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 | $Log$ |
e7f62f16 |
18 | Revision 1.26 2007/01/23 19:20:03 acolla |
19 | Removed old ldif files, added TOF, MCH ldif files. Added some options in |
20 | AliShuttleConfig::Print. Added in Ali Shuttle: SetShuttleTempDir and |
21 | SetShuttleLogDir |
22 | |
36c99a6a |
23 | Revision 1.25 2007/01/15 19:13:52 acolla |
24 | Moved some AliInfo to AliDebug in SendMail function |
25 | |
fc5a4708 |
26 | Revision 1.21 2006/12/07 08:51:26 jgrosseo |
27 | update (alberto): |
28 | table, db names in ldap configuration |
29 | added GRP preprocessor |
30 | DCS data can also be retrieved by data point |
31 | |
2c15234c |
32 | Revision 1.20 2006/11/16 16:16:48 jgrosseo |
33 | introducing strict run ordering flag |
34 | removed giving preprocessor name to preprocessor, they have to know their name themselves ;-) |
35 | |
be48e3ea |
36 | Revision 1.19 2006/11/06 14:23:04 jgrosseo |
37 | major update (Alberto) |
38 | o) reading of run parameters from the logbook |
39 | o) online offline naming conversion |
40 | o) standalone DCSclient package |
41 | |
eba76848 |
42 | Revision 1.18 2006/10/20 15:22:59 jgrosseo |
43 | o) Adding time out to the execution of the preprocessors: The Shuttle forks and the parent process monitors the child |
44 | o) Merging Collect, CollectAll, CollectNew function |
45 | o) Removing implementation of empty copy constructors (declaration still there!) |
46 | |
cb343cfd |
47 | Revision 1.17 2006/10/05 16:20:55 jgrosseo |
48 | adapting to new CDB classes |
49 | |
6ec0e06c |
50 | Revision 1.16 2006/10/05 15:46:26 jgrosseo |
51 | applying to the new interface |
52 | |
481441a2 |
53 | Revision 1.15 2006/10/02 16:38:39 jgrosseo |
54 | update (alberto): |
55 | fixed memory leaks |
56 | storing of objects that failed to be stored to the grid before |
57 | interfacing of shuttle status table in daq system |
58 | |
2bb7b766 |
59 | Revision 1.14 2006/08/29 09:16:05 jgrosseo |
60 | small update |
61 | |
85a80aa9 |
62 | Revision 1.13 2006/08/15 10:50:00 jgrosseo |
63 | effc++ corrections (alberto) |
64 | |
4f0ab988 |
65 | Revision 1.12 2006/08/08 14:19:29 jgrosseo |
66 | Update to shuttle classes (Alberto) |
67 | |
68 | - Possibility to set the full object's path in the Preprocessor's and |
69 | Shuttle's Store functions |
70 | - Possibility to extend the object's run validity in the same classes |
71 | ("startValidity" and "validityInfinite" parameters) |
72 | - Implementation of the StoreReferenceData function to store reference |
73 | data in a dedicated CDB storage. |
74 | |
84090f85 |
75 | Revision 1.11 2006/07/21 07:37:20 jgrosseo |
76 | last run is stored after each run |
77 | |
7bfb2090 |
78 | Revision 1.10 2006/07/20 09:54:40 jgrosseo |
79 | introducing status management: The processing per subdetector is divided into several steps, |
80 | after each step the status is stored on disk. If the system crashes in any of the steps the Shuttle |
81 | can keep track of the number of failures and skips further processing after a certain threshold is |
82 | exceeded. These thresholds can be configured in LDAP. |
83 | |
5164a766 |
84 | Revision 1.9 2006/07/19 10:09:55 jgrosseo |
85 | new configuration, accesst to DAQ FES (Alberto) |
86 | |
57f50b3c |
87 | Revision 1.8 2006/07/11 12:44:36 jgrosseo |
88 | adding parameters for extended validity range of data produced by preprocessor |
89 | |
17111222 |
90 | Revision 1.7 2006/07/10 14:37:09 jgrosseo |
91 | small fix + todo comment |
92 | |
e090413b |
93 | Revision 1.6 2006/07/10 13:01:41 jgrosseo |
94 | enhanced storing of last sucessfully processed run (alberto) |
95 | |
a7160fe9 |
96 | Revision 1.5 2006/07/04 14:59:57 jgrosseo |
97 | revision of AliDCSValue: Removed wrapper classes, reduced storage size per value by factor 2 |
98 | |
45a493ce |
99 | Revision 1.4 2006/06/12 09:11:16 jgrosseo |
100 | coding conventions (Alberto) |
101 | |
58bc3020 |
102 | Revision 1.3 2006/06/06 14:26:40 jgrosseo |
103 | o) removed files that were moved to STEER |
104 | o) shuttle updated to follow the new interface (Alberto) |
105 | |
b948db8d |
106 | Revision 1.2 2006/03/07 07:52:34 hristov |
107 | New version (B.Yordanov) |
108 | |
d477ad88 |
109 | Revision 1.6 2005/11/19 17:19:14 byordano |
110 | RetrieveDATEEntries and RetrieveConditionsData added |
111 | |
112 | Revision 1.5 2005/11/19 11:09:27 byordano |
113 | AliShuttle declaration added |
114 | |
115 | Revision 1.4 2005/11/17 17:47:34 byordano |
116 | TList changed to TObjArray |
117 | |
118 | Revision 1.3 2005/11/17 14:43:23 byordano |
119 | import to local CVS |
120 | |
121 | Revision 1.1.1.1 2005/10/28 07:33:58 hristov |
122 | Initial import as subdirectory in AliRoot |
123 | |
73abe331 |
124 | Revision 1.2 2005/09/13 08:41:15 byordano |
125 | default startTime endTime added |
126 | |
127 | Revision 1.4 2005/08/30 09:13:02 byordano |
128 | some docs added |
129 | |
130 | Revision 1.3 2005/08/29 21:15:47 byordano |
131 | some docs added |
132 | |
133 | */ |
134 | |
135 | // |
136 | // This class is the main manager for AliShuttle. |
137 | // It organizes the data retrieval from DCS and call the |
b948db8d |
138 | // interface methods of AliPreprocessor. |
73abe331 |
139 | // For every detector in AliShuttleConfgi (see AliShuttleConfig), |
140 | // data for its set of aliases is retrieved. If there is registered |
b948db8d |
141 | // AliPreprocessor for this detector then it will be used |
142 | // accroding to the schema (see AliPreprocessor). |
143 | // If there isn't registered AliPreprocessor than the retrieved |
73abe331 |
144 | // data is stored automatically to the undelying AliCDBStorage. |
145 | // For detSpec is used the alias name. |
146 | // |
147 | |
148 | #include "AliShuttle.h" |
149 | |
150 | #include "AliCDBManager.h" |
151 | #include "AliCDBStorage.h" |
152 | #include "AliCDBId.h" |
84090f85 |
153 | #include "AliCDBRunRange.h" |
154 | #include "AliCDBPath.h" |
5164a766 |
155 | #include "AliCDBEntry.h" |
73abe331 |
156 | #include "AliShuttleConfig.h" |
eba76848 |
157 | #include "DCSClient/AliDCSClient.h" |
73abe331 |
158 | #include "AliLog.h" |
b948db8d |
159 | #include "AliPreprocessor.h" |
5164a766 |
160 | #include "AliShuttleStatus.h" |
2bb7b766 |
161 | #include "AliShuttleLogbookEntry.h" |
73abe331 |
162 | |
57f50b3c |
163 | #include <TSystem.h> |
58bc3020 |
164 | #include <TObject.h> |
b948db8d |
165 | #include <TString.h> |
57f50b3c |
166 | #include <TTimeStamp.h> |
73abe331 |
167 | #include <TObjString.h> |
57f50b3c |
168 | #include <TSQLServer.h> |
169 | #include <TSQLResult.h> |
170 | #include <TSQLRow.h> |
cb343cfd |
171 | #include <TMutex.h> |
73abe331 |
172 | |
e7f62f16 |
173 | #include <TMonaLisaWriter.h> |
174 | |
5164a766 |
175 | #include <fstream> |
176 | |
cb343cfd |
177 | #include <sys/types.h> |
178 | #include <sys/wait.h> |
179 | |
73abe331 |
180 | ClassImp(AliShuttle) |
181 | |
2bb7b766 |
182 | TString AliShuttle::fgkMainCDB("alien://folder=ShuttleCDB"); |
84090f85 |
183 | TString AliShuttle::fgkLocalCDB("local://LocalShuttleCDB"); |
2bb7b766 |
184 | TString AliShuttle::fgkMainRefStorage("alien://folder=ShuttleReference"); |
84090f85 |
185 | TString AliShuttle::fgkLocalRefStorage("local://LocalReferenceStorage"); |
186 | |
4f0ab988 |
187 | Bool_t AliShuttle::fgkProcessDCS(kTRUE); |
188 | |
36c99a6a |
189 | TString AliShuttle::fgkShuttleTempDir = gSystem->ExpandPathName("$ALICE_ROOT/SHUTTLE/temp"); |
190 | TString AliShuttle::fgkShuttleLogDir = gSystem->ExpandPathName("$ALICE_ROOT/SHUTTLE/log"); |
57f50b3c |
191 | |
b948db8d |
192 | //______________________________________________________________________________________________ |
193 | AliShuttle::AliShuttle(const AliShuttleConfig* config, |
194 | UInt_t timeout, Int_t retries): |
4f0ab988 |
195 | fConfig(config), |
196 | fTimeout(timeout), fRetries(retries), |
197 | fPreprocessorMap(), |
2bb7b766 |
198 | fLogbookEntry(0), |
eba76848 |
199 | fCurrentDetector(), |
85a80aa9 |
200 | fStatusEntry(0), |
cb343cfd |
201 | fGridError(kFALSE), |
202 | fMonitoringMutex(0), |
eba76848 |
203 | fLastActionTime(0), |
e7f62f16 |
204 | fLastAction(), |
205 | fMonaLisa(0) |
73abe331 |
206 | { |
207 | // |
208 | // config: AliShuttleConfig used |
73abe331 |
209 | // timeout: timeout used for AliDCSClient connection |
210 | // retries: the number of retries in case of connection error. |
211 | // |
212 | |
57f50b3c |
213 | if (!fConfig->IsValid()) AliFatal("********** !!!!! Invalid configuration !!!!! **********"); |
be48e3ea |
214 | for(int iSys=0;iSys<4;iSys++) { |
57f50b3c |
215 | fServer[iSys]=0; |
be48e3ea |
216 | if (iSys < 3) |
2c15234c |
217 | fFXSlist[iSys].SetOwner(kTRUE); |
57f50b3c |
218 | } |
2bb7b766 |
219 | fPreprocessorMap.SetOwner(kTRUE); |
be48e3ea |
220 | |
221 | for (UInt_t iDet=0; iDet<NDetectors(); iDet++) |
222 | fFirstUnprocessed[iDet] = kFALSE; |
223 | |
cb343cfd |
224 | fMonitoringMutex = new TMutex(); |
58bc3020 |
225 | } |
226 | |
b948db8d |
227 | //______________________________________________________________________________________________ |
57f50b3c |
228 | AliShuttle::~AliShuttle() |
58bc3020 |
229 | { |
230 | // destructor |
231 | |
b948db8d |
232 | fPreprocessorMap.DeleteAll(); |
be48e3ea |
233 | for(int iSys=0;iSys<4;iSys++) |
57f50b3c |
234 | if(fServer[iSys]) { |
235 | fServer[iSys]->Close(); |
236 | delete fServer[iSys]; |
eba76848 |
237 | fServer[iSys] = 0; |
57f50b3c |
238 | } |
2bb7b766 |
239 | |
240 | if (fStatusEntry){ |
241 | delete fStatusEntry; |
242 | fStatusEntry = 0; |
243 | } |
cb343cfd |
244 | |
245 | if (fMonitoringMutex) |
246 | { |
247 | delete fMonitoringMutex; |
248 | fMonitoringMutex = 0; |
249 | } |
73abe331 |
250 | } |
251 | |
b948db8d |
252 | //______________________________________________________________________________________________ |
57f50b3c |
253 | void AliShuttle::RegisterPreprocessor(AliPreprocessor* preprocessor) |
58bc3020 |
254 | { |
73abe331 |
255 | // |
b948db8d |
256 | // Registers new AliPreprocessor. |
73abe331 |
257 | // It uses GetName() for indentificator of the pre processor. |
258 | // The pre processor is registered it there isn't any other |
259 | // with the same identificator (GetName()). |
260 | // |
261 | |
eba76848 |
262 | const char* detName = preprocessor->GetName(); |
263 | if(GetDetPos(detName) < 0) |
264 | AliFatal(Form("********** !!!!! Invalid detector name: %s !!!!! **********", detName)); |
265 | |
266 | if (fPreprocessorMap.GetValue(detName)) { |
267 | AliWarning(Form("AliPreprocessor %s is already registered!", detName)); |
73abe331 |
268 | return; |
269 | } |
270 | |
eba76848 |
271 | fPreprocessorMap.Add(new TObjString(detName), preprocessor); |
73abe331 |
272 | } |
b948db8d |
273 | //______________________________________________________________________________________________ |
84090f85 |
274 | UInt_t AliShuttle::Store(const AliCDBPath& path, TObject* object, |
275 | AliCDBMetaData* metaData, Int_t validityStart, Bool_t validityInfinite) |
73abe331 |
276 | { |
84090f85 |
277 | // Stores a CDB object in the storage for offline reconstruction. Objects that are not needed for |
278 | // offline reconstruction, but should be stored anyway (e.g. for debugging) should NOT be stored |
279 | // using this function. Use StoreReferenceData instead! |
85a80aa9 |
280 | // It calls WriteToCDB function which perform actual storage |
b948db8d |
281 | |
85a80aa9 |
282 | return WriteToCDB(fgkMainCDB, fgkLocalCDB, path, object, |
283 | metaData, validityStart, validityInfinite); |
84090f85 |
284 | |
285 | } |
286 | |
287 | //______________________________________________________________________________________________ |
481441a2 |
288 | UInt_t AliShuttle::StoreReferenceData(const AliCDBPath& path, TObject* object, AliCDBMetaData* metaData) |
84090f85 |
289 | { |
290 | // Stores a CDB object in the storage for reference data. This objects will not be available during |
291 | // offline reconstrunction. Use this function for reference data only! |
85a80aa9 |
292 | // It calls WriteToCDB function which perform actual storage |
293 | |
481441a2 |
294 | return WriteToCDB(fgkMainRefStorage, fgkLocalRefStorage, path, object, metaData); |
84090f85 |
295 | |
85a80aa9 |
296 | } |
297 | |
298 | //______________________________________________________________________________________________ |
299 | UInt_t AliShuttle::WriteToCDB(const char* mainUri, const char* localUri, |
300 | const AliCDBPath& path, TObject* object, AliCDBMetaData* metaData, |
301 | Int_t validityStart, Bool_t validityInfinite) |
302 | { |
303 | // write object into the CDB. Parameters are passed by Store and StoreReferenceData functions. |
304 | // The parameters are: |
305 | // 1) Uri of the main storage (Grid) |
306 | // 2) Uri of the backup storage (Local) |
307 | // 3) the object's path. |
308 | // 4) the object to be stored |
309 | // 5) the metaData to be associated with the object |
310 | // 6) the validity start run number w.r.t. the current run, |
84090f85 |
311 | // if the data is valid only for this run leave the default 0 |
85a80aa9 |
312 | // 7) specifies if the calibration data is valid for infinity (this means until updated), |
84090f85 |
313 | // typical for calibration runs, the default is kFALSE |
314 | // |
84090f85 |
315 | // returns 0 if fail |
85a80aa9 |
316 | // 1 if stored in main (Grid) storage |
317 | // 2 if stored in backup (Local) storage |
84090f85 |
318 | |
85a80aa9 |
319 | const char* cdbType = (mainUri == fgkMainCDB) ? "CDB" : "Reference"; |
2bb7b766 |
320 | |
85a80aa9 |
321 | Int_t firstRun = GetCurrentRun() - validityStart; |
84090f85 |
322 | if(firstRun < 0) { |
2bb7b766 |
323 | AliError("First valid run happens to be less than 0! Setting it to 0."); |
84090f85 |
324 | firstRun=0; |
325 | } |
326 | |
327 | Int_t lastRun = -1; |
328 | if(validityInfinite) { |
329 | lastRun = AliCDBRunRange::Infinity(); |
330 | } else { |
331 | lastRun = GetCurrentRun(); |
332 | } |
333 | |
2bb7b766 |
334 | AliCDBId id(path, firstRun, lastRun, -1, -1); |
335 | |
336 | if(! dynamic_cast<TObjString*> (metaData->GetProperty("RunUsed(TObjString)"))){ |
337 | TObjString runUsed = Form("%d", GetCurrentRun()); |
9e080f92 |
338 | metaData->SetProperty("RunUsed(TObjString)", runUsed.Clone()); |
2bb7b766 |
339 | } |
84090f85 |
340 | |
341 | UInt_t result = 0; |
342 | |
85a80aa9 |
343 | if (!(AliCDBManager::Instance()->GetStorage(mainUri))) { |
2bb7b766 |
344 | AliError(Form("WriteToCDB - Cannot activate main %s storage", cdbType)); |
84090f85 |
345 | } else { |
85a80aa9 |
346 | result = (UInt_t) AliCDBManager::Instance()->GetStorage(mainUri) |
84090f85 |
347 | ->Put(object, id, metaData); |
348 | } |
349 | |
350 | if(!result) { |
351 | |
352 | Log(fCurrentDetector, |
2bb7b766 |
353 | Form("WriteToCDB - Problem with main %s storage. Putting <%s> into backup storage", |
354 | cdbType, path.GetPath().Data())); |
355 | |
356 | // Set Grid version to current run number, to ease retrieval later |
357 | id.SetVersion(GetCurrentRun()); |
84090f85 |
358 | |
85a80aa9 |
359 | result = AliCDBManager::Instance()->GetStorage(localUri) |
84090f85 |
360 | ->Put(object, id, metaData); |
361 | |
362 | if(result) { |
363 | result = 2; |
85a80aa9 |
364 | fGridError = kTRUE; |
84090f85 |
365 | }else{ |
2bb7b766 |
366 | Log(fCurrentDetector, "WriteToCDB - Can't store data!"); |
b948db8d |
367 | } |
368 | } |
2bb7b766 |
369 | |
b948db8d |
370 | return result; |
371 | |
73abe331 |
372 | } |
373 | |
b948db8d |
374 | //______________________________________________________________________________________________ |
5164a766 |
375 | AliShuttleStatus* AliShuttle::ReadShuttleStatus() |
376 | { |
2bb7b766 |
377 | // Reads the AliShuttleStatus from the CDB |
5164a766 |
378 | |
2bb7b766 |
379 | if (fStatusEntry){ |
380 | delete fStatusEntry; |
381 | fStatusEntry = 0; |
382 | } |
5164a766 |
383 | |
2bb7b766 |
384 | fStatusEntry = AliCDBManager::Instance()->GetStorage(AliShuttle::GetLocalCDB()) |
385 | ->Get(Form("/SHUTTLE/STATUS/%s", fCurrentDetector.Data()), GetCurrentRun()); |
5164a766 |
386 | |
2bb7b766 |
387 | if (!fStatusEntry) return 0; |
388 | fStatusEntry->SetOwner(1); |
5164a766 |
389 | |
2bb7b766 |
390 | AliShuttleStatus* status = dynamic_cast<AliShuttleStatus*> (fStatusEntry->GetObject()); |
391 | if (!status) { |
392 | AliError("Invalid object stored to CDB!"); |
393 | return 0; |
394 | } |
5164a766 |
395 | |
2bb7b766 |
396 | return status; |
5164a766 |
397 | } |
398 | |
399 | //______________________________________________________________________________________________ |
7bfb2090 |
400 | Bool_t AliShuttle::WriteShuttleStatus(AliShuttleStatus* status) |
5164a766 |
401 | { |
2bb7b766 |
402 | // writes the status for one subdetector |
403 | |
404 | if (fStatusEntry){ |
405 | delete fStatusEntry; |
406 | fStatusEntry = 0; |
407 | } |
5164a766 |
408 | |
2bb7b766 |
409 | Int_t run = GetCurrentRun(); |
5164a766 |
410 | |
2bb7b766 |
411 | AliCDBId id(AliCDBPath("SHUTTLE", "STATUS", fCurrentDetector), run, run); |
5164a766 |
412 | |
2bb7b766 |
413 | fStatusEntry = new AliCDBEntry(status, id, new AliCDBMetaData); |
414 | fStatusEntry->SetOwner(1); |
5164a766 |
415 | |
2bb7b766 |
416 | UInt_t result = AliCDBManager::Instance()->GetStorage(fgkLocalCDB)->Put(fStatusEntry); |
7bfb2090 |
417 | |
2bb7b766 |
418 | if (!result) { |
419 | AliError(Form("WriteShuttleStatus for %s, run %d failed", fCurrentDetector.Data(), run)); |
420 | return kFALSE; |
421 | } |
e7f62f16 |
422 | |
423 | SendMLInfo(); |
7bfb2090 |
424 | |
2bb7b766 |
425 | return kTRUE; |
5164a766 |
426 | } |
427 | |
428 | //______________________________________________________________________________________________ |
429 | void AliShuttle::UpdateShuttleStatus(AliShuttleStatus::Status newStatus, Bool_t increaseCount) |
430 | { |
431 | // changes the AliShuttleStatus for the given detector and run to the given status |
432 | |
2bb7b766 |
433 | if (!fStatusEntry){ |
434 | AliError("UNEXPECTED: fStatusEntry empty"); |
435 | return; |
436 | } |
5164a766 |
437 | |
2bb7b766 |
438 | AliShuttleStatus* status = dynamic_cast<AliShuttleStatus*> (fStatusEntry->GetObject()); |
5164a766 |
439 | |
2bb7b766 |
440 | if (!status){ |
441 | AliError("UNEXPECTED: status could not be read from current CDB entry"); |
442 | return; |
443 | } |
5164a766 |
444 | |
2c15234c |
445 | TString actionStr = Form("UpdateShuttleStatus - %s: Changing state from %s to %s", |
eba76848 |
446 | fCurrentDetector.Data(), |
36c99a6a |
447 | status->GetStatusName(), |
eba76848 |
448 | status->GetStatusName(newStatus)); |
cb343cfd |
449 | Log("SHUTTLE", actionStr); |
450 | SetLastAction(actionStr); |
5164a766 |
451 | |
2bb7b766 |
452 | status->SetStatus(newStatus); |
453 | if (increaseCount) status->IncreaseCount(); |
5164a766 |
454 | |
2bb7b766 |
455 | AliCDBManager::Instance()->GetStorage(fgkLocalCDB)->Put(fStatusEntry); |
e7f62f16 |
456 | |
457 | SendMLInfo(); |
5164a766 |
458 | } |
e7f62f16 |
459 | |
460 | //______________________________________________________________________________________________ |
461 | void AliShuttle::SendMLInfo() |
462 | { |
463 | // |
464 | // sends ML information about the current status of the current detector being processed |
465 | // |
466 | |
467 | AliShuttleStatus* status = dynamic_cast<AliShuttleStatus*> (fStatusEntry->GetObject()); |
468 | |
469 | if (!status){ |
470 | AliError("UNEXPECTED: status could not be read from current CDB entry"); |
471 | return; |
472 | } |
473 | |
474 | TMonaLisaText mlStatus(Form("%s_status", fCurrentDetector.Data()), status->GetStatusName()); |
475 | TMonaLisaValue mlRetryCount(Form("%s_count", fCurrentDetector.Data()), status->GetCount()); |
476 | |
477 | TList mlList; |
478 | mlList.Add(&mlStatus); |
479 | mlList.Add(&mlRetryCount); |
480 | |
481 | fMonaLisa->SendParameters(&mlList); |
482 | } |
483 | |
5164a766 |
484 | //______________________________________________________________________________________________ |
485 | Bool_t AliShuttle::ContinueProcessing() |
486 | { |
2bb7b766 |
487 | // this function reads the AliShuttleStatus information from CDB and |
488 | // checks if the processing should be continued |
489 | // if yes it returns kTRUE and updates the AliShuttleStatus with nextStatus |
490 | |
57c1a579 |
491 | if (!fConfig->HostProcessDetector(fCurrentDetector)) return kFALSE; |
492 | |
493 | AliPreprocessor* aPreprocessor = |
494 | dynamic_cast<AliPreprocessor*> (fPreprocessorMap.GetValue(fCurrentDetector)); |
495 | if (!aPreprocessor) |
496 | { |
497 | AliInfo(Form("%s: no preprocessor registered", fCurrentDetector.Data())); |
498 | return kFALSE; |
499 | } |
500 | |
2bb7b766 |
501 | AliShuttleLogbookEntry::Status entryStatus = |
eba76848 |
502 | fLogbookEntry->GetDetectorStatus(fCurrentDetector); |
2bb7b766 |
503 | |
504 | if(entryStatus != AliShuttleLogbookEntry::kUnprocessed) { |
9e080f92 |
505 | AliInfo(Form("ContinueProcessing - %s is %s", |
2bb7b766 |
506 | fCurrentDetector.Data(), |
507 | fLogbookEntry->GetDetectorStatusName(entryStatus))); |
508 | return kFALSE; |
509 | } |
510 | |
511 | // if we get here, according to Shuttle logbook subdetector is in UNPROCESSED state |
be48e3ea |
512 | |
513 | // check if current run is first unprocessed run for current detector |
514 | if (fConfig->StrictRunOrder(fCurrentDetector) && |
515 | !fFirstUnprocessed[GetDetPos(fCurrentDetector)]) |
516 | { |
517 | Log("SHUTTLE", Form("ContinueProcessing - %s requires strict run ordering but this is not the first unprocessed run!")); |
518 | return kFALSE; |
519 | } |
520 | |
2bb7b766 |
521 | AliShuttleStatus* status = ReadShuttleStatus(); |
522 | if (!status) { |
523 | // first time |
524 | Log("SHUTTLE", Form("ContinueProcessing - %s: Processing first time", |
525 | fCurrentDetector.Data())); |
526 | status = new AliShuttleStatus(AliShuttleStatus::kStarted); |
527 | return WriteShuttleStatus(status); |
528 | } |
529 | |
530 | // The following two cases shouldn't happen if Shuttle Logbook was correctly updated. |
531 | // If it happens it may mean Logbook updating failed... let's do it now! |
532 | if (status->GetStatus() == AliShuttleStatus::kDone || |
533 | status->GetStatus() == AliShuttleStatus::kFailed){ |
534 | Log("SHUTTLE", Form("ContinueProcessing - %s is already %s. Updating Shuttle Logbook", |
535 | fCurrentDetector.Data(), |
536 | status->GetStatusName(status->GetStatus()))); |
537 | UpdateShuttleLogbook(fCurrentDetector.Data(), |
538 | status->GetStatusName(status->GetStatus())); |
539 | return kFALSE; |
540 | } |
541 | |
542 | if (status->GetStatus() == AliShuttleStatus::kStoreFailed) { |
543 | Log("SHUTTLE", |
544 | Form("ContinueProcessing - %s: Grid storage of one or more objects failed. Trying again now", |
545 | fCurrentDetector.Data())); |
546 | if(TryToStoreAgain()){ |
547 | Log(fCurrentDetector.Data(), "ContinueProcessing - All objects successfully stored into OCDB"); |
548 | UpdateShuttleStatus(AliShuttleStatus::kDone); |
549 | UpdateShuttleLogbook(fCurrentDetector.Data(), "DONE"); |
550 | } else { |
551 | Log("SHUTTLE", |
552 | Form("ContinueProcessing - %s: Grid storage failed again", |
553 | fCurrentDetector.Data())); |
e7f62f16 |
554 | // trigger ML information manually because we do not had a status change |
555 | SendMLInfo(); |
2bb7b766 |
556 | } |
557 | return kFALSE; |
558 | } |
559 | |
560 | // if we get here, there is a restart |
57c1a579 |
561 | Bool_t cont = kFALSE; |
2bb7b766 |
562 | |
563 | // abort conditions |
cb343cfd |
564 | if (status->GetCount() >= fConfig->GetMaxRetries()) { |
57c1a579 |
565 | Log("SHUTTLE", Form("ContinueProcessing - %s failed %d times in status %s - " |
566 | "Updating Shuttle Logbook", fCurrentDetector.Data(), |
2bb7b766 |
567 | status->GetCount(), status->GetStatusName())); |
568 | UpdateShuttleLogbook(fCurrentDetector.Data(), "FAILED"); |
e7f62f16 |
569 | UpdateShuttleStatus(AliShuttleStatus::kFailed); |
57c1a579 |
570 | } else { |
571 | Log("SHUTTLE", Form("ContinueProcessing - %s: restarting. " |
572 | "Aborted before with %s. Retry number %d.", fCurrentDetector.Data(), |
573 | status->GetStatusName(), status->GetCount())); |
574 | UpdateShuttleStatus(AliShuttleStatus::kStarted, kTRUE); |
575 | cont = kTRUE; |
2bb7b766 |
576 | } |
577 | |
57c1a579 |
578 | // Send mail to detector expert! |
579 | AliInfo(Form("Sending mail to %s expert...", fCurrentDetector.Data())); |
580 | if (!SendMail()) |
581 | Log("SHUTTLE", Form("ContinueProcessing - Could not send mail to %s expert", |
582 | fCurrentDetector.Data())); |
2bb7b766 |
583 | |
57c1a579 |
584 | return cont; |
5164a766 |
585 | } |
586 | |
587 | //______________________________________________________________________________________________ |
2bb7b766 |
588 | Bool_t AliShuttle::Process(AliShuttleLogbookEntry* entry) |
58bc3020 |
589 | { |
73abe331 |
590 | // |
b948db8d |
591 | // Makes data retrieval for all detectors in the configuration. |
2bb7b766 |
592 | // entry: Shuttle logbook entry, contains run paramenters and status of detectors |
593 | // (Unprocessed, Inactive, Failed or Done). |
d477ad88 |
594 | // Returns kFALSE in case of error occured and kTRUE otherwise |
73abe331 |
595 | // |
596 | |
2bb7b766 |
597 | if(!entry) return kFALSE; |
598 | |
599 | fLogbookEntry = entry; |
600 | |
e7f62f16 |
601 | if (fLogbookEntry->IsDone()) |
602 | { |
2bb7b766 |
603 | Log("SHUTTLE","Process - Shuttle is already DONE. Updating logbook"); |
604 | UpdateShuttleLogbook("shuttle_done"); |
605 | fLogbookEntry = 0; |
606 | return kTRUE; |
607 | } |
608 | |
e7f62f16 |
609 | // create ML instance that monitors this run |
610 | fMonaLisa = new TMonaLisaWriter(Form("%d", GetCurrentRun()), "SHUTTLE", "aliendb1.cern.ch"); |
611 | // disable monitoring of other parameters that come e.g. from TFile |
612 | gMonitoringWriter = 0; |
2bb7b766 |
613 | |
614 | AliInfo(Form("\n\n \t\t\t^*^*^*^*^*^*^*^*^*^*^*^* run %d: START ^*^*^*^*^*^*^*^*^*^*^*^* \n", |
615 | GetCurrentRun())); |
616 | |
e7f62f16 |
617 | |
618 | // Send the information to ML |
619 | TMonaLisaText mlStatus("SHUTTLE_status", "Processing"); |
620 | |
621 | TList mlList; |
622 | mlList.Add(&mlStatus); |
623 | |
624 | fMonaLisa->SendParameters(&mlList); |
625 | |
eba76848 |
626 | fLogbookEntry->Print("all"); |
57f50b3c |
627 | |
628 | // Initialization |
d477ad88 |
629 | Bool_t hasError = kFALSE; |
2c15234c |
630 | for(Int_t iSys=0;iSys<3;iSys++) fFXSCalled[iSys]=kFALSE; |
5164a766 |
631 | |
2bb7b766 |
632 | AliCDBStorage *mainCDBSto = AliCDBManager::Instance()->GetStorage(fgkMainCDB); |
633 | if(mainCDBSto) mainCDBSto->QueryCDB(GetCurrentRun()); |
634 | AliCDBStorage *mainRefSto = AliCDBManager::Instance()->GetStorage(fgkMainRefStorage); |
635 | if(mainRefSto) mainRefSto->QueryCDB(GetCurrentRun()); |
d477ad88 |
636 | |
57f50b3c |
637 | // Loop on detectors in the configuration |
b948db8d |
638 | TIter iter(fConfig->GetDetectors()); |
2bb7b766 |
639 | TObjString* aDetector = 0; |
b948db8d |
640 | |
be48e3ea |
641 | while ((aDetector = (TObjString*) iter.Next())) |
642 | { |
7bfb2090 |
643 | fCurrentDetector = aDetector->String(); |
5164a766 |
644 | |
9e080f92 |
645 | if (ContinueProcessing() == kFALSE) continue; |
646 | |
2bb7b766 |
647 | AliInfo(Form("\n\n \t\t\t****** run %d - %s: START ******", |
648 | GetCurrentRun(), aDetector->GetName())); |
649 | |
e7f62f16 |
650 | Log(fCurrentDetector.Data(), "Starting processing"); |
85a80aa9 |
651 | |
be48e3ea |
652 | Int_t pid = fork(); |
653 | |
654 | if (pid < 0) |
655 | { |
656 | Log("SHUTTLE", "ERROR: Forking failed"); |
657 | } |
658 | else if (pid > 0) |
659 | { |
660 | // parent |
661 | AliInfo(Form("In parent process of %d - %s: Starting monitoring", |
662 | GetCurrentRun(), aDetector->GetName())); |
663 | |
664 | Long_t begin = time(0); |
665 | |
666 | int status; // to be used with waitpid, on purpose an int (not Int_t)! |
667 | while (waitpid(pid, &status, WNOHANG) == 0) |
668 | { |
669 | Long_t expiredTime = time(0) - begin; |
670 | |
671 | if (expiredTime > fConfig->GetPPTimeOut()) |
672 | { |
673 | Log("SHUTTLE", Form("Process time out. Run time: %d seconds. Killing...", |
674 | expiredTime)); |
675 | |
676 | kill(pid, 9); |
677 | |
678 | hasError = kTRUE; |
679 | |
680 | gSystem->Sleep(1000); |
681 | } |
682 | else |
683 | { |
684 | if (expiredTime % 60 == 0) |
685 | Log("SHUTTLE", Form("Checked process. Run time: %d seconds.", |
686 | expiredTime)); |
687 | gSystem->Sleep(1000); |
688 | } |
689 | } |
690 | |
691 | AliInfo(Form("In parent process of %d - %s: Client has terminated.", |
692 | GetCurrentRun(), aDetector->GetName())); |
693 | |
694 | if (WIFEXITED(status)) |
695 | { |
696 | Int_t returnCode = WEXITSTATUS(status); |
697 | |
698 | Log("SHUTTLE", Form("The return code is %d", returnCode)); |
699 | |
700 | if (returnCode != 0) |
701 | hasError = kTRUE; |
702 | } |
703 | } |
704 | else if (pid == 0) |
705 | { |
706 | // client |
707 | AliInfo(Form("In client process of %d - %s", GetCurrentRun(), aDetector->GetName())); |
708 | |
709 | UInt_t result = ProcessCurrentDetector(); |
710 | |
711 | Int_t returnCode = 0; // will be set to 1 in case of an error |
712 | |
713 | if (!result) |
714 | { |
715 | returnCode = 1; |
716 | AliInfo(Form("\n \t\t\t****** run %d - %s: PREPROCESSOR ERROR ****** \n\n", |
717 | GetCurrentRun(), aDetector->GetName())); |
718 | } |
719 | else if (result == 2) |
720 | { |
721 | AliInfo(Form("\n \t\t\t****** run %d - %s: STORAGE ERROR ****** \n\n", |
722 | GetCurrentRun(), aDetector->GetName())); |
723 | } else |
724 | { |
725 | AliInfo(Form("\n \t\t\t****** run %d - %s: DONE ****** \n\n", |
726 | GetCurrentRun(), aDetector->GetName())); |
727 | } |
728 | |
729 | if (result > 0) |
730 | { |
2c15234c |
731 | // Process successful: Update time_processed field in FXS logbooks! |
732 | if (fFXSCalled[kDAQ]) |
be48e3ea |
733 | { |
734 | if (UpdateDAQTable() == kFALSE) |
735 | returnCode = 1; |
2c15234c |
736 | fFXSlist[kDAQ].Clear(); |
be48e3ea |
737 | } |
2c15234c |
738 | //if(fFXSCalled[kDCS]) { |
be48e3ea |
739 | // if (UpdateDCSTable(aDetector->GetName()) == kFALSE) |
740 | // returnCode = 1; |
2c15234c |
741 | // fFXSlist[kDCS].Clear(); |
be48e3ea |
742 | //} |
9e080f92 |
743 | if (fFXSCalled[kHLT]) |
744 | { |
745 | if (UpdateHLTTable() == kFALSE) |
746 | returnCode = 1; |
747 | fFXSlist[kHLT].Clear(); |
748 | } |
be48e3ea |
749 | } |
750 | |
751 | AliInfo(Form("Client process of %d - %s is exiting now with %d.", |
752 | GetCurrentRun(), aDetector->GetName(), returnCode)); |
753 | |
754 | // the client exits here |
755 | gSystem->Exit(returnCode); |
756 | |
757 | AliError("We should never get here!!!"); |
758 | } |
7bfb2090 |
759 | } |
5164a766 |
760 | |
2bb7b766 |
761 | AliInfo(Form("\n\n \t\t\t^*^*^*^*^*^*^*^*^*^*^*^* run %d: FINISH ^*^*^*^*^*^*^*^*^*^*^*^* \n", |
762 | GetCurrentRun())); |
763 | |
764 | //check if shuttle is done for this run, if so update logbook |
765 | TObjArray checkEntryArray; |
766 | checkEntryArray.SetOwner(1); |
9e080f92 |
767 | TString whereClause = Form("where run=%d", GetCurrentRun()); |
768 | if (!QueryShuttleLogbook(whereClause.Data(), checkEntryArray) || checkEntryArray.GetEntries() == 0) { |
769 | Log("SHUTTLE", Form("Process - Warning: Cannot check status of run %d on Shuttle logbook!", |
770 | GetCurrentRun())); |
771 | return hasError == kFALSE; |
772 | } |
b948db8d |
773 | |
9e080f92 |
774 | AliShuttleLogbookEntry* checkEntry = dynamic_cast<AliShuttleLogbookEntry*> |
775 | (checkEntryArray.At(0)); |
2bb7b766 |
776 | |
9e080f92 |
777 | if (checkEntry) |
778 | { |
779 | if (checkEntry->IsDone()) |
be48e3ea |
780 | { |
9e080f92 |
781 | Log("SHUTTLE","Process - Shuttle is DONE. Updating logbook"); |
782 | UpdateShuttleLogbook("shuttle_done"); |
783 | } |
784 | else |
785 | { |
786 | for (UInt_t iDet=0; iDet<NDetectors(); iDet++) |
be48e3ea |
787 | { |
9e080f92 |
788 | if (checkEntry->GetDetectorStatus(iDet) == AliShuttleLogbookEntry::kUnprocessed) |
be48e3ea |
789 | { |
9e080f92 |
790 | AliDebug(2, Form("Run %d: setting %s as \"not first time unprocessed\"", |
791 | checkEntry->GetRun(), GetDetName(iDet))); |
792 | fFirstUnprocessed[iDet] = kFALSE; |
be48e3ea |
793 | } |
794 | } |
2bb7b766 |
795 | } |
796 | } |
797 | |
e7f62f16 |
798 | // remove ML instance |
799 | delete fMonaLisa; |
800 | fMonaLisa = 0; |
801 | |
2bb7b766 |
802 | fLogbookEntry = 0; |
85a80aa9 |
803 | |
a7160fe9 |
804 | return hasError == kFALSE; |
73abe331 |
805 | } |
806 | |
b948db8d |
807 | //______________________________________________________________________________________________ |
2bb7b766 |
808 | UInt_t AliShuttle::ProcessCurrentDetector() |
73abe331 |
809 | { |
810 | // |
2bb7b766 |
811 | // Makes data retrieval just for a specific detector (fCurrentDetector). |
73abe331 |
812 | // Threre should be a configuration for this detector. |
73abe331 |
813 | |
2bb7b766 |
814 | AliInfo(Form("Retrieving values for %s, run %d", fCurrentDetector.Data(), GetCurrentRun())); |
73abe331 |
815 | |
7bfb2090 |
816 | UpdateShuttleStatus(AliShuttleStatus::kDCSStarted); |
73abe331 |
817 | |
2c15234c |
818 | TMap dcsMap; |
819 | dcsMap.SetOwner(1); |
73abe331 |
820 | |
85a80aa9 |
821 | Bool_t aDCSError = kFALSE; |
822 | fGridError = kFALSE; |
d477ad88 |
823 | |
2c15234c |
824 | // TODO Test only... I've added a flag that allows to |
825 | // exclude DCS archive DB query |
826 | if (!fgkProcessDCS) |
827 | { |
828 | AliInfo("Skipping DCS processing!"); |
829 | aDCSError = kFALSE; |
830 | } else { |
831 | TString host(fConfig->GetDCSHost(fCurrentDetector)); |
832 | Int_t port = fConfig->GetDCSPort(fCurrentDetector); |
833 | |
834 | // Retrieval of Aliases |
835 | TObjString* anAlias = 0; |
36c99a6a |
836 | Int_t iAlias = 1; |
837 | Int_t nTotAliases= ((TMap*)fConfig->GetDCSAliases(fCurrentDetector))->GetEntries(); |
2c15234c |
838 | TIter iterAliases(fConfig->GetDCSAliases(fCurrentDetector)); |
839 | while ((anAlias = (TObjString*) iterAliases.Next())) |
840 | { |
841 | TObjArray *valueSet = new TObjArray(); |
842 | valueSet->SetOwner(1); |
843 | |
36c99a6a |
844 | if (((iAlias-1) % 500) == 0 || iAlias == nTotAliases) |
845 | AliInfo(Form("Querying DCS archive: alias %s (%d of %d)", |
846 | anAlias->GetName(), iAlias++, nTotAliases)); |
2c15234c |
847 | aDCSError = (GetValueSet(host, port, anAlias->String(), valueSet, kAlias) == 0); |
848 | |
849 | if(!aDCSError) |
850 | { |
851 | dcsMap.Add(anAlias->Clone(), valueSet); |
852 | } else { |
853 | Log(fCurrentDetector, |
854 | Form("ProcessCurrentDetector - Error while retrieving alias %s", |
855 | anAlias->GetName())); |
856 | UpdateShuttleStatus(AliShuttleStatus::kDCSError); |
857 | dcsMap.DeleteAll(); |
858 | return 0; |
859 | } |
4f0ab988 |
860 | } |
2c15234c |
861 | |
862 | // Retrieval of Data Points |
863 | TObjString* aDP = 0; |
36c99a6a |
864 | Int_t iDP = 0; |
865 | Int_t nTotDPs= ((TMap*)fConfig->GetDCSDataPoints(fCurrentDetector))->GetEntries(); |
2c15234c |
866 | TIter iterDP(fConfig->GetDCSDataPoints(fCurrentDetector)); |
867 | while ((aDP = (TObjString*) iterDP.Next())) |
868 | { |
869 | TObjArray *valueSet = new TObjArray(); |
870 | valueSet->SetOwner(1); |
36c99a6a |
871 | if (((iDP-1) % 500) == 0 || iDP == nTotDPs) |
872 | AliInfo(Form("Querying DCS archive: DP %s (%d of %d)", |
873 | aDP->GetName(), iDP++, nTotDPs)); |
2c15234c |
874 | aDCSError = (GetValueSet(host, port, aDP->String(), valueSet, kDP) == 0); |
875 | |
876 | if(!aDCSError) |
877 | { |
878 | dcsMap.Add(aDP->Clone(), valueSet); |
879 | } else { |
880 | Log(fCurrentDetector, |
881 | Form("ProcessCurrentDetector - Error while retrieving data point %s", |
882 | aDP->GetName())); |
883 | UpdateShuttleStatus(AliShuttleStatus::kDCSError); |
884 | dcsMap.DeleteAll(); |
885 | return 0; |
886 | } |
73abe331 |
887 | } |
888 | } |
b948db8d |
889 | |
2bb7b766 |
890 | // DCS Archive DB processing successful. Call Preprocessor! |
85a80aa9 |
891 | UpdateShuttleStatus(AliShuttleStatus::kPPStarted); |
a7160fe9 |
892 | |
85a80aa9 |
893 | AliPreprocessor* aPreprocessor = |
5164a766 |
894 | dynamic_cast<AliPreprocessor*> (fPreprocessorMap.GetValue(fCurrentDetector)); |
b948db8d |
895 | |
2bb7b766 |
896 | aPreprocessor->Initialize(GetCurrentRun(), GetCurrentStartTime(), GetCurrentEndTime()); |
2c15234c |
897 | UInt_t aPPResult = aPreprocessor->Process(&dcsMap); |
2bb7b766 |
898 | |
899 | UInt_t returnValue = 0; |
85a80aa9 |
900 | if (aPPResult == 0) { // Preprocessor error |
cb343cfd |
901 | UpdateShuttleStatus(AliShuttleStatus::kPPError); |
2bb7b766 |
902 | returnValue = 0; |
85a80aa9 |
903 | } else if (fGridError == kFALSE) { // process and Grid storage ok! |
904 | UpdateShuttleStatus(AliShuttleStatus::kDone); |
2bb7b766 |
905 | UpdateShuttleLogbook(fCurrentDetector, "DONE"); |
906 | Log(fCurrentDetector.Data(), |
907 | "ProcessCurrentDetector - Preprocessor and Grid storage ended successfully"); |
908 | returnValue = 1; |
85a80aa9 |
909 | } else { // Grid storage error (process ok, but object put in local storage) |
910 | UpdateShuttleStatus(AliShuttleStatus::kStoreFailed); |
2bb7b766 |
911 | returnValue = 2; |
85a80aa9 |
912 | } |
b948db8d |
913 | |
2c15234c |
914 | dcsMap.DeleteAll(); |
b948db8d |
915 | |
2bb7b766 |
916 | return returnValue; |
917 | } |
918 | |
919 | //______________________________________________________________________________________________ |
920 | Bool_t AliShuttle::QueryShuttleLogbook(const char* whereClause, |
921 | TObjArray& entries) |
922 | { |
923 | // Query DAQ's Shuttle logbook and fills detector status object. |
924 | // Call QueryRunParameters to query DAQ logbook for run parameters. |
925 | |
fc5a4708 |
926 | entries.SetOwner(1); |
927 | |
2bb7b766 |
928 | // check connection, in case connect |
be48e3ea |
929 | if(!Connect(3)) return kFALSE; |
2bb7b766 |
930 | |
931 | TString sqlQuery; |
932 | sqlQuery = Form("select * from logbook_shuttle %s order by run", whereClause); |
933 | |
be48e3ea |
934 | TSQLResult* aResult = fServer[3]->Query(sqlQuery); |
2bb7b766 |
935 | if (!aResult) { |
936 | AliError(Form("Can't execute query <%s>!", sqlQuery.Data())); |
937 | return kFALSE; |
938 | } |
939 | |
fc5a4708 |
940 | AliDebug(2,Form("Query = %s", sqlQuery.Data())); |
941 | |
2bb7b766 |
942 | if(aResult->GetRowCount() == 0) { |
9e080f92 |
943 | // if(sqlQuery.EndsWith("where shuttle_done=0 order by run")){ |
944 | // Log("SHUTTLE", "QueryShuttleLogbook - All runs in Shuttle Logbook are already DONE"); |
945 | // delete aResult; |
946 | // return kTRUE; |
947 | // } else { |
948 | AliInfo("No entries in Shuttle Logbook match request"); |
2bb7b766 |
949 | delete aResult; |
950 | return kTRUE; |
9e080f92 |
951 | // } |
2bb7b766 |
952 | } |
953 | |
954 | // TODO Check field count! |
fc5a4708 |
955 | const UInt_t nCols = 22; |
2bb7b766 |
956 | if (aResult->GetFieldCount() != (Int_t) nCols) { |
957 | AliError("Invalid SQL result field number!"); |
958 | delete aResult; |
959 | return kFALSE; |
960 | } |
961 | |
2bb7b766 |
962 | TSQLRow* aRow; |
963 | while ((aRow = aResult->Next())) { |
964 | TString runString(aRow->GetField(0), aRow->GetFieldLength(0)); |
965 | Int_t run = runString.Atoi(); |
966 | |
eba76848 |
967 | AliShuttleLogbookEntry *entry = QueryRunParameters(run); |
968 | if (!entry) |
969 | continue; |
2bb7b766 |
970 | |
971 | // loop on detectors |
eba76848 |
972 | for(UInt_t ii = 0; ii < nCols; ii++) |
973 | entry->SetDetectorStatus(aResult->GetFieldName(ii), aRow->GetField(ii)); |
2bb7b766 |
974 | |
eba76848 |
975 | entries.AddLast(entry); |
2bb7b766 |
976 | delete aRow; |
977 | } |
978 | |
9e080f92 |
979 | // if(sqlQuery.EndsWith("where shuttle_done=0 order by run")) |
980 | // Log("SHUTTLE", Form("QueryShuttleLogbook - Found %d unprocessed runs in Shuttle Logbook", |
981 | // entries.GetEntriesFast())); |
2bb7b766 |
982 | delete aResult; |
983 | return kTRUE; |
984 | } |
985 | |
986 | //______________________________________________________________________________________________ |
eba76848 |
987 | AliShuttleLogbookEntry* AliShuttle::QueryRunParameters(Int_t run) |
2bb7b766 |
988 | { |
eba76848 |
989 | // |
990 | // Retrieve run parameters written in the DAQ logbook and sets them into AliShuttleLogbookEntry object |
991 | // |
2bb7b766 |
992 | |
993 | // check connection, in case connect |
be48e3ea |
994 | if (!Connect(3)) |
eba76848 |
995 | return 0; |
2bb7b766 |
996 | |
997 | TString sqlQuery; |
2c15234c |
998 | sqlQuery.Form("select * from %s where run=%d", fConfig->GetDAQlbTable(), run); |
2bb7b766 |
999 | |
be48e3ea |
1000 | TSQLResult* aResult = fServer[3]->Query(sqlQuery); |
2bb7b766 |
1001 | if (!aResult) { |
1002 | AliError(Form("Can't execute query <%s>!", sqlQuery.Data())); |
eba76848 |
1003 | return 0; |
2bb7b766 |
1004 | } |
1005 | |
eba76848 |
1006 | if (aResult->GetRowCount() == 0) { |
2bb7b766 |
1007 | Log("SHUTTLE", Form("QueryRunParameters - No entry in DAQ Logbook for run %d. Skipping", run)); |
1008 | delete aResult; |
eba76848 |
1009 | return 0; |
2bb7b766 |
1010 | } |
1011 | |
eba76848 |
1012 | if (aResult->GetRowCount() > 1) { |
2bb7b766 |
1013 | AliError(Form("More than one entry in DAQ Logbook for run %d. Skipping", run)); |
1014 | delete aResult; |
eba76848 |
1015 | return 0; |
2bb7b766 |
1016 | } |
1017 | |
eba76848 |
1018 | TSQLRow* aRow = aResult->Next(); |
1019 | if (!aRow) |
1020 | { |
1021 | AliError(Form("Could not retrieve row for run %d. Skipping", run)); |
1022 | delete aResult; |
1023 | return 0; |
1024 | } |
2bb7b766 |
1025 | |
eba76848 |
1026 | AliShuttleLogbookEntry* entry = new AliShuttleLogbookEntry(run); |
2bb7b766 |
1027 | |
eba76848 |
1028 | for (Int_t ii = 0; ii < aResult->GetFieldCount(); ii++) |
1029 | entry->SetRunParameter(aResult->GetFieldName(ii), aRow->GetField(ii)); |
2bb7b766 |
1030 | |
eba76848 |
1031 | UInt_t startTime = entry->GetStartTime(); |
1032 | UInt_t endTime = entry->GetEndTime(); |
1033 | |
1034 | if (!startTime || !endTime || startTime > endTime) { |
1035 | Log("SHUTTLE", |
1036 | Form("QueryRunParameters - Invalid parameters for Run %d: startTime = %d, endTime = %d", |
1037 | run, startTime, endTime)); |
1038 | delete entry; |
2bb7b766 |
1039 | delete aRow; |
eba76848 |
1040 | delete aResult; |
1041 | return 0; |
2bb7b766 |
1042 | } |
1043 | |
eba76848 |
1044 | delete aRow; |
2bb7b766 |
1045 | delete aResult; |
eba76848 |
1046 | |
1047 | return entry; |
2bb7b766 |
1048 | } |
1049 | |
1050 | //______________________________________________________________________________________________ |
1051 | Bool_t AliShuttle::TryToStoreAgain() |
1052 | { |
1053 | // Called in case the detector failed to store the object in Grid OCDB |
1054 | // It tries to store the object again, if it does not find more recent and overlapping objects |
1055 | // Calls underlying TryToStoreAgain(const char*) function twice, for OCDB and Reference storage. |
1056 | |
1057 | AliInfo("Trying to store OCDB data again..."); |
1058 | Bool_t resultCDB = TryToStoreAgain(fgkMainCDB); |
1059 | |
1060 | AliInfo("Trying to store reference data again..."); |
1061 | Bool_t resultRef = TryToStoreAgain(fgkMainRefStorage); |
1062 | |
1063 | return resultCDB && resultRef; |
1064 | } |
1065 | |
1066 | //______________________________________________________________________________________________ |
1067 | Bool_t AliShuttle::TryToStoreAgain(TString& gridURI) |
1068 | { |
1069 | // Called by TryToStoreAgain(), performs actual storage retry |
1070 | |
6ec0e06c |
1071 | TObjArray* gridIds=0; |
2bb7b766 |
1072 | |
1073 | Bool_t result = kTRUE; |
1074 | |
1075 | const char* type = 0; |
1076 | TString backupURI; |
1077 | if(gridURI == fgkMainCDB) { |
1078 | type = "OCDB"; |
1079 | backupURI = fgkLocalCDB; |
1080 | } else if(gridURI == fgkMainRefStorage) { |
1081 | type = "reference"; |
1082 | backupURI = fgkLocalRefStorage; |
1083 | } else { |
1084 | AliError(Form("Invalid storage URI: %s", gridURI.Data())); |
1085 | return kFALSE; |
1086 | } |
1087 | |
1088 | AliCDBManager* man = AliCDBManager::Instance(); |
1089 | |
1090 | AliCDBStorage *gridSto = man->GetStorage(gridURI); |
1091 | if(!gridSto) { |
1092 | Log(fCurrentDetector.Data(), |
1093 | Form("TryToStoreAgain - cannot activate main %s storage", type)); |
1094 | return kFALSE; |
1095 | } |
1096 | |
1097 | gridIds = gridSto->GetQueryCDBList(); |
1098 | |
1099 | // get objects previously stored in local CDB |
1100 | AliCDBStorage *backupSto = man->GetStorage(backupURI); |
eba76848 |
1101 | AliCDBPath aPath(GetOfflineDetName(fCurrentDetector.Data()),"*","*"); |
2bb7b766 |
1102 | // Local objects were stored with current run as Grid version! |
1103 | TList* localEntries = backupSto->GetAll(aPath.GetPath(), GetCurrentRun(), GetCurrentRun()); |
1104 | localEntries->SetOwner(1); |
1105 | |
1106 | // loop on local stored objects |
1107 | TIter localIter(localEntries); |
1108 | AliCDBEntry *aLocEntry = 0; |
1109 | while((aLocEntry = dynamic_cast<AliCDBEntry*> (localIter.Next()))){ |
1110 | aLocEntry->SetOwner(1); |
1111 | AliCDBId aLocId = aLocEntry->GetId(); |
1112 | aLocEntry->SetVersion(-1); |
1113 | aLocEntry->SetSubVersion(-1); |
1114 | |
1115 | // loop on Grid valid Id's |
1116 | Bool_t store = kTRUE; |
1117 | TIter gridIter(gridIds); |
1118 | AliCDBId* aGridId = 0; |
1119 | while((aGridId = dynamic_cast<AliCDBId*> (gridIter.Next()))){ |
be48e3ea |
1120 | // If local object is valid up to infinity we store it only if it is |
1121 | // the first unprocessed run! |
1122 | if (aLocId.GetLastRun() == AliCDBRunRange::Infinity()) |
1123 | { |
1124 | if (!fFirstUnprocessed[GetDetPos(fCurrentDetector)]) |
1125 | { |
1126 | Log(fCurrentDetector.Data(), |
2c15234c |
1127 | ("TryToStoreAgain - This object has validity infinite but " |
1128 | "there are previous unprocessed runs!")); |
be48e3ea |
1129 | continue; |
1130 | } else { |
1131 | break; |
1132 | } |
2bb7b766 |
1133 | } |
1134 | if(aGridId->GetPath() != aLocId.GetPath()) continue; |
1135 | // skip all objects valid up to infinity |
1136 | if(aGridId->GetLastRun() == AliCDBRunRange::Infinity()) continue; |
1137 | // if we get here, it means there's already some more recent object stored on Grid! |
1138 | store = kFALSE; |
1139 | break; |
1140 | } |
1141 | |
1142 | if(!store){ |
1143 | Log(fCurrentDetector.Data(), |
1144 | Form("TryToStoreAgain - A more recent object already exists in %s storage: <%s>", |
1145 | type, aGridId->ToString().Data())); |
1146 | // removing local filename... |
1147 | // TODO maybe it's better not to remove it, it was not copied to the Grid! |
1148 | TString filename; |
1149 | backupSto->IdToFilename(aLocId, filename); |
1150 | AliInfo(Form("Removing local file %s", filename.Data())); |
1151 | gSystem->Exec(Form("rm %s",filename.Data())); |
1152 | continue; |
1153 | } |
1154 | |
1155 | // If we get here, the file can be stored! |
1156 | Bool_t storeOk = gridSto->Put(aLocEntry); |
1157 | if(storeOk){ |
1158 | Log(fCurrentDetector.Data(), |
1159 | Form("TryToStoreAgain - Object <%s> successfully put into %s storage", |
1160 | aLocId.ToString().Data(), type)); |
1161 | |
1162 | // removing local filename... |
1163 | TString filename; |
1164 | backupSto->IdToFilename(aLocId, filename); |
1165 | AliInfo(Form("Removing local file %s", filename.Data())); |
1166 | gSystem->Exec(Form("rm %s", filename.Data())); |
1167 | continue; |
1168 | } else { |
1169 | Log(fCurrentDetector.Data(), |
1170 | Form("TryToStoreAgain - Grid %s storage of object <%s> failed again", |
1171 | type, aLocId.ToString().Data())); |
1172 | result = kFALSE; |
1173 | } |
1174 | } |
1175 | localEntries->Clear(); |
1176 | |
1177 | return result; |
73abe331 |
1178 | } |
1179 | |
b948db8d |
1180 | //______________________________________________________________________________________________ |
2c15234c |
1181 | Bool_t AliShuttle::GetValueSet(const char* host, Int_t port, const char* entry, |
1182 | TObjArray* valueSet, DCSType type) |
73abe331 |
1183 | { |
2c15234c |
1184 | // Retrieve all "entry" data points from the DCS server |
58bc3020 |
1185 | // host, port: TSocket connection parameters |
2c15234c |
1186 | // entry: name of the alias or data point |
2bb7b766 |
1187 | // valueSet: array of retrieved AliDCSValue's |
2c15234c |
1188 | // type: kAlias or kDP |
58bc3020 |
1189 | |
73abe331 |
1190 | AliDCSClient client(host, port, fTimeout, fRetries); |
2c15234c |
1191 | if (!client.IsConnected()) |
1192 | { |
b948db8d |
1193 | return kFALSE; |
73abe331 |
1194 | } |
1195 | |
2c15234c |
1196 | Int_t result=0; |
73abe331 |
1197 | |
2c15234c |
1198 | if (type == kAlias) |
1199 | { |
1200 | result = client.GetAliasValues(entry, |
1201 | GetCurrentStartTime(), GetCurrentEndTime(), valueSet); |
1202 | } else |
1203 | if (type == kDP) |
1204 | { |
1205 | result = client.GetDPValues(entry, |
1206 | GetCurrentStartTime(), GetCurrentEndTime(), valueSet); |
1207 | } |
1208 | |
1209 | if (result < 0) |
1210 | { |
2bb7b766 |
1211 | Log(fCurrentDetector.Data(), Form("GetValueSet - Can't get '%s'! Reason: %s", |
2c15234c |
1212 | entry, AliDCSClient::GetErrorString(result))); |
73abe331 |
1213 | |
2c15234c |
1214 | if (result == AliDCSClient::fgkServerError) |
1215 | { |
2bb7b766 |
1216 | Log(fCurrentDetector.Data(), Form("GetValueSet - Server error: %s", |
73abe331 |
1217 | client.GetServerError().Data())); |
1218 | } |
1219 | |
1220 | return kFALSE; |
1221 | } |
1222 | |
1223 | return kTRUE; |
1224 | } |
b948db8d |
1225 | |
1226 | //______________________________________________________________________________________________ |
57f50b3c |
1227 | const char* AliShuttle::GetFile(Int_t system, const char* detector, |
1228 | const char* id, const char* source) |
b948db8d |
1229 | { |
57f50b3c |
1230 | // Get calibration file from file exchange servers |
1231 | // calls specific getter according to system index (kDAQ, kDCS, kHLT) |
1232 | |
1233 | switch(system){ |
1234 | case kDAQ: |
1235 | return GetDAQFileName(detector, id, source); |
1236 | break; |
1237 | case kDCS: |
1238 | return GetDCSFileName(detector, id, source); |
1239 | break; |
1240 | case kHLT: |
1241 | return GetHLTFileName(detector, id, source); |
1242 | break; |
1243 | default: |
1244 | AliError(Form("No valid system index: %d",system)); |
1245 | } |
b948db8d |
1246 | |
b948db8d |
1247 | return 0; |
1248 | } |
1249 | |
b948db8d |
1250 | //______________________________________________________________________________________________ |
57f50b3c |
1251 | TList* AliShuttle::GetFileSources(Int_t system, const char* detector, const char* id) |
b948db8d |
1252 | { |
57f50b3c |
1253 | // Get sources producing the condition file Id from file exchange servers |
1254 | // calls specific getter according to system index (kDAQ, kDCS, kHLT) |
1255 | |
1256 | switch(system){ |
1257 | case kDAQ: |
1258 | return GetDAQFileSources(detector, id); |
1259 | break; |
1260 | case kDCS: |
1261 | return GetDCSFileSources(detector, id); |
1262 | break; |
1263 | case kHLT: |
1264 | return GetHLTFileSources(detector, id); |
1265 | break; |
1266 | default: |
1267 | AliError(Form("No valid system index: %d",system)); |
1268 | } |
1269 | |
1270 | return NULL; |
1271 | } |
1272 | |
1273 | //______________________________________________________________________________________________ |
2bb7b766 |
1274 | Bool_t AliShuttle::Connect(Int_t system) |
1275 | { |
2c15234c |
1276 | // Connect to MySQL Server of the system's FXS MySQL databases |
1277 | // DAQ Logbook, Shuttle Logbook and DAQ FXS db are on the same host |
57f50b3c |
1278 | |
1279 | // check connection: if already connected return |
1280 | if(fServer[system] && fServer[system]->IsConnected()) return kTRUE; |
1281 | |
2c15234c |
1282 | TString dbHost, dbUser, dbPass, dbName; |
be48e3ea |
1283 | |
2c15234c |
1284 | if (system < 3) // FXS db servers |
be48e3ea |
1285 | { |
fc5a4708 |
1286 | dbHost = Form("mysql://%s:%d", fConfig->GetFXSdbHost(system), fConfig->GetFXSdbPort(system)); |
2c15234c |
1287 | dbUser = fConfig->GetFXSdbUser(system); |
1288 | dbPass = fConfig->GetFXSdbPass(system); |
1289 | dbName = fConfig->GetFXSdbName(system); |
be48e3ea |
1290 | } else { // Run & Shuttle logbook servers |
1291 | // TODO Will the Shuttle logbook server be the same as the Run logbook server ??? |
fc5a4708 |
1292 | dbHost = Form("mysql://%s:%d", fConfig->GetDAQlbHost(), fConfig->GetDAQlbPort()); |
2c15234c |
1293 | dbUser = fConfig->GetDAQlbUser(); |
1294 | dbPass = fConfig->GetDAQlbPass(); |
1295 | dbName = fConfig->GetDAQlbDB(); |
be48e3ea |
1296 | } |
57f50b3c |
1297 | |
2c15234c |
1298 | fServer[system] = TSQLServer::Connect(dbHost.Data(), dbUser.Data(), dbPass.Data()); |
57f50b3c |
1299 | if (!fServer[system] || !fServer[system]->IsConnected()) { |
be48e3ea |
1300 | if(system < 3) |
1301 | { |
2c15234c |
1302 | AliError(Form("Can't establish connection to FXS database for %s", |
eba76848 |
1303 | AliShuttleInterface::GetSystemName(system))); |
be48e3ea |
1304 | } else { |
1305 | AliError("Can't establish connection to Run logbook."); |
1306 | } |
2bb7b766 |
1307 | if(fServer[system]) delete fServer[system]; |
57f50b3c |
1308 | return kFALSE; |
1309 | } |
1310 | |
1311 | // Get tables |
1312 | // TODO in the configuration should the table name be there too? |
2bb7b766 |
1313 | TSQLResult* aResult=0; |
57f50b3c |
1314 | switch(system){ |
1315 | case kDAQ: |
2c15234c |
1316 | aResult = fServer[kDAQ]->GetTables(dbName.Data()); |
57f50b3c |
1317 | break; |
1318 | case kDCS: |
2c15234c |
1319 | //aResult = fServer[kDCS]->GetTables(dbName.Data()); |
57f50b3c |
1320 | break; |
1321 | case kHLT: |
9e080f92 |
1322 | aResult = fServer[kHLT]->GetTables(dbName.Data()); |
57f50b3c |
1323 | break; |
1324 | default: |
2c15234c |
1325 | aResult = fServer[3]->GetTables(dbName.Data()); |
57f50b3c |
1326 | break; |
1327 | } |
1328 | |
2bb7b766 |
1329 | delete aResult; |
57f50b3c |
1330 | return kTRUE; |
1331 | } |
1332 | |
1333 | //______________________________________________________________________________________________ |
2bb7b766 |
1334 | const char* AliShuttle::GetDAQFileName(const char* detector, const char* id, const char* source) |
1335 | { |
2c15234c |
1336 | // Retrieves a file from the DAQ FXS. |
1337 | // First queris the DAQ FXS database for the DAQ file name, using the run, detector, id and source info |
57f50b3c |
1338 | // then calls RetrieveDAQFile(DAQfilename) for actual copy to local disk |
2bb7b766 |
1339 | // run: current run being processed (given by Logbook entry fLogbookEntry) |
eba76848 |
1340 | // detector: the Preprocessor name |
57f50b3c |
1341 | // id: provided as a parameter by the Preprocessor |
1342 | // source: provided by the Preprocessor through GetFileSources function |
1343 | |
1344 | // check connection, in case connect |
eba76848 |
1345 | if (!Connect(kDAQ)) |
1346 | { |
2c15234c |
1347 | Log(detector, "GetDAQFileName - Couldn't connect to DAQ FXS database"); |
57f50b3c |
1348 | return 0; |
1349 | } |
1350 | |
1351 | // Query preparation |
2c15234c |
1352 | TString sqlQueryStart = Form("select filePath from %s where", fConfig->GetFXSdbTable(kDAQ)); |
57f50b3c |
1353 | TString whereClause = Form("run=%d and detector=\"%s\" and fileId=\"%s\" and DAQsource=\"%s\"", |
eba76848 |
1354 | GetCurrentRun(), detector, id, source); |
57f50b3c |
1355 | TString sqlQuery = Form("%s %s", sqlQueryStart.Data(), whereClause.Data()); |
1356 | |
84090f85 |
1357 | AliDebug(2, Form("SQL query: \n%s",sqlQuery.Data())); |
57f50b3c |
1358 | |
1359 | // Query execution |
2bb7b766 |
1360 | TSQLResult* aResult = 0; |
1361 | aResult = dynamic_cast<TSQLResult*> (fServer[kDAQ]->Query(sqlQuery)); |
57f50b3c |
1362 | if (!aResult) { |
2bb7b766 |
1363 | Log(detector, Form("GetDAQFileName - Can't execute SQL query for: id = %s, source = %s", |
1364 | id, source)); |
57f50b3c |
1365 | return 0; |
1366 | } |
1367 | |
eba76848 |
1368 | if(aResult->GetRowCount() == 0) |
1369 | { |
57f50b3c |
1370 | Log(detector, |
2c15234c |
1371 | Form("GetDAQFileName - No entry in FXS table for: id = %s, source = %s", |
2bb7b766 |
1372 | id, source)); |
57f50b3c |
1373 | delete aResult; |
1374 | return 0; |
1375 | } |
1376 | |
eba76848 |
1377 | if (aResult->GetRowCount() > 1) { |
57f50b3c |
1378 | Log(detector, |
2c15234c |
1379 | Form("GetDAQFileName - More than one entry in FXS table for: id = %s, source = %s", |
2bb7b766 |
1380 | id, source)); |
57f50b3c |
1381 | delete aResult; |
1382 | return 0; |
1383 | } |
1384 | |
2bb7b766 |
1385 | TSQLRow* aRow = dynamic_cast<TSQLRow*> (aResult->Next()); |
57f50b3c |
1386 | |
eba76848 |
1387 | if (!aRow){ |
2bb7b766 |
1388 | Log(detector, Form("GetDAQFileName - Empty set result from query: id = %s, source = %s", |
1389 | id, source)); |
57f50b3c |
1390 | delete aResult; |
1391 | return 0; |
1392 | } |
1393 | |
1394 | TString filePath(aRow->GetField(0), aRow->GetFieldLength(0)); |
1395 | |
1396 | delete aResult; |
2bb7b766 |
1397 | delete aRow; |
57f50b3c |
1398 | |
84090f85 |
1399 | AliDebug(2, Form("filePath = %s",filePath.Data())); |
57f50b3c |
1400 | |
1401 | // retrieved file is renamed to make it unique |
9e080f92 |
1402 | TString localFileName = Form("DAQ_%s_%d_%s_%s.shuttle", |
2bb7b766 |
1403 | detector, GetCurrentRun(), id, source); |
57f50b3c |
1404 | |
2c15234c |
1405 | // file retrieval from DAQ FXS |
57f50b3c |
1406 | Bool_t result = RetrieveDAQFile(filePath.Data(), localFileName.Data()); |
1407 | if(!result) { |
2c15234c |
1408 | Log(detector, Form("GetDAQFileName - Copy of file %s from DAQ FXS failed", filePath.Data())); |
57f50b3c |
1409 | return 0; |
1410 | } else { |
2c15234c |
1411 | AliInfo(Form("File %s copied from DAQ FXS into %s/%s", |
36c99a6a |
1412 | filePath.Data(), GetShuttleTempDir(), localFileName.Data())); |
57f50b3c |
1413 | } |
1414 | |
2c15234c |
1415 | fFXSCalled[kDAQ]=kTRUE; |
9e080f92 |
1416 | TObjString *fileParams = new TObjString(Form("%s#!?!#%s", id, source)); |
2c15234c |
1417 | fFXSlist[kDAQ].Add(fileParams); |
57f50b3c |
1418 | |
9e080f92 |
1419 | static TString fullLocalFileName; |
36c99a6a |
1420 | fullLocalFileName = TString::Format("%s/%s", GetShuttleTempDir(), localFileName.Data()); |
1421 | |
9e080f92 |
1422 | AliInfo(Form("fullLocalFileName = %s", fullLocalFileName.Data())); |
1423 | |
1424 | return fullLocalFileName.Data(); |
57f50b3c |
1425 | |
1426 | } |
1427 | |
1428 | //______________________________________________________________________________________________ |
2bb7b766 |
1429 | Bool_t AliShuttle::RetrieveDAQFile(const char* daqFileName, const char* localFileName) |
1430 | { |
9e080f92 |
1431 | // Copies file from DAQ FXS to local Shuttle machine |
57f50b3c |
1432 | |
1433 | // check temp directory: trying to cd to temp; if it does not exist, create it |
9e080f92 |
1434 | AliDebug(2, Form("Copy file %s from DAQ FXS into %s/%s", |
36c99a6a |
1435 | daqFileName, GetShuttleTempDir(), localFileName)); |
57f50b3c |
1436 | |
36c99a6a |
1437 | void* dir = gSystem->OpenDirectory(GetShuttleTempDir()); |
57f50b3c |
1438 | if (dir == NULL) { |
36c99a6a |
1439 | if (gSystem->mkdir(GetShuttleTempDir(), kTRUE)) { |
1440 | AliError(Form("Can't open directory <%s>", GetShuttleTempDir())); |
57f50b3c |
1441 | return kFALSE; |
1442 | } |
1443 | |
1444 | } else { |
1445 | gSystem->FreeDirectory(dir); |
1446 | } |
1447 | |
9e080f92 |
1448 | TString baseDAQFXSFolder = "FES"; |
fc5a4708 |
1449 | TString command = Form("scp -oPort=%d -2 %s@%s:%s/%s %s/%s", |
1450 | fConfig->GetFXSPort(kDAQ), |
2c15234c |
1451 | fConfig->GetFXSUser(kDAQ), |
1452 | fConfig->GetFXSHost(kDAQ), |
1453 | baseDAQFXSFolder.Data(), |
57f50b3c |
1454 | daqFileName, |
36c99a6a |
1455 | GetShuttleTempDir(), |
57f50b3c |
1456 | localFileName); |
1457 | |
84090f85 |
1458 | AliDebug(2, Form("%s",command.Data())); |
57f50b3c |
1459 | |
1460 | UInt_t nRetries = 0; |
1461 | UInt_t maxRetries = 3; |
1462 | |
1463 | // copy!! if successful TSystem::Exec returns 0 |
1464 | while(nRetries++ < maxRetries) { |
84090f85 |
1465 | AliDebug(2, Form("Trying to copy file. Retry # %d", nRetries)); |
57f50b3c |
1466 | if(gSystem->Exec(command.Data()) == 0) return kTRUE; |
1467 | } |
1468 | |
1469 | return kFALSE; |
1470 | |
1471 | } |
1472 | |
1473 | //______________________________________________________________________________________________ |
2bb7b766 |
1474 | TList* AliShuttle::GetDAQFileSources(const char* detector, const char* id) |
1475 | { |
9e080f92 |
1476 | // Retrieves list of DAQ sources of file Id |
57f50b3c |
1477 | |
1478 | // check connection, in case connect |
1479 | if(!Connect(kDAQ)){ |
2c15234c |
1480 | Log(detector, "GetDAQFileSources - Couldn't connect to DAQ FXS database"); |
57f50b3c |
1481 | return 0; |
1482 | } |
1483 | |
1484 | // Query preparation |
2c15234c |
1485 | TString sqlQueryStart = Form("select DAQsource from %s where", fConfig->GetFXSdbTable(kDAQ)); |
57f50b3c |
1486 | TString whereClause = Form("run=%d and detector=\"%s\" and fileId=\"%s\"", |
eba76848 |
1487 | GetCurrentRun(), detector, id); |
57f50b3c |
1488 | TString sqlQuery = Form("%s %s", sqlQueryStart.Data(), whereClause.Data()); |
1489 | |
84090f85 |
1490 | AliDebug(2, Form("SQL query: \n%s",sqlQuery.Data())); |
57f50b3c |
1491 | |
1492 | // Query execution |
1493 | TSQLResult* aResult; |
1494 | aResult = fServer[kDAQ]->Query(sqlQuery); |
1495 | if (!aResult) { |
2bb7b766 |
1496 | Log(detector, Form("GetDAQFileSources - Can't execute SQL query for id: %s", id)); |
57f50b3c |
1497 | return 0; |
1498 | } |
1499 | |
1500 | if (aResult->GetRowCount() == 0) { |
1501 | Log(detector, |
2c15234c |
1502 | Form("GetDAQFileSources - No entry in FXS table for id: %s", id)); |
57f50b3c |
1503 | delete aResult; |
1504 | return 0; |
1505 | } |
1506 | |
1507 | TSQLRow* aRow; |
1508 | TList *list = new TList(); |
1509 | list->SetOwner(1); |
1510 | |
1511 | while((aRow = aResult->Next())){ |
1512 | |
1513 | TString daqSource(aRow->GetField(0), aRow->GetFieldLength(0)); |
84090f85 |
1514 | AliDebug(2, Form("daqSource = %s", daqSource.Data())); |
57f50b3c |
1515 | list->Add(new TObjString(daqSource)); |
2bb7b766 |
1516 | delete aRow; |
57f50b3c |
1517 | } |
1518 | delete aResult; |
1519 | |
1520 | return list; |
1521 | |
1522 | } |
1523 | |
1524 | //______________________________________________________________________________________________ |
2bb7b766 |
1525 | const char* AliShuttle::GetDCSFileName(const char* /*detector*/, const char* /*id*/, const char* /*source*/){ |
2c15234c |
1526 | // Retrieves a file from the DCS FXS. |
2bb7b766 |
1527 | |
1528 | return "You're in DCS"; |
1529 | |
1530 | } |
1531 | |
1532 | //______________________________________________________________________________________________ |
1533 | TList* AliShuttle::GetDCSFileSources(const char* /*detector*/, const char* /*id*/){ |
2c15234c |
1534 | // Retrieves file sources from the DCS FXS. |
2bb7b766 |
1535 | |
1536 | return NULL; |
1537 | |
1538 | } |
1539 | |
1540 | //______________________________________________________________________________________________ |
9e080f92 |
1541 | const char* AliShuttle::GetHLTFileName(const char* detector, const char* id, const char* source){ |
2c15234c |
1542 | // Retrieves a file from the HLT FXS. |
9e080f92 |
1543 | // First queris the HLT FXS database for the HLT file name, using the run, detector, id and source info |
1544 | // then calls RetrieveDAQFile(DAQfilename) for actual copy to local disk |
1545 | // run: current run being processed (given by Logbook entry fLogbookEntry) |
1546 | // detector: the Preprocessor name |
1547 | // id: provided as a parameter by the Preprocessor |
1548 | // source: provided by the Preprocessor through GetFileSources function |
1549 | |
1550 | // check connection, in case connect |
1551 | if (!Connect(kHLT)) |
1552 | { |
1553 | Log(detector, "GetHLTFileName - Couldn't connect to HLT FXS database"); |
1554 | return 0; |
1555 | } |
1556 | |
1557 | // Query preparation |
1558 | TString sqlQueryStart = Form("select filePath,fileSize,fileChecksum from %s where", |
1559 | fConfig->GetFXSdbTable(kHLT)); |
1560 | TString whereClause = Form("run=%d and detector=\"%s\" and fileId=\"%s\" and DDLnumbers=\"%s\"", |
1561 | GetCurrentRun(), detector, id, source); |
1562 | TString sqlQuery = Form("%s %s", sqlQueryStart.Data(), whereClause.Data()); |
1563 | |
1564 | AliDebug(2, Form("SQL query: \n%s",sqlQuery.Data())); |
1565 | |
1566 | // Query execution |
1567 | TSQLResult* aResult = 0; |
1568 | aResult = dynamic_cast<TSQLResult*> (fServer[kHLT]->Query(sqlQuery)); |
1569 | if (!aResult) { |
1570 | Log(detector, Form("GetHLTFileName - Can't execute SQL query for: id = %s, source = %s", |
1571 | id, source)); |
1572 | return 0; |
1573 | } |
1574 | |
1575 | if(aResult->GetRowCount() == 0) |
1576 | { |
1577 | Log(detector, |
1578 | Form("GetHLTFileName - No entry in FXS table for: id = %s, source = %s", |
1579 | id, source)); |
1580 | delete aResult; |
1581 | return 0; |
1582 | } |
2bb7b766 |
1583 | |
9e080f92 |
1584 | if (aResult->GetRowCount() > 1) { |
1585 | Log(detector, |
1586 | Form("GetHLTFileName - More than one entry in FXS table for: id = %s, source = %s", |
1587 | id, source)); |
1588 | delete aResult; |
1589 | return 0; |
1590 | } |
1591 | |
1592 | if (aResult->GetFieldCount() != 3) { |
1593 | Log(detector, |
1594 | Form("GetHLTFileName - Wrong field count in FXS table for: id = %s, source = %s", |
1595 | id, source)); |
1596 | delete aResult; |
1597 | return 0; |
1598 | } |
1599 | |
1600 | TSQLRow* aRow = dynamic_cast<TSQLRow*> (aResult->Next()); |
1601 | |
1602 | if (!aRow){ |
1603 | Log(detector, Form("GetHLTFileName - Empty set result from query: id = %s, source = %s", |
1604 | id, source)); |
1605 | delete aResult; |
1606 | return 0; |
1607 | } |
1608 | |
1609 | TString filePath(aRow->GetField(0), aRow->GetFieldLength(0)); |
1610 | TString fileSize(aRow->GetField(1), aRow->GetFieldLength(1)); |
1611 | TString fileMd5Sum(aRow->GetField(2), aRow->GetFieldLength(2)); |
1612 | |
1613 | delete aResult; |
1614 | delete aRow; |
1615 | |
1616 | AliDebug(2, Form("filePath = %s",filePath.Data())); |
1617 | |
1618 | // The full file path in HLT FXS is runNb/DET/DDLnumber/filePath |
1619 | // TString fullFilePath = Form("%d/%s/%s/%s", GetCurrentRun(), detector, source, filePath.Data()); |
1620 | |
1621 | // retrieved file is renamed to make it unique |
1622 | TString localFileName = Form("HLT_%s_%d_%s_%s.shuttle", |
1623 | detector, GetCurrentRun(), id, source); |
1624 | |
1625 | // file retrieval from HLT FXS |
1626 | Bool_t result = RetrieveHLTFile(filePath.Data(), localFileName.Data()); |
1627 | if(!result) |
1628 | { |
1629 | Log(detector, Form("GetHLTFileName - Copy of file %s from HLT FXS failed", filePath.Data())); |
1630 | return 0; |
1631 | } else { |
1632 | AliInfo(Form("File %s copied from HLT FXS into %s/%s", |
36c99a6a |
1633 | filePath.Data(), GetShuttleTempDir(), localFileName.Data())); |
9e080f92 |
1634 | } |
1635 | |
1636 | // compare md5sum of local file with the one stored in the HLT DB |
1637 | Int_t md5Comp = gSystem->Exec(Form("md5sum %s/%s |grep %s 2>&1 > /dev/null", |
36c99a6a |
1638 | GetShuttleTempDir(), localFileName.Data(), fileMd5Sum.Data())); |
9e080f92 |
1639 | |
1640 | if (md5Comp != 0) |
1641 | { |
1642 | Log(detector, Form("GetHLTFileName - md5sum of file %s does not match with local copy!", filePath.Data())); |
1643 | return 0; |
1644 | } |
1645 | |
1646 | fFXSCalled[kHLT]=kTRUE; |
1647 | TObjString *fileParams = new TObjString(Form("%s#!?!#%s", id, source)); |
1648 | fFXSlist[kHLT].Add(fileParams); |
1649 | |
1650 | static TString fullLocalFileName; |
36c99a6a |
1651 | fullLocalFileName = TString::Format("%s/%s", GetShuttleTempDir(), localFileName.Data()); |
1652 | |
9e080f92 |
1653 | AliInfo(Form("fullLocalFileName = %s", fullLocalFileName.Data())); |
1654 | |
1655 | return fullLocalFileName.Data(); |
2bb7b766 |
1656 | |
1657 | } |
1658 | |
1659 | //______________________________________________________________________________________________ |
9e080f92 |
1660 | Bool_t AliShuttle::RetrieveHLTFile(const char* hltFileName, const char* localFileName) |
1661 | { |
1662 | // Copies file from HLT FXS to local Shuttle machine |
2bb7b766 |
1663 | |
9e080f92 |
1664 | // check temp directory: trying to cd to temp; if it does not exist, create it |
1665 | AliDebug(2, Form("Copy file %s from HLT FXS into %s/%s", |
36c99a6a |
1666 | hltFileName, GetShuttleTempDir(), localFileName)); |
9e080f92 |
1667 | |
36c99a6a |
1668 | void* dir = gSystem->OpenDirectory(GetShuttleTempDir()); |
9e080f92 |
1669 | if (dir == NULL) { |
36c99a6a |
1670 | if (gSystem->mkdir(GetShuttleTempDir(), kTRUE)) { |
1671 | AliError(Form("Can't open directory <%s>", GetShuttleTempDir())); |
9e080f92 |
1672 | return kFALSE; |
1673 | } |
1674 | |
1675 | } else { |
1676 | gSystem->FreeDirectory(dir); |
1677 | } |
1678 | |
1679 | TString baseHLTFXSFolder = "~"; |
1680 | TString command = Form("scp -oPort=%d %s@%s:%s/%s %s/%s", |
1681 | fConfig->GetFXSPort(kHLT), |
1682 | fConfig->GetFXSUser(kHLT), |
1683 | fConfig->GetFXSHost(kHLT), |
1684 | baseHLTFXSFolder.Data(), |
1685 | hltFileName, |
36c99a6a |
1686 | GetShuttleTempDir(), |
9e080f92 |
1687 | localFileName); |
1688 | |
1689 | AliDebug(2, Form("%s",command.Data())); |
1690 | |
1691 | UInt_t nRetries = 0; |
1692 | UInt_t maxRetries = 3; |
1693 | |
1694 | // copy!! if successful TSystem::Exec returns 0 |
1695 | while(nRetries++ < maxRetries) { |
1696 | AliDebug(2, Form("Trying to copy file. Retry # %d", nRetries)); |
1697 | if(gSystem->Exec(command.Data()) == 0) return kTRUE; |
1698 | } |
1699 | |
1700 | return kFALSE; |
1701 | |
1702 | } |
1703 | |
1704 | //______________________________________________________________________________________________ |
1705 | TList* AliShuttle::GetHLTFileSources(const char* detector, const char* id){ |
1706 | // Retrieves list of HLT sources (DDLnumbers) of file Id |
1707 | |
1708 | // check connection, in case connect |
1709 | if(!Connect(kHLT)){ |
1710 | Log(detector, "GetHLTFileSources - Couldn't connect to HLT FXS database"); |
1711 | return 0; |
1712 | } |
1713 | |
1714 | // Query preparation |
1715 | TString sqlQueryStart = Form("select DDLnumbers from %s where", fConfig->GetFXSdbTable(kHLT)); |
1716 | TString whereClause = Form("run=%d and detector=\"%s\" and fileId=\"%s\"", |
1717 | GetCurrentRun(), detector, id); |
1718 | TString sqlQuery = Form("%s %s", sqlQueryStart.Data(), whereClause.Data()); |
1719 | |
1720 | AliDebug(2, Form("SQL query: \n%s",sqlQuery.Data())); |
1721 | |
1722 | // Query execution |
1723 | TSQLResult* aResult; |
1724 | aResult = fServer[kHLT]->Query(sqlQuery); |
1725 | if (!aResult) { |
1726 | Log(detector, Form("GetHLTFileSources - Can't execute SQL query for id: %s", id)); |
1727 | return 0; |
1728 | } |
1729 | |
1730 | if (aResult->GetRowCount() == 0) { |
1731 | Log(detector, |
1732 | Form("GetHLTFileSources - No entry in FXS table for id: %s", id)); |
1733 | delete aResult; |
1734 | return 0; |
1735 | } |
1736 | |
1737 | TSQLRow* aRow; |
1738 | TList *list = new TList(); |
1739 | list->SetOwner(1); |
1740 | |
1741 | while((aRow = aResult->Next())){ |
1742 | |
1743 | TString ddlNumbers(aRow->GetField(0), aRow->GetFieldLength(0)); |
1744 | AliDebug(2, Form("DDLnumbers = %s", ddlNumbers.Data())); |
1745 | list->Add(new TObjString(ddlNumbers)); |
1746 | delete aRow; |
1747 | } |
1748 | delete aResult; |
1749 | |
1750 | return list; |
2bb7b766 |
1751 | |
1752 | } |
1753 | |
1754 | //______________________________________________________________________________________________ |
1755 | Bool_t AliShuttle::UpdateDAQTable() |
1756 | { |
57f50b3c |
1757 | // Update DAQ table filling time_processed field in all rows corresponding to current run and detector |
1758 | |
1759 | // check connection, in case connect |
1760 | if(!Connect(kDAQ)){ |
2c15234c |
1761 | Log(fCurrentDetector, "UpdateDAQTable - Couldn't connect to DAQ FXS database"); |
57f50b3c |
1762 | return kFALSE; |
1763 | } |
1764 | |
1765 | TTimeStamp now; // now |
1766 | |
2c15234c |
1767 | // Loop on FXS list entries |
1768 | TIter iter(&fFXSlist[kDAQ]); |
1769 | TObjString *aFXSentry=0; |
1770 | while((aFXSentry = dynamic_cast<TObjString*> (iter.Next()))){ |
1771 | TString aFXSentrystr = aFXSentry->String(); |
9e080f92 |
1772 | TObjArray *aFXSarray = aFXSentrystr.Tokenize("#!?!#"); |
2c15234c |
1773 | if(!aFXSarray || aFXSarray->GetEntries() != 2 ) { |
1774 | Log(fCurrentDetector, Form("UpdateDAQTable - error updating FXS entry. Check string: <%s>", |
1775 | aFXSentrystr.Data())); |
1776 | if(aFXSarray) delete aFXSarray; |
57f50b3c |
1777 | return kFALSE; |
1778 | } |
2c15234c |
1779 | const char* fileId = ((TObjString*) aFXSarray->At(0))->GetName(); |
1780 | const char* daqSource = ((TObjString*) aFXSarray->At(1))->GetName(); |
57f50b3c |
1781 | TString whereClause = Form("where run=%d and detector=\"%s\" and fileId=\"%s\" and DAQsource=\"%s\";", |
eba76848 |
1782 | GetCurrentRun(), fCurrentDetector.Data(), fileId, daqSource); |
57f50b3c |
1783 | |
2c15234c |
1784 | delete aFXSarray; |
57f50b3c |
1785 | |
2c15234c |
1786 | TString sqlQuery = Form("update %s set time_processed=%d %s", fConfig->GetFXSdbTable(kDAQ), |
1787 | now.GetSec(), whereClause.Data()); |
57f50b3c |
1788 | |
84090f85 |
1789 | AliDebug(2, Form("SQL query: \n%s",sqlQuery.Data())); |
57f50b3c |
1790 | |
1791 | // Query execution |
1792 | TSQLResult* aResult; |
9e080f92 |
1793 | aResult = dynamic_cast<TSQLResult*> (fServer[kDAQ]->Query(sqlQuery)); |
57f50b3c |
1794 | if (!aResult) { |
2bb7b766 |
1795 | Log(fCurrentDetector, Form("UpdateDAQTable - Can't execute SQL query <%s>", sqlQuery.Data())); |
57f50b3c |
1796 | return kFALSE; |
1797 | } |
1798 | delete aResult; |
2bb7b766 |
1799 | } |
57f50b3c |
1800 | |
2bb7b766 |
1801 | return kTRUE; |
1802 | } |
57f50b3c |
1803 | |
9e080f92 |
1804 | //______________________________________________________________________________________________ |
1805 | Bool_t AliShuttle::UpdateHLTTable() |
1806 | { |
1807 | // Update HLT table filling time_processed field in all rows corresponding to current run and detector |
1808 | |
1809 | // check connection, in case connect |
1810 | if(!Connect(kHLT)){ |
1811 | Log(fCurrentDetector, "UpdateHLTTable - Couldn't connect to HLT FXS database"); |
1812 | return kFALSE; |
1813 | } |
1814 | |
1815 | TTimeStamp now; // now |
1816 | |
1817 | // Loop on FXS list entries |
1818 | TIter iter(&fFXSlist[kHLT]); |
1819 | TObjString *aFXSentry=0; |
1820 | while((aFXSentry = dynamic_cast<TObjString*> (iter.Next()))){ |
1821 | TString aFXSentrystr = aFXSentry->String(); |
1822 | TObjArray *aFXSarray = aFXSentrystr.Tokenize("#!?!#"); |
1823 | if(!aFXSarray || aFXSarray->GetEntries() != 2 ) { |
1824 | Log(fCurrentDetector, Form("UpdateHLTTable - error updating FXS entry. Check string: <%s>", |
1825 | aFXSentrystr.Data())); |
1826 | if(aFXSarray) delete aFXSarray; |
1827 | return kFALSE; |
1828 | } |
1829 | const char* fileId = ((TObjString*) aFXSarray->At(0))->GetName(); |
1830 | const char* hltSource = ((TObjString*) aFXSarray->At(1))->GetName(); |
1831 | TString whereClause = Form("where run=%d and detector=\"%s\" and fileId=\"%s\" and DDLnumbers=\"%s\";", |
1832 | GetCurrentRun(), fCurrentDetector.Data(), fileId, hltSource); |
1833 | |
1834 | delete aFXSarray; |
1835 | |
1836 | TString sqlQuery = Form("update %s set time_processed=%d %s", fConfig->GetFXSdbTable(kHLT), |
1837 | now.GetSec(), whereClause.Data()); |
1838 | |
1839 | AliDebug(2, Form("SQL query: \n%s",sqlQuery.Data())); |
1840 | |
1841 | // Query execution |
1842 | TSQLResult* aResult; |
1843 | aResult = dynamic_cast<TSQLResult*> (fServer[kHLT]->Query(sqlQuery)); |
1844 | if (!aResult) { |
1845 | Log(fCurrentDetector, Form("UpdateHLTTable - Can't execute SQL query <%s>", sqlQuery.Data())); |
1846 | return kFALSE; |
1847 | } |
1848 | delete aResult; |
1849 | } |
1850 | |
1851 | return kTRUE; |
1852 | } |
57f50b3c |
1853 | |
2bb7b766 |
1854 | //______________________________________________________________________________________________ |
1855 | Bool_t AliShuttle::UpdateShuttleLogbook(const char* detector, const char* status) |
1856 | { |
e7f62f16 |
1857 | // |
1858 | // Update Shuttle logbook filling detector or shuttle_done column |
1859 | // ex. of usage: UpdateShuttleLogbook("PHOS", "DONE") or UpdateShuttleLogbook("shuttle_done") |
1860 | // |
57f50b3c |
1861 | |
2bb7b766 |
1862 | // check connection, in case connect |
be48e3ea |
1863 | if(!Connect(3)){ |
2bb7b766 |
1864 | Log("SHUTTLE", "UpdateShuttleLogbook - Couldn't connect to DAQ Logbook."); |
1865 | return kFALSE; |
57f50b3c |
1866 | } |
1867 | |
2bb7b766 |
1868 | TString detName(detector); |
1869 | TString setClause; |
e7f62f16 |
1870 | if(detName == "shuttle_done") |
1871 | { |
2bb7b766 |
1872 | setClause = "set shuttle_done=1"; |
e7f62f16 |
1873 | |
1874 | // Send the information to ML |
1875 | TMonaLisaText mlStatus("SHUTTLE_status", "Done"); |
1876 | |
1877 | TList mlList; |
1878 | mlList.Add(&mlStatus); |
1879 | |
1880 | fMonaLisa->SendParameters(&mlList); |
2bb7b766 |
1881 | } else { |
2bb7b766 |
1882 | TString statusStr(status); |
1883 | if(statusStr.Contains("done", TString::kIgnoreCase) || |
1884 | statusStr.Contains("failed", TString::kIgnoreCase)){ |
eba76848 |
1885 | setClause = Form("set %s=\"%s\"", detector, status); |
2bb7b766 |
1886 | } else { |
1887 | Log("SHUTTLE", |
1888 | Form("UpdateShuttleLogbook - Invalid status <%s> for detector %s", |
1889 | status, detector)); |
1890 | return kFALSE; |
1891 | } |
1892 | } |
57f50b3c |
1893 | |
2bb7b766 |
1894 | TString whereClause = Form("where run=%d", GetCurrentRun()); |
1895 | |
1896 | TString sqlQuery = Form("update logbook_shuttle %s %s", |
1897 | setClause.Data(), whereClause.Data()); |
57f50b3c |
1898 | |
2bb7b766 |
1899 | AliDebug(2, Form("SQL query: \n%s",sqlQuery.Data())); |
1900 | |
1901 | // Query execution |
1902 | TSQLResult* aResult; |
be48e3ea |
1903 | aResult = dynamic_cast<TSQLResult*> (fServer[3]->Query(sqlQuery)); |
2bb7b766 |
1904 | if (!aResult) { |
1905 | Log("SHUTTLE", Form("UpdateShuttleLogbook - Can't execute query <%s>", sqlQuery.Data())); |
1906 | return kFALSE; |
57f50b3c |
1907 | } |
2bb7b766 |
1908 | delete aResult; |
57f50b3c |
1909 | |
1910 | return kTRUE; |
1911 | } |
1912 | |
1913 | //______________________________________________________________________________________________ |
2bb7b766 |
1914 | Int_t AliShuttle::GetCurrentRun() const |
1915 | { |
1916 | // Get current run from logbook entry |
57f50b3c |
1917 | |
2bb7b766 |
1918 | return fLogbookEntry ? fLogbookEntry->GetRun() : -1; |
57f50b3c |
1919 | } |
1920 | |
1921 | //______________________________________________________________________________________________ |
2bb7b766 |
1922 | UInt_t AliShuttle::GetCurrentStartTime() const |
1923 | { |
1924 | // get current start time |
57f50b3c |
1925 | |
2bb7b766 |
1926 | return fLogbookEntry ? fLogbookEntry->GetStartTime() : 0; |
57f50b3c |
1927 | } |
1928 | |
1929 | //______________________________________________________________________________________________ |
2bb7b766 |
1930 | UInt_t AliShuttle::GetCurrentEndTime() const |
1931 | { |
1932 | // get current end time from logbook entry |
57f50b3c |
1933 | |
2bb7b766 |
1934 | return fLogbookEntry ? fLogbookEntry->GetEndTime() : 0; |
57f50b3c |
1935 | } |
1936 | |
b948db8d |
1937 | //______________________________________________________________________________________________ |
1938 | void AliShuttle::Log(const char* detector, const char* message) |
1939 | { |
58bc3020 |
1940 | // Fill log string with a message |
b948db8d |
1941 | |
36c99a6a |
1942 | void* dir = gSystem->OpenDirectory(GetShuttleLogDir()); |
84090f85 |
1943 | if (dir == NULL) { |
36c99a6a |
1944 | if (gSystem->mkdir(GetShuttleLogDir(), kTRUE)) { |
1945 | AliError(Form("Can't open directory <%s>", GetShuttleLogDir())); |
84090f85 |
1946 | return; |
1947 | } |
b948db8d |
1948 | |
84090f85 |
1949 | } else { |
1950 | gSystem->FreeDirectory(dir); |
1951 | } |
b948db8d |
1952 | |
cb343cfd |
1953 | TString toLog = Form("%s (%d): %s - ", TTimeStamp(time(0)).AsString("s"), getpid(), detector); |
e7f62f16 |
1954 | if (GetCurrentRun() >= 0) |
1955 | toLog += Form("run %d - ", GetCurrentRun()); |
2bb7b766 |
1956 | toLog += Form("%s", message); |
1957 | |
84090f85 |
1958 | AliInfo(toLog.Data()); |
b948db8d |
1959 | |
84090f85 |
1960 | TString fileName; |
e7f62f16 |
1961 | if (GetCurrentRun() >= 0) |
1962 | fileName.Form("%s/%s_%d.log", GetShuttleLogDir(), detector, GetCurrentRun()); |
1963 | else |
1964 | fileName.Form("%s/%s.log", GetShuttleLogDir(), detector); |
1965 | |
84090f85 |
1966 | gSystem->ExpandPathName(fileName); |
1967 | |
1968 | ofstream logFile; |
1969 | logFile.open(fileName, ofstream::out | ofstream::app); |
1970 | |
1971 | if (!logFile.is_open()) { |
1972 | AliError(Form("Could not open file %s", fileName.Data())); |
1973 | return; |
1974 | } |
7bfb2090 |
1975 | |
84090f85 |
1976 | logFile << toLog.Data() << "\n"; |
b948db8d |
1977 | |
84090f85 |
1978 | logFile.close(); |
b948db8d |
1979 | } |
2bb7b766 |
1980 | |
2bb7b766 |
1981 | //______________________________________________________________________________________________ |
1982 | Bool_t AliShuttle::Collect(Int_t run) |
1983 | { |
eba76848 |
1984 | // |
1985 | // Collects conditions data for all UNPROCESSED run written to DAQ LogBook in case of run = -1 (default) |
1986 | // If a dedicated run is given this run is processed |
1987 | // |
1988 | // In operational mode, this is the Shuttle function triggered by the EOR signal. |
1989 | // |
2bb7b766 |
1990 | |
eba76848 |
1991 | if (run == -1) |
1992 | Log("SHUTTLE","Collect - Shuttle called. Collecting conditions data for unprocessed runs"); |
1993 | else |
1994 | Log("SHUTTLE", Form("Collect - Shuttle called. Collecting conditions data for run %d", run)); |
cb343cfd |
1995 | |
1996 | SetLastAction("Starting"); |
2bb7b766 |
1997 | |
1998 | TString whereClause("where shuttle_done=0"); |
eba76848 |
1999 | if (run != -1) |
2000 | whereClause += Form(" and run=%d", run); |
2bb7b766 |
2001 | |
2002 | TObjArray shuttleLogbookEntries; |
be48e3ea |
2003 | if (!QueryShuttleLogbook(whereClause, shuttleLogbookEntries)) |
2004 | { |
cb343cfd |
2005 | Log("SHUTTLE", "Collect - Can't retrieve entries from Shuttle logbook"); |
2bb7b766 |
2006 | return kFALSE; |
2007 | } |
2008 | |
9e080f92 |
2009 | if (shuttleLogbookEntries.GetEntries() == 0) |
2010 | { |
2011 | if (run == -1) |
2012 | Log("SHUTTLE","Collect - Found no UNPROCESSED runs in Shuttle logbook"); |
2013 | else |
2014 | Log("SHUTTLE", Form("Collect - Run %d is already DONE " |
2015 | "or it does not exist in Shuttle logbook", run)); |
2016 | return kTRUE; |
2017 | } |
2018 | |
be48e3ea |
2019 | for (UInt_t iDet=0; iDet<NDetectors(); iDet++) |
2020 | fFirstUnprocessed[iDet] = kTRUE; |
2021 | |
fc5a4708 |
2022 | if (run != -1) |
be48e3ea |
2023 | { |
2024 | // query Shuttle logbook for earlier runs, check if some detectors are unprocessed, |
2025 | // flag them into fFirstUnprocessed array |
2026 | TString whereClause(Form("where shuttle_done=0 and run < %d", run)); |
2027 | TObjArray tmpLogbookEntries; |
2028 | if (!QueryShuttleLogbook(whereClause, tmpLogbookEntries)) |
2029 | { |
2030 | Log("SHUTTLE", "Collect - Can't retrieve entries from Shuttle logbook"); |
2031 | return kFALSE; |
2032 | } |
2033 | |
2034 | TIter iter(&tmpLogbookEntries); |
2035 | AliShuttleLogbookEntry* anEntry = 0; |
2036 | while ((anEntry = dynamic_cast<AliShuttleLogbookEntry*> (iter.Next()))) |
2037 | { |
2038 | for (UInt_t iDet=0; iDet<NDetectors(); iDet++) |
2039 | { |
2040 | if (anEntry->GetDetectorStatus(iDet) == AliShuttleLogbookEntry::kUnprocessed) |
2041 | { |
2042 | AliDebug(2, Form("Run %d: setting %s as \"not first time unprocessed\"", |
2043 | anEntry->GetRun(), GetDetName(iDet))); |
2044 | fFirstUnprocessed[iDet] = kFALSE; |
2045 | } |
2046 | } |
2047 | |
2048 | } |
2049 | |
2050 | } |
2051 | |
2052 | if (!RetrieveConditionsData(shuttleLogbookEntries)) |
2053 | { |
cb343cfd |
2054 | Log("SHUTTLE", "Collect - Process of at least one run failed"); |
2bb7b766 |
2055 | return kFALSE; |
2056 | } |
2057 | |
36c99a6a |
2058 | Log("SHUTTLE", "Collect - Requested run(s) successfully processed"); |
eba76848 |
2059 | return kTRUE; |
2bb7b766 |
2060 | } |
2061 | |
2bb7b766 |
2062 | //______________________________________________________________________________________________ |
2063 | Bool_t AliShuttle::RetrieveConditionsData(const TObjArray& dateEntries) |
2064 | { |
2065 | // Retrieve conditions data for all runs that aren't processed yet |
2066 | |
2067 | Bool_t hasError = kFALSE; |
2068 | |
2069 | TIter iter(&dateEntries); |
2070 | AliShuttleLogbookEntry* anEntry; |
2071 | |
2072 | while ((anEntry = (AliShuttleLogbookEntry*) iter.Next())){ |
2073 | if (!Process(anEntry)){ |
2074 | hasError = kTRUE; |
2075 | } |
2076 | } |
2077 | |
2078 | return hasError == kFALSE; |
2079 | } |
cb343cfd |
2080 | |
2081 | //______________________________________________________________________________________________ |
2082 | ULong_t AliShuttle::GetTimeOfLastAction() const |
2083 | { |
2084 | ULong_t tmp; |
36c99a6a |
2085 | |
cb343cfd |
2086 | fMonitoringMutex->Lock(); |
be48e3ea |
2087 | |
cb343cfd |
2088 | tmp = fLastActionTime; |
36c99a6a |
2089 | |
cb343cfd |
2090 | fMonitoringMutex->UnLock(); |
36c99a6a |
2091 | |
cb343cfd |
2092 | return tmp; |
2093 | } |
2094 | |
2095 | //______________________________________________________________________________________________ |
2096 | const TString AliShuttle::GetLastAction() const |
2097 | { |
2098 | // returns a string description of the last action |
2099 | |
2100 | TString tmp; |
36c99a6a |
2101 | |
cb343cfd |
2102 | fMonitoringMutex->Lock(); |
2103 | |
2104 | tmp = fLastAction; |
2105 | |
2106 | fMonitoringMutex->UnLock(); |
2107 | |
36c99a6a |
2108 | return tmp; |
cb343cfd |
2109 | } |
2110 | |
2111 | //______________________________________________________________________________________________ |
2112 | void AliShuttle::SetLastAction(const char* action) |
2113 | { |
2114 | // updates the monitoring variables |
36c99a6a |
2115 | |
cb343cfd |
2116 | fMonitoringMutex->Lock(); |
36c99a6a |
2117 | |
cb343cfd |
2118 | fLastAction = action; |
2119 | fLastActionTime = time(0); |
2120 | |
2121 | fMonitoringMutex->UnLock(); |
2122 | } |
eba76848 |
2123 | |
2124 | //______________________________________________________________________________________________ |
2125 | const char* AliShuttle::GetRunParameter(const char* param) |
2126 | { |
2127 | // returns run parameter read from DAQ logbook |
2128 | |
2129 | if(!fLogbookEntry) { |
2130 | AliError("No logbook entry!"); |
2131 | return 0; |
2132 | } |
2133 | |
2134 | return fLogbookEntry->GetRunParameter(param); |
2135 | } |
57c1a579 |
2136 | |
2137 | //______________________________________________________________________________________________ |
2138 | Bool_t AliShuttle::SendMail() |
2139 | { |
2140 | // sends a mail to the subdetector expert in case of preprocessor error |
2141 | |
36c99a6a |
2142 | void* dir = gSystem->OpenDirectory(GetShuttleLogDir()); |
57c1a579 |
2143 | if (dir == NULL) |
2144 | { |
36c99a6a |
2145 | if (gSystem->mkdir(GetShuttleLogDir(), kTRUE)) |
57c1a579 |
2146 | { |
36c99a6a |
2147 | AliError(Form("Can't open directory <%s>", GetShuttleLogDir())); |
57c1a579 |
2148 | return kFALSE; |
2149 | } |
2150 | |
2151 | } else { |
2152 | gSystem->FreeDirectory(dir); |
2153 | } |
2154 | |
2155 | TString bodyFileName; |
36c99a6a |
2156 | bodyFileName.Form("%s/mail.body", GetShuttleLogDir()); |
57c1a579 |
2157 | gSystem->ExpandPathName(bodyFileName); |
2158 | |
2159 | ofstream mailBody; |
2160 | mailBody.open(bodyFileName, ofstream::out); |
2161 | |
2162 | if (!mailBody.is_open()) |
2163 | { |
2164 | AliError(Form("Could not open mail body file %s", bodyFileName.Data())); |
2165 | return kFALSE; |
2166 | } |
2167 | |
2168 | TString to=""; |
2169 | TIter iterExperts(fConfig->GetResponsibles(fCurrentDetector)); |
2170 | TObjString *anExpert=0; |
2171 | while ((anExpert = (TObjString*) iterExperts.Next())) |
2172 | { |
2173 | to += Form("%s,", anExpert->GetName()); |
2174 | } |
2175 | to.Remove(to.Length()-1); |
909732f7 |
2176 | AliDebug(2, Form("to: %s",to.Data())); |
57c1a579 |
2177 | |
36c99a6a |
2178 | // TODO this will be removed... |
2179 | if (to.Contains("not_yet_set")) { |
2180 | AliInfo("List of detector responsibles not yet set!"); |
2181 | return kFALSE; |
2182 | } |
2183 | |
57c1a579 |
2184 | TString cc="alberto.colla@cern.ch"; |
2185 | |
2186 | TString subject = Form("%s Shuttle preprocessor error in run %d !", |
2187 | fCurrentDetector.Data(), GetCurrentRun()); |
909732f7 |
2188 | AliDebug(2, Form("subject: %s", subject.Data())); |
57c1a579 |
2189 | |
2190 | TString body = Form("Dear %s expert(s), \n\n", fCurrentDetector.Data()); |
2191 | body += Form("SHUTTLE just detected that your preprocessor " |
36c99a6a |
2192 | "exited with ERROR state in run %d!!\n\n", GetCurrentRun()); |
57c1a579 |
2193 | body += Form("Please check %s status on the web page asap!\n\n", fCurrentDetector.Data()); |
2194 | body += Form("The last 10 lines of %s log file are following:\n\n"); |
2195 | |
909732f7 |
2196 | AliDebug(2, Form("Body begin: %s", body.Data())); |
57c1a579 |
2197 | |
2198 | mailBody << body.Data(); |
2199 | mailBody.close(); |
2200 | mailBody.open(bodyFileName, ofstream::out | ofstream::app); |
2201 | |
36c99a6a |
2202 | TString logFileName = Form("%s/%s.log", GetShuttleLogDir(), fCurrentDetector.Data()); |
57c1a579 |
2203 | TString tailCommand = Form("tail -n 10 %s >> %s", logFileName.Data(), bodyFileName.Data()); |
2204 | if (gSystem->Exec(tailCommand.Data())) |
2205 | { |
2206 | mailBody << Form("%s log file not found ...\n\n", fCurrentDetector.Data()); |
2207 | } |
2208 | |
2209 | TString endBody = Form("------------------------------------------------------\n\n"); |
36c99a6a |
2210 | endBody += Form("In case of problems please contact the SHUTTLE core team.\n\n"); |
2211 | endBody += "Please do not answer this message directly, it is automatically generated.\n\n"; |
57c1a579 |
2212 | endBody += "Sincerely yours,\n\n \t\t\tthe SHUTTLE\n"; |
2213 | |
909732f7 |
2214 | AliDebug(2, Form("Body end: %s", endBody.Data())); |
57c1a579 |
2215 | |
2216 | mailBody << endBody.Data(); |
2217 | |
2218 | mailBody.close(); |
2219 | |
2220 | // send mail! |
2221 | TString mailCommand = Form("mail -s \"%s\" -c %s %s < %s", |
2222 | subject.Data(), |
2223 | cc.Data(), |
2224 | to.Data(), |
2225 | bodyFileName.Data()); |
909732f7 |
2226 | AliDebug(2, Form("mail command: %s", mailCommand.Data())); |
57c1a579 |
2227 | |
2228 | Bool_t result = gSystem->Exec(mailCommand.Data()); |
2229 | |
2230 | return result == 0; |
2231 | } |
36c99a6a |
2232 | |
2233 | //______________________________________________________________________________________________ |
2234 | void AliShuttle::SetShuttleTempDir(const char* tmpDir) |
2235 | { |
2236 | // sets Shuttle temp directory |
2237 | |
2238 | fgkShuttleTempDir = gSystem->ExpandPathName(tmpDir); |
2239 | } |
2240 | |
2241 | //______________________________________________________________________________________________ |
2242 | void AliShuttle::SetShuttleLogDir(const char* logDir) |
2243 | { |
2244 | // sets Shuttle log directory |
2245 | |
2246 | fgkShuttleLogDir = gSystem->ExpandPathName(logDir); |
2247 | } |