]>
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 | ||
16 | /* | |
17 | $Log$ | |
fc5a4708 | 18 | Revision 1.21 2006/12/07 08:51:26 jgrosseo |
19 | update (alberto): | |
20 | table, db names in ldap configuration | |
21 | added GRP preprocessor | |
22 | DCS data can also be retrieved by data point | |
23 | ||
2c15234c | 24 | Revision 1.20 2006/11/16 16:16:48 jgrosseo |
25 | introducing strict run ordering flag | |
26 | removed giving preprocessor name to preprocessor, they have to know their name themselves ;-) | |
27 | ||
be48e3ea | 28 | Revision 1.19 2006/11/06 14:23:04 jgrosseo |
29 | major update (Alberto) | |
30 | o) reading of run parameters from the logbook | |
31 | o) online offline naming conversion | |
32 | o) standalone DCSclient package | |
33 | ||
eba76848 | 34 | Revision 1.18 2006/10/20 15:22:59 jgrosseo |
35 | o) Adding time out to the execution of the preprocessors: The Shuttle forks and the parent process monitors the child | |
36 | o) Merging Collect, CollectAll, CollectNew function | |
37 | o) Removing implementation of empty copy constructors (declaration still there!) | |
38 | ||
cb343cfd | 39 | Revision 1.17 2006/10/05 16:20:55 jgrosseo |
40 | adapting to new CDB classes | |
41 | ||
6ec0e06c | 42 | Revision 1.16 2006/10/05 15:46:26 jgrosseo |
43 | applying to the new interface | |
44 | ||
481441a2 | 45 | Revision 1.15 2006/10/02 16:38:39 jgrosseo |
46 | update (alberto): | |
47 | fixed memory leaks | |
48 | storing of objects that failed to be stored to the grid before | |
49 | interfacing of shuttle status table in daq system | |
50 | ||
2bb7b766 | 51 | Revision 1.14 2006/08/29 09:16:05 jgrosseo |
52 | small update | |
53 | ||
85a80aa9 | 54 | Revision 1.13 2006/08/15 10:50:00 jgrosseo |
55 | effc++ corrections (alberto) | |
56 | ||
4f0ab988 | 57 | Revision 1.12 2006/08/08 14:19:29 jgrosseo |
58 | Update to shuttle classes (Alberto) | |
59 | ||
60 | - Possibility to set the full object's path in the Preprocessor's and | |
61 | Shuttle's Store functions | |
62 | - Possibility to extend the object's run validity in the same classes | |
63 | ("startValidity" and "validityInfinite" parameters) | |
64 | - Implementation of the StoreReferenceData function to store reference | |
65 | data in a dedicated CDB storage. | |
66 | ||
84090f85 | 67 | Revision 1.11 2006/07/21 07:37:20 jgrosseo |
68 | last run is stored after each run | |
69 | ||
7bfb2090 | 70 | Revision 1.10 2006/07/20 09:54:40 jgrosseo |
71 | introducing status management: The processing per subdetector is divided into several steps, | |
72 | after each step the status is stored on disk. If the system crashes in any of the steps the Shuttle | |
73 | can keep track of the number of failures and skips further processing after a certain threshold is | |
74 | exceeded. These thresholds can be configured in LDAP. | |
75 | ||
5164a766 | 76 | Revision 1.9 2006/07/19 10:09:55 jgrosseo |
77 | new configuration, accesst to DAQ FES (Alberto) | |
78 | ||
57f50b3c | 79 | Revision 1.8 2006/07/11 12:44:36 jgrosseo |
80 | adding parameters for extended validity range of data produced by preprocessor | |
81 | ||
17111222 | 82 | Revision 1.7 2006/07/10 14:37:09 jgrosseo |
83 | small fix + todo comment | |
84 | ||
e090413b | 85 | Revision 1.6 2006/07/10 13:01:41 jgrosseo |
86 | enhanced storing of last sucessfully processed run (alberto) | |
87 | ||
a7160fe9 | 88 | Revision 1.5 2006/07/04 14:59:57 jgrosseo |
89 | revision of AliDCSValue: Removed wrapper classes, reduced storage size per value by factor 2 | |
90 | ||
45a493ce | 91 | Revision 1.4 2006/06/12 09:11:16 jgrosseo |
92 | coding conventions (Alberto) | |
93 | ||
58bc3020 | 94 | Revision 1.3 2006/06/06 14:26:40 jgrosseo |
95 | o) removed files that were moved to STEER | |
96 | o) shuttle updated to follow the new interface (Alberto) | |
97 | ||
b948db8d | 98 | Revision 1.2 2006/03/07 07:52:34 hristov |
99 | New version (B.Yordanov) | |
100 | ||
d477ad88 | 101 | Revision 1.6 2005/11/19 17:19:14 byordano |
102 | RetrieveDATEEntries and RetrieveConditionsData added | |
103 | ||
104 | Revision 1.5 2005/11/19 11:09:27 byordano | |
105 | AliShuttle declaration added | |
106 | ||
107 | Revision 1.4 2005/11/17 17:47:34 byordano | |
108 | TList changed to TObjArray | |
109 | ||
110 | Revision 1.3 2005/11/17 14:43:23 byordano | |
111 | import to local CVS | |
112 | ||
113 | Revision 1.1.1.1 2005/10/28 07:33:58 hristov | |
114 | Initial import as subdirectory in AliRoot | |
115 | ||
73abe331 | 116 | Revision 1.2 2005/09/13 08:41:15 byordano |
117 | default startTime endTime added | |
118 | ||
119 | Revision 1.4 2005/08/30 09:13:02 byordano | |
120 | some docs added | |
121 | ||
122 | Revision 1.3 2005/08/29 21:15:47 byordano | |
123 | some docs added | |
124 | ||
125 | */ | |
126 | ||
127 | // | |
128 | // This class is the main manager for AliShuttle. | |
129 | // It organizes the data retrieval from DCS and call the | |
b948db8d | 130 | // interface methods of AliPreprocessor. |
73abe331 | 131 | // For every detector in AliShuttleConfgi (see AliShuttleConfig), |
132 | // data for its set of aliases is retrieved. If there is registered | |
b948db8d | 133 | // AliPreprocessor for this detector then it will be used |
134 | // accroding to the schema (see AliPreprocessor). | |
135 | // If there isn't registered AliPreprocessor than the retrieved | |
73abe331 | 136 | // data is stored automatically to the undelying AliCDBStorage. |
137 | // For detSpec is used the alias name. | |
138 | // | |
139 | ||
140 | #include "AliShuttle.h" | |
141 | ||
142 | #include "AliCDBManager.h" | |
143 | #include "AliCDBStorage.h" | |
144 | #include "AliCDBId.h" | |
84090f85 | 145 | #include "AliCDBRunRange.h" |
146 | #include "AliCDBPath.h" | |
5164a766 | 147 | #include "AliCDBEntry.h" |
73abe331 | 148 | #include "AliShuttleConfig.h" |
eba76848 | 149 | #include "DCSClient/AliDCSClient.h" |
73abe331 | 150 | #include "AliLog.h" |
b948db8d | 151 | #include "AliPreprocessor.h" |
5164a766 | 152 | #include "AliShuttleStatus.h" |
2bb7b766 | 153 | #include "AliShuttleLogbookEntry.h" |
73abe331 | 154 | |
57f50b3c | 155 | #include <TSystem.h> |
58bc3020 | 156 | #include <TObject.h> |
b948db8d | 157 | #include <TString.h> |
57f50b3c | 158 | #include <TTimeStamp.h> |
73abe331 | 159 | #include <TObjString.h> |
57f50b3c | 160 | #include <TSQLServer.h> |
161 | #include <TSQLResult.h> | |
162 | #include <TSQLRow.h> | |
cb343cfd | 163 | #include <TMutex.h> |
73abe331 | 164 | |
5164a766 | 165 | #include <fstream> |
166 | ||
cb343cfd | 167 | #include <sys/types.h> |
168 | #include <sys/wait.h> | |
169 | ||
73abe331 | 170 | ClassImp(AliShuttle) |
171 | ||
2bb7b766 | 172 | TString AliShuttle::fgkMainCDB("alien://folder=ShuttleCDB"); |
84090f85 | 173 | TString AliShuttle::fgkLocalCDB("local://LocalShuttleCDB"); |
2bb7b766 | 174 | TString AliShuttle::fgkMainRefStorage("alien://folder=ShuttleReference"); |
84090f85 | 175 | TString AliShuttle::fgkLocalRefStorage("local://LocalReferenceStorage"); |
176 | ||
4f0ab988 | 177 | Bool_t AliShuttle::fgkProcessDCS(kTRUE); |
178 | ||
84090f85 | 179 | const char* AliShuttle::fgkShuttleTempDir = gSystem->ExpandPathName("$ALICE_ROOT/SHUTTLE/temp"); |
180 | const char* AliShuttle::fgkShuttleLogDir = gSystem->ExpandPathName("$ALICE_ROOT/SHUTTLE/log"); | |
57f50b3c | 181 | |
b948db8d | 182 | //______________________________________________________________________________________________ |
183 | AliShuttle::AliShuttle(const AliShuttleConfig* config, | |
184 | UInt_t timeout, Int_t retries): | |
4f0ab988 | 185 | fConfig(config), |
186 | fTimeout(timeout), fRetries(retries), | |
187 | fPreprocessorMap(), | |
2bb7b766 | 188 | fLogbookEntry(0), |
eba76848 | 189 | fCurrentDetector(), |
85a80aa9 | 190 | fStatusEntry(0), |
cb343cfd | 191 | fGridError(kFALSE), |
192 | fMonitoringMutex(0), | |
eba76848 | 193 | fLastActionTime(0), |
194 | fLastAction() | |
73abe331 | 195 | { |
196 | // | |
197 | // config: AliShuttleConfig used | |
73abe331 | 198 | // timeout: timeout used for AliDCSClient connection |
199 | // retries: the number of retries in case of connection error. | |
200 | // | |
201 | ||
57f50b3c | 202 | if (!fConfig->IsValid()) AliFatal("********** !!!!! Invalid configuration !!!!! **********"); |
be48e3ea | 203 | for(int iSys=0;iSys<4;iSys++) { |
57f50b3c | 204 | fServer[iSys]=0; |
be48e3ea | 205 | if (iSys < 3) |
2c15234c | 206 | fFXSlist[iSys].SetOwner(kTRUE); |
57f50b3c | 207 | } |
2bb7b766 | 208 | fPreprocessorMap.SetOwner(kTRUE); |
be48e3ea | 209 | |
210 | for (UInt_t iDet=0; iDet<NDetectors(); iDet++) | |
211 | fFirstUnprocessed[iDet] = kFALSE; | |
212 | ||
cb343cfd | 213 | fMonitoringMutex = new TMutex(); |
58bc3020 | 214 | } |
215 | ||
b948db8d | 216 | //______________________________________________________________________________________________ |
57f50b3c | 217 | AliShuttle::~AliShuttle() |
58bc3020 | 218 | { |
219 | // destructor | |
220 | ||
b948db8d | 221 | fPreprocessorMap.DeleteAll(); |
be48e3ea | 222 | for(int iSys=0;iSys<4;iSys++) |
57f50b3c | 223 | if(fServer[iSys]) { |
224 | fServer[iSys]->Close(); | |
225 | delete fServer[iSys]; | |
eba76848 | 226 | fServer[iSys] = 0; |
57f50b3c | 227 | } |
2bb7b766 | 228 | |
229 | if (fStatusEntry){ | |
230 | delete fStatusEntry; | |
231 | fStatusEntry = 0; | |
232 | } | |
cb343cfd | 233 | |
234 | if (fMonitoringMutex) | |
235 | { | |
236 | delete fMonitoringMutex; | |
237 | fMonitoringMutex = 0; | |
238 | } | |
73abe331 | 239 | } |
240 | ||
b948db8d | 241 | //______________________________________________________________________________________________ |
57f50b3c | 242 | void AliShuttle::RegisterPreprocessor(AliPreprocessor* preprocessor) |
58bc3020 | 243 | { |
73abe331 | 244 | // |
b948db8d | 245 | // Registers new AliPreprocessor. |
73abe331 | 246 | // It uses GetName() for indentificator of the pre processor. |
247 | // The pre processor is registered it there isn't any other | |
248 | // with the same identificator (GetName()). | |
249 | // | |
250 | ||
eba76848 | 251 | const char* detName = preprocessor->GetName(); |
252 | if(GetDetPos(detName) < 0) | |
253 | AliFatal(Form("********** !!!!! Invalid detector name: %s !!!!! **********", detName)); | |
254 | ||
255 | if (fPreprocessorMap.GetValue(detName)) { | |
256 | AliWarning(Form("AliPreprocessor %s is already registered!", detName)); | |
73abe331 | 257 | return; |
258 | } | |
259 | ||
eba76848 | 260 | fPreprocessorMap.Add(new TObjString(detName), preprocessor); |
73abe331 | 261 | } |
b948db8d | 262 | //______________________________________________________________________________________________ |
84090f85 | 263 | UInt_t AliShuttle::Store(const AliCDBPath& path, TObject* object, |
264 | AliCDBMetaData* metaData, Int_t validityStart, Bool_t validityInfinite) | |
73abe331 | 265 | { |
84090f85 | 266 | // Stores a CDB object in the storage for offline reconstruction. Objects that are not needed for |
267 | // offline reconstruction, but should be stored anyway (e.g. for debugging) should NOT be stored | |
268 | // using this function. Use StoreReferenceData instead! | |
85a80aa9 | 269 | // It calls WriteToCDB function which perform actual storage |
b948db8d | 270 | |
85a80aa9 | 271 | return WriteToCDB(fgkMainCDB, fgkLocalCDB, path, object, |
272 | metaData, validityStart, validityInfinite); | |
84090f85 | 273 | |
274 | } | |
275 | ||
276 | //______________________________________________________________________________________________ | |
481441a2 | 277 | UInt_t AliShuttle::StoreReferenceData(const AliCDBPath& path, TObject* object, AliCDBMetaData* metaData) |
84090f85 | 278 | { |
279 | // Stores a CDB object in the storage for reference data. This objects will not be available during | |
280 | // offline reconstrunction. Use this function for reference data only! | |
85a80aa9 | 281 | // It calls WriteToCDB function which perform actual storage |
282 | ||
481441a2 | 283 | return WriteToCDB(fgkMainRefStorage, fgkLocalRefStorage, path, object, metaData); |
84090f85 | 284 | |
85a80aa9 | 285 | } |
286 | ||
287 | //______________________________________________________________________________________________ | |
288 | UInt_t AliShuttle::WriteToCDB(const char* mainUri, const char* localUri, | |
289 | const AliCDBPath& path, TObject* object, AliCDBMetaData* metaData, | |
290 | Int_t validityStart, Bool_t validityInfinite) | |
291 | { | |
292 | // write object into the CDB. Parameters are passed by Store and StoreReferenceData functions. | |
293 | // The parameters are: | |
294 | // 1) Uri of the main storage (Grid) | |
295 | // 2) Uri of the backup storage (Local) | |
296 | // 3) the object's path. | |
297 | // 4) the object to be stored | |
298 | // 5) the metaData to be associated with the object | |
299 | // 6) the validity start run number w.r.t. the current run, | |
84090f85 | 300 | // if the data is valid only for this run leave the default 0 |
85a80aa9 | 301 | // 7) specifies if the calibration data is valid for infinity (this means until updated), |
84090f85 | 302 | // typical for calibration runs, the default is kFALSE |
303 | // | |
84090f85 | 304 | // returns 0 if fail |
85a80aa9 | 305 | // 1 if stored in main (Grid) storage |
306 | // 2 if stored in backup (Local) storage | |
84090f85 | 307 | |
85a80aa9 | 308 | const char* cdbType = (mainUri == fgkMainCDB) ? "CDB" : "Reference"; |
2bb7b766 | 309 | |
85a80aa9 | 310 | Int_t firstRun = GetCurrentRun() - validityStart; |
84090f85 | 311 | if(firstRun < 0) { |
2bb7b766 | 312 | AliError("First valid run happens to be less than 0! Setting it to 0."); |
84090f85 | 313 | firstRun=0; |
314 | } | |
315 | ||
316 | Int_t lastRun = -1; | |
317 | if(validityInfinite) { | |
318 | lastRun = AliCDBRunRange::Infinity(); | |
319 | } else { | |
320 | lastRun = GetCurrentRun(); | |
321 | } | |
322 | ||
2bb7b766 | 323 | AliCDBId id(path, firstRun, lastRun, -1, -1); |
324 | ||
325 | if(! dynamic_cast<TObjString*> (metaData->GetProperty("RunUsed(TObjString)"))){ | |
326 | TObjString runUsed = Form("%d", GetCurrentRun()); | |
9e080f92 | 327 | metaData->SetProperty("RunUsed(TObjString)", runUsed.Clone()); |
2bb7b766 | 328 | } |
84090f85 | 329 | |
330 | UInt_t result = 0; | |
331 | ||
85a80aa9 | 332 | if (!(AliCDBManager::Instance()->GetStorage(mainUri))) { |
2bb7b766 | 333 | AliError(Form("WriteToCDB - Cannot activate main %s storage", cdbType)); |
84090f85 | 334 | } else { |
85a80aa9 | 335 | result = (UInt_t) AliCDBManager::Instance()->GetStorage(mainUri) |
84090f85 | 336 | ->Put(object, id, metaData); |
337 | } | |
338 | ||
339 | if(!result) { | |
340 | ||
341 | Log(fCurrentDetector, | |
2bb7b766 | 342 | Form("WriteToCDB - Problem with main %s storage. Putting <%s> into backup storage", |
343 | cdbType, path.GetPath().Data())); | |
344 | ||
345 | // Set Grid version to current run number, to ease retrieval later | |
346 | id.SetVersion(GetCurrentRun()); | |
84090f85 | 347 | |
85a80aa9 | 348 | result = AliCDBManager::Instance()->GetStorage(localUri) |
84090f85 | 349 | ->Put(object, id, metaData); |
350 | ||
351 | if(result) { | |
352 | result = 2; | |
85a80aa9 | 353 | fGridError = kTRUE; |
84090f85 | 354 | }else{ |
2bb7b766 | 355 | Log(fCurrentDetector, "WriteToCDB - Can't store data!"); |
b948db8d | 356 | } |
357 | } | |
2bb7b766 | 358 | |
b948db8d | 359 | return result; |
360 | ||
73abe331 | 361 | } |
362 | ||
b948db8d | 363 | //______________________________________________________________________________________________ |
5164a766 | 364 | AliShuttleStatus* AliShuttle::ReadShuttleStatus() |
365 | { | |
2bb7b766 | 366 | // Reads the AliShuttleStatus from the CDB |
5164a766 | 367 | |
2bb7b766 | 368 | if (fStatusEntry){ |
369 | delete fStatusEntry; | |
370 | fStatusEntry = 0; | |
371 | } | |
5164a766 | 372 | |
2bb7b766 | 373 | fStatusEntry = AliCDBManager::Instance()->GetStorage(AliShuttle::GetLocalCDB()) |
374 | ->Get(Form("/SHUTTLE/STATUS/%s", fCurrentDetector.Data()), GetCurrentRun()); | |
5164a766 | 375 | |
2bb7b766 | 376 | if (!fStatusEntry) return 0; |
377 | fStatusEntry->SetOwner(1); | |
5164a766 | 378 | |
2bb7b766 | 379 | AliShuttleStatus* status = dynamic_cast<AliShuttleStatus*> (fStatusEntry->GetObject()); |
380 | if (!status) { | |
381 | AliError("Invalid object stored to CDB!"); | |
382 | return 0; | |
383 | } | |
5164a766 | 384 | |
2bb7b766 | 385 | return status; |
5164a766 | 386 | } |
387 | ||
388 | //______________________________________________________________________________________________ | |
7bfb2090 | 389 | Bool_t AliShuttle::WriteShuttleStatus(AliShuttleStatus* status) |
5164a766 | 390 | { |
2bb7b766 | 391 | // writes the status for one subdetector |
392 | ||
393 | if (fStatusEntry){ | |
394 | delete fStatusEntry; | |
395 | fStatusEntry = 0; | |
396 | } | |
5164a766 | 397 | |
2bb7b766 | 398 | Int_t run = GetCurrentRun(); |
5164a766 | 399 | |
2bb7b766 | 400 | AliCDBId id(AliCDBPath("SHUTTLE", "STATUS", fCurrentDetector), run, run); |
5164a766 | 401 | |
2bb7b766 | 402 | fStatusEntry = new AliCDBEntry(status, id, new AliCDBMetaData); |
403 | fStatusEntry->SetOwner(1); | |
5164a766 | 404 | |
2bb7b766 | 405 | UInt_t result = AliCDBManager::Instance()->GetStorage(fgkLocalCDB)->Put(fStatusEntry); |
7bfb2090 | 406 | |
2bb7b766 | 407 | if (!result) { |
408 | AliError(Form("WriteShuttleStatus for %s, run %d failed", fCurrentDetector.Data(), run)); | |
409 | return kFALSE; | |
410 | } | |
7bfb2090 | 411 | |
2bb7b766 | 412 | return kTRUE; |
5164a766 | 413 | } |
414 | ||
415 | //______________________________________________________________________________________________ | |
416 | void AliShuttle::UpdateShuttleStatus(AliShuttleStatus::Status newStatus, Bool_t increaseCount) | |
417 | { | |
418 | // changes the AliShuttleStatus for the given detector and run to the given status | |
419 | ||
2bb7b766 | 420 | if (!fStatusEntry){ |
421 | AliError("UNEXPECTED: fStatusEntry empty"); | |
422 | return; | |
423 | } | |
5164a766 | 424 | |
2bb7b766 | 425 | AliShuttleStatus* status = dynamic_cast<AliShuttleStatus*> (fStatusEntry->GetObject()); |
5164a766 | 426 | |
2bb7b766 | 427 | if (!status){ |
428 | AliError("UNEXPECTED: status could not be read from current CDB entry"); | |
429 | return; | |
430 | } | |
5164a766 | 431 | |
2c15234c | 432 | TString actionStr = Form("UpdateShuttleStatus - %s: Changing state from %s to %s", |
eba76848 | 433 | fCurrentDetector.Data(), |
434 | status->GetStatusName(), | |
435 | status->GetStatusName(newStatus)); | |
cb343cfd | 436 | Log("SHUTTLE", actionStr); |
437 | SetLastAction(actionStr); | |
5164a766 | 438 | |
2bb7b766 | 439 | status->SetStatus(newStatus); |
440 | if (increaseCount) status->IncreaseCount(); | |
5164a766 | 441 | |
2bb7b766 | 442 | AliCDBManager::Instance()->GetStorage(fgkLocalCDB)->Put(fStatusEntry); |
5164a766 | 443 | } |
5164a766 | 444 | //______________________________________________________________________________________________ |
445 | Bool_t AliShuttle::ContinueProcessing() | |
446 | { | |
2bb7b766 | 447 | // this function reads the AliShuttleStatus information from CDB and |
448 | // checks if the processing should be continued | |
449 | // if yes it returns kTRUE and updates the AliShuttleStatus with nextStatus | |
450 | ||
57c1a579 | 451 | if (!fConfig->HostProcessDetector(fCurrentDetector)) return kFALSE; |
452 | ||
453 | AliPreprocessor* aPreprocessor = | |
454 | dynamic_cast<AliPreprocessor*> (fPreprocessorMap.GetValue(fCurrentDetector)); | |
455 | if (!aPreprocessor) | |
456 | { | |
457 | AliInfo(Form("%s: no preprocessor registered", fCurrentDetector.Data())); | |
458 | return kFALSE; | |
459 | } | |
460 | ||
2bb7b766 | 461 | AliShuttleLogbookEntry::Status entryStatus = |
eba76848 | 462 | fLogbookEntry->GetDetectorStatus(fCurrentDetector); |
2bb7b766 | 463 | |
464 | if(entryStatus != AliShuttleLogbookEntry::kUnprocessed) { | |
9e080f92 | 465 | AliInfo(Form("ContinueProcessing - %s is %s", |
2bb7b766 | 466 | fCurrentDetector.Data(), |
467 | fLogbookEntry->GetDetectorStatusName(entryStatus))); | |
468 | return kFALSE; | |
469 | } | |
470 | ||
471 | // if we get here, according to Shuttle logbook subdetector is in UNPROCESSED state | |
be48e3ea | 472 | |
473 | // check if current run is first unprocessed run for current detector | |
474 | if (fConfig->StrictRunOrder(fCurrentDetector) && | |
475 | !fFirstUnprocessed[GetDetPos(fCurrentDetector)]) | |
476 | { | |
477 | Log("SHUTTLE", Form("ContinueProcessing - %s requires strict run ordering but this is not the first unprocessed run!")); | |
478 | return kFALSE; | |
479 | } | |
480 | ||
2bb7b766 | 481 | AliShuttleStatus* status = ReadShuttleStatus(); |
482 | if (!status) { | |
483 | // first time | |
484 | Log("SHUTTLE", Form("ContinueProcessing - %s: Processing first time", | |
485 | fCurrentDetector.Data())); | |
486 | status = new AliShuttleStatus(AliShuttleStatus::kStarted); | |
487 | return WriteShuttleStatus(status); | |
488 | } | |
489 | ||
490 | // The following two cases shouldn't happen if Shuttle Logbook was correctly updated. | |
491 | // If it happens it may mean Logbook updating failed... let's do it now! | |
492 | if (status->GetStatus() == AliShuttleStatus::kDone || | |
493 | status->GetStatus() == AliShuttleStatus::kFailed){ | |
494 | Log("SHUTTLE", Form("ContinueProcessing - %s is already %s. Updating Shuttle Logbook", | |
495 | fCurrentDetector.Data(), | |
496 | status->GetStatusName(status->GetStatus()))); | |
497 | UpdateShuttleLogbook(fCurrentDetector.Data(), | |
498 | status->GetStatusName(status->GetStatus())); | |
499 | return kFALSE; | |
500 | } | |
501 | ||
502 | if (status->GetStatus() == AliShuttleStatus::kStoreFailed) { | |
503 | Log("SHUTTLE", | |
504 | Form("ContinueProcessing - %s: Grid storage of one or more objects failed. Trying again now", | |
505 | fCurrentDetector.Data())); | |
506 | if(TryToStoreAgain()){ | |
507 | Log(fCurrentDetector.Data(), "ContinueProcessing - All objects successfully stored into OCDB"); | |
508 | UpdateShuttleStatus(AliShuttleStatus::kDone); | |
509 | UpdateShuttleLogbook(fCurrentDetector.Data(), "DONE"); | |
510 | } else { | |
511 | Log("SHUTTLE", | |
512 | Form("ContinueProcessing - %s: Grid storage failed again", | |
513 | fCurrentDetector.Data())); | |
514 | } | |
515 | return kFALSE; | |
516 | } | |
517 | ||
518 | // if we get here, there is a restart | |
57c1a579 | 519 | Bool_t cont = kFALSE; |
2bb7b766 | 520 | |
521 | // abort conditions | |
cb343cfd | 522 | if (status->GetCount() >= fConfig->GetMaxRetries()) { |
57c1a579 | 523 | Log("SHUTTLE", Form("ContinueProcessing - %s failed %d times in status %s - " |
524 | "Updating Shuttle Logbook", fCurrentDetector.Data(), | |
2bb7b766 | 525 | status->GetCount(), status->GetStatusName())); |
526 | UpdateShuttleLogbook(fCurrentDetector.Data(), "FAILED"); | |
57c1a579 | 527 | } else { |
528 | Log("SHUTTLE", Form("ContinueProcessing - %s: restarting. " | |
529 | "Aborted before with %s. Retry number %d.", fCurrentDetector.Data(), | |
530 | status->GetStatusName(), status->GetCount())); | |
531 | UpdateShuttleStatus(AliShuttleStatus::kStarted, kTRUE); | |
532 | cont = kTRUE; | |
2bb7b766 | 533 | } |
534 | ||
57c1a579 | 535 | // Send mail to detector expert! |
536 | AliInfo(Form("Sending mail to %s expert...", fCurrentDetector.Data())); | |
537 | if (!SendMail()) | |
538 | Log("SHUTTLE", Form("ContinueProcessing - Could not send mail to %s expert", | |
539 | fCurrentDetector.Data())); | |
2bb7b766 | 540 | |
57c1a579 | 541 | return cont; |
5164a766 | 542 | } |
543 | ||
544 | //______________________________________________________________________________________________ | |
2bb7b766 | 545 | Bool_t AliShuttle::Process(AliShuttleLogbookEntry* entry) |
58bc3020 | 546 | { |
73abe331 | 547 | // |
b948db8d | 548 | // Makes data retrieval for all detectors in the configuration. |
2bb7b766 | 549 | // entry: Shuttle logbook entry, contains run paramenters and status of detectors |
550 | // (Unprocessed, Inactive, Failed or Done). | |
d477ad88 | 551 | // Returns kFALSE in case of error occured and kTRUE otherwise |
73abe331 | 552 | // |
553 | ||
2bb7b766 | 554 | if(!entry) return kFALSE; |
555 | ||
556 | fLogbookEntry = entry; | |
557 | ||
558 | if(fLogbookEntry->IsDone()){ | |
559 | Log("SHUTTLE","Process - Shuttle is already DONE. Updating logbook"); | |
560 | UpdateShuttleLogbook("shuttle_done"); | |
561 | fLogbookEntry = 0; | |
562 | return kTRUE; | |
563 | } | |
564 | ||
565 | ||
566 | AliInfo(Form("\n\n \t\t\t^*^*^*^*^*^*^*^*^*^*^*^* run %d: START ^*^*^*^*^*^*^*^*^*^*^*^* \n", | |
567 | GetCurrentRun())); | |
568 | ||
eba76848 | 569 | fLogbookEntry->Print("all"); |
57f50b3c | 570 | |
571 | // Initialization | |
d477ad88 | 572 | Bool_t hasError = kFALSE; |
2c15234c | 573 | for(Int_t iSys=0;iSys<3;iSys++) fFXSCalled[iSys]=kFALSE; |
5164a766 | 574 | |
2bb7b766 | 575 | AliCDBStorage *mainCDBSto = AliCDBManager::Instance()->GetStorage(fgkMainCDB); |
576 | if(mainCDBSto) mainCDBSto->QueryCDB(GetCurrentRun()); | |
577 | AliCDBStorage *mainRefSto = AliCDBManager::Instance()->GetStorage(fgkMainRefStorage); | |
578 | if(mainRefSto) mainRefSto->QueryCDB(GetCurrentRun()); | |
d477ad88 | 579 | |
57f50b3c | 580 | // Loop on detectors in the configuration |
b948db8d | 581 | TIter iter(fConfig->GetDetectors()); |
2bb7b766 | 582 | TObjString* aDetector = 0; |
b948db8d | 583 | |
be48e3ea | 584 | while ((aDetector = (TObjString*) iter.Next())) |
585 | { | |
7bfb2090 | 586 | fCurrentDetector = aDetector->String(); |
5164a766 | 587 | |
9e080f92 | 588 | if (ContinueProcessing() == kFALSE) continue; |
589 | ||
2bb7b766 | 590 | AliInfo(Form("\n\n \t\t\t****** run %d - %s: START ******", |
591 | GetCurrentRun(), aDetector->GetName())); | |
592 | ||
85a80aa9 | 593 | |
be48e3ea | 594 | Int_t pid = fork(); |
595 | ||
596 | if (pid < 0) | |
597 | { | |
598 | Log("SHUTTLE", "ERROR: Forking failed"); | |
599 | } | |
600 | else if (pid > 0) | |
601 | { | |
602 | // parent | |
603 | AliInfo(Form("In parent process of %d - %s: Starting monitoring", | |
604 | GetCurrentRun(), aDetector->GetName())); | |
605 | ||
606 | Long_t begin = time(0); | |
607 | ||
608 | int status; // to be used with waitpid, on purpose an int (not Int_t)! | |
609 | while (waitpid(pid, &status, WNOHANG) == 0) | |
610 | { | |
611 | Long_t expiredTime = time(0) - begin; | |
612 | ||
613 | if (expiredTime > fConfig->GetPPTimeOut()) | |
614 | { | |
615 | Log("SHUTTLE", Form("Process time out. Run time: %d seconds. Killing...", | |
616 | expiredTime)); | |
617 | ||
618 | kill(pid, 9); | |
619 | ||
620 | hasError = kTRUE; | |
621 | ||
622 | gSystem->Sleep(1000); | |
623 | } | |
624 | else | |
625 | { | |
626 | if (expiredTime % 60 == 0) | |
627 | Log("SHUTTLE", Form("Checked process. Run time: %d seconds.", | |
628 | expiredTime)); | |
629 | gSystem->Sleep(1000); | |
630 | } | |
631 | } | |
632 | ||
633 | AliInfo(Form("In parent process of %d - %s: Client has terminated.", | |
634 | GetCurrentRun(), aDetector->GetName())); | |
635 | ||
636 | if (WIFEXITED(status)) | |
637 | { | |
638 | Int_t returnCode = WEXITSTATUS(status); | |
639 | ||
640 | Log("SHUTTLE", Form("The return code is %d", returnCode)); | |
641 | ||
642 | if (returnCode != 0) | |
643 | hasError = kTRUE; | |
644 | } | |
645 | } | |
646 | else if (pid == 0) | |
647 | { | |
648 | // client | |
649 | AliInfo(Form("In client process of %d - %s", GetCurrentRun(), aDetector->GetName())); | |
650 | ||
651 | UInt_t result = ProcessCurrentDetector(); | |
652 | ||
653 | Int_t returnCode = 0; // will be set to 1 in case of an error | |
654 | ||
655 | if (!result) | |
656 | { | |
657 | returnCode = 1; | |
658 | AliInfo(Form("\n \t\t\t****** run %d - %s: PREPROCESSOR ERROR ****** \n\n", | |
659 | GetCurrentRun(), aDetector->GetName())); | |
660 | } | |
661 | else if (result == 2) | |
662 | { | |
663 | AliInfo(Form("\n \t\t\t****** run %d - %s: STORAGE ERROR ****** \n\n", | |
664 | GetCurrentRun(), aDetector->GetName())); | |
665 | } else | |
666 | { | |
667 | AliInfo(Form("\n \t\t\t****** run %d - %s: DONE ****** \n\n", | |
668 | GetCurrentRun(), aDetector->GetName())); | |
669 | } | |
670 | ||
671 | if (result > 0) | |
672 | { | |
2c15234c | 673 | // Process successful: Update time_processed field in FXS logbooks! |
674 | if (fFXSCalled[kDAQ]) | |
be48e3ea | 675 | { |
676 | if (UpdateDAQTable() == kFALSE) | |
677 | returnCode = 1; | |
2c15234c | 678 | fFXSlist[kDAQ].Clear(); |
be48e3ea | 679 | } |
2c15234c | 680 | //if(fFXSCalled[kDCS]) { |
be48e3ea | 681 | // if (UpdateDCSTable(aDetector->GetName()) == kFALSE) |
682 | // returnCode = 1; | |
2c15234c | 683 | // fFXSlist[kDCS].Clear(); |
be48e3ea | 684 | //} |
9e080f92 | 685 | if (fFXSCalled[kHLT]) |
686 | { | |
687 | if (UpdateHLTTable() == kFALSE) | |
688 | returnCode = 1; | |
689 | fFXSlist[kHLT].Clear(); | |
690 | } | |
be48e3ea | 691 | } |
692 | ||
693 | AliInfo(Form("Client process of %d - %s is exiting now with %d.", | |
694 | GetCurrentRun(), aDetector->GetName(), returnCode)); | |
695 | ||
696 | // the client exits here | |
697 | gSystem->Exit(returnCode); | |
698 | ||
699 | AliError("We should never get here!!!"); | |
700 | } | |
7bfb2090 | 701 | } |
5164a766 | 702 | |
2bb7b766 | 703 | AliInfo(Form("\n\n \t\t\t^*^*^*^*^*^*^*^*^*^*^*^* run %d: FINISH ^*^*^*^*^*^*^*^*^*^*^*^* \n", |
704 | GetCurrentRun())); | |
705 | ||
706 | //check if shuttle is done for this run, if so update logbook | |
707 | TObjArray checkEntryArray; | |
708 | checkEntryArray.SetOwner(1); | |
9e080f92 | 709 | TString whereClause = Form("where run=%d", GetCurrentRun()); |
710 | if (!QueryShuttleLogbook(whereClause.Data(), checkEntryArray) || checkEntryArray.GetEntries() == 0) { | |
711 | Log("SHUTTLE", Form("Process - Warning: Cannot check status of run %d on Shuttle logbook!", | |
712 | GetCurrentRun())); | |
713 | return hasError == kFALSE; | |
714 | } | |
b948db8d | 715 | |
9e080f92 | 716 | AliShuttleLogbookEntry* checkEntry = dynamic_cast<AliShuttleLogbookEntry*> |
717 | (checkEntryArray.At(0)); | |
2bb7b766 | 718 | |
9e080f92 | 719 | if (checkEntry) |
720 | { | |
721 | if (checkEntry->IsDone()) | |
be48e3ea | 722 | { |
9e080f92 | 723 | Log("SHUTTLE","Process - Shuttle is DONE. Updating logbook"); |
724 | UpdateShuttleLogbook("shuttle_done"); | |
725 | } | |
726 | else | |
727 | { | |
728 | for (UInt_t iDet=0; iDet<NDetectors(); iDet++) | |
be48e3ea | 729 | { |
9e080f92 | 730 | if (checkEntry->GetDetectorStatus(iDet) == AliShuttleLogbookEntry::kUnprocessed) |
be48e3ea | 731 | { |
9e080f92 | 732 | AliDebug(2, Form("Run %d: setting %s as \"not first time unprocessed\"", |
733 | checkEntry->GetRun(), GetDetName(iDet))); | |
734 | fFirstUnprocessed[iDet] = kFALSE; | |
be48e3ea | 735 | } |
736 | } | |
2bb7b766 | 737 | } |
738 | } | |
739 | ||
740 | fLogbookEntry = 0; | |
85a80aa9 | 741 | |
a7160fe9 | 742 | return hasError == kFALSE; |
73abe331 | 743 | } |
744 | ||
b948db8d | 745 | //______________________________________________________________________________________________ |
2bb7b766 | 746 | UInt_t AliShuttle::ProcessCurrentDetector() |
73abe331 | 747 | { |
748 | // | |
2bb7b766 | 749 | // Makes data retrieval just for a specific detector (fCurrentDetector). |
73abe331 | 750 | // Threre should be a configuration for this detector. |
73abe331 | 751 | |
2bb7b766 | 752 | AliInfo(Form("Retrieving values for %s, run %d", fCurrentDetector.Data(), GetCurrentRun())); |
73abe331 | 753 | |
7bfb2090 | 754 | UpdateShuttleStatus(AliShuttleStatus::kDCSStarted); |
73abe331 | 755 | |
2c15234c | 756 | TMap dcsMap; |
757 | dcsMap.SetOwner(1); | |
73abe331 | 758 | |
85a80aa9 | 759 | Bool_t aDCSError = kFALSE; |
760 | fGridError = kFALSE; | |
d477ad88 | 761 | |
2c15234c | 762 | // TODO Test only... I've added a flag that allows to |
763 | // exclude DCS archive DB query | |
764 | if (!fgkProcessDCS) | |
765 | { | |
766 | AliInfo("Skipping DCS processing!"); | |
767 | aDCSError = kFALSE; | |
768 | } else { | |
769 | TString host(fConfig->GetDCSHost(fCurrentDetector)); | |
770 | Int_t port = fConfig->GetDCSPort(fCurrentDetector); | |
771 | ||
772 | // Retrieval of Aliases | |
773 | TObjString* anAlias = 0; | |
774 | TIter iterAliases(fConfig->GetDCSAliases(fCurrentDetector)); | |
775 | while ((anAlias = (TObjString*) iterAliases.Next())) | |
776 | { | |
777 | TObjArray *valueSet = new TObjArray(); | |
778 | valueSet->SetOwner(1); | |
779 | ||
780 | AliInfo("Querying DCS archive DB (Aliases)..."); | |
781 | aDCSError = (GetValueSet(host, port, anAlias->String(), valueSet, kAlias) == 0); | |
782 | ||
783 | if(!aDCSError) | |
784 | { | |
785 | dcsMap.Add(anAlias->Clone(), valueSet); | |
786 | } else { | |
787 | Log(fCurrentDetector, | |
788 | Form("ProcessCurrentDetector - Error while retrieving alias %s", | |
789 | anAlias->GetName())); | |
790 | UpdateShuttleStatus(AliShuttleStatus::kDCSError); | |
791 | dcsMap.DeleteAll(); | |
792 | return 0; | |
793 | } | |
4f0ab988 | 794 | } |
2c15234c | 795 | |
796 | // Retrieval of Data Points | |
797 | TObjString* aDP = 0; | |
798 | TIter iterDP(fConfig->GetDCSDataPoints(fCurrentDetector)); | |
799 | while ((aDP = (TObjString*) iterDP.Next())) | |
800 | { | |
801 | TObjArray *valueSet = new TObjArray(); | |
802 | valueSet->SetOwner(1); | |
803 | AliInfo("Querying DCS archive DB (Data Points)..."); | |
804 | aDCSError = (GetValueSet(host, port, aDP->String(), valueSet, kDP) == 0); | |
805 | ||
806 | if(!aDCSError) | |
807 | { | |
808 | dcsMap.Add(aDP->Clone(), valueSet); | |
809 | } else { | |
810 | Log(fCurrentDetector, | |
811 | Form("ProcessCurrentDetector - Error while retrieving data point %s", | |
812 | aDP->GetName())); | |
813 | UpdateShuttleStatus(AliShuttleStatus::kDCSError); | |
814 | dcsMap.DeleteAll(); | |
815 | return 0; | |
816 | } | |
73abe331 | 817 | } |
818 | } | |
b948db8d | 819 | |
2bb7b766 | 820 | // DCS Archive DB processing successful. Call Preprocessor! |
85a80aa9 | 821 | UpdateShuttleStatus(AliShuttleStatus::kPPStarted); |
a7160fe9 | 822 | |
85a80aa9 | 823 | AliPreprocessor* aPreprocessor = |
5164a766 | 824 | dynamic_cast<AliPreprocessor*> (fPreprocessorMap.GetValue(fCurrentDetector)); |
b948db8d | 825 | |
2bb7b766 | 826 | aPreprocessor->Initialize(GetCurrentRun(), GetCurrentStartTime(), GetCurrentEndTime()); |
2c15234c | 827 | UInt_t aPPResult = aPreprocessor->Process(&dcsMap); |
2bb7b766 | 828 | |
829 | UInt_t returnValue = 0; | |
85a80aa9 | 830 | if (aPPResult == 0) { // Preprocessor error |
cb343cfd | 831 | UpdateShuttleStatus(AliShuttleStatus::kPPError); |
2bb7b766 | 832 | returnValue = 0; |
85a80aa9 | 833 | } else if (fGridError == kFALSE) { // process and Grid storage ok! |
834 | UpdateShuttleStatus(AliShuttleStatus::kDone); | |
2bb7b766 | 835 | UpdateShuttleLogbook(fCurrentDetector, "DONE"); |
836 | Log(fCurrentDetector.Data(), | |
837 | "ProcessCurrentDetector - Preprocessor and Grid storage ended successfully"); | |
838 | returnValue = 1; | |
85a80aa9 | 839 | } else { // Grid storage error (process ok, but object put in local storage) |
840 | UpdateShuttleStatus(AliShuttleStatus::kStoreFailed); | |
2bb7b766 | 841 | returnValue = 2; |
85a80aa9 | 842 | } |
b948db8d | 843 | |
2c15234c | 844 | dcsMap.DeleteAll(); |
b948db8d | 845 | |
2bb7b766 | 846 | return returnValue; |
847 | } | |
848 | ||
849 | //______________________________________________________________________________________________ | |
850 | Bool_t AliShuttle::QueryShuttleLogbook(const char* whereClause, | |
851 | TObjArray& entries) | |
852 | { | |
853 | // Query DAQ's Shuttle logbook and fills detector status object. | |
854 | // Call QueryRunParameters to query DAQ logbook for run parameters. | |
855 | ||
fc5a4708 | 856 | entries.SetOwner(1); |
857 | ||
2bb7b766 | 858 | // check connection, in case connect |
be48e3ea | 859 | if(!Connect(3)) return kFALSE; |
2bb7b766 | 860 | |
861 | TString sqlQuery; | |
862 | sqlQuery = Form("select * from logbook_shuttle %s order by run", whereClause); | |
863 | ||
be48e3ea | 864 | TSQLResult* aResult = fServer[3]->Query(sqlQuery); |
2bb7b766 | 865 | if (!aResult) { |
866 | AliError(Form("Can't execute query <%s>!", sqlQuery.Data())); | |
867 | return kFALSE; | |
868 | } | |
869 | ||
fc5a4708 | 870 | AliDebug(2,Form("Query = %s", sqlQuery.Data())); |
871 | ||
2bb7b766 | 872 | if(aResult->GetRowCount() == 0) { |
9e080f92 | 873 | // if(sqlQuery.EndsWith("where shuttle_done=0 order by run")){ |
874 | // Log("SHUTTLE", "QueryShuttleLogbook - All runs in Shuttle Logbook are already DONE"); | |
875 | // delete aResult; | |
876 | // return kTRUE; | |
877 | // } else { | |
878 | AliInfo("No entries in Shuttle Logbook match request"); | |
2bb7b766 | 879 | delete aResult; |
880 | return kTRUE; | |
9e080f92 | 881 | // } |
2bb7b766 | 882 | } |
883 | ||
884 | // TODO Check field count! | |
fc5a4708 | 885 | const UInt_t nCols = 22; |
2bb7b766 | 886 | if (aResult->GetFieldCount() != (Int_t) nCols) { |
887 | AliError("Invalid SQL result field number!"); | |
888 | delete aResult; | |
889 | return kFALSE; | |
890 | } | |
891 | ||
2bb7b766 | 892 | TSQLRow* aRow; |
893 | while ((aRow = aResult->Next())) { | |
894 | TString runString(aRow->GetField(0), aRow->GetFieldLength(0)); | |
895 | Int_t run = runString.Atoi(); | |
896 | ||
eba76848 | 897 | AliShuttleLogbookEntry *entry = QueryRunParameters(run); |
898 | if (!entry) | |
899 | continue; | |
2bb7b766 | 900 | |
901 | // loop on detectors | |
eba76848 | 902 | for(UInt_t ii = 0; ii < nCols; ii++) |
903 | entry->SetDetectorStatus(aResult->GetFieldName(ii), aRow->GetField(ii)); | |
2bb7b766 | 904 | |
eba76848 | 905 | entries.AddLast(entry); |
2bb7b766 | 906 | delete aRow; |
907 | } | |
908 | ||
9e080f92 | 909 | // if(sqlQuery.EndsWith("where shuttle_done=0 order by run")) |
910 | // Log("SHUTTLE", Form("QueryShuttleLogbook - Found %d unprocessed runs in Shuttle Logbook", | |
911 | // entries.GetEntriesFast())); | |
2bb7b766 | 912 | delete aResult; |
913 | return kTRUE; | |
914 | } | |
915 | ||
916 | //______________________________________________________________________________________________ | |
eba76848 | 917 | AliShuttleLogbookEntry* AliShuttle::QueryRunParameters(Int_t run) |
2bb7b766 | 918 | { |
eba76848 | 919 | // |
920 | // Retrieve run parameters written in the DAQ logbook and sets them into AliShuttleLogbookEntry object | |
921 | // | |
2bb7b766 | 922 | |
923 | // check connection, in case connect | |
be48e3ea | 924 | if (!Connect(3)) |
eba76848 | 925 | return 0; |
2bb7b766 | 926 | |
927 | TString sqlQuery; | |
2c15234c | 928 | sqlQuery.Form("select * from %s where run=%d", fConfig->GetDAQlbTable(), run); |
2bb7b766 | 929 | |
be48e3ea | 930 | TSQLResult* aResult = fServer[3]->Query(sqlQuery); |
2bb7b766 | 931 | if (!aResult) { |
932 | AliError(Form("Can't execute query <%s>!", sqlQuery.Data())); | |
eba76848 | 933 | return 0; |
2bb7b766 | 934 | } |
935 | ||
eba76848 | 936 | if (aResult->GetRowCount() == 0) { |
2bb7b766 | 937 | Log("SHUTTLE", Form("QueryRunParameters - No entry in DAQ Logbook for run %d. Skipping", run)); |
938 | delete aResult; | |
eba76848 | 939 | return 0; |
2bb7b766 | 940 | } |
941 | ||
eba76848 | 942 | if (aResult->GetRowCount() > 1) { |
2bb7b766 | 943 | AliError(Form("More than one entry in DAQ Logbook for run %d. Skipping", run)); |
944 | delete aResult; | |
eba76848 | 945 | return 0; |
2bb7b766 | 946 | } |
947 | ||
eba76848 | 948 | TSQLRow* aRow = aResult->Next(); |
949 | if (!aRow) | |
950 | { | |
951 | AliError(Form("Could not retrieve row for run %d. Skipping", run)); | |
952 | delete aResult; | |
953 | return 0; | |
954 | } | |
2bb7b766 | 955 | |
eba76848 | 956 | AliShuttleLogbookEntry* entry = new AliShuttleLogbookEntry(run); |
2bb7b766 | 957 | |
eba76848 | 958 | for (Int_t ii = 0; ii < aResult->GetFieldCount(); ii++) |
959 | entry->SetRunParameter(aResult->GetFieldName(ii), aRow->GetField(ii)); | |
2bb7b766 | 960 | |
eba76848 | 961 | UInt_t startTime = entry->GetStartTime(); |
962 | UInt_t endTime = entry->GetEndTime(); | |
963 | ||
964 | if (!startTime || !endTime || startTime > endTime) { | |
965 | Log("SHUTTLE", | |
966 | Form("QueryRunParameters - Invalid parameters for Run %d: startTime = %d, endTime = %d", | |
967 | run, startTime, endTime)); | |
968 | delete entry; | |
2bb7b766 | 969 | delete aRow; |
eba76848 | 970 | delete aResult; |
971 | return 0; | |
2bb7b766 | 972 | } |
973 | ||
eba76848 | 974 | delete aRow; |
2bb7b766 | 975 | delete aResult; |
eba76848 | 976 | |
977 | return entry; | |
2bb7b766 | 978 | } |
979 | ||
980 | //______________________________________________________________________________________________ | |
981 | Bool_t AliShuttle::TryToStoreAgain() | |
982 | { | |
983 | // Called in case the detector failed to store the object in Grid OCDB | |
984 | // It tries to store the object again, if it does not find more recent and overlapping objects | |
985 | // Calls underlying TryToStoreAgain(const char*) function twice, for OCDB and Reference storage. | |
986 | ||
987 | AliInfo("Trying to store OCDB data again..."); | |
988 | Bool_t resultCDB = TryToStoreAgain(fgkMainCDB); | |
989 | ||
990 | AliInfo("Trying to store reference data again..."); | |
991 | Bool_t resultRef = TryToStoreAgain(fgkMainRefStorage); | |
992 | ||
993 | return resultCDB && resultRef; | |
994 | } | |
995 | ||
996 | //______________________________________________________________________________________________ | |
997 | Bool_t AliShuttle::TryToStoreAgain(TString& gridURI) | |
998 | { | |
999 | // Called by TryToStoreAgain(), performs actual storage retry | |
1000 | ||
6ec0e06c | 1001 | TObjArray* gridIds=0; |
2bb7b766 | 1002 | |
1003 | Bool_t result = kTRUE; | |
1004 | ||
1005 | const char* type = 0; | |
1006 | TString backupURI; | |
1007 | if(gridURI == fgkMainCDB) { | |
1008 | type = "OCDB"; | |
1009 | backupURI = fgkLocalCDB; | |
1010 | } else if(gridURI == fgkMainRefStorage) { | |
1011 | type = "reference"; | |
1012 | backupURI = fgkLocalRefStorage; | |
1013 | } else { | |
1014 | AliError(Form("Invalid storage URI: %s", gridURI.Data())); | |
1015 | return kFALSE; | |
1016 | } | |
1017 | ||
1018 | AliCDBManager* man = AliCDBManager::Instance(); | |
1019 | ||
1020 | AliCDBStorage *gridSto = man->GetStorage(gridURI); | |
1021 | if(!gridSto) { | |
1022 | Log(fCurrentDetector.Data(), | |
1023 | Form("TryToStoreAgain - cannot activate main %s storage", type)); | |
1024 | return kFALSE; | |
1025 | } | |
1026 | ||
1027 | gridIds = gridSto->GetQueryCDBList(); | |
1028 | ||
1029 | // get objects previously stored in local CDB | |
1030 | AliCDBStorage *backupSto = man->GetStorage(backupURI); | |
eba76848 | 1031 | AliCDBPath aPath(GetOfflineDetName(fCurrentDetector.Data()),"*","*"); |
2bb7b766 | 1032 | // Local objects were stored with current run as Grid version! |
1033 | TList* localEntries = backupSto->GetAll(aPath.GetPath(), GetCurrentRun(), GetCurrentRun()); | |
1034 | localEntries->SetOwner(1); | |
1035 | ||
1036 | // loop on local stored objects | |
1037 | TIter localIter(localEntries); | |
1038 | AliCDBEntry *aLocEntry = 0; | |
1039 | while((aLocEntry = dynamic_cast<AliCDBEntry*> (localIter.Next()))){ | |
1040 | aLocEntry->SetOwner(1); | |
1041 | AliCDBId aLocId = aLocEntry->GetId(); | |
1042 | aLocEntry->SetVersion(-1); | |
1043 | aLocEntry->SetSubVersion(-1); | |
1044 | ||
1045 | // loop on Grid valid Id's | |
1046 | Bool_t store = kTRUE; | |
1047 | TIter gridIter(gridIds); | |
1048 | AliCDBId* aGridId = 0; | |
1049 | while((aGridId = dynamic_cast<AliCDBId*> (gridIter.Next()))){ | |
be48e3ea | 1050 | // If local object is valid up to infinity we store it only if it is |
1051 | // the first unprocessed run! | |
1052 | if (aLocId.GetLastRun() == AliCDBRunRange::Infinity()) | |
1053 | { | |
1054 | if (!fFirstUnprocessed[GetDetPos(fCurrentDetector)]) | |
1055 | { | |
1056 | Log(fCurrentDetector.Data(), | |
2c15234c | 1057 | ("TryToStoreAgain - This object has validity infinite but " |
1058 | "there are previous unprocessed runs!")); | |
be48e3ea | 1059 | continue; |
1060 | } else { | |
1061 | break; | |
1062 | } | |
2bb7b766 | 1063 | } |
1064 | if(aGridId->GetPath() != aLocId.GetPath()) continue; | |
1065 | // skip all objects valid up to infinity | |
1066 | if(aGridId->GetLastRun() == AliCDBRunRange::Infinity()) continue; | |
1067 | // if we get here, it means there's already some more recent object stored on Grid! | |
1068 | store = kFALSE; | |
1069 | break; | |
1070 | } | |
1071 | ||
1072 | if(!store){ | |
1073 | Log(fCurrentDetector.Data(), | |
1074 | Form("TryToStoreAgain - A more recent object already exists in %s storage: <%s>", | |
1075 | type, aGridId->ToString().Data())); | |
1076 | // removing local filename... | |
1077 | // TODO maybe it's better not to remove it, it was not copied to the Grid! | |
1078 | TString filename; | |
1079 | backupSto->IdToFilename(aLocId, filename); | |
1080 | AliInfo(Form("Removing local file %s", filename.Data())); | |
1081 | gSystem->Exec(Form("rm %s",filename.Data())); | |
1082 | continue; | |
1083 | } | |
1084 | ||
1085 | // If we get here, the file can be stored! | |
1086 | Bool_t storeOk = gridSto->Put(aLocEntry); | |
1087 | if(storeOk){ | |
1088 | Log(fCurrentDetector.Data(), | |
1089 | Form("TryToStoreAgain - Object <%s> successfully put into %s storage", | |
1090 | aLocId.ToString().Data(), type)); | |
1091 | ||
1092 | // removing local filename... | |
1093 | TString filename; | |
1094 | backupSto->IdToFilename(aLocId, filename); | |
1095 | AliInfo(Form("Removing local file %s", filename.Data())); | |
1096 | gSystem->Exec(Form("rm %s", filename.Data())); | |
1097 | continue; | |
1098 | } else { | |
1099 | Log(fCurrentDetector.Data(), | |
1100 | Form("TryToStoreAgain - Grid %s storage of object <%s> failed again", | |
1101 | type, aLocId.ToString().Data())); | |
1102 | result = kFALSE; | |
1103 | } | |
1104 | } | |
1105 | localEntries->Clear(); | |
1106 | ||
1107 | return result; | |
73abe331 | 1108 | } |
1109 | ||
b948db8d | 1110 | //______________________________________________________________________________________________ |
2c15234c | 1111 | Bool_t AliShuttle::GetValueSet(const char* host, Int_t port, const char* entry, |
1112 | TObjArray* valueSet, DCSType type) | |
73abe331 | 1113 | { |
2c15234c | 1114 | // Retrieve all "entry" data points from the DCS server |
58bc3020 | 1115 | // host, port: TSocket connection parameters |
2c15234c | 1116 | // entry: name of the alias or data point |
2bb7b766 | 1117 | // valueSet: array of retrieved AliDCSValue's |
2c15234c | 1118 | // type: kAlias or kDP |
58bc3020 | 1119 | |
73abe331 | 1120 | AliDCSClient client(host, port, fTimeout, fRetries); |
2c15234c | 1121 | if (!client.IsConnected()) |
1122 | { | |
b948db8d | 1123 | return kFALSE; |
73abe331 | 1124 | } |
1125 | ||
2c15234c | 1126 | Int_t result=0; |
73abe331 | 1127 | |
2c15234c | 1128 | if (type == kAlias) |
1129 | { | |
1130 | result = client.GetAliasValues(entry, | |
1131 | GetCurrentStartTime(), GetCurrentEndTime(), valueSet); | |
1132 | } else | |
1133 | if (type == kDP) | |
1134 | { | |
1135 | result = client.GetDPValues(entry, | |
1136 | GetCurrentStartTime(), GetCurrentEndTime(), valueSet); | |
1137 | } | |
1138 | ||
1139 | if (result < 0) | |
1140 | { | |
2bb7b766 | 1141 | Log(fCurrentDetector.Data(), Form("GetValueSet - Can't get '%s'! Reason: %s", |
2c15234c | 1142 | entry, AliDCSClient::GetErrorString(result))); |
73abe331 | 1143 | |
2c15234c | 1144 | if (result == AliDCSClient::fgkServerError) |
1145 | { | |
2bb7b766 | 1146 | Log(fCurrentDetector.Data(), Form("GetValueSet - Server error: %s", |
73abe331 | 1147 | client.GetServerError().Data())); |
1148 | } | |
1149 | ||
1150 | return kFALSE; | |
1151 | } | |
1152 | ||
1153 | return kTRUE; | |
1154 | } | |
b948db8d | 1155 | |
1156 | //______________________________________________________________________________________________ | |
57f50b3c | 1157 | const char* AliShuttle::GetFile(Int_t system, const char* detector, |
1158 | const char* id, const char* source) | |
b948db8d | 1159 | { |
57f50b3c | 1160 | // Get calibration file from file exchange servers |
1161 | // calls specific getter according to system index (kDAQ, kDCS, kHLT) | |
1162 | ||
1163 | switch(system){ | |
1164 | case kDAQ: | |
1165 | return GetDAQFileName(detector, id, source); | |
1166 | break; | |
1167 | case kDCS: | |
1168 | return GetDCSFileName(detector, id, source); | |
1169 | break; | |
1170 | case kHLT: | |
1171 | return GetHLTFileName(detector, id, source); | |
1172 | break; | |
1173 | default: | |
1174 | AliError(Form("No valid system index: %d",system)); | |
1175 | } | |
b948db8d | 1176 | |
b948db8d | 1177 | return 0; |
1178 | } | |
1179 | ||
b948db8d | 1180 | //______________________________________________________________________________________________ |
57f50b3c | 1181 | TList* AliShuttle::GetFileSources(Int_t system, const char* detector, const char* id) |
b948db8d | 1182 | { |
57f50b3c | 1183 | // Get sources producing the condition file Id from file exchange servers |
1184 | // calls specific getter according to system index (kDAQ, kDCS, kHLT) | |
1185 | ||
1186 | switch(system){ | |
1187 | case kDAQ: | |
1188 | return GetDAQFileSources(detector, id); | |
1189 | break; | |
1190 | case kDCS: | |
1191 | return GetDCSFileSources(detector, id); | |
1192 | break; | |
1193 | case kHLT: | |
1194 | return GetHLTFileSources(detector, id); | |
1195 | break; | |
1196 | default: | |
1197 | AliError(Form("No valid system index: %d",system)); | |
1198 | } | |
1199 | ||
1200 | return NULL; | |
1201 | } | |
1202 | ||
1203 | //______________________________________________________________________________________________ | |
2bb7b766 | 1204 | Bool_t AliShuttle::Connect(Int_t system) |
1205 | { | |
2c15234c | 1206 | // Connect to MySQL Server of the system's FXS MySQL databases |
1207 | // DAQ Logbook, Shuttle Logbook and DAQ FXS db are on the same host | |
57f50b3c | 1208 | |
1209 | // check connection: if already connected return | |
1210 | if(fServer[system] && fServer[system]->IsConnected()) return kTRUE; | |
1211 | ||
2c15234c | 1212 | TString dbHost, dbUser, dbPass, dbName; |
be48e3ea | 1213 | |
2c15234c | 1214 | if (system < 3) // FXS db servers |
be48e3ea | 1215 | { |
fc5a4708 | 1216 | dbHost = Form("mysql://%s:%d", fConfig->GetFXSdbHost(system), fConfig->GetFXSdbPort(system)); |
2c15234c | 1217 | dbUser = fConfig->GetFXSdbUser(system); |
1218 | dbPass = fConfig->GetFXSdbPass(system); | |
1219 | dbName = fConfig->GetFXSdbName(system); | |
be48e3ea | 1220 | } else { // Run & Shuttle logbook servers |
1221 | // TODO Will the Shuttle logbook server be the same as the Run logbook server ??? | |
fc5a4708 | 1222 | dbHost = Form("mysql://%s:%d", fConfig->GetDAQlbHost(), fConfig->GetDAQlbPort()); |
2c15234c | 1223 | dbUser = fConfig->GetDAQlbUser(); |
1224 | dbPass = fConfig->GetDAQlbPass(); | |
1225 | dbName = fConfig->GetDAQlbDB(); | |
be48e3ea | 1226 | } |
57f50b3c | 1227 | |
2c15234c | 1228 | fServer[system] = TSQLServer::Connect(dbHost.Data(), dbUser.Data(), dbPass.Data()); |
57f50b3c | 1229 | if (!fServer[system] || !fServer[system]->IsConnected()) { |
be48e3ea | 1230 | if(system < 3) |
1231 | { | |
2c15234c | 1232 | AliError(Form("Can't establish connection to FXS database for %s", |
eba76848 | 1233 | AliShuttleInterface::GetSystemName(system))); |
be48e3ea | 1234 | } else { |
1235 | AliError("Can't establish connection to Run logbook."); | |
1236 | } | |
2bb7b766 | 1237 | if(fServer[system]) delete fServer[system]; |
57f50b3c | 1238 | return kFALSE; |
1239 | } | |
1240 | ||
1241 | // Get tables | |
1242 | // TODO in the configuration should the table name be there too? | |
2bb7b766 | 1243 | TSQLResult* aResult=0; |
57f50b3c | 1244 | switch(system){ |
1245 | case kDAQ: | |
2c15234c | 1246 | aResult = fServer[kDAQ]->GetTables(dbName.Data()); |
57f50b3c | 1247 | break; |
1248 | case kDCS: | |
2c15234c | 1249 | //aResult = fServer[kDCS]->GetTables(dbName.Data()); |
57f50b3c | 1250 | break; |
1251 | case kHLT: | |
9e080f92 | 1252 | aResult = fServer[kHLT]->GetTables(dbName.Data()); |
57f50b3c | 1253 | break; |
1254 | default: | |
2c15234c | 1255 | aResult = fServer[3]->GetTables(dbName.Data()); |
57f50b3c | 1256 | break; |
1257 | } | |
1258 | ||
2bb7b766 | 1259 | delete aResult; |
57f50b3c | 1260 | return kTRUE; |
1261 | } | |
1262 | ||
1263 | //______________________________________________________________________________________________ | |
2bb7b766 | 1264 | const char* AliShuttle::GetDAQFileName(const char* detector, const char* id, const char* source) |
1265 | { | |
2c15234c | 1266 | // Retrieves a file from the DAQ FXS. |
1267 | // First queris the DAQ FXS database for the DAQ file name, using the run, detector, id and source info | |
57f50b3c | 1268 | // then calls RetrieveDAQFile(DAQfilename) for actual copy to local disk |
2bb7b766 | 1269 | // run: current run being processed (given by Logbook entry fLogbookEntry) |
eba76848 | 1270 | // detector: the Preprocessor name |
57f50b3c | 1271 | // id: provided as a parameter by the Preprocessor |
1272 | // source: provided by the Preprocessor through GetFileSources function | |
1273 | ||
1274 | // check connection, in case connect | |
eba76848 | 1275 | if (!Connect(kDAQ)) |
1276 | { | |
2c15234c | 1277 | Log(detector, "GetDAQFileName - Couldn't connect to DAQ FXS database"); |
57f50b3c | 1278 | return 0; |
1279 | } | |
1280 | ||
1281 | // Query preparation | |
2c15234c | 1282 | TString sqlQueryStart = Form("select filePath from %s where", fConfig->GetFXSdbTable(kDAQ)); |
57f50b3c | 1283 | TString whereClause = Form("run=%d and detector=\"%s\" and fileId=\"%s\" and DAQsource=\"%s\"", |
eba76848 | 1284 | GetCurrentRun(), detector, id, source); |
57f50b3c | 1285 | TString sqlQuery = Form("%s %s", sqlQueryStart.Data(), whereClause.Data()); |
1286 | ||
84090f85 | 1287 | AliDebug(2, Form("SQL query: \n%s",sqlQuery.Data())); |
57f50b3c | 1288 | |
1289 | // Query execution | |
2bb7b766 | 1290 | TSQLResult* aResult = 0; |
1291 | aResult = dynamic_cast<TSQLResult*> (fServer[kDAQ]->Query(sqlQuery)); | |
57f50b3c | 1292 | if (!aResult) { |
2bb7b766 | 1293 | Log(detector, Form("GetDAQFileName - Can't execute SQL query for: id = %s, source = %s", |
1294 | id, source)); | |
57f50b3c | 1295 | return 0; |
1296 | } | |
1297 | ||
eba76848 | 1298 | if(aResult->GetRowCount() == 0) |
1299 | { | |
57f50b3c | 1300 | Log(detector, |
2c15234c | 1301 | Form("GetDAQFileName - No entry in FXS table for: id = %s, source = %s", |
2bb7b766 | 1302 | id, source)); |
57f50b3c | 1303 | delete aResult; |
1304 | return 0; | |
1305 | } | |
1306 | ||
eba76848 | 1307 | if (aResult->GetRowCount() > 1) { |
57f50b3c | 1308 | Log(detector, |
2c15234c | 1309 | Form("GetDAQFileName - More than one entry in FXS table for: id = %s, source = %s", |
2bb7b766 | 1310 | id, source)); |
57f50b3c | 1311 | delete aResult; |
1312 | return 0; | |
1313 | } | |
1314 | ||
2bb7b766 | 1315 | TSQLRow* aRow = dynamic_cast<TSQLRow*> (aResult->Next()); |
57f50b3c | 1316 | |
eba76848 | 1317 | if (!aRow){ |
2bb7b766 | 1318 | Log(detector, Form("GetDAQFileName - Empty set result from query: id = %s, source = %s", |
1319 | id, source)); | |
57f50b3c | 1320 | delete aResult; |
1321 | return 0; | |
1322 | } | |
1323 | ||
1324 | TString filePath(aRow->GetField(0), aRow->GetFieldLength(0)); | |
1325 | ||
1326 | delete aResult; | |
2bb7b766 | 1327 | delete aRow; |
57f50b3c | 1328 | |
84090f85 | 1329 | AliDebug(2, Form("filePath = %s",filePath.Data())); |
57f50b3c | 1330 | |
1331 | // retrieved file is renamed to make it unique | |
9e080f92 | 1332 | TString localFileName = Form("DAQ_%s_%d_%s_%s.shuttle", |
2bb7b766 | 1333 | detector, GetCurrentRun(), id, source); |
57f50b3c | 1334 | |
2c15234c | 1335 | // file retrieval from DAQ FXS |
57f50b3c | 1336 | Bool_t result = RetrieveDAQFile(filePath.Data(), localFileName.Data()); |
1337 | if(!result) { | |
2c15234c | 1338 | Log(detector, Form("GetDAQFileName - Copy of file %s from DAQ FXS failed", filePath.Data())); |
57f50b3c | 1339 | return 0; |
1340 | } else { | |
2c15234c | 1341 | AliInfo(Form("File %s copied from DAQ FXS into %s/%s", |
57f50b3c | 1342 | filePath.Data(), fgkShuttleTempDir, localFileName.Data())); |
1343 | } | |
1344 | ||
2c15234c | 1345 | fFXSCalled[kDAQ]=kTRUE; |
9e080f92 | 1346 | TObjString *fileParams = new TObjString(Form("%s#!?!#%s", id, source)); |
2c15234c | 1347 | fFXSlist[kDAQ].Add(fileParams); |
57f50b3c | 1348 | |
9e080f92 | 1349 | static TString fullLocalFileName; |
1350 | fullLocalFileName = TString::Format("%s/%s", fgkShuttleTempDir, localFileName.Data()); | |
1351 | ||
1352 | AliInfo(Form("fullLocalFileName = %s", fullLocalFileName.Data())); | |
1353 | ||
1354 | return fullLocalFileName.Data(); | |
57f50b3c | 1355 | |
1356 | } | |
1357 | ||
1358 | //______________________________________________________________________________________________ | |
2bb7b766 | 1359 | Bool_t AliShuttle::RetrieveDAQFile(const char* daqFileName, const char* localFileName) |
1360 | { | |
9e080f92 | 1361 | // Copies file from DAQ FXS to local Shuttle machine |
57f50b3c | 1362 | |
1363 | // check temp directory: trying to cd to temp; if it does not exist, create it | |
9e080f92 | 1364 | AliDebug(2, Form("Copy file %s from DAQ FXS into %s/%s", |
1365 | daqFileName, fgkShuttleTempDir, localFileName)); | |
57f50b3c | 1366 | |
1367 | void* dir = gSystem->OpenDirectory(fgkShuttleTempDir); | |
1368 | if (dir == NULL) { | |
1369 | if (gSystem->mkdir(fgkShuttleTempDir, kTRUE)) { | |
2bb7b766 | 1370 | AliError(Form("Can't open directory <%s>", fgkShuttleTempDir)); |
57f50b3c | 1371 | return kFALSE; |
1372 | } | |
1373 | ||
1374 | } else { | |
1375 | gSystem->FreeDirectory(dir); | |
1376 | } | |
1377 | ||
9e080f92 | 1378 | TString baseDAQFXSFolder = "FES"; |
fc5a4708 | 1379 | TString command = Form("scp -oPort=%d -2 %s@%s:%s/%s %s/%s", |
1380 | fConfig->GetFXSPort(kDAQ), | |
2c15234c | 1381 | fConfig->GetFXSUser(kDAQ), |
1382 | fConfig->GetFXSHost(kDAQ), | |
1383 | baseDAQFXSFolder.Data(), | |
57f50b3c | 1384 | daqFileName, |
1385 | fgkShuttleTempDir, | |
1386 | localFileName); | |
1387 | ||
84090f85 | 1388 | AliDebug(2, Form("%s",command.Data())); |
57f50b3c | 1389 | |
1390 | UInt_t nRetries = 0; | |
1391 | UInt_t maxRetries = 3; | |
1392 | ||
1393 | // copy!! if successful TSystem::Exec returns 0 | |
1394 | while(nRetries++ < maxRetries) { | |
84090f85 | 1395 | AliDebug(2, Form("Trying to copy file. Retry # %d", nRetries)); |
57f50b3c | 1396 | if(gSystem->Exec(command.Data()) == 0) return kTRUE; |
1397 | } | |
1398 | ||
1399 | return kFALSE; | |
1400 | ||
1401 | } | |
1402 | ||
1403 | //______________________________________________________________________________________________ | |
2bb7b766 | 1404 | TList* AliShuttle::GetDAQFileSources(const char* detector, const char* id) |
1405 | { | |
9e080f92 | 1406 | // Retrieves list of DAQ sources of file Id |
57f50b3c | 1407 | |
1408 | // check connection, in case connect | |
1409 | if(!Connect(kDAQ)){ | |
2c15234c | 1410 | Log(detector, "GetDAQFileSources - Couldn't connect to DAQ FXS database"); |
57f50b3c | 1411 | return 0; |
1412 | } | |
1413 | ||
1414 | // Query preparation | |
2c15234c | 1415 | TString sqlQueryStart = Form("select DAQsource from %s where", fConfig->GetFXSdbTable(kDAQ)); |
57f50b3c | 1416 | TString whereClause = Form("run=%d and detector=\"%s\" and fileId=\"%s\"", |
eba76848 | 1417 | GetCurrentRun(), detector, id); |
57f50b3c | 1418 | TString sqlQuery = Form("%s %s", sqlQueryStart.Data(), whereClause.Data()); |
1419 | ||
84090f85 | 1420 | AliDebug(2, Form("SQL query: \n%s",sqlQuery.Data())); |
57f50b3c | 1421 | |
1422 | // Query execution | |
1423 | TSQLResult* aResult; | |
1424 | aResult = fServer[kDAQ]->Query(sqlQuery); | |
1425 | if (!aResult) { | |
2bb7b766 | 1426 | Log(detector, Form("GetDAQFileSources - Can't execute SQL query for id: %s", id)); |
57f50b3c | 1427 | return 0; |
1428 | } | |
1429 | ||
1430 | if (aResult->GetRowCount() == 0) { | |
1431 | Log(detector, | |
2c15234c | 1432 | Form("GetDAQFileSources - No entry in FXS table for id: %s", id)); |
57f50b3c | 1433 | delete aResult; |
1434 | return 0; | |
1435 | } | |
1436 | ||
1437 | TSQLRow* aRow; | |
1438 | TList *list = new TList(); | |
1439 | list->SetOwner(1); | |
1440 | ||
1441 | while((aRow = aResult->Next())){ | |
1442 | ||
1443 | TString daqSource(aRow->GetField(0), aRow->GetFieldLength(0)); | |
84090f85 | 1444 | AliDebug(2, Form("daqSource = %s", daqSource.Data())); |
57f50b3c | 1445 | list->Add(new TObjString(daqSource)); |
2bb7b766 | 1446 | delete aRow; |
57f50b3c | 1447 | } |
1448 | delete aResult; | |
1449 | ||
1450 | return list; | |
1451 | ||
1452 | } | |
1453 | ||
1454 | //______________________________________________________________________________________________ | |
2bb7b766 | 1455 | const char* AliShuttle::GetDCSFileName(const char* /*detector*/, const char* /*id*/, const char* /*source*/){ |
2c15234c | 1456 | // Retrieves a file from the DCS FXS. |
2bb7b766 | 1457 | |
1458 | return "You're in DCS"; | |
1459 | ||
1460 | } | |
1461 | ||
1462 | //______________________________________________________________________________________________ | |
1463 | TList* AliShuttle::GetDCSFileSources(const char* /*detector*/, const char* /*id*/){ | |
2c15234c | 1464 | // Retrieves file sources from the DCS FXS. |
2bb7b766 | 1465 | |
1466 | return NULL; | |
1467 | ||
1468 | } | |
1469 | ||
1470 | //______________________________________________________________________________________________ | |
9e080f92 | 1471 | const char* AliShuttle::GetHLTFileName(const char* detector, const char* id, const char* source){ |
2c15234c | 1472 | // Retrieves a file from the HLT FXS. |
9e080f92 | 1473 | // First queris the HLT FXS database for the HLT file name, using the run, detector, id and source info |
1474 | // then calls RetrieveDAQFile(DAQfilename) for actual copy to local disk | |
1475 | // run: current run being processed (given by Logbook entry fLogbookEntry) | |
1476 | // detector: the Preprocessor name | |
1477 | // id: provided as a parameter by the Preprocessor | |
1478 | // source: provided by the Preprocessor through GetFileSources function | |
1479 | ||
1480 | // check connection, in case connect | |
1481 | if (!Connect(kHLT)) | |
1482 | { | |
1483 | Log(detector, "GetHLTFileName - Couldn't connect to HLT FXS database"); | |
1484 | return 0; | |
1485 | } | |
1486 | ||
1487 | // Query preparation | |
1488 | TString sqlQueryStart = Form("select filePath,fileSize,fileChecksum from %s where", | |
1489 | fConfig->GetFXSdbTable(kHLT)); | |
1490 | TString whereClause = Form("run=%d and detector=\"%s\" and fileId=\"%s\" and DDLnumbers=\"%s\"", | |
1491 | GetCurrentRun(), detector, id, source); | |
1492 | TString sqlQuery = Form("%s %s", sqlQueryStart.Data(), whereClause.Data()); | |
1493 | ||
1494 | AliDebug(2, Form("SQL query: \n%s",sqlQuery.Data())); | |
1495 | ||
1496 | // Query execution | |
1497 | TSQLResult* aResult = 0; | |
1498 | aResult = dynamic_cast<TSQLResult*> (fServer[kHLT]->Query(sqlQuery)); | |
1499 | if (!aResult) { | |
1500 | Log(detector, Form("GetHLTFileName - Can't execute SQL query for: id = %s, source = %s", | |
1501 | id, source)); | |
1502 | return 0; | |
1503 | } | |
1504 | ||
1505 | if(aResult->GetRowCount() == 0) | |
1506 | { | |
1507 | Log(detector, | |
1508 | Form("GetHLTFileName - No entry in FXS table for: id = %s, source = %s", | |
1509 | id, source)); | |
1510 | delete aResult; | |
1511 | return 0; | |
1512 | } | |
2bb7b766 | 1513 | |
9e080f92 | 1514 | if (aResult->GetRowCount() > 1) { |
1515 | Log(detector, | |
1516 | Form("GetHLTFileName - More than one entry in FXS table for: id = %s, source = %s", | |
1517 | id, source)); | |
1518 | delete aResult; | |
1519 | return 0; | |
1520 | } | |
1521 | ||
1522 | if (aResult->GetFieldCount() != 3) { | |
1523 | Log(detector, | |
1524 | Form("GetHLTFileName - Wrong field count in FXS table for: id = %s, source = %s", | |
1525 | id, source)); | |
1526 | delete aResult; | |
1527 | return 0; | |
1528 | } | |
1529 | ||
1530 | TSQLRow* aRow = dynamic_cast<TSQLRow*> (aResult->Next()); | |
1531 | ||
1532 | if (!aRow){ | |
1533 | Log(detector, Form("GetHLTFileName - Empty set result from query: id = %s, source = %s", | |
1534 | id, source)); | |
1535 | delete aResult; | |
1536 | return 0; | |
1537 | } | |
1538 | ||
1539 | TString filePath(aRow->GetField(0), aRow->GetFieldLength(0)); | |
1540 | TString fileSize(aRow->GetField(1), aRow->GetFieldLength(1)); | |
1541 | TString fileMd5Sum(aRow->GetField(2), aRow->GetFieldLength(2)); | |
1542 | ||
1543 | delete aResult; | |
1544 | delete aRow; | |
1545 | ||
1546 | AliDebug(2, Form("filePath = %s",filePath.Data())); | |
1547 | ||
1548 | // The full file path in HLT FXS is runNb/DET/DDLnumber/filePath | |
1549 | // TString fullFilePath = Form("%d/%s/%s/%s", GetCurrentRun(), detector, source, filePath.Data()); | |
1550 | ||
1551 | // retrieved file is renamed to make it unique | |
1552 | TString localFileName = Form("HLT_%s_%d_%s_%s.shuttle", | |
1553 | detector, GetCurrentRun(), id, source); | |
1554 | ||
1555 | // file retrieval from HLT FXS | |
1556 | Bool_t result = RetrieveHLTFile(filePath.Data(), localFileName.Data()); | |
1557 | if(!result) | |
1558 | { | |
1559 | Log(detector, Form("GetHLTFileName - Copy of file %s from HLT FXS failed", filePath.Data())); | |
1560 | return 0; | |
1561 | } else { | |
1562 | AliInfo(Form("File %s copied from HLT FXS into %s/%s", | |
1563 | filePath.Data(), fgkShuttleTempDir, localFileName.Data())); | |
1564 | } | |
1565 | ||
1566 | // compare md5sum of local file with the one stored in the HLT DB | |
1567 | Int_t md5Comp = gSystem->Exec(Form("md5sum %s/%s |grep %s 2>&1 > /dev/null", | |
1568 | fgkShuttleTempDir, localFileName.Data(), fileMd5Sum.Data())); | |
1569 | ||
1570 | if (md5Comp != 0) | |
1571 | { | |
1572 | Log(detector, Form("GetHLTFileName - md5sum of file %s does not match with local copy!", filePath.Data())); | |
1573 | return 0; | |
1574 | } | |
1575 | ||
1576 | fFXSCalled[kHLT]=kTRUE; | |
1577 | TObjString *fileParams = new TObjString(Form("%s#!?!#%s", id, source)); | |
1578 | fFXSlist[kHLT].Add(fileParams); | |
1579 | ||
1580 | static TString fullLocalFileName; | |
1581 | fullLocalFileName = TString::Format("%s/%s", fgkShuttleTempDir, localFileName.Data()); | |
1582 | ||
1583 | AliInfo(Form("fullLocalFileName = %s", fullLocalFileName.Data())); | |
1584 | ||
1585 | return fullLocalFileName.Data(); | |
2bb7b766 | 1586 | |
1587 | } | |
1588 | ||
1589 | //______________________________________________________________________________________________ | |
9e080f92 | 1590 | Bool_t AliShuttle::RetrieveHLTFile(const char* hltFileName, const char* localFileName) |
1591 | { | |
1592 | // Copies file from HLT FXS to local Shuttle machine | |
2bb7b766 | 1593 | |
9e080f92 | 1594 | // check temp directory: trying to cd to temp; if it does not exist, create it |
1595 | AliDebug(2, Form("Copy file %s from HLT FXS into %s/%s", | |
1596 | hltFileName, fgkShuttleTempDir, localFileName)); | |
1597 | ||
1598 | void* dir = gSystem->OpenDirectory(fgkShuttleTempDir); | |
1599 | if (dir == NULL) { | |
1600 | if (gSystem->mkdir(fgkShuttleTempDir, kTRUE)) { | |
1601 | AliError(Form("Can't open directory <%s>", fgkShuttleTempDir)); | |
1602 | return kFALSE; | |
1603 | } | |
1604 | ||
1605 | } else { | |
1606 | gSystem->FreeDirectory(dir); | |
1607 | } | |
1608 | ||
1609 | TString baseHLTFXSFolder = "~"; | |
1610 | TString command = Form("scp -oPort=%d %s@%s:%s/%s %s/%s", | |
1611 | fConfig->GetFXSPort(kHLT), | |
1612 | fConfig->GetFXSUser(kHLT), | |
1613 | fConfig->GetFXSHost(kHLT), | |
1614 | baseHLTFXSFolder.Data(), | |
1615 | hltFileName, | |
1616 | fgkShuttleTempDir, | |
1617 | localFileName); | |
1618 | ||
1619 | AliDebug(2, Form("%s",command.Data())); | |
1620 | ||
1621 | UInt_t nRetries = 0; | |
1622 | UInt_t maxRetries = 3; | |
1623 | ||
1624 | // copy!! if successful TSystem::Exec returns 0 | |
1625 | while(nRetries++ < maxRetries) { | |
1626 | AliDebug(2, Form("Trying to copy file. Retry # %d", nRetries)); | |
1627 | if(gSystem->Exec(command.Data()) == 0) return kTRUE; | |
1628 | } | |
1629 | ||
1630 | return kFALSE; | |
1631 | ||
1632 | } | |
1633 | ||
1634 | //______________________________________________________________________________________________ | |
1635 | TList* AliShuttle::GetHLTFileSources(const char* detector, const char* id){ | |
1636 | // Retrieves list of HLT sources (DDLnumbers) of file Id | |
1637 | ||
1638 | // check connection, in case connect | |
1639 | if(!Connect(kHLT)){ | |
1640 | Log(detector, "GetHLTFileSources - Couldn't connect to HLT FXS database"); | |
1641 | return 0; | |
1642 | } | |
1643 | ||
1644 | // Query preparation | |
1645 | TString sqlQueryStart = Form("select DDLnumbers from %s where", fConfig->GetFXSdbTable(kHLT)); | |
1646 | TString whereClause = Form("run=%d and detector=\"%s\" and fileId=\"%s\"", | |
1647 | GetCurrentRun(), detector, id); | |
1648 | TString sqlQuery = Form("%s %s", sqlQueryStart.Data(), whereClause.Data()); | |
1649 | ||
1650 | AliDebug(2, Form("SQL query: \n%s",sqlQuery.Data())); | |
1651 | ||
1652 | // Query execution | |
1653 | TSQLResult* aResult; | |
1654 | aResult = fServer[kHLT]->Query(sqlQuery); | |
1655 | if (!aResult) { | |
1656 | Log(detector, Form("GetHLTFileSources - Can't execute SQL query for id: %s", id)); | |
1657 | return 0; | |
1658 | } | |
1659 | ||
1660 | if (aResult->GetRowCount() == 0) { | |
1661 | Log(detector, | |
1662 | Form("GetHLTFileSources - No entry in FXS table for id: %s", id)); | |
1663 | delete aResult; | |
1664 | return 0; | |
1665 | } | |
1666 | ||
1667 | TSQLRow* aRow; | |
1668 | TList *list = new TList(); | |
1669 | list->SetOwner(1); | |
1670 | ||
1671 | while((aRow = aResult->Next())){ | |
1672 | ||
1673 | TString ddlNumbers(aRow->GetField(0), aRow->GetFieldLength(0)); | |
1674 | AliDebug(2, Form("DDLnumbers = %s", ddlNumbers.Data())); | |
1675 | list->Add(new TObjString(ddlNumbers)); | |
1676 | delete aRow; | |
1677 | } | |
1678 | delete aResult; | |
1679 | ||
1680 | return list; | |
2bb7b766 | 1681 | |
1682 | } | |
1683 | ||
1684 | //______________________________________________________________________________________________ | |
1685 | Bool_t AliShuttle::UpdateDAQTable() | |
1686 | { | |
57f50b3c | 1687 | // Update DAQ table filling time_processed field in all rows corresponding to current run and detector |
1688 | ||
1689 | // check connection, in case connect | |
1690 | if(!Connect(kDAQ)){ | |
2c15234c | 1691 | Log(fCurrentDetector, "UpdateDAQTable - Couldn't connect to DAQ FXS database"); |
57f50b3c | 1692 | return kFALSE; |
1693 | } | |
1694 | ||
1695 | TTimeStamp now; // now | |
1696 | ||
2c15234c | 1697 | // Loop on FXS list entries |
1698 | TIter iter(&fFXSlist[kDAQ]); | |
1699 | TObjString *aFXSentry=0; | |
1700 | while((aFXSentry = dynamic_cast<TObjString*> (iter.Next()))){ | |
1701 | TString aFXSentrystr = aFXSentry->String(); | |
9e080f92 | 1702 | TObjArray *aFXSarray = aFXSentrystr.Tokenize("#!?!#"); |
2c15234c | 1703 | if(!aFXSarray || aFXSarray->GetEntries() != 2 ) { |
1704 | Log(fCurrentDetector, Form("UpdateDAQTable - error updating FXS entry. Check string: <%s>", | |
1705 | aFXSentrystr.Data())); | |
1706 | if(aFXSarray) delete aFXSarray; | |
57f50b3c | 1707 | return kFALSE; |
1708 | } | |
2c15234c | 1709 | const char* fileId = ((TObjString*) aFXSarray->At(0))->GetName(); |
1710 | const char* daqSource = ((TObjString*) aFXSarray->At(1))->GetName(); | |
57f50b3c | 1711 | TString whereClause = Form("where run=%d and detector=\"%s\" and fileId=\"%s\" and DAQsource=\"%s\";", |
eba76848 | 1712 | GetCurrentRun(), fCurrentDetector.Data(), fileId, daqSource); |
57f50b3c | 1713 | |
2c15234c | 1714 | delete aFXSarray; |
57f50b3c | 1715 | |
2c15234c | 1716 | TString sqlQuery = Form("update %s set time_processed=%d %s", fConfig->GetFXSdbTable(kDAQ), |
1717 | now.GetSec(), whereClause.Data()); | |
57f50b3c | 1718 | |
84090f85 | 1719 | AliDebug(2, Form("SQL query: \n%s",sqlQuery.Data())); |
57f50b3c | 1720 | |
1721 | // Query execution | |
1722 | TSQLResult* aResult; | |
9e080f92 | 1723 | aResult = dynamic_cast<TSQLResult*> (fServer[kDAQ]->Query(sqlQuery)); |
57f50b3c | 1724 | if (!aResult) { |
2bb7b766 | 1725 | Log(fCurrentDetector, Form("UpdateDAQTable - Can't execute SQL query <%s>", sqlQuery.Data())); |
57f50b3c | 1726 | return kFALSE; |
1727 | } | |
1728 | delete aResult; | |
2bb7b766 | 1729 | } |
57f50b3c | 1730 | |
2bb7b766 | 1731 | return kTRUE; |
1732 | } | |
57f50b3c | 1733 | |
9e080f92 | 1734 | //______________________________________________________________________________________________ |
1735 | Bool_t AliShuttle::UpdateHLTTable() | |
1736 | { | |
1737 | // Update HLT table filling time_processed field in all rows corresponding to current run and detector | |
1738 | ||
1739 | // check connection, in case connect | |
1740 | if(!Connect(kHLT)){ | |
1741 | Log(fCurrentDetector, "UpdateHLTTable - Couldn't connect to HLT FXS database"); | |
1742 | return kFALSE; | |
1743 | } | |
1744 | ||
1745 | TTimeStamp now; // now | |
1746 | ||
1747 | // Loop on FXS list entries | |
1748 | TIter iter(&fFXSlist[kHLT]); | |
1749 | TObjString *aFXSentry=0; | |
1750 | while((aFXSentry = dynamic_cast<TObjString*> (iter.Next()))){ | |
1751 | TString aFXSentrystr = aFXSentry->String(); | |
1752 | TObjArray *aFXSarray = aFXSentrystr.Tokenize("#!?!#"); | |
1753 | if(!aFXSarray || aFXSarray->GetEntries() != 2 ) { | |
1754 | Log(fCurrentDetector, Form("UpdateHLTTable - error updating FXS entry. Check string: <%s>", | |
1755 | aFXSentrystr.Data())); | |
1756 | if(aFXSarray) delete aFXSarray; | |
1757 | return kFALSE; | |
1758 | } | |
1759 | const char* fileId = ((TObjString*) aFXSarray->At(0))->GetName(); | |
1760 | const char* hltSource = ((TObjString*) aFXSarray->At(1))->GetName(); | |
1761 | TString whereClause = Form("where run=%d and detector=\"%s\" and fileId=\"%s\" and DDLnumbers=\"%s\";", | |
1762 | GetCurrentRun(), fCurrentDetector.Data(), fileId, hltSource); | |
1763 | ||
1764 | delete aFXSarray; | |
1765 | ||
1766 | TString sqlQuery = Form("update %s set time_processed=%d %s", fConfig->GetFXSdbTable(kHLT), | |
1767 | now.GetSec(), whereClause.Data()); | |
1768 | ||
1769 | AliDebug(2, Form("SQL query: \n%s",sqlQuery.Data())); | |
1770 | ||
1771 | // Query execution | |
1772 | TSQLResult* aResult; | |
1773 | aResult = dynamic_cast<TSQLResult*> (fServer[kHLT]->Query(sqlQuery)); | |
1774 | if (!aResult) { | |
1775 | Log(fCurrentDetector, Form("UpdateHLTTable - Can't execute SQL query <%s>", sqlQuery.Data())); | |
1776 | return kFALSE; | |
1777 | } | |
1778 | delete aResult; | |
1779 | } | |
1780 | ||
1781 | return kTRUE; | |
1782 | } | |
57f50b3c | 1783 | |
2bb7b766 | 1784 | //______________________________________________________________________________________________ |
1785 | Bool_t AliShuttle::UpdateShuttleLogbook(const char* detector, const char* status) | |
1786 | { | |
1787 | // Update Shuttle logbook filling detector or shuttle_done column | |
1788 | // ex. of usage: UpdateShuttleLogbook("PHOS", "DONE") or UpdateShuttleLogbook("shuttle_done") | |
57f50b3c | 1789 | |
2bb7b766 | 1790 | // check connection, in case connect |
be48e3ea | 1791 | if(!Connect(3)){ |
2bb7b766 | 1792 | Log("SHUTTLE", "UpdateShuttleLogbook - Couldn't connect to DAQ Logbook."); |
1793 | return kFALSE; | |
57f50b3c | 1794 | } |
1795 | ||
2bb7b766 | 1796 | TString detName(detector); |
1797 | TString setClause; | |
1798 | if(detName == "shuttle_done") { | |
1799 | setClause = "set shuttle_done=1"; | |
1800 | } else { | |
2bb7b766 | 1801 | TString statusStr(status); |
1802 | if(statusStr.Contains("done", TString::kIgnoreCase) || | |
1803 | statusStr.Contains("failed", TString::kIgnoreCase)){ | |
eba76848 | 1804 | setClause = Form("set %s=\"%s\"", detector, status); |
2bb7b766 | 1805 | } else { |
1806 | Log("SHUTTLE", | |
1807 | Form("UpdateShuttleLogbook - Invalid status <%s> for detector %s", | |
1808 | status, detector)); | |
1809 | return kFALSE; | |
1810 | } | |
1811 | } | |
57f50b3c | 1812 | |
2bb7b766 | 1813 | TString whereClause = Form("where run=%d", GetCurrentRun()); |
1814 | ||
1815 | TString sqlQuery = Form("update logbook_shuttle %s %s", | |
1816 | setClause.Data(), whereClause.Data()); | |
57f50b3c | 1817 | |
2bb7b766 | 1818 | AliDebug(2, Form("SQL query: \n%s",sqlQuery.Data())); |
1819 | ||
1820 | // Query execution | |
1821 | TSQLResult* aResult; | |
be48e3ea | 1822 | aResult = dynamic_cast<TSQLResult*> (fServer[3]->Query(sqlQuery)); |
2bb7b766 | 1823 | if (!aResult) { |
1824 | Log("SHUTTLE", Form("UpdateShuttleLogbook - Can't execute query <%s>", sqlQuery.Data())); | |
1825 | return kFALSE; | |
57f50b3c | 1826 | } |
2bb7b766 | 1827 | delete aResult; |
57f50b3c | 1828 | |
1829 | return kTRUE; | |
1830 | } | |
1831 | ||
1832 | //______________________________________________________________________________________________ | |
2bb7b766 | 1833 | Int_t AliShuttle::GetCurrentRun() const |
1834 | { | |
1835 | // Get current run from logbook entry | |
57f50b3c | 1836 | |
2bb7b766 | 1837 | return fLogbookEntry ? fLogbookEntry->GetRun() : -1; |
57f50b3c | 1838 | } |
1839 | ||
1840 | //______________________________________________________________________________________________ | |
2bb7b766 | 1841 | UInt_t AliShuttle::GetCurrentStartTime() const |
1842 | { | |
1843 | // get current start time | |
57f50b3c | 1844 | |
2bb7b766 | 1845 | return fLogbookEntry ? fLogbookEntry->GetStartTime() : 0; |
57f50b3c | 1846 | } |
1847 | ||
1848 | //______________________________________________________________________________________________ | |
2bb7b766 | 1849 | UInt_t AliShuttle::GetCurrentEndTime() const |
1850 | { | |
1851 | // get current end time from logbook entry | |
57f50b3c | 1852 | |
2bb7b766 | 1853 | return fLogbookEntry ? fLogbookEntry->GetEndTime() : 0; |
57f50b3c | 1854 | } |
1855 | ||
b948db8d | 1856 | //______________________________________________________________________________________________ |
1857 | void AliShuttle::Log(const char* detector, const char* message) | |
1858 | { | |
58bc3020 | 1859 | // Fill log string with a message |
b948db8d | 1860 | |
84090f85 | 1861 | void* dir = gSystem->OpenDirectory(fgkShuttleLogDir); |
1862 | if (dir == NULL) { | |
1863 | if (gSystem->mkdir(fgkShuttleLogDir, kTRUE)) { | |
2bb7b766 | 1864 | AliError(Form("Can't open directory <%s>", fgkShuttleTempDir)); |
84090f85 | 1865 | return; |
1866 | } | |
b948db8d | 1867 | |
84090f85 | 1868 | } else { |
1869 | gSystem->FreeDirectory(dir); | |
1870 | } | |
b948db8d | 1871 | |
cb343cfd | 1872 | TString toLog = Form("%s (%d): %s - ", TTimeStamp(time(0)).AsString("s"), getpid(), detector); |
2bb7b766 | 1873 | if(GetCurrentRun()>=0 ) toLog += Form("run %d - ", GetCurrentRun()); |
1874 | toLog += Form("%s", message); | |
1875 | ||
84090f85 | 1876 | AliInfo(toLog.Data()); |
b948db8d | 1877 | |
84090f85 | 1878 | TString fileName; |
1879 | fileName.Form("%s/%s.log", fgkShuttleLogDir, detector); | |
1880 | gSystem->ExpandPathName(fileName); | |
1881 | ||
1882 | ofstream logFile; | |
1883 | logFile.open(fileName, ofstream::out | ofstream::app); | |
1884 | ||
1885 | if (!logFile.is_open()) { | |
1886 | AliError(Form("Could not open file %s", fileName.Data())); | |
1887 | return; | |
1888 | } | |
7bfb2090 | 1889 | |
84090f85 | 1890 | logFile << toLog.Data() << "\n"; |
b948db8d | 1891 | |
84090f85 | 1892 | logFile.close(); |
b948db8d | 1893 | } |
2bb7b766 | 1894 | |
2bb7b766 | 1895 | //______________________________________________________________________________________________ |
1896 | Bool_t AliShuttle::Collect(Int_t run) | |
1897 | { | |
eba76848 | 1898 | // |
1899 | // Collects conditions data for all UNPROCESSED run written to DAQ LogBook in case of run = -1 (default) | |
1900 | // If a dedicated run is given this run is processed | |
1901 | // | |
1902 | // In operational mode, this is the Shuttle function triggered by the EOR signal. | |
1903 | // | |
2bb7b766 | 1904 | |
eba76848 | 1905 | if (run == -1) |
1906 | Log("SHUTTLE","Collect - Shuttle called. Collecting conditions data for unprocessed runs"); | |
1907 | else | |
1908 | Log("SHUTTLE", Form("Collect - Shuttle called. Collecting conditions data for run %d", run)); | |
cb343cfd | 1909 | |
1910 | SetLastAction("Starting"); | |
2bb7b766 | 1911 | |
1912 | TString whereClause("where shuttle_done=0"); | |
eba76848 | 1913 | if (run != -1) |
1914 | whereClause += Form(" and run=%d", run); | |
2bb7b766 | 1915 | |
1916 | TObjArray shuttleLogbookEntries; | |
be48e3ea | 1917 | if (!QueryShuttleLogbook(whereClause, shuttleLogbookEntries)) |
1918 | { | |
cb343cfd | 1919 | Log("SHUTTLE", "Collect - Can't retrieve entries from Shuttle logbook"); |
2bb7b766 | 1920 | return kFALSE; |
1921 | } | |
1922 | ||
9e080f92 | 1923 | if (shuttleLogbookEntries.GetEntries() == 0) |
1924 | { | |
1925 | if (run == -1) | |
1926 | Log("SHUTTLE","Collect - Found no UNPROCESSED runs in Shuttle logbook"); | |
1927 | else | |
1928 | Log("SHUTTLE", Form("Collect - Run %d is already DONE " | |
1929 | "or it does not exist in Shuttle logbook", run)); | |
1930 | return kTRUE; | |
1931 | } | |
1932 | ||
be48e3ea | 1933 | for (UInt_t iDet=0; iDet<NDetectors(); iDet++) |
1934 | fFirstUnprocessed[iDet] = kTRUE; | |
1935 | ||
fc5a4708 | 1936 | if (run != -1) |
be48e3ea | 1937 | { |
1938 | // query Shuttle logbook for earlier runs, check if some detectors are unprocessed, | |
1939 | // flag them into fFirstUnprocessed array | |
1940 | TString whereClause(Form("where shuttle_done=0 and run < %d", run)); | |
1941 | TObjArray tmpLogbookEntries; | |
1942 | if (!QueryShuttleLogbook(whereClause, tmpLogbookEntries)) | |
1943 | { | |
1944 | Log("SHUTTLE", "Collect - Can't retrieve entries from Shuttle logbook"); | |
1945 | return kFALSE; | |
1946 | } | |
1947 | ||
1948 | TIter iter(&tmpLogbookEntries); | |
1949 | AliShuttleLogbookEntry* anEntry = 0; | |
1950 | while ((anEntry = dynamic_cast<AliShuttleLogbookEntry*> (iter.Next()))) | |
1951 | { | |
1952 | for (UInt_t iDet=0; iDet<NDetectors(); iDet++) | |
1953 | { | |
1954 | if (anEntry->GetDetectorStatus(iDet) == AliShuttleLogbookEntry::kUnprocessed) | |
1955 | { | |
1956 | AliDebug(2, Form("Run %d: setting %s as \"not first time unprocessed\"", | |
1957 | anEntry->GetRun(), GetDetName(iDet))); | |
1958 | fFirstUnprocessed[iDet] = kFALSE; | |
1959 | } | |
1960 | } | |
1961 | ||
1962 | } | |
1963 | ||
1964 | } | |
1965 | ||
1966 | if (!RetrieveConditionsData(shuttleLogbookEntries)) | |
1967 | { | |
cb343cfd | 1968 | Log("SHUTTLE", "Collect - Process of at least one run failed"); |
2bb7b766 | 1969 | return kFALSE; |
1970 | } | |
1971 | ||
eba76848 | 1972 | return kTRUE; |
2bb7b766 | 1973 | } |
1974 | ||
2bb7b766 | 1975 | //______________________________________________________________________________________________ |
1976 | Bool_t AliShuttle::RetrieveConditionsData(const TObjArray& dateEntries) | |
1977 | { | |
1978 | // Retrieve conditions data for all runs that aren't processed yet | |
1979 | ||
1980 | Bool_t hasError = kFALSE; | |
1981 | ||
1982 | TIter iter(&dateEntries); | |
1983 | AliShuttleLogbookEntry* anEntry; | |
1984 | ||
1985 | while ((anEntry = (AliShuttleLogbookEntry*) iter.Next())){ | |
1986 | if (!Process(anEntry)){ | |
1987 | hasError = kTRUE; | |
1988 | } | |
1989 | } | |
1990 | ||
1991 | return hasError == kFALSE; | |
1992 | } | |
cb343cfd | 1993 | |
1994 | //______________________________________________________________________________________________ | |
1995 | ULong_t AliShuttle::GetTimeOfLastAction() const | |
1996 | { | |
1997 | ULong_t tmp; | |
1998 | ||
1999 | fMonitoringMutex->Lock(); | |
be48e3ea | 2000 | |
cb343cfd | 2001 | tmp = fLastActionTime; |
2002 | ||
2003 | fMonitoringMutex->UnLock(); | |
2004 | ||
2005 | return tmp; | |
2006 | } | |
2007 | ||
2008 | //______________________________________________________________________________________________ | |
2009 | const TString AliShuttle::GetLastAction() const | |
2010 | { | |
2011 | // returns a string description of the last action | |
2012 | ||
2013 | TString tmp; | |
2014 | ||
2015 | fMonitoringMutex->Lock(); | |
2016 | ||
2017 | tmp = fLastAction; | |
2018 | ||
2019 | fMonitoringMutex->UnLock(); | |
2020 | ||
2021 | return tmp; | |
2022 | } | |
2023 | ||
2024 | //______________________________________________________________________________________________ | |
2025 | void AliShuttle::SetLastAction(const char* action) | |
2026 | { | |
2027 | // updates the monitoring variables | |
2028 | ||
2029 | fMonitoringMutex->Lock(); | |
2030 | ||
2031 | fLastAction = action; | |
2032 | fLastActionTime = time(0); | |
2033 | ||
2034 | fMonitoringMutex->UnLock(); | |
2035 | } | |
eba76848 | 2036 | |
2037 | //______________________________________________________________________________________________ | |
2038 | const char* AliShuttle::GetRunParameter(const char* param) | |
2039 | { | |
2040 | // returns run parameter read from DAQ logbook | |
2041 | ||
2042 | if(!fLogbookEntry) { | |
2043 | AliError("No logbook entry!"); | |
2044 | return 0; | |
2045 | } | |
2046 | ||
2047 | return fLogbookEntry->GetRunParameter(param); | |
2048 | } | |
57c1a579 | 2049 | |
2050 | //______________________________________________________________________________________________ | |
2051 | Bool_t AliShuttle::SendMail() | |
2052 | { | |
2053 | // sends a mail to the subdetector expert in case of preprocessor error | |
2054 | ||
2055 | void* dir = gSystem->OpenDirectory(fgkShuttleLogDir); | |
2056 | if (dir == NULL) | |
2057 | { | |
2058 | if (gSystem->mkdir(fgkShuttleLogDir, kTRUE)) | |
2059 | { | |
2060 | AliError(Form("Can't open directory <%s>", fgkShuttleTempDir)); | |
2061 | return kFALSE; | |
2062 | } | |
2063 | ||
2064 | } else { | |
2065 | gSystem->FreeDirectory(dir); | |
2066 | } | |
2067 | ||
2068 | TString bodyFileName; | |
2069 | bodyFileName.Form("%s/mail.body", fgkShuttleLogDir); | |
2070 | gSystem->ExpandPathName(bodyFileName); | |
2071 | ||
2072 | ofstream mailBody; | |
2073 | mailBody.open(bodyFileName, ofstream::out); | |
2074 | ||
2075 | if (!mailBody.is_open()) | |
2076 | { | |
2077 | AliError(Form("Could not open mail body file %s", bodyFileName.Data())); | |
2078 | return kFALSE; | |
2079 | } | |
2080 | ||
2081 | TString to=""; | |
2082 | TIter iterExperts(fConfig->GetResponsibles(fCurrentDetector)); | |
2083 | TObjString *anExpert=0; | |
2084 | while ((anExpert = (TObjString*) iterExperts.Next())) | |
2085 | { | |
2086 | to += Form("%s,", anExpert->GetName()); | |
2087 | } | |
2088 | to.Remove(to.Length()-1); | |
909732f7 | 2089 | AliDebug(2, Form("to: %s",to.Data())); |
57c1a579 | 2090 | |
2091 | TString cc="alberto.colla@cern.ch"; | |
2092 | ||
2093 | TString subject = Form("%s Shuttle preprocessor error in run %d !", | |
2094 | fCurrentDetector.Data(), GetCurrentRun()); | |
909732f7 | 2095 | AliDebug(2, Form("subject: %s", subject.Data())); |
57c1a579 | 2096 | |
2097 | TString body = Form("Dear %s expert(s), \n\n", fCurrentDetector.Data()); | |
2098 | body += Form("SHUTTLE just detected that your preprocessor " | |
2099 | "exited with ERROR state in run %d !!\n\n", GetCurrentRun()); | |
2100 | body += Form("Please check %s status on the web page asap!\n\n", fCurrentDetector.Data()); | |
2101 | body += Form("The last 10 lines of %s log file are following:\n\n"); | |
2102 | ||
909732f7 | 2103 | AliDebug(2, Form("Body begin: %s", body.Data())); |
57c1a579 | 2104 | |
2105 | mailBody << body.Data(); | |
2106 | mailBody.close(); | |
2107 | mailBody.open(bodyFileName, ofstream::out | ofstream::app); | |
2108 | ||
2109 | TString logFileName = Form("%s/%s.log", fgkShuttleLogDir, fCurrentDetector.Data()); | |
2110 | TString tailCommand = Form("tail -n 10 %s >> %s", logFileName.Data(), bodyFileName.Data()); | |
2111 | if (gSystem->Exec(tailCommand.Data())) | |
2112 | { | |
2113 | mailBody << Form("%s log file not found ...\n\n", fCurrentDetector.Data()); | |
2114 | } | |
2115 | ||
2116 | TString endBody = Form("------------------------------------------------------\n\n"); | |
2117 | endBody += Form("In case of problems please contact the SHUTTLE core team!\n\n"); | |
2118 | endBody += "Please do not answer this message directly, it is automatically generated!\n\n"; | |
2119 | endBody += "Sincerely yours,\n\n \t\t\tthe SHUTTLE\n"; | |
2120 | ||
909732f7 | 2121 | AliDebug(2, Form("Body end: %s", endBody.Data())); |
57c1a579 | 2122 | |
2123 | mailBody << endBody.Data(); | |
2124 | ||
2125 | mailBody.close(); | |
2126 | ||
2127 | // send mail! | |
2128 | TString mailCommand = Form("mail -s \"%s\" -c %s %s < %s", | |
2129 | subject.Data(), | |
2130 | cc.Data(), | |
2131 | to.Data(), | |
2132 | bodyFileName.Data()); | |
909732f7 | 2133 | AliDebug(2, Form("mail command: %s", mailCommand.Data())); |
57c1a579 | 2134 | |
2135 | Bool_t result = gSystem->Exec(mailCommand.Data()); | |
2136 | ||
2137 | return result == 0; | |
2138 | } |