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