]>
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> | |
503ff24f | 60 | #include <TMap.h> |
73abe331 | 61 | |
e7f62f16 | 62 | #include <TMonaLisaWriter.h> |
63 | ||
5164a766 | 64 | #include <fstream> |
65 | ||
cb343cfd | 66 | #include <sys/types.h> |
67 | #include <sys/wait.h> | |
68 | ||
1bd183ba | 69 | #include <signal.h> |
70 | ||
73abe331 | 71 | ClassImp(AliShuttle) |
72 | ||
b948db8d | 73 | //______________________________________________________________________________________________ |
74 | AliShuttle::AliShuttle(const AliShuttleConfig* config, | |
75 | UInt_t timeout, Int_t retries): | |
4f0ab988 | 76 | fConfig(config), |
77 | fTimeout(timeout), fRetries(retries), | |
78 | fPreprocessorMap(), | |
2bb7b766 | 79 | fLogbookEntry(0), |
eba76848 | 80 | fCurrentDetector(), |
926cbc0e | 81 | fFirstProcessing(0), |
339adafa | 82 | fFXSError(-1), |
85a80aa9 | 83 | fStatusEntry(0), |
cb343cfd | 84 | fMonitoringMutex(0), |
eba76848 | 85 | fLastActionTime(0), |
e7f62f16 | 86 | fLastAction(), |
9827400b | 87 | fMonaLisa(0), |
88 | fTestMode(kNone), | |
ffa29e93 | 89 | fReadTestMode(kFALSE), |
90 | fOutputRedirected(kFALSE) | |
73abe331 | 91 | { |
92 | // | |
93 | // config: AliShuttleConfig used | |
73abe331 | 94 | // timeout: timeout used for AliDCSClient connection |
95 | // retries: the number of retries in case of connection error. | |
96 | // | |
97 | ||
57f50b3c | 98 | if (!fConfig->IsValid()) AliFatal("********** !!!!! Invalid configuration !!!!! **********"); |
be48e3ea | 99 | for(int iSys=0;iSys<4;iSys++) { |
57f50b3c | 100 | fServer[iSys]=0; |
be48e3ea | 101 | if (iSys < 3) |
2c15234c | 102 | fFXSlist[iSys].SetOwner(kTRUE); |
57f50b3c | 103 | } |
2bb7b766 | 104 | fPreprocessorMap.SetOwner(kTRUE); |
be48e3ea | 105 | |
106 | for (UInt_t iDet=0; iDet<NDetectors(); iDet++) | |
107 | fFirstUnprocessed[iDet] = kFALSE; | |
108 | ||
cb343cfd | 109 | fMonitoringMutex = new TMutex(); |
58bc3020 | 110 | } |
111 | ||
b948db8d | 112 | //______________________________________________________________________________________________ |
57f50b3c | 113 | AliShuttle::~AliShuttle() |
58bc3020 | 114 | { |
9827400b | 115 | // |
116 | // destructor | |
117 | // | |
58bc3020 | 118 | |
b948db8d | 119 | fPreprocessorMap.DeleteAll(); |
be48e3ea | 120 | for(int iSys=0;iSys<4;iSys++) |
57f50b3c | 121 | if(fServer[iSys]) { |
122 | fServer[iSys]->Close(); | |
123 | delete fServer[iSys]; | |
eba76848 | 124 | fServer[iSys] = 0; |
57f50b3c | 125 | } |
2bb7b766 | 126 | |
127 | if (fStatusEntry){ | |
128 | delete fStatusEntry; | |
129 | fStatusEntry = 0; | |
130 | } | |
cb343cfd | 131 | |
132 | if (fMonitoringMutex) | |
133 | { | |
134 | delete fMonitoringMutex; | |
135 | fMonitoringMutex = 0; | |
136 | } | |
73abe331 | 137 | } |
138 | ||
b948db8d | 139 | //______________________________________________________________________________________________ |
57f50b3c | 140 | void AliShuttle::RegisterPreprocessor(AliPreprocessor* preprocessor) |
58bc3020 | 141 | { |
73abe331 | 142 | // |
b948db8d | 143 | // Registers new AliPreprocessor. |
73abe331 | 144 | // It uses GetName() for indentificator of the pre processor. |
145 | // The pre processor is registered it there isn't any other | |
146 | // with the same identificator (GetName()). | |
147 | // | |
148 | ||
eba76848 | 149 | const char* detName = preprocessor->GetName(); |
150 | if(GetDetPos(detName) < 0) | |
151 | AliFatal(Form("********** !!!!! Invalid detector name: %s !!!!! **********", detName)); | |
152 | ||
153 | if (fPreprocessorMap.GetValue(detName)) { | |
154 | AliWarning(Form("AliPreprocessor %s is already registered!", detName)); | |
73abe331 | 155 | return; |
156 | } | |
157 | ||
eba76848 | 158 | fPreprocessorMap.Add(new TObjString(detName), preprocessor); |
73abe331 | 159 | } |
b948db8d | 160 | //______________________________________________________________________________________________ |
3301427a | 161 | Bool_t AliShuttle::Store(const AliCDBPath& path, TObject* object, |
84090f85 | 162 | AliCDBMetaData* metaData, Int_t validityStart, Bool_t validityInfinite) |
73abe331 | 163 | { |
9827400b | 164 | // Stores a CDB object in the storage for offline reconstruction. Objects that are not needed for |
165 | // offline reconstruction, but should be stored anyway (e.g. for debugging) should NOT be stored | |
166 | // using this function. Use StoreReferenceData instead! | |
167 | // It calls StoreLocally function which temporarily stores the data locally; when the preprocessor | |
168 | // finishes the data are transferred to the main storage (Grid). | |
b948db8d | 169 | |
3301427a | 170 | return StoreLocally(fgkLocalCDB, path, object, metaData, validityStart, validityInfinite); |
84090f85 | 171 | } |
172 | ||
173 | //______________________________________________________________________________________________ | |
3301427a | 174 | Bool_t AliShuttle::StoreReferenceData(const AliCDBPath& path, TObject* object, AliCDBMetaData* metaData) |
84090f85 | 175 | { |
9827400b | 176 | // Stores a CDB object in the storage for reference data. This objects will not be available during |
177 | // offline reconstrunction. Use this function for reference data only! | |
178 | // It calls StoreLocally function which temporarily stores the data locally; when the preprocessor | |
179 | // finishes the data are transferred to the main storage (Grid). | |
85a80aa9 | 180 | |
3301427a | 181 | return StoreLocally(fgkLocalRefStorage, path, object, metaData); |
85a80aa9 | 182 | } |
183 | ||
184 | //______________________________________________________________________________________________ | |
3301427a | 185 | Bool_t AliShuttle::StoreLocally(const TString& localUri, |
85a80aa9 | 186 | const AliCDBPath& path, TObject* object, AliCDBMetaData* metaData, |
187 | Int_t validityStart, Bool_t validityInfinite) | |
188 | { | |
9827400b | 189 | // Store object temporarily in local storage. Parameters are passed by Store and StoreReferenceData functions. |
190 | // when the preprocessor finishes the data are transferred to the main storage (Grid). | |
191 | // The parameters are: | |
192 | // 1) Uri of the backup storage (Local) | |
193 | // 2) the object's path. | |
194 | // 3) the object to be stored | |
195 | // 4) the metaData to be associated with the object | |
196 | // 5) the validity start run number w.r.t. the current run, | |
197 | // if the data is valid only for this run leave the default 0 | |
198 | // 6) specifies if the calibration data is valid for infinity (this means until updated), | |
199 | // typical for calibration runs, the default is kFALSE | |
200 | // | |
201 | // returns 0 if fail, 1 otherwise | |
84090f85 | 202 | |
924fb6e9 | 203 | |
9827400b | 204 | if (fTestMode & kErrorStorage) |
205 | { | |
206 | Log(fCurrentDetector, "StoreLocally - In TESTMODE - Simulating error while storing locally"); | |
207 | return kFALSE; | |
208 | } | |
209 | ||
3301427a | 210 | const char* cdbType = (localUri == fgkLocalCDB) ? "CDB" : "Reference"; |
2bb7b766 | 211 | |
85a80aa9 | 212 | Int_t firstRun = GetCurrentRun() - validityStart; |
84090f85 | 213 | if(firstRun < 0) { |
9827400b | 214 | AliWarning("First valid run happens to be less than 0! Setting it to 0."); |
84090f85 | 215 | firstRun=0; |
216 | } | |
217 | ||
218 | Int_t lastRun = -1; | |
219 | if(validityInfinite) { | |
220 | lastRun = AliCDBRunRange::Infinity(); | |
221 | } else { | |
222 | lastRun = GetCurrentRun(); | |
223 | } | |
224 | ||
3301427a | 225 | // Version is set to current run, it will be used later to transfer data to Grid |
226 | AliCDBId id(path, firstRun, lastRun, GetCurrentRun(), -1); | |
2bb7b766 | 227 | |
228 | if(! dynamic_cast<TObjString*> (metaData->GetProperty("RunUsed(TObjString)"))){ | |
229 | TObjString runUsed = Form("%d", GetCurrentRun()); | |
9e080f92 | 230 | metaData->SetProperty("RunUsed(TObjString)", runUsed.Clone()); |
2bb7b766 | 231 | } |
84090f85 | 232 | |
3301427a | 233 | Bool_t result = kFALSE; |
84090f85 | 234 | |
3301427a | 235 | if (!(AliCDBManager::Instance()->GetStorage(localUri))) { |
236 | Log("SHUTTLE", Form("StoreLocally - Cannot activate local %s storage", cdbType)); | |
84090f85 | 237 | } else { |
924fb6e9 | 238 | Int_t logLevel = AliLog::GetGlobalLogLevel(); |
239 | AliLog::SetGlobalLogLevel(AliLog::kError); | |
3301427a | 240 | result = AliCDBManager::Instance()->GetStorage(localUri) |
84090f85 | 241 | ->Put(object, id, metaData); |
924fb6e9 | 242 | AliLog::SetGlobalLogLevel((AliLog::EType_t)logLevel); |
84090f85 | 243 | } |
244 | ||
245 | if(!result) { | |
246 | ||
9827400b | 247 | Log(fCurrentDetector, Form("StoreLocally - Can't store object <%s>!", id.ToString().Data())); |
3301427a | 248 | } |
2bb7b766 | 249 | |
924fb6e9 | 250 | |
3301427a | 251 | return result; |
252 | } | |
84090f85 | 253 | |
3301427a | 254 | //______________________________________________________________________________________________ |
255 | Bool_t AliShuttle::StoreOCDB() | |
256 | { | |
9827400b | 257 | // |
258 | // Called when preprocessor ends successfully or when previous storage attempt failed (kStoreError status) | |
259 | // Calls underlying StoreOCDB(const char*) function twice, for OCDB and Reference storage. | |
260 | // Then calls StoreRefFilesToGrid to store reference files. | |
261 | // | |
262 | ||
5e993b6f | 263 | UpdateShuttleStatus(AliShuttleStatus::kStoreStarted); |
56d3e9e4 | 264 | |
9827400b | 265 | if (fTestMode & kErrorGrid) |
266 | { | |
267 | Log("SHUTTLE", "StoreOCDB - In TESTMODE - Simulating error while storing in the Grid"); | |
268 | Log(fCurrentDetector, "StoreOCDB - In TESTMODE - Simulating error while storing in the Grid"); | |
269 | return kFALSE; | |
270 | } | |
271 | ||
c88ad5db | 272 | Log("SHUTTLE","StoreOCDB - Storing OCDB data ..."); |
5e993b6f | 273 | Int_t resultCDB = StoreOCDB(fgkMainCDB); |
86aa42c3 | 274 | |
c88ad5db | 275 | Log("SHUTTLE","StoreOCDB - Storing reference data ..."); |
5e993b6f | 276 | Int_t resultRef = StoreOCDB(fgkMainRefStorage); |
9827400b | 277 | |
c88ad5db | 278 | Log("SHUTTLE","StoreOCDB - Storing reference files ..."); |
279 | Bool_t resultRefFiles = CopyFilesToGrid("reference"); | |
280 | ||
281 | Bool_t resultMetadata = kTRUE; | |
282 | if(fCurrentDetector == "GRP") | |
283 | { | |
7802e884 | 284 | Log("SHUTTLE","StoreOCDB - Storing Run Metadata file ..."); |
c88ad5db | 285 | resultMetadata = CopyFilesToGrid("metadata"); |
286 | } | |
9827400b | 287 | |
5e993b6f | 288 | Int_t storeResult = 0; |
289 | ||
290 | if (resultCDB < 0 || resultRef < 0 || resultRefFiles == kFALSE || resultMetadata == kFALSE) | |
291 | storeResult = -1; | |
292 | else if (resultCDB > 0 || resultRef > 0) | |
293 | storeResult = 1; | |
294 | ||
295 | if (storeResult < 0) | |
296 | { | |
297 | Log("SHUTTLE", | |
298 | Form("\t\t\t****** run %d - %s: STORAGE ERROR ******", | |
299 | GetCurrentRun(), fCurrentDetector.Data())); | |
300 | UpdateShuttleStatus(AliShuttleStatus::kStoreError); | |
301 | } | |
302 | else if (storeResult > 0) | |
303 | { | |
304 | Log("SHUTTLE", | |
305 | Form("\t\t\t****** run %d - %s: STORAGE DELAYED ******", | |
306 | GetCurrentRun(), fCurrentDetector.Data())); | |
307 | UpdateShuttleStatus(AliShuttleStatus::kStoreDelayed); | |
308 | } | |
309 | else if (storeResult == 0) | |
310 | { | |
311 | Log("SHUTTLE", | |
312 | Form("\t\t\t****** run %d - %s: DONE ******", | |
313 | GetCurrentRun(), fCurrentDetector.Data())); | |
314 | UpdateShuttleStatus(AliShuttleStatus::kDone); | |
315 | UpdateShuttleLogbook(fCurrentDetector, "DONE"); | |
316 | } | |
317 | ||
318 | return (storeResult == 0); | |
3301427a | 319 | } |
320 | ||
321 | //______________________________________________________________________________________________ | |
5e993b6f | 322 | Int_t AliShuttle::StoreOCDB(const TString& gridURI) |
3301427a | 323 | { |
324 | // | |
325 | // Called by StoreOCDB(), performs actual storage to the main OCDB and reference storages (Grid) | |
326 | // | |
5e993b6f | 327 | // Return code: |
328 | // -2 initialization error | |
329 | // -1 storage error | |
330 | // 0 success | |
331 | // 1 storage delayed (e.g. previous unprocessed runs) | |
332 | // | |
3301427a | 333 | |
334 | TObjArray* gridIds=0; | |
335 | ||
336 | Bool_t result = kTRUE; | |
5e993b6f | 337 | Bool_t delayed = kFALSE; |
3301427a | 338 | |
339 | const char* type = 0; | |
340 | TString localURI; | |
341 | if(gridURI == fgkMainCDB) { | |
342 | type = "OCDB"; | |
343 | localURI = fgkLocalCDB; | |
344 | } else if(gridURI == fgkMainRefStorage) { | |
345 | type = "reference"; | |
346 | localURI = fgkLocalRefStorage; | |
347 | } else { | |
348 | AliError(Form("Invalid storage URI: %s", gridURI.Data())); | |
5e993b6f | 349 | return -2; |
3301427a | 350 | } |
351 | ||
352 | AliCDBManager* man = AliCDBManager::Instance(); | |
353 | ||
354 | AliCDBStorage *gridSto = man->GetStorage(gridURI); | |
355 | if(!gridSto) { | |
356 | Log("SHUTTLE", | |
357 | Form("StoreOCDB - cannot activate main %s storage", type)); | |
5e993b6f | 358 | return -2; |
3301427a | 359 | } |
360 | ||
361 | gridIds = gridSto->GetQueryCDBList(); | |
362 | ||
363 | // get objects previously stored in local CDB | |
364 | AliCDBStorage *localSto = man->GetStorage(localURI); | |
365 | if(!localSto) { | |
366 | Log("SHUTTLE", | |
367 | Form("StoreOCDB - cannot activate local %s storage", type)); | |
5e993b6f | 368 | return -2; |
3301427a | 369 | } |
370 | AliCDBPath aPath(GetOfflineDetName(fCurrentDetector.Data()),"*","*"); | |
371 | // Local objects were stored with current run as Grid version! | |
372 | TList* localEntries = localSto->GetAll(aPath.GetPath(), GetCurrentRun(), GetCurrentRun()); | |
373 | localEntries->SetOwner(1); | |
374 | ||
375 | // loop on local stored objects | |
376 | TIter localIter(localEntries); | |
377 | AliCDBEntry *aLocEntry = 0; | |
378 | while((aLocEntry = dynamic_cast<AliCDBEntry*> (localIter.Next()))){ | |
379 | aLocEntry->SetOwner(1); | |
380 | AliCDBId aLocId = aLocEntry->GetId(); | |
381 | aLocEntry->SetVersion(-1); | |
382 | aLocEntry->SetSubVersion(-1); | |
6c0c9948 | 383 | |
384 | Log(fCurrentDetector.Data(), Form("Attempting to store %s", aLocId.ToString().Data())); | |
3301427a | 385 | |
386 | // If local object is valid up to infinity we store it only if it is | |
387 | // the first unprocessed run! | |
388 | if (aLocId.GetLastRun() == AliCDBRunRange::Infinity() && | |
389 | !fFirstUnprocessed[GetDetPos(fCurrentDetector)]) | |
390 | { | |
391 | Log("SHUTTLE", Form("StoreOCDB - %s: object %s has validity infinite but " | |
392 | "there are previous unprocessed runs!", | |
393 | fCurrentDetector.Data(), aLocId.GetPath().Data())); | |
5e993b6f | 394 | Log(fCurrentDetector.Data(), Form("StoreOCDB - %s: object %s has validity infinite but " |
395 | "there are previous unprocessed runs!", | |
396 | fCurrentDetector.Data(), aLocId.GetPath().Data())); | |
397 | delayed = kTRUE; | |
3301427a | 398 | continue; |
399 | } | |
400 | ||
401 | // loop on Grid valid Id's | |
402 | Bool_t store = kTRUE; | |
403 | TIter gridIter(gridIds); | |
404 | AliCDBId* aGridId = 0; | |
6c0c9948 | 405 | while ((aGridId = dynamic_cast<AliCDBId*> (gridIter.Next()))) { |
406 | if (aGridId->GetPath() != aLocId.GetPath()) | |
407 | continue; | |
3301427a | 408 | // skip all objects valid up to infinity |
6c0c9948 | 409 | if (aGridId->GetLastRun() == AliCDBRunRange::Infinity()) |
410 | continue; | |
411 | ||
3301427a | 412 | // if we get here, it means there's already some more recent object stored on Grid! |
6c0c9948 | 413 | Log(fCurrentDetector.Data(), |
414 | Form("StoreOCDB - A more recent object already exists in %s storage: <%s>", | |
415 | type, aGridId->ToString().Data())); | |
416 | ||
3301427a | 417 | store = kFALSE; |
418 | break; | |
419 | } | |
420 | ||
6c0c9948 | 421 | Bool_t storeOk = kFALSE; |
422 | if (store) | |
423 | { | |
424 | Log(fCurrentDetector.Data(), Form("Prechecks succeeded. Ready to store %s", aLocId.ToString().Data())); | |
425 | storeOk = gridSto->Put(aLocEntry); | |
426 | if (storeOk) { | |
3301427a | 427 | Log("SHUTTLE", |
6c0c9948 | 428 | Form("StoreOCDB - Object <%s> successfully put into %s storage", |
429 | aLocId.ToString().Data(), type)); | |
2d9019b4 | 430 | Log(fCurrentDetector.Data(), |
431 | Form("StoreOCDB - Object <%s> successfully put into %s storage", | |
6c0c9948 | 432 | aLocId.ToString().Data(), type)); |
433 | } else { | |
434 | Log("SHUTTLE", | |
435 | Form("StoreOCDB - Grid %s storage of object <%s> failed", | |
436 | type, aLocId.ToString().Data())); | |
437 | Log(fCurrentDetector.Data(), | |
438 | Form("StoreOCDB - Grid %s storage of object <%s> failed", | |
439 | type, aLocId.ToString().Data())); | |
440 | result = kFALSE; | |
3301427a | 441 | } |
6c0c9948 | 442 | } |
443 | ||
444 | if (!store || storeOk) { | |
445 | // removing local file... | |
3301427a | 446 | TString filename; |
447 | localSto->IdToFilename(aLocId, filename); | |
c88ad5db | 448 | Log("SHUTTLE", Form("StoreOCDB - Removing local file %s", filename.Data())); |
3301427a | 449 | RemoveFile(filename.Data()); |
b948db8d | 450 | } |
451 | } | |
3301427a | 452 | localEntries->Clear(); |
2bb7b766 | 453 | |
5e993b6f | 454 | Int_t returnCode = 0; |
455 | ||
456 | if (result == kFALSE) | |
457 | returnCode = -1; | |
458 | else if (delayed != kFALSE) | |
459 | returnCode = 1; | |
460 | ||
461 | Log("SHUTTLE", Form("StoreOCDB - Returning with %d (result = %d, delayed = %d)", returnCode, result, delayed)); | |
462 | Log(fCurrentDetector.Data(), Form("StoreOCDB - Returning with %d (result = %d, delayed = %d)", returnCode, result, delayed)); | |
463 | ||
464 | return returnCode; | |
3301427a | 465 | } |
466 | ||
546242fb | 467 | //______________________________________________________________________________________________ |
468 | Bool_t AliShuttle::CleanReferenceStorage(const char* detector) | |
469 | { | |
2d9019b4 | 470 | // clears the directory used to store reference files of a given subdetector |
546242fb | 471 | |
472 | AliCDBManager* man = AliCDBManager::Instance(); | |
473 | AliCDBStorage* sto = man->GetStorage(fgkLocalRefStorage); | |
2d9019b4 | 474 | TString localBaseFolder = sto->GetBaseFolder(); |
475 | ||
476 | TString targetDir = GetRefFilePrefix(localBaseFolder.Data(), detector); | |
477 | ||
d524ade6 | 478 | Log("SHUTTLE", Form("CleanReferenceStorage - Cleaning %s", targetDir.Data())); |
2d9019b4 | 479 | |
480 | TString begin; | |
481 | begin.Form("%d_", GetCurrentRun()); | |
482 | ||
483 | TSystemDirectory* baseDir = new TSystemDirectory("/", targetDir); | |
484 | if (!baseDir) | |
485 | return kTRUE; | |
486 | ||
487 | TList* dirList = baseDir->GetListOfFiles(); | |
488 | delete baseDir; | |
489 | ||
490 | if (!dirList) return kTRUE; | |
491 | ||
492 | if (dirList->GetEntries() < 3) | |
493 | { | |
494 | delete dirList; | |
495 | return kTRUE; | |
496 | } | |
497 | ||
498 | Int_t nDirs = 0, nDel = 0; | |
499 | TIter dirIter(dirList); | |
500 | TSystemFile* entry = 0; | |
546242fb | 501 | |
2d9019b4 | 502 | Bool_t success = kTRUE; |
546242fb | 503 | |
2d9019b4 | 504 | while ((entry = dynamic_cast<TSystemFile*> (dirIter.Next()))) |
505 | { | |
506 | if (entry->IsDirectory()) | |
507 | continue; | |
508 | ||
509 | TString fileName(entry->GetName()); | |
510 | if (!fileName.BeginsWith(begin)) | |
511 | continue; | |
512 | ||
513 | nDirs++; | |
514 | ||
515 | // delete file | |
516 | Int_t result = gSystem->Unlink(fileName.Data()); | |
517 | ||
518 | if (result) | |
519 | { | |
d524ade6 | 520 | Log("SHUTTLE", Form("CleanReferenceStorage - Could not delete file %s!", fileName.Data())); |
2d9019b4 | 521 | success = kFALSE; |
522 | } else { | |
523 | nDel++; | |
524 | } | |
525 | } | |
526 | ||
527 | if(nDirs > 0) | |
528 | Log("SHUTTLE", Form("CleanReferenceStorage - %d (over %d) reference files in folder %s were deleted.", | |
529 | nDel, nDirs, targetDir.Data())); | |
530 | ||
531 | ||
532 | delete dirList; | |
533 | return success; | |
534 | ||
535 | ||
536 | ||
537 | ||
538 | ||
546242fb | 539 | |
540 | Int_t result = gSystem->GetPathInfo(targetDir, 0, (Long64_t*) 0, 0, 0); | |
541 | if (result == 0) | |
542 | { | |
543 | // delete directory | |
d524ade6 | 544 | result = gSystem->Exec(Form("rm -rf %s", targetDir.Data())); |
546242fb | 545 | if (result != 0) |
546 | { | |
d524ade6 | 547 | Log("SHUTTLE", Form("CleanReferenceStorage - Could not clean directory %s", targetDir.Data())); |
546242fb | 548 | return kFALSE; |
549 | } | |
550 | } | |
551 | ||
552 | result = gSystem->mkdir(targetDir, kTRUE); | |
553 | if (result != 0) | |
554 | { | |
c88ad5db | 555 | Log("SHUTTLE", Form("CleanReferenceStorage - Error creating base directory %s", targetDir.Data())); |
546242fb | 556 | return kFALSE; |
557 | } | |
558 | ||
559 | return kTRUE; | |
560 | } | |
561 | ||
9827400b | 562 | //______________________________________________________________________________________________ |
563 | Bool_t AliShuttle::StoreReferenceFile(const char* detector, const char* localFile, const char* gridFileName) | |
564 | { | |
565 | // | |
3c2a21c8 | 566 | // Stores reference file directly (without opening it). This function stores the file locally. |
9827400b | 567 | // |
3c2a21c8 | 568 | // The file is stored under the following location: |
569 | // <base folder of local reference storage>/<DET>/<RUN#>_<gridFileName> | |
570 | // where <gridFileName> is the second parameter given to the function | |
571 | // | |
9827400b | 572 | |
573 | if (fTestMode & kErrorStorage) | |
574 | { | |
575 | Log(fCurrentDetector, "StoreReferenceFile - In TESTMODE - Simulating error while storing locally"); | |
576 | return kFALSE; | |
577 | } | |
578 | ||
579 | AliCDBManager* man = AliCDBManager::Instance(); | |
580 | AliCDBStorage* sto = man->GetStorage(fgkLocalRefStorage); | |
581 | ||
582 | TString localBaseFolder = sto->GetBaseFolder(); | |
583 | ||
d524ade6 | 584 | TString target = GetRefFilePrefix(localBaseFolder.Data(), detector); |
585 | target.Append(Form("/%d_%s", GetCurrentRun(), gridFileName)); | |
9827400b | 586 | |
d524ade6 | 587 | return CopyFileLocally(localFile, target); |
c88ad5db | 588 | } |
589 | ||
590 | //______________________________________________________________________________________________ | |
591 | Bool_t AliShuttle::StoreRunMetadataFile(const char* localFile, const char* gridFileName) | |
592 | { | |
593 | // | |
594 | // Stores Run metadata file to the Grid, in the run folder | |
595 | // | |
596 | // Only GRP can call this function. | |
597 | ||
598 | if (fTestMode & kErrorStorage) | |
599 | { | |
600 | Log(fCurrentDetector, "StoreRunMetaDataFile - In TESTMODE - Simulating error while storing locally"); | |
601 | return kFALSE; | |
602 | } | |
603 | ||
604 | AliCDBManager* man = AliCDBManager::Instance(); | |
605 | AliCDBStorage* sto = man->GetStorage(fgkLocalRefStorage); | |
606 | ||
607 | TString localBaseFolder = sto->GetBaseFolder(); | |
608 | ||
609 | // Build Run level folder | |
955d2abc | 610 | // folder = /alice/data/year/lhcPeriod/runNb/raw |
c88ad5db | 611 | |
c88ad5db | 612 | |
675f64cd | 613 | TString lhcPeriod = GetLHCPeriod(); |
c88ad5db | 614 | if (lhcPeriod.Length() == 0) |
615 | { | |
616 | Log("SHUTTLE","StoreRunMetaDataFile - LHCPeriod not found in logbook!"); | |
617 | return 0; | |
618 | } | |
a1c5723b | 619 | |
620 | // TODO partitions with one detector only write data into LHCperiod_DET | |
2694f253 | 621 | TString partition = GetRunParameter("detector"); |
a1c5723b | 622 | |
623 | if (partition.Length() > 0 && partition != "ALICE") | |
624 | { | |
625 | lhcPeriod.Append(Form("_%s", partition.Data())); | |
626 | Log(fCurrentDetector, Form("Run data tags merged file will be written in %s", | |
627 | lhcPeriod.Data())); | |
628 | } | |
8884d71e | 629 | |
b8cf5e0f | 630 | TString target = Form("%s/GRP/RunMetadata%s%d/%s/%09d/raw/%s", |
631 | localBaseFolder.Data(), fConfig->GetAlienPath(), GetCurrentYear(), | |
d524ade6 | 632 | lhcPeriod.Data(), GetCurrentRun(), gridFileName); |
c88ad5db | 633 | |
d524ade6 | 634 | return CopyFileLocally(localFile, target); |
c88ad5db | 635 | } |
636 | ||
637 | //______________________________________________________________________________________________ | |
d524ade6 | 638 | Bool_t AliShuttle::CopyFileLocally(const char* localFile, const TString& target) |
c88ad5db | 639 | { |
640 | // | |
641 | // Stores file locally. Called by StoreReferenceFile and StoreRunMetadataFile | |
d524ade6 | 642 | // Files are temporarily stored in the local reference storage. When the preprocessor |
643 | // finishes, the Shuttle calls CopyFilesToGrid to transfer the files to AliEn | |
644 | // (in reference or run level folders) | |
c88ad5db | 645 | // |
646 | ||
d524ade6 | 647 | TString targetDir(target(0, target.Last('/'))); |
648 | ||
649 | //try to open base dir folder, if it does not exist | |
2d9019b4 | 650 | void* dir = gSystem->OpenDirectory(targetDir.Data()); |
651 | if (dir == NULL) { | |
652 | if (gSystem->mkdir(targetDir.Data(), kTRUE)) { | |
a1c5723b | 653 | Log("SHUTTLE", Form("CopyFileLocally - Can't open directory <%s>", targetDir.Data())); |
2d9019b4 | 654 | return kFALSE; |
655 | } | |
656 | ||
657 | } else { | |
658 | gSystem->FreeDirectory(dir); | |
659 | } | |
9827400b | 660 | |
7d43a416 | 661 | Int_t result = 0; |
662 | ||
663 | result = gSystem->GetPathInfo(localFile, 0, (Long64_t*) 0, 0, 0); | |
9827400b | 664 | if (result) |
665 | { | |
a1c5723b | 666 | Log("SHUTTLE", Form("CopyFileLocally - %s does not exist", localFile)); |
546242fb | 667 | return kFALSE; |
9827400b | 668 | } |
546242fb | 669 | |
7d43a416 | 670 | result = gSystem->GetPathInfo(target, 0, (Long64_t*) 0, 0, 0); |
671 | if (!result) | |
672 | { | |
a1c5723b | 673 | Log("SHUTTLE", Form("CopyFileLocally - target file %s already exist, removing...", target.Data())); |
7d43a416 | 674 | if (gSystem->Unlink(target.Data())) |
675 | { | |
a1c5723b | 676 | Log("SHUTTLE", Form("CopyFileLocally - Could not remove existing target file %s!", target.Data())); |
7d43a416 | 677 | return kFALSE; |
678 | } | |
679 | } | |
680 | ||
9827400b | 681 | result = gSystem->CopyFile(localFile, target); |
682 | ||
683 | if (result == 0) | |
684 | { | |
a1c5723b | 685 | Log("SHUTTLE", Form("CopyFileLocally - File %s stored locally to %s", localFile, target.Data())); |
9827400b | 686 | return kTRUE; |
687 | } | |
688 | else | |
689 | { | |
a1c5723b | 690 | Log("SHUTTLE", Form("CopyFileLocally - Could not store file %s to %s! Error code = %d", |
546242fb | 691 | localFile, target.Data(), result)); |
9827400b | 692 | return kFALSE; |
693 | } | |
c88ad5db | 694 | |
695 | ||
696 | ||
9827400b | 697 | } |
698 | ||
699 | //______________________________________________________________________________________________ | |
c88ad5db | 700 | Bool_t AliShuttle::CopyFilesToGrid(const char* type) |
9827400b | 701 | { |
702 | // | |
c88ad5db | 703 | // Transfers local files to the Grid. Local files can be reference files |
704 | // or run metadata file (from GRP only). | |
9827400b | 705 | // |
c88ad5db | 706 | // According to the type (ref, metadata) the files are stored under the following location: |
707 | // ref --> <base folder of reference storage>/<DET>/<RUN#>_<gridFileName> | |
708 | // metadata --> <run data folder>/<MetadataFileName> | |
86aa42c3 | 709 | // |
c88ad5db | 710 | |
9827400b | 711 | AliCDBManager* man = AliCDBManager::Instance(); |
712 | AliCDBStorage* sto = man->GetStorage(fgkLocalRefStorage); | |
713 | if (!sto) | |
714 | return kFALSE; | |
715 | TString localBaseFolder = sto->GetBaseFolder(); | |
9827400b | 716 | |
c88ad5db | 717 | TString dir; |
718 | TString alienDir; | |
9827400b | 719 | TString begin; |
9827400b | 720 | |
c88ad5db | 721 | if (strcmp(type, "reference") == 0) |
722 | { | |
723 | dir = GetRefFilePrefix(localBaseFolder.Data(), fCurrentDetector.Data()); | |
724 | AliCDBStorage* gridSto = man->GetStorage(fgkMainRefStorage); | |
725 | if (!gridSto) | |
726 | return kFALSE; | |
727 | TString gridBaseFolder = gridSto->GetBaseFolder(); | |
728 | alienDir = GetRefFilePrefix(gridBaseFolder.Data(), fCurrentDetector.Data()); | |
729 | begin = Form("%d_", GetCurrentRun()); | |
730 | } | |
731 | else if (strcmp(type, "metadata") == 0) | |
732 | { | |
c88ad5db | 733 | |
675f64cd | 734 | TString lhcPeriod = GetLHCPeriod(); |
c88ad5db | 735 | |
736 | if (lhcPeriod.Length() == 0) | |
737 | { | |
738 | Log("SHUTTLE","CopyFilesToGrid - LHCPeriod not found in logbook!"); | |
739 | return 0; | |
740 | } | |
741 | ||
a1c5723b | 742 | // TODO partitions with one detector only write data into LHCperiod_DET |
2694f253 | 743 | TString partition = GetRunParameter("detector"); |
a1c5723b | 744 | |
745 | if (partition.Length() > 0 && partition != "ALICE") | |
746 | { | |
747 | lhcPeriod.Append(Form("_%s", partition.Data())); | |
748 | } | |
749 | ||
b8cf5e0f | 750 | dir = Form("%s/GRP/RunMetadata%s%d/%s/%09d/raw", |
751 | localBaseFolder.Data(), fConfig->GetAlienPath(), GetCurrentYear(), | |
c88ad5db | 752 | lhcPeriod.Data(), GetCurrentRun()); |
b8cf5e0f | 753 | alienDir = dir(dir.Index(fConfig->GetAlienPath()), dir.Length()); |
675f64cd | 754 | |
c88ad5db | 755 | begin = ""; |
756 | } | |
757 | else | |
758 | { | |
759 | Log("SHUTTLE", "CopyFilesToGrid - Unexpected: type label must be reference or metadata!"); | |
760 | return kFALSE; | |
761 | } | |
762 | ||
9827400b | 763 | TSystemDirectory* baseDir = new TSystemDirectory("/", dir); |
3d8bc902 | 764 | if (!baseDir) |
765 | return kTRUE; | |
766 | ||
2d9019b4 | 767 | TList* dirList = baseDir->GetListOfFiles(); |
768 | delete baseDir; | |
769 | ||
770 | if (!dirList) return kTRUE; | |
771 | ||
772 | if (dirList->GetEntries() < 3) | |
3d8bc902 | 773 | { |
2d9019b4 | 774 | delete dirList; |
9827400b | 775 | return kTRUE; |
3d8bc902 | 776 | } |
2d9019b4 | 777 | |
546242fb | 778 | if (!gGrid) |
779 | { | |
c88ad5db | 780 | Log("SHUTTLE", "CopyFilesToGrid - Connection to Grid failed: Cannot continue!"); |
2d9019b4 | 781 | delete dirList; |
546242fb | 782 | return kFALSE; |
783 | } | |
784 | ||
2d9019b4 | 785 | Int_t nDirs = 0, nTransfer = 0; |
786 | TIter dirIter(dirList); | |
787 | TSystemFile* entry = 0; | |
788 | ||
9827400b | 789 | Bool_t success = kTRUE; |
3d8bc902 | 790 | Bool_t first = kTRUE; |
9827400b | 791 | |
2d9019b4 | 792 | while ((entry = dynamic_cast<TSystemFile*> (dirIter.Next()))) |
793 | { | |
9827400b | 794 | if (entry->IsDirectory()) |
795 | continue; | |
796 | ||
797 | TString fileName(entry->GetName()); | |
798 | if (!fileName.BeginsWith(begin)) | |
799 | continue; | |
800 | ||
2d9019b4 | 801 | nDirs++; |
802 | ||
3d8bc902 | 803 | if (first) |
804 | { | |
805 | first = kFALSE; | |
c88ad5db | 806 | // check that folder exists, otherwise create it |
3d8bc902 | 807 | TGridResult* result = gGrid->Ls(alienDir.Data(), "a"); |
808 | ||
809 | if (!result) | |
2d9019b4 | 810 | { |
811 | delete dirList; | |
3d8bc902 | 812 | return kFALSE; |
2d9019b4 | 813 | } |
3d8bc902 | 814 | |
546242fb | 815 | if (!result->GetFileName(1)) // TODO: It looks like element 0 is always 0!! |
3d8bc902 | 816 | { |
675f64cd | 817 | // TODO It does not work currently! Bug in TAliEn::Mkdir |
818 | // TODO Manually fixed in local root v5-16-00 | |
c88ad5db | 819 | if (!gGrid->Mkdir(alienDir.Data(),"-p",0)) |
3d8bc902 | 820 | { |
c88ad5db | 821 | Log("SHUTTLE", Form("CopyFilesToGrid - Cannot create directory %s", |
3d8bc902 | 822 | alienDir.Data())); |
2d9019b4 | 823 | delete dirList; |
3d8bc902 | 824 | return kFALSE; |
546242fb | 825 | } else { |
c88ad5db | 826 | Log("SHUTTLE",Form("CopyFilesToGrid - Folder %s created", alienDir.Data())); |
3d8bc902 | 827 | } |
828 | ||
546242fb | 829 | } else { |
c88ad5db | 830 | Log("SHUTTLE",Form("CopyFilesToGrid - Folder %s found", alienDir.Data())); |
3d8bc902 | 831 | } |
832 | } | |
833 | ||
9827400b | 834 | TString fullLocalPath; |
835 | fullLocalPath.Form("%s/%s", dir.Data(), fileName.Data()); | |
836 | ||
837 | TString fullGridPath; | |
838 | fullGridPath.Form("alien://%s/%s", alienDir.Data(), fileName.Data()); | |
839 | ||
a986b218 | 840 | Bool_t result = TFile::Cp(fullLocalPath, fullGridPath); |
9827400b | 841 | |
842 | if (result) | |
843 | { | |
c88ad5db | 844 | Log("SHUTTLE", Form("CopyFilesToGrid - Copying local file %s to %s succeeded!", |
845 | fullLocalPath.Data(), fullGridPath.Data())); | |
9827400b | 846 | RemoveFile(fullLocalPath); |
2d9019b4 | 847 | nTransfer++; |
9827400b | 848 | } |
849 | else | |
850 | { | |
c88ad5db | 851 | Log("SHUTTLE", Form("CopyFilesToGrid - Copying local file %s to %s FAILED!", |
852 | fullLocalPath.Data(), fullGridPath.Data())); | |
9827400b | 853 | success = kFALSE; |
854 | } | |
855 | } | |
2d9019b4 | 856 | |
c88ad5db | 857 | Log("SHUTTLE", Form("CopyFilesToGrid - %d (over %d) files in folder %s copied to Grid.", |
858 | nTransfer, nDirs, dir.Data())); | |
2d9019b4 | 859 | |
860 | ||
861 | delete dirList; | |
9827400b | 862 | return success; |
863 | } | |
864 | ||
2d9019b4 | 865 | //______________________________________________________________________________________________ |
866 | const char* AliShuttle::GetRefFilePrefix(const char* base, const char* detector) | |
867 | { | |
868 | // | |
869 | // Get folder name of reference files | |
870 | // | |
871 | ||
872 | TString offDetStr(GetOfflineDetName(detector)); | |
3dd835d8 | 873 | static TString dir; |
2d9019b4 | 874 | if (offDetStr == "ITS" || offDetStr == "MUON" || offDetStr == "PHOS") |
875 | { | |
876 | dir.Form("%s/%s/%s", base, offDetStr.Data(), detector); | |
877 | } else { | |
878 | dir.Form("%s/%s", base, offDetStr.Data()); | |
879 | } | |
880 | ||
881 | return dir.Data(); | |
2d9019b4 | 882 | } |
c88ad5db | 883 | |
3301427a | 884 | //______________________________________________________________________________________________ |
885 | void AliShuttle::CleanLocalStorage(const TString& uri) | |
886 | { | |
9827400b | 887 | // |
888 | // Called in case the preprocessor is declared failed. Remove remaining objects from the local storages. | |
889 | // | |
3301427a | 890 | |
891 | const char* type = 0; | |
892 | if(uri == fgkLocalCDB) { | |
893 | type = "OCDB"; | |
894 | } else if(uri == fgkLocalRefStorage) { | |
546242fb | 895 | type = "Reference"; |
3301427a | 896 | } else { |
897 | AliError(Form("Invalid storage URI: %s", uri.Data())); | |
898 | return; | |
899 | } | |
900 | ||
901 | AliCDBManager* man = AliCDBManager::Instance(); | |
b948db8d | 902 | |
3301427a | 903 | // open local storage |
904 | AliCDBStorage *localSto = man->GetStorage(uri); | |
905 | if(!localSto) { | |
906 | Log("SHUTTLE", | |
907 | Form("CleanLocalStorage - cannot activate local %s storage", type)); | |
908 | return; | |
909 | } | |
910 | ||
911 | TString filename(Form("%s/%s/*/Run*_v%d_s*.root", | |
546242fb | 912 | localSto->GetBaseFolder().Data(), GetOfflineDetName(fCurrentDetector.Data()), GetCurrentRun())); |
3301427a | 913 | |
c88ad5db | 914 | AliDebug(2, Form("filename = %s", filename.Data())); |
3301427a | 915 | |
c88ad5db | 916 | Log("SHUTTLE", Form("Removing remaining local files for run %d and detector %s ...", |
3301427a | 917 | GetCurrentRun(), fCurrentDetector.Data())); |
918 | ||
919 | RemoveFile(filename.Data()); | |
920 | ||
921 | } | |
922 | ||
923 | //______________________________________________________________________________________________ | |
924 | void AliShuttle::RemoveFile(const char* filename) | |
925 | { | |
9827400b | 926 | // |
927 | // removes local file | |
928 | // | |
3301427a | 929 | |
930 | TString command(Form("rm -f %s", filename)); | |
931 | ||
932 | Int_t result = gSystem->Exec(command.Data()); | |
933 | if(result != 0) | |
934 | { | |
935 | Log("SHUTTLE", Form("RemoveFile - %s: Cannot remove file %s!", | |
936 | fCurrentDetector.Data(), filename)); | |
937 | } | |
73abe331 | 938 | } |
939 | ||
b948db8d | 940 | //______________________________________________________________________________________________ |
5164a766 | 941 | AliShuttleStatus* AliShuttle::ReadShuttleStatus() |
942 | { | |
9827400b | 943 | // |
944 | // Reads the AliShuttleStatus from the CDB | |
945 | // | |
5164a766 | 946 | |
2bb7b766 | 947 | if (fStatusEntry){ |
948 | delete fStatusEntry; | |
949 | fStatusEntry = 0; | |
950 | } | |
5164a766 | 951 | |
3ebbf3c4 | 952 | Int_t path1 = GetCurrentRun()/10000; |
10a5a932 | 953 | fStatusEntry = AliCDBManager::Instance()->GetStorage(GetLocalCDB()) |
3ebbf3c4 | 954 | ->Get(Form("/SHUTTLE/%s/%d", fCurrentDetector.Data(), path1), GetCurrentRun()); |
5164a766 | 955 | |
2bb7b766 | 956 | if (!fStatusEntry) return 0; |
957 | fStatusEntry->SetOwner(1); | |
5164a766 | 958 | |
2bb7b766 | 959 | AliShuttleStatus* status = dynamic_cast<AliShuttleStatus*> (fStatusEntry->GetObject()); |
960 | if (!status) { | |
961 | AliError("Invalid object stored to CDB!"); | |
962 | return 0; | |
963 | } | |
5164a766 | 964 | |
2bb7b766 | 965 | return status; |
5164a766 | 966 | } |
967 | ||
968 | //______________________________________________________________________________________________ | |
7bfb2090 | 969 | Bool_t AliShuttle::WriteShuttleStatus(AliShuttleStatus* status) |
5164a766 | 970 | { |
9827400b | 971 | // |
972 | // writes the status for one subdetector | |
973 | // | |
2bb7b766 | 974 | |
975 | if (fStatusEntry){ | |
976 | delete fStatusEntry; | |
977 | fStatusEntry = 0; | |
978 | } | |
5164a766 | 979 | |
2bb7b766 | 980 | Int_t run = GetCurrentRun(); |
3ebbf3c4 | 981 | Int_t path1 = run/10000; |
982 | TString path1_string = Form("%d",path1); | |
5164a766 | 983 | |
3ebbf3c4 | 984 | AliCDBId id(AliCDBPath("SHUTTLE", fCurrentDetector, path1_string), run, run); |
5164a766 | 985 | |
2bb7b766 | 986 | fStatusEntry = new AliCDBEntry(status, id, new AliCDBMetaData); |
987 | fStatusEntry->SetOwner(1); | |
5164a766 | 988 | |
924fb6e9 | 989 | Int_t logLevel = AliLog::GetGlobalLogLevel(); |
990 | AliLog::SetGlobalLogLevel(AliLog::kError); | |
991 | ||
2bb7b766 | 992 | UInt_t result = AliCDBManager::Instance()->GetStorage(fgkLocalCDB)->Put(fStatusEntry); |
7bfb2090 | 993 | |
924fb6e9 | 994 | AliLog::SetGlobalLogLevel((AliLog::EType_t)logLevel); |
995 | ||
996 | if (!result) { | |
3301427a | 997 | Log("SHUTTLE", Form("WriteShuttleStatus - Failed for %s, run %d", |
998 | fCurrentDetector.Data(), run)); | |
2bb7b766 | 999 | return kFALSE; |
1000 | } | |
e7f62f16 | 1001 | |
8e828d39 | 1002 | SendMLDetInfo(); |
7bfb2090 | 1003 | |
2bb7b766 | 1004 | return kTRUE; |
5164a766 | 1005 | } |
1006 | ||
1007 | //______________________________________________________________________________________________ | |
1008 | void AliShuttle::UpdateShuttleStatus(AliShuttleStatus::Status newStatus, Bool_t increaseCount) | |
1009 | { | |
9827400b | 1010 | // |
1011 | // changes the AliShuttleStatus for the given detector and run to the given status | |
1012 | // | |
5164a766 | 1013 | |
2bb7b766 | 1014 | if (!fStatusEntry){ |
1015 | AliError("UNEXPECTED: fStatusEntry empty"); | |
1016 | return; | |
1017 | } | |
5164a766 | 1018 | |
2bb7b766 | 1019 | AliShuttleStatus* status = dynamic_cast<AliShuttleStatus*> (fStatusEntry->GetObject()); |
5164a766 | 1020 | |
2bb7b766 | 1021 | if (!status){ |
c88ad5db | 1022 | Log("SHUTTLE", "UpdateShuttleStatus - UNEXPECTED: status could not be read from current CDB entry"); |
2bb7b766 | 1023 | return; |
1024 | } | |
5164a766 | 1025 | |
2c15234c | 1026 | TString actionStr = Form("UpdateShuttleStatus - %s: Changing state from %s to %s", |
eba76848 | 1027 | fCurrentDetector.Data(), |
36c99a6a | 1028 | status->GetStatusName(), |
eba76848 | 1029 | status->GetStatusName(newStatus)); |
cb343cfd | 1030 | Log("SHUTTLE", actionStr); |
1031 | SetLastAction(actionStr); | |
5164a766 | 1032 | |
2bb7b766 | 1033 | status->SetStatus(newStatus); |
1034 | if (increaseCount) status->IncreaseCount(); | |
5164a766 | 1035 | |
924fb6e9 | 1036 | Int_t logLevel = AliLog::GetGlobalLogLevel(); |
1037 | AliLog::SetGlobalLogLevel(AliLog::kError); | |
1038 | ||
2bb7b766 | 1039 | AliCDBManager::Instance()->GetStorage(fgkLocalCDB)->Put(fStatusEntry); |
e7f62f16 | 1040 | |
924fb6e9 | 1041 | AliLog::SetGlobalLogLevel((AliLog::EType_t)logLevel); |
1042 | ||
8e828d39 | 1043 | SendMLDetInfo(); |
5164a766 | 1044 | } |
e7f62f16 | 1045 | |
1046 | //______________________________________________________________________________________________ | |
8e828d39 | 1047 | void AliShuttle::SendMLDetInfo() |
e7f62f16 | 1048 | { |
1049 | // | |
1050 | // sends ML information about the current status of the current detector being processed | |
1051 | // | |
1052 | ||
1053 | AliShuttleStatus* status = dynamic_cast<AliShuttleStatus*> (fStatusEntry->GetObject()); | |
1054 | ||
1055 | if (!status){ | |
8e828d39 | 1056 | Log("SHUTTLE", "SendMLDetInfo - UNEXPECTED: status could not be read from current CDB entry"); |
e7f62f16 | 1057 | return; |
1058 | } | |
1059 | ||
1060 | TMonaLisaText mlStatus(Form("%s_status", fCurrentDetector.Data()), status->GetStatusName()); | |
1061 | TMonaLisaValue mlRetryCount(Form("%s_count", fCurrentDetector.Data()), status->GetCount()); | |
1062 | ||
1063 | TList mlList; | |
1064 | mlList.Add(&mlStatus); | |
1065 | mlList.Add(&mlRetryCount); | |
1066 | ||
ee6f7523 | 1067 | TString mlID; |
1068 | mlID.Form("%d", GetCurrentRun()); | |
1069 | fMonaLisa->SendParameters(&mlList, mlID); | |
e7f62f16 | 1070 | } |
1071 | ||
5164a766 | 1072 | //______________________________________________________________________________________________ |
1073 | Bool_t AliShuttle::ContinueProcessing() | |
1074 | { | |
9827400b | 1075 | // this function reads the AliShuttleStatus information from CDB and |
1076 | // checks if the processing should be continued | |
1077 | // if yes it returns kTRUE and updates the AliShuttleStatus with nextStatus | |
2bb7b766 | 1078 | |
926cbc0e | 1079 | if (!fConfig->HostProcessDetector(fCurrentDetector)) |
1080 | return kFALSE; | |
57c1a579 | 1081 | |
1082 | AliPreprocessor* aPreprocessor = | |
1083 | dynamic_cast<AliPreprocessor*> (fPreprocessorMap.GetValue(fCurrentDetector)); | |
1084 | if (!aPreprocessor) | |
1085 | { | |
c88ad5db | 1086 | Log("SHUTTLE", Form("ContinueProcessing - %s: no preprocessor registered", fCurrentDetector.Data())); |
57c1a579 | 1087 | return kFALSE; |
1088 | } | |
1089 | ||
2bb7b766 | 1090 | AliShuttleLogbookEntry::Status entryStatus = |
eba76848 | 1091 | fLogbookEntry->GetDetectorStatus(fCurrentDetector); |
2bb7b766 | 1092 | |
5e993b6f | 1093 | if (entryStatus != AliShuttleLogbookEntry::kUnprocessed) { |
c88ad5db | 1094 | Log("SHUTTLE", Form("ContinueProcessing - %s is %s", |
2bb7b766 | 1095 | fCurrentDetector.Data(), |
1096 | fLogbookEntry->GetDetectorStatusName(entryStatus))); | |
1097 | return kFALSE; | |
1098 | } | |
1099 | ||
1100 | // if we get here, according to Shuttle logbook subdetector is in UNPROCESSED state | |
be48e3ea | 1101 | |
1102 | // check if current run is first unprocessed run for current detector | |
1103 | if (fConfig->StrictRunOrder(fCurrentDetector) && | |
1104 | !fFirstUnprocessed[GetDetPos(fCurrentDetector)]) | |
1105 | { | |
86aa42c3 | 1106 | if (fTestMode == kNone) |
1107 | { | |
c88ad5db | 1108 | Log("SHUTTLE", Form("ContinueProcessing - %s requires strict run ordering" |
0a0b087a | 1109 | " but this is not the first unprocessed run!",fCurrentDetector.Data())); |
86aa42c3 | 1110 | return kFALSE; |
1111 | } | |
1112 | else | |
1113 | { | |
c88ad5db | 1114 | Log("SHUTTLE", Form("ContinueProcessing - In TESTMODE - " |
1115 | "Although %s requires strict run ordering " | |
1116 | "and this is not the first unprocessed run, " | |
0a0b087a | 1117 | "the SHUTTLE continues",fCurrentDetector.Data())); |
86aa42c3 | 1118 | } |
be48e3ea | 1119 | } |
1120 | ||
926cbc0e | 1121 | // Is the subdetector processed first time for this run? |
1122 | fFirstProcessing = kFALSE; | |
1123 | ||
2bb7b766 | 1124 | AliShuttleStatus* status = ReadShuttleStatus(); |
1125 | if (!status) { | |
1126 | // first time | |
1127 | Log("SHUTTLE", Form("ContinueProcessing - %s: Processing first time", | |
1128 | fCurrentDetector.Data())); | |
1129 | status = new AliShuttleStatus(AliShuttleStatus::kStarted); | |
926cbc0e | 1130 | fFirstProcessing = kTRUE; |
2bb7b766 | 1131 | return WriteShuttleStatus(status); |
1132 | } | |
1133 | ||
ffa25f20 | 1134 | // The following case shouldn't happen if Shuttle Logbook was correctly updated. |
2bb7b766 | 1135 | // If it happens it may mean Logbook updating failed... let's do it now! |
1136 | if (status->GetStatus() == AliShuttleStatus::kDone || | |
ffa25f20 | 1137 | status->GetStatus() == AliShuttleStatus::kFailed || |
1138 | status->GetStatus() == AliShuttleStatus::kSkipped) { | |
2bb7b766 | 1139 | Log("SHUTTLE", Form("ContinueProcessing - %s is already %s. Updating Shuttle Logbook", |
1140 | fCurrentDetector.Data(), | |
1141 | status->GetStatusName(status->GetStatus()))); | |
ffa25f20 | 1142 | |
1143 | if (status->GetStatus() == AliShuttleStatus::kSkipped) | |
1144 | { | |
1145 | UpdateShuttleLogbook(fCurrentDetector.Data(), "DONE"); | |
1146 | } | |
1147 | else | |
1148 | UpdateShuttleLogbook(fCurrentDetector.Data(), status->GetStatusName(status->GetStatus())); | |
1149 | ||
2bb7b766 | 1150 | return kFALSE; |
1151 | } | |
1152 | ||
5e993b6f | 1153 | if (status->GetStatus() == AliShuttleStatus::kStoreStarted || status->GetStatus() == AliShuttleStatus::kStoreDelayed ||status->GetStatus() == AliShuttleStatus::kStoreError) { |
2bb7b766 | 1154 | Log("SHUTTLE", |
c88ad5db | 1155 | Form("ContinueProcessing - %s: Grid storage of one or more " |
1156 | "objects failed. Trying again now", | |
2bb7b766 | 1157 | fCurrentDetector.Data())); |
5e993b6f | 1158 | StoreOCDB(); |
2bb7b766 | 1159 | return kFALSE; |
1160 | } | |
1161 | ||
1162 | // if we get here, there is a restart | |
57c1a579 | 1163 | Bool_t cont = kFALSE; |
2bb7b766 | 1164 | |
1165 | // abort conditions | |
cb343cfd | 1166 | if (status->GetCount() >= fConfig->GetMaxRetries()) { |
57c1a579 | 1167 | Log("SHUTTLE", Form("ContinueProcessing - %s failed %d times in status %s - " |
1168 | "Updating Shuttle Logbook", fCurrentDetector.Data(), | |
2bb7b766 | 1169 | status->GetCount(), status->GetStatusName())); |
1170 | UpdateShuttleLogbook(fCurrentDetector.Data(), "FAILED"); | |
e7f62f16 | 1171 | UpdateShuttleStatus(AliShuttleStatus::kFailed); |
3301427a | 1172 | |
1173 | // there may still be objects in local OCDB and reference storage | |
1174 | // and FXS databases may be not updated: do it now! | |
9827400b | 1175 | |
1176 | // TODO Currently disabled, we want to keep files in case of failure! | |
1177 | // CleanLocalStorage(fgkLocalCDB); | |
1178 | // CleanLocalStorage(fgkLocalRefStorage); | |
1179 | // UpdateTableFailCase(); | |
1180 | ||
1181 | // Send mail to detector expert! | |
c88ad5db | 1182 | Log("SHUTTLE", Form("ContinueProcessing - Sending mail to %s expert...", |
926cbc0e | 1183 | fCurrentDetector.Data())); |
fe6a3257 | 1184 | // det experts in to |
1185 | TString to=""; | |
1186 | TIter *iterExperts = 0; | |
1187 | iterExperts = new TIter(fConfig->GetResponsibles(fCurrentDetector)); | |
1188 | TObjString *anExpert=0; | |
1189 | while ((anExpert = (TObjString*) iterExperts->Next())) | |
1190 | { | |
1191 | to += Form("%s, \n", anExpert->GetName()); | |
1192 | } | |
1193 | delete iterExperts; | |
1194 | ||
1195 | if (to.Length() > 0) | |
1196 | to.Remove(to.Length()-3); | |
1197 | AliDebug(2, Form("to: %s",to.Data())); | |
1198 | ||
1199 | if (to.IsNull()) { | |
1200 | Log("SHUTTLE", Form("List of %s responsibles not set!", fCurrentDetector.Data())); | |
1201 | return kFALSE; | |
1202 | } | |
1203 | ||
1204 | Log(fCurrentDetector.Data(), Form("ContinueProcessing - Sending mail to %s expert(s):", | |
1205 | fCurrentDetector.Data())); | |
1206 | Log(fCurrentDetector.Data(), Form("\n%s", to.Data())); | |
339adafa | 1207 | if (!SendMail(kPPEMail)) |
9827400b | 1208 | Log("SHUTTLE", Form("ContinueProcessing - Could not send mail to %s expert", |
926cbc0e | 1209 | fCurrentDetector.Data())); |
3301427a | 1210 | |
57c1a579 | 1211 | } else { |
1212 | Log("SHUTTLE", Form("ContinueProcessing - %s: restarting. " | |
1213 | "Aborted before with %s. Retry number %d.", fCurrentDetector.Data(), | |
1214 | status->GetStatusName(), status->GetCount())); | |
9827400b | 1215 | Bool_t increaseCount = kTRUE; |
c88ad5db | 1216 | if (status->GetStatus() == AliShuttleStatus::kDCSError || |
8da81210 | 1217 | status->GetStatus() == AliShuttleStatus::kDCSStarted || |
5ca8f352 | 1218 | status->GetStatus() == AliShuttleStatus::kFXSError || |
1219 | status->GetStatus() == AliShuttleStatus::kOCDBError) | |
c88ad5db | 1220 | increaseCount = kFALSE; |
675f64cd | 1221 | |
9827400b | 1222 | UpdateShuttleStatus(AliShuttleStatus::kStarted, increaseCount); |
57c1a579 | 1223 | cont = kTRUE; |
2bb7b766 | 1224 | } |
1225 | ||
57c1a579 | 1226 | return cont; |
5164a766 | 1227 | } |
1228 | ||
8e828d39 | 1229 | //______________________________________________________________________________________________ |
1230 | void AliShuttle::SendMLRunInfo(const char* status) | |
1231 | { | |
1232 | // | |
1233 | // Send information about this run to ML | |
1234 | ||
1235 | TMonaLisaText mlStatus("SHUTTLE_status", status); | |
1236 | TString runType(fLogbookEntry->GetRunType()); | |
1237 | if (strlen(fLogbookEntry->GetRunParameter("log")) > 0){ | |
1238 | ||
1239 | runType += "("; | |
1240 | runType += fLogbookEntry->GetRunParameter("log"); | |
1241 | runType += ")"; | |
1242 | } | |
728290a4 | 1243 | if (fLogbookEntry->GetDATestMode()){ |
1244 | runType += " (DATest)"; | |
1245 | } | |
8e828d39 | 1246 | TMonaLisaText mlRunType("SHUTTLE_runtype", runType); |
1247 | ||
1248 | TList mlList; | |
1249 | mlList.Add(&mlStatus); | |
1250 | mlList.Add(&mlRunType); | |
1251 | ||
1252 | TString mlID; | |
1253 | mlID.Form("%d", GetCurrentRun()); | |
1254 | fMonaLisa->SendParameters(&mlList, mlID); | |
1255 | } | |
1256 | ||
ef83773f | 1257 | //______________________________________________________________________________________________ |
1258 | Int_t AliShuttle::GetMem(Int_t pid) | |
1259 | { | |
1260 | // invokes ps to get the memory consumption of the process <pid> | |
1261 | // returns -1 in case of error | |
1262 | ||
1263 | TString checkStr; | |
1264 | checkStr.Form("ps -o vsize --pid %d | tail -n 1", pid); | |
1265 | FILE* pipe = gSystem->OpenPipe(checkStr, "r"); | |
1266 | if (!pipe) | |
1267 | { | |
1268 | Log("SHUTTLE", Form("Process - Error: " | |
1269 | "Could not open pipe to %s", checkStr.Data())); | |
1270 | return -1; | |
1271 | } | |
1272 | ||
1273 | char buffer[100]; | |
1274 | if (!fgets(buffer, 100, pipe)) | |
1275 | { | |
1276 | Log("SHUTTLE", "Process - Error: ps did not return anything"); | |
1277 | gSystem->ClosePipe(pipe); | |
1278 | return -1; | |
1279 | } | |
1280 | gSystem->ClosePipe(pipe); | |
1281 | ||
1282 | //Log("SHUTTLE", Form("ps returned %s", buffer)); | |
1283 | ||
1284 | Int_t mem = 0; | |
1285 | if ((sscanf(buffer, "%d\n", &mem) != 1) || !mem) | |
1286 | { | |
1287 | Log("SHUTTLE", "Process - Error: Could not parse output of ps"); | |
1288 | return -1; | |
1289 | } | |
1290 | ||
1291 | return mem; | |
1292 | } | |
1293 | ||
5164a766 | 1294 | //______________________________________________________________________________________________ |
2bb7b766 | 1295 | Bool_t AliShuttle::Process(AliShuttleLogbookEntry* entry) |
58bc3020 | 1296 | { |
73abe331 | 1297 | // |
b948db8d | 1298 | // Makes data retrieval for all detectors in the configuration. |
2bb7b766 | 1299 | // entry: Shuttle logbook entry, contains run paramenters and status of detectors |
1300 | // (Unprocessed, Inactive, Failed or Done). | |
d477ad88 | 1301 | // Returns kFALSE in case of error occured and kTRUE otherwise |
73abe331 | 1302 | // |
1303 | ||
9827400b | 1304 | if (!entry) return kFALSE; |
2bb7b766 | 1305 | |
1306 | fLogbookEntry = entry; | |
1307 | ||
c88ad5db | 1308 | Log("SHUTTLE", Form("\t\t\t^*^*^*^*^*^*^*^*^*^*^*^* run %d: START ^*^*^*^*^*^*^*^*^*^*^*^*", |
9827400b | 1309 | GetCurrentRun())); |
2bb7b766 | 1310 | |
339adafa | 1311 | CountOpenRuns(); |
1312 | ||
8e828d39 | 1313 | // Send the information to ML |
1314 | SendMLRunInfo("Processing"); | |
3301427a | 1315 | |
9827400b | 1316 | if (fLogbookEntry->IsDone()) |
1317 | { | |
1318 | Log("SHUTTLE","Process - Shuttle is already DONE. Updating logbook"); | |
1319 | UpdateShuttleLogbook("shuttle_done"); | |
1320 | fLogbookEntry = 0; | |
1321 | return kTRUE; | |
1322 | } | |
1323 | ||
1324 | // read test mode if flag is set | |
1325 | if (fReadTestMode) | |
1326 | { | |
3d8bc902 | 1327 | fTestMode = kNone; |
9827400b | 1328 | TString logEntry(entry->GetRunParameter("log")); |
1329 | //printf("log entry = %s\n", logEntry.Data()); | |
1330 | TString searchStr("Testmode: "); | |
1331 | Int_t pos = logEntry.Index(searchStr.Data()); | |
1332 | //printf("%d\n", pos); | |
1333 | if (pos >= 0) | |
1334 | { | |
1335 | TSubString subStr = logEntry(pos + searchStr.Length(), logEntry.Length()); | |
1336 | //printf("%s\n", subStr.String().Data()); | |
1337 | TString newStr(subStr.Data()); | |
1338 | TObjArray* token = newStr.Tokenize(' '); | |
1339 | if (token) | |
1340 | { | |
1341 | //token->Print(); | |
1342 | TObjString* tmpStr = dynamic_cast<TObjString*> (token->First()); | |
1343 | if (tmpStr) | |
1344 | { | |
1345 | Int_t testMode = tmpStr->String().Atoi(); | |
1346 | if (testMode > 0) | |
1347 | { | |
c88ad5db | 1348 | Log("SHUTTLE", Form("Process - Enabling test mode %d", testMode)); |
9827400b | 1349 | SetTestMode((TestMode) testMode); |
1350 | } | |
1351 | } | |
1352 | delete token; | |
1353 | } | |
1354 | } | |
1355 | } | |
c88ad5db | 1356 | |
eba76848 | 1357 | fLogbookEntry->Print("all"); |
57f50b3c | 1358 | |
1359 | // Initialization | |
d477ad88 | 1360 | Bool_t hasError = kFALSE; |
5164a766 | 1361 | |
728290a4 | 1362 | // Set the CDB and Reference folders according to the year |
1363 | ||
1364 | // build cdb paths (repeat each time, run might be a DATest run) | |
1365 | if (!fLogbookEntry->GetDATestMode()){ | |
1366 | fgkMainCDB.Form("alien://folder=%s%d/OCDB?user=alidaq?cacheFold=/tmp/OCDBCache", | |
1367 | fConfig->GetAlienPath(), GetCurrentYear()); | |
1368 | ||
1369 | fgkMainRefStorage.Form("alien://folder=%s%d/Reference?user=alidaq?cacheFold=/tmp/OCDBCache", | |
1370 | fConfig->GetAlienPath(), GetCurrentYear()); | |
1371 | } | |
1372 | else { | |
1373 | fgkMainCDB.Form("alien://folder=%s%d/DATest/OCDB?user=alidaq?cacheFold=/tmp/OCDBCache", | |
1374 | fConfig->GetAlienPath(), GetCurrentYear()); | |
1375 | ||
1376 | fgkMainRefStorage.Form("alien://folder=%s%d/DATest/Reference?user=alidaq?cacheFold=/tmp/OCDBCache", | |
1377 | fConfig->GetAlienPath(), GetCurrentYear()); | |
1378 | } | |
1379 | ||
1380 | AliDebug(2,Form("Main CDB storage = %s",fgkMainCDB.Data())); | |
1381 | AliDebug(2,Form("Main Reference storage = %s",fgkMainRefStorage.Data())); | |
1382 | ||
57f50b3c | 1383 | // Loop on detectors in the configuration |
b948db8d | 1384 | TIter iter(fConfig->GetDetectors()); |
2bb7b766 | 1385 | TObjString* aDetector = 0; |
b948db8d | 1386 | |
4508ab8b | 1387 | Bool_t first = kTRUE; |
1388 | ||
be48e3ea | 1389 | while ((aDetector = (TObjString*) iter.Next())) |
1390 | { | |
7bfb2090 | 1391 | fCurrentDetector = aDetector->String(); |
5164a766 | 1392 | |
5e993b6f | 1393 | if (ContinueProcessing() == kFALSE) |
1394 | continue; | |
4508ab8b | 1395 | |
1396 | if (first) | |
1397 | { | |
1398 | // only read QueryCDB when needed and only once | |
1399 | AliCDBStorage *mainCDBSto = AliCDBManager::Instance()->GetStorage(fgkMainCDB); | |
1400 | if(mainCDBSto) mainCDBSto->QueryCDB(GetCurrentRun()); | |
1401 | AliCDBStorage *mainRefSto = AliCDBManager::Instance()->GetStorage(fgkMainRefStorage); | |
1402 | if(mainRefSto) mainRefSto->QueryCDB(GetCurrentRun()); | |
1403 | first = kFALSE; | |
1404 | } | |
9e080f92 | 1405 | |
c88ad5db | 1406 | Log("SHUTTLE", Form("\t\t\t****** run %d - %s: START ******", |
2bb7b766 | 1407 | GetCurrentRun(), aDetector->GetName())); |
1408 | ||
9d733021 | 1409 | for(Int_t iSys=0;iSys<3;iSys++) fFXSCalled[iSys]=kFALSE; |
ef83773f | 1410 | |
1411 | Int_t initialMem = GetMem(getpid()); | |
1412 | Log("SHUTTLE", Form("Memory consumption before forking is %d", initialMem)); | |
9d733021 | 1413 | |
c88ad5db | 1414 | Log(fCurrentDetector.Data(), "Process - Starting processing"); |
85a80aa9 | 1415 | |
be48e3ea | 1416 | Int_t pid = fork(); |
1417 | ||
1418 | if (pid < 0) | |
1419 | { | |
c88ad5db | 1420 | Log("SHUTTLE", "Process - ERROR: Forking failed"); |
be48e3ea | 1421 | } |
1422 | else if (pid > 0) | |
1423 | { | |
1424 | // parent | |
c88ad5db | 1425 | Log("SHUTTLE", Form("Process - In parent process of %d - %s: Starting monitoring", |
be48e3ea | 1426 | GetCurrentRun(), aDetector->GetName())); |
1427 | ||
1428 | Long_t begin = time(0); | |
1429 | ||
1430 | int status; // to be used with waitpid, on purpose an int (not Int_t)! | |
1431 | while (waitpid(pid, &status, WNOHANG) == 0) | |
1432 | { | |
1433 | Long_t expiredTime = time(0) - begin; | |
1434 | ||
1435 | if (expiredTime > fConfig->GetPPTimeOut()) | |
1436 | { | |
56d3e9e4 | 1437 | TString logMsg; |
1438 | AliShuttleStatus *currentStatus = ReadShuttleStatus(); | |
1439 | AliShuttleStatus::Status newStatus = AliShuttleStatus::kInvalid; | |
1440 | ||
838d9827 | 1441 | if (currentStatus->GetStatus() == AliShuttleStatus::kDCSStarted) |
1442 | { | |
1443 | // in case the pp goes in TimeOut while retrieving the DCS DPs | |
1444 | // set status to kDCSError | |
1445 | ||
0a0b087a | 1446 | logMsg.Form("Process - Process of %s timed out while retrieving the DCS DataPoints. Run time: %ld seconds. Killing... and setting status to DCSError.", |
838d9827 | 1447 | fCurrentDetector.Data(), expiredTime); |
1448 | newStatus = AliShuttleStatus::kDCSError; | |
1449 | } | |
1450 | else if (currentStatus->GetStatus() <= AliShuttleStatus::kPPDone) | |
56d3e9e4 | 1451 | { |
1452 | // in case pp not yet done set status to kPPTimeOut | |
1453 | ||
0a0b087a | 1454 | logMsg.Form("Process - Process of %s timed out. Run time: %ld seconds. Killing...", |
56d3e9e4 | 1455 | fCurrentDetector.Data(), expiredTime); |
1456 | newStatus = AliShuttleStatus::kPPTimeOut; | |
1457 | } | |
1458 | else if (currentStatus->GetStatus() == AliShuttleStatus::kStoreStarted) | |
1459 | { | |
1460 | // in case the pp goes in TimeOut while storing the objects in the OCDB | |
1461 | // set status to kStoreError | |
1462 | ||
0a0b087a | 1463 | logMsg.Form("Process - Process of %s timed out while storing the OCDB object. Run time: %ld seconds. Killing... and setting status to StoreError.", |
56d3e9e4 | 1464 | fCurrentDetector.Data(), expiredTime); |
1465 | newStatus = AliShuttleStatus::kStoreError; | |
1466 | } | |
1467 | else | |
1468 | { | |
1469 | // in other cases don't change the status | |
1470 | ||
0a0b087a | 1471 | logMsg.Form("Process - Process of %s timed out in status = %s. Run time: %ld seconds. Killing... without changing the status", |
56d3e9e4 | 1472 | fCurrentDetector.Data(), currentStatus->GetStatusName(), expiredTime); |
1473 | } | |
1474 | ||
1475 | Log("SHUTTLE", logMsg); | |
1476 | Log(fCurrentDetector, logMsg); | |
be48e3ea | 1477 | |
1478 | kill(pid, 9); | |
1479 | ||
56d3e9e4 | 1480 | if (newStatus != AliShuttleStatus::kInvalid) |
1481 | UpdateShuttleStatus(newStatus); | |
be48e3ea | 1482 | hasError = kTRUE; |
1483 | ||
1484 | gSystem->Sleep(1000); | |
1485 | } | |
1486 | else | |
1487 | { | |
be48e3ea | 1488 | gSystem->Sleep(1000); |
9827400b | 1489 | |
ef83773f | 1490 | Int_t mem = GetMem(pid); |
1491 | ||
1492 | if (mem < 0) | |
9827400b | 1493 | continue; |
9827400b | 1494 | |
ef83773f | 1495 | mem -= initialMem; |
1496 | if (mem < 0) | |
1497 | mem = 0; | |
9827400b | 1498 | |
1499 | if (expiredTime % 60 == 0) | |
ee6f7523 | 1500 | { |
c88ad5db | 1501 | Log("SHUTTLE", Form("Process - %s: Checking process. " |
0a0b087a | 1502 | "Run time: %ld seconds - Memory consumption: %d KB", |
c88ad5db | 1503 | fCurrentDetector.Data(), expiredTime, mem)); |
ee6f7523 | 1504 | SendAlive(); |
1505 | } | |
9827400b | 1506 | |
1507 | if (mem > fConfig->GetPPMaxMem()) | |
1508 | { | |
1509 | TString tmp; | |
c88ad5db | 1510 | tmp.Form("Process - Process exceeds maximum allowed memory " |
1511 | "(%d KB > %d KB). Killing...", | |
9827400b | 1512 | mem, fConfig->GetPPMaxMem()); |
1513 | Log("SHUTTLE", tmp); | |
1514 | Log(fCurrentDetector, tmp); | |
1515 | ||
1516 | kill(pid, 9); | |
1517 | ||
1518 | UpdateShuttleStatus(AliShuttleStatus::kPPOutOfMemory); | |
1519 | hasError = kTRUE; | |
1520 | ||
1521 | gSystem->Sleep(1000); | |
1522 | } | |
be48e3ea | 1523 | } |
1524 | } | |
1525 | ||
c88ad5db | 1526 | Log("SHUTTLE", Form("Process - In parent process of %d - %s: Client has terminated.", |
be48e3ea | 1527 | GetCurrentRun(), aDetector->GetName())); |
1528 | ||
1529 | if (WIFEXITED(status)) | |
1530 | { | |
1531 | Int_t returnCode = WEXITSTATUS(status); | |
1532 | ||
c88ad5db | 1533 | Log("SHUTTLE", Form("Process - %s: the return code is %d", fCurrentDetector.Data(), |
3301427a | 1534 | returnCode)); |
be48e3ea | 1535 | |
9827400b | 1536 | if (returnCode == 0) hasError = kTRUE; |
be48e3ea | 1537 | } |
1538 | } | |
1539 | else if (pid == 0) | |
1540 | { | |
26b1866e | 1541 | // child |
1542 | Log("SHUTTLE", Form("Process - In child process of %d - %s", GetCurrentRun(), | |
c88ad5db | 1543 | aDetector->GetName())); |
be48e3ea | 1544 | |
c88ad5db | 1545 | Log("SHUTTLE", Form("Process - Redirecting output to %s log",fCurrentDetector.Data())); |
ffa29e93 | 1546 | |
546242fb | 1547 | if ((freopen(GetLogFileName(fCurrentDetector), "a", stdout)) == 0) |
ffa29e93 | 1548 | { |
c88ad5db | 1549 | Log("SHUTTLE", "Process - Could not freopen stdout"); |
ffa29e93 | 1550 | } |
1551 | else | |
1552 | { | |
1553 | fOutputRedirected = kTRUE; | |
1554 | if ((dup2(fileno(stdout), fileno(stderr))) < 0) | |
c88ad5db | 1555 | Log("SHUTTLE", "Process - Could not redirect stderr"); |
ffa29e93 | 1556 | |
1557 | } | |
51c3744e | 1558 | |
1559 | Log("SHUTTLE", "Executing TGrid::Connect"); | |
1560 | TGrid::Connect("alien://"); | |
ffa29e93 | 1561 | |
5bac2bde | 1562 | TString wd = gSystem->WorkingDirectory(); |
3ebbf3c4 | 1563 | Int_t dir_lev1 = GetCurrentRun()/10000; |
1564 | TString tmpDir = Form("%s/%d/%d/%s_process", GetShuttleTempDir(), | |
1565 | dir_lev1, GetCurrentRun(), fCurrentDetector.Data()); | |
5bac2bde | 1566 | |
d524ade6 | 1567 | Int_t result = gSystem->GetPathInfo(tmpDir.Data(), 0, (Long64_t*) 0, 0, 0); |
1568 | if (!result) // temp dir already exists! | |
1569 | { | |
1570 | Log(fCurrentDetector.Data(), | |
1571 | Form("Process - %s dir already exists! Removing...", tmpDir.Data())); | |
1572 | gSystem->Exec(Form("rm -rf %s",tmpDir.Data())); | |
675f64cd | 1573 | } |
1574 | ||
1575 | if (gSystem->mkdir(tmpDir.Data(), 1)) | |
1576 | { | |
1577 | Log(fCurrentDetector.Data(), "Process - could not make temp directory!!"); | |
1578 | gSystem->Exit(1); | |
d524ade6 | 1579 | } |
1580 | ||
1581 | if (!gSystem->ChangeDirectory(tmpDir.Data())) | |
1582 | { | |
1583 | Log(fCurrentDetector.Data(), "Process - could not change directory!!"); | |
1584 | gSystem->Exit(1); | |
1585 | } | |
5bac2bde | 1586 | |
ffa25f20 | 1587 | Int_t success = ProcessCurrentDetector(); |
26b1866e | 1588 | |
5bac2bde | 1589 | gSystem->ChangeDirectory(wd.Data()); |
b1d18693 | 1590 | |
ffa25f20 | 1591 | if (success == 1) // Preprocessor finished successfully! |
9827400b | 1592 | { |
fb40ead4 | 1593 | // remove temporary folder or DCS map |
1594 | if (!fConfig->KeepTempFolder()) | |
1595 | { | |
1596 | gSystem->Exec(Form("rm -rf %s",tmpDir.Data())); | |
1597 | } else if (!fConfig->KeepDCSMap()) | |
1598 | { | |
1599 | gSystem->Exec(Form("rm -f %s/DCSMap.root",tmpDir.Data())); | |
1600 | } | |
b1d18693 | 1601 | |
3301427a | 1602 | // Update time_processed field in FXS DB |
1603 | if (UpdateTable() == kFALSE) | |
5bac2bde | 1604 | Log("SHUTTLE", Form("Process - %s: Could not update FXS databases!", |
1605 | fCurrentDetector.Data())); | |
3301427a | 1606 | |
1607 | // Transfer the data from local storage to main storage (Grid) | |
3301427a | 1608 | if (StoreOCDB() == kFALSE) |
9827400b | 1609 | success = kFALSE; |
5e993b6f | 1610 | } |
ffa25f20 | 1611 | else if (success == 0) |
c88ad5db | 1612 | { |
1613 | Log("SHUTTLE", | |
5ca8f352 | 1614 | Form("\t\t\t****** run %d - %s: ERROR ******", |
c88ad5db | 1615 | GetCurrentRun(), aDetector->GetName())); |
be48e3ea | 1616 | } |
1617 | ||
4b95672b | 1618 | for (UInt_t iSys=0; iSys<3; iSys++) |
1619 | { | |
1620 | if (fFXSCalled[iSys]) fFXSlist[iSys].Clear(); | |
1621 | } | |
1622 | ||
c88ad5db | 1623 | Log("SHUTTLE", Form("Process - Client process of %d - %s is exiting now with %d.", |
9827400b | 1624 | GetCurrentRun(), aDetector->GetName(), success)); |
be48e3ea | 1625 | |
1626 | // the client exits here | |
9827400b | 1627 | gSystem->Exit(success); |
be48e3ea | 1628 | |
1629 | AliError("We should never get here!!!"); | |
1630 | } | |
7bfb2090 | 1631 | } |
5164a766 | 1632 | |
c88ad5db | 1633 | Log("SHUTTLE", Form("\t\t\t^*^*^*^*^*^*^*^*^*^*^*^* run %d: FINISH ^*^*^*^*^*^*^*^*^*^*^*^*", |
2bb7b766 | 1634 | GetCurrentRun())); |
1635 | ||
1636 | //check if shuttle is done for this run, if so update logbook | |
1637 | TObjArray checkEntryArray; | |
1638 | checkEntryArray.SetOwner(1); | |
9e080f92 | 1639 | TString whereClause = Form("where run=%d", GetCurrentRun()); |
b0e53b15 | 1640 | if (!QueryShuttleLogbook(whereClause.Data(), checkEntryArray) || |
1641 | checkEntryArray.GetEntries() == 0) { | |
9e080f92 | 1642 | Log("SHUTTLE", Form("Process - Warning: Cannot check status of run %d on Shuttle logbook!", |
1643 | GetCurrentRun())); | |
1644 | return hasError == kFALSE; | |
1645 | } | |
b948db8d | 1646 | |
9e080f92 | 1647 | AliShuttleLogbookEntry* checkEntry = dynamic_cast<AliShuttleLogbookEntry*> |
1648 | (checkEntryArray.At(0)); | |
2bb7b766 | 1649 | |
9e080f92 | 1650 | if (checkEntry) |
1651 | { | |
1652 | if (checkEntry->IsDone()) | |
be48e3ea | 1653 | { |
9e080f92 | 1654 | Log("SHUTTLE","Process - Shuttle is DONE. Updating logbook"); |
1655 | UpdateShuttleLogbook("shuttle_done"); | |
1656 | } | |
1657 | else | |
1658 | { | |
1659 | for (UInt_t iDet=0; iDet<NDetectors(); iDet++) | |
be48e3ea | 1660 | { |
9e080f92 | 1661 | if (checkEntry->GetDetectorStatus(iDet) == AliShuttleLogbookEntry::kUnprocessed) |
be48e3ea | 1662 | { |
9e080f92 | 1663 | AliDebug(2, Form("Run %d: setting %s as \"not first time unprocessed\"", |
1664 | checkEntry->GetRun(), GetDetName(iDet))); | |
1665 | fFirstUnprocessed[iDet] = kFALSE; | |
be48e3ea | 1666 | } |
1667 | } | |
8e828d39 | 1668 | SendMLRunInfo("Pending"); |
2bb7b766 | 1669 | } |
1670 | } | |
1671 | ||
1672 | fLogbookEntry = 0; | |
85a80aa9 | 1673 | |
a7160fe9 | 1674 | return hasError == kFALSE; |
73abe331 | 1675 | } |
1676 | ||
b948db8d | 1677 | //______________________________________________________________________________________________ |
ffa25f20 | 1678 | Int_t AliShuttle::ProcessCurrentDetector() |
73abe331 | 1679 | { |
1680 | // | |
2bb7b766 | 1681 | // Makes data retrieval just for a specific detector (fCurrentDetector). |
73abe331 | 1682 | // Threre should be a configuration for this detector. |
73abe331 | 1683 | |
1d172743 | 1684 | Log("SHUTTLE", Form("ProcessCurrentDetector - Retrieving values for %s, run %d", |
1685 | fCurrentDetector.Data(), GetCurrentRun())); | |
73abe331 | 1686 | |
d524ade6 | 1687 | TString wd = gSystem->WorkingDirectory(); |
1688 | ||
2d9019b4 | 1689 | if (!CleanReferenceStorage(fCurrentDetector.Data())) |
ffa25f20 | 1690 | return 0; |
d524ade6 | 1691 | |
1692 | gSystem->ChangeDirectory(wd.Data()); | |
1693 | ||
3301427a | 1694 | // call preprocessor |
1695 | AliPreprocessor* aPreprocessor = | |
1696 | dynamic_cast<AliPreprocessor*> (fPreprocessorMap.GetValue(fCurrentDetector)); | |
1697 | ||
8579c6fb | 1698 | // check if the preprocessor wants to process this run type |
1699 | if (aPreprocessor->ProcessRunType() == kFALSE) | |
ffa25f20 | 1700 | { |
1701 | UpdateShuttleStatus(AliShuttleStatus::kSkipped); | |
1702 | UpdateShuttleLogbook(fCurrentDetector, "DONE"); | |
bd23d48b | 1703 | if (!UpdateTableSkippedCase(fCurrentDetector.Data())) |
1704 | { | |
1705 | AliError(Form("Could not update FXS tables for run %d !", GetCurrentRun())); | |
1706 | } | |
ffa25f20 | 1707 | Log(fCurrentDetector, Form("ProcessCurrentDetector - %s preprocessor is not interested in this run type", fCurrentDetector.Data())); |
1708 | ||
1709 | return 2; | |
1710 | } | |
1711 | ||
5ca8f352 | 1712 | // checking if OCDB is reachable |
1713 | AliCDBEntry* testEntry = GetFromOCDB("SHUTTLE","GRP/CTP/DummyConfig"); | |
1714 | if (!testEntry){ | |
1715 | // OCDB is not accessible, going in OCDBError for current detector | |
1716 | AliError("OCDB Test entry not accessible"); | |
1717 | UpdateShuttleStatus(AliShuttleStatus::kOCDBError); | |
1718 | return 0; | |
1719 | } | |
1720 | ||
ffa25f20 | 1721 | TMap* dcsMap = new TMap(); |
1722 | ||
3301427a | 1723 | aPreprocessor->Initialize(GetCurrentRun(), GetCurrentStartTime(), GetCurrentEndTime()); |
1724 | ||
1725 | Bool_t processDCS = aPreprocessor->ProcessDCS(); | |
d477ad88 | 1726 | |
651fdaab | 1727 | if (!processDCS) |
1728 | { | |
1d172743 | 1729 | Log(fCurrentDetector, "ProcessCurrentDetector -" |
1730 | " The preprocessor requested to skip the retrieval of DCS values"); | |
651fdaab | 1731 | } |
8b739301 | 1732 | else if (fTestMode & kSkipDCS) |
2c15234c | 1733 | { |
1d172743 | 1734 | Log(fCurrentDetector, "ProcessCurrentDetector - In TESTMODE: Skipping DCS processing"); |
9827400b | 1735 | } |
1736 | else if (fTestMode & kErrorDCS) | |
1737 | { | |
1d172743 | 1738 | Log(fCurrentDetector, "ProcessCurrentDetector - In TESTMODE: Simulating DCS error"); |
3d8bc902 | 1739 | UpdateShuttleStatus(AliShuttleStatus::kDCSStarted); |
9827400b | 1740 | UpdateShuttleStatus(AliShuttleStatus::kDCSError); |
1d172743 | 1741 | delete dcsMap; |
ffa25f20 | 1742 | return 0; |
2c15234c | 1743 | } else { |
3301427a | 1744 | |
1745 | UpdateShuttleStatus(AliShuttleStatus::kDCSStarted); | |
1746 | ||
1d172743 | 1747 | // Query DCS archive |
1748 | Int_t nServers = fConfig->GetNServers(fCurrentDetector); | |
a038aa70 | 1749 | |
1d172743 | 1750 | for (int iServ=0; iServ<nServers; iServ++) |
2c15234c | 1751 | { |
1d172743 | 1752 | |
1753 | TString host(fConfig->GetDCSHost(fCurrentDetector, iServ)); | |
1754 | Int_t port = fConfig->GetDCSPort(fCurrentDetector, iServ); | |
542b6cc8 | 1755 | Int_t multiSplit = fConfig->GetMultiSplit(fCurrentDetector, iServ); |
1756 | ||
1790d4b7 | 1757 | Log(fCurrentDetector, Form("ProcessCurrentDetector -" |
1758 | " Querying DCS Amanda server %s:%d (%d of %d)", | |
1759 | host.Data(), port, iServ+1, nServers)); | |
1d172743 | 1760 | |
1761 | TMap* aliasMap = 0; | |
1762 | TMap* dpMap = 0; | |
26b1866e | 1763 | |
1d172743 | 1764 | if (fConfig->GetDCSAliases(fCurrentDetector, iServ)->GetEntries() > 0) |
2c15234c | 1765 | { |
5ed060b1 | 1766 | Log(fCurrentDetector, Form("Querying %d DCS aliases", fConfig->GetDCSAliases(fCurrentDetector, iServ)->GetEntries())); |
1d172743 | 1767 | aliasMap = GetValueSet(host, port, |
542b6cc8 | 1768 | fConfig->GetDCSAliases(fCurrentDetector, iServ), |
1769 | kAlias, multiSplit); | |
1d172743 | 1770 | if (!aliasMap) |
1771 | { | |
1772 | Log(fCurrentDetector, | |
1773 | Form("ProcessCurrentDetector -" | |
675f64cd | 1774 | " Error retrieving DCS aliases from server %s." |
1775 | " Sending mail to DCS experts!", host.Data())); | |
1d172743 | 1776 | UpdateShuttleStatus(AliShuttleStatus::kDCSError); |
675f64cd | 1777 | |
339adafa | 1778 | if (!SendMail(kDCSEMail)) |
fb40ead4 | 1779 | Log("SHUTTLE", Form("ProcessCurrentDetector - " |
95d4aaef | 1780 | "Could not send mail to DCS experts!")); |
675f64cd | 1781 | |
a038aa70 | 1782 | delete dcsMap; |
ffa25f20 | 1783 | return 0; |
1d172743 | 1784 | } |
2c15234c | 1785 | } |
a038aa70 | 1786 | |
1d172743 | 1787 | if (fConfig->GetDCSDataPoints(fCurrentDetector, iServ)->GetEntries() > 0) |
a038aa70 | 1788 | { |
5ed060b1 | 1789 | Log(fCurrentDetector, Form("Querying %d DCS data points", fConfig->GetDCSDataPoints(fCurrentDetector, iServ)->GetEntries())); |
1d172743 | 1790 | dpMap = GetValueSet(host, port, |
542b6cc8 | 1791 | fConfig->GetDCSDataPoints(fCurrentDetector, iServ), |
1792 | kDP, multiSplit); | |
1d172743 | 1793 | if (!dpMap) |
1794 | { | |
1795 | Log(fCurrentDetector, | |
1796 | Form("ProcessCurrentDetector -" | |
675f64cd | 1797 | " Error retrieving DCS data points from server %s." |
1798 | " Sending mail to DCS experts!", host.Data())); | |
1d172743 | 1799 | UpdateShuttleStatus(AliShuttleStatus::kDCSError); |
675f64cd | 1800 | |
339adafa | 1801 | if (!SendMail(kDCSEMail)) |
fb40ead4 | 1802 | Log("SHUTTLE", Form("ProcessCurrentDetector - " |
95d4aaef | 1803 | "Could not send mail to DCS experts!")); |
675f64cd | 1804 | |
1d172743 | 1805 | if (aliasMap) delete aliasMap; |
1806 | delete dcsMap; | |
ffa25f20 | 1807 | return 0; |
1d172743 | 1808 | } |
a038aa70 | 1809 | } |
1d172743 | 1810 | |
1811 | // merge aliasMap and dpMap into dcsMap | |
1812 | if(aliasMap) { | |
1813 | TIter iter(aliasMap); | |
a038aa70 | 1814 | TObjString* key = 0; |
1815 | while ((key = (TObjString*) iter.Next())) | |
1d172743 | 1816 | dcsMap->Add(key, aliasMap->GetValue(key->String())); |
1817 | ||
1818 | aliasMap->SetOwner(kFALSE); | |
1819 | delete aliasMap; | |
1820 | } | |
1821 | ||
1822 | if(dpMap) { | |
1823 | TIter iter(dpMap); | |
1824 | TObjString* key = 0; | |
1825 | while ((key = (TObjString*) iter.Next())) | |
1826 | dcsMap->Add(key, dpMap->GetValue(key->String())); | |
1827 | ||
1828 | dpMap->SetOwner(kFALSE); | |
1829 | delete dpMap; | |
a038aa70 | 1830 | } |
73abe331 | 1831 | } |
1832 | } | |
dc25836b | 1833 | |
b1d18693 | 1834 | // save map into file, to help debugging in case of preprocessor error |
fbc112e3 | 1835 | TFile* f = TFile::Open("DCSMap.root","recreate"); |
b1d18693 | 1836 | f->cd(); |
1837 | dcsMap->Write("DCSMap", TObject::kSingleKey); | |
1838 | f->Close(); | |
fbc112e3 | 1839 | delete f; |
b1d18693 | 1840 | |
2bb7b766 | 1841 | // DCS Archive DB processing successful. Call Preprocessor! |
85a80aa9 | 1842 | UpdateShuttleStatus(AliShuttleStatus::kPPStarted); |
a7160fe9 | 1843 | |
339adafa | 1844 | fFXSError = -1; // this variable is kTRUE after ::Process if an FXS error occured |
8da81210 | 1845 | |
a038aa70 | 1846 | UInt_t returnValue = aPreprocessor->Process(dcsMap); |
8da81210 | 1847 | |
339adafa | 1848 | if (fFXSError!=-1) { |
8da81210 | 1849 | UpdateShuttleStatus(AliShuttleStatus::kFXSError); |
339adafa | 1850 | SendMail(kFXSEMail, fFXSError); |
8da81210 | 1851 | dcsMap->DeleteAll(); |
1852 | delete dcsMap; | |
ffa25f20 | 1853 | return 0; |
8da81210 | 1854 | } |
b948db8d | 1855 | |
3301427a | 1856 | if (returnValue > 0) // Preprocessor error! |
1857 | { | |
c88ad5db | 1858 | Log(fCurrentDetector, Form("ProcessCurrentDetector - " |
1859 | "Preprocessor failed. Process returned %d.", returnValue)); | |
cb343cfd | 1860 | UpdateShuttleStatus(AliShuttleStatus::kPPError); |
a038aa70 | 1861 | dcsMap->DeleteAll(); |
1862 | delete dcsMap; | |
ffa25f20 | 1863 | return 0; |
9827400b | 1864 | } |
1865 | ||
1866 | // preprocessor ok! | |
1867 | UpdateShuttleStatus(AliShuttleStatus::kPPDone); | |
1868 | Log(fCurrentDetector, Form("ProcessCurrentDetector - %s preprocessor returned success", | |
1869 | fCurrentDetector.Data())); | |
b948db8d | 1870 | |
a038aa70 | 1871 | dcsMap->DeleteAll(); |
1872 | delete dcsMap; | |
b948db8d | 1873 | |
ffa25f20 | 1874 | return 1; |
2bb7b766 | 1875 | } |
1876 | ||
fbc112e3 | 1877 | //______________________________________________________________________________________________ |
1878 | void AliShuttle::CountOpenRuns() | |
1879 | { | |
1880 | // Query DAQ's Shuttle logbook and sends the number of open runs to ML | |
1881 | ||
982278d6 | 1882 | SendAlive(); |
1883 | ||
fbc112e3 | 1884 | // check connection, in case connect |
1885 | if (!Connect(3)) | |
1886 | return; | |
1887 | ||
1888 | TString sqlQuery; | |
1889 | sqlQuery = Form("select count(*) from %s where shuttle_done=0", fConfig->GetShuttlelbTable()); | |
1890 | ||
1891 | TSQLResult* aResult = fServer[3]->Query(sqlQuery); | |
1892 | if (!aResult) { | |
1893 | AliError(Form("Can't execute query <%s>!", sqlQuery.Data())); | |
1894 | return; | |
1895 | } | |
1896 | ||
1897 | AliDebug(2,Form("Query = %s", sqlQuery.Data())); | |
1898 | ||
1899 | if (aResult->GetRowCount() == 0) { | |
1900 | AliError(Form("No result for query %s received", sqlQuery.Data())); | |
1901 | return; | |
1902 | } | |
1903 | ||
1904 | if (aResult->GetFieldCount() != 1) { | |
1905 | AliError(Form("Invalid field count for query %s received", sqlQuery.Data())); | |
1906 | return; | |
1907 | } | |
1908 | ||
1909 | TSQLRow* aRow = aResult->Next(); | |
1910 | if (!aRow) { | |
1911 | AliError(Form("Could not receive result of query %s", sqlQuery.Data())); | |
1912 | return; | |
1913 | } | |
1914 | ||
1915 | TString result(aRow->GetField(0), aRow->GetFieldLength(0)); | |
1916 | Int_t count = result.Atoi(); | |
1917 | ||
1918 | Log("SHUTTLE", Form("%d unprocessed runs", count)); | |
1919 | ||
1920 | delete aRow; | |
1921 | delete aResult; | |
1922 | ||
1923 | TMonaLisaValue mlStatus("SHUTTLE_openruns", count); | |
1924 | ||
1925 | TList mlList; | |
1926 | mlList.Add(&mlStatus); | |
1927 | ||
1928 | fMonaLisa->SendParameters(&mlList, "__PROCESSINGINFO__"); | |
1929 | } | |
1930 | ||
2bb7b766 | 1931 | //______________________________________________________________________________________________ |
1932 | Bool_t AliShuttle::QueryShuttleLogbook(const char* whereClause, | |
1933 | TObjArray& entries) | |
1934 | { | |
9827400b | 1935 | // Query DAQ's Shuttle logbook and fills detector status object. |
1936 | // Call QueryRunParameters to query DAQ logbook for run parameters. | |
1937 | // | |
2bb7b766 | 1938 | |
fc5a4708 | 1939 | entries.SetOwner(1); |
1940 | ||
2bb7b766 | 1941 | // check connection, in case connect |
fbc112e3 | 1942 | if (!Connect(3)) return kFALSE; |
2bb7b766 | 1943 | |
1944 | TString sqlQuery; | |
441b0e9c | 1945 | sqlQuery = Form("select * from %s %s order by run", fConfig->GetShuttlelbTable(), whereClause); |
2bb7b766 | 1946 | |
be48e3ea | 1947 | TSQLResult* aResult = fServer[3]->Query(sqlQuery); |
2bb7b766 | 1948 | if (!aResult) { |
1949 | AliError(Form("Can't execute query <%s>!", sqlQuery.Data())); | |
1950 | return kFALSE; | |
1951 | } | |
1952 | ||
fc5a4708 | 1953 | AliDebug(2,Form("Query = %s", sqlQuery.Data())); |
1954 | ||
2bb7b766 | 1955 | if(aResult->GetRowCount() == 0) { |
c88ad5db | 1956 | Log("SHUTTLE", "No entries in Shuttle Logbook match request"); |
9827400b | 1957 | delete aResult; |
1958 | return kTRUE; | |
2bb7b766 | 1959 | } |
1960 | ||
1961 | // TODO Check field count! | |
5b23e2da | 1962 | const UInt_t nCols = 26; |
2bb7b766 | 1963 | if (aResult->GetFieldCount() != (Int_t) nCols) { |
c88ad5db | 1964 | Log("SHUTTLE", "Invalid SQL result field number!"); |
2bb7b766 | 1965 | delete aResult; |
1966 | return kFALSE; | |
1967 | } | |
1968 | ||
2bb7b766 | 1969 | TSQLRow* aRow; |
1970 | while ((aRow = aResult->Next())) { | |
1971 | TString runString(aRow->GetField(0), aRow->GetFieldLength(0)); | |
1972 | Int_t run = runString.Atoi(); | |
1973 | ||
eba76848 | 1974 | AliShuttleLogbookEntry *entry = QueryRunParameters(run); |
1975 | if (!entry) | |
1976 | continue; | |
2bb7b766 | 1977 | |
728290a4 | 1978 | // DA test mode flag |
1979 | TString daTestModeString(aRow->GetField(2), aRow->GetFieldLength(2)); // field 2 = DA test mode flag | |
1980 | Bool_t daTestMode = (Bool_t)daTestModeString.Atoi(); | |
1981 | entry->SetDATestMode(daTestMode); | |
1982 | ||
2bb7b766 | 1983 | // loop on detectors |
eba76848 | 1984 | for(UInt_t ii = 0; ii < nCols; ii++) |
1985 | entry->SetDetectorStatus(aResult->GetFieldName(ii), aRow->GetField(ii)); | |
2bb7b766 | 1986 | |
eba76848 | 1987 | entries.AddLast(entry); |
2bb7b766 | 1988 | delete aRow; |
1989 | } | |
1990 | ||
2bb7b766 | 1991 | delete aResult; |
1992 | return kTRUE; | |
1993 | } | |
1994 | ||
1995 | //______________________________________________________________________________________________ | |
eba76848 | 1996 | AliShuttleLogbookEntry* AliShuttle::QueryRunParameters(Int_t run) |
2bb7b766 | 1997 | { |
eba76848 | 1998 | // |
1999 | // Retrieve run parameters written in the DAQ logbook and sets them into AliShuttleLogbookEntry object | |
2000 | // | |
2bb7b766 | 2001 | |
2002 | // check connection, in case connect | |
be48e3ea | 2003 | if (!Connect(3)) |
eba76848 | 2004 | return 0; |
2bb7b766 | 2005 | |
2006 | TString sqlQuery; | |
2c15234c | 2007 | sqlQuery.Form("select * from %s where run=%d", fConfig->GetDAQlbTable(), run); |
2bb7b766 | 2008 | |
be48e3ea | 2009 | TSQLResult* aResult = fServer[3]->Query(sqlQuery); |
2bb7b766 | 2010 | if (!aResult) { |
c88ad5db | 2011 | Log("SHUTTLE", Form("Can't execute query <%s>!", sqlQuery.Data())); |
eba76848 | 2012 | return 0; |
2bb7b766 | 2013 | } |
2014 | ||
eba76848 | 2015 | if (aResult->GetRowCount() == 0) { |
2bb7b766 | 2016 | Log("SHUTTLE", Form("QueryRunParameters - No entry in DAQ Logbook for run %d. Skipping", run)); |
2017 | delete aResult; | |
eba76848 | 2018 | return 0; |
2bb7b766 | 2019 | } |
2020 | ||
eba76848 | 2021 | if (aResult->GetRowCount() > 1) { |
c88ad5db | 2022 | Log("SHUTTLE", Form("QueryRunParameters - UNEXPECTED: " |
2023 | "more than one entry in DAQ Logbook for run %d!", run)); | |
2bb7b766 | 2024 | delete aResult; |
eba76848 | 2025 | return 0; |
2bb7b766 | 2026 | } |
2027 | ||
eba76848 | 2028 | TSQLRow* aRow = aResult->Next(); |
2029 | if (!aRow) | |
2030 | { | |
c88ad5db | 2031 | Log("SHUTTLE", Form("QueryRunParameters - Could not retrieve row for run %d. Skipping", run)); |
eba76848 | 2032 | delete aResult; |
2033 | return 0; | |
2034 | } | |
2bb7b766 | 2035 | |
eba76848 | 2036 | AliShuttleLogbookEntry* entry = new AliShuttleLogbookEntry(run); |
2bb7b766 | 2037 | |
eba76848 | 2038 | for (Int_t ii = 0; ii < aResult->GetFieldCount(); ii++) |
2039 | entry->SetRunParameter(aResult->GetFieldName(ii), aRow->GetField(ii)); | |
2bb7b766 | 2040 | |
c9db68c7 | 2041 | delete aRow; |
2042 | delete aResult; | |
2043 | ||
eba76848 | 2044 | UInt_t startTime = entry->GetStartTime(); |
2045 | UInt_t endTime = entry->GetEndTime(); | |
89bcf3b4 | 2046 | Bool_t ecsSuccess = entry->GetECSSuccess(); |
1415d3f9 | 2047 | TString runType = entry->GetRunType(); |
2048 | TString tmpdaqstartTime = entry->GetRunParameter("DAQ_time_start"); | |
f38b266c | 2049 | TString recordingFlagString = entry->GetRunParameter("GDCmStreamRecording"); |
2050 | UInt_t recordingFlag = recordingFlagString.Atoi(); | |
1415d3f9 | 2051 | UInt_t daqstartTime = tmpdaqstartTime.Atoi(); |
c9db68c7 | 2052 | |
af93e5d7 | 2053 | UInt_t now = time(0); |
2eb98462 | 2054 | Int_t dcsDelay = fConfig->GetDCSDelay()+fConfig->GetDCSQueryOffset(); |
8e828d39 | 2055 | |
2056 | Bool_t skip = kFALSE; | |
2057 | ||
1415d3f9 | 2058 | // runs are processed if |
2059 | // a) runType is PHYSICS and ecsSuccess is set | |
2060 | // b) runType is not PHYSICS and (ecsSuccess is set or DAQ_time_start is non-0) | |
2061 | // effectively this means that all runs are processed that started properly (ecsSucess behaviour is different for PHYSICS and non-PHYSICS runs (check with ECS!) | |
1bbcc21b | 2062 | if (startTime != 0 && endTime != 0) { |
2063 | if (endTime > startTime) { | |
2064 | if (endTime >= now - dcsDelay) { | |
2065 | Log("SHUTTLE", Form("Skipping run %d for now, because DCS buffer time is not yet expired", run)); | |
2066 | } else { | |
f38b266c | 2067 | if ((runType == "PHYSICS" || runType == "STANDALONE") && recordingFlag == 0){ |
2068 | Log("SHUTTLE", Form("QueryRunParameters - Run type for run %d is %s but the recording is OFF - Skipping!", run, runType.Data())); | |
2069 | skip = kTRUE; | |
2070 | } | |
2071 | else { | |
2072 | if (runType == "PHYSICS") { | |
2073 | if (ecsSuccess) { | |
2074 | return entry; | |
2075 | } else { | |
2076 | Log("SHUTTLE", Form("QueryRunParameters - Run type for run %d is PHYSICS but ECS success flag not set (Reason = %s) - Skipping!", run, entry->GetRunParameter("eor_reason"))); | |
2077 | skip = kTRUE; | |
2078 | } | |
1bbcc21b | 2079 | } else { |
f38b266c | 2080 | if (ecsSuccess || daqstartTime > 0) { |
2081 | if (ecsSuccess == kFALSE) | |
2082 | Log("SHUTTLE", Form("Processing run %d although in status ECS failure (Reason: %s), since run type != PHYSICS and DAQ_time_start != 0", run, entry->GetRunParameter("eor_reason"))); | |
2083 | return entry; | |
2084 | } else { | |
2085 | Log("SHUTTLE", Form("QueryRunParameters - Run type for run %d is %s, ECS success flag was not set (Reason = %s) and DAQ_time_start was NULL - Skipping!", run, runType.Data(), entry->GetRunParameter("eor_reason"))); | |
2086 | skip = kTRUE; | |
2087 | } | |
1bbcc21b | 2088 | } |
2089 | } | |
2090 | } | |
2091 | } else { | |
2092 | Log("SHUTTLE", Form("QueryRunParameters - Invalid parameters for run %d: startTime equal to endTime: %d %d - Skipping!", run, startTime, endTime)); | |
2093 | skip = kTRUE; | |
2094 | } | |
1415d3f9 | 2095 | } else { |
2096 | Log("SHUTTLE", Form("QueryRunParameters - Invalid parameters for Run %d: " | |
2097 | "startTime = %d, endTime = %d. Skipping (Shuttle won't be marked as DONE)!", | |
2098 | run, startTime, endTime)); | |
89bcf3b4 | 2099 | } |
8e828d39 | 2100 | |
2101 | if (skip) | |
2102 | { | |
2103 | Log("SHUTTLE", Form("Marking SHUTTLE skipped for run %d", run)); | |
2104 | fLogbookEntry = entry; | |
2105 | if (!UpdateShuttleLogbook("shuttle_skipped")) | |
2106 | { | |
2107 | AliError(Form("Could not update logbook for run %d !", run)); | |
2108 | } | |
bd23d48b | 2109 | if (!UpdateTableSkippedCase("ALL")) |
2110 | { | |
2111 | AliError(Form("Could not update FXS tables for run %d !", run)); | |
2112 | } | |
8e828d39 | 2113 | fLogbookEntry = 0; |
2114 | } | |
c9db68c7 | 2115 | |
2116 | delete entry; | |
2117 | return 0; | |
2bb7b766 | 2118 | } |
2119 | ||
a038aa70 | 2120 | //______________________________________________________________________________________________ |
2121 | TMap* AliShuttle::GetValueSet(const char* host, Int_t port, const TSeqCollection* entries, | |
542b6cc8 | 2122 | DCSType type, Int_t multiSplit) |
a038aa70 | 2123 | { |
2124 | // Retrieve all "entry" data points from the DCS server | |
2125 | // host, port: TSocket connection parameters | |
2126 | // entries: list of name of the alias or data point | |
2127 | // type: kAlias or kDP | |
2128 | // returns TMap of values, 0 when failure | |
542b6cc8 | 2129 | |
2130 | AliDCSClient client(host, port, fTimeout, fRetries, multiSplit); | |
b41b252a | 2131 | |
a038aa70 | 2132 | TMap* result = 0; |
b41b252a | 2133 | if (type == kAlias) |
a038aa70 | 2134 | { |
5f3fe74e | 2135 | //result = client.GetAliasValues(entries, GetCurrentStartTime()-offset, |
2136 | // GetCurrentEndTime()+offset); | |
eff647d5 | 2137 | result = client.GetAliasValues(entries, GetStartTimeDCSQuery(), |
2138 | GetEndTimeDCSQuery()); | |
b41b252a | 2139 | } |
2140 | else if (type == kDP) | |
2141 | { | |
5f3fe74e | 2142 | //result = client.GetDPValues(entries, GetCurrentStartTime()-offset, |
2143 | // GetCurrentEndTime()+offset); | |
eff647d5 | 2144 | result = client.GetDPValues(entries, GetStartTimeDCSQuery(), |
2145 | GetEndTimeDCSQuery()); | |
b41b252a | 2146 | } |
a038aa70 | 2147 | |
b41b252a | 2148 | if (result == 0) |
2149 | { | |
2150 | Log(fCurrentDetector.Data(), Form("GetValueSet - Can't get entries! Reason: %s", | |
1790d4b7 | 2151 | client.GetErrorString(client.GetResultErrorCode()))); |
2152 | if (client.GetResultErrorCode() == AliDCSClient::fgkServerError) | |
2153 | Log(fCurrentDetector.Data(), Form("GetValueSet - Server error code: %s", | |
2154 | client.GetServerError().Data())); | |
a038aa70 | 2155 | |
b41b252a | 2156 | return 0; |
a038aa70 | 2157 | } |
b41b252a | 2158 | |
a038aa70 | 2159 | return result; |
2160 | } | |
b41b252a | 2161 | |
b948db8d | 2162 | //______________________________________________________________________________________________ |
57f50b3c | 2163 | const char* AliShuttle::GetFile(Int_t system, const char* detector, |
2164 | const char* id, const char* source) | |
b948db8d | 2165 | { |
9827400b | 2166 | // Get calibration file from file exchange servers |
2167 | // First queris the FXS database for the file name, using the run, detector, id and source info | |
2168 | // then calls RetrieveFile(filename) for actual copy to local disk | |
2169 | // run: current run being processed (given by Logbook entry fLogbookEntry) | |
2170 | // detector: the Preprocessor name | |
2171 | // id: provided as a parameter by the Preprocessor | |
2172 | // source: provided by the Preprocessor through GetFileSources function | |
2173 | ||
2174 | // check if test mode should simulate a FXS error | |
2175 | if (fTestMode & kErrorFXSFiles) | |
2176 | { | |
2177 | Log(detector, Form("GetFile - In TESTMODE - Simulating error while connecting to %s FXS", GetSystemName(system))); | |
2178 | return 0; | |
2179 | } | |
2180 | ||
57f50b3c | 2181 | // check connection, in case connect |
9d733021 | 2182 | if (!Connect(system)) |
eba76848 | 2183 | { |
9d733021 | 2184 | Log(detector, Form("GetFile - Couldn't connect to %s FXS database", GetSystemName(system))); |
339adafa | 2185 | fFXSError = system; |
57f50b3c | 2186 | return 0; |
2187 | } | |
2188 | ||
2189 | // Query preparation | |
9d733021 | 2190 | TString sourceName(source); |
d386d623 | 2191 | Int_t nFields = 3; |
2192 | TString sqlQueryStart = Form("select filePath,size,fileChecksum from %s where", | |
2193 | fConfig->GetFXSdbTable(system)); | |
2194 | TString whereClause = Form("run=%d and detector=\"%s\" and fileId=\"%s\"", | |
2195 | GetCurrentRun(), detector, id); | |
2196 | ||
9d733021 | 2197 | if (system == kDAQ) |
2198 | { | |
d386d623 | 2199 | whereClause += Form(" and DAQsource=\"%s\"", source); |
57f50b3c | 2200 | } |
9d733021 | 2201 | else if (system == kDCS) |
eba76848 | 2202 | { |
9d733021 | 2203 | sourceName="none"; |
57f50b3c | 2204 | } |
9d733021 | 2205 | else if (system == kHLT) |
9e080f92 | 2206 | { |
d386d623 | 2207 | whereClause += Form(" and DDLnumbers=\"%s\"", source); |
9e080f92 | 2208 | } |
2209 | ||
9e080f92 | 2210 | TString sqlQuery = Form("%s %s", sqlQueryStart.Data(), whereClause.Data()); |
2211 | ||
2212 | AliDebug(2, Form("SQL query: \n%s",sqlQuery.Data())); | |
2213 | ||
2214 | // Query execution | |
2215 | TSQLResult* aResult = 0; | |
9d733021 | 2216 | aResult = dynamic_cast<TSQLResult*> (fServer[system]->Query(sqlQuery)); |
9e080f92 | 2217 | if (!aResult) { |
8da81210 | 2218 | Log(detector, Form("GetFile - Can't execute SQL query to %s database for: id = %s, source = %s", |
9d733021 | 2219 | GetSystemName(system), id, sourceName.Data())); |
339adafa | 2220 | fFXSError = system; |
9e080f92 | 2221 | return 0; |
2222 | } | |
2223 | ||
8da81210 | 2224 | if (aResult->GetRowCount() == 0) |
9e080f92 | 2225 | { |
2226 | Log(detector, | |
8da81210 | 2227 | Form("GetFile - No entry in %s FXS db for: id = %s, source = %s", |
9d733021 | 2228 | GetSystemName(system), id, sourceName.Data())); |
9e080f92 | 2229 | delete aResult; |
2230 | return 0; | |
2231 | } | |
2bb7b766 | 2232 | |
9e080f92 | 2233 | if (aResult->GetRowCount() > 1) { |
2234 | Log(detector, | |
8da81210 | 2235 | Form("GetFile - More than one entry in %s FXS db for: id = %s, source = %s", |
9d733021 | 2236 | GetSystemName(system), id, sourceName.Data())); |
339adafa | 2237 | fFXSError = system; |
9e080f92 | 2238 | delete aResult; |
2239 | return 0; | |
2240 | } | |
2241 | ||
9d733021 | 2242 | if (aResult->GetFieldCount() != nFields) { |
9e080f92 | 2243 | Log(detector, |
9d733021 | 2244 | Form("GetFileName - Wrong field count in %s FXS db for: id = %s, source = %s", |
2245 | GetSystemName(system), id, sourceName.Data())); | |
339adafa | 2246 | fFXSError = system; |
9e080f92 | 2247 | delete aResult; |
2248 | return 0; | |
2249 | } | |
2250 | ||
2251 | TSQLRow* aRow = dynamic_cast<TSQLRow*> (aResult->Next()); | |
2252 | ||
2253 | if (!aRow){ | |
8da81210 | 2254 | Log(detector, Form("GetFile - Empty set result in %s FXS db from query: id = %s, source = %s", |
9d733021 | 2255 | GetSystemName(system), id, sourceName.Data())); |
339adafa | 2256 | fFXSError = system; |
9e080f92 | 2257 | delete aResult; |
2258 | return 0; | |
2259 | } | |
2260 | ||
2261 | TString filePath(aRow->GetField(0), aRow->GetFieldLength(0)); | |
2262 | TString fileSize(aRow->GetField(1), aRow->GetFieldLength(1)); | |
d386d623 | 2263 | TString fileChecksum(aRow->GetField(2), aRow->GetFieldLength(2)); |
9e080f92 | 2264 | |
2265 | delete aResult; | |
2266 | delete aRow; | |
2267 | ||
d386d623 | 2268 | AliDebug(2, Form("filePath = %s; size = %s, fileChecksum = %s", |
2269 | filePath.Data(), fileSize.Data(), fileChecksum.Data())); | |
9e080f92 | 2270 | |
9e080f92 | 2271 | // retrieved file is renamed to make it unique |
3ebbf3c4 | 2272 | Int_t dir_lev1 = GetCurrentRun()/10000; |
2273 | TString localFileName = Form("%s/%d/%d/%s_process/%s_%s_%d_%s_%s.shuttle", | |
2274 | GetShuttleTempDir(), dir_lev1, GetCurrentRun(), detector, | |
d524ade6 | 2275 | GetSystemName(system), detector, GetCurrentRun(), |
2276 | id, sourceName.Data()); | |
3ebbf3c4 | 2277 | Log("SHUTTLE",Form("file from FXS = %s",localFileName.Data())); |
9d733021 | 2278 | |
9e080f92 | 2279 | |
9d733021 | 2280 | // file retrieval from FXS |
4b95672b | 2281 | UInt_t nRetries = 0; |
2282 | UInt_t maxRetries = 3; | |
2283 | Bool_t result = kFALSE; | |
2284 | ||
2285 | // copy!! if successful TSystem::Exec returns 0 | |
8da81210 | 2286 | while (nRetries++ < maxRetries) { |
4b95672b | 2287 | AliDebug(2, Form("Trying to copy file. Retry # %d", nRetries)); |
2288 | result = RetrieveFile(system, filePath.Data(), localFileName.Data()); | |
8da81210 | 2289 | if (!result) |
4b95672b | 2290 | { |
8da81210 | 2291 | Log(detector, Form("GetFile - Copy of file %s from %s FXS failed", |
9d733021 | 2292 | filePath.Data(), GetSystemName(system))); |
4b95672b | 2293 | continue; |
4f0749a8 | 2294 | } |
9e080f92 | 2295 | |
c98417a1 | 2296 | if (fileSize.Length()>0) |
2297 | { | |
2298 | // compare filesize of local file with the one stored in the FXS DB | |
926cbc0e | 2299 | Long_t size = -1; |
2300 | Int_t sizeComp = gSystem->GetPathInfo(localFileName.Data(), 0, &size, 0, 0); | |
c98417a1 | 2301 | |
926cbc0e | 2302 | if (sizeComp != 0 || size != fileSize.Atoi()) |
c98417a1 | 2303 | { |
8da81210 | 2304 | Log(detector, Form("GetFile - size of file %s does not match with local copy!", |
c98417a1 | 2305 | filePath.Data())); |
2306 | result = kFALSE; | |
2307 | continue; | |
2308 | } | |
2309 | ||
2310 | } else { | |
2311 | Log(fCurrentDetector, Form("GetFile - size of file %s not set in %s database, skipping comparison", | |
2312 | filePath.Data(), GetSystemName(system))); | |
2313 | } | |
2314 | ||
d386d623 | 2315 | if (fileChecksum.Length()>0) |
4b95672b | 2316 | { |
2317 | // compare md5sum of local file with the one stored in the FXS DB | |
95d4aaef | 2318 | if(fileChecksum.Contains(' ')) fileChecksum.Resize(fileChecksum.First(' ')); |
339adafa | 2319 | Int_t md5Comp = gSystem->Exec(Form("md5sum %s |grep %s > /dev/null 2> /dev/null", |
d524ade6 | 2320 | localFileName.Data(), fileChecksum.Data())); |
9e080f92 | 2321 | |
4b95672b | 2322 | if (md5Comp != 0) |
2323 | { | |
8da81210 | 2324 | Log(detector, Form("GetFile - md5sum of file %s does not match with local copy!", |
4b95672b | 2325 | filePath.Data())); |
2326 | result = kFALSE; | |
2327 | continue; | |
2328 | } | |
d386d623 | 2329 | } else { |
2330 | Log(fCurrentDetector, Form("GetFile - md5sum of file %s not set in %s database, skipping comparison", | |
2331 | filePath.Data(), GetSystemName(system))); | |
9d733021 | 2332 | } |
4b95672b | 2333 | if (result) break; |
9e080f92 | 2334 | } |
2335 | ||
8da81210 | 2336 | if (!result) |
2337 | { | |
339adafa | 2338 | fFXSError = system; |
8da81210 | 2339 | return 0; |
2340 | } | |
4b95672b | 2341 | |
9d733021 | 2342 | fFXSCalled[system]=kTRUE; |
2343 | TObjString *fileParams = new TObjString(Form("%s#!?!#%s", id, sourceName.Data())); | |
2344 | fFXSlist[system].Add(fileParams); | |
9e080f92 | 2345 | |
675f64cd | 2346 | static TString staticLocalFileName; |
2347 | staticLocalFileName.Form("%s", localFileName.Data()); | |
2348 | ||
c88ad5db | 2349 | Log(fCurrentDetector, Form("GetFile - Retrieved file with id %s and " |
2350 | "source %s from %s to %s", id, source, | |
d524ade6 | 2351 | GetSystemName(system), localFileName.Data())); |
675f64cd | 2352 | |
d524ade6 | 2353 | return staticLocalFileName.Data(); |
2bb7b766 | 2354 | } |
2355 | ||
2356 | //______________________________________________________________________________________________ | |
9d733021 | 2357 | Bool_t AliShuttle::RetrieveFile(UInt_t system, const char* fxsFileName, const char* localFileName) |
9e080f92 | 2358 | { |
9827400b | 2359 | // |
2360 | // Copies file from FXS to local Shuttle machine | |
2361 | // | |
2bb7b766 | 2362 | |
9e080f92 | 2363 | // check temp directory: trying to cd to temp; if it does not exist, create it |
d524ade6 | 2364 | AliDebug(2, Form("Copy file %s from %s FXS into %s", |
2365 | GetSystemName(system), fxsFileName, localFileName)); | |
2366 | ||
2367 | TString tmpDir(localFileName); | |
2368 | ||
2369 | tmpDir = tmpDir(0,tmpDir.Last('/')); | |
9e080f92 | 2370 | |
d524ade6 | 2371 | Int_t noDir = gSystem->GetPathInfo(tmpDir.Data(), 0, (Long64_t*) 0, 0, 0); |
2372 | if (noDir) // temp dir does not exists! | |
2373 | { | |
2374 | if (gSystem->mkdir(tmpDir.Data(), 1)) | |
2375 | { | |
2376 | Log(fCurrentDetector.Data(), "RetrieveFile - could not make temp directory!!"); | |
9e080f92 | 2377 | return kFALSE; |
2378 | } | |
9e080f92 | 2379 | } |
2380 | ||
339adafa | 2381 | TString command = Form("scp -oPort=%d -2 %s@%s:%s/%s %s", |
9d733021 | 2382 | fConfig->GetFXSPort(system), |
2383 | fConfig->GetFXSUser(system), | |
2384 | fConfig->GetFXSHost(system), | |
339adafa | 2385 | fConfig->GetFXSBaseFolder(system), |
9d733021 | 2386 | fxsFileName, |
9e080f92 | 2387 | localFileName); |
2388 | ||
2389 | AliDebug(2, Form("%s",command.Data())); | |
2390 | ||
4b95672b | 2391 | Bool_t result = (gSystem->Exec(command.Data()) == 0); |
9e080f92 | 2392 | |
4b95672b | 2393 | return result; |
9e080f92 | 2394 | } |
2395 | ||
2396 | //______________________________________________________________________________________________ | |
9d733021 | 2397 | TList* AliShuttle::GetFileSources(Int_t system, const char* detector, const char* id) |
2398 | { | |
9827400b | 2399 | // |
2400 | // Get sources producing the condition file Id from file exchange servers | |
4a33bdd9 | 2401 | // if id is NULL all sources are returned (distinct) |
9827400b | 2402 | // |
1bcd28db | 2403 | |
8884d71e | 2404 | if (id) |
2405 | { | |
2406 | Log(detector, Form("GetFileSources - Querying %s FXS for files with id %s produced by %s", GetSystemName(system), id, detector)); | |
2407 | } else { | |
2408 | Log(detector, Form("GetFileSources - Querying %s FXS for files produced by %s", GetSystemName(system), detector)); | |
2409 | } | |
9827400b | 2410 | |
2411 | // check if test mode should simulate a FXS error | |
2412 | if (fTestMode & kErrorFXSSources) | |
2413 | { | |
2414 | Log(detector, Form("GetFileSources - In TESTMODE - Simulating error while connecting to %s FXS", GetSystemName(system))); | |
2415 | return 0; | |
2416 | } | |
2417 | ||
9d733021 | 2418 | if (system == kDCS) |
2419 | { | |
c88ad5db | 2420 | Log(detector, "GetFileSources - WARNING: DCS system has only one source of data!"); |
6297b37d | 2421 | TList *list = new TList(); |
2422 | list->SetOwner(1); | |
2423 | list->Add(new TObjString(" ")); | |
2424 | return list; | |
9d733021 | 2425 | } |
9e080f92 | 2426 | |
2427 | // check connection, in case connect | |
9d733021 | 2428 | if (!Connect(system)) |
2429 | { | |
4a33bdd9 | 2430 | Log(detector, Form("GetFileSources - Couldn't connect to %s FXS database", GetSystemName(system))); |
339adafa | 2431 | fFXSError = system; |
9d733021 | 2432 | return NULL; |
9e080f92 | 2433 | } |
2434 | ||
b827f957 | 2435 | TString sourceName = ""; |
9d733021 | 2436 | if (system == kDAQ) |
2437 | { | |
2438 | sourceName = "DAQsource"; | |
2439 | } else if (system == kHLT) | |
2440 | { | |
2441 | sourceName = "DDLnumbers"; | |
2442 | } | |
2443 | ||
4a33bdd9 | 2444 | TString sqlQueryStart = Form("select distinct %s from %s where", sourceName.Data(), fConfig->GetFXSdbTable(system)); |
2445 | TString whereClause = Form("run=%d and detector=\"%s\"", | |
2446 | GetCurrentRun(), detector); | |
2447 | if (id) | |
2448 | whereClause += Form(" and fileId=\"%s\"", id); | |
9e080f92 | 2449 | TString sqlQuery = Form("%s %s", sqlQueryStart.Data(), whereClause.Data()); |
2450 | ||
2451 | AliDebug(2, Form("SQL query: \n%s",sqlQuery.Data())); | |
2452 | ||
2453 | // Query execution | |
2454 | TSQLResult* aResult; | |
9d733021 | 2455 | aResult = fServer[system]->Query(sqlQuery); |
9e080f92 | 2456 | if (!aResult) { |
9d733021 | 2457 | Log(detector, Form("GetFileSources - Can't execute SQL query to %s database for id: %s", |
2458 | GetSystemName(system), id)); | |
339adafa | 2459 | fFXSError = system; |
9e080f92 | 2460 | return 0; |
2461 | } | |
2462 | ||
86aa42c3 | 2463 | TList *list = new TList(); |
2464 | list->SetOwner(1); | |
2465 | ||
9d733021 | 2466 | if (aResult->GetRowCount() == 0) |
2467 | { | |
9e080f92 | 2468 | Log(detector, |
9d733021 | 2469 | Form("GetFileSources - No entry in %s FXS table for id: %s", GetSystemName(system), id)); |
9e080f92 | 2470 | delete aResult; |
86aa42c3 | 2471 | return list; |
9e080f92 | 2472 | } |
2473 | ||
1bcd28db | 2474 | Log(detector, Form("GetFileSources - Found %d sources", aResult->GetRowCount())); |
9e080f92 | 2475 | |
1bcd28db | 2476 | TSQLRow* aRow; |
9d733021 | 2477 | while ((aRow = aResult->Next())) |
2478 | { | |
9e080f92 | 2479 | |
9d733021 | 2480 | TString source(aRow->GetField(0), aRow->GetFieldLength(0)); |
2481 | AliDebug(2, Form("%s = %s", sourceName.Data(), source.Data())); | |
2482 | list->Add(new TObjString(source)); | |
9e080f92 | 2483 | delete aRow; |
2484 | } | |
9d733021 | 2485 | |
9e080f92 | 2486 | delete aResult; |
2487 | ||
2488 | return list; | |
2bb7b766 | 2489 | } |
2490 | ||
4a33bdd9 | 2491 | //______________________________________________________________________________________________ |
2492 | TList* AliShuttle::GetFileIDs(Int_t system, const char* detector, const char* source) | |
2493 | { | |
2494 | // | |
2495 | // Get all ids of condition files produced by a given source from file exchange servers | |
2496 | // | |
2497 | ||
1bcd28db | 2498 | Log(detector, Form("GetFileIDs - Retrieving ids with source %s with %s", source, GetSystemName(system))); |
2499 | ||
4a33bdd9 | 2500 | // check if test mode should simulate a FXS error |
2501 | if (fTestMode & kErrorFXSSources) | |
2502 | { | |
2503 | Log(detector, Form("GetFileIDs - In TESTMODE - Simulating error while connecting to %s FXS", GetSystemName(system))); | |
2504 | return 0; | |
2505 | } | |
2506 | ||
2507 | // check connection, in case connect | |
2508 | if (!Connect(system)) | |
2509 | { | |
2510 | Log(detector, Form("GetFileIDs - Couldn't connect to %s FXS database", GetSystemName(system))); | |
2511 | return NULL; | |
2512 | } | |
2513 | ||
b827f957 | 2514 | TString sourceName = ""; |
4a33bdd9 | 2515 | if (system == kDAQ) |
2516 | { | |
2517 | sourceName = "DAQsource"; | |
2518 | } else if (system == kHLT) | |
2519 | { | |
2520 | sourceName = "DDLnumbers"; | |
2521 | } | |
2522 | ||
2523 | TString sqlQueryStart = Form("select fileId from %s where", fConfig->GetFXSdbTable(system)); | |
2524 | TString whereClause = Form("run=%d and detector=\"%s\"", | |
2525 | GetCurrentRun(), detector); | |
2526 | if (sourceName.Length() > 0 && source) | |
2527 | whereClause += Form(" and %s=\"%s\"", sourceName.Data(), source); | |
2528 | TString sqlQuery = Form("%s %s", sqlQueryStart.Data(), whereClause.Data()); | |
2529 | ||
2530 | AliDebug(2, Form("SQL query: \n%s",sqlQuery.Data())); | |
2531 | ||
2532 | // Query execution | |
2533 | TSQLResult* aResult; | |
2534 | aResult = fServer[system]->Query(sqlQuery); | |
2535 | if (!aResult) { | |
2536 | Log(detector, Form("GetFileIDs - Can't execute SQL query to %s database for source: %s", | |
2537 | GetSystemName(system), source)); | |
2538 | return 0; | |
2539 | } | |
2540 | ||
2541 | TList *list = new TList(); | |
2542 | list->SetOwner(1); | |
2543 | ||
2544 | if (aResult->GetRowCount() == 0) | |
2545 | { | |
2546 | Log(detector, | |
2547 | Form("GetFileIDs - No entry in %s FXS table for source: %s", GetSystemName(system), source)); | |
2548 | delete aResult; | |
2549 | return list; | |
2550 | } | |
2551 | ||
1bcd28db | 2552 | Log(detector, Form("GetFileIDs - Found %d ids", aResult->GetRowCount())); |
2553 | ||
4a33bdd9 | 2554 | TSQLRow* aRow; |
2555 | ||
2556 | while ((aRow = aResult->Next())) | |
2557 | { | |
2558 | ||
2559 | TString id(aRow->GetField(0), aRow->GetFieldLength(0)); | |
2560 | AliDebug(2, Form("fileId = %s", id.Data())); | |
2561 | list->Add(new TObjString(id)); | |
2562 | delete aRow; | |
2563 | } | |
2564 | ||
2565 | delete aResult; | |
2566 | ||
2567 | return list; | |
2568 | } | |
2569 | ||
2bb7b766 | 2570 | //______________________________________________________________________________________________ |
9d733021 | 2571 | Bool_t AliShuttle::Connect(Int_t system) |
2bb7b766 | 2572 | { |
9827400b | 2573 | // Connect to MySQL Server of the system's FXS MySQL databases |
2574 | // DAQ Logbook, Shuttle Logbook and DAQ FXS db are on the same host | |
2575 | // | |
57f50b3c | 2576 | |
9d733021 | 2577 | // check connection: if already connected return |
dd84c76d | 2578 | |
3a4ddf0e | 2579 | if(fServer[system] && fServer[system]->IsConnected()) { |
dd84c76d | 2580 | // ping the server |
2581 | if (fServer[system]->PingVerify()==kTRUE){ // connection is still alive | |
2582 | return kTRUE; | |
2583 | } | |
2584 | else{ | |
2585 | AliWarning(Form("Connection got lost to FXS database for %s. Closing and reconnecting.", | |
2586 | AliShuttleInterface::GetSystemName(system))); | |
2587 | fServer[system]->Close(); | |
2588 | delete fServer[system]; | |
2589 | fServer[system] = 0x0; | |
2590 | } | |
3a4ddf0e | 2591 | } |
57f50b3c | 2592 | |
9d733021 | 2593 | TString dbHost, dbUser, dbPass, dbName; |
57f50b3c | 2594 | |
9d733021 | 2595 | if (system < 3) // FXS db servers |
2596 | { | |
2597 | dbHost = Form("mysql://%s:%d", fConfig->GetFXSdbHost(system), fConfig->GetFXSdbPort(system)); | |
2598 | dbUser = fConfig->GetFXSdbUser(system); | |
2599 | dbPass = fConfig->GetFXSdbPass(system); | |
2600 | dbName = fConfig->GetFXSdbName(system); | |
2601 | } else { // Run & Shuttle logbook servers | |
2602 | // TODO Will the Shuttle logbook server be the same as the Run logbook server ??? | |
2603 | dbHost = Form("mysql://%s:%d", fConfig->GetDAQlbHost(), fConfig->GetDAQlbPort()); | |
2604 | dbUser = fConfig->GetDAQlbUser(); | |
2605 | dbPass = fConfig->GetDAQlbPass(); | |
2606 | dbName = fConfig->GetDAQlbDB(); | |
2607 | } | |
57f50b3c | 2608 | |
9d733021 | 2609 | fServer[system] = TSQLServer::Connect(dbHost.Data(), dbUser.Data(), dbPass.Data()); |
2610 | if (!fServer[system] || !fServer[system]->IsConnected()) { | |
2611 | if(system < 3) | |
2612 | { | |
2613 | AliError(Form("Can't establish connection to FXS database for %s", | |
2614 | AliShuttleInterface::GetSystemName(system))); | |
2615 | } else { | |
2616 | AliError("Can't establish connection to Run logbook."); | |
57f50b3c | 2617 | } |
9d733021 | 2618 | if(fServer[system]) delete fServer[system]; |
2619 | return kFALSE; | |
2bb7b766 | 2620 | } |
57f50b3c | 2621 | |
9d733021 | 2622 | // Get tables |
2623 | TSQLResult* aResult=0; | |
2624 | switch(system){ | |
2625 | case kDAQ: | |
2626 | aResult = fServer[kDAQ]->GetTables(dbName.Data()); | |
2627 | break; | |
2628 | case kDCS: | |
2629 | aResult = fServer[kDCS]->GetTables(dbName.Data()); | |
2630 | break; | |
2631 | case kHLT: | |
2632 | aResult = fServer[kHLT]->GetTables(dbName.Data()); | |
2633 | break; | |
2634 | default: | |
2635 | aResult = fServer[3]->GetTables(dbName.Data()); | |
2636 | break; | |
2637 | } | |
2638 | ||
2639 | delete aResult; | |
2bb7b766 | 2640 | return kTRUE; |
2641 | } | |
57f50b3c | 2642 | |
9e080f92 | 2643 | //______________________________________________________________________________________________ |
9d733021 | 2644 | Bool_t AliShuttle::UpdateTable() |
9e080f92 | 2645 | { |
9827400b | 2646 | // |
2647 | // Update FXS table filling time_processed field in all rows corresponding to current run and detector | |
2648 | // | |
9e080f92 | 2649 | |
9d733021 | 2650 | Bool_t result = kTRUE; |
9e080f92 | 2651 | |
9d733021 | 2652 | for (UInt_t system=0; system<3; system++) |
2653 | { | |
2654 | if(!fFXSCalled[system]) continue; | |
9e080f92 | 2655 | |
9d733021 | 2656 | // check connection, in case connect |
2657 | if (!Connect(system)) | |
2658 | { | |
2659 | Log(fCurrentDetector, Form("UpdateTable - Couldn't connect to %s FXS database", GetSystemName(system))); | |
2660 | result = kFALSE; | |
2661 | continue; | |
9e080f92 | 2662 | } |
9e080f92 | 2663 | |
9d733021 | 2664 | TTimeStamp now; // now |
2665 | ||
2666 | // Loop on FXS list entries | |
2667 | TIter iter(&fFXSlist[system]); | |
2668 | TObjString *aFXSentry=0; | |
2669 | while ((aFXSentry = dynamic_cast<TObjString*> (iter.Next()))) | |
2670 | { | |
2671 | TString aFXSentrystr = aFXSentry->String(); | |
2672 | TObjArray *aFXSarray = aFXSentrystr.Tokenize("#!?!#"); | |
2673 | if (!aFXSarray || aFXSarray->GetEntries() != 2 ) | |
2674 | { | |
2675 | Log(fCurrentDetector, Form("UpdateTable - error updating %s FXS entry. Check string: <%s>", | |
2676 | GetSystemName(system), aFXSentrystr.Data())); | |
2677 | if(aFXSarray) delete aFXSarray; | |
2678 | result = kFALSE; | |
2679 | continue; | |
2680 | } | |
2681 | const char* fileId = ((TObjString*) aFXSarray->At(0))->GetName(); | |
2682 | const char* source = ((TObjString*) aFXSarray->At(1))->GetName(); | |
2683 | ||
2684 | TString whereClause; | |
2685 | if (system == kDAQ) | |
2686 | { | |
2687 | whereClause = Form("where run=%d and detector=\"%s\" and fileId=\"%s\" and DAQsource=\"%s\";", | |
2688 | GetCurrentRun(), fCurrentDetector.Data(), fileId, source); | |
2689 | } | |
2690 | else if (system == kDCS) | |
2691 | { | |
2692 | whereClause = Form("where run=%d and detector=\"%s\" and fileId=\"%s\";", | |
2693 | GetCurrentRun(), fCurrentDetector.Data(), fileId); | |
2694 | } | |
2695 | else if (system == kHLT) | |
2696 | { | |
2697 | whereClause = Form("where run=%d and detector=\"%s\" and fileId=\"%s\" and DDLnumbers=\"%s\";", | |
2698 | GetCurrentRun(), fCurrentDetector.Data(), fileId, source); | |
2699 | } | |
2700 | ||
2701 | delete aFXSarray; | |
9e080f92 | 2702 | |
0a0b087a | 2703 | TString sqlQuery = Form("update %s set time_processed=%ld %s", fConfig->GetFXSdbTable(system), |
2704 | (ULong_t)now.GetSec(), whereClause.Data()); | |
9e080f92 | 2705 | |
9d733021 | 2706 | AliDebug(2, Form("SQL query: \n%s",sqlQuery.Data())); |
9e080f92 | 2707 | |
9d733021 | 2708 | // Query execution |
2709 | TSQLResult* aResult; | |
2710 | aResult = dynamic_cast<TSQLResult*> (fServer[system]->Query(sqlQuery)); | |
2711 | if (!aResult) | |
2712 | { | |
2713 | Log(fCurrentDetector, Form("UpdateTable - %s db: can't execute SQL query <%s>", | |
2714 | GetSystemName(system), sqlQuery.Data())); | |
2715 | result = kFALSE; | |
2716 | continue; | |
2717 | } | |
2718 | delete aResult; | |
9e080f92 | 2719 | } |
9e080f92 | 2720 | } |
2721 | ||
9d733021 | 2722 | return result; |
9e080f92 | 2723 | } |
57f50b3c | 2724 | |
bd23d48b | 2725 | //_______________________________________________________________________________ |
2726 | Bool_t AliShuttle::UpdateTableSkippedCase(const char* detector) | |
2727 | { | |
2728 | // | |
2729 | // Update FXS table filling time_processed field in all rows corresponding to current run and detector | |
2730 | // if detector = "ALL" update all detectors | |
2731 | // | |
2732 | ||
2733 | Bool_t result = kTRUE; | |
2734 | ||
7a782903 | 2735 | TString detName(detector); |
2736 | ||
bd23d48b | 2737 | for (UInt_t system=0; system<3; system++) |
2738 | { | |
2739 | ||
2740 | // check connection, in case connect | |
2741 | if (!Connect(system)) | |
2742 | { | |
2743 | Log(fCurrentDetector, Form("UpdateTableSkippedCase - Couldn't connect to %s FXS database", GetSystemName(system))); | |
2744 | result = kFALSE; | |
2745 | continue; | |
2746 | } | |
2747 | ||
2748 | TTimeStamp now; // now | |
2749 | ||
2750 | // Loop on FXS list entries | |
2751 | TIter iter(&fFXSlist[system]); | |
2752 | ||
2753 | TString whereClause; | |
7a782903 | 2754 | if (detName == "ALL") whereClause = Form("where run=%d and time_processed IS NULL;",GetCurrentRun()); |
bd23d48b | 2755 | else whereClause = Form("where run=%d and detector=\"%s\" and time_processed IS NULL;",GetCurrentRun(), detector); |
2756 | ||
4be2f262 | 2757 | //Log("SHUTTLE",Form(" whereClause = %s ",whereClause.Data())); |
bd23d48b | 2758 | |
0a0b087a | 2759 | TString sqlQuery = Form("update %s set time_processed=%ld %s", fConfig->GetFXSdbTable(system), |
2760 | (ULong_t)now.GetSec(), whereClause.Data()); | |
bd23d48b | 2761 | |
2762 | AliDebug(2, Form("SQL query: \n%s",sqlQuery.Data())); | |
2763 | ||
2764 | // Query execution | |
2765 | TSQLResult* aResult; | |
2766 | aResult = dynamic_cast<TSQLResult*> (fServer[system]->Query(sqlQuery)); | |
2767 | if (!aResult) | |
2768 | { | |
2769 | Log("SHUTTLE", Form("UpdateTableSkippedCase - %s db: can't execute SQL query <%s>", | |
2770 | GetSystemName(system), sqlQuery.Data())); | |
2771 | result = kFALSE; | |
2772 | continue; | |
2773 | } | |
2774 | delete aResult; | |
2775 | ||
2776 | } | |
2777 | ||
2778 | return result; | |
2779 | } | |
3301427a | 2780 | //______________________________________________________________________________________________ |
2781 | Bool_t AliShuttle::UpdateTableFailCase() | |
2782 | { | |
9827400b | 2783 | // Update FXS table filling time_processed field in all rows corresponding to current run and detector |
2784 | // this is called in case the preprocessor is declared failed for the current run, because | |
2785 | // the fields are updated only in case of success | |
3301427a | 2786 | |
2787 | Bool_t result = kTRUE; | |
2788 | ||
2789 | for (UInt_t system=0; system<3; system++) | |
2790 | { | |
2791 | // check connection, in case connect | |
2792 | if (!Connect(system)) | |
2793 | { | |
2794 | Log(fCurrentDetector, Form("UpdateTableFailCase - Couldn't connect to %s FXS database", | |
2795 | GetSystemName(system))); | |
2796 | result = kFALSE; | |
2797 | continue; | |
2798 | } | |
2799 | ||
2800 | TTimeStamp now; // now | |
2801 | ||
2802 | // Loop on FXS list entries | |
2803 | ||
2804 | TString whereClause = Form("where run=%d and detector=\"%s\";", | |
2805 | GetCurrentRun(), fCurrentDetector.Data()); | |
2806 | ||
2807 | ||
0a0b087a | 2808 | TString sqlQuery = Form("update %s set time_processed=%ld %s", fConfig->GetFXSdbTable(system), |
2809 | (ULong_t)now.GetSec(), whereClause.Data()); | |
3301427a | 2810 | |
2811 | AliDebug(2, Form("SQL query: \n%s",sqlQuery.Data())); | |
2812 | ||
2813 | // Query execution | |
2814 | TSQLResult* aResult; | |
2815 | aResult = dynamic_cast<TSQLResult*> (fServer[system]->Query(sqlQuery)); | |
2816 | if (!aResult) | |
2817 | { | |
2818 | Log(fCurrentDetector, Form("UpdateTableFailCase - %s db: can't execute SQL query <%s>", | |
2819 | GetSystemName(system), sqlQuery.Data())); | |
2820 | result = kFALSE; | |
2821 | continue; | |
2822 | } | |
2823 | delete aResult; | |
2824 | } | |
2825 | ||
2826 | return result; | |
2827 | } | |
2828 | ||
2bb7b766 | 2829 | //______________________________________________________________________________________________ |
2830 | Bool_t AliShuttle::UpdateShuttleLogbook(const char* detector, const char* status) | |
2831 | { | |
e7f62f16 | 2832 | // |
2833 | // Update Shuttle logbook filling detector or shuttle_done column | |
2834 | // ex. of usage: UpdateShuttleLogbook("PHOS", "DONE") or UpdateShuttleLogbook("shuttle_done") | |
2835 | // | |
57f50b3c | 2836 | |
2bb7b766 | 2837 | // check connection, in case connect |
be48e3ea | 2838 | if(!Connect(3)){ |
2bb7b766 | 2839 | Log("SHUTTLE", "UpdateShuttleLogbook - Couldn't connect to DAQ Logbook."); |
2840 | return kFALSE; | |
57f50b3c | 2841 | } |
2842 | ||
2bb7b766 | 2843 | TString detName(detector); |
2844 | TString setClause; | |
8e828d39 | 2845 | if (detName == "shuttle_done" || detName == "shuttle_skipped") |
e7f62f16 | 2846 | { |
2bb7b766 | 2847 | setClause = "set shuttle_done=1"; |
8e828d39 | 2848 | |
ee6f7523 | 2849 | if (detName == "shuttle_done") |
b0e53b15 | 2850 | { |
8e828d39 | 2851 | if (TouchFile() != kTRUE) |
2156c755 | 2852 | { |
2853 | SendMLRunInfo("Pending"); | |
503ff24f | 2854 | return kFALSE; |
2156c755 | 2855 | } |
8e828d39 | 2856 | |
2857 | SendMLRunInfo("Done"); | |
b0e53b15 | 2858 | } |
8e828d39 | 2859 | else |
2860 | SendMLRunInfo("Skipped"); | |
2861 | } | |
2862 | else { | |
2bb7b766 | 2863 | TString statusStr(status); |
2864 | if(statusStr.Contains("done", TString::kIgnoreCase) || | |
2865 | statusStr.Contains("failed", TString::kIgnoreCase)){ | |
eba76848 | 2866 | setClause = Form("set %s=\"%s\"", detector, status); |
2bb7b766 | 2867 | } else { |
2868 | Log("SHUTTLE", | |
2869 | Form("UpdateShuttleLogbook - Invalid status <%s> for detector %s", | |
2870 | status, detector)); | |
2871 | return kFALSE; | |
2872 | } | |
2873 | } | |
57f50b3c | 2874 | |
2bb7b766 | 2875 | TString whereClause = Form("where run=%d", GetCurrentRun()); |
2876 | ||
441b0e9c | 2877 | TString sqlQuery = Form("update %s %s %s", |
2878 | fConfig->GetShuttlelbTable(), setClause.Data(), whereClause.Data()); | |
57f50b3c | 2879 | |
2bb7b766 | 2880 | AliDebug(2, Form("SQL query: \n%s",sqlQuery.Data())); |
2881 | ||
2882 | // Query execution | |
2883 | TSQLResult* aResult; | |
be48e3ea | 2884 | aResult = dynamic_cast<TSQLResult*> (fServer[3]->Query(sqlQuery)); |
2bb7b766 | 2885 | if (!aResult) { |
2886 | Log("SHUTTLE", Form("UpdateShuttleLogbook - Can't execute query <%s>", sqlQuery.Data())); | |
2887 | return kFALSE; | |
57f50b3c | 2888 | } |
2bb7b766 | 2889 | delete aResult; |
57f50b3c | 2890 | |
2891 | return kTRUE; | |
2892 | } | |
2893 | ||
2894 | //______________________________________________________________________________________________ | |
2bb7b766 | 2895 | Int_t AliShuttle::GetCurrentRun() const |
2896 | { | |
9827400b | 2897 | // |
2898 | // Get current run from logbook entry | |
2899 | // | |
57f50b3c | 2900 | |
2bb7b766 | 2901 | return fLogbookEntry ? fLogbookEntry->GetRun() : -1; |
57f50b3c | 2902 | } |
2903 | ||
2904 | //______________________________________________________________________________________________ | |
2bb7b766 | 2905 | UInt_t AliShuttle::GetCurrentStartTime() const |
2906 | { | |
9827400b | 2907 | // |
2908 | // get current start time | |
2909 | // | |
57f50b3c | 2910 | |
2bb7b766 | 2911 | return fLogbookEntry ? fLogbookEntry->GetStartTime() : 0; |
57f50b3c | 2912 | } |
2913 | ||
2914 | //______________________________________________________________________________________________ | |
2bb7b766 | 2915 | UInt_t AliShuttle::GetCurrentEndTime() const |
2916 | { | |
9827400b | 2917 | // |
2918 | // get current end time from logbook entry | |
2919 | // | |
57f50b3c | 2920 | |
2bb7b766 | 2921 | return fLogbookEntry ? fLogbookEntry->GetEndTime() : 0; |
57f50b3c | 2922 | } |
675f64cd | 2923 | //______________________________________________________________________________________________ |
2924 | UInt_t AliShuttle::GetCurrentYear() const | |
2925 | { | |
2926 | // | |
2927 | // Get current year from logbook entry | |
2928 | // | |
2929 | ||
2930 | if (!fLogbookEntry) return 0; | |
2931 | ||
2932 | TTimeStamp startTime(GetCurrentStartTime()); | |
2933 | TString year = Form("%d",startTime.GetDate()); | |
2934 | year = year(0,4); | |
2935 | ||
2936 | return year.Atoi(); | |
2937 | } | |
2938 | ||
2939 | //______________________________________________________________________________________________ | |
2940 | const char* AliShuttle::GetLHCPeriod() const | |
2941 | { | |
2942 | // | |
2943 | // Get current LHC period from logbook entry | |
2944 | // | |
2945 | ||
2946 | if (!fLogbookEntry) return 0; | |
2947 | ||
2948 | return fLogbookEntry->GetRunParameter("LHCperiod"); | |
2949 | } | |
2950 | ||
b948db8d | 2951 | //______________________________________________________________________________________________ |
6a150f83 | 2952 | void AliShuttle::Log(const char* detector, const char* message, UInt_t level) |
b948db8d | 2953 | { |
9827400b | 2954 | // |
2955 | // Fill log string with a message | |
2956 | // | |
7d4cf768 | 2957 | |
3ebbf3c4 | 2958 | TString logRunDir = GetShuttleLogDir(); |
2959 | if (GetCurrentRun() >=0) { | |
2960 | Int_t logDir_lev1 = GetCurrentRun()/10000; | |
2961 | logRunDir += Form("/%d/%d", logDir_lev1, GetCurrentRun()); | |
2962 | } | |
7d4cf768 | 2963 | void* dir = gSystem->OpenDirectory(logRunDir.Data()); |
84090f85 | 2964 | if (dir == NULL) { |
7d4cf768 | 2965 | if (gSystem->mkdir(logRunDir.Data(), kTRUE)) { |
36c99a6a | 2966 | AliError(Form("Can't open directory <%s>", GetShuttleLogDir())); |
84090f85 | 2967 | return; |
2968 | } | |
b948db8d | 2969 | |
84090f85 | 2970 | } else { |
2971 | gSystem->FreeDirectory(dir); | |
2972 | } | |
b948db8d | 2973 | |
0b025a52 | 2974 | TString toLog = Form("%s UTC (%d): %s - ", TTimeStamp(time(0)).AsString("s"), getpid(), detector); |
e7f62f16 | 2975 | if (GetCurrentRun() >= 0) |
2976 | toLog += Form("run %d - ", GetCurrentRun()); | |
2bb7b766 | 2977 | toLog += Form("%s", message); |
2978 | ||
6a150f83 | 2979 | AliLog::Message(level, toLog, MODULENAME(), ClassName(), FUNCTIONNAME(), __FILE__, __LINE__); |
ffa29e93 | 2980 | |
2981 | // if we redirect the log output already to the file, leave here | |
2982 | if (fOutputRedirected && strcmp(detector, "SHUTTLE") != 0) | |
2983 | return; | |
b948db8d | 2984 | |
ffa29e93 | 2985 | TString fileName = GetLogFileName(detector); |
e7f62f16 | 2986 | |
84090f85 | 2987 | gSystem->ExpandPathName(fileName); |
2988 | ||
2989 | ofstream logFile; | |
2990 | logFile.open(fileName, ofstream::out | ofstream::app); | |
2991 | ||
2992 | if (!logFile.is_open()) { | |
2993 | AliError(Form("Could not open file %s", fileName.Data())); | |
2994 | return; | |
2995 | } | |
7bfb2090 | 2996 | |
84090f85 | 2997 | logFile << toLog.Data() << "\n"; |
b948db8d | 2998 | |
84090f85 | 2999 | logFile.close(); |
b948db8d | 3000 | } |
2bb7b766 | 3001 | |
ffa29e93 | 3002 | //______________________________________________________________________________________________ |
3003 | TString AliShuttle::GetLogFileName(const char* detector) const | |
3004 | { | |
3005 | // | |
3006 | // returns the name of the log file for a given sub detector | |
3007 | // | |
3008 | ||
3009 | TString fileName; | |
3010 | ||
3011 | if (GetCurrentRun() >= 0) | |
7d4cf768 | 3012 | { |
3ebbf3c4 | 3013 | Int_t logDir_lev1 = GetCurrentRun()/10000; |
3014 | fileName.Form("%s/%d/%d/%s.log", GetShuttleLogDir(), logDir_lev1, GetCurrentRun(), | |
3015 | detector); | |
7d4cf768 | 3016 | } else { |
ffa29e93 | 3017 | fileName.Form("%s/%s.log", GetShuttleLogDir(), detector); |
7d4cf768 | 3018 | } |
ffa29e93 | 3019 | |
3020 | return fileName; | |
3021 | } | |
3022 | ||
ee6f7523 | 3023 | //______________________________________________________________________________________________ |
3024 | void AliShuttle::SendAlive() | |
3025 | { | |
3026 | // sends alive message to ML | |
3027 | ||
3028 | TMonaLisaText mlStatus("SHUTTLE_status", "Alive"); | |
3029 | ||
3030 | TList mlList; | |
3031 | mlList.Add(&mlStatus); | |
3032 | ||
3033 | fMonaLisa->SendParameters(&mlList, "__PROCESSINGINFO__"); | |
3034 | } | |
3035 | ||
2bb7b766 | 3036 | //______________________________________________________________________________________________ |
3037 | Bool_t AliShuttle::Collect(Int_t run) | |
3038 | { | |
9827400b | 3039 | // |
3040 | // Collects conditions data for all UNPROCESSED run written to DAQ LogBook in case of run = -1 (default) | |
3041 | // If a dedicated run is given this run is processed | |
3042 | // | |
3043 | // In operational mode, this is the Shuttle function triggered by the EOR signal. | |
3044 | // | |
2bb7b766 | 3045 | |
eba76848 | 3046 | if (run == -1) |
3047 | Log("SHUTTLE","Collect - Shuttle called. Collecting conditions data for unprocessed runs"); | |
3048 | else | |
3049 | Log("SHUTTLE", Form("Collect - Shuttle called. Collecting conditions data for run %d", run)); | |
cb343cfd | 3050 | |
3051 | SetLastAction("Starting"); | |
2bb7b766 | 3052 | |
ee6f7523 | 3053 | // create ML instance |
3054 | if (!fMonaLisa) | |
3055 | fMonaLisa = new TMonaLisaWriter(fConfig->GetMonitorHost(), fConfig->GetMonitorTable()); | |
3056 | ||
fbc112e3 | 3057 | CountOpenRuns(); |
ee6f7523 | 3058 | |
2bb7b766 | 3059 | TString whereClause("where shuttle_done=0"); |
eba76848 | 3060 | if (run != -1) |
3061 | whereClause += Form(" and run=%d", run); | |
2bb7b766 | 3062 | |
3063 | TObjArray shuttleLogbookEntries; | |
be48e3ea | 3064 | if (!QueryShuttleLogbook(whereClause, shuttleLogbookEntries)) |
3065 | { | |
cb343cfd | 3066 | Log("SHUTTLE", "Collect - Can't retrieve entries from Shuttle logbook"); |
2bb7b766 | 3067 | return kFALSE; |
3068 | } | |
3069 | ||
9e080f92 | 3070 | if (shuttleLogbookEntries.GetEntries() == 0) |
3071 | { | |
3072 | if (run == -1) | |
3073 | Log("SHUTTLE","Collect - Found no UNPROCESSED runs in Shuttle logbook"); | |
3074 | else | |
3075 | Log("SHUTTLE", Form("Collect - Run %d is already DONE " | |
3076 | "or it does not exist in Shuttle logbook", run)); | |
3077 | return kTRUE; | |
3078 | } | |
3079 | ||
be48e3ea | 3080 | for (UInt_t iDet=0; iDet<NDetectors(); iDet++) |
3081 | fFirstUnprocessed[iDet] = kTRUE; | |
3082 | ||
fc5a4708 | 3083 | if (run != -1) |
be48e3ea | 3084 | { |
3085 | // query Shuttle logbook for earlier runs, check if some detectors are unprocessed, | |
3086 | // flag them into fFirstUnprocessed array | |
e3ae8d33 | 3087 | TString whereClauseBis(Form("where shuttle_done=0 and run < %d", run)); |
be48e3ea | 3088 | TObjArray tmpLogbookEntries; |
e3ae8d33 | 3089 | if (!QueryShuttleLogbook(whereClauseBis, tmpLogbookEntries)) |
be48e3ea | 3090 | { |
3091 | Log("SHUTTLE", "Collect - Can't retrieve entries from Shuttle logbook"); | |
3092 | return kFALSE; | |
3093 | } | |
3094 | ||
3095 | TIter iter(&tmpLogbookEntries); | |
3096 | AliShuttleLogbookEntry* anEntry = 0; | |
3097 | while ((anEntry = dynamic_cast<AliShuttleLogbookEntry*> (iter.Next()))) | |
3098 | { | |
3099 | for (UInt_t iDet=0; iDet<NDetectors(); iDet++) | |
3100 | { | |
3101 | if (anEntry->GetDetectorStatus(iDet) == AliShuttleLogbookEntry::kUnprocessed) | |
3102 | { | |
3103 | AliDebug(2, Form("Run %d: setting %s as \"not first time unprocessed\"", | |
3104 | anEntry->GetRun(), GetDetName(iDet))); | |
3105 | fFirstUnprocessed[iDet] = kFALSE; | |
3106 | } | |
3107 | } | |
3108 | ||
3109 | } | |
3110 | ||
3111 | } | |
3112 | ||
3113 | if (!RetrieveConditionsData(shuttleLogbookEntries)) | |
3114 | { | |
cb343cfd | 3115 | Log("SHUTTLE", "Collect - Process of at least one run failed"); |
a3ca69fe | 3116 | CountOpenRuns(); |
2bb7b766 | 3117 | return kFALSE; |
3118 | } | |
3119 | ||
36c99a6a | 3120 | Log("SHUTTLE", "Collect - Requested run(s) successfully processed"); |
a3ca69fe | 3121 | CountOpenRuns(); |
eba76848 | 3122 | return kTRUE; |
2bb7b766 | 3123 | } |
3124 | ||
2bb7b766 | 3125 | //______________________________________________________________________________________________ |
3126 | Bool_t AliShuttle::RetrieveConditionsData(const TObjArray& dateEntries) | |
3127 | { | |
9827400b | 3128 | // |
3129 | // Retrieve conditions data for all runs that aren't processed yet | |
3130 | // | |
2bb7b766 | 3131 | |
3132 | Bool_t hasError = kFALSE; | |
3133 | ||
3134 | TIter iter(&dateEntries); | |
3135 | AliShuttleLogbookEntry* anEntry; | |
3136 | ||
3137 | while ((anEntry = (AliShuttleLogbookEntry*) iter.Next())){ | |
3138 | if (!Process(anEntry)){ | |
3139 | hasError = kTRUE; | |
3140 | } | |
4b95672b | 3141 | |
3142 | // clean SHUTTLE temp directory | |
d524ade6 | 3143 | //TString filename = Form("%s/*.shuttle", GetShuttleTempDir()); |
3144 | //RemoveFile(filename.Data()); | |
2bb7b766 | 3145 | } |
3146 | ||
3147 | return hasError == kFALSE; | |
3148 | } | |
cb343cfd | 3149 | |
3150 | //______________________________________________________________________________________________ | |
3151 | ULong_t AliShuttle::GetTimeOfLastAction() const | |
3152 | { | |
9827400b | 3153 | // |
3154 | // Gets time of last action | |
3155 | // | |
3156 | ||
cb343cfd | 3157 | ULong_t tmp; |
36c99a6a | 3158 | |
cb343cfd | 3159 | fMonitoringMutex->Lock(); |
be48e3ea | 3160 | |
cb343cfd | 3161 | tmp = fLastActionTime; |
36c99a6a | 3162 | |
cb343cfd | 3163 | fMonitoringMutex->UnLock(); |
36c99a6a | 3164 | |
cb343cfd | 3165 | return tmp; |
3166 | } | |
3167 | ||
3168 | //______________________________________________________________________________________________ | |
3169 | const TString AliShuttle::GetLastAction() const | |
3170 | { | |
9827400b | 3171 | // |
cb343cfd | 3172 | // returns a string description of the last action |
9827400b | 3173 | // |
cb343cfd | 3174 | |
3175 | TString tmp; | |
36c99a6a | 3176 | |
cb343cfd | 3177 | fMonitoringMutex->Lock(); |
3178 | ||
3179 | tmp = fLastAction; | |
3180 | ||
3181 | fMonitoringMutex->UnLock(); | |
3182 | ||
36c99a6a | 3183 | return tmp; |
cb343cfd | 3184 | } |
3185 | ||
3186 | //______________________________________________________________________________________________ | |
3187 | void AliShuttle::SetLastAction(const char* action) | |
3188 | { | |
9827400b | 3189 | // |
cb343cfd | 3190 | // updates the monitoring variables |
9827400b | 3191 | // |
36c99a6a | 3192 | |
cb343cfd | 3193 | fMonitoringMutex->Lock(); |
36c99a6a | 3194 | |
cb343cfd | 3195 | fLastAction = action; |
3196 | fLastActionTime = time(0); | |
3197 | ||
3198 | fMonitoringMutex->UnLock(); | |
3199 | } | |
eba76848 | 3200 | |
3201 | //______________________________________________________________________________________________ | |
3202 | const char* AliShuttle::GetRunParameter(const char* param) | |
3203 | { | |
9827400b | 3204 | // |
3205 | // returns run parameter read from DAQ logbook | |
3206 | // | |
eba76848 | 3207 | |
3208 | if(!fLogbookEntry) { | |
3209 | AliError("No logbook entry!"); | |
3210 | return 0; | |
3211 | } | |
3212 | ||
3213 | return fLogbookEntry->GetRunParameter(param); | |
3214 | } | |
57c1a579 | 3215 | |
d386d623 | 3216 | //______________________________________________________________________________________________ |
9827400b | 3217 | AliCDBEntry* AliShuttle::GetFromOCDB(const char* detector, const AliCDBPath& path) |
d386d623 | 3218 | { |
9827400b | 3219 | // |
3220 | // returns object from OCDB valid for current run | |
3221 | // | |
d386d623 | 3222 | |
9827400b | 3223 | if (fTestMode & kErrorOCDB) |
3224 | { | |
3225 | Log(detector, "GetFromOCDB - In TESTMODE - Simulating error with OCDB"); | |
3226 | return 0; | |
3227 | } | |
3228 | ||
d386d623 | 3229 | AliCDBStorage *sto = AliCDBManager::Instance()->GetStorage(fgkMainCDB); |
3230 | if (!sto) | |
3231 | { | |
9827400b | 3232 | Log(detector, "GetFromOCDB - Cannot activate main OCDB for query!"); |
d386d623 | 3233 | return 0; |
3234 | } | |
3235 | ||
3236 | return dynamic_cast<AliCDBEntry*> (sto->Get(path, GetCurrentRun())); | |
3237 | } | |
3238 | ||
57c1a579 | 3239 | //______________________________________________________________________________________________ |
339adafa | 3240 | Bool_t AliShuttle::SendMail(EMailTarget target, Int_t system) |
57c1a579 | 3241 | { |
9827400b | 3242 | // |
3243 | // sends a mail to the subdetector expert in case of preprocessor error | |
3244 | // | |
3245 | ||
3246 | if (fTestMode != kNone) | |
3247 | return kTRUE; | |
fb40ead4 | 3248 | |
926cbc0e | 3249 | if (!fConfig->SendMail()) |
3250 | return kTRUE; | |
57c1a579 | 3251 | |
339adafa | 3252 | if (target == kDCSEMail || target == kFXSEMail) { |
3253 | if (!fFirstProcessing) | |
3a4ddf0e | 3254 | return kTRUE; |
36c99a6a | 3255 | } |
3256 | ||
7bdfe1fa | 3257 | Int_t runMode = (Int_t)fConfig->GetRunMode(); |
3258 | TString tmpStr; | |
3259 | if (runMode == 0) tmpStr = " Nightly Test:"; | |
3260 | else tmpStr = " Data Taking:"; | |
fbc112e3 | 3261 | void* dir = gSystem->OpenDirectory(GetShuttleLogDir()); |
3262 | if (dir == NULL) | |
3263 | { | |
3264 | if (gSystem->mkdir(GetShuttleLogDir(), kTRUE)) | |
3265 | { | |
3266 | Log("SHUTTLE", Form("SendMail - Can't open directory <%s>", GetShuttleLogDir())); | |
3267 | return kFALSE; | |
3268 | } | |
3269 | ||
3270 | } else { | |
3271 | gSystem->FreeDirectory(dir); | |
3272 | } | |
3273 | ||
339adafa | 3274 | // det experts in to |
3275 | TString to=""; | |
3835a613 | 3276 | TIter *iterExperts = 0; |
339adafa | 3277 | if (target == kDCSEMail) { |
3278 | iterExperts = new TIter(fConfig->GetAdmins(AliShuttleConfig::kAmanda)); | |
3279 | } | |
3280 | else if (target == kFXSEMail) { | |
3281 | iterExperts = new TIter(fConfig->GetAdmins(system)); | |
3282 | } | |
3835a613 | 3283 | if (iterExperts) { |
3284 | TObjString *anExpert=0; | |
3285 | while ((anExpert = (TObjString*) iterExperts->Next())) | |
3286 | { | |
3287 | to += Form("%s,", anExpert->GetName()); | |
3288 | } | |
3289 | delete iterExperts; | |
339adafa | 3290 | } |
3835a613 | 3291 | |
3292 | // add subdetector experts | |
3293 | iterExperts = new TIter(fConfig->GetResponsibles(fCurrentDetector)); | |
339adafa | 3294 | TObjString *anExpert=0; |
3295 | while ((anExpert = (TObjString*) iterExperts->Next())) | |
4508ab8b | 3296 | { |
339adafa | 3297 | to += Form("%s,", anExpert->GetName()); |
3298 | } | |
3299 | delete iterExperts; | |
3835a613 | 3300 | |
339adafa | 3301 | if (to.Length() > 0) |
3302 | to.Remove(to.Length()-1); | |
3303 | AliDebug(2, Form("to: %s",to.Data())); | |
4508ab8b | 3304 | |
339adafa | 3305 | if (to.IsNull()) { |
3306 | Log("SHUTTLE", Form("List of %d responsibles not set!", (Int_t) target)); | |
3307 | return kFALSE; | |
3308 | } | |
3309 | ||
3310 | // SHUTTLE responsibles in cc | |
fb40ead4 | 3311 | TString cc=""; |
3312 | TIter iterAdmins(fConfig->GetAdmins(AliShuttleConfig::kGlobal)); | |
3313 | TObjString *anAdmin=0; | |
3314 | while ((anAdmin = (TObjString*) iterAdmins.Next())) | |
3315 | { | |
3316 | cc += Form("%s,", anAdmin->GetName()); | |
3317 | } | |
3318 | if (cc.Length() > 0) | |
c19963db | 3319 | cc.Remove(cc.Length()-1); |
fb40ead4 | 3320 | AliDebug(2, Form("cc: %s",to.Data())); |
57c1a579 | 3321 | |
339adafa | 3322 | // mail body |
675f64cd | 3323 | TString bodyFileName; |
3324 | bodyFileName.Form("%s/mail.body", GetShuttleLogDir()); | |
3325 | gSystem->ExpandPathName(bodyFileName); | |
3326 | ||
3327 | ofstream mailBody; | |
3328 | mailBody.open(bodyFileName, ofstream::out); | |
3329 | ||
3330 | if (!mailBody.is_open()) | |
3331 | { | |
339adafa | 3332 | Log("SHUTTLE", Form("Could not open mail body file %s", bodyFileName.Data())); |
675f64cd | 3333 | return kFALSE; |
3334 | } | |
3335 | ||
675f64cd | 3336 | |
339adafa | 3337 | TString subject; |
3338 | TString body; | |
675f64cd | 3339 | |
339adafa | 3340 | if (target == kDCSEMail){ |
af93e5d7 | 3341 | subject = Form("%s CRITICAL Retrieval of data points for %s FAILED in run %d !", |
7bdfe1fa | 3342 | tmpStr.Data(), fCurrentDetector.Data(), GetCurrentRun()); |
339adafa | 3343 | AliDebug(2, Form("subject: %s", subject.Data())); |
3344 | ||
3345 | body = Form("Dear DCS experts, \n\n"); | |
3346 | body += Form("SHUTTLE couldn\'t retrieve the data points for detector %s " | |
3347 | "in run %d!!\n\n", fCurrentDetector.Data(), GetCurrentRun()); | |
fb40ead4 | 3348 | } |
339adafa | 3349 | else if (target == kFXSEMail){ |
af93e5d7 | 3350 | subject = Form("%s CRITICAL FXS communication for %s FAILED in run %d !", |
7bdfe1fa | 3351 | tmpStr.Data(), fCurrentDetector.Data(), GetCurrentRun()); |
339adafa | 3352 | AliDebug(2, Form("subject: %s", subject.Data())); |
3353 | TString sys; | |
3354 | if (system == kDAQ) sys="DAQ"; | |
3355 | else if (system == kDCS) sys="DCS"; | |
3356 | else if (system == kHLT) sys="HLT"; | |
3357 | else return kFALSE; | |
9c0e2d64 | 3358 | body = Form("Dear %s FXS experts, \n\n",sys.Data()); |
339adafa | 3359 | body += Form("SHUTTLE couldn\'t retrieve data from the FXS for detector %s " |
9c0e2d64 | 3360 | "in run %d!!\n\n", fCurrentDetector.Data(), GetCurrentRun()); |
8662f0a8 | 3361 | body += Form("The contacted server was:\nDB: %s\nFXS:%s\n\n", fConfig->GetFXSdbHost(system), fConfig->GetFXSHost(system)); |
339adafa | 3362 | } |
3363 | else { | |
7bdfe1fa | 3364 | subject = Form("%s %s Shuttle preprocessor FAILED in run %d (run type = %s)!", |
3365 | tmpStr.Data(), fCurrentDetector.Data(), GetCurrentRun(), GetRunType()); | |
339adafa | 3366 | AliDebug(2, Form("subject: %s", subject.Data())); |
3367 | ||
3368 | body = Form("Dear %s expert(s), \n\n", fCurrentDetector.Data()); | |
3369 | body += Form("SHUTTLE just detected that your preprocessor " | |
3370 | "failed processing run %d (run type = %s)!!\n\n", | |
3371 | GetCurrentRun(), GetRunType()); | |
3372 | } | |
675f64cd | 3373 | |
7d4cf768 | 3374 | body += Form("Please check %s status on the SHUTTLE monitoring page: \n\n", |
3375 | fCurrentDetector.Data()); | |
b0e53b15 | 3376 | if (fConfig->GetRunMode() == AliShuttleConfig::kTest) |
3377 | { | |
8e828d39 | 3378 | body += Form("\thttp://pcalimonitor.cern.ch/shuttle.jsp?time=24 \n\n"); |
b0e53b15 | 3379 | } else { |
8e828d39 | 3380 | body += Form("\thttp://pcalimonitor.cern.ch/shuttle.jsp?instance=PROD&time=24 \n\n"); |
b0e53b15 | 3381 | } |
339adafa | 3382 | |
3383 | ||
7d4cf768 | 3384 | TString logFolder = "logs"; |
3385 | if (fConfig->GetRunMode() == AliShuttleConfig::kProd) | |
3386 | logFolder += "_PROD"; | |
3387 | ||
3388 | ||
675f64cd | 3389 | body += Form("Find the %s log for the current run on \n\n" |
29fb87f4 | 3390 | "\thttp://pcalishuttle02.cern.ch/%s/%d/%d/%s.log \n\n", |
3ebbf3c4 | 3391 | fCurrentDetector.Data(), logFolder.Data(), GetCurrentRun()/10000, |
3392 | GetCurrentRun(), fCurrentDetector.Data()); | |
fe6a3257 | 3393 | body += Form("The last 15 lines of %s log file are following:\n\n", fCurrentDetector.Data()); |
675f64cd | 3394 | |
3395 | AliDebug(2, Form("Body begin: %s", body.Data())); | |
3396 | ||
3397 | mailBody << body.Data(); | |
3398 | mailBody.close(); | |
3399 | mailBody.open(bodyFileName, ofstream::out | ofstream::app); | |
3400 | ||
3ebbf3c4 | 3401 | TString logFileName = Form("%s/%d/%d/%s.log", GetShuttleLogDir(), |
3402 | GetCurrentRun()/10000, GetCurrentRun(), fCurrentDetector.Data()); | |
fe6a3257 | 3403 | TString tailCommand = Form("tail -n 15 %s >> %s", logFileName.Data(), bodyFileName.Data()); |
675f64cd | 3404 | if (gSystem->Exec(tailCommand.Data())) |
3405 | { | |
3406 | mailBody << Form("%s log file not found ...\n\n", fCurrentDetector.Data()); | |
3407 | } | |
3408 | ||
3409 | TString endBody = Form("------------------------------------------------------\n\n"); | |
36c99a6a | 3410 | endBody += Form("In case of problems please contact the SHUTTLE core team.\n\n"); |
3411 | endBody += "Please do not answer this message directly, it is automatically generated.\n\n"; | |
546242fb | 3412 | endBody += "Greetings,\n\n \t\t\tthe SHUTTLE\n"; |
57c1a579 | 3413 | |
909732f7 | 3414 | AliDebug(2, Form("Body end: %s", endBody.Data())); |
57c1a579 | 3415 | |
3416 | mailBody << endBody.Data(); | |
3417 | ||
3418 | mailBody.close(); | |
3419 | ||
3420 | // send mail! | |
3421 | TString mailCommand = Form("mail -s \"%s\" -c %s %s < %s", | |
3422 | subject.Data(), | |
3423 | cc.Data(), | |
3424 | to.Data(), | |
3425 | bodyFileName.Data()); | |
909732f7 | 3426 | AliDebug(2, Form("mail command: %s", mailCommand.Data())); |
57c1a579 | 3427 | |
3428 | Bool_t result = gSystem->Exec(mailCommand.Data()); | |
3429 | ||
3430 | return result == 0; | |
3431 | } | |
441b0e9c | 3432 | //______________________________________________________________________________________________ |
9827400b | 3433 | const char* AliShuttle::GetRunType() |
441b0e9c | 3434 | { |
9827400b | 3435 | // |
3436 | // returns run type read from "run type" logbook | |
3437 | // | |
441b0e9c | 3438 | |
3439 | if(!fLogbookEntry) { | |
3440 | AliError("No logbook entry!"); | |
3441 | return 0; | |
3442 | } | |
3443 | ||
9827400b | 3444 | return fLogbookEntry->GetRunType(); |
441b0e9c | 3445 | } |
3446 | ||
4859271b | 3447 | //______________________________________________________________________________________________ |
3448 | Bool_t AliShuttle::GetHLTStatus() | |
3449 | { | |
3450 | // Return HLT status (ON=1 OFF=0) | |
af76b20d | 3451 | // Converts the HLT status from the mode string read in the run logbook (not just a bool) |
4859271b | 3452 | |
3453 | if(!fLogbookEntry) { | |
3454 | AliError("No logbook entry!"); | |
3455 | return 0; | |
3456 | } | |
3457 | ||
af76b20d | 3458 | // TODO implement when HLTMode is inserted in run logbook |
b4ebc2d5 | 3459 | TString hltMode = fLogbookEntry->GetRunParameter("HLTmode"); |
af76b20d | 3460 | TSubString firstChar = hltMode(0,1); |
3461 | AliDebug(2,Form("First char = %s ",firstChar.Data())); | |
3462 | if (firstChar == "A") { | |
3463 | return kFALSE; | |
3464 | } | |
3465 | else if ((firstChar == "B") || (firstChar == "C") || (firstChar == "D") || (firstChar == "E")) { | |
3466 | return kTRUE; | |
3467 | } | |
3468 | else { | |
3469 | Log("SHUTTLE","Unexpected HLT mode! Returning 0...."); | |
3470 | return kFALSE; | |
3471 | } | |
4859271b | 3472 | } |
ca8ea066 | 3473 | |
3474 | //______________________________________________________________________________________________ | |
3475 | const char* AliShuttle::GetTriggerConfiguration() | |
3476 | { | |
3477 | // Receives the trigger configuration from the DAQ logbook for the current run | |
3478 | ||
3479 | // check connection, if needed reconnect | |
3480 | if (!Connect(3)) | |
3481 | return 0; | |
3482 | ||
3483 | TString sqlQuery; | |
3484 | sqlQuery.Form("SELECT configFile FROM logbook_trigger_config WHERE run = %d", GetCurrentRun()); | |
3485 | TSQLResult* result = fServer[3]->Query(sqlQuery); | |
3486 | if (!result) | |
3487 | { | |
3488 | Log("SHUTTLE", Form("ERROR: Can't execute query <%s>!", sqlQuery.Data())); | |
3489 | return 0; | |
3490 | } | |
3491 | ||
3492 | if (result->GetRowCount() == 0) | |
3493 | { | |
ff0191c7 | 3494 | Log("SHUTTLE", "WARNING: Trigger configuration not found in logbook_trigger_config"); |
ca8ea066 | 3495 | delete result; |
3496 | return 0; | |
3497 | } | |
3498 | ||
3499 | TSQLRow* row = result->Next(); | |
3500 | if (!row) | |
3501 | { | |
3502 | Log("SHUTTLE", "ERROR: Could not receive logbook_trigger_config data"); | |
3503 | delete result; | |
3504 | return 0; | |
3505 | } | |
3506 | ||
3507 | // static, so that pointer remains valid when it is returned to the calling class | |
3508 | static TString triggerConfig(row->GetField(0)); | |
3509 | ||
3510 | delete row; | |
3511 | row = 0; | |
3512 | ||
3513 | delete result; | |
3514 | result = 0; | |
3515 | ||
3516 | Log("SHUTTLE", Form("Found trigger configuration: %s", triggerConfig.Data())); | |
3517 | ||
3518 | return triggerConfig; | |
3519 | } | |
3520 | ||
2c5f9d06 | 3521 | //______________________________________________________________________________________________ |
3522 | const char* AliShuttle::GetCTPTimeParams() | |
3523 | { | |
3524 | // Receives the CTP time parameters from the DAQ logbook for the current run | |
3525 | ||
3526 | // check connection, if needed reconnect | |
3527 | if (!Connect(3)) | |
3528 | return 0; | |
3529 | ||
3530 | TString sqlQuery; | |
3531 | sqlQuery.Form("SELECT alignmentFile FROM logbook_trigger_config WHERE run = %d", GetCurrentRun()); | |
3532 | TSQLResult* result = fServer[3]->Query(sqlQuery); | |
3533 | if (!result) | |
3534 | { | |
3535 | Log("SHUTTLE", Form("ERROR: Can't execute query <%s>!", sqlQuery.Data())); | |
3536 | return 0; | |
3537 | } | |
3538 | ||
3539 | if (result->GetRowCount() == 0) | |
3540 | { | |
ff0191c7 | 3541 | Log("SHUTTLE", "WARNING: CTP time params not found in logbook_trigger_config"); |
2c5f9d06 | 3542 | delete result; |
3543 | return 0; | |
3544 | } | |
3545 | ||
3546 | TSQLRow* row = result->Next(); | |
3547 | if (!row) | |
3548 | { | |
3549 | Log("SHUTTLE", "ERROR: Could not receive logbook_trigger_config data"); | |
3550 | delete result; | |
3551 | return 0; | |
3552 | } | |
3553 | ||
3554 | // static, so that pointer remains valid when it is returned to the calling class | |
3555 | static TString triggerTimeParams(row->GetField(0)); | |
3556 | ||
3557 | delete row; | |
3558 | row = 0; | |
3559 | ||
3560 | delete result; | |
3561 | result = 0; | |
3562 | ||
a191c286 | 3563 | Log("SHUTTLE", Form("Found trigger time parameters: %s", triggerTimeParams.Data())); |
2c5f9d06 | 3564 | |
3565 | return triggerTimeParams; | |
3566 | } | |
3567 | ||
d386d623 | 3568 | //______________________________________________________________________________________________ |
b48d2542 | 3569 | const char* AliShuttle::GetTriggerDetectorMask() |
3570 | { | |
3571 | // Receives the trigger detector mask from DAQ logbook | |
3572 | ||
3573 | // check connection, if needed reconnect | |
3574 | if (!Connect(3)) | |
3575 | return 0; | |
3576 | ||
3577 | TString sqlQuery; | |
3578 | sqlQuery.Form("SELECT BIN(BIT_OR(inputDetectorMask)) from logbook_trigger_clusters WHERE run = %d;", GetCurrentRun()); | |
3579 | TSQLResult* result = fServer[3]->Query(sqlQuery); | |
3580 | if (!result) | |
3581 | { | |
3582 | Log("SHUTTLE", Form("ERROR: Can't execute query <%s>!", sqlQuery.Data())); | |
3583 | return 0; | |
3584 | } | |
3585 | ||
3586 | if (result->GetRowCount() == 0) | |
3587 | { | |
3588 | Log("SHUTTLE", "ERROR: Trigger Detector Mask not found in logbook_trigger_clusters"); | |
3589 | delete result; | |
3590 | return 0; | |
3591 | } | |
3592 | ||
3593 | TSQLRow* row = result->Next(); | |
3594 | if (!row) | |
3595 | { | |
3596 | Log("SHUTTLE", "ERROR: Could not receive logbook_trigger_clusters data"); | |
3597 | delete result; | |
3598 | return 0; | |
3599 | } | |
3600 | ||
3601 | // static, so that pointer remains valid when it is returned to the calling class | |
3602 | static TString triggerDetectorMask(row->GetField(0)); | |
3603 | ||
3604 | delete row; | |
3605 | row = 0; | |
3606 | ||
3607 | delete result; | |
3608 | result = 0; | |
3609 | ||
2e7d9963 | 3610 | Log("SHUTTLE", Form("Found Trigger Detector Mask: %s", triggerDetectorMask.Data())); |
b48d2542 | 3611 | |
3612 | return triggerDetectorMask; | |
3613 | } | |
3614 | ||
3615 | //______________________________________________________________________________________________ | |
d386d623 | 3616 | void AliShuttle::SetShuttleTempDir(const char* tmpDir) |
3617 | { | |
9827400b | 3618 | // |
3619 | // sets Shuttle temp directory | |
3620 | // | |
d386d623 | 3621 | |
3622 | fgkShuttleTempDir = gSystem->ExpandPathName(tmpDir); | |
3623 | } | |
3624 | ||
3625 | //______________________________________________________________________________________________ | |
3626 | void AliShuttle::SetShuttleLogDir(const char* logDir) | |
3627 | { | |
9827400b | 3628 | // |
3629 | // sets Shuttle log directory | |
3630 | // | |
d386d623 | 3631 | |
3632 | fgkShuttleLogDir = gSystem->ExpandPathName(logDir); | |
3633 | } | |
503ff24f | 3634 | //______________________________________________________________________________________________ |
3635 | Bool_t AliShuttle::TouchFile() | |
3636 | { | |
3637 | // | |
3638 | // touching a file on the grid if run has been DONE | |
3639 | // | |
3640 | ||
3641 | if (!gGrid) | |
3642 | { | |
3643 | Log("SHUTTLE",Form("No TGrid connection estabilished!")); | |
3644 | Log("SHUTTLE",Form("Could not touch file for run %i",GetCurrentRun())); | |
3645 | return kFALSE; | |
3646 | } | |
3647 | ||
b8cf5e0f | 3648 | TString dir; |
728290a4 | 3649 | dir.Form("%s%d/SHUTTLE_DONE", fConfig->GetAlienPath(), GetCurrentYear()); |
b8cf5e0f | 3650 | // checking whether directory for touch command exists |
3651 | TString commandLs; | |
3652 | commandLs.Form("ls %s",dir.Data()); | |
3653 | TGridResult *resultLs = dynamic_cast<TGridResult*>(gGrid->Command(commandLs)); | |
3654 | if (!resultLs){ | |
3655 | Log("SHUTTLE",Form("No result for %s command, returning without touching",commandLs.Data())); | |
3656 | return kFALSE; | |
3657 | } | |
3658 | TMap *mapLs = dynamic_cast<TMap*>(resultLs->At(0)); | |
3659 | if (!mapLs){ | |
3660 | Log("SHUTTLE",Form("No map for %s command, returning without touching",commandLs.Data())); | |
97d38a5a | 3661 | delete resultLs; |
3662 | resultLs = 0x0; | |
b8cf5e0f | 3663 | return kFALSE; |
3664 | } | |
3665 | TObjString *valueLsPath = dynamic_cast<TObjString*>(mapLs->GetValue("path")); | |
eff13ba0 | 3666 | if (!valueLsPath || (valueLsPath->GetString()).CompareTo(dir)!=1){ |
b8cf5e0f | 3667 | Log("SHUTTLE",Form("No directory %s found, creating it",dir.Data())); |
3668 | ||
3669 | // creating the directory | |
3670 | ||
3671 | Bool_t boolMkdir = gGrid->Mkdir(dir.Data()); | |
3672 | if (!boolMkdir) { | |
3673 | Log("SHUTTLE",Form("Impossible to create dir %s in alien catalogue for run %i!",dir.Data(),GetCurrentRun())); | |
97d38a5a | 3674 | delete resultLs; |
3675 | resultLs = 0x0; | |
b8cf5e0f | 3676 | return kFALSE; |
3677 | } | |
3678 | Log("SHUTTLE",Form("Directory %s successfully created in alien catalogue for run %i",dir.Data(),GetCurrentRun())); | |
3679 | } | |
3680 | else { | |
3681 | Log("SHUTTLE",Form("Directory %s correctly found for run %i",dir.Data(),GetCurrentRun())); | |
3682 | } | |
3683 | ||
97d38a5a | 3684 | delete resultLs; |
3685 | resultLs = 0x0; | |
3686 | ||
503ff24f | 3687 | TString command; |
b8cf5e0f | 3688 | command.Form("touch %s/%i", dir.Data(), GetCurrentRun()); |
e660ff60 | 3689 | Log("SHUTTLE", Form("Creating entry in file catalog: %s", command.Data())); |
503ff24f | 3690 | TGridResult *resultTouch = dynamic_cast<TGridResult*>(gGrid->Command(command)); |
b8cf5e0f | 3691 | if (!resultTouch){ |
3692 | Log("SHUTTLE",Form("No result for touching command, returning without touching for run %i",GetCurrentRun())); | |
3693 | return kFALSE; | |
503ff24f | 3694 | } |
b8cf5e0f | 3695 | TMap *mapTouch = dynamic_cast<TMap*>(resultTouch->At(0)); |
3696 | if (!mapTouch){ | |
3697 | Log("SHUTTLE",Form("No map for touching command, returning without touching for run %i",GetCurrentRun())); | |
97d38a5a | 3698 | delete resultTouch; |
3699 | resultTouch = 0x0; | |
b8cf5e0f | 3700 | return kFALSE; |
3701 | } | |
3702 | TObjString *valueTouch = dynamic_cast<TObjString*>(mapTouch->GetValue("__result__")); | |
3703 | if (!valueTouch){ | |
3704 | Log("SHUTTLE",Form("No value for \"__result__\" key set in the map for touching command, returning without touching for run %i",GetCurrentRun())); | |
97d38a5a | 3705 | delete resultTouch; |
3706 | resultTouch = 0x0; | |
b8cf5e0f | 3707 | return kFALSE; |
503ff24f | 3708 | } |
b8cf5e0f | 3709 | if (valueTouch->GetString()!="1"){ |
3710 | Log("SHUTTLE",Form("Failing the touching command, returning without touching for run %i",GetCurrentRun())); | |
97d38a5a | 3711 | delete resultTouch; |
3712 | resultTouch = 0x0; | |
b8cf5e0f | 3713 | return kFALSE; |
3714 | } | |
97d38a5a | 3715 | delete resultTouch; |
3716 | resultTouch = 0x0; | |
98400bd9 | 3717 | Log("SHUTTLE", "Sucessfully touched the file"); |
b8cf5e0f | 3718 | return kTRUE; |
503ff24f | 3719 | } |
2eb98462 | 3720 | //______________________________________________________________________________________________ |
f100adeb | 3721 | UInt_t AliShuttle::GetStartTimeDCSQuery() |
2eb98462 | 3722 | { |
3723 | // Return Start Time for the DCS query | |
3724 | // | |
3725 | // The call is delegated to AliShuttleInterface | |
3726 | ||
3d24ad5a | 3727 | return GetCurrentStartTime()-fConfig->GetDCSQueryOffset(); |
2eb98462 | 3728 | } |
3729 | //______________________________________________________________________________________________ | |
f100adeb | 3730 | UInt_t AliShuttle::GetEndTimeDCSQuery() |
2eb98462 | 3731 | { |
3732 | // Return End Time for the DCS query | |
3733 | // | |
3734 | // The call is delegated to AliShuttleInterface | |
3735 | ||
3736 | return GetCurrentEndTime()+fConfig->GetDCSQueryOffset(); | |
3737 | } | |
503ff24f | 3738 |