Coverage for python/lsst/ctrl/bps/transform.py: 8%

334 statements  

« prev     ^ index     » next       coverage.py v6.4.1, created at 2022-06-15 02:15 -0700

1# This file is part of ctrl_bps. 

2# 

3# Developed for the LSST Data Management System. 

4# This product includes software developed by the LSST Project 

5# (https://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 <https://www.gnu.org/licenses/>. 

21 

22"""Driver for the transformation of a QuantumGraph into a generic workflow. 

23""" 

24 

25import copy 

26import dataclasses 

27import logging 

28import math 

29import os 

30import re 

31 

32from lsst.utils.logging import VERBOSE 

33from lsst.utils.timer import time_this, timeMethod 

34 

35from . import ( 

36 DEFAULT_MEM_RETRIES, 

37 BpsConfig, 

38 GenericWorkflow, 

39 GenericWorkflowExec, 

40 GenericWorkflowFile, 

41 GenericWorkflowJob, 

42) 

43from .bps_utils import ( 

44 WhenToSaveQuantumGraphs, 

45 _create_execution_butler, 

46 create_job_quantum_graph_filename, 

47 save_qg_subgraph, 

48) 

49 

50# All available job attributes. 

51_ATTRS_ALL = frozenset([field.name for field in dataclasses.fields(GenericWorkflowJob)]) 

52 

53# Job attributes that need to be set to their maximal value in the cluster. 

54_ATTRS_MAX = frozenset( 

55 { 

56 "memory_multiplier", 

57 "number_of_retries", 

58 "request_cpus", 

59 "request_memory", 

60 "request_memory_max", 

61 } 

62) 

63 

64# Job attributes that need to be set to sum of their values in the cluster. 

65_ATTRS_SUM = frozenset( 

66 { 

67 "request_disk", 

68 "request_walltime", 

69 } 

70) 

71 

72# Job attributes do not fall into a specific category 

73_ATTRS_MISC = frozenset( 

74 { 

75 "cmdline", 

76 "cmdvals", 

77 "environment", 

78 "pre_cmdline", 

79 "post_cmdline", 

80 "profile", 

81 "attrs", 

82 } 

83) 

84 

85# Attributes that need to be the same for each quanta in the cluster. 

86_ATTRS_UNIVERSAL = frozenset(_ATTRS_ALL - (_ATTRS_MAX | _ATTRS_MISC | _ATTRS_SUM)) 

87 

88_LOG = logging.getLogger(__name__) 

89 

90 

91@timeMethod(logger=_LOG, logLevel=VERBOSE) 

92def transform(config, cqgraph, prefix): 

93 """Transform a ClusteredQuantumGraph to a GenericWorkflow. 

94 

95 Parameters 

96 ---------- 

97 config : `lsst.ctrl.bps.BpsConfig` 

98 BPS configuration. 

99 cqgraph : `lsst.ctrl.bps.ClusteredQuantumGraph` 

100 A clustered quantum graph to transform into a generic workflow. 

101 prefix : `str` 

102 Root path for any output files. 

103 

104 Returns 

105 ------- 

106 generic_workflow : `lsst.ctrl.bps.GenericWorkflow` 

107 The generic workflow transformed from the clustered quantum graph. 

108 generic_workflow_config : `lsst.ctrl.bps.BpsConfig` 

109 Configuration to accompany GenericWorkflow. 

110 """ 

111 _, when_create = config.search(".executionButler.whenCreate") 

112 if when_create.upper() == "TRANSFORM": 

113 _, execution_butler_dir = config.search(".bps_defined.executionButlerDir") 

114 _LOG.info("Creating execution butler in '%s'", execution_butler_dir) 

115 with time_this(log=_LOG, level=logging.INFO, prefix=None, msg="Creating execution butler completed"): 

116 _create_execution_butler(config, config["runQgraphFile"], execution_butler_dir, prefix) 

117 

118 if cqgraph.name is not None: 

119 name = cqgraph.name 

120 else: 

121 _, name = config.search("uniqProcName", opt={"required": True}) 

122 

123 generic_workflow = create_generic_workflow(config, cqgraph, name, prefix) 

124 generic_workflow_config = create_generic_workflow_config(config, prefix) 

125 

126 return generic_workflow, generic_workflow_config 

127 

128 

129def add_workflow_init_nodes(config, qgraph, generic_workflow): 

130 """Add nodes to workflow graph that perform initialization steps. 

131 

132 Assumes that all of the initialization should be executed prior to any 

133 of the current workflow. 

134 

135 Parameters 

136 ---------- 

137 config : `lsst.ctrl.bps.BpsConfig` 

138 BPS configuration. 

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

140 The quantum graph the generic workflow represents. 

141 generic_workflow : `lsst.ctrl.bps.GenericWorkflow` 

142 Generic workflow to which the initialization steps should be added. 

143 """ 

144 # Create a workflow graph that will have task and file nodes necessary for 

145 # initializing the pipeline execution 

146 init_workflow = create_init_workflow(config, qgraph, generic_workflow.get_file("runQgraphFile")) 

147 _LOG.debug("init_workflow nodes = %s", init_workflow.nodes()) 

148 generic_workflow.add_workflow_source(init_workflow) 

149 

150 

151def create_init_workflow(config, qgraph, qgraph_gwfile): 

152 """Create workflow for running initialization job(s). 

153 

154 Parameters 

155 ---------- 

156 config : `lsst.ctrl.bps.BpsConfig` 

157 BPS configuration. 

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

159 The quantum graph the generic workflow represents. 

160 qgraph_gwfile : `lsst.ctrl.bps.GenericWorkflowFile` 

161 File object for the full run QuantumGraph file. 

162 

163 Returns 

164 ------- 

165 init_workflow : `lsst.ctrl.bps.GenericWorkflow` 

166 GenericWorkflow consisting of job(s) to initialize workflow. 

167 """ 

168 _LOG.debug("creating init subgraph") 

169 _LOG.debug("creating init task input(s)") 

170 search_opt = { 

171 "curvals": {"curr_pipetask": "pipetaskInit"}, 

172 "replaceVars": False, 

173 "expandEnvVars": False, 

174 "replaceEnvVars": True, 

175 "required": False, 

176 } 

177 found, value = config.search("computeSite", opt=search_opt) 

178 if found: 

179 search_opt["curvals"]["curr_site"] = value 

180 found, value = config.search("computeCloud", opt=search_opt) 

181 if found: 

182 search_opt["curvals"]["curr_cloud"] = value 

183 

184 init_workflow = GenericWorkflow("init") 

185 init_workflow.add_file(qgraph_gwfile) 

186 

187 # create job for executing --init-only 

188 gwjob = GenericWorkflowJob("pipetaskInit") 

189 

190 job_values = _get_job_values(config, search_opt, "runQuantumCommand") 

191 job_values["name"] = "pipetaskInit" 

192 job_values["label"] = "pipetaskInit" 

193 

194 # Adjust job attributes values if necessary. 

195 _handle_job_values(job_values, gwjob) 

196 

197 # Pick a node id for each task (not quantum!) to avoid reading the entire 

198 # quantum graph during the initialization stage. 

199 node_ids = [] 

200 for task in qgraph.iterTaskGraph(): 

201 task_def = qgraph.findTaskDefByLabel(task.label) 

202 node = next(iter(qgraph.getNodesForTask(task_def))) 

203 node_ids.append(node.nodeId) 

204 gwjob.cmdvals["qgraphId"] = qgraph.graphID 

205 gwjob.cmdvals["qgraphNodeId"] = ",".join(sorted([f"{node_id}" for node_id in node_ids])) 

206 

207 init_workflow.add_job(gwjob) 

208 

209 # Lookup butler values 

210 _, when_create = config.search(".executionButler.whenCreate", opt=search_opt) 

211 _, butler_config = config.search("butlerConfig", opt=search_opt) 

212 _, execution_butler_dir = config.search(".bps_defined.executionButlerDir", opt=search_opt) 

213 prefix = config["submitPath"] 

214 butler_gwfile = _get_butler_gwfile(prefix, when_create, butler_config, execution_butler_dir) 

215 

216 init_workflow.add_job_inputs(gwjob.name, [qgraph_gwfile, butler_gwfile]) 

217 _enhance_command(config, init_workflow, gwjob, {}) 

218 

219 return init_workflow 

220 

221 

222def _enhance_command(config, generic_workflow, gwjob, cached_job_values): 

223 """Enhance command line with env and file placeholders 

224 and gather command line values. 

225 

226 Parameters 

227 ---------- 

228 config : `lsst.ctrl.bps.BpsConfig` 

229 BPS configuration. 

230 generic_workflow : `lsst.ctrl.bps.GenericWorkflow` 

231 Generic workflow that contains the job. 

232 gwjob : `lsst.ctrl.bps.GenericWorkflowJob` 

233 Generic workflow job to which the updated executable, arguments, 

234 and values should be saved. 

235 cached_job_values : `dict` [`str`, dict[`str`, `Any`]] 

236 Cached values common across jobs with same label. Updated if values 

237 aren't already saved for given gwjob's label. 

238 """ 

239 _LOG.debug("gwjob given to _enhance_command: %s", gwjob) 

240 

241 search_opt = { 

242 "curvals": {"curr_pipetask": gwjob.label}, 

243 "replaceVars": False, 

244 "expandEnvVars": False, 

245 "replaceEnvVars": True, 

246 "required": False, 

247 } 

248 

249 if gwjob.label not in cached_job_values: 

250 cached_job_values[gwjob.label] = {} 

251 # Allowing whenSaveJobQgraph and useLazyCommands per pipetask label. 

252 key = "whenSaveJobQgraph" 

253 _, when_save = config.search(key, opt=search_opt) 

254 cached_job_values[gwjob.label][key] = WhenToSaveQuantumGraphs[when_save.upper()] 

255 

256 key = "useLazyCommands" 

257 search_opt["default"] = True 

258 _, cached_job_values[gwjob.label][key] = config.search(key, opt=search_opt) 

259 del search_opt["default"] 

260 

261 # Change qgraph variable to match whether using run or per-job qgraph 

262 # Note: these are lookup keys, not actual physical filenames. 

263 if cached_job_values[gwjob.label]["whenSaveJobQgraph"] == WhenToSaveQuantumGraphs.NEVER: 

264 gwjob.arguments = gwjob.arguments.replace("{qgraphFile}", "{runQgraphFile}") 

265 elif gwjob.name == "pipetaskInit": 

266 gwjob.arguments = gwjob.arguments.replace("{qgraphFile}", "{runQgraphFile}") 

267 else: # Needed unique file keys for per-job QuantumGraphs 

268 gwjob.arguments = gwjob.arguments.replace("{qgraphFile}", f"{{qgraphFile_{gwjob.name}}}") 

269 

270 # Replace files with special placeholders 

271 for gwfile in generic_workflow.get_job_inputs(gwjob.name): 

272 gwjob.arguments = gwjob.arguments.replace(f"{{{gwfile.name}}}", f"<FILE:{gwfile.name}>") 

273 for gwfile in generic_workflow.get_job_outputs(gwjob.name): 

274 gwjob.arguments = gwjob.arguments.replace(f"{{{gwfile.name}}}", f"<FILE:{gwfile.name}>") 

275 

276 # Save dict of other values needed to complete command line. 

277 # (Be careful to not replace env variables as they may 

278 # be different in compute job.) 

279 search_opt["replaceVars"] = True 

280 

281 for key in re.findall(r"{([^}]+)}", gwjob.arguments): 

282 if key not in gwjob.cmdvals: 

283 if key not in cached_job_values[gwjob.label]: 

284 _, cached_job_values[gwjob.label][key] = config.search(key, opt=search_opt) 

285 gwjob.cmdvals[key] = cached_job_values[gwjob.label][key] 

286 

287 # backwards compatibility 

288 if not cached_job_values[gwjob.label]["useLazyCommands"]: 

289 if "bpsUseShared" not in cached_job_values[gwjob.label]: 

290 key = "bpsUseShared" 

291 search_opt["default"] = True 

292 _, cached_job_values[gwjob.label][key] = config.search(key, opt=search_opt) 

293 del search_opt["default"] 

294 

295 gwjob.arguments = _fill_arguments( 

296 cached_job_values[gwjob.label]["bpsUseShared"], generic_workflow, gwjob.arguments, gwjob.cmdvals 

297 ) 

298 

299 

300def _fill_arguments(use_shared, generic_workflow, arguments, cmdvals): 

301 """Replace placeholders in command line string in job. 

302 

303 Parameters 

304 ---------- 

305 use_shared : `bool` 

306 Whether using shared filesystem. 

307 generic_workflow : `lsst.ctrl.bps.GenericWorkflow` 

308 Generic workflow containing the job. 

309 arguments : `str` 

310 String containing placeholders. 

311 cmdvals : `dict` [`str`, `Any`] 

312 Any command line values that can be used to replace placeholders. 

313 

314 Returns 

315 ------- 

316 arguments : `str` 

317 Command line with FILE and ENV placeholders replaced. 

318 """ 

319 # Replace file placeholders 

320 for file_key in re.findall(r"<FILE:([^>]+)>", arguments): 

321 gwfile = generic_workflow.get_file(file_key) 

322 if not gwfile.wms_transfer: 

323 # Must assume full URI if in command line and told WMS is not 

324 # responsible for transferring file. 

325 uri = gwfile.src_uri 

326 elif use_shared: 

327 if gwfile.job_shared: 

328 # Have shared filesystems and jobs can share file. 

329 uri = gwfile.src_uri 

330 else: 

331 # Taking advantage of inside knowledge. Not future-proof. 

332 # Temporary fix until have job wrapper that pulls files 

333 # within job. 

334 if gwfile.name == "butlerConfig" and os.path.splitext(gwfile.src_uri)[1] != ".yaml": 

335 uri = "butler.yaml" 

336 else: 

337 uri = os.path.basename(gwfile.src_uri) 

338 else: # Using push transfer 

339 uri = os.path.basename(gwfile.src_uri) 

340 

341 arguments = arguments.replace(f"<FILE:{file_key}>", uri) 

342 

343 # Replace env placeholder with submit-side values 

344 arguments = re.sub(r"<ENV:([^>]+)>", r"$\1", arguments) 

345 arguments = os.path.expandvars(arguments) 

346 

347 # Replace remaining vars 

348 arguments = arguments.format(**cmdvals) 

349 

350 return arguments 

351 

352 

353def _get_butler_gwfile(prefix, when_create, butler_config, execution_butler_dir): 

354 """Get butler location to be used by job. 

355 

356 Parameters 

357 ---------- 

358 prefix : `str` 

359 Root path for any output files. 

360 when_create : `str` 

361 When to create the execution butler used to determine whether job is 

362 using execution butler or not. 

363 butler_config : `str` 

364 Location of central butler repositories config file. 

365 execution_butler_dir : `str` 

366 Location of execution butler repository. 

367 

368 Returns 

369 ------- 

370 gwfile : `lsst.ctrl.bps.GenericWorkflowFile` 

371 Representation of butler location. 

372 """ 

373 if when_create.upper() == "NEVER": 

374 wms_transfer = False 

375 job_access_remote = True 

376 job_shared = True 

377 else: 

378 butler_config = execution_butler_dir 

379 if not butler_config.startswith("/"): 

380 butler_config = f"{prefix}/{butler_config}" 

381 wms_transfer = True 

382 job_access_remote = False 

383 job_shared = False 

384 

385 gwfile = GenericWorkflowFile( 

386 "butlerConfig", 

387 src_uri=butler_config, 

388 wms_transfer=wms_transfer, 

389 job_access_remote=job_access_remote, 

390 job_shared=job_shared, 

391 ) 

392 

393 return gwfile 

394 

395 

396def _get_qgraph_gwfile(config, save_qgraph_per_job, gwjob, run_qgraph_file, prefix): 

397 """Get qgraph location to be used by job. 

398 

399 Parameters 

400 ---------- 

401 config : `lsst.ctrl.bps.BpsConfig` 

402 Bps configuration. 

403 save_qgraph_per_job: `lsst.ctrl.bps.bps_utils.WhenToSaveQuantumGraphs` 

404 What submission stage to save per-job qgraph files (or NEVER) 

405 gwjob : `lsst.ctrl.bps.GenericWorkflowJob` 

406 Job for which determining QuantumGraph file. 

407 run_qgraph_file : `lsst.ctrl.bps.GenericWorkflowFile` 

408 File representation of the full run QuantumGraph. 

409 prefix : `str` 

410 Path prefix for any files written. 

411 

412 Returns 

413 ------- 

414 gwfile : `lsst.ctrl.bps.GenericWorkflowFile` 

415 Representation of butler location (may not include filename). 

416 """ 

417 qgraph_gwfile = None 

418 if save_qgraph_per_job != WhenToSaveQuantumGraphs.NEVER: 

419 qgraph_gwfile = GenericWorkflowFile( 

420 f"qgraphFile_{gwjob.name}", 

421 src_uri=create_job_quantum_graph_filename(config, gwjob, prefix), 

422 wms_transfer=True, 

423 job_access_remote=True, 

424 job_shared=True, 

425 ) 

426 else: 

427 qgraph_gwfile = run_qgraph_file 

428 

429 return qgraph_gwfile 

430 

431 

432def _get_job_values(config, search_opt, cmd_line_key): 

433 """Gather generic workflow job values from the bps config. 

434 

435 Parameters 

436 ---------- 

437 config : `lsst.ctrl.bps.BpsConfig` 

438 Bps configuration. 

439 search_opt : `dict` [`str`, `Any`] 

440 Search options to be used when searching config. 

441 cmd_line_key : `str` or None 

442 Which command line key to search for (e.g., "runQuantumCommand"). 

443 

444 Returns 

445 ------- 

446 job_values : `dict` [ `str`, `Any` ]` 

447 A mapping between job attributes and their values. 

448 """ 

449 _LOG.debug("cmd_line_key=%s, search_opt=%s", cmd_line_key, search_opt) 

450 job_values = {} 

451 for attr in _ATTRS_ALL: 

452 # Variable names in yaml are camel case instead of snake case. 

453 yaml_name = re.sub(r"_(\S)", lambda match: match.group(1).upper(), attr) 

454 found, value = config.search(yaml_name, opt=search_opt) 

455 if found: 

456 job_values[attr] = value 

457 else: 

458 job_values[attr] = None 

459 

460 # If the automatic memory scaling is enabled (i.e. the memory multiplier 

461 # is set and it is a positive number greater than 1.0), adjust number 

462 # of retries when necessary. If the memory multiplier is invalid, disable 

463 # automatic memory scaling. 

464 if job_values["memory_multiplier"] is not None: 

465 if math.ceil(float(job_values["memory_multiplier"])) > 1: 

466 if job_values["number_of_retries"] is None: 

467 job_values["number_of_retries"] = DEFAULT_MEM_RETRIES 

468 else: 

469 job_values["memory_multiplier"] = None 

470 

471 if cmd_line_key: 

472 found, cmdline = config.search(cmd_line_key, opt=search_opt) 

473 # Make sure cmdline isn't None as that could be sent in as a 

474 # default value in search_opt. 

475 if found and cmdline: 

476 cmd, args = cmdline.split(" ", 1) 

477 job_values["executable"] = GenericWorkflowExec(os.path.basename(cmd), cmd, False) 

478 if args: 

479 job_values["arguments"] = args 

480 

481 return job_values 

482 

483 

484def _handle_job_values(quantum_job_values, gwjob, attributes=_ATTRS_ALL): 

485 """Set the job attributes in the cluster to their correct values. 

486 

487 Parameters 

488 ---------- 

489 quantum_job_values : `dict` [`str`, Any] 

490 Job values for running single Quantum. 

491 gwjob : `lsst.ctrl.bps.GenericWorkflowJob` 

492 Generic workflow job in which to store the universal values. 

493 attributes: `Iterable` [`str`], optional 

494 Job attributes to be set in the job following different rules. 

495 The default value is _ATTRS_ALL. 

496 """ 

497 _LOG.debug("Call to _handle_job_values") 

498 _handle_job_values_universal(quantum_job_values, gwjob, attributes) 

499 _handle_job_values_max(quantum_job_values, gwjob, attributes) 

500 _handle_job_values_sum(quantum_job_values, gwjob, attributes) 

501 

502 

503def _handle_job_values_universal(quantum_job_values, gwjob, attributes=_ATTRS_UNIVERSAL): 

504 """Handle job attributes that must have the same value for every quantum 

505 in the cluster. 

506 

507 Parameters 

508 ---------- 

509 quantum_job_values : `dict` [`str`, Any] 

510 Job values for running single Quantum. 

511 gwjob : `lsst.ctrl.bps.GenericWorkflowJob` 

512 Generic workflow job in which to store the universal values. 

513 attributes: `Iterable` [`str`], optional 

514 Job attributes to be set in the job following different rules. 

515 The default value is _ATTRS_UNIVERSAL. 

516 """ 

517 for attr in _ATTRS_UNIVERSAL & set(attributes): 

518 _LOG.debug( 

519 "Handling job %s (job=%s, quantum=%s)", 

520 attr, 

521 getattr(gwjob, attr), 

522 quantum_job_values.get(attr, "MISSING"), 

523 ) 

524 current_value = getattr(gwjob, attr) 

525 try: 

526 quantum_value = quantum_job_values[attr] 

527 except KeyError: 

528 continue 

529 else: 

530 if not current_value: 

531 setattr(gwjob, attr, quantum_value) 

532 elif current_value != quantum_value: 

533 _LOG.error( 

534 "Inconsistent value for %s in Cluster %s Quantum Number %s\n" 

535 "Current cluster value: %s\n" 

536 "Quantum value: %s", 

537 attr, 

538 gwjob.name, 

539 quantum_job_values.get("qgraphNodeId", "MISSING"), 

540 current_value, 

541 quantum_value, 

542 ) 

543 raise RuntimeError(f"Inconsistent value for {attr} in cluster {gwjob.name}.") 

544 

545 

546def _handle_job_values_max(quantum_job_values, gwjob, attributes=_ATTRS_MAX): 

547 """Handle job attributes that should be set to their maximum value in 

548 the in cluster. 

549 

550 Parameters 

551 ---------- 

552 quantum_job_values : `dict` [`str`, `Any`] 

553 Job values for running single Quantum. 

554 gwjob : `lsst.ctrl.bps.GenericWorkflowJob` 

555 Generic workflow job in which to store the aggregate values. 

556 attributes: `Iterable` [`str`], optional 

557 Job attributes to be set in the job following different rules. 

558 The default value is _ATTR_MAX. 

559 """ 

560 for attr in _ATTRS_MAX & set(attributes): 

561 current_value = getattr(gwjob, attr) 

562 try: 

563 quantum_value = quantum_job_values[attr] 

564 except KeyError: 

565 continue 

566 else: 

567 needs_update = False 

568 if current_value is None: 

569 if quantum_value is not None: 

570 needs_update = True 

571 else: 

572 if quantum_value is not None and current_value < quantum_value: 

573 needs_update = True 

574 if needs_update: 

575 setattr(gwjob, attr, quantum_value) 

576 

577 # When updating memory requirements for a job, check if memory 

578 # autoscaling is enabled. If it is, always use the memory 

579 # multiplier and the number of retries which comes with the 

580 # quantum. 

581 # 

582 # Note that as a result, the quantum with the biggest memory 

583 # requirements will determine whether the memory autoscaling 

584 # will be enabled (or disabled) depending on the value of its 

585 # memory multiplier. 

586 if attr == "request_memory": 

587 gwjob.memory_multiplier = quantum_job_values["memory_multiplier"] 

588 if gwjob.memory_multiplier is not None: 

589 gwjob.number_of_retries = quantum_job_values["number_of_retries"] 

590 

591 

592def _handle_job_values_sum(quantum_job_values, gwjob, attributes=_ATTRS_SUM): 

593 """Handle job attributes that are the sum of their values in the cluster. 

594 

595 Parameters 

596 ---------- 

597 quantum_job_values : `dict` [`str`, `Any`] 

598 Job values for running single Quantum. 

599 gwjob : `lsst.ctrl.bps.GenericWorkflowJob` 

600 Generic workflow job in which to store the aggregate values. 

601 attributes: `Iterable` [`str`], optional 

602 Job attributes to be set in the job following different rules. 

603 The default value is _ATTRS_SUM. 

604 """ 

605 for attr in _ATTRS_SUM & set(attributes): 

606 current_value = getattr(gwjob, attr) 

607 if not current_value: 

608 setattr(gwjob, attr, quantum_job_values[attr]) 

609 else: 

610 setattr(gwjob, attr, current_value + quantum_job_values[attr]) 

611 

612 

613def create_generic_workflow(config, cqgraph, name, prefix): 

614 """Create a generic workflow from a ClusteredQuantumGraph such that it 

615 has information needed for WMS (e.g., command lines). 

616 

617 Parameters 

618 ---------- 

619 config : `lsst.ctrl.bps.BpsConfig` 

620 BPS configuration. 

621 cqgraph : `lsst.ctrl.bps.ClusteredQuantumGraph` 

622 ClusteredQuantumGraph for running a specific pipeline on a specific 

623 payload. 

624 name : `str` 

625 Name for the workflow (typically unique). 

626 prefix : `str` 

627 Root path for any output files. 

628 

629 Returns 

630 ------- 

631 generic_workflow : `lsst.ctrl.bps.GenericWorkflow` 

632 Generic workflow for the given ClusteredQuantumGraph + config. 

633 """ 

634 # Determine whether saving per-job QuantumGraph files in the loop. 

635 _, when_save = config.search("whenSaveJobQgraph", {"default": WhenToSaveQuantumGraphs.TRANSFORM.name}) 

636 save_qgraph_per_job = WhenToSaveQuantumGraphs[when_save.upper()] 

637 

638 search_opt = {"replaceVars": False, "expandEnvVars": False, "replaceEnvVars": True, "required": False} 

639 

640 # Lookup butler values once 

641 _, when_create = config.search(".executionButler.whenCreate", opt=search_opt) 

642 _, butler_config = config.search("butlerConfig", opt=search_opt) 

643 _, execution_butler_dir = config.search(".bps_defined.executionButlerDir", opt=search_opt) 

644 

645 generic_workflow = GenericWorkflow(name) 

646 

647 # Save full run QuantumGraph for use by jobs 

648 generic_workflow.add_file( 

649 GenericWorkflowFile( 

650 "runQgraphFile", 

651 src_uri=config["runQgraphFile"], 

652 wms_transfer=True, 

653 job_access_remote=True, 

654 job_shared=True, 

655 ) 

656 ) 

657 

658 # Cache pipetask specific or more generic job values to minimize number 

659 # on config searches. 

660 cached_job_values = {} 

661 cached_pipetask_values = {} 

662 

663 for cluster in cqgraph.clusters(): 

664 _LOG.debug("Loop over clusters: %s, %s", cluster, type(cluster)) 

665 _LOG.debug( 

666 "cqgraph: name=%s, len=%s, label=%s, ids=%s", 

667 cluster.name, 

668 len(cluster.qgraph_node_ids), 

669 cluster.label, 

670 cluster.qgraph_node_ids, 

671 ) 

672 

673 gwjob = GenericWorkflowJob(cluster.name) 

674 

675 # First get job values from cluster or cluster config 

676 search_opt["curvals"] = {"curr_cluster": cluster.label} 

677 found, value = config.search("computeSite", opt=search_opt) 

678 if found: 

679 search_opt["curvals"]["curr_site"] = value 

680 found, value = config.search("computeCloud", opt=search_opt) 

681 if found: 

682 search_opt["curvals"]["curr_cloud"] = value 

683 

684 # If some config values are set for this cluster 

685 if cluster.label not in cached_job_values: 

686 _LOG.debug("config['cluster'][%s] = %s", cluster.label, config["cluster"][cluster.label]) 

687 cached_job_values[cluster.label] = {} 

688 

689 # Allowing whenSaveJobQgraph and useLazyCommands per cluster label. 

690 key = "whenSaveJobQgraph" 

691 _, when_save = config.search(key, opt=search_opt) 

692 cached_job_values[cluster.label][key] = WhenToSaveQuantumGraphs[when_save.upper()] 

693 

694 key = "useLazyCommands" 

695 search_opt["default"] = True 

696 _, cached_job_values[cluster.label][key] = config.search(key, opt=search_opt) 

697 del search_opt["default"] 

698 

699 if cluster.label in config["cluster"]: 

700 # Don't want to get global defaults here so only look in 

701 # cluster section. 

702 cached_job_values[cluster.label].update( 

703 _get_job_values(config["cluster"][cluster.label], search_opt, "runQuantumCommand") 

704 ) 

705 cluster_job_values = copy.copy(cached_job_values[cluster.label]) 

706 

707 cluster_job_values["name"] = cluster.name 

708 cluster_job_values["label"] = cluster.label 

709 cluster_job_values["quanta_counts"] = cluster.quanta_counts 

710 cluster_job_values["tags"] = cluster.tags 

711 _LOG.debug("cluster_job_values = %s", cluster_job_values) 

712 _handle_job_values(cluster_job_values, gwjob, cluster_job_values.keys()) 

713 

714 # For purposes of whether to continue searching for a value is whether 

715 # the value evaluates to False. 

716 unset_attributes = {attr for attr in _ATTRS_ALL if not getattr(gwjob, attr)} 

717 

718 _LOG.debug("unset_attributes=%s", unset_attributes) 

719 _LOG.debug("set=%s", _ATTRS_ALL - unset_attributes) 

720 

721 # For job info not defined at cluster level, attempt to get job info 

722 # either common or aggregate for all Quanta in cluster. 

723 for node_id in iter(cluster.qgraph_node_ids): 

724 _LOG.debug("node_id=%s", node_id) 

725 qnode = cqgraph.get_quantum_node(node_id) 

726 

727 if qnode.taskDef.label not in cached_pipetask_values: 

728 search_opt["curvals"]["curr_pipetask"] = qnode.taskDef.label 

729 cached_pipetask_values[qnode.taskDef.label] = _get_job_values( 

730 config, search_opt, "runQuantumCommand" 

731 ) 

732 

733 _handle_job_values(cached_pipetask_values[qnode.taskDef.label], gwjob, unset_attributes) 

734 

735 # Update job with workflow attribute and profile values. 

736 qgraph_gwfile = _get_qgraph_gwfile( 

737 config, save_qgraph_per_job, gwjob, generic_workflow.get_file("runQgraphFile"), prefix 

738 ) 

739 butler_gwfile = _get_butler_gwfile(prefix, when_create, butler_config, execution_butler_dir) 

740 

741 generic_workflow.add_job(gwjob) 

742 generic_workflow.add_job_inputs(gwjob.name, [qgraph_gwfile, butler_gwfile]) 

743 

744 gwjob.cmdvals["qgraphId"] = cqgraph.qgraph.graphID 

745 gwjob.cmdvals["qgraphNodeId"] = ",".join( 

746 sorted([f"{node_id}" for node_id in cluster.qgraph_node_ids]) 

747 ) 

748 _enhance_command(config, generic_workflow, gwjob, cached_job_values) 

749 

750 # If writing per-job QuantumGraph files during TRANSFORM stage, 

751 # write it now while in memory. 

752 if save_qgraph_per_job == WhenToSaveQuantumGraphs.TRANSFORM: 

753 save_qg_subgraph(cqgraph.qgraph, qgraph_gwfile.src_uri, cluster.qgraph_node_ids) 

754 

755 # Create job dependencies. 

756 for parent in cqgraph.clusters(): 

757 for child in cqgraph.successors(parent): 

758 generic_workflow.add_job_relationships(parent.name, child.name) 

759 

760 # Add initial workflow. 

761 if config.get("runInit", "{default: False}"): 

762 add_workflow_init_nodes(config, cqgraph.qgraph, generic_workflow) 

763 

764 generic_workflow.run_attrs.update( 

765 { 

766 "bps_isjob": "True", 

767 "bps_project": config["project"], 

768 "bps_campaign": config["campaign"], 

769 "bps_run": generic_workflow.name, 

770 "bps_operator": config["operator"], 

771 "bps_payload": config["payloadName"], 

772 "bps_runsite": config["computeSite"], 

773 } 

774 ) 

775 

776 # Add final job 

777 add_final_job(config, generic_workflow, prefix) 

778 

779 return generic_workflow 

780 

781 

782def create_generic_workflow_config(config, prefix): 

783 """Create generic workflow configuration. 

784 

785 Parameters 

786 ---------- 

787 config : `lsst.ctrl.bps.BpsConfig` 

788 Bps configuration. 

789 prefix : `str` 

790 Root path for any output files. 

791 

792 Returns 

793 ------- 

794 generic_workflow_config : `lsst.ctrl.bps.BpsConfig` 

795 Configuration accompanying the GenericWorkflow. 

796 """ 

797 generic_workflow_config = BpsConfig(config) 

798 generic_workflow_config["workflowName"] = config["uniqProcName"] 

799 generic_workflow_config["workflowPath"] = prefix 

800 return generic_workflow_config 

801 

802 

803def add_final_job(config, generic_workflow, prefix): 

804 """Add final workflow job depending upon configuration. 

805 

806 Parameters 

807 ---------- 

808 config : `lsst.ctrl.bps.BpsConfig` 

809 Bps configuration. 

810 generic_workflow : `lsst.ctrl.bps.GenericWorkflow` 

811 Generic workflow to which attributes should be added. 

812 prefix : `str` 

813 Directory in which to output final script. 

814 """ 

815 _, when_create = config.search(".executionButler.whenCreate") 

816 _, when_merge = config.search(".executionButler.whenMerge") 

817 

818 search_opt = {"searchobj": config[".executionButler"], "curvals": {}, "default": None} 

819 found, value = config.search("computeSite", opt=search_opt) 

820 if found: 

821 search_opt["curvals"]["curr_site"] = value 

822 found, value = config.search("computeCloud", opt=search_opt) 

823 if found: 

824 search_opt["curvals"]["curr_cloud"] = value 

825 

826 if when_create.upper() != "NEVER" and when_merge.upper() != "NEVER": 

827 # create gwjob 

828 gwjob = GenericWorkflowJob("mergeExecutionButler") 

829 gwjob.label = "mergeExecutionButler" 

830 

831 job_values = _get_job_values(config, search_opt, None) 

832 for attr in _ATTRS_ALL: 

833 if not getattr(gwjob, attr) and job_values.get(attr, None): 

834 setattr(gwjob, attr, job_values[attr]) 

835 

836 # Create script and add command line to job. 

837 gwjob.executable, gwjob.arguments = _create_final_command(config, prefix) 

838 

839 # Determine inputs from command line. 

840 for file_key in re.findall(r"<FILE:([^>]+)>", gwjob.arguments): 

841 gwfile = generic_workflow.get_file(file_key) 

842 generic_workflow.add_job_inputs(gwjob.name, gwfile) 

843 

844 _enhance_command(config, generic_workflow, gwjob, {}) 

845 

846 # Put transfer repo job in appropriate location in workflow. 

847 if when_merge.upper() == "ALWAYS": 

848 # add as special final job 

849 generic_workflow.add_final(gwjob) 

850 elif when_merge.upper() == "SUCCESS": 

851 # add as regular sink node 

852 add_final_job_as_sink(generic_workflow, gwjob) 

853 else: 

854 raise ValueError(f"Invalid value for executionButler.when_merge {when_merge}") 

855 

856 

857def _create_final_command(config, prefix): 

858 """Create the command and shell script for the final job. 

859 

860 Parameters 

861 ---------- 

862 config : `lsst.ctrl.bps.BpsConfig` 

863 Bps configuration. 

864 prefix : `str` 

865 Directory in which to output final script. 

866 

867 Returns 

868 ------- 

869 executable : `lsst.ctrl.bps.GenericWorkflowExec` 

870 Executable object for the final script. 

871 arguments : `str` 

872 Command line needed to call the final script. 

873 """ 

874 search_opt = { 

875 "replaceVars": False, 

876 "replaceEnvVars": False, 

877 "expandEnvVars": False, 

878 "searchobj": config["executionButler"], 

879 } 

880 

881 script_file = os.path.join(prefix, "final_job.bash") 

882 with open(script_file, "w") as fh: 

883 print("#!/bin/bash\n", file=fh) 

884 print("set -e", file=fh) 

885 print("set -x", file=fh) 

886 

887 print("butlerConfig=$1", file=fh) 

888 print("executionButlerDir=$2", file=fh) 

889 

890 i = 1 

891 found, command = config.search(f".executionButler.command{i}", opt=search_opt) 

892 while found: 

893 # Temporarily replace any env vars so formatter doesn't try to 

894 # replace them. 

895 command = re.sub(r"\${([^}]+)}", r"<BPSTMP:\1>", command) 

896 

897 # executionButlerDir and butlerConfig will be args to script and 

898 # set to env vars 

899 command = command.replace("{executionButlerDir}", "<BPSTMP:executionButlerDir>") 

900 command = command.replace("{butlerConfig}", "<BPSTMP:butlerConfig>") 

901 

902 # Replace all other vars in command string 

903 search_opt["replaceVars"] = True 

904 command = config.formatter.format(command, config, search_opt) 

905 search_opt["replaceVars"] = False 

906 

907 # Replace any temporary env place holders. 

908 command = re.sub(r"<BPSTMP:([^>]+)>", r"${\1}", command) 

909 

910 print(command, file=fh) 

911 i += 1 

912 found, command = config.search(f".executionButler.command{i}", opt=search_opt) 

913 os.chmod(script_file, 0o755) 

914 executable = GenericWorkflowExec(os.path.basename(script_file), script_file, True) 

915 

916 _, orig_butler = config.search("butlerConfig") 

917 # The execution butler was saved as butlerConfig in the workflow. 

918 return executable, f"{orig_butler} <FILE:butlerConfig>" 

919 

920 

921def add_final_job_as_sink(generic_workflow, final_job): 

922 """Add final job as the single sink for the workflow. 

923 

924 Parameters 

925 ---------- 

926 generic_workflow : `lsst.ctrl.bps.GenericWorkflow` 

927 Generic workflow to which attributes should be added. 

928 final_job : `lsst.ctrl.bps.GenericWorkflowJob` 

929 Job to add as new sink node depending upon all previous sink nodes. 

930 """ 

931 # Find sink nodes of generic workflow graph. 

932 gw_sinks = [n for n in generic_workflow if generic_workflow.out_degree(n) == 0] 

933 _LOG.debug("gw_sinks = %s", gw_sinks) 

934 

935 generic_workflow.add_job(final_job) 

936 generic_workflow.add_job_relationships(gw_sinks, final_job.name)