); |
e7f62f16 | 1033 | return; |
1034 | } | |
1035 | ||
1036 | TMonaLisaText mlStatus(Form("%s_status", fCurrentDetector.Data()), status->GetStatusName()); | |
1037 | TMonaLisaValue mlRetryCount(Form("%s_count", fCurrentDetector.Data()), status->GetCount()); | |
1038 | ||
1039 | TList mlList; | |
1040 | mlList.Add(&mlStatus); | |
1041 | mlList.Add(&mlRetryCount); | |
1042 | ||
ee6f7523 | 1043 | TString mlID; |
1044 | mlID.Form("%d", GetCurrentRun()); | |
1045 | fMonaLisa->SendParameters(&mlList, mlID); | |
e7f62f16 | 1046 | } |
1047 | ||
5164a766 | 1048 | //______________________________________________________________________________________________ |
1049 | Bool_t AliShuttle::ContinueProcessing() | |
1050 | { | |
9827400b | 1051 | // this function reads the AliShuttleStatus information from CDB and |
1052 | // checks if the processing should be continued | |
1053 | // if yes it returns kTRUE and updates the AliShuttleStatus with nextStatus | |
2bb7b766 | 1054 | |
926cbc0e | 1055 | if (!fConfig->HostProcessDetector(fCurrentDetector)) |
1056 | return kFALSE; | |
57c1a579 | 1057 | |
1058 | AliPreprocessor* aPreprocessor = | |
1059 | dynamic_cast<AliPreprocessor*> (fPreprocessorMap.GetValue(fCurrentDetector)); | |
1060 | if (!aPreprocessor) | |
1061 | { | |
c88ad5db | 1062 | Log("SHUTTLE", Form("ContinueProcessing - %s: no preprocessor registered", fCurrentDetector.Data())); |
57c1a579 | 1063 | return kFALSE; |
1064 | } | |
1065 | ||
2bb7b766 | 1066 | AliShuttleLogbookEntry::Status entryStatus = |
eba76848 | 1067 | fLogbookEntry->GetDetectorStatus(fCurrentDetector); |
2bb7b766 | 1068 | |
5e993b6f | 1069 | if (entryStatus != AliShuttleLogbookEntry::kUnprocessed) { |
c88ad5db | 1070 | Log("SHUTTLE", Form("ContinueProcessing - %s is %s", |
2bb7b766 | 1071 | fCurrentDetector.Data(), |
1072 | fLogbookEntry->GetDetectorStatusName(entryStatus))); | |
1073 | return kFALSE; | |
1074 | } | |
1075 | ||
1076 | // if we get here, according to Shuttle logbook subdetector is in UNPROCESSED state | |
be48e3ea | 1077 | |
1078 | // check if current run is first unprocessed run for current detector | |
1079 | if (fConfig->StrictRunOrder(fCurrentDetector) && | |
1080 | !fFirstUnprocessed[GetDetPos(fCurrentDetector)]) | |
1081 | { | |
86aa42c3 | 1082 | if (fTestMode == kNone) |
1083 | { | |
c88ad5db | 1084 | Log("SHUTTLE", Form("ContinueProcessing - %s requires strict run ordering" |
1085 | " but this is not the first unprocessed run!")); | |
86aa42c3 | 1086 | return kFALSE; |
1087 | } | |
1088 | else | |
1089 | { | |
c88ad5db | 1090 | Log("SHUTTLE", Form("ContinueProcessing - In TESTMODE - " |
1091 | "Although %s requires strict run ordering " | |
1092 | "and this is not the first unprocessed run, " | |
1093 | "the SHUTTLE continues")); | |
86aa42c3 | 1094 | } |
be48e3ea | 1095 | } |
1096 | ||
926cbc0e | 1097 | // Is the subdetector processed first time for this run? |
1098 | fFirstProcessing = kFALSE; | |
1099 | ||
2bb7b766 | 1100 | AliShuttleStatus* status = ReadShuttleStatus(); |
1101 | if (!status) { | |
1102 | // first time | |
1103 | Log("SHUTTLE", Form("ContinueProcessing - %s: Processing first time", | |
1104 | fCurrentDetector.Data())); | |
1105 | status = new AliShuttleStatus(AliShuttleStatus::kStarted); | |
926cbc0e | 1106 | fFirstProcessing = kTRUE; |
2bb7b766 | 1107 | return WriteShuttleStatus(status); |
1108 | } | |
1109 | ||
1110 | // The following two cases shouldn't happen if Shuttle Logbook was correctly updated. | |
1111 | // If it happens it may mean Logbook updating failed... let's do it now! | |
1112 | if (status->GetStatus() == AliShuttleStatus::kDone || | |
1113 | status->GetStatus() == AliShuttleStatus::kFailed){ | |
1114 | Log("SHUTTLE", Form("ContinueProcessing - %s is already %s. Updating Shuttle Logbook", | |
1115 | fCurrentDetector.Data(), | |
1116 | status->GetStatusName(status->GetStatus()))); | |
1117 | UpdateShuttleLogbook(fCurrentDetector.Data(), | |
1118 | status->GetStatusName(status->GetStatus())); | |
1119 | return kFALSE; | |
1120 | } | |
1121 | ||
5e993b6f | 1122 | if (status->GetStatus() == AliShuttleStatus::kStoreStarted || status->GetStatus() == AliShuttleStatus::kStoreDelayed ||status->GetStatus() == AliShuttleStatus::kStoreError) { |
2bb7b766 | 1123 | Log("SHUTTLE", |
c88ad5db | 1124 | Form("ContinueProcessing - %s: Grid storage of one or more " |
1125 | "objects failed. Trying again now", | |
2bb7b766 | 1126 | fCurrentDetector.Data())); |
5e993b6f | 1127 | StoreOCDB(); |
2bb7b766 | 1128 | return kFALSE; |
1129 | } | |
1130 | ||
1131 | // if we get here, there is a restart | |
57c1a579 | 1132 | Bool_t cont = kFALSE; |
2bb7b766 | 1133 | |
1134 | // abort conditions | |
cb343cfd | 1135 | if (status->GetCount() >= fConfig->GetMaxRetries()) { |
57c1a579 | 1136 | Log("SHUTTLE", Form("ContinueProcessing - %s failed %d times in status %s - " |
1137 | "Updating Shuttle Logbook", fCurrentDetector.Data(), | |
2bb7b766 | 1138 | status->GetCount(), status->GetStatusName())); |
1139 | UpdateShuttleLogbook(fCurrentDetector.Data(), "FAILED"); | |
e7f62f16 | 1140 | UpdateShuttleStatus(AliShuttleStatus::kFailed); |
3301427a | 1141 | |
1142 | // there may still be objects in local OCDB and reference storage | |
1143 | // and FXS databases may be not updated: do it now! | |
9827400b | 1144 | |
1145 | // TODO Currently disabled, we want to keep files in case of failure! | |
1146 | // CleanLocalStorage(fgkLocalCDB); | |
1147 | // CleanLocalStorage(fgkLocalRefStorage); | |
1148 | // UpdateTableFailCase(); | |
1149 | ||
1150 | // Send mail to detector expert! | |
c88ad5db | 1151 | Log("SHUTTLE", Form("ContinueProcessing - Sending mail to %s expert...", |
926cbc0e | 1152 | fCurrentDetector.Data())); |
9827400b | 1153 | if (!SendMail()) |
1154 | Log("SHUTTLE", Form("ContinueProcessing - Could not send mail to %s expert", | |
926cbc0e | 1155 | fCurrentDetector.Data())); |
3301427a | 1156 | |
57c1a579 | 1157 | } else { |
1158 | Log("SHUTTLE", Form("ContinueProcessing - %s: restarting. " | |
1159 | "Aborted before with %s. Retry number %d.", fCurrentDetector.Data(), | |
1160 | status->GetStatusName(), status->GetCount())); | |
9827400b | 1161 | Bool_t increaseCount = kTRUE; |
c88ad5db | 1162 | if (status->GetStatus() == AliShuttleStatus::kDCSError || |
1163 | status->GetStatus() == AliShuttleStatus::kDCSStarted) | |
1164 | increaseCount = kFALSE; | |
675f64cd | 1165 | |
9827400b | 1166 | UpdateShuttleStatus(AliShuttleStatus::kStarted, increaseCount); |
57c1a579 | 1167 | cont = kTRUE; |
2bb7b766 | 1168 | } |
1169 | ||
57c1a579 | 1170 | return cont; |
5164a766 | 1171 | } |
1172 | ||
1173 | //______________________________________________________________________________________________ | |
2bb7b766 | 1174 | Bool_t AliShuttle::Process(AliShuttleLogbookEntry* entry) |
58bc3020 | 1175 | { |
73abe331 | 1176 | // |
b948db8d | 1177 | // Makes data retrieval for all detectors in the configuration. |
2bb7b766 | 1178 | // entry: Shuttle logbook entry, contains run paramenters and status of detectors |
1179 | // (Unprocessed, Inactive, Failed or Done). | |
d477ad88 | 1180 | // Returns kFALSE in case of error occured and kTRUE otherwise |
73abe331 | 1181 | // |
1182 | ||
9827400b | 1183 | if (!entry) return kFALSE; |
2bb7b766 | 1184 | |
1185 | fLogbookEntry = entry; | |
1186 | ||
c88ad5db | 1187 | Log("SHUTTLE", Form("\t\t\t^*^*^*^*^*^*^*^*^*^*^*^* run %d: START ^*^*^*^*^*^*^*^*^*^*^*^*", |
9827400b | 1188 | GetCurrentRun())); |
2bb7b766 | 1189 | |
e7f62f16 | 1190 | // Send the information to ML |
1191 | TMonaLisaText mlStatus("SHUTTLE_status", "Processing"); | |
9827400b | 1192 | TMonaLisaText mlRunType("SHUTTLE_runtype", Form("%s (%s)", entry->GetRunType(), entry->GetRunParameter("log"))); |
e7f62f16 | 1193 | |
1194 | TList mlList; | |
1195 | mlList.Add(&mlStatus); | |
9827400b | 1196 | mlList.Add(&mlRunType); |
e7f62f16 | 1197 | |
ee6f7523 | 1198 | TString mlID; |
1199 | mlID.Form("%d", GetCurrentRun()); | |
1200 | fMonaLisa->SendParameters(&mlList, mlID); | |
3301427a | 1201 | |
9827400b | 1202 | if (fLogbookEntry->IsDone()) |
1203 | { | |
1204 | Log("SHUTTLE","Process - Shuttle is already DONE. Updating logbook"); | |
1205 | UpdateShuttleLogbook("shuttle_done"); | |
1206 | fLogbookEntry = 0; | |
1207 | return kTRUE; | |
1208 | } | |
1209 | ||
1210 | // read test mode if flag is set | |
1211 | if (fReadTestMode) | |
1212 | { | |
3d8bc902 | 1213 | fTestMode = kNone; |
9827400b | 1214 | TString logEntry(entry->GetRunParameter("log")); |
1215 | //printf("log entry = %s\n", logEntry.Data()); | |
1216 | TString searchStr("Testmode: "); | |
1217 | Int_t pos = logEntry.Index(searchStr.Data()); | |
1218 | //printf("%d\n", pos); | |
1219 | if (pos >= 0) | |
1220 | { | |
1221 | TSubString subStr = logEntry(pos + searchStr.Length(), logEntry.Length()); | |
1222 | //printf("%s\n", subStr.String().Data()); | |
1223 | TString newStr(subStr.Data()); | |
1224 | TObjArray* token = newStr.Tokenize(' '); | |
1225 | if (token) | |
1226 | { | |
1227 | //token->Print(); | |
1228 | TObjString* tmpStr = dynamic_cast<TObjString*> (token->First()); | |
1229 | if (tmpStr) | |
1230 | { | |
1231 | Int_t testMode = tmpStr->String().Atoi(); | |
1232 | if (testMode > 0) | |
1233 | { | |
c88ad5db | 1234 | Log("SHUTTLE", Form("Process - Enabling test mode %d", testMode)); |
9827400b | 1235 | SetTestMode((TestMode) testMode); |
1236 | } | |
1237 | } | |
1238 | delete token; | |
1239 | } | |
1240 | } | |
1241 | } | |
c88ad5db | 1242 | |
eba76848 | 1243 | fLogbookEntry->Print("all"); |
57f50b3c | 1244 | |
1245 | // Initialization | |
d477ad88 | 1246 | Bool_t hasError = kFALSE; |
5164a766 | 1247 | |
675f64cd | 1248 | // Set the CDB and Reference folders according to the year and LHC period |
1249 | TString lhcPeriod(GetLHCPeriod()); | |
1250 | if (lhcPeriod.Length() == 0) | |
1251 | { | |
7d4cf768 | 1252 | Log("SHUTTLE","Process - LHCPeriod not found in logbook!"); |
1253 | return 0; | |
675f64cd | 1254 | } |
1255 | ||
1256 | if (fgkMainCDB.Length() == 0) | |
1257 | fgkMainCDB = Form("alien://folder=/alice/data/%d/%s/OCDB?user=alidaq?cacheFold=/tmp/OCDBCache", | |
1258 | GetCurrentYear(), lhcPeriod.Data()); | |
1259 | ||
1260 | if (fgkMainRefStorage.Length() == 0) | |
1261 | fgkMainRefStorage = Form("alien://folder=/alice/data/%d/%s/Reference?user=alidaq?cacheFold=/tmp/OCDBCache", | |
1262 | GetCurrentYear(), lhcPeriod.Data()); | |
1263 | ||
57f50b3c | 1264 | // Loop on detectors in the configuration |
b948db8d | 1265 | TIter iter(fConfig->GetDetectors()); |
2bb7b766 | 1266 | TObjString* aDetector = 0; |
b948db8d | 1267 | |
4508ab8b | 1268 | Bool_t first = kTRUE; |
1269 | ||
be48e3ea | 1270 | while ((aDetector = (TObjString*) iter.Next())) |
1271 | { | |
7bfb2090 | 1272 | fCurrentDetector = aDetector->String(); |
5164a766 | 1273 | |
5e993b6f | 1274 | if (ContinueProcessing() == kFALSE) |
1275 | continue; | |
4508ab8b | 1276 | |
1277 | if (first) | |
1278 | { | |
1279 | // only read QueryCDB when needed and only once | |
1280 | AliCDBStorage *mainCDBSto = AliCDBManager::Instance()->GetStorage(fgkMainCDB); | |
1281 | if(mainCDBSto) mainCDBSto->QueryCDB(GetCurrentRun()); | |
1282 | AliCDBStorage *mainRefSto = AliCDBManager::Instance()->GetStorage(fgkMainRefStorage); | |
1283 | if(mainRefSto) mainRefSto->QueryCDB(GetCurrentRun()); | |
1284 | first = kFALSE; | |
1285 | } | |
9e080f92 | 1286 | |
c88ad5db | 1287 | Log("SHUTTLE", Form("\t\t\t****** run %d - %s: START ******", |
2bb7b766 | 1288 | GetCurrentRun(), aDetector->GetName())); |
1289 | ||
9d733021 | 1290 | for(Int_t iSys=0;iSys<3;iSys++) fFXSCalled[iSys]=kFALSE; |
1291 | ||
c88ad5db | 1292 | Log(fCurrentDetector.Data(), "Process - Starting processing"); |
85a80aa9 | 1293 | |
be48e3ea | 1294 | Int_t pid = fork(); |
1295 | ||
1296 | if (pid < 0) | |
1297 | { | |
c88ad5db | 1298 | Log("SHUTTLE", "Process - ERROR: Forking failed"); |
be48e3ea | 1299 | } |
1300 | else if (pid > 0) | |
1301 | { | |
1302 | // parent | |
c88ad5db | 1303 | Log("SHUTTLE", Form("Process - In parent process of %d - %s: Starting monitoring", |
be48e3ea | 1304 | GetCurrentRun(), aDetector->GetName())); |
1305 | ||
1306 | Long_t begin = time(0); | |
1307 | ||
1308 | int status; // to be used with waitpid, on purpose an int (not Int_t)! | |
1309 | while (waitpid(pid, &status, WNOHANG) == 0) | |
1310 | { | |
1311 | Long_t expiredTime = time(0) - begin; | |
1312 | ||
1313 | if (expiredTime > fConfig->GetPPTimeOut()) | |
1314 | { | |
9827400b | 1315 | TString tmp; |
c88ad5db | 1316 | tmp.Form("Process - Process of %s time out. " |
1317 | "Run time: %d seconds. Killing...", | |
1318 | fCurrentDetector.Data(), expiredTime); | |
9827400b | 1319 | Log("SHUTTLE", tmp); |
1320 | Log(fCurrentDetector, tmp); | |
be48e3ea | 1321 | |
1322 | kill(pid, 9); | |
1323 | ||
3301427a | 1324 | UpdateShuttleStatus(AliShuttleStatus::kPPTimeOut); |
be48e3ea | 1325 | hasError = kTRUE; |
1326 | ||
1327 | gSystem->Sleep(1000); | |
1328 | } | |
1329 | else | |
1330 | { | |
be48e3ea | 1331 | gSystem->Sleep(1000); |
9827400b | 1332 | |
1333 | TString checkStr; | |
1334 | checkStr.Form("ps -o vsize --pid %d | tail -n 1", pid); | |
1335 | FILE* pipe = gSystem->OpenPipe(checkStr, "r"); | |
1336 | if (!pipe) | |
1337 | { | |
c88ad5db | 1338 | Log("SHUTTLE", Form("Process - Error: " |
1339 | "Could not open pipe to %s", checkStr.Data())); | |
9827400b | 1340 | continue; |
1341 | } | |
1342 | ||
1343 | char buffer[100]; | |
1344 | if (!fgets(buffer, 100, pipe)) | |
1345 | { | |
c88ad5db | 1346 | Log("SHUTTLE", "Process - Error: ps did not return anything"); |
9827400b | 1347 | gSystem->ClosePipe(pipe); |
1348 | continue; | |
1349 | } | |
1350 | gSystem->ClosePipe(pipe); | |
1351 | ||
1352 | //Log("SHUTTLE", Form("ps returned %s", buffer)); | |
1353 | ||
1354 | Int_t mem = 0; | |
1355 | if ((sscanf(buffer, "%d\n", &mem) != 1) || !mem) | |
1356 | { | |
c88ad5db | 1357 | Log("SHUTTLE", "Process - Error: Could not parse output of ps"); |
9827400b | 1358 | continue; |
1359 | } | |
1360 | ||
1361 | if (expiredTime % 60 == 0) | |
ee6f7523 | 1362 | { |
c88ad5db | 1363 | Log("SHUTTLE", Form("Process - %s: Checking process. " |
1364 | "Run time: %d seconds - Memory consumption: %d KB", | |
1365 | fCurrentDetector.Data(), expiredTime, mem)); | |
ee6f7523 | 1366 | SendAlive(); |
1367 | } | |
9827400b | 1368 | |
1369 | if (mem > fConfig->GetPPMaxMem()) | |
1370 | { | |
1371 | TString tmp; | |
c88ad5db | 1372 | tmp.Form("Process - Process exceeds maximum allowed memory " |
1373 | "(%d KB > %d KB). Killing...", | |
9827400b | 1374 | mem, fConfig->GetPPMaxMem()); |
1375 | Log("SHUTTLE", tmp); | |
1376 | Log(fCurrentDetector, tmp); | |
1377 | ||
1378 | kill(pid, 9); | |
1379 | ||
1380 | UpdateShuttleStatus(AliShuttleStatus::kPPOutOfMemory); | |
1381 | hasError = kTRUE; | |
1382 | ||
1383 | gSystem->Sleep(1000); | |
1384 | } | |
be48e3ea | 1385 | } |
1386 | } | |
1387 | ||
c88ad5db | 1388 | Log("SHUTTLE", Form("Process - In parent process of %d - %s: Client has terminated.", |
be48e3ea | 1389 | GetCurrentRun(), aDetector->GetName())); |
1390 | ||
1391 | if (WIFEXITED(status)) | |
1392 | { | |
1393 | Int_t returnCode = WEXITSTATUS(status); | |
1394 | ||
c88ad5db | 1395 | Log("SHUTTLE", Form("Process - %s: the return code is %d", fCurrentDetector.Data(), |
3301427a | 1396 | returnCode)); |
be48e3ea | 1397 | |
9827400b | 1398 | if (returnCode == 0) hasError = kTRUE; |
be48e3ea | 1399 | } |
1400 | } | |
1401 | else if (pid == 0) | |
1402 | { | |
1403 | // client | |
c88ad5db | 1404 | Log("SHUTTLE", Form("Process - In client process of %d - %s", GetCurrentRun(), |
1405 | aDetector->GetName())); | |
be48e3ea | 1406 | |
c88ad5db | 1407 | Log("SHUTTLE", Form("Process - Redirecting output to %s log",fCurrentDetector.Data())); |
ffa29e93 | 1408 | |
546242fb | 1409 | if ((freopen(GetLogFileName(fCurrentDetector), "a", stdout)) == 0) |
ffa29e93 | 1410 | { |
c88ad5db | 1411 | Log("SHUTTLE", "Process - Could not freopen stdout"); |
ffa29e93 | 1412 | } |
1413 | else | |
1414 | { | |
1415 | fOutputRedirected = kTRUE; | |
1416 | if ((dup2(fileno(stdout), fileno(stderr))) < 0) | |
c88ad5db | 1417 | Log("SHUTTLE", "Process - Could not redirect stderr"); |
ffa29e93 | 1418 | |
1419 | } | |
1420 | ||
5bac2bde | 1421 | TString wd = gSystem->WorkingDirectory(); |
675f64cd | 1422 | TString tmpDir = Form("%s/%s_%d_process", GetShuttleTempDir(), |
1423 | fCurrentDetector.Data(), GetCurrentRun()); | |
5bac2bde | 1424 | |
d524ade6 | 1425 | Int_t result = gSystem->GetPathInfo(tmpDir.Data(), 0, (Long64_t*) 0, 0, 0); |
1426 | if (!result) // temp dir already exists! | |
1427 | { | |
1428 | Log(fCurrentDetector.Data(), | |
1429 | Form("Process - %s dir already exists! Removing...", tmpDir.Data())); | |
1430 | gSystem->Exec(Form("rm -rf %s",tmpDir.Data())); | |
675f64cd | 1431 | } |
1432 | ||
1433 | if (gSystem->mkdir(tmpDir.Data(), 1)) | |
1434 | { | |
1435 | Log(fCurrentDetector.Data(), "Process - could not make temp directory!!"); | |
1436 | gSystem->Exit(1); | |
d524ade6 | 1437 | } |
1438 | ||
1439 | if (!gSystem->ChangeDirectory(tmpDir.Data())) | |
1440 | { | |
1441 | Log(fCurrentDetector.Data(), "Process - could not change directory!!"); | |
1442 | gSystem->Exit(1); | |
1443 | } | |
5bac2bde | 1444 | |
9827400b | 1445 | Bool_t success = ProcessCurrentDetector(); |
5bac2bde | 1446 | |
1447 | gSystem->ChangeDirectory(wd.Data()); | |
b1d18693 | 1448 | |
9827400b | 1449 | if (success) // Preprocessor finished successfully! |
1450 | { | |
fb40ead4 | 1451 | // remove temporary folder or DCS map |
1452 | if (!fConfig->KeepTempFolder()) | |
1453 | { | |
1454 | gSystem->Exec(Form("rm -rf %s",tmpDir.Data())); | |
1455 | } else if (!fConfig->KeepDCSMap()) | |
1456 | { | |
1457 | gSystem->Exec(Form("rm -f %s/DCSMap.root",tmpDir.Data())); | |
1458 | } | |
b1d18693 | 1459 | |
3301427a | 1460 | // Update time_processed field in FXS DB |
1461 | if (UpdateTable() == kFALSE) | |
5bac2bde | 1462 | Log("SHUTTLE", Form("Process - %s: Could not update FXS databases!", |
1463 | fCurrentDetector.Data())); | |
3301427a | 1464 | |
1465 | // Transfer the data from local storage to main storage (Grid) | |
3301427a | 1466 | if (StoreOCDB() == kFALSE) |
9827400b | 1467 | success = kFALSE; |
5e993b6f | 1468 | } |
1469 | else | |
c88ad5db | 1470 | { |
1471 | Log("SHUTTLE", | |
1472 | Form("\t\t\t****** run %d - %s: PP ERROR ******", | |
1473 | GetCurrentRun(), aDetector->GetName())); | |
be48e3ea | 1474 | } |
1475 | ||
4b95672b | 1476 | for (UInt_t iSys=0; iSys<3; iSys++) |
1477 | { | |
1478 | if (fFXSCalled[iSys]) fFXSlist[iSys].Clear(); | |
1479 | } | |
1480 | ||
c88ad5db | 1481 | Log("SHUTTLE", Form("Process - Client process of %d - %s is exiting now with %d.", |
9827400b | 1482 | GetCurrentRun(), aDetector->GetName(), success)); |
be48e3ea | 1483 | |
1484 | // the client exits here | |
9827400b | 1485 | gSystem->Exit(success); |
be48e3ea | 1486 | |
1487 | AliError("We should never get here!!!"); | |
1488 | } | |
7bfb2090 | 1489 | } |
5164a766 | 1490 | |
c88ad5db | 1491 | Log("SHUTTLE", Form("\t\t\t^*^*^*^*^*^*^*^*^*^*^*^* run %d: FINISH ^*^*^*^*^*^*^*^*^*^*^*^*", |
2bb7b766 | 1492 | GetCurrentRun())); |
1493 | ||
1494 | //check if shuttle is done for this run, if so update logbook | |
1495 | TObjArray checkEntryArray; | |
1496 | checkEntryArray.SetOwner(1); | |
9e080f92 | 1497 | TString whereClause = Form("where run=%d", GetCurrentRun()); |
b0e53b15 | 1498 | if (!QueryShuttleLogbook(whereClause.Data(), checkEntryArray) || |
1499 | checkEntryArray.GetEntries() == 0) { | |
9e080f92 | 1500 | Log("SHUTTLE", Form("Process - Warning: Cannot check status of run %d on Shuttle logbook!", |
1501 | GetCurrentRun())); | |
1502 | return hasError == kFALSE; | |
1503 | } | |
b948db8d | 1504 | |
9e080f92 | 1505 | AliShuttleLogbookEntry* checkEntry = dynamic_cast<AliShuttleLogbookEntry*> |
1506 | (checkEntryArray.At(0)); | |
2bb7b766 | 1507 | |
9e080f92 | 1508 | if (checkEntry) |
1509 | { | |
1510 | if (checkEntry->IsDone()) | |
be48e3ea | 1511 | { |
9e080f92 | 1512 | Log("SHUTTLE","Process - Shuttle is DONE. Updating logbook"); |
1513 | UpdateShuttleLogbook("shuttle_done"); | |
1514 | } | |
1515 | else | |
1516 | { | |
1517 | for (UInt_t iDet=0; iDet<NDetectors(); iDet++) | |
be48e3ea | 1518 | { |
9e080f92 | 1519 | if (checkEntry->GetDetectorStatus(iDet) == AliShuttleLogbookEntry::kUnprocessed) |
be48e3ea | 1520 | { |
9e080f92 | 1521 | AliDebug(2, Form("Run %d: setting %s as \"not first time unprocessed\"", |
1522 | checkEntry->GetRun(), GetDetName(iDet))); | |
1523 | fFirstUnprocessed[iDet] = kFALSE; | |
be48e3ea | 1524 | } |
1525 | } | |
2bb7b766 | 1526 | } |
1527 | } | |
1528 | ||
1529 | fLogbookEntry = 0; | |
85a80aa9 | 1530 | |
a7160fe9 | 1531 | return hasError == kFALSE; |
73abe331 | 1532 | } |
1533 | ||
b948db8d | 1534 | //______________________________________________________________________________________________ |
9827400b | 1535 | Bool_t AliShuttle::ProcessCurrentDetector() |
73abe331 | 1536 | { |
1537 | // | |
2bb7b766 | 1538 | // Makes data retrieval just for a specific detector (fCurrentDetector). |
73abe331 | 1539 | // Threre should be a configuration for this detector. |
73abe331 | 1540 | |
1d172743 | 1541 | Log("SHUTTLE", Form("ProcessCurrentDetector - Retrieving values for %s, run %d", |
1542 | fCurrentDetector.Data(), GetCurrentRun())); | |
73abe331 | 1543 | |
d524ade6 | 1544 | TString wd = gSystem->WorkingDirectory(); |
1545 | ||
2d9019b4 | 1546 | if (!CleanReferenceStorage(fCurrentDetector.Data())) |
546242fb | 1547 | return kFALSE; |
d524ade6 | 1548 | |
1549 | gSystem->ChangeDirectory(wd.Data()); | |
1550 | ||
1d172743 | 1551 | TMap* dcsMap = new TMap(); |
3301427a | 1552 | |
1553 | // call preprocessor | |
1554 | AliPreprocessor* aPreprocessor = | |
1555 | dynamic_cast<AliPreprocessor*> (fPreprocessorMap.GetValue(fCurrentDetector)); | |
1556 | ||
1557 | aPreprocessor->Initialize(GetCurrentRun(), GetCurrentStartTime(), GetCurrentEndTime()); | |
1558 | ||
1559 | Bool_t processDCS = aPreprocessor->ProcessDCS(); | |
d477ad88 | 1560 | |
651fdaab | 1561 | if (!processDCS) |
1562 | { | |
1d172743 | 1563 | Log(fCurrentDetector, "ProcessCurrentDetector -" |
1564 | " The preprocessor requested to skip the retrieval of DCS values"); | |
651fdaab | 1565 | } |
8b739301 | 1566 | else if (fTestMode & kSkipDCS) |
2c15234c | 1567 | { |
1d172743 | 1568 | Log(fCurrentDetector, "ProcessCurrentDetector - In TESTMODE: Skipping DCS processing"); |
9827400b | 1569 | } |
1570 | else if (fTestMode & kErrorDCS) | |
1571 | { | |
1d172743 | 1572 | Log(fCurrentDetector, "ProcessCurrentDetector - In TESTMODE: Simulating DCS error"); |
3d8bc902 | 1573 | UpdateShuttleStatus(AliShuttleStatus::kDCSStarted); |
9827400b | 1574 | UpdateShuttleStatus(AliShuttleStatus::kDCSError); |
1d172743 | 1575 | delete dcsMap; |
9827400b | 1576 | return kFALSE; |
2c15234c | 1577 | } else { |
3301427a | 1578 | |
1579 | UpdateShuttleStatus(AliShuttleStatus::kDCSStarted); | |
1580 | ||
1d172743 | 1581 | // Query DCS archive |
1582 | Int_t nServers = fConfig->GetNServers(fCurrentDetector); | |
a038aa70 | 1583 | |
1d172743 | 1584 | for (int iServ=0; iServ<nServers; iServ++) |
2c15234c | 1585 | { |
1d172743 | 1586 | |
1587 | TString host(fConfig->GetDCSHost(fCurrentDetector, iServ)); | |
1588 | Int_t port = fConfig->GetDCSPort(fCurrentDetector, iServ); | |
542b6cc8 | 1589 | Int_t multiSplit = fConfig->GetMultiSplit(fCurrentDetector, iServ); |
1590 | ||
1790d4b7 | 1591 | Log(fCurrentDetector, Form("ProcessCurrentDetector -" |
1592 | " Querying DCS Amanda server %s:%d (%d of %d)", | |
1593 | host.Data(), port, iServ+1, nServers)); | |
1d172743 | 1594 | |
1595 | TMap* aliasMap = 0; | |
1596 | TMap* dpMap = 0; | |
1597 | ||
1598 | if (fConfig->GetDCSAliases(fCurrentDetector, iServ)->GetEntries() > 0) | |
2c15234c | 1599 | { |
1d172743 | 1600 | aliasMap = GetValueSet(host, port, |
542b6cc8 | 1601 | fConfig->GetDCSAliases(fCurrentDetector, iServ), |
1602 | kAlias, multiSplit); | |
1d172743 | 1603 | if (!aliasMap) |
1604 | { | |
1605 | Log(fCurrentDetector, | |
1606 | Form("ProcessCurrentDetector -" | |
675f64cd | 1607 | " Error retrieving DCS aliases from server %s." |
1608 | " Sending mail to DCS experts!", host.Data())); | |
1d172743 | 1609 | UpdateShuttleStatus(AliShuttleStatus::kDCSError); |
675f64cd | 1610 | |
fb40ead4 | 1611 | if (!SendMailToDCS()) |
1612 | Log("SHUTTLE", Form("ProcessCurrentDetector - " | |
95d4aaef | 1613 | "Could not send mail to DCS experts!")); |
675f64cd | 1614 | |
a038aa70 | 1615 | delete dcsMap; |
1d172743 | 1616 | return kFALSE; |
1617 | } | |
2c15234c | 1618 | } |
a038aa70 | 1619 | |
1d172743 | 1620 | if (fConfig->GetDCSDataPoints(fCurrentDetector, iServ)->GetEntries() > 0) |
a038aa70 | 1621 | { |
1d172743 | 1622 | dpMap = GetValueSet(host, port, |
542b6cc8 | 1623 | fConfig->GetDCSDataPoints(fCurrentDetector, iServ), |
1624 | kDP, multiSplit); | |
1d172743 | 1625 | if (!dpMap) |
1626 | { | |
1627 | Log(fCurrentDetector, | |
1628 | Form("ProcessCurrentDetector -" | |
675f64cd | 1629 | " Error retrieving DCS data points from server %s." |
1630 | " Sending mail to DCS experts!", host.Data())); | |
1d172743 | 1631 | UpdateShuttleStatus(AliShuttleStatus::kDCSError); |
675f64cd | 1632 | |
fb40ead4 | 1633 | if (!SendMailToDCS()) |
1634 | Log("SHUTTLE", Form("ProcessCurrentDetector - " | |
95d4aaef | 1635 | "Could not send mail to DCS experts!")); |
675f64cd | 1636 | |
1d172743 | 1637 | if (aliasMap) delete aliasMap; |
1638 | delete dcsMap; | |
1639 | return kFALSE; | |
1640 | } | |
a038aa70 | 1641 | } |
1d172743 | 1642 | |
1643 | // merge aliasMap and dpMap into dcsMap | |
1644 | if(aliasMap) { | |
1645 | TIter iter(aliasMap); | |
a038aa70 | 1646 | TObjString* key = 0; |
1647 | while ((key = (TObjString*) iter.Next())) | |
1d172743 | 1648 | dcsMap->Add(key, aliasMap->GetValue(key->String())); |
1649 | ||
1650 | aliasMap->SetOwner(kFALSE); | |
1651 | delete aliasMap; | |
1652 | } | |
1653 | ||
1654 | if(dpMap) { | |
1655 | TIter iter(dpMap); | |
1656 | TObjString* key = 0; | |
1657 | while ((key = (TObjString*) iter.Next())) | |
1658 | dcsMap->Add(key, dpMap->GetValue(key->String())); | |
1659 | ||
1660 | dpMap->SetOwner(kFALSE); | |
1661 | delete dpMap; | |
a038aa70 | 1662 | } |
73abe331 | 1663 | } |
1664 | } | |
dc25836b | 1665 | |
b1d18693 | 1666 | // save map into file, to help debugging in case of preprocessor error |
fbc112e3 | 1667 | TFile* f = TFile::Open("DCSMap.root","recreate"); |
b1d18693 | 1668 | f->cd(); |
1669 | dcsMap->Write("DCSMap", TObject::kSingleKey); | |
1670 | f->Close(); | |
fbc112e3 | 1671 | delete f; |
b1d18693 | 1672 | |
2bb7b766 | 1673 | // DCS Archive DB processing successful. Call Preprocessor! |
85a80aa9 | 1674 | UpdateShuttleStatus(AliShuttleStatus::kPPStarted); |
a7160fe9 | 1675 | |
a038aa70 | 1676 | UInt_t returnValue = aPreprocessor->Process(dcsMap); |
b948db8d | 1677 | |
3301427a | 1678 | if (returnValue > 0) // Preprocessor error! |
1679 | { | |
c88ad5db | 1680 | Log(fCurrentDetector, Form("ProcessCurrentDetector - " |
1681 | "Preprocessor failed. Process returned %d.", returnValue)); | |
cb343cfd | 1682 | UpdateShuttleStatus(AliShuttleStatus::kPPError); |
a038aa70 | 1683 | dcsMap->DeleteAll(); |
1684 | delete dcsMap; | |
9827400b | 1685 | return kFALSE; |
1686 | } | |
1687 | ||
1688 | // preprocessor ok! | |
1689 | UpdateShuttleStatus(AliShuttleStatus::kPPDone); | |
1690 | Log(fCurrentDetector, Form("ProcessCurrentDetector - %s preprocessor returned success", | |
1691 | fCurrentDetector.Data())); | |
b948db8d | 1692 | |
a038aa70 | 1693 | dcsMap->DeleteAll(); |
1694 | delete dcsMap; | |
b948db8d | 1695 | |
9827400b | 1696 | return kTRUE; |
2bb7b766 | 1697 | } |
1698 | ||
fbc112e3 | 1699 | //______________________________________________________________________________________________ |
1700 | void AliShuttle::CountOpenRuns() | |
1701 | { | |
1702 | // Query DAQ's Shuttle logbook and sends the number of open runs to ML | |
1703 | ||
1704 | // check connection, in case connect | |
1705 | if (!Connect(3)) | |
1706 | return; | |
1707 | ||
1708 | TString sqlQuery; | |
1709 | sqlQuery = Form("select count(*) from %s where shuttle_done=0", fConfig->GetShuttlelbTable()); | |
1710 | ||
1711 | TSQLResult* aResult = fServer[3]->Query(sqlQuery); | |
1712 | if (!aResult) { | |
1713 | AliError(Form("Can't execute query <%s>!", sqlQuery.Data())); | |
1714 | return; | |
1715 | } | |
1716 | ||
1717 | AliDebug(2,Form("Query = %s", sqlQuery.Data())); | |
1718 | ||
1719 | if (aResult->GetRowCount() == 0) { | |
1720 | AliError(Form("No result for query %s received", sqlQuery.Data())); | |
1721 | return; | |
1722 | } | |
1723 | ||
1724 | if (aResult->GetFieldCount() != 1) { | |
1725 | AliError(Form("Invalid field count for query %s received", sqlQuery.Data())); | |
1726 | return; | |
1727 | } | |
1728 | ||
1729 | TSQLRow* aRow = aResult->Next(); | |
1730 | if (!aRow) { | |
1731 | AliError(Form("Could not receive result of query %s", sqlQuery.Data())); | |
1732 | return; | |
1733 | } | |
1734 | ||
1735 | TString result(aRow->GetField(0), aRow->GetFieldLength(0)); | |
1736 | Int_t count = result.Atoi(); | |
1737 | ||
1738 | Log("SHUTTLE", Form("%d unprocessed runs", count)); | |
1739 | ||
1740 | delete aRow; | |
1741 | delete aResult; | |
1742 | ||
1743 | TMonaLisaValue mlStatus("SHUTTLE_openruns", count); | |
1744 | ||
1745 | TList mlList; | |
1746 | mlList.Add(&mlStatus); | |
1747 | ||
1748 | fMonaLisa->SendParameters(&mlList, "__PROCESSINGINFO__"); | |
1749 | } | |
1750 | ||
2bb7b766 | 1751 | //______________________________________________________________________________________________ |
1752 | Bool_t AliShuttle::QueryShuttleLogbook(const char* whereClause, | |
1753 | TObjArray& entries) | |
1754 | { | |
9827400b | 1755 | // Query DAQ's Shuttle logbook and fills detector status object. |
1756 | // Call QueryRunParameters to query DAQ logbook for run parameters. | |
1757 | // | |
2bb7b766 | 1758 | |
fc5a4708 | 1759 | entries.SetOwner(1); |
1760 | ||
2bb7b766 | 1761 | // check connection, in case connect |
fbc112e3 | 1762 | if (!Connect(3)) return kFALSE; |
2bb7b766 | 1763 | |
1764 | TString sqlQuery; | |
441b0e9c | 1765 | sqlQuery = Form("select * from %s %s order by run", fConfig->GetShuttlelbTable(), whereClause); |
2bb7b766 | 1766 | |
be48e3ea | 1767 | TSQLResult* aResult = fServer[3]->Query(sqlQuery); |
2bb7b766 | 1768 | if (!aResult) { |
1769 | AliError(Form("Can't execute query <%s>!", sqlQuery.Data())); | |
1770 | return kFALSE; | |
1771 | } | |
1772 | ||
fc5a4708 | 1773 | AliDebug(2,Form("Query = %s", sqlQuery.Data())); |
1774 | ||
2bb7b766 | 1775 | if(aResult->GetRowCount() == 0) { |
c88ad5db | 1776 | Log("SHUTTLE", "No entries in Shuttle Logbook match request"); |
9827400b | 1777 | delete aResult; |
1778 | return kTRUE; | |
2bb7b766 | 1779 | } |
1780 | ||
1781 | // TODO Check field count! | |
db99d43e | 1782 | const UInt_t nCols = 23; |
2bb7b766 | 1783 | if (aResult->GetFieldCount() != (Int_t) nCols) { |
c88ad5db | 1784 | Log("SHUTTLE", "Invalid SQL result field number!"); |
2bb7b766 | 1785 | delete aResult; |
1786 | return kFALSE; | |
1787 | } | |
1788 | ||
2bb7b766 | 1789 | TSQLRow* aRow; |
1790 | while ((aRow = aResult->Next())) { | |
1791 | TString runString(aRow->GetField(0), aRow->GetFieldLength(0)); | |
1792 | Int_t run = runString.Atoi(); | |
1793 | ||
eba76848 | 1794 | AliShuttleLogbookEntry *entry = QueryRunParameters(run); |
1795 | if (!entry) | |
1796 | continue; | |
2bb7b766 | 1797 | |
1798 | // loop on detectors | |
eba76848 | 1799 | for(UInt_t ii = 0; ii < nCols; ii++) |
1800 | entry->SetDetectorStatus(aResult->GetFieldName(ii), aRow->GetField(ii)); | |
2bb7b766 | 1801 | |
eba76848 | 1802 | entries.AddLast(entry); |
2bb7b766 | 1803 | delete aRow; |
1804 | } | |
1805 | ||
2bb7b766 | 1806 | delete aResult; |
1807 | return kTRUE; | |
1808 | } | |
1809 | ||
1810 | //______________________________________________________________________________________________ | |
eba76848 | 1811 | AliShuttleLogbookEntry* AliShuttle::QueryRunParameters(Int_t run) |
2bb7b766 | 1812 | { |
eba76848 | 1813 | // |
1814 | // Retrieve run parameters written in the DAQ logbook and sets them into AliShuttleLogbookEntry object | |
1815 | // | |
2bb7b766 | 1816 | |
1817 | // check connection, in case connect | |
be48e3ea | 1818 | if (!Connect(3)) |
eba76848 | 1819 | return 0; |
2bb7b766 | 1820 | |
1821 | TString sqlQuery; | |
2c15234c | 1822 | sqlQuery.Form("select * from %s where run=%d", fConfig->GetDAQlbTable(), run); |
2bb7b766 | 1823 | |
be48e3ea | 1824 | TSQLResult* aResult = fServer[3]->Query(sqlQuery); |
2bb7b766 | 1825 | if (!aResult) { |
c88ad5db | 1826 | Log("SHUTTLE", Form("Can't execute query <%s>!", sqlQuery.Data())); |
eba76848 | 1827 | return 0; |
2bb7b766 | 1828 | } |
1829 | ||
eba76848 | 1830 | if (aResult->GetRowCount() == 0) { |
2bb7b766 | 1831 | Log("SHUTTLE", Form("QueryRunParameters - No entry in DAQ Logbook for run %d. Skipping", run)); |
1832 | delete aResult; | |
eba76848 | 1833 | return 0; |
2bb7b766 | 1834 | } |
1835 | ||
eba76848 | 1836 | if (aResult->GetRowCount() > 1) { |
c88ad5db | 1837 | Log("SHUTTLE", Form("QueryRunParameters - UNEXPECTED: " |
1838 | "more than one entry in DAQ Logbook for run %d!", run)); | |
2bb7b766 | 1839 | delete aResult; |
eba76848 | 1840 | return 0; |
2bb7b766 | 1841 | } |
1842 | ||
eba76848 | 1843 | TSQLRow* aRow = aResult->Next(); |
1844 | if (!aRow) | |
1845 | { | |
c88ad5db | 1846 | Log("SHUTTLE", Form("QueryRunParameters - Could not retrieve row for run %d. Skipping", run)); |
eba76848 | 1847 | delete aResult; |
1848 | return 0; | |
1849 | } | |
2bb7b766 | 1850 | |
eba76848 | 1851 | AliShuttleLogbookEntry* entry = new AliShuttleLogbookEntry(run); |
2bb7b766 | 1852 | |
eba76848 | 1853 | for (Int_t ii = 0; ii < aResult->GetFieldCount(); ii++) |
1854 | entry->SetRunParameter(aResult->GetFieldName(ii), aRow->GetField(ii)); | |
2bb7b766 | 1855 | |
eba76848 | 1856 | UInt_t startTime = entry->GetStartTime(); |
1857 | UInt_t endTime = entry->GetEndTime(); | |
1858 | ||
1abfbb60 | 1859 | // if (!startTime || !endTime || startTime > endTime) |
1860 | // { | |
1861 | // Log("SHUTTLE", | |
1862 | // Form("QueryRunParameters - Invalid parameters for Run %d: startTime = %d, endTime = %d. Skipping!", | |
1863 | // run, startTime, endTime)); | |
1864 | // | |
1865 | // Log("SHUTTLE", Form("Marking SHUTTLE done for run %d", run)); | |
1866 | // fLogbookEntry = entry; | |
1867 | // if (!UpdateShuttleLogbook("shuttle_done")) | |
1868 | // { | |
1869 | // AliError(Form("Could not update logbook for run %d !", run)); | |
1870 | // } | |
1871 | // fLogbookEntry = 0; | |
1872 | // | |
1873 | // delete entry; | |
1874 | // delete aRow; | |
1875 | // delete aResult; | |
1876 | // return 0; | |
1877 | // } | |
1878 | ||
1879 | if (!startTime) | |
b0e53b15 | 1880 | { |
1881 | Log("SHUTTLE", | |
1abfbb60 | 1882 | Form("QueryRunParameters - Invalid parameters for Run %d: " |
1883 | "startTime = %d, endTime = %d. Skipping!", | |
1884 | run, startTime, endTime)); | |
b0e53b15 | 1885 | |
1886 | Log("SHUTTLE", Form("Marking SHUTTLE done for run %d", run)); | |
1887 | fLogbookEntry = entry; | |
ee6f7523 | 1888 | if (!UpdateShuttleLogbook("shuttle_ignored")) |
b0e53b15 | 1889 | { |
1890 | AliError(Form("Could not update logbook for run %d !", run)); | |
1891 | } | |
1892 | fLogbookEntry = 0; | |
1893 | ||
1894 | delete entry; | |
1895 | delete aRow; | |
1896 | delete aResult; | |
1897 | return 0; | |
1898 | } | |
1899 | ||
1abfbb60 | 1900 | if (startTime && !endTime) |
1901 | { | |
1902 | // TODO Here we don't mark SHUTTLE done, because this may mean | |
1903 | //the run is still ongoing!! | |
1904 | Log("SHUTTLE", | |
1905 | Form("QueryRunParameters - Invalid parameters for Run %d: " | |
1906 | "startTime = %d, endTime = %d. Skipping (Shuttle won't be marked as DONE)!", | |
1907 | run, startTime, endTime)); | |
1908 | ||
1909 | //Log("SHUTTLE", Form("Marking SHUTTLE done for run %d", run)); | |
1910 | //fLogbookEntry = entry; | |
1911 | //if (!UpdateShuttleLogbook("shuttle_done")) | |
1912 | //{ | |
1913 | // AliError(Form("Could not update logbook for run %d !", run)); | |
1914 | //} | |
1915 | //fLogbookEntry = 0; | |
1916 | ||
1917 | delete entry; | |
1918 | delete aRow; | |
1919 | delete aResult; | |
1920 | return 0; | |
1921 | } | |
1922 | ||
1923 | if (startTime && endTime && (startTime > endTime)) | |
1924 | { | |
1925 | Log("SHUTTLE", | |
1926 | Form("QueryRunParameters - Invalid parameters for Run %d: " | |
1927 | "startTime = %d, endTime = %d. Skipping!", | |
1928 | run, startTime, endTime)); | |
1929 | ||
1930 | Log("SHUTTLE", Form("Marking SHUTTLE done for run %d", run)); | |
1931 | fLogbookEntry = entry; | |
ee6f7523 | 1932 | if (!UpdateShuttleLogbook("shuttle_ignored")) |
1abfbb60 | 1933 | { |
1934 | AliError(Form("Could not update logbook for run %d !", run)); | |
1935 | } | |
1936 | fLogbookEntry = 0; | |
1937 | ||
1938 | delete entry; | |
1939 | delete aRow; | |
1940 | delete aResult; | |
1941 | return 0; | |
1942 | } | |
1943 | ||
b0e53b15 | 1944 | TString totEventsStr = entry->GetRunParameter("totalEvents"); |
1945 | Int_t totEvents = totEventsStr.Atoi(); | |
1946 | if (totEvents < 1) | |
1947 | { | |
eba76848 | 1948 | Log("SHUTTLE", |
b0e53b15 | 1949 | Form("QueryRunParameters - Run %d has 0 events - Skipping!", run)); |
1950 | ||
1951 | Log("SHUTTLE", Form("Marking SHUTTLE done for run %d", run)); | |
1952 | fLogbookEntry = entry; | |
97e3c167 | 1953 | if (!UpdateShuttleLogbook("shuttle_ignored")) |
b0e53b15 | 1954 | { |
1955 | AliError(Form("Could not update logbook for run %d !", run)); | |
1956 | } | |
1957 | fLogbookEntry = 0; | |
1958 | ||
eba76848 | 1959 | delete entry; |
2bb7b766 | 1960 | delete aRow; |
eba76848 | 1961 | delete aResult; |
1962 | return 0; | |
2bb7b766 | 1963 | } |
1964 | ||
eba76848 | 1965 | delete aRow; |
2bb7b766 | 1966 | delete aResult; |
eba76848 | 1967 | |
1968 | return entry; | |
2bb7b766 | 1969 | } |
1970 | ||
a038aa70 | 1971 | //______________________________________________________________________________________________ |
1972 | TMap* AliShuttle::GetValueSet(const char* host, Int_t port, const TSeqCollection* entries, | |
542b6cc8 | 1973 | DCSType type, Int_t multiSplit) |
a038aa70 | 1974 | { |
1975 | // Retrieve all "entry" data points from the DCS server | |
1976 | // host, port: TSocket connection parameters | |
1977 | // entries: list of name of the alias or data point | |
1978 | // type: kAlias or kDP | |
1979 | // returns TMap of values, 0 when failure | |
542b6cc8 | 1980 | |
1981 | AliDCSClient client(host, port, fTimeout, fRetries, multiSplit); | |
b41b252a | 1982 | |
a038aa70 | 1983 | TMap* result = 0; |
b41b252a | 1984 | if (type == kAlias) |
a038aa70 | 1985 | { |
b41b252a | 1986 | result = client.GetAliasValues(entries, GetCurrentStartTime(), |
1987 | GetCurrentEndTime()); | |
1988 | } | |
1989 | else if (type == kDP) | |
1990 | { | |
1991 | result = client.GetDPValues(entries, GetCurrentStartTime(), | |
1992 | GetCurrentEndTime()); | |
1993 | } | |
a038aa70 | 1994 | |
b41b252a | 1995 | if (result == 0) |
1996 | { | |
1997 | Log(fCurrentDetector.Data(), Form("GetValueSet - Can't get entries! Reason: %s", | |
1790d4b7 | 1998 | client.GetErrorString(client.GetResultErrorCode()))); |
1999 | if (client.GetResultErrorCode() == AliDCSClient::fgkServerError) | |
2000 | Log(fCurrentDetector.Data(), Form("GetValueSet - Server error code: %s", | |
2001 | client.GetServerError().Data())); | |
a038aa70 | 2002 | |
b41b252a | 2003 | return 0; |
a038aa70 | 2004 | } |
b41b252a | 2005 | |
a038aa70 | 2006 | return result; |
2007 | } | |
b41b252a | 2008 | |
b948db8d | 2009 | //______________________________________________________________________________________________ |
57f50b3c | 2010 | const char* AliShuttle::GetFile(Int_t system, const char* detector, |
2011 | const char* id, const char* source) | |
b948db8d | 2012 | { |
9827400b | 2013 | // Get calibration file from file exchange servers |
2014 | // First queris the FXS database for the file name, using the run, detector, id and source info | |
2015 | // then calls RetrieveFile(filename) for actual copy to local disk | |
2016 | // run: current run being processed (given by Logbook entry fLogbookEntry) | |
2017 | // detector: the Preprocessor name | |
2018 | // id: provided as a parameter by the Preprocessor | |
2019 | // source: provided by the Preprocessor through GetFileSources function | |
2020 | ||
2021 | // check if test mode should simulate a FXS error | |
2022 | if (fTestMode & kErrorFXSFiles) | |
2023 | { | |
2024 | Log(detector, Form("GetFile - In TESTMODE - Simulating error while connecting to %s FXS", GetSystemName(system))); | |
2025 | return 0; | |
2026 | } | |
2027 | ||
57f50b3c | 2028 | // check connection, in case connect |
9d733021 | 2029 | if (!Connect(system)) |
eba76848 | 2030 | { |
9d733021 | 2031 | Log(detector, Form("GetFile - Couldn't connect to %s FXS database", GetSystemName(system))); |
57f50b3c | 2032 | return 0; |
2033 | } | |
2034 | ||
2035 | // Query preparation | |
9d733021 | 2036 | TString sourceName(source); |
d386d623 | 2037 | Int_t nFields = 3; |
2038 | TString sqlQueryStart = Form("select filePath,size,fileChecksum from %s where", | |
2039 | fConfig->GetFXSdbTable(system)); | |
2040 | TString whereClause = Form("run=%d and detector=\"%s\" and fileId=\"%s\"", | |
2041 | GetCurrentRun(), detector, id); | |
2042 | ||
9d733021 | 2043 | if (system == kDAQ) |
2044 | { | |
d386d623 | 2045 | whereClause += Form(" and DAQsource=\"%s\"", source); |
57f50b3c | 2046 | } |
9d733021 | 2047 | else if (system == kDCS) |
eba76848 | 2048 | { |
9d733021 | 2049 | sourceName="none"; |
57f50b3c | 2050 | } |
9d733021 | 2051 | else if (system == kHLT) |
9e080f92 | 2052 | { |
d386d623 | 2053 | whereClause += Form(" and DDLnumbers=\"%s\"", source); |
9d733021 | 2054 | nFields = 3; |
9e080f92 | 2055 | } |
2056 | ||
9e080f92 | 2057 | TString sqlQuery = Form("%s %s", sqlQueryStart.Data(), whereClause.Data()); |
2058 | ||
2059 | AliDebug(2, Form("SQL query: \n%s",sqlQuery.Data())); | |
2060 | ||
2061 | // Query execution | |
2062 | TSQLResult* aResult = 0; | |
9d733021 | 2063 | aResult = dynamic_cast<TSQLResult*> (fServer[system]->Query(sqlQuery)); |
9e080f92 | 2064 | if (!aResult) { |
9d733021 | 2065 | Log(detector, Form("GetFileName - Can't execute SQL query to %s database for: id = %s, source = %s", |
2066 | GetSystemName(system), id, sourceName.Data())); | |
9e080f92 | 2067 | return 0; |
2068 | } | |
2069 | ||
2070 | if(aResult->GetRowCount() == 0) | |
2071 | { | |
2072 | Log(detector, | |
9d733021 | 2073 | Form("GetFileName - No entry in %s FXS db for: id = %s, source = %s", |
2074 | GetSystemName(system), id, sourceName.Data())); | |
9e080f92 | 2075 | delete aResult; |
2076 | return 0; | |
2077 | } | |
2bb7b766 | 2078 | |
9e080f92 | 2079 | if (aResult->GetRowCount() > 1) { |
2080 | Log(detector, | |
9d733021 | 2081 | Form("GetFileName - More than one entry in %s FXS db for: id = %s, source = %s", |
2082 | GetSystemName(system), id, sourceName.Data())); | |
9e080f92 | 2083 | delete aResult; |
2084 | return 0; | |
2085 | } | |
2086 | ||
9d733021 | 2087 | if (aResult->GetFieldCount() != nFields) { |
9e080f92 | 2088 | Log(detector, |
9d733021 | 2089 | Form("GetFileName - Wrong field count in %s FXS db for: id = %s, source = %s", |
2090 | GetSystemName(system), id, sourceName.Data())); | |
9e080f92 | 2091 | delete aResult; |
2092 | return 0; | |
2093 | } | |
2094 | ||
2095 | TSQLRow* aRow = dynamic_cast<TSQLRow*> (aResult->Next()); | |
2096 | ||
2097 | if (!aRow){ | |
9d733021 | 2098 | Log(detector, Form("GetFileName - Empty set result in %s FXS db from query: id = %s, source = %s", |
2099 | GetSystemName(system), id, sourceName.Data())); | |
9e080f92 | 2100 | delete aResult; |
2101 | return 0; | |
2102 | } | |
2103 | ||
2104 | TString filePath(aRow->GetField(0), aRow->GetFieldLength(0)); | |
2105 | TString fileSize(aRow->GetField(1), aRow->GetFieldLength(1)); | |
d386d623 | 2106 | TString fileChecksum(aRow->GetField(2), aRow->GetFieldLength(2)); |
9e080f92 | 2107 | |
2108 | delete aResult; | |
2109 | delete aRow; | |
2110 | ||
d386d623 | 2111 | AliDebug(2, Form("filePath = %s; size = %s, fileChecksum = %s", |
2112 | filePath.Data(), fileSize.Data(), fileChecksum.Data())); | |
9e080f92 | 2113 | |
9e080f92 | 2114 | // retrieved file is renamed to make it unique |
675f64cd | 2115 | TString localFileName = Form("%s/%s_%d_process/%s_%s_%d_%s_%s.shuttle", |
2116 | GetShuttleTempDir(), detector, GetCurrentRun(), | |
d524ade6 | 2117 | GetSystemName(system), detector, GetCurrentRun(), |
2118 | id, sourceName.Data()); | |
9d733021 | 2119 | |
9e080f92 | 2120 | |
9d733021 | 2121 | // file retrieval from FXS |
4b95672b | 2122 | UInt_t nRetries = 0; |
2123 | UInt_t maxRetries = 3; | |
2124 | Bool_t result = kFALSE; | |
2125 | ||
2126 | // copy!! if successful TSystem::Exec returns 0 | |
2127 | while(nRetries++ < maxRetries) { | |
2128 | AliDebug(2, Form("Trying to copy file. Retry # %d", nRetries)); | |
2129 | result = RetrieveFile(system, filePath.Data(), localFileName.Data()); | |
2130 | if(!result) | |
2131 | { | |
2132 | Log(detector, Form("GetFileName - Copy of file %s from %s FXS failed", | |
9d733021 | 2133 | filePath.Data(), GetSystemName(system))); |
4b95672b | 2134 | continue; |
4f0749a8 | 2135 | } |
9e080f92 | 2136 | |
c98417a1 | 2137 | if (fileSize.Length()>0) |
2138 | { | |
2139 | // compare filesize of local file with the one stored in the FXS DB | |
926cbc0e | 2140 | Long_t size = -1; |
2141 | Int_t sizeComp = gSystem->GetPathInfo(localFileName.Data(), 0, &size, 0, 0); | |
c98417a1 | 2142 | |
926cbc0e | 2143 | if (sizeComp != 0 || size != fileSize.Atoi()) |
c98417a1 | 2144 | { |
2145 | Log(detector, Form("GetFileName - size of file %s does not match with local copy!", | |
2146 | filePath.Data())); | |
2147 | result = kFALSE; | |
2148 | continue; | |
2149 | } | |
2150 | ||
2151 | } else { | |
2152 | Log(fCurrentDetector, Form("GetFile - size of file %s not set in %s database, skipping comparison", | |
2153 | filePath.Data(), GetSystemName(system))); | |
2154 | } | |
2155 | ||
d386d623 | 2156 | if (fileChecksum.Length()>0) |
4b95672b | 2157 | { |
2158 | // compare md5sum of local file with the one stored in the FXS DB | |
95d4aaef | 2159 | if(fileChecksum.Contains(' ')) fileChecksum.Resize(fileChecksum.First(' ')); |
d524ade6 | 2160 | Int_t md5Comp = gSystem->Exec(Form("md5sum %s |grep %s 2>&1 > /dev/null", |
2161 | localFileName.Data(), fileChecksum.Data())); | |
9e080f92 | 2162 | |
4b95672b | 2163 | if (md5Comp != 0) |
2164 | { | |
2165 | Log(detector, Form("GetFileName - md5sum of file %s does not match with local copy!", | |
2166 | filePath.Data())); | |
2167 | result = kFALSE; | |
2168 | continue; | |
2169 | } | |
d386d623 | 2170 | } else { |
2171 | Log(fCurrentDetector, Form("GetFile - md5sum of file %s not set in %s database, skipping comparison", | |
2172 | filePath.Data(), GetSystemName(system))); | |
9d733021 | 2173 | } |
4b95672b | 2174 | if (result) break; |
9e080f92 | 2175 | } |
2176 | ||
4b95672b | 2177 | if(!result) return 0; |
2178 | ||
9d733021 | 2179 | fFXSCalled[system]=kTRUE; |
2180 | TObjString *fileParams = new TObjString(Form("%s#!?!#%s", id, sourceName.Data())); | |
2181 | fFXSlist[system].Add(fileParams); | |
9e080f92 | 2182 | |
675f64cd | 2183 | static TString staticLocalFileName; |
2184 | staticLocalFileName.Form("%s", localFileName.Data()); | |
2185 | ||
c88ad5db | 2186 | Log(fCurrentDetector, Form("GetFile - Retrieved file with id %s and " |
2187 | "source %s from %s to %s", id, source, | |
d524ade6 | 2188 | GetSystemName(system), localFileName.Data())); |
675f64cd | 2189 | |
d524ade6 | 2190 | return staticLocalFileName.Data(); |
2bb7b766 | 2191 | } |
2192 | ||
2193 | //______________________________________________________________________________________________ | |
9d733021 | 2194 | Bool_t AliShuttle::RetrieveFile(UInt_t system, const char* fxsFileName, const char* localFileName) |
9e080f92 | 2195 | { |
9827400b | 2196 | // |
2197 | // Copies file from FXS to local Shuttle machine | |
2198 | // | |
2bb7b766 | 2199 | |
9e080f92 | 2200 | // check temp directory: trying to cd to temp; if it does not exist, create it |
d524ade6 | 2201 | AliDebug(2, Form("Copy file %s from %s FXS into %s", |
2202 | GetSystemName(system), fxsFileName, localFileName)); | |
2203 | ||
2204 | TString tmpDir(localFileName); | |
2205 | ||
2206 | tmpDir = tmpDir(0,tmpDir.Last('/')); | |
9e080f92 | 2207 | |
d524ade6 | 2208 | Int_t noDir = gSystem->GetPathInfo(tmpDir.Data(), 0, (Long64_t*) 0, 0, 0); |
2209 | if (noDir) // temp dir does not exists! | |
2210 | { | |
2211 | if (gSystem->mkdir(tmpDir.Data(), 1)) | |
2212 | { | |
2213 | Log(fCurrentDetector.Data(), "RetrieveFile - could not make temp directory!!"); | |
9e080f92 | 2214 | return kFALSE; |
2215 | } | |
9e080f92 | 2216 | } |
2217 | ||
9d733021 | 2218 | TString baseFXSFolder; |
2219 | if (system == kDAQ) | |
2220 | { | |
2221 | baseFXSFolder = "FES/"; | |
2222 | } | |
2223 | else if (system == kDCS) | |
2224 | { | |
2225 | baseFXSFolder = ""; | |
2226 | } | |
2227 | else if (system == kHLT) | |
2228 | { | |
42fde080 | 2229 | baseFXSFolder = "/opt/FXS/"; |
9d733021 | 2230 | } |
2231 | ||
2232 | ||
d524ade6 | 2233 | TString command = Form("scp -oPort=%d -2 %s@%s:%s%s %s", |
9d733021 | 2234 | fConfig->GetFXSPort(system), |
2235 | fConfig->GetFXSUser(system), | |
2236 | fConfig->GetFXSHost(system), | |
2237 | baseFXSFolder.Data(), | |
2238 | fxsFileName, | |
9e080f92 | 2239 | localFileName); |
2240 | ||
2241 | AliDebug(2, Form("%s",command.Data())); | |
2242 | ||
4b95672b | 2243 | Bool_t result = (gSystem->Exec(command.Data()) == 0); |
9e080f92 | 2244 | |
4b95672b | 2245 | return result; |
9e080f92 | 2246 | } |
2247 | ||
2248 | //______________________________________________________________________________________________ | |
9d733021 | 2249 | TList* AliShuttle::GetFileSources(Int_t system, const char* detector, const char* id) |
2250 | { | |
9827400b | 2251 | // |
2252 | // Get sources producing the condition file Id from file exchange servers | |
4a33bdd9 | 2253 | // if id is NULL all sources are returned (distinct) |
9827400b | 2254 | // |
1bcd28db | 2255 | |
8884d71e | 2256 | if (id) |
2257 | { | |
2258 | Log(detector, Form("GetFileSources - Querying %s FXS for files with id %s produced by %s", GetSystemName(system), id, detector)); | |
2259 | } else { | |
2260 | Log(detector, Form("GetFileSources - Querying %s FXS for files produced by %s", GetSystemName(system), detector)); | |
2261 | } | |
9827400b | 2262 | |
2263 | // check if test mode should simulate a FXS error | |
2264 | if (fTestMode & kErrorFXSSources) | |
2265 | { | |
2266 | Log(detector, Form("GetFileSources - In TESTMODE - Simulating error while connecting to %s FXS", GetSystemName(system))); | |
2267 | return 0; | |
2268 | } | |
2269 | ||
9d733021 | 2270 | if (system == kDCS) |
2271 | { | |
c88ad5db | 2272 | Log(detector, "GetFileSources - WARNING: DCS system has only one source of data!"); |
6297b37d | 2273 | TList *list = new TList(); |
2274 | list->SetOwner(1); | |
2275 | list->Add(new TObjString(" ")); | |
2276 | return list; | |
9d733021 | 2277 | } |
9e080f92 | 2278 | |
2279 | // check connection, in case connect | |
9d733021 | 2280 | if (!Connect(system)) |
2281 | { | |
4a33bdd9 | 2282 | Log(detector, Form("GetFileSources - Couldn't connect to %s FXS database", GetSystemName(system))); |
9d733021 | 2283 | return NULL; |
9e080f92 | 2284 | } |
2285 | ||
9d733021 | 2286 | TString sourceName = 0; |
2287 | if (system == kDAQ) | |
2288 | { | |
2289 | sourceName = "DAQsource"; | |
2290 | } else if (system == kHLT) | |
2291 | { | |
2292 | sourceName = "DDLnumbers"; | |
2293 | } | |
2294 | ||
4a33bdd9 | 2295 | TString sqlQueryStart = Form("select distinct %s from %s where", sourceName.Data(), fConfig->GetFXSdbTable(system)); |
2296 | TString whereClause = Form("run=%d and detector=\"%s\"", | |
2297 | GetCurrentRun(), detector); | |
2298 | if (id) | |
2299 | whereClause += Form(" and fileId=\"%s\"", id); | |
9e080f92 | 2300 | TString sqlQuery = Form("%s %s", sqlQueryStart.Data(), whereClause.Data()); |
2301 | ||
2302 | AliDebug(2, Form("SQL query: \n%s",sqlQuery.Data())); | |
2303 | ||
2304 | // Query execution | |
2305 | TSQLResult* aResult; | |
9d733021 | 2306 | aResult = fServer[system]->Query(sqlQuery); |
9e080f92 | 2307 | if (!aResult) { |
9d733021 | 2308 | Log(detector, Form("GetFileSources - Can't execute SQL query to %s database for id: %s", |
2309 | GetSystemName(system), id)); | |
9e080f92 | 2310 | return 0; |
2311 | } | |
2312 | ||
86aa42c3 | 2313 | TList *list = new TList(); |
2314 | list->SetOwner(1); | |
2315 | ||
9d733021 | 2316 | if (aResult->GetRowCount() == 0) |
2317 | { | |
9e080f92 | 2318 | Log(detector, |
9d733021 | 2319 | Form("GetFileSources - No entry in %s FXS table for id: %s", GetSystemName(system), id)); |
9e080f92 | 2320 | delete aResult; |
86aa42c3 | 2321 | return list; |
9e080f92 | 2322 | } |
2323 | ||
1bcd28db | 2324 | Log(detector, Form("GetFileSources - Found %d sources", aResult->GetRowCount())); |
9e080f92 | 2325 | |
1bcd28db | 2326 | TSQLRow* aRow; |
9d733021 | 2327 | while ((aRow = aResult->Next())) |
2328 | { | |
9e080f92 | 2329 | |
9d733021 | 2330 | TString source(aRow->GetField(0), aRow->GetFieldLength(0)); |
2331 | AliDebug(2, Form("%s = %s", sourceName.Data(), source.Data())); | |
2332 | list->Add(new TObjString(source)); | |
9e080f92 | 2333 | delete aRow; |
2334 | } | |
9d733021 | 2335 | |
9e080f92 | 2336 | delete aResult; |
2337 | ||
2338 | return list; | |
2bb7b766 | 2339 | } |
2340 | ||
4a33bdd9 | 2341 | //______________________________________________________________________________________________ |
2342 | TList* AliShuttle::GetFileIDs(Int_t system, const char* detector, const char* source) | |
2343 | { | |
2344 | // | |
2345 | // Get all ids of condition files produced by a given source from file exchange servers | |
2346 | // | |
2347 | ||
1bcd28db | 2348 | Log(detector, Form("GetFileIDs - Retrieving ids with source %s with %s", source, GetSystemName(system))); |
2349 | ||
4a33bdd9 | 2350 | // check if test mode should simulate a FXS error |
2351 | if (fTestMode & kErrorFXSSources) | |
2352 | { | |
2353 | Log(detector, Form("GetFileIDs - In TESTMODE - Simulating error while connecting to %s FXS", GetSystemName(system))); | |
2354 | return 0; | |
2355 | } | |
2356 | ||
2357 | // check connection, in case connect | |
2358 | if (!Connect(system)) | |
2359 | { | |
2360 | Log(detector, Form("GetFileIDs - Couldn't connect to %s FXS database", GetSystemName(system))); | |
2361 | return NULL; | |
2362 | } | |
2363 | ||
2364 | TString sourceName = 0; | |
2365 | if (system == kDAQ) | |
2366 | { | |
2367 | sourceName = "DAQsource"; | |
2368 | } else if (system == kHLT) | |
2369 | { | |
2370 | sourceName = "DDLnumbers"; | |
2371 | } | |
2372 | ||
2373 | TString sqlQueryStart = Form("select fileId from %s where", fConfig->GetFXSdbTable(system)); | |
2374 | TString whereClause = Form("run=%d and detector=\"%s\"", | |
2375 | GetCurrentRun(), detector); | |
2376 | if (sourceName.Length() > 0 && source) | |
2377 | whereClause += Form(" and %s=\"%s\"", sourceName.Data(), source); | |
2378 | TString sqlQuery = Form("%s %s", sqlQueryStart.Data(), whereClause.Data()); | |
2379 | ||
2380 | AliDebug(2, Form("SQL query: \n%s",sqlQuery.Data())); | |
2381 | ||
2382 | // Query execution | |
2383 | TSQLResult* aResult; | |
2384 | aResult = fServer[system]->Query(sqlQuery); | |
2385 | if (!aResult) { | |
2386 | Log(detector, Form("GetFileIDs - Can't execute SQL query to %s database for source: %s", | |
2387 | GetSystemName(system), source)); | |
2388 | return 0; | |
2389 | } | |
2390 | ||
2391 | TList *list = new TList(); | |
2392 | list->SetOwner(1); | |
2393 | ||
2394 | if (aResult->GetRowCount() == 0) | |
2395 | { | |
2396 | Log(detector, | |
2397 | Form("GetFileIDs - No entry in %s FXS table for source: %s", GetSystemName(system), source)); | |
2398 | delete aResult; | |
2399 | return list; | |
2400 | } | |
2401 | ||
1bcd28db | 2402 | Log(detector, Form("GetFileIDs - Found %d ids", aResult->GetRowCount())); |
2403 | ||
4a33bdd9 | 2404 | TSQLRow* aRow; |
2405 | ||
2406 | while ((aRow = aResult->Next())) | |
2407 | { | |
2408 | ||
2409 | TString id(aRow->GetField(0), aRow->GetFieldLength(0)); | |
2410 | AliDebug(2, Form("fileId = %s", id.Data())); | |
2411 | list->Add(new TObjString(id)); | |
2412 | delete aRow; | |
2413 | } | |
2414 | ||
2415 | delete aResult; | |
2416 | ||
2417 | return list; | |
2418 | } | |
2419 | ||
2bb7b766 | 2420 | //______________________________________________________________________________________________ |
9d733021 | 2421 | Bool_t AliShuttle::Connect(Int_t system) |
2bb7b766 | 2422 | { |
9827400b | 2423 | // Connect to MySQL Server of the system's FXS MySQL databases |
2424 | // DAQ Logbook, Shuttle Logbook and DAQ FXS db are on the same host | |
2425 | // | |
57f50b3c | 2426 | |
9d733021 | 2427 | // check connection: if already connected return |
2428 | if(fServer[system] && fServer[system]->IsConnected()) return kTRUE; | |
57f50b3c | 2429 | |
9d733021 | 2430 | TString dbHost, dbUser, dbPass, dbName; |
57f50b3c | 2431 | |
9d733021 | 2432 | if (system < 3) // FXS db servers |
2433 | { | |
2434 | dbHost = Form("mysql://%s:%d", fConfig->GetFXSdbHost(system), fConfig->GetFXSdbPort(system)); | |
2435 | dbUser = fConfig->GetFXSdbUser(system); | |
2436 | dbPass = fConfig->GetFXSdbPass(system); | |
2437 | dbName = fConfig->GetFXSdbName(system); | |
2438 | } else { // Run & Shuttle logbook servers | |
2439 | // TODO Will the Shuttle logbook server be the same as the Run logbook server ??? | |
2440 | dbHost = Form("mysql://%s:%d", fConfig->GetDAQlbHost(), fConfig->GetDAQlbPort()); | |
2441 | dbUser = fConfig->GetDAQlbUser(); | |
2442 | dbPass = fConfig->GetDAQlbPass(); | |
2443 | dbName = fConfig->GetDAQlbDB(); | |
2444 | } | |
57f50b3c | 2445 | |
9d733021 | 2446 | fServer[system] = TSQLServer::Connect(dbHost.Data(), dbUser.Data(), dbPass.Data()); |
2447 | if (!fServer[system] || !fServer[system]->IsConnected()) { | |
2448 | if(system < 3) | |
2449 | { | |
2450 | AliError(Form("Can't establish connection to FXS database for %s", | |
2451 | AliShuttleInterface::GetSystemName(system))); | |
2452 | } else { | |
2453 | AliError("Can't establish connection to Run logbook."); | |
57f50b3c | 2454 | } |
9d733021 | 2455 | if(fServer[system]) delete fServer[system]; |
2456 | return kFALSE; | |
2bb7b766 | 2457 | } |
57f50b3c | 2458 | |
9d733021 | 2459 | // Get tables |
2460 | TSQLResult* aResult=0; | |
2461 | switch(system){ | |
2462 | case kDAQ: | |
2463 | aResult = fServer[kDAQ]->GetTables(dbName.Data()); | |
2464 | break; | |
2465 | case kDCS: | |
2466 | aResult = fServer[kDCS]->GetTables(dbName.Data()); | |
2467 | break; | |
2468 | case kHLT: | |
2469 | aResult = fServer[kHLT]->GetTables(dbName.Data()); | |
2470 | break; | |
2471 | default: | |
2472 | aResult = fServer[3]->GetTables(dbName.Data()); | |
2473 | break; | |
2474 | } | |
2475 | ||
2476 | delete aResult; | |
2bb7b766 | 2477 | return kTRUE; |
2478 | } | |
57f50b3c | 2479 | |
9e080f92 | 2480 | //______________________________________________________________________________________________ |
9d733021 | 2481 | Bool_t AliShuttle::UpdateTable() |
9e080f92 | 2482 | { |
9827400b | 2483 | // |
2484 | // Update FXS table filling time_processed field in all rows corresponding to current run and detector | |
2485 | // | |
9e080f92 | 2486 | |
9d733021 | 2487 | Bool_t result = kTRUE; |
9e080f92 | 2488 | |
9d733021 | 2489 | for (UInt_t system=0; system<3; system++) |
2490 | { | |
2491 | if(!fFXSCalled[system]) continue; | |
9e080f92 | 2492 | |
9d733021 | 2493 | // check connection, in case connect |
2494 | if (!Connect(system)) | |
2495 | { | |
2496 | Log(fCurrentDetector, Form("UpdateTable - Couldn't connect to %s FXS database", GetSystemName(system))); | |
2497 | result = kFALSE; | |
2498 | continue; | |
9e080f92 | 2499 | } |
9e080f92 | 2500 | |
9d733021 | 2501 | TTimeStamp now; // now |
2502 | ||
2503 | // Loop on FXS list entries | |
2504 | TIter iter(&fFXSlist[system]); | |
2505 | TObjString *aFXSentry=0; | |
2506 | while ((aFXSentry = dynamic_cast<TObjString*> (iter.Next()))) | |
2507 | { | |
2508 | TString aFXSentrystr = aFXSentry->String(); | |
2509 | TObjArray *aFXSarray = aFXSentrystr.Tokenize("#!?!#"); | |
2510 | if (!aFXSarray || aFXSarray->GetEntries() != 2 ) | |
2511 | { | |
2512 | Log(fCurrentDetector, Form("UpdateTable - error updating %s FXS entry. Check string: <%s>", | |
2513 | GetSystemName(system), aFXSentrystr.Data())); | |
2514 | if(aFXSarray) delete aFXSarray; | |
2515 | result = kFALSE; | |
2516 | continue; | |
2517 | } | |
2518 | const char* fileId = ((TObjString*) aFXSarray->At(0))->GetName(); | |
2519 | const char* source = ((TObjString*) aFXSarray->At(1))->GetName(); | |
2520 | ||
2521 | TString whereClause; | |
2522 | if (system == kDAQ) | |
2523 | { | |
2524 | whereClause = Form("where run=%d and detector=\"%s\" and fileId=\"%s\" and DAQsource=\"%s\";", | |
2525 | GetCurrentRun(), fCurrentDetector.Data(), fileId, source); | |
2526 | } | |
2527 | else if (system == kDCS) | |
2528 | { | |
2529 | whereClause = Form("where run=%d and detector=\"%s\" and fileId=\"%s\";", | |
2530 | GetCurrentRun(), fCurrentDetector.Data(), fileId); | |
2531 | } | |
2532 | else if (system == kHLT) | |
2533 | { | |
2534 | whereClause = Form("where run=%d and detector=\"%s\" and fileId=\"%s\" and DDLnumbers=\"%s\";", | |
2535 | GetCurrentRun(), fCurrentDetector.Data(), fileId, source); | |
2536 | } | |
2537 | ||
2538 | delete aFXSarray; | |
9e080f92 | 2539 | |
9d733021 | 2540 | TString sqlQuery = Form("update %s set time_processed=%d %s", fConfig->GetFXSdbTable(system), |
2541 | now.GetSec(), whereClause.Data()); | |
9e080f92 | 2542 | |
9d733021 | 2543 | AliDebug(2, Form("SQL query: \n%s",sqlQuery.Data())); |
9e080f92 | 2544 | |
9d733021 | 2545 | // Query execution |
2546 | TSQLResult* aResult; | |
2547 | aResult = dynamic_cast<TSQLResult*> (fServer[system]->Query(sqlQuery)); | |
2548 | if (!aResult) | |
2549 | { | |
2550 | Log(fCurrentDetector, Form("UpdateTable - %s db: can't execute SQL query <%s>", | |
2551 | GetSystemName(system), sqlQuery.Data())); | |
2552 | result = kFALSE; | |
2553 | continue; | |
2554 | } | |
2555 | delete aResult; | |
9e080f92 | 2556 | } |
9e080f92 | 2557 | } |
2558 | ||
9d733021 | 2559 | return result; |
9e080f92 | 2560 | } |
57f50b3c | 2561 | |
3301427a | 2562 | //______________________________________________________________________________________________ |
2563 | Bool_t AliShuttle::UpdateTableFailCase() | |
2564 | { | |
9827400b | 2565 | // Update FXS table filling time_processed field in all rows corresponding to current run and detector |
2566 | // this is called in case the preprocessor is declared failed for the current run, because | |
2567 | // the fields are updated only in case of success | |
3301427a | 2568 | |
2569 | Bool_t result = kTRUE; | |
2570 | ||
2571 | for (UInt_t system=0; system<3; system++) | |
2572 | { | |
2573 | // check connection, in case connect | |
2574 | if (!Connect(system)) | |
2575 | { | |
2576 | Log(fCurrentDetector, Form("UpdateTableFailCase - Couldn't connect to %s FXS database", | |
2577 | GetSystemName(system))); | |
2578 | result = kFALSE; | |
2579 | continue; | |
2580 | } | |
2581 | ||
2582 | TTimeStamp now; // now | |
2583 | ||
2584 | // Loop on FXS list entries | |
2585 | ||
2586 | TString whereClause = Form("where run=%d and detector=\"%s\";", | |
2587 | GetCurrentRun(), fCurrentDetector.Data()); | |
2588 | ||
2589 | ||
2590 | TString sqlQuery = Form("update %s set time_processed=%d %s", fConfig->GetFXSdbTable(system), | |
2591 | now.GetSec(), whereClause.Data()); | |
2592 | ||
2593 | AliDebug(2, Form("SQL query: \n%s",sqlQuery.Data())); | |
2594 | ||
2595 | // Query execution | |
2596 | TSQLResult* aResult; | |
2597 | aResult = dynamic_cast<TSQLResult*> (fServer[system]->Query(sqlQuery)); | |
2598 | if (!aResult) | |
2599 | { | |
2600 | Log(fCurrentDetector, Form("UpdateTableFailCase - %s db: can't execute SQL query <%s>", | |
2601 | GetSystemName(system), sqlQuery.Data())); | |
2602 | result = kFALSE; | |
2603 | continue; | |
2604 | } | |
2605 | delete aResult; | |
2606 | } | |
2607 | ||
2608 | return result; | |
2609 | } | |
2610 | ||
2bb7b766 | 2611 | //______________________________________________________________________________________________ |
2612 | Bool_t AliShuttle::UpdateShuttleLogbook(const char* detector, const char* status) | |
2613 | { | |
e7f62f16 | 2614 | // |
2615 | // Update Shuttle logbook filling detector or shuttle_done column | |
2616 | // ex. of usage: UpdateShuttleLogbook("PHOS", "DONE") or UpdateShuttleLogbook("shuttle_done") | |
2617 | // | |
57f50b3c | 2618 | |
2bb7b766 | 2619 | // check connection, in case connect |
be48e3ea | 2620 | if(!Connect(3)){ |
2bb7b766 | 2621 | Log("SHUTTLE", "UpdateShuttleLogbook - Couldn't connect to DAQ Logbook."); |
2622 | return kFALSE; | |
57f50b3c | 2623 | } |
2624 | ||
2bb7b766 | 2625 | TString detName(detector); |
2626 | TString setClause; | |
ee6f7523 | 2627 | if (detName == "shuttle_done" || detName == "shuttle_ignored") |
e7f62f16 | 2628 | { |
2bb7b766 | 2629 | setClause = "set shuttle_done=1"; |
e7f62f16 | 2630 | |
ee6f7523 | 2631 | if (detName == "shuttle_done") |
b0e53b15 | 2632 | { |
2633 | // Send the information to ML | |
2634 | TMonaLisaText mlStatus("SHUTTLE_status", "Done"); | |
e7f62f16 | 2635 | |
b0e53b15 | 2636 | TList mlList; |
2637 | mlList.Add(&mlStatus); | |
2638 | ||
ee6f7523 | 2639 | TString mlID; |
2640 | mlID.Form("%d", GetCurrentRun()); | |
2641 | fMonaLisa->SendParameters(&mlList, mlID); | |
b0e53b15 | 2642 | } |
2bb7b766 | 2643 | } else { |
2bb7b766 | 2644 | TString statusStr(status); |
2645 | if(statusStr.Contains("done", TString::kIgnoreCase) || | |
2646 | statusStr.Contains("failed", TString::kIgnoreCase)){ | |
eba76848 | 2647 | setClause = Form("set %s=\"%s\"", detector, status); |
2bb7b766 | 2648 | } else { |
2649 | Log("SHUTTLE", | |
2650 | Form("UpdateShuttleLogbook - Invalid status <%s> for detector %s", | |
2651 | status, detector)); | |
2652 | return kFALSE; | |
2653 | } | |
2654 | } | |
57f50b3c | 2655 | |
2bb7b766 | 2656 | TString whereClause = Form("where run=%d", GetCurrentRun()); |
2657 | ||
441b0e9c | 2658 | TString sqlQuery = Form("update %s %s %s", |
2659 | fConfig->GetShuttlelbTable(), setClause.Data(), whereClause.Data()); | |
57f50b3c | 2660 | |
2bb7b766 | 2661 | AliDebug(2, Form("SQL query: \n%s",sqlQuery.Data())); |
2662 | ||
2663 | // Query execution | |
2664 | TSQLResult* aResult; | |
be48e3ea | 2665 | aResult = dynamic_cast<TSQLResult*> (fServer[3]->Query(sqlQuery)); |
2bb7b766 | 2666 | if (!aResult) { |
2667 | Log("SHUTTLE", Form("UpdateShuttleLogbook - Can't execute query <%s>", sqlQuery.Data())); | |
2668 | return kFALSE; | |
57f50b3c | 2669 | } |
2bb7b766 | 2670 | delete aResult; |
57f50b3c | 2671 | |
2672 | return kTRUE; | |
2673 | } | |
2674 | ||
2675 | //______________________________________________________________________________________________ | |
2bb7b766 | 2676 | Int_t AliShuttle::GetCurrentRun() const |
2677 | { | |
9827400b | 2678 | // |
2679 | // Get current run from logbook entry | |
2680 | // | |
57f50b3c | 2681 | |
2bb7b766 | 2682 | return fLogbookEntry ? fLogbookEntry->GetRun() : -1; |
57f50b3c | 2683 | } |
2684 | ||
2685 | //______________________________________________________________________________________________ | |
2bb7b766 | 2686 | UInt_t AliShuttle::GetCurrentStartTime() const |
2687 | { | |
9827400b | 2688 | // |
2689 | // get current start time | |
2690 | // | |
57f50b3c | 2691 | |
2bb7b766 | 2692 | return fLogbookEntry ? fLogbookEntry->GetStartTime() : 0; |
57f50b3c | 2693 | } |
2694 | ||
2695 | //______________________________________________________________________________________________ | |
2bb7b766 | 2696 | UInt_t AliShuttle::GetCurrentEndTime() const |
2697 | { | |
9827400b | 2698 | // |
2699 | // get current end time from logbook entry | |
2700 | // | |
57f50b3c | 2701 | |
2bb7b766 | 2702 | return fLogbookEntry ? fLogbookEntry->GetEndTime() : 0; |
57f50b3c | 2703 | } |
2704 | ||
675f64cd | 2705 | //______________________________________________________________________________________________ |
2706 | UInt_t AliShuttle::GetCurrentYear() const | |
2707 | { | |
2708 | // | |
2709 | // Get current year from logbook entry | |
2710 | // | |
2711 | ||
2712 | if (!fLogbookEntry) return 0; | |
2713 | ||
2714 | TTimeStamp startTime(GetCurrentStartTime()); | |
2715 | TString year = Form("%d",startTime.GetDate()); | |
2716 | year = year(0,4); | |
2717 | ||
2718 | return year.Atoi(); | |
2719 | } | |
2720 | ||
2721 | //______________________________________________________________________________________________ | |
2722 | const char* AliShuttle::GetLHCPeriod() const | |
2723 | { | |
2724 | // | |
2725 | // Get current LHC period from logbook entry | |
2726 | // | |
2727 | ||
2728 | if (!fLogbookEntry) return 0; | |
2729 | ||
2730 | return fLogbookEntry->GetRunParameter("LHCperiod"); | |
2731 | } | |
2732 | ||
b948db8d | 2733 | //______________________________________________________________________________________________ |
2734 | void AliShuttle::Log(const char* detector, const char* message) | |
2735 | { | |
9827400b | 2736 | // |
2737 | // Fill log string with a message | |
2738 | // | |
b948db8d | 2739 | |
7d4cf768 | 2740 | TString logRunDir = GetShuttleLogDir(); |
2741 | if (GetCurrentRun() >=0) | |
2742 | logRunDir += Form("/%d", GetCurrentRun()); | |
2743 | ||
2744 | void* dir = gSystem->OpenDirectory(logRunDir.Data()); | |
84090f85 | 2745 | if (dir == NULL) { |
7d4cf768 | 2746 | if (gSystem->mkdir(logRunDir.Data(), kTRUE)) { |
36c99a6a | 2747 | AliError(Form("Can't open directory <%s>", GetShuttleLogDir())); |
84090f85 | 2748 | return; |
2749 | } | |
b948db8d | 2750 | |
84090f85 | 2751 | } else { |
2752 | gSystem->FreeDirectory(dir); | |
2753 | } | |
b948db8d | 2754 | |
cb343cfd | 2755 | TString toLog = Form("%s (%d): %s - ", TTimeStamp(time(0)).AsString("s"), getpid(), detector); |
e7f62f16 | 2756 | if (GetCurrentRun() >= 0) |
2757 | toLog += Form("run %d - ", GetCurrentRun()); | |
2bb7b766 | 2758 | toLog += Form("%s", message); |
2759 | ||
84090f85 | 2760 | AliInfo(toLog.Data()); |
ffa29e93 | 2761 | |
2762 | // if we redirect the log output already to the file, leave here | |
2763 | if (fOutputRedirected && strcmp(detector, "SHUTTLE") != 0) | |
2764 | return; | |
b948db8d | 2765 | |
ffa29e93 | 2766 | TString fileName = GetLogFileName(detector); |
e7f62f16 | 2767 | |
84090f85 | 2768 | gSystem->ExpandPathName(fileName); |
2769 | ||
2770 | ofstream logFile; | |
2771 | logFile.open(fileName, ofstream::out | ofstream::app); | |
2772 | ||
2773 | if (!logFile.is_open()) { | |
2774 | AliError(Form("Could not open file %s", fileName.Data())); | |
2775 | return; | |
2776 | } | |
7bfb2090 | 2777 | |
84090f85 | 2778 | logFile << toLog.Data() << "\n"; |
b948db8d | 2779 | |
84090f85 | 2780 | logFile.close(); |
b948db8d | 2781 | } |
2bb7b766 | 2782 | |
ffa29e93 | 2783 | //______________________________________________________________________________________________ |
2784 | TString AliShuttle::GetLogFileName(const char* detector) const | |
2785 | { | |
2786 | // | |
2787 | // returns the name of the log file for a given sub detector | |
2788 | // | |
2789 | ||
2790 | TString fileName; | |
2791 | ||
2792 | if (GetCurrentRun() >= 0) | |
7d4cf768 | 2793 | { |
2794 | fileName.Form("%s/%d/%s_%d.log", GetShuttleLogDir(), GetCurrentRun(), | |
2795 | detector, GetCurrentRun()); | |
2796 | } else { | |
ffa29e93 | 2797 | fileName.Form("%s/%s.log", GetShuttleLogDir(), detector); |
7d4cf768 | 2798 | } |
ffa29e93 | 2799 | |
2800 | return fileName; | |
2801 | } | |
2802 | ||
ee6f7523 | 2803 | //______________________________________________________________________________________________ |
2804 | void AliShuttle::SendAlive() | |
2805 | { | |
2806 | // sends alive message to ML | |
2807 | ||
2808 | TMonaLisaText mlStatus("SHUTTLE_status", "Alive"); | |
2809 | ||
2810 | TList mlList; | |
2811 | mlList.Add(&mlStatus); | |
2812 | ||
2813 | fMonaLisa->SendParameters(&mlList, "__PROCESSINGINFO__"); | |
2814 | } | |
2815 | ||
2bb7b766 | 2816 | //______________________________________________________________________________________________ |
2817 | Bool_t AliShuttle::Collect(Int_t run) | |
2818 | { | |
9827400b | 2819 | // |
2820 | // Collects conditions data for all UNPROCESSED run written to DAQ LogBook in case of run = -1 (default) | |
2821 | // If a dedicated run is given this run is processed | |
2822 | // | |
2823 | // In operational mode, this is the Shuttle function triggered by the EOR signal. | |
2824 | // | |
2bb7b766 | 2825 | |
eba76848 | 2826 | if (run == -1) |
2827 | Log("SHUTTLE","Collect - Shuttle called. Collecting conditions data for unprocessed runs"); | |
2828 | else | |
2829 | Log("SHUTTLE", Form("Collect - Shuttle called. Collecting conditions data for run %d", run)); | |
cb343cfd | 2830 | |
2831 | SetLastAction("Starting"); | |
2bb7b766 | 2832 | |
ee6f7523 | 2833 | // create ML instance |
2834 | if (!fMonaLisa) | |
2835 | fMonaLisa = new TMonaLisaWriter(fConfig->GetMonitorHost(), fConfig->GetMonitorTable()); | |
2836 | ||
ee6f7523 | 2837 | SendAlive(); |
fbc112e3 | 2838 | CountOpenRuns(); |
ee6f7523 | 2839 | |
2bb7b766 | 2840 | TString whereClause("where shuttle_done=0"); |
eba76848 | 2841 | if (run != -1) |
2842 | whereClause += Form(" and run=%d", run); | |
2bb7b766 | 2843 | |
2844 | TObjArray shuttleLogbookEntries; | |
be48e3ea | 2845 | if (!QueryShuttleLogbook(whereClause, shuttleLogbookEntries)) |
2846 | { | |
cb343cfd | 2847 | Log("SHUTTLE", "Collect - Can't retrieve entries from Shuttle logbook"); |
2bb7b766 | 2848 | return kFALSE; |
2849 | } | |
2850 | ||
9e080f92 | 2851 | if (shuttleLogbookEntries.GetEntries() == 0) |
2852 | { | |
2853 | if (run == -1) | |
2854 | Log("SHUTTLE","Collect - Found no UNPROCESSED runs in Shuttle logbook"); | |
2855 | else | |
2856 | Log("SHUTTLE", Form("Collect - Run %d is already DONE " | |
2857 | "or it does not exist in Shuttle logbook", run)); | |
2858 | return kTRUE; | |
2859 | } | |
2860 | ||
be48e3ea | 2861 | for (UInt_t iDet=0; iDet<NDetectors(); iDet++) |
2862 | fFirstUnprocessed[iDet] = kTRUE; | |
2863 | ||
fc5a4708 | 2864 | if (run != -1) |
be48e3ea | 2865 | { |
2866 | // query Shuttle logbook for earlier runs, check if some detectors are unprocessed, | |
2867 | // flag them into fFirstUnprocessed array | |
2868 | TString whereClause(Form("where shuttle_done=0 and run < %d", run)); | |
2869 | TObjArray tmpLogbookEntries; | |
2870 | if (!QueryShuttleLogbook(whereClause, tmpLogbookEntries)) | |
2871 | { | |
2872 | Log("SHUTTLE", "Collect - Can't retrieve entries from Shuttle logbook"); | |
2873 | return kFALSE; | |
2874 | } | |
2875 | ||
2876 | TIter iter(&tmpLogbookEntries); | |
2877 | AliShuttleLogbookEntry* anEntry = 0; | |
2878 | while ((anEntry = dynamic_cast<AliShuttleLogbookEntry*> (iter.Next()))) | |
2879 | { | |
2880 | for (UInt_t iDet=0; iDet<NDetectors(); iDet++) | |
2881 | { | |
2882 | if (anEntry->GetDetectorStatus(iDet) == AliShuttleLogbookEntry::kUnprocessed) | |
2883 | { | |
2884 | AliDebug(2, Form("Run %d: setting %s as \"not first time unprocessed\"", | |
2885 | anEntry->GetRun(), GetDetName(iDet))); | |
2886 | fFirstUnprocessed[iDet] = kFALSE; | |
2887 | } | |
2888 | } | |
2889 | ||
2890 | } | |
2891 | ||
2892 | } | |
2893 | ||
2894 | if (!RetrieveConditionsData(shuttleLogbookEntries)) | |
2895 | { | |
cb343cfd | 2896 | Log("SHUTTLE", "Collect - Process of at least one run failed"); |
a3ca69fe | 2897 | CountOpenRuns(); |
2bb7b766 | 2898 | return kFALSE; |
2899 | } | |
2900 | ||
36c99a6a | 2901 | Log("SHUTTLE", "Collect - Requested run(s) successfully processed"); |
a3ca69fe | 2902 | CountOpenRuns(); |
eba76848 | 2903 | return kTRUE; |
2bb7b766 | 2904 | } |
2905 | ||
2bb7b766 | 2906 | //______________________________________________________________________________________________ |
2907 | Bool_t AliShuttle::RetrieveConditionsData(const TObjArray& dateEntries) | |
2908 | { | |
9827400b | 2909 | // |
2910 | // Retrieve conditions data for all runs that aren't processed yet | |
2911 | // | |
2bb7b766 | 2912 | |
2913 | Bool_t hasError = kFALSE; | |
2914 | ||
2915 | TIter iter(&dateEntries); | |
2916 | AliShuttleLogbookEntry* anEntry; | |
2917 | ||
2918 | while ((anEntry = (AliShuttleLogbookEntry*) iter.Next())){ | |
2919 | if (!Process(anEntry)){ | |
2920 | hasError = kTRUE; | |
2921 | } | |
4b95672b | 2922 | |
2923 | // clean SHUTTLE temp directory | |
d524ade6 | 2924 | //TString filename = Form("%s/*.shuttle", GetShuttleTempDir()); |
2925 | //RemoveFile(filename.Data()); | |
2bb7b766 | 2926 | } |
2927 | ||
2928 | return hasError == kFALSE; | |
2929 | } | |
cb343cfd | 2930 | |
2931 | //______________________________________________________________________________________________ | |
2932 | ULong_t AliShuttle::GetTimeOfLastAction() const | |
2933 | { | |
9827400b | 2934 | // |
2935 | // Gets time of last action | |
2936 | // | |
2937 | ||
cb343cfd | 2938 | ULong_t tmp; |
36c99a6a | 2939 | |
cb343cfd | 2940 | fMonitoringMutex->Lock(); |
be48e3ea | 2941 | |
cb343cfd | 2942 | tmp = fLastActionTime; |
36c99a6a | 2943 | |
cb343cfd | 2944 | fMonitoringMutex->UnLock(); |
36c99a6a | 2945 | |
cb343cfd | 2946 | return tmp; |
2947 | } | |
2948 | ||
2949 | //______________________________________________________________________________________________ | |
2950 | const TString AliShuttle::GetLastAction() const | |
2951 | { | |
9827400b | 2952 | // |
cb343cfd | 2953 | // returns a string description of the last action |
9827400b | 2954 | // |
cb343cfd | 2955 | |
2956 | TString tmp; | |
36c99a6a | 2957 | |
cb343cfd | 2958 | fMonitoringMutex->Lock(); |
2959 | ||
2960 | tmp = fLastAction; | |
2961 | ||
2962 | fMonitoringMutex->UnLock(); | |
2963 | ||
36c99a6a | 2964 | return tmp; |
cb343cfd | 2965 | } |
2966 | ||
2967 | //______________________________________________________________________________________________ | |
2968 | void AliShuttle::SetLastAction(const char* action) | |
2969 | { | |
9827400b | 2970 | // |
cb343cfd | 2971 | // updates the monitoring variables |
9827400b | 2972 | // |
36c99a6a | 2973 | |
cb343cfd | 2974 | fMonitoringMutex->Lock(); |
36c99a6a | 2975 | |
cb343cfd | 2976 | fLastAction = action; |
2977 | fLastActionTime = time(0); | |
2978 | ||
2979 | fMonitoringMutex->UnLock(); | |
2980 | } | |
eba76848 | 2981 | |
2982 | //______________________________________________________________________________________________ | |
2983 | const char* AliShuttle::GetRunParameter(const char* param) | |
2984 | { | |
9827400b | 2985 | // |
2986 | // returns run parameter read from DAQ logbook | |
2987 | // | |
eba76848 | 2988 | |
2989 | if(!fLogbookEntry) { | |
2990 | AliError("No logbook entry!"); | |
2991 | return 0; | |
2992 | } | |
2993 | ||
2994 | return fLogbookEntry->GetRunParameter(param); | |
2995 | } | |
57c1a579 | 2996 | |
d386d623 | 2997 | //______________________________________________________________________________________________ |
9827400b | 2998 | AliCDBEntry* AliShuttle::GetFromOCDB(const char* detector, const AliCDBPath& path) |
d386d623 | 2999 | { |
9827400b | 3000 | // |
3001 | // returns object from OCDB valid for current run | |
3002 | // | |
d386d623 | 3003 | |
9827400b | 3004 | if (fTestMode & kErrorOCDB) |
3005 | { | |
3006 | Log(detector, "GetFromOCDB - In TESTMODE - Simulating error with OCDB"); | |
3007 | return 0; | |
3008 | } | |
3009 | ||
d386d623 | 3010 | AliCDBStorage *sto = AliCDBManager::Instance()->GetStorage(fgkMainCDB); |
3011 | if (!sto) | |
3012 | { | |
9827400b | 3013 | Log(detector, "GetFromOCDB - Cannot activate main OCDB for query!"); |
d386d623 | 3014 | return 0; |
3015 | } | |
3016 | ||
3017 | return dynamic_cast<AliCDBEntry*> (sto->Get(path, GetCurrentRun())); | |
3018 | } | |
3019 | ||
57c1a579 | 3020 | //______________________________________________________________________________________________ |
3021 | Bool_t AliShuttle::SendMail() | |
3022 | { | |
9827400b | 3023 | // |
3024 | // sends a mail to the subdetector expert in case of preprocessor error | |
3025 | // | |
3026 | ||
3027 | if (fTestMode != kNone) | |
3028 | return kTRUE; | |
fb40ead4 | 3029 | |
926cbc0e | 3030 | if (!fConfig->SendMail()) |
3031 | return kTRUE; | |
57c1a579 | 3032 | |
57c1a579 | 3033 | TString to=""; |
3034 | TIter iterExperts(fConfig->GetResponsibles(fCurrentDetector)); | |
3035 | TObjString *anExpert=0; | |
3036 | while ((anExpert = (TObjString*) iterExperts.Next())) | |
3037 | { | |
3038 | to += Form("%s,", anExpert->GetName()); | |
3039 | } | |
4508ab8b | 3040 | if (to.Length() > 0) |
3041 | to.Remove(to.Length()-1); | |
909732f7 | 3042 | AliDebug(2, Form("to: %s",to.Data())); |
57c1a579 | 3043 | |
86aa42c3 | 3044 | if (to.IsNull()) { |
fb40ead4 | 3045 | Log("SHUTTLE", "List of detector responsibles not set!"); |
36c99a6a | 3046 | return kFALSE; |
3047 | } | |
3048 | ||
fbc112e3 | 3049 | void* dir = gSystem->OpenDirectory(GetShuttleLogDir()); |
3050 | if (dir == NULL) | |
3051 | { | |
3052 | if (gSystem->mkdir(GetShuttleLogDir(), kTRUE)) | |
3053 | { | |
3054 | Log("SHUTTLE", Form("SendMail - Can't open directory <%s>", GetShuttleLogDir())); | |
3055 | return kFALSE; | |
3056 | } | |
3057 | ||
3058 | } else { | |
3059 | gSystem->FreeDirectory(dir); | |
3060 | } | |
3061 | ||
4508ab8b | 3062 | TString bodyFileName; |
3063 | bodyFileName.Form("%s/mail.body", GetShuttleLogDir()); | |
3064 | gSystem->ExpandPathName(bodyFileName); | |
3065 | ||
3066 | ofstream mailBody; | |
3067 | mailBody.open(bodyFileName, ofstream::out); | |
3068 | ||
3069 | if (!mailBody.is_open()) | |
3070 | { | |
3071 | Log("SHUTTLE", Form("Could not open mail body file %s", bodyFileName.Data())); | |
3072 | return kFALSE; | |
3073 | } | |
3074 | ||
fb40ead4 | 3075 | TString cc=""; |
3076 | TIter iterAdmins(fConfig->GetAdmins(AliShuttleConfig::kGlobal)); | |
3077 | TObjString *anAdmin=0; | |
3078 | while ((anAdmin = (TObjString*) iterAdmins.Next())) | |
3079 | { | |
3080 | cc += Form("%s,", anAdmin->GetName()); | |
3081 | } | |
3082 | if (cc.Length() > 0) | |
c19963db | 3083 | cc.Remove(cc.Length()-1); |
fb40ead4 | 3084 | AliDebug(2, Form("cc: %s",to.Data())); |
57c1a579 | 3085 | |
6a1146c4 | 3086 | TString subject = Form("%s Shuttle preprocessor FAILED in run %d (run type = %s)!", |
3087 | fCurrentDetector.Data(), GetCurrentRun(), GetRunType()); | |
909732f7 | 3088 | AliDebug(2, Form("subject: %s", subject.Data())); |
57c1a579 | 3089 | |
3090 | TString body = Form("Dear %s expert(s), \n\n", fCurrentDetector.Data()); | |
3091 | body += Form("SHUTTLE just detected that your preprocessor " | |
6a1146c4 | 3092 | "failed processing run %d (run type = %s)!!\n\n", |
3093 | GetCurrentRun(), GetRunType()); | |
7d4cf768 | 3094 | body += Form("Please check %s status on the SHUTTLE monitoring page: \n\n", |
3095 | fCurrentDetector.Data()); | |
b0e53b15 | 3096 | if (fConfig->GetRunMode() == AliShuttleConfig::kTest) |
3097 | { | |
3098 | body += Form("\thttp://pcalimonitor.cern.ch:8889/shuttle.jsp?time=168 \n\n"); | |
3099 | } else { | |
4508ab8b | 3100 | body += Form("\thttp://pcalimonitor.cern.ch/shuttle.jsp?instance=PROD&time=168 \n\n"); |
b0e53b15 | 3101 | } |
3102 | ||
7d4cf768 | 3103 | |
3104 | TString logFolder = "logs"; | |
3105 | if (fConfig->GetRunMode() == AliShuttleConfig::kProd) | |
3106 | logFolder += "_PROD"; | |
3107 | ||
3108 | ||
546242fb | 3109 | body += Form("Find the %s log for the current run on \n\n" |
7d4cf768 | 3110 | "\thttp://pcalishuttle01.cern.ch:8880/%s/%d/%s_%d.log \n\n", |
3111 | fCurrentDetector.Data(), logFolder.Data(), GetCurrentRun(), | |
3112 | fCurrentDetector.Data(), GetCurrentRun()); | |
3113 | body += Form("The last 10 lines of %s log file are following:\n\n", fCurrentDetector.Data()); | |
57c1a579 | 3114 | |
909732f7 | 3115 | AliDebug(2, Form("Body begin: %s", body.Data())); |
57c1a579 | 3116 | |
3117 | mailBody << body.Data(); | |
3118 | mailBody.close(); | |
3119 | mailBody.open(bodyFileName, ofstream::out | ofstream::app); | |
3120 | ||
7d4cf768 | 3121 | TString logFileName = Form("%s/%d/%s_%d.log", GetShuttleLogDir(), |
3122 | GetCurrentRun(), fCurrentDetector.Data(), GetCurrentRun()); | |
57c1a579 | 3123 | TString tailCommand = Form("tail -n 10 %s >> %s", logFileName.Data(), bodyFileName.Data()); |
3124 | if (gSystem->Exec(tailCommand.Data())) | |
3125 | { | |
3126 | mailBody << Form("%s log file not found ...\n\n", fCurrentDetector.Data()); | |
3127 | } | |
3128 | ||
3129 | TString endBody = Form("------------------------------------------------------\n\n"); | |
675f64cd | 3130 | endBody += Form("In case of problems please contact the SHUTTLE core team.\n\n"); |
3131 | endBody += "Please do not answer this message directly, it is automatically generated.\n\n"; | |
3132 | endBody += "Greetings,\n\n \t\t\tthe SHUTTLE\n"; | |
3133 | ||
3134 | AliDebug(2, Form("Body end: %s", endBody.Data())); | |
3135 | ||
3136 | mailBody << endBody.Data(); | |
3137 | ||
3138 | mailBody.close(); | |
3139 | ||
3140 | // send mail! | |
3141 | TString mailCommand = Form("mail -s \"%s\" -c %s %s < %s", | |
3142 | subject.Data(), | |
3143 | cc.Data(), | |
3144 | to.Data(), | |
3145 | bodyFileName.Data()); | |
3146 | AliDebug(2, Form("mail command: %s", mailCommand.Data())); | |
3147 | ||
3148 | Bool_t result = gSystem->Exec(mailCommand.Data()); | |
3149 | ||
3150 | return result == 0; | |
3151 | } | |
3152 | ||
3153 | //______________________________________________________________________________________________ | |
3154 | Bool_t AliShuttle::SendMailToDCS() | |
3155 | { | |
3156 | // | |
fb40ead4 | 3157 | // sends a mail to the DCS Amanda experts in case of DCS data point retrieval error |
675f64cd | 3158 | // |
3159 | ||
3160 | if (fTestMode != kNone) | |
3161 | return kTRUE; | |
3162 | ||
926cbc0e | 3163 | if (!fConfig->SendMail()) |
3164 | return kTRUE; | |
3165 | ||
3166 | if (!fFirstProcessing) | |
3167 | return kTRUE; | |
fb40ead4 | 3168 | |
675f64cd | 3169 | void* dir = gSystem->OpenDirectory(GetShuttleLogDir()); |
3170 | if (dir == NULL) | |
3171 | { | |
3172 | if (gSystem->mkdir(GetShuttleLogDir(), kTRUE)) | |
3173 | { | |
3174 | Log("SHUTTLE", Form("SendMailToDCS - Can't open directory <%s>", GetShuttleLogDir())); | |
3175 | return kFALSE; | |
3176 | } | |
3177 | ||
3178 | } else { | |
3179 | gSystem->FreeDirectory(dir); | |
3180 | } | |
3181 | ||
3182 | TString bodyFileName; | |
3183 | bodyFileName.Form("%s/mail.body", GetShuttleLogDir()); | |
3184 | gSystem->ExpandPathName(bodyFileName); | |
3185 | ||
3186 | ofstream mailBody; | |
3187 | mailBody.open(bodyFileName, ofstream::out); | |
3188 | ||
3189 | if (!mailBody.is_open()) | |
3190 | { | |
3191 | Log("SHUTTLE", Form("SendMailToDCS - Could not open mail body file %s", bodyFileName.Data())); | |
3192 | return kFALSE; | |
3193 | } | |
3194 | ||
fb40ead4 | 3195 | TString to=""; |
3196 | TIter iterExperts(fConfig->GetAdmins(AliShuttleConfig::kAmanda)); | |
3197 | TObjString *anExpert=0; | |
3198 | while ((anExpert = (TObjString*) iterExperts.Next())) | |
3199 | { | |
3200 | to += Form("%s,", anExpert->GetName()); | |
3201 | } | |
3202 | if (to.Length() > 0) | |
3203 | to.Remove(to.Length()-1); | |
675f64cd | 3204 | AliDebug(2, Form("to: %s",to.Data())); |
3205 | ||
3206 | if (to.IsNull()) { | |
fb40ead4 | 3207 | Log("SHUTTLE", "List of Amanda server administrators not set!"); |
675f64cd | 3208 | return kFALSE; |
3209 | } | |
3210 | ||
fb40ead4 | 3211 | TString cc=""; |
3212 | TIter iterAdmins(fConfig->GetAdmins(AliShuttleConfig::kGlobal)); | |
3213 | TObjString *anAdmin=0; | |
3214 | while ((anAdmin = (TObjString*) iterAdmins.Next())) | |
3215 | { | |
3216 | cc += Form("%s,", anAdmin->GetName()); | |
3217 | } | |
3218 | if (cc.Length() > 0) | |
c19963db | 3219 | cc.Remove(cc.Length()-1); |
fb40ead4 | 3220 | AliDebug(2, Form("cc: %s",to.Data())); |
675f64cd | 3221 | |
3222 | TString subject = Form("Retrieval of data points for %s FAILED in run %d !", | |
3223 | fCurrentDetector.Data(), GetCurrentRun()); | |
3224 | AliDebug(2, Form("subject: %s", subject.Data())); | |
3225 | ||
3226 | TString body = Form("Dear DCS experts, \n\n"); | |
3227 | body += Form("SHUTTLE couldn\'t retrieve the data points for detector %s " | |
3228 | "in run %d!!\n\n", fCurrentDetector.Data(), GetCurrentRun()); | |
7d4cf768 | 3229 | body += Form("Please check %s status on the SHUTTLE monitoring page: \n\n", |
3230 | fCurrentDetector.Data()); | |
b0e53b15 | 3231 | if (fConfig->GetRunMode() == AliShuttleConfig::kTest) |
3232 | { | |
3233 | body += Form("\thttp://pcalimonitor.cern.ch:8889/shuttle.jsp?time=168 \n\n"); | |
3234 | } else { | |
3235 | body += Form("\thttp://pcalimonitor.cern.ch/shuttle.jsp?instance=PROD?time=168 \n\n"); | |
3236 | } | |
7d4cf768 | 3237 | |
3238 | TString logFolder = "logs"; | |
3239 | if (fConfig->GetRunMode() == AliShuttleConfig::kProd) | |
3240 | logFolder += "_PROD"; | |
3241 | ||
3242 | ||
675f64cd | 3243 | body += Form("Find the %s log for the current run on \n\n" |
7d4cf768 | 3244 | "\thttp://pcalishuttle01.cern.ch:8880/%s/%d/%s_%d.log \n\n", |
3245 | fCurrentDetector.Data(), logFolder.Data(), GetCurrentRun(), | |
3246 | fCurrentDetector.Data(), GetCurrentRun()); | |
3247 | body += Form("The last 10 lines of %s log file are following:\n\n", fCurrentDetector.Data()); | |
675f64cd | 3248 | |
3249 | AliDebug(2, Form("Body begin: %s", body.Data())); | |
3250 | ||
3251 | mailBody << body.Data(); | |
3252 | mailBody.close(); | |
3253 | mailBody.open(bodyFileName, ofstream::out | ofstream::app); | |
3254 | ||
7d4cf768 | 3255 | TString logFileName = Form("%s/%d/%s_%d.log", GetShuttleLogDir(), GetCurrentRun(), |
3256 | fCurrentDetector.Data(), GetCurrentRun()); | |
675f64cd | 3257 | TString tailCommand = Form("tail -n 10 %s >> %s", logFileName.Data(), bodyFileName.Data()); |
3258 | if (gSystem->Exec(tailCommand.Data())) | |
3259 | { | |
3260 | mailBody << Form("%s log file not found ...\n\n", fCurrentDetector.Data()); | |
3261 | } | |
3262 | ||
3263 | TString endBody = Form("------------------------------------------------------\n\n"); | |
36c99a6a | 3264 | endBody += Form("In case of problems please contact the SHUTTLE core team.\n\n"); |
3265 | endBody += "Please do not answer this message directly, it is automatically generated.\n\n"; | |
546242fb | 3266 | endBody += "Greetings,\n\n \t\t\tthe SHUTTLE\n"; |
57c1a579 | 3267 | |
909732f7 | 3268 | AliDebug(2, Form("Body end: %s", endBody.Data())); |
57c1a579 | 3269 | |
3270 | mailBody << endBody.Data(); | |
3271 | ||
3272 | mailBody.close(); | |
3273 | ||
3274 | // send mail! | |
3275 | TString mailCommand = Form("mail -s \"%s\" -c %s %s < %s", | |
3276 | subject.Data(), | |
3277 | cc.Data(), | |
3278 | to.Data(), | |
3279 | bodyFileName.Data()); | |
909732f7 | 3280 | AliDebug(2, Form("mail command: %s", mailCommand.Data())); |
57c1a579 | 3281 | |
3282 | Bool_t result = gSystem->Exec(mailCommand.Data()); | |
3283 | ||
3284 | return result == 0; | |
3285 | } | |
d386d623 | 3286 | |
441b0e9c | 3287 | //______________________________________________________________________________________________ |
9827400b | 3288 | const char* AliShuttle::GetRunType() |
441b0e9c | 3289 | { |
9827400b | 3290 | // |
3291 | // returns run type read from "run type" logbook | |
3292 | // | |
441b0e9c | 3293 | |
3294 | if(!fLogbookEntry) { | |
3295 | AliError("No logbook entry!"); | |
3296 | return 0; | |
3297 | } | |
3298 | ||
9827400b | 3299 | return fLogbookEntry->GetRunType(); |
441b0e9c | 3300 | } |
3301 | ||
4859271b | 3302 | //______________________________________________________________________________________________ |
3303 | Bool_t AliShuttle::GetHLTStatus() | |
3304 | { | |
3305 | // Return HLT status (ON=1 OFF=0) | |
3306 | // Converts the HLT status from the status string read in the run logbook (not just a bool) | |
3307 | ||
3308 | if(!fLogbookEntry) { | |
3309 | AliError("No logbook entry!"); | |
3310 | return 0; | |
3311 | } | |
3312 | ||
3313 | // TODO implement when HLTStatus is inserted in run logbook | |
3314 | //TString hltStatus = fLogbookEntry->GetRunParameter("HLTStatus"); | |
3315 | //if(hltStatus == "OFF") {return kFALSE}; | |
3316 | ||
3317 | return kTRUE; | |
3318 | } | |
3319 | ||
d386d623 | 3320 | //______________________________________________________________________________________________ |
3321 | void AliShuttle::SetShuttleTempDir(const char* tmpDir) | |
3322 | { | |
9827400b | 3323 | // |
3324 | // sets Shuttle temp directory | |
3325 | // | |
d386d623 | 3326 | |
3327 | fgkShuttleTempDir = gSystem->ExpandPathName(tmpDir); | |
3328 | } | |
3329 | ||
3330 | //______________________________________________________________________________________________ | |
3331 | void AliShuttle::SetShuttleLogDir(const char* logDir) | |
3332 | { | |
9827400b | 3333 | // |
3334 | // sets Shuttle log directory | |
3335 | // | |
d386d623 | 3336 | |
3337 | fgkShuttleLogDir = gSystem->ExpandPathName(logDir); | |
3338 | } |