Hide keyboard shortcuts

Hot-keys on this page

r m x p   toggle line displays

j k   next/prev highlighted chunk

0   (zero) top of page

1   (one) first highlighted chunk

1# This file is part of ctrl_mpexec. 

2# 

3# Developed for the LSST Data Management System. 

4# This product includes software developed by the LSST Project 

5# (http://www.lsst.org). 

6# See the COPYRIGHT file at the top-level directory of this distribution 

7# for details of code ownership. 

8# 

9# This program is free software: you can redistribute it and/or modify 

10# it under the terms of the GNU General Public License as published by 

11# the Free Software Foundation, either version 3 of the License, or 

12# (at your option) any later version. 

13# 

14# This program is distributed in the hope that it will be useful, 

15# but WITHOUT ANY WARRANTY; without even the implied warranty of 

16# MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the 

17# GNU General Public License for more details. 

18# 

19# You should have received a copy of the GNU General Public License 

20# along with this program. If not, see <http://www.gnu.org/licenses/>. 

21 

22__all__ = ['SingleQuantumExecutor'] 

23 

24# ------------------------------- 

25# Imports of standard modules -- 

26# ------------------------------- 

27import logging 

28import os 

29import shutil 

30import sys 

31import tempfile 

32import time 

33from contextlib import contextmanager 

34from collections import defaultdict 

35from itertools import chain 

36from logging import FileHandler 

37from typing import List 

38 

39# ----------------------------- 

40# Imports for other modules -- 

41# ----------------------------- 

42from .quantumGraphExecutor import QuantumExecutor 

43from lsst.daf.base import PropertyList, PropertySet 

44from lsst.obs.base import Instrument 

45from lsst.pipe.base import ( 

46 AdjustQuantumHelper, 

47 ButlerQuantumContext, 

48 InvalidQuantumError, 

49 NoWorkFound, 

50 RepeatableQuantumError, 

51 logInfo, 

52) 

53from lsst.daf.butler import ( 

54 DatasetRef, 

55 DatasetType, 

56 FileDataset, 

57 NamedKeyDict, 

58 Quantum, 

59) 

60from lsst.daf.butler.core.logging import ( 

61 ButlerLogRecordHandler, 

62 ButlerLogRecords, 

63 ButlerMDC, 

64 JsonLogFormatter, 

65) 

66# ---------------------------------- 

67# Local non-exported definitions -- 

68# ---------------------------------- 

69 

70_LOG = logging.getLogger(__name__.partition(".")[2]) 

71 

72 

73class _LogCaptureFlag: 

74 """Simple flag to enable/disable log-to-butler saving. 

75 """ 

76 store: bool = True 

77 

78 

79class SingleQuantumExecutor(QuantumExecutor): 

80 """Executor class which runs one Quantum at a time. 

81 

82 Parameters 

83 ---------- 

84 butler : `~lsst.daf.butler.Butler` 

85 Data butler. 

86 taskFactory : `~lsst.pipe.base.TaskFactory` 

87 Instance of a task factory. 

88 skipExistingIn : `list` [ `str` ], optional 

89 Accepts list of collections, if all Quantum outputs already exist in 

90 the specified list of collections then that Quantum will not be rerun. 

91 clobberOutputs : `bool`, optional 

92 If `True`, then existing outputs in output run collection will be 

93 overwritten. If ``skipExistingIn`` is defined, only outputs from 

94 failed quanta will be overwritten. 

95 enableLsstDebug : `bool`, optional 

96 Enable debugging with ``lsstDebug`` facility for a task. 

97 exitOnKnownError : `bool`, optional 

98 If `True`, call `sys.exit` with the appropriate exit code for special 

99 known exceptions, after printing a traceback, instead of letting the 

100 exception propagate up to calling. This is always the behavior for 

101 InvalidQuantumError. 

102 """ 

103 

104 stream_json_logs = True 

105 """If True each log record is written to a temporary file and ingested 

106 when quantum completes. If False the records are accumulated in memory 

107 and stored in butler on quantum completion.""" 

108 

109 def __init__(self, taskFactory, skipExistingIn=None, clobberOutputs=False, enableLsstDebug=False, 

110 exitOnKnownError=False): 

111 self.taskFactory = taskFactory 

112 self.skipExistingIn = skipExistingIn 

113 self.enableLsstDebug = enableLsstDebug 

114 self.clobberOutputs = clobberOutputs 

115 self.exitOnKnownError = exitOnKnownError 

116 self.log_handler = None 

117 

118 def execute(self, taskDef, quantum, butler): 

119 # Docstring inherited from QuantumExecutor.execute 

120 startTime = time.time() 

121 

122 with self.captureLogging(taskDef, quantum, butler) as captureLog: 

123 

124 # Save detailed resource usage before task start to metadata. 

125 quantumMetadata = PropertyList() 

126 logInfo(None, "prep", metadata=quantumMetadata) 

127 

128 taskClass, label, config = taskDef.taskClass, taskDef.label, taskDef.config 

129 

130 # check whether to skip or delete old outputs, if it returns True 

131 # or raises an exception do not try to store logs, as they may be 

132 # already in butler. 

133 captureLog.store = False 

134 if self.checkExistingOutputs(quantum, butler, taskDef): 

135 _LOG.info("Skipping already-successful quantum for label=%s dataId=%s.", label, 

136 quantum.dataId) 

137 return 

138 captureLog.store = True 

139 

140 try: 

141 quantum = self.updatedQuantumInputs(quantum, butler, taskDef) 

142 except NoWorkFound as exc: 

143 _LOG.info("Nothing to do for task '%s' on quantum %s; saving metadata and skipping: %s", 

144 taskDef.label, quantum.dataId, str(exc)) 

145 # Make empty metadata that looks something like what a 

146 # do-nothing task would write (but we don't bother with empty 

147 # nested PropertySets for subtasks). This is slightly 

148 # duplicative with logic in pipe_base that we can't easily call 

149 # from here; we'll fix this on DM-29761. 

150 logInfo(None, "end", metadata=quantumMetadata) 

151 fullMetadata = PropertySet() 

152 fullMetadata[taskDef.label] = PropertyList() 

153 fullMetadata["quantum"] = quantumMetadata 

154 self.writeMetadata(quantum, fullMetadata, taskDef, butler) 

155 return 

156 

157 # enable lsstDebug debugging 

158 if self.enableLsstDebug: 

159 try: 

160 _LOG.debug("Will try to import debug.py") 

161 import debug # noqa:F401 

162 except ImportError: 

163 _LOG.warn("No 'debug' module found.") 

164 

165 # initialize global state 

166 self.initGlobals(quantum, butler) 

167 

168 # Ensure that we are executing a frozen config 

169 config.freeze() 

170 logInfo(None, "init", metadata=quantumMetadata) 

171 task = self.makeTask(taskClass, label, config, butler) 

172 logInfo(None, "start", metadata=quantumMetadata) 

173 try: 

174 self.runQuantum(task, quantum, taskDef, butler) 

175 except Exception: 

176 _LOG.exception("Execution of task '%s' on quantum %s failed", 

177 taskDef.label, quantum.dataId) 

178 raise 

179 logInfo(None, "end", metadata=quantumMetadata) 

180 fullMetadata = task.getFullMetadata() 

181 fullMetadata["quantum"] = quantumMetadata 

182 self.writeMetadata(quantum, fullMetadata, taskDef, butler) 

183 stopTime = time.time() 

184 _LOG.info("Execution of task '%s' on quantum %s took %.3f seconds", 

185 taskDef.label, quantum.dataId, stopTime - startTime) 

186 

187 @contextmanager 

188 def captureLogging(self, taskDef, quantum, butler): 

189 """Configure logging system to capture logs for execution of this task. 

190 

191 Parameters 

192 ---------- 

193 taskDef : `lsst.pipe.base.TaskDef` 

194 The task definition. 

195 quantum : `~lsst.daf.butler.Quantum` 

196 Single Quantum instance. 

197 butler : `~lsst.daf.butler.Butler` 

198 Butler to write logs to. 

199 

200 Notes 

201 ----- 

202 Expected to be used as a context manager to ensure that logging 

203 records are inserted into the butler once the quantum has been 

204 executed: 

205 

206 .. code-block:: py 

207 

208 with self.captureLogging(taskDef, quantum, butler): 

209 # Run quantum and capture logs. 

210 

211 Ths method can also setup logging to attach task- or 

212 quantum-specific information to log messages. Potentially this can 

213 take into account some info from task configuration as well. 

214 """ 

215 # Add a handler to the root logger to capture execution log output. 

216 # How does it get removed reliably? 

217 if taskDef.logOutputDatasetName is not None: 

218 # Either accumulate into ButlerLogRecords or stream 

219 # JSON records to file and ingest that. 

220 tmpdir = None 

221 if self.stream_json_logs: 

222 # Create the log file in a temporary directory rather than 

223 # creating a temporary file. This is necessary because 

224 # temporary files are created with restrictive permissions 

225 # and during file ingest these permissions persist in the 

226 # datastore. Using a temp directory allows us to create 

227 # a file with umask default permissions. 

228 tmpdir = tempfile.mkdtemp(prefix="butler-temp-logs-") 

229 

230 # Construct a file to receive the log records and "touch" it. 

231 log_file = os.path.join(tmpdir, f"butler-log-{taskDef.label}.json") 

232 with open(log_file, "w"): 

233 pass 

234 self.log_handler = FileHandler(log_file) 

235 self.log_handler.setFormatter(JsonLogFormatter()) 

236 else: 

237 self.log_handler = ButlerLogRecordHandler() 

238 

239 logging.getLogger().addHandler(self.log_handler) 

240 

241 # include quantum dataId and task label into MDC 

242 label = taskDef.label 

243 if quantum.dataId: 

244 label += f":{quantum.dataId}" 

245 

246 ctx = _LogCaptureFlag() 

247 try: 

248 with ButlerMDC.set_mdc({"LABEL": label, "RUN": butler.run}): 

249 yield ctx 

250 finally: 

251 # Ensure that the logs are stored in butler. 

252 self.writeLogRecords(quantum, taskDef, butler, ctx.store) 

253 if tmpdir: 

254 shutil.rmtree(tmpdir, ignore_errors=True) 

255 

256 def checkExistingOutputs(self, quantum, butler, taskDef): 

257 """Decide whether this quantum needs to be executed. 

258 

259 If only partial outputs exist then they are removed if 

260 ``clobberOutputs`` is True, otherwise an exception is raised. 

261 

262 Parameters 

263 ---------- 

264 quantum : `~lsst.daf.butler.Quantum` 

265 Quantum to check for existing outputs 

266 butler : `~lsst.daf.butler.Butler` 

267 Data butler. 

268 taskDef : `~lsst.pipe.base.TaskDef` 

269 Task definition structure. 

270 

271 Returns 

272 ------- 

273 exist : `bool` 

274 `True` if ``self.skipExistingIn`` is defined, and a previous 

275 execution of this quanta appears to have completed successfully 

276 (either because metadata was written or all datasets were written). 

277 `False` otherwise. 

278 

279 Raises 

280 ------ 

281 RuntimeError 

282 Raised if some outputs exist and some not. 

283 """ 

284 if self.skipExistingIn and taskDef.metadataDatasetName is not None: 

285 # Metadata output exists; this is sufficient to assume the previous 

286 # run was successful and should be skipped. 

287 ref = butler.registry.findDataset(taskDef.metadataDatasetName, quantum.dataId, 

288 collections=self.skipExistingIn) 

289 if ref is not None: 

290 if butler.datastore.exists(ref): 

291 return True 

292 

293 # Previously we always checked for existing outputs in `butler.run`, 

294 # now logic gets more complicated as we only want to skip quantum 

295 # whose outputs exist in `self.skipExistingIn` but pruning should only 

296 # be done for outputs existing in `butler.run`. 

297 

298 def findOutputs(collections): 

299 """Find quantum outputs in specified collections. 

300 """ 

301 existingRefs = [] 

302 missingRefs = [] 

303 for datasetRefs in quantum.outputs.values(): 

304 for datasetRef in datasetRefs: 

305 ref = butler.registry.findDataset(datasetRef.datasetType, datasetRef.dataId, 

306 collections=collections) 

307 if ref is not None and butler.datastore.exists(ref): 

308 existingRefs.append(ref) 

309 else: 

310 missingRefs.append(datasetRef) 

311 return existingRefs, missingRefs 

312 

313 existingRefs, missingRefs = findOutputs(self.skipExistingIn) 

314 if self.skipExistingIn: 

315 if existingRefs and not missingRefs: 

316 # everything is already there 

317 return True 

318 

319 # If we are to re-run quantum then prune datasets that exists in 

320 # output run collection, only if `self.clobberOutputs` is set. 

321 if existingRefs: 

322 existingRefs, missingRefs = findOutputs(butler.run) 

323 if existingRefs and missingRefs: 

324 _LOG.debug("Partial outputs exist for task %s dataId=%s collection=%s " 

325 "existingRefs=%s missingRefs=%s", 

326 taskDef, quantum.dataId, butler.run, existingRefs, missingRefs) 

327 if self.clobberOutputs: 

328 # only prune 

329 _LOG.info("Removing partial outputs for task %s: %s", taskDef, existingRefs) 

330 butler.pruneDatasets(existingRefs, disassociate=True, unstore=True, purge=True) 

331 return False 

332 else: 

333 raise RuntimeError(f"Registry inconsistency while checking for existing outputs:" 

334 f" collection={butler.run} existingRefs={existingRefs}" 

335 f" missingRefs={missingRefs}") 

336 

337 # need to re-run 

338 return False 

339 

340 def makeTask(self, taskClass, name, config, butler): 

341 """Make new task instance. 

342 

343 Parameters 

344 ---------- 

345 taskClass : `type` 

346 Sub-class of `~lsst.pipe.base.PipelineTask`. 

347 name : `str` 

348 Name for this task. 

349 config : `~lsst.pipe.base.PipelineTaskConfig` 

350 Configuration object for this task 

351 

352 Returns 

353 ------- 

354 task : `~lsst.pipe.base.PipelineTask` 

355 Instance of ``taskClass`` type. 

356 butler : `~lsst.daf.butler.Butler` 

357 Data butler. 

358 """ 

359 # call task factory for that 

360 return self.taskFactory.makeTask(taskClass, name, config, None, butler) 

361 

362 def updatedQuantumInputs(self, quantum, butler, taskDef): 

363 """Update quantum with extra information, returns a new updated 

364 Quantum. 

365 

366 Some methods may require input DatasetRefs to have non-None 

367 ``dataset_id``, but in case of intermediate dataset it may not be 

368 filled during QuantumGraph construction. This method will retrieve 

369 missing info from registry. 

370 

371 Parameters 

372 ---------- 

373 quantum : `~lsst.daf.butler.Quantum` 

374 Single Quantum instance. 

375 butler : `~lsst.daf.butler.Butler` 

376 Data butler. 

377 taskDef : `~lsst.pipe.base.TaskDef` 

378 Task definition structure. 

379 

380 Returns 

381 ------- 

382 update : `~lsst.daf.butler.Quantum` 

383 Updated Quantum instance 

384 """ 

385 anyChanges = False 

386 updatedInputs = defaultdict(list) 

387 for key, refsForDatasetType in quantum.inputs.items(): 

388 newRefsForDatasetType = updatedInputs[key] 

389 for ref in refsForDatasetType: 

390 if ref.id is None: 

391 resolvedRef = butler.registry.findDataset(ref.datasetType, ref.dataId, 

392 collections=butler.collections) 

393 if resolvedRef is None: 

394 _LOG.info("No dataset found for %s", ref) 

395 continue 

396 else: 

397 _LOG.debug("Updated dataset ID for %s", ref) 

398 else: 

399 resolvedRef = ref 

400 # We need to ask datastore if the dataset actually exists 

401 # because the Registry of a local "execution butler" cannot 

402 # know this (because we prepopulate it with all of the datasets 

403 # that might be created). 

404 if butler.datastore.exists(resolvedRef): 

405 newRefsForDatasetType.append(resolvedRef) 

406 if len(newRefsForDatasetType) != len(refsForDatasetType): 

407 anyChanges = True 

408 # If we removed any input datasets, let the task check if it has enough 

409 # to proceed and/or prune related datasets that it also doesn't 

410 # need/produce anymore. It will raise NoWorkFound if it can't run, 

411 # which we'll let propagate up. This is exactly what we run during QG 

412 # generation, because a task shouldn't care whether an input is missing 

413 # because some previous task didn't produce it, or because it just 

414 # wasn't there during QG generation. 

415 updatedInputs = NamedKeyDict[DatasetType, List[DatasetRef]](updatedInputs.items()) 

416 helper = AdjustQuantumHelper(updatedInputs, quantum.outputs) 

417 if anyChanges: 

418 helper.adjust_in_place(taskDef.connections, label=taskDef.label, data_id=quantum.dataId) 

419 return Quantum(taskName=quantum.taskName, 

420 taskClass=quantum.taskClass, 

421 dataId=quantum.dataId, 

422 initInputs=quantum.initInputs, 

423 inputs=helper.inputs, 

424 outputs=helper.outputs 

425 ) 

426 

427 def runQuantum(self, task, quantum, taskDef, butler): 

428 """Execute task on a single quantum. 

429 

430 Parameters 

431 ---------- 

432 task : `~lsst.pipe.base.PipelineTask` 

433 Task object. 

434 quantum : `~lsst.daf.butler.Quantum` 

435 Single Quantum instance. 

436 taskDef : `~lsst.pipe.base.TaskDef` 

437 Task definition structure. 

438 butler : `~lsst.daf.butler.Butler` 

439 Data butler. 

440 """ 

441 # Create a butler that operates in the context of a quantum 

442 butlerQC = ButlerQuantumContext(butler, quantum) 

443 

444 # Get the input and output references for the task 

445 inputRefs, outputRefs = taskDef.connections.buildDatasetRefs(quantum) 

446 

447 # Call task runQuantum() method. Catch a few known failure modes and 

448 # translate them into specific 

449 try: 

450 task.runQuantum(butlerQC, inputRefs, outputRefs) 

451 except NoWorkFound as err: 

452 # Not an error, just an early exit. 

453 _LOG.info("Task '%s' on quantum %s exited early: %s", 

454 taskDef.label, quantum.dataId, str(err)) 

455 pass 

456 except RepeatableQuantumError as err: 

457 if self.exitOnKnownError: 

458 _LOG.warning("Caught repeatable quantum error for %s (%s):", taskDef, quantum.dataId) 

459 _LOG.warning(err, exc_info=True) 

460 sys.exit(err.EXIT_CODE) 

461 else: 

462 raise 

463 except InvalidQuantumError as err: 

464 _LOG.fatal("Invalid quantum error for %s (%s): %s", taskDef, quantum.dataId) 

465 _LOG.fatal(err, exc_info=True) 

466 sys.exit(err.EXIT_CODE) 

467 

468 def writeMetadata(self, quantum, metadata, taskDef, butler): 

469 if taskDef.metadataDatasetName is not None: 

470 # DatasetRef has to be in the Quantum outputs, can lookup by name 

471 try: 

472 ref = quantum.outputs[taskDef.metadataDatasetName] 

473 except LookupError as exc: 

474 raise InvalidQuantumError( 

475 f"Quantum outputs is missing metadata dataset type {taskDef.metadataDatasetName};" 

476 f" this could happen due to inconsistent options between QuantumGraph generation" 

477 f" and execution") from exc 

478 butler.put(metadata, ref[0]) 

479 

480 def writeLogRecords(self, quantum, taskDef, butler, store): 

481 # If we are logging to an external file we must always try to 

482 # close it. 

483 filename = None 

484 if isinstance(self.log_handler, FileHandler): 

485 filename = self.log_handler.stream.name 

486 self.log_handler.close() 

487 

488 if self.log_handler is not None: 

489 # Remove the handler so we stop accumulating log messages. 

490 logging.getLogger().removeHandler(self.log_handler) 

491 

492 try: 

493 if store and taskDef.logOutputDatasetName is not None and self.log_handler is not None: 

494 # DatasetRef has to be in the Quantum outputs, can lookup by 

495 # name 

496 try: 

497 ref = quantum.outputs[taskDef.logOutputDatasetName] 

498 except LookupError as exc: 

499 raise InvalidQuantumError( 

500 f"Quantum outputs is missing log output dataset type {taskDef.logOutputDatasetName};" 

501 f" this could happen due to inconsistent options between QuantumGraph generation" 

502 f" and execution") from exc 

503 

504 if isinstance(self.log_handler, ButlerLogRecordHandler): 

505 butler.put(self.log_handler.records, ref[0]) 

506 

507 # Clear the records in case the handler is reused. 

508 self.log_handler.records.clear() 

509 else: 

510 assert filename is not None, "Somehow unable to extract filename from file handler" 

511 

512 # Need to ingest this file directly into butler. 

513 dataset = FileDataset(path=filename, refs=ref[0]) 

514 try: 

515 butler.ingest(dataset, transfer="move") 

516 filename = None 

517 except NotImplementedError: 

518 # Some datastores can't receive files (e.g. in-memory 

519 # datastore when testing), we store empty list for 

520 # those just to have a dataset. Alternative is to read 

521 # the file as a ButlerLogRecords object and put it. 

522 _LOG.info("Log records could not be stored in this butler because the" 

523 " datastore can not ingest files, empty record list is stored instead.") 

524 records = ButlerLogRecords.from_records([]) 

525 butler.put(records, ref[0]) 

526 finally: 

527 # remove file if it is not ingested 

528 if filename is not None: 

529 try: 

530 os.remove(filename) 

531 except OSError: 

532 pass 

533 

534 def initGlobals(self, quantum, butler): 

535 """Initialize global state needed for task execution. 

536 

537 Parameters 

538 ---------- 

539 quantum : `~lsst.daf.butler.Quantum` 

540 Single Quantum instance. 

541 butler : `~lsst.daf.butler.Butler` 

542 Data butler. 

543 

544 Notes 

545 ----- 

546 There is an issue with initializing filters singleton which is done 

547 by instrument, to avoid requiring tasks to do it in runQuantum() 

548 we do it here when any dataId has an instrument dimension. Also for 

549 now we only allow single instrument, verify that all instrument 

550 names in all dataIds are identical. 

551 

552 This will need revision when filter singleton disappears. 

553 """ 

554 oneInstrument = None 

555 for datasetRefs in chain(quantum.inputs.values(), quantum.outputs.values()): 

556 for datasetRef in datasetRefs: 

557 dataId = datasetRef.dataId 

558 instrument = dataId.get("instrument") 

559 if instrument is not None: 

560 if oneInstrument is not None: 

561 assert instrument == oneInstrument, \ 

562 "Currently require that only one instrument is used per graph" 

563 else: 

564 oneInstrument = instrument 

565 Instrument.fromName(instrument, butler.registry)