Coverage for python/lsst/ctrl/mpexec/cmdLineFwk.py: 14%

361 statements  

« prev     ^ index     » next       coverage.py v7.2.7, created at 2023-06-14 09:14 +0000

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"""Module defining CmdLineFwk class and related methods. 

23""" 

24 

25from __future__ import annotations 

26 

27__all__ = ["CmdLineFwk"] 

28 

29import atexit 

30import copy 

31import datetime 

32import getpass 

33import logging 

34import shutil 

35from collections.abc import Iterable, Mapping, Sequence 

36from types import SimpleNamespace 

37from typing import TYPE_CHECKING 

38 

39from astropy.table import Table 

40from lsst.daf.butler import ( 

41 Butler, 

42 CollectionType, 

43 DatasetId, 

44 DatasetRef, 

45 DatastoreCacheManager, 

46 QuantumBackedButler, 

47) 

48from lsst.daf.butler.registry import MissingCollectionError, RegistryDefaults 

49from lsst.daf.butler.registry.wildcards import CollectionWildcard 

50from lsst.pipe.base import ( 

51 GraphBuilder, 

52 Instrument, 

53 Pipeline, 

54 PipelineDatasetTypes, 

55 QuantumGraph, 

56 buildExecutionButler, 

57) 

58from lsst.utils import doImportType 

59from lsst.utils.threads import disable_implicit_threading 

60 

61from . import util 

62from .dotTools import graph2dot, pipeline2dot 

63from .executionGraphFixup import ExecutionGraphFixup 

64from .mpGraphExecutor import MPGraphExecutor 

65from .preExecInit import PreExecInit, PreExecInitLimited 

66from .singleQuantumExecutor import SingleQuantumExecutor 

67 

68if TYPE_CHECKING: 

69 from lsst.daf.butler import ( 

70 Config, 

71 DatasetType, 

72 DatastoreRecordData, 

73 DimensionUniverse, 

74 LimitedButler, 

75 Quantum, 

76 Registry, 

77 ) 

78 from lsst.pipe.base import TaskDef, TaskFactory 

79 

80 

81# ---------------------------------- 

82# Local non-exported definitions -- 

83# ---------------------------------- 

84 

85_LOG = logging.getLogger(__name__) 

86 

87 

88class _OutputChainedCollectionInfo: 

89 """A helper class for handling command-line arguments related to an output 

90 `~lsst.daf.butler.CollectionType.CHAINED` collection. 

91 

92 Parameters 

93 ---------- 

94 registry : `lsst.daf.butler.Registry` 

95 Butler registry that collections will be added to and/or queried from. 

96 name : `str` 

97 Name of the collection given on the command line. 

98 """ 

99 

100 def __init__(self, registry: Registry, name: str): 

101 self.name = name 

102 try: 

103 self.chain = tuple(registry.getCollectionChain(name)) 

104 self.exists = True 

105 except MissingCollectionError: 

106 self.chain = () 

107 self.exists = False 

108 

109 def __str__(self) -> str: 

110 return self.name 

111 

112 name: str 

113 """Name of the collection provided on the command line (`str`). 

114 """ 

115 

116 exists: bool 

117 """Whether this collection already exists in the registry (`bool`). 

118 """ 

119 

120 chain: tuple[str, ...] 

121 """The definition of the collection, if it already exists (`tuple`[`str`]). 

122 

123 Empty if the collection does not already exist. 

124 """ 

125 

126 

127class _OutputRunCollectionInfo: 

128 """A helper class for handling command-line arguments related to an output 

129 `~lsst.daf.butler.CollectionType.RUN` collection. 

130 

131 Parameters 

132 ---------- 

133 registry : `lsst.daf.butler.Registry` 

134 Butler registry that collections will be added to and/or queried from. 

135 name : `str` 

136 Name of the collection given on the command line. 

137 """ 

138 

139 def __init__(self, registry: Registry, name: str): 

140 self.name = name 

141 try: 

142 actualType = registry.getCollectionType(name) 

143 if actualType is not CollectionType.RUN: 

144 raise TypeError(f"Collection '{name}' exists but has type {actualType.name}, not RUN.") 

145 self.exists = True 

146 except MissingCollectionError: 

147 self.exists = False 

148 

149 name: str 

150 """Name of the collection provided on the command line (`str`). 

151 """ 

152 

153 exists: bool 

154 """Whether this collection already exists in the registry (`bool`). 

155 """ 

156 

157 

158class _ButlerFactory: 

159 """A helper class for processing command-line arguments related to input 

160 and output collections. 

161 

162 Parameters 

163 ---------- 

164 registry : `lsst.daf.butler.Registry` 

165 Butler registry that collections will be added to and/or queried from. 

166 

167 args : `types.SimpleNamespace` 

168 Parsed command-line arguments. The following attributes are used, 

169 either at construction or in later methods. 

170 

171 ``output`` 

172 The name of a `~lsst.daf.butler.CollectionType.CHAINED` 

173 input/output collection. 

174 

175 ``output_run`` 

176 The name of a `~lsst.daf.butler.CollectionType.RUN` input/output 

177 collection. 

178 

179 ``extend_run`` 

180 A boolean indicating whether ``output_run`` should already exist 

181 and be extended. 

182 

183 ``replace_run`` 

184 A boolean indicating that (if `True`) ``output_run`` should already 

185 exist but will be removed from the output chained collection and 

186 replaced with a new one. 

187 

188 ``prune_replaced`` 

189 A boolean indicating whether to prune the replaced run (requires 

190 ``replace_run``). 

191 

192 ``inputs`` 

193 Input collections of any type; see 

194 :ref:`daf_butler_ordered_collection_searches` for details. 

195 

196 ``butler_config`` 

197 Path to a data repository root or configuration file. 

198 

199 writeable : `bool` 

200 If `True`, a `~lsst.daf.butler.Butler` is being initialized in a 

201 context where actual writes should happens, and hence no output run 

202 is necessary. 

203 

204 Raises 

205 ------ 

206 ValueError 

207 Raised if ``writeable is True`` but there are no output collections. 

208 """ 

209 

210 def __init__(self, registry: Registry, args: SimpleNamespace, writeable: bool): 

211 if args.output is not None: 

212 self.output = _OutputChainedCollectionInfo(registry, args.output) 

213 else: 

214 self.output = None 

215 if args.output_run is not None: 

216 self.outputRun = _OutputRunCollectionInfo(registry, args.output_run) 

217 elif self.output is not None: 

218 if args.extend_run: 

219 if not self.output.chain: 

220 raise ValueError("Cannot use --extend-run option with non-existing or empty output chain") 

221 runName = self.output.chain[0] 

222 else: 

223 runName = "{}/{}".format(self.output, Instrument.makeCollectionTimestamp()) 

224 self.outputRun = _OutputRunCollectionInfo(registry, runName) 

225 elif not writeable: 

226 # If we're not writing yet, ok to have no output run. 

227 self.outputRun = None 

228 else: 

229 raise ValueError("Cannot write without at least one of (--output, --output-run).") 

230 # Recursively flatten any input CHAINED collections. We do this up 

231 # front so we can tell if the user passes the same inputs on subsequent 

232 # calls, even though we also flatten when we define the output CHAINED 

233 # collection. 

234 self.inputs = tuple(registry.queryCollections(args.input, flattenChains=True)) if args.input else () 

235 

236 def check(self, args: SimpleNamespace) -> None: 

237 """Check command-line options for consistency with each other and the 

238 data repository. 

239 

240 Parameters 

241 ---------- 

242 args : `types.SimpleNamespace` 

243 Parsed command-line arguments. See class documentation for the 

244 construction parameter of the same name. 

245 """ 

246 assert not (args.extend_run and args.replace_run), "In mutually-exclusive group in ArgumentParser." 

247 if self.inputs and self.output is not None and self.output.exists: 

248 # Passing the same inputs that were used to initialize the output 

249 # collection is allowed; this means they must _end_ with the same 

250 # collections, because we push new runs to the front of the chain. 

251 for c1, c2 in zip(self.inputs[::-1], self.output.chain[::-1]): 

252 if c1 != c2: 

253 raise ValueError( 

254 f"Output CHAINED collection {self.output.name!r} exists, but it ends with " 

255 "a different sequence of input collections than those given: " 

256 f"{c1!r} != {c2!r} in inputs={self.inputs} vs " 

257 f"{self.output.name}={self.output.chain}." 

258 ) 

259 if len(self.inputs) > len(self.output.chain): 

260 nNew = len(self.inputs) - len(self.output.chain) 

261 raise ValueError( 

262 f"Cannot add new input collections {self.inputs[:nNew]} after " 

263 "output collection is first created." 

264 ) 

265 if args.extend_run: 

266 if self.outputRun is None: 

267 raise ValueError("Cannot --extend-run when no output collection is given.") 

268 elif not self.outputRun.exists: 

269 raise ValueError( 

270 f"Cannot --extend-run; output collection '{self.outputRun.name}' does not exist." 

271 ) 

272 if not args.extend_run and self.outputRun is not None and self.outputRun.exists: 

273 raise ValueError( 

274 f"Output run '{self.outputRun.name}' already exists, but --extend-run was not given." 

275 ) 

276 if args.prune_replaced and not args.replace_run: 

277 raise ValueError("--prune-replaced requires --replace-run.") 

278 if args.replace_run and (self.output is None or not self.output.exists): 

279 raise ValueError("--output must point to an existing CHAINED collection for --replace-run.") 

280 

281 @classmethod 

282 def _makeReadParts(cls, args: SimpleNamespace) -> tuple[Butler, Sequence[str], _ButlerFactory]: 

283 """Parse arguments to support implementations of `makeReadButler` and 

284 `makeButlerAndCollections`. 

285 

286 Parameters 

287 ---------- 

288 args : `types.SimpleNamespace` 

289 Parsed command-line arguments. See class documentation for the 

290 construction parameter of the same name. 

291 

292 Returns 

293 ------- 

294 butler : `lsst.daf.butler.Butler` 

295 A read-only butler constructed from the repo at 

296 ``args.butler_config``, but with no default collections. 

297 inputs : `Sequence` [ `str` ] 

298 A collection search path constructed according to ``args``. 

299 self : `_ButlerFactory` 

300 A new `_ButlerFactory` instance representing the processed version 

301 of ``args``. 

302 """ 

303 butler = Butler(args.butler_config, writeable=False) 

304 self = cls(butler.registry, args, writeable=False) 

305 self.check(args) 

306 if self.output and self.output.exists: 

307 if args.replace_run: 

308 replaced = self.output.chain[0] 

309 inputs = list(self.output.chain[1:]) 

310 _LOG.debug( 

311 "Simulating collection search in '%s' after removing '%s'.", self.output.name, replaced 

312 ) 

313 else: 

314 inputs = [self.output.name] 

315 else: 

316 inputs = list(self.inputs) 

317 if args.extend_run: 

318 assert self.outputRun is not None, "Output collection has to be specified." 

319 inputs.insert(0, self.outputRun.name) 

320 collSearch = CollectionWildcard.from_expression(inputs).require_ordered() 

321 return butler, collSearch, self 

322 

323 @classmethod 

324 def makeReadButler(cls, args: SimpleNamespace) -> Butler: 

325 """Construct a read-only butler according to the given command-line 

326 arguments. 

327 

328 Parameters 

329 ---------- 

330 args : `types.SimpleNamespace` 

331 Parsed command-line arguments. See class documentation for the 

332 construction parameter of the same name. 

333 

334 Returns 

335 ------- 

336 butler : `lsst.daf.butler.Butler` 

337 A read-only butler initialized with the collections specified by 

338 ``args``. 

339 """ 

340 cls.defineDatastoreCache() # Ensure that this butler can use a shared cache. 

341 butler, inputs, _ = cls._makeReadParts(args) 

342 _LOG.debug("Preparing butler to read from %s.", inputs) 

343 return Butler(butler=butler, collections=inputs) 

344 

345 @classmethod 

346 def makeButlerAndCollections(cls, args: SimpleNamespace) -> tuple[Butler, Sequence[str], str | None]: 

347 """Return a read-only registry, a collection search path, and the name 

348 of the run to be used for future writes. 

349 

350 Parameters 

351 ---------- 

352 args : `types.SimpleNamespace` 

353 Parsed command-line arguments. See class documentation for the 

354 construction parameter of the same name. 

355 

356 Returns 

357 ------- 

358 butler : `lsst.daf.butler.Butler` 

359 A read-only butler that collections will be added to and/or queried 

360 from. 

361 inputs : `Sequence` [ `str` ] 

362 Collections to search for datasets. 

363 run : `str` or `None` 

364 Name of the output `~lsst.daf.butler.CollectionType.RUN` collection 

365 if it already exists, or `None` if it does not. 

366 """ 

367 butler, inputs, self = cls._makeReadParts(args) 

368 run: str | None = None 

369 if args.extend_run: 

370 assert self.outputRun is not None, "Output collection has to be specified." 

371 if self.outputRun is not None: 

372 run = self.outputRun.name 

373 _LOG.debug("Preparing registry to read from %s and expect future writes to '%s'.", inputs, run) 

374 return butler, inputs, run 

375 

376 @staticmethod 

377 def defineDatastoreCache() -> None: 

378 """Define where datastore cache directories should be found. 

379 

380 Notes 

381 ----- 

382 All the jobs should share a datastore cache if applicable. This 

383 method asks for a shared fallback cache to be defined and then 

384 configures an exit handler to clean it up. 

385 """ 

386 defined, cache_dir = DatastoreCacheManager.set_fallback_cache_directory_if_unset() 

387 if defined: 

388 atexit.register(shutil.rmtree, cache_dir, ignore_errors=True) 

389 _LOG.debug("Defining shared datastore cache directory to %s", cache_dir) 

390 

391 @classmethod 

392 def makeWriteButler(cls, args: SimpleNamespace, taskDefs: Iterable[TaskDef] | None = None) -> Butler: 

393 """Return a read-write butler initialized to write to and read from 

394 the collections specified by the given command-line arguments. 

395 

396 Parameters 

397 ---------- 

398 args : `types.SimpleNamespace` 

399 Parsed command-line arguments. See class documentation for the 

400 construction parameter of the same name. 

401 taskDefs : iterable of `TaskDef`, optional 

402 Definitions for tasks in a pipeline. This argument is only needed 

403 if ``args.replace_run`` is `True` and ``args.prune_replaced`` is 

404 "unstore". 

405 

406 Returns 

407 ------- 

408 butler : `lsst.daf.butler.Butler` 

409 A read-write butler initialized according to the given arguments. 

410 """ 

411 cls.defineDatastoreCache() # Ensure that this butler can use a shared cache. 

412 butler = Butler(args.butler_config, writeable=True) 

413 self = cls(butler.registry, args, writeable=True) 

414 self.check(args) 

415 assert self.outputRun is not None, "Output collection has to be specified." # for mypy 

416 if self.output is not None: 

417 chainDefinition = list(self.output.chain if self.output.exists else self.inputs) 

418 if args.replace_run: 

419 replaced = chainDefinition.pop(0) 

420 if args.prune_replaced == "unstore": 

421 # Remove datasets from datastore 

422 with butler.transaction(): 

423 refs: Iterable[DatasetRef] = butler.registry.queryDatasets(..., collections=replaced) 

424 # we want to remove regular outputs but keep 

425 # initOutputs, configs, and versions. 

426 if taskDefs is not None: 

427 initDatasetNames = set(PipelineDatasetTypes.initOutputNames(taskDefs)) 

428 refs = [ref for ref in refs if ref.datasetType.name not in initDatasetNames] 

429 butler.pruneDatasets(refs, unstore=True, disassociate=False) 

430 elif args.prune_replaced == "purge": 

431 # Erase entire collection and all datasets, need to remove 

432 # collection from its chain collection first. 

433 with butler.transaction(): 

434 butler.registry.setCollectionChain(self.output.name, chainDefinition, flatten=True) 

435 butler.removeRuns([replaced], unstore=True) 

436 elif args.prune_replaced is not None: 

437 raise NotImplementedError(f"Unsupported --prune-replaced option '{args.prune_replaced}'.") 

438 if not self.output.exists: 

439 butler.registry.registerCollection(self.output.name, CollectionType.CHAINED) 

440 if not args.extend_run: 

441 butler.registry.registerCollection(self.outputRun.name, CollectionType.RUN) 

442 chainDefinition.insert(0, self.outputRun.name) 

443 butler.registry.setCollectionChain(self.output.name, chainDefinition, flatten=True) 

444 _LOG.debug( 

445 "Preparing butler to write to '%s' and read from '%s'=%s", 

446 self.outputRun.name, 

447 self.output.name, 

448 chainDefinition, 

449 ) 

450 butler.registry.defaults = RegistryDefaults(run=self.outputRun.name, collections=self.output.name) 

451 else: 

452 inputs = (self.outputRun.name,) + self.inputs 

453 _LOG.debug("Preparing butler to write to '%s' and read from %s.", self.outputRun.name, inputs) 

454 butler.registry.defaults = RegistryDefaults(run=self.outputRun.name, collections=inputs) 

455 return butler 

456 

457 output: _OutputChainedCollectionInfo | None 

458 """Information about the output chained collection, if there is or will be 

459 one (`_OutputChainedCollectionInfo` or `None`). 

460 """ 

461 

462 outputRun: _OutputRunCollectionInfo | None 

463 """Information about the output run collection, if there is or will be 

464 one (`_OutputRunCollectionInfo` or `None`). 

465 """ 

466 

467 inputs: tuple[str, ...] 

468 """Input collections provided directly by the user (`tuple` [ `str` ]). 

469 """ 

470 

471 

472class _QBBFactory: 

473 """Class which is a callable for making QBB instances.""" 

474 

475 def __init__( 

476 self, butler_config: Config, dimensions: DimensionUniverse, dataset_types: Mapping[str, DatasetType] 

477 ): 

478 self.butler_config = butler_config 

479 self.dimensions = dimensions 

480 self.dataset_types = dataset_types 

481 

482 def __call__(self, quantum: Quantum) -> LimitedButler: 

483 """Return freshly initialized `~lsst.daf.butler.QuantumBackedButler`. 

484 

485 Factory method to create QuantumBackedButler instances. 

486 """ 

487 return QuantumBackedButler.initialize( 

488 config=self.butler_config, 

489 quantum=quantum, 

490 dimensions=self.dimensions, 

491 dataset_types=self.dataset_types, 

492 ) 

493 

494 

495# ------------------------ 

496# Exported definitions -- 

497# ------------------------ 

498 

499 

500class CmdLineFwk: 

501 """PipelineTask framework which executes tasks from command line. 

502 

503 In addition to executing tasks this activator provides additional methods 

504 for task management like dumping configuration or execution chain. 

505 """ 

506 

507 MP_TIMEOUT = 3600 * 24 * 30 # Default timeout (sec) for multiprocessing 

508 

509 def __init__(self) -> None: 

510 pass 

511 

512 def makePipeline(self, args: SimpleNamespace) -> Pipeline: 

513 """Build a pipeline from command line arguments. 

514 

515 Parameters 

516 ---------- 

517 args : `types.SimpleNamespace` 

518 Parsed command line 

519 

520 Returns 

521 ------- 

522 pipeline : `~lsst.pipe.base.Pipeline` 

523 """ 

524 if args.pipeline: 

525 pipeline = Pipeline.from_uri(args.pipeline) 

526 else: 

527 pipeline = Pipeline("anonymous") 

528 

529 # loop over all pipeline actions and apply them in order 

530 for action in args.pipeline_actions: 

531 if action.action == "add_instrument": 

532 pipeline.addInstrument(action.value) 

533 

534 elif action.action == "new_task": 

535 pipeline.addTask(action.value, action.label) 

536 

537 elif action.action == "delete_task": 

538 pipeline.removeTask(action.label) 

539 

540 elif action.action == "config": 

541 # action value string is "field=value", split it at '=' 

542 field, _, value = action.value.partition("=") 

543 pipeline.addConfigOverride(action.label, field, value) 

544 

545 elif action.action == "configfile": 

546 pipeline.addConfigFile(action.label, action.value) 

547 

548 else: 

549 raise ValueError(f"Unexpected pipeline action: {action.action}") 

550 

551 if args.save_pipeline: 

552 pipeline.write_to_uri(args.save_pipeline) 

553 

554 if args.pipeline_dot: 

555 pipeline2dot(pipeline, args.pipeline_dot) 

556 

557 return pipeline 

558 

559 def makeGraph(self, pipeline: Pipeline, args: SimpleNamespace) -> QuantumGraph | None: 

560 """Build a graph from command line arguments. 

561 

562 Parameters 

563 ---------- 

564 pipeline : `~lsst.pipe.base.Pipeline` 

565 Pipeline, can be empty or ``None`` if graph is read from a file. 

566 args : `types.SimpleNamespace` 

567 Parsed command line 

568 

569 Returns 

570 ------- 

571 graph : `~lsst.pipe.base.QuantumGraph` or `None` 

572 If resulting graph is empty then `None` is returned. 

573 """ 

574 # make sure that --extend-run always enables --skip-existing 

575 if args.extend_run: 

576 args.skip_existing = True 

577 

578 butler, collections, run = _ButlerFactory.makeButlerAndCollections(args) 

579 

580 if args.skip_existing and run: 

581 args.skip_existing_in += (run,) 

582 

583 if args.qgraph: 

584 # click passes empty tuple as default value for qgraph_node_id 

585 nodes = args.qgraph_node_id or None 

586 qgraph = QuantumGraph.loadUri(args.qgraph, butler.dimensions, nodes=nodes, graphID=args.qgraph_id) 

587 

588 # pipeline can not be provided in this case 

589 if pipeline: 

590 raise ValueError("Pipeline must not be given when quantum graph is read from file.") 

591 if args.show_qgraph_header: 

592 print(QuantumGraph.readHeader(args.qgraph)) 

593 else: 

594 task_defs = list(pipeline.toExpandedPipeline()) 

595 if args.mock: 

596 from lsst.pipe.base.tests.mocks import mock_task_defs 

597 

598 task_defs = mock_task_defs(task_defs, unmocked_dataset_types=args.unmocked_dataset_types) 

599 # make execution plan (a.k.a. DAG) for pipeline 

600 graphBuilder = GraphBuilder( 

601 butler.registry, 

602 skipExistingIn=args.skip_existing_in, 

603 clobberOutputs=args.clobber_outputs, 

604 datastore=butler.datastore if args.qgraph_datastore_records else None, 

605 ) 

606 # accumulate metadata 

607 metadata = { 

608 "input": args.input, 

609 "output": args.output, 

610 "butler_argument": args.butler_config, 

611 "output_run": run, 

612 "extend_run": args.extend_run, 

613 "skip_existing_in": args.skip_existing_in, 

614 "skip_existing": args.skip_existing, 

615 "data_query": args.data_query, 

616 "user": getpass.getuser(), 

617 "time": f"{datetime.datetime.now()}", 

618 } 

619 assert run is not None, "Butler output run collection must be defined" 

620 qgraph = graphBuilder.makeGraph( 

621 task_defs, 

622 collections, 

623 run, 

624 args.data_query, 

625 metadata=metadata, 

626 datasetQueryConstraint=args.dataset_query_constraint, 

627 dataId=pipeline.get_data_id(butler.dimensions), 

628 ) 

629 if args.show_qgraph_header: 

630 qgraph.buildAndPrintHeader() 

631 

632 # Count quanta in graph; give a warning if it's empty and return None. 

633 nQuanta = len(qgraph) 

634 if nQuanta == 0: 

635 return None 

636 else: 

637 if _LOG.isEnabledFor(logging.INFO): 

638 qg_task_table = self._generateTaskTable(qgraph) 

639 qg_task_table_formatted = "\n".join(qg_task_table.pformat_all()) 

640 _LOG.info( 

641 "QuantumGraph contains %d quanta for %d tasks, graph ID: %r\n%s", 

642 nQuanta, 

643 len(qgraph.taskGraph), 

644 qgraph.graphID, 

645 qg_task_table_formatted, 

646 ) 

647 

648 if args.save_qgraph: 

649 qgraph.saveUri(args.save_qgraph) 

650 

651 if args.save_single_quanta: 

652 for quantumNode in qgraph: 

653 sqgraph = qgraph.subset(quantumNode) 

654 uri = args.save_single_quanta.format(quantumNode) 

655 sqgraph.saveUri(uri) 

656 

657 if args.qgraph_dot: 

658 graph2dot(qgraph, args.qgraph_dot) 

659 

660 if args.execution_butler_location: 

661 butler = Butler(args.butler_config) 

662 newArgs = copy.deepcopy(args) 

663 

664 def builderShim(butler: Butler) -> Butler: 

665 newArgs.butler_config = butler._config 

666 # Calling makeWriteButler is done for the side effects of 

667 # calling that method, maining parsing all the args into 

668 # collection names, creating collections, etc. 

669 newButler = _ButlerFactory.makeWriteButler(newArgs) 

670 return newButler 

671 

672 # Include output collection in collections for input 

673 # files if it exists in the repo. 

674 all_inputs = args.input 

675 if args.output is not None: 

676 try: 

677 all_inputs += (next(iter(butler.registry.queryCollections(args.output))),) 

678 except MissingCollectionError: 

679 pass 

680 

681 _LOG.debug("Calling buildExecutionButler with collections=%s", all_inputs) 

682 buildExecutionButler( 

683 butler, 

684 qgraph, 

685 args.execution_butler_location, 

686 run, 

687 butlerModifier=builderShim, 

688 collections=all_inputs, 

689 clobber=args.clobber_execution_butler, 

690 datastoreRoot=args.target_datastore_root, 

691 transfer=args.transfer, 

692 ) 

693 

694 return qgraph 

695 

696 def runPipeline( 

697 self, 

698 graph: QuantumGraph, 

699 taskFactory: TaskFactory, 

700 args: SimpleNamespace, 

701 butler: Butler | None = None, 

702 ) -> None: 

703 """Execute complete QuantumGraph. 

704 

705 Parameters 

706 ---------- 

707 graph : `~lsst.pipe.base.QuantumGraph` 

708 Execution graph. 

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

710 Task factory 

711 args : `types.SimpleNamespace` 

712 Parsed command line 

713 butler : `~lsst.daf.butler.Butler`, optional 

714 Data Butler instance, if not defined then new instance is made 

715 using command line options. 

716 """ 

717 # Check that output run defined on command line is consistent with 

718 # quantum graph. 

719 if args.output_run and graph.metadata: 

720 graph_output_run = graph.metadata.get("output_run", args.output_run) 

721 if graph_output_run != args.output_run: 

722 raise ValueError( 

723 f"Output run defined on command line ({args.output_run}) has to be " 

724 f"identical to graph metadata ({graph_output_run}). " 

725 "To update graph metadata run `pipetask update-graph-run` command." 

726 ) 

727 

728 # Make sure that --extend-run always enables --skip-existing, 

729 # clobbering should be disabled if --extend-run is not specified. 

730 if args.extend_run: 

731 args.skip_existing = True 

732 else: 

733 args.clobber_outputs = False 

734 

735 if not args.enable_implicit_threading: 

736 disable_implicit_threading() 

737 

738 # Make butler instance. QuantumGraph should have an output run defined, 

739 # but we ignore it here and let command line decide actual output run. 

740 if butler is None: 

741 butler = _ButlerFactory.makeWriteButler(args, graph.iterTaskGraph()) 

742 

743 if args.skip_existing: 

744 args.skip_existing_in += (butler.run,) 

745 

746 # Enable lsstDebug debugging. Note that this is done once in the 

747 # main process before PreExecInit and it is also repeated before 

748 # running each task in SingleQuantumExecutor (which may not be 

749 # needed if `multipocessing` always uses fork start method). 

750 if args.enableLsstDebug: 

751 try: 

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

753 import debug # type: ignore # noqa:F401 

754 except ImportError: 

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

756 

757 # Save all InitOutputs, configs, etc. 

758 preExecInit = PreExecInit(butler, taskFactory, extendRun=args.extend_run) 

759 preExecInit.initialize( 

760 graph, 

761 saveInitOutputs=not args.skip_init_writes, 

762 registerDatasetTypes=args.register_dataset_types, 

763 saveVersions=not args.no_versions, 

764 ) 

765 

766 if not args.init_only: 

767 graphFixup = self._importGraphFixup(args) 

768 quantumExecutor = SingleQuantumExecutor( 

769 butler, 

770 taskFactory, 

771 skipExistingIn=args.skip_existing_in, 

772 clobberOutputs=args.clobber_outputs, 

773 enableLsstDebug=args.enableLsstDebug, 

774 exitOnKnownError=args.fail_fast, 

775 ) 

776 

777 timeout = self.MP_TIMEOUT if args.timeout is None else args.timeout 

778 executor = MPGraphExecutor( 

779 numProc=args.processes, 

780 timeout=timeout, 

781 startMethod=args.start_method, 

782 quantumExecutor=quantumExecutor, 

783 failFast=args.fail_fast, 

784 pdb=args.pdb, 

785 executionGraphFixup=graphFixup, 

786 ) 

787 # Have to reset connection pool to avoid sharing connections with 

788 # forked processes. 

789 butler.registry.resetConnectionPool() 

790 try: 

791 with util.profile(args.profile, _LOG): 

792 executor.execute(graph) 

793 finally: 

794 if args.summary: 

795 report = executor.getReport() 

796 if report: 

797 with open(args.summary, "w") as out: 

798 # Do not save fields that are not set. 

799 out.write(report.json(exclude_none=True, indent=2)) 

800 

801 def _generateTaskTable(self, qgraph: QuantumGraph) -> Table: 

802 """Generate astropy table listing the number of quanta per task for a 

803 given quantum graph. 

804 

805 Parameters 

806 ---------- 

807 qgraph : `lsst.pipe.base.graph.graph.QuantumGraph` 

808 A QuantumGraph object. 

809 

810 Returns 

811 ------- 

812 qg_task_table : `astropy.table.table.Table` 

813 An astropy table containing columns: Quanta and Tasks. 

814 """ 

815 qg_quanta, qg_tasks = [], [] 

816 for task_def in qgraph.iterTaskGraph(): 

817 num_qnodes = qgraph.getNumberOfQuantaForTask(task_def) 

818 qg_quanta.append(num_qnodes) 

819 qg_tasks.append(task_def.label) 

820 qg_task_table = Table(dict(Quanta=qg_quanta, Tasks=qg_tasks)) 

821 return qg_task_table 

822 

823 def _importGraphFixup(self, args: SimpleNamespace) -> ExecutionGraphFixup | None: 

824 """Import/instantiate graph fixup object. 

825 

826 Parameters 

827 ---------- 

828 args : `types.SimpleNamespace` 

829 Parsed command line. 

830 

831 Returns 

832 ------- 

833 fixup : `ExecutionGraphFixup` or `None` 

834 

835 Raises 

836 ------ 

837 ValueError 

838 Raised if import fails, method call raises exception, or returned 

839 instance has unexpected type. 

840 """ 

841 if args.graph_fixup: 

842 try: 

843 factory = doImportType(args.graph_fixup) 

844 except Exception as exc: 

845 raise ValueError("Failed to import graph fixup class/method") from exc 

846 try: 

847 fixup = factory() 

848 except Exception as exc: 

849 raise ValueError("Failed to make instance of graph fixup") from exc 

850 if not isinstance(fixup, ExecutionGraphFixup): 

851 raise ValueError("Graph fixup is not an instance of ExecutionGraphFixup class") 

852 return fixup 

853 return None 

854 

855 def preExecInitQBB(self, task_factory: TaskFactory, args: SimpleNamespace) -> None: 

856 # Load quantum graph. We do not really need individual Quanta here, 

857 # but we need datastore records for initInputs, and those are only 

858 # available from Quanta, so load the whole thing. 

859 qgraph = QuantumGraph.loadUri(args.qgraph, graphID=args.qgraph_id) 

860 universe = qgraph.universe 

861 

862 # Collect all init input/output dataset IDs. 

863 predicted_inputs: set[DatasetId] = set() 

864 predicted_outputs: set[DatasetId] = set() 

865 for taskDef in qgraph.iterTaskGraph(): 

866 if (refs := qgraph.initInputRefs(taskDef)) is not None: 

867 predicted_inputs.update(ref.id for ref in refs) 

868 if (refs := qgraph.initOutputRefs(taskDef)) is not None: 

869 predicted_outputs.update(ref.id for ref in refs) 

870 predicted_outputs.update(ref.id for ref in qgraph.globalInitOutputRefs()) 

871 # remove intermediates from inputs 

872 predicted_inputs -= predicted_outputs 

873 

874 # Very inefficient way to extract datastore records from quantum graph, 

875 # we have to scan all quanta and look at their datastore records. 

876 datastore_records: dict[str, DatastoreRecordData] = {} 

877 for quantum_node in qgraph: 

878 for store_name, records in quantum_node.quantum.datastore_records.items(): 

879 subset = records.subset(predicted_inputs) 

880 if subset is not None: 

881 datastore_records.setdefault(store_name, DatastoreRecordData()).update(subset) 

882 

883 dataset_types = {dstype.name: dstype for dstype in qgraph.registryDatasetTypes()} 

884 

885 # Make butler from everything. 

886 butler = QuantumBackedButler.from_predicted( 

887 config=args.butler_config, 

888 predicted_inputs=predicted_inputs, 

889 predicted_outputs=predicted_outputs, 

890 dimensions=universe, 

891 datastore_records=datastore_records, 

892 search_paths=args.config_search_path, 

893 dataset_types=dataset_types, 

894 ) 

895 

896 # Save all InitOutputs, configs, etc. 

897 preExecInit = PreExecInitLimited(butler, task_factory) 

898 preExecInit.initialize(qgraph) 

899 

900 def runGraphQBB(self, task_factory: TaskFactory, args: SimpleNamespace) -> None: 

901 # Load quantum graph. 

902 nodes = args.qgraph_node_id or None 

903 qgraph = QuantumGraph.loadUri(args.qgraph, nodes=nodes, graphID=args.qgraph_id) 

904 

905 if qgraph.metadata is None: 

906 raise ValueError("QuantumGraph is missing metadata, cannot ") 

907 

908 dataset_types = {dstype.name: dstype for dstype in qgraph.registryDatasetTypes()} 

909 

910 _butler_factory = _QBBFactory( 

911 butler_config=args.butler_config, 

912 dimensions=qgraph.universe, 

913 dataset_types=dataset_types, 

914 ) 

915 

916 # make special quantum executor 

917 quantumExecutor = SingleQuantumExecutor( 

918 butler=None, 

919 taskFactory=task_factory, 

920 enableLsstDebug=args.enableLsstDebug, 

921 exitOnKnownError=args.fail_fast, 

922 limited_butler_factory=_butler_factory, 

923 ) 

924 

925 timeout = self.MP_TIMEOUT if args.timeout is None else args.timeout 

926 executor = MPGraphExecutor( 

927 numProc=args.processes, 

928 timeout=timeout, 

929 startMethod=args.start_method, 

930 quantumExecutor=quantumExecutor, 

931 failFast=args.fail_fast, 

932 pdb=args.pdb, 

933 ) 

934 try: 

935 with util.profile(args.profile, _LOG): 

936 executor.execute(qgraph) 

937 finally: 

938 if args.summary: 

939 report = executor.getReport() 

940 if report: 

941 with open(args.summary, "w") as out: 

942 # Do not save fields that are not set. 

943 out.write(report.json(exclude_none=True, indent=2))