Coverage for python/lsst/ctrl/mpexec/preExecInit.py: 18%
159 statements
« prev ^ index » next coverage.py v7.4.1, created at 2024-02-07 12:18 +0000
« prev ^ index » next coverage.py v7.4.1, created at 2024-02-07 12:18 +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 software is dual licensed under the GNU General Public License and also
10# under a 3-clause BSD license. Recipients may choose which of these licenses
11# to use; please see the files gpl-3.0.txt and/or bsd_license.txt,
12# respectively. If you choose the GPL option then the following text applies
13# (but note that there is still no warranty even if you opt for BSD instead):
14#
15# This program is free software: you can redistribute it and/or modify
16# it under the terms of the GNU General Public License as published by
17# the Free Software Foundation, either version 3 of the License, or
18# (at your option) any later version.
19#
20# This program is distributed in the hope that it will be useful,
21# but WITHOUT ANY WARRANTY; without even the implied warranty of
22# MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
23# GNU General Public License for more details.
24#
25# You should have received a copy of the GNU General Public License
26# along with this program. If not, see <http://www.gnu.org/licenses/>.
28from __future__ import annotations
30__all__ = ["PreExecInit"]
32# -------------------------------
33# Imports of standard modules --
34# -------------------------------
35import abc
36import logging
37from collections.abc import Iterable, Iterator
38from contextlib import contextmanager
39from typing import TYPE_CHECKING, Any
41# -----------------------------
42# Imports for other modules --
43# -----------------------------
44from lsst.daf.butler import DatasetRef, DatasetType
45from lsst.daf.butler.registry import ConflictingDefinitionError
46from lsst.pipe.base import PipelineDatasetTypes
47from lsst.pipe.base import automatic_connection_constants as acc
48from lsst.utils.packages import Packages
50if TYPE_CHECKING:
51 from lsst.daf.butler import Butler, LimitedButler
52 from lsst.pipe.base import QuantumGraph, TaskDef, TaskFactory
54_LOG = logging.getLogger(__name__)
57class MissingReferenceError(Exception):
58 """Exception raised when resolved reference is missing from graph."""
60 pass
63def _compare_packages(old_packages: Packages, new_packages: Packages) -> None:
64 """Compare two versions of Packages.
66 Parameters
67 ----------
68 old_packages : `Packages`
69 Previously recorded package versions.
70 new_packages : `Packages`
71 New set of package versions.
73 Raises
74 ------
75 TypeError
76 Raised if parameters are inconsistent.
77 """
78 diff = new_packages.difference(old_packages)
79 if diff:
80 versions_str = "; ".join(f"{pkg}: {diff[pkg][1]} vs {diff[pkg][0]}" for pkg in diff)
81 raise TypeError(f"Package versions mismatch: ({versions_str})")
82 else:
83 _LOG.debug("new packages are consistent with old")
86class PreExecInitBase(abc.ABC):
87 """Common part of the implementation of PreExecInit classes that does not
88 depend on Butler type.
90 Parameters
91 ----------
92 butler : `~lsst.daf.butler.LimitedButler`
93 Butler to use.
94 taskFactory : `lsst.pipe.base.TaskFactory`
95 Task factory.
96 extendRun : `bool`
97 Whether extend run parameter is in use.
98 """
100 def __init__(self, butler: LimitedButler, taskFactory: TaskFactory, extendRun: bool):
101 self.butler = butler
102 self.taskFactory = taskFactory
103 self.extendRun = extendRun
105 def initialize(
106 self,
107 graph: QuantumGraph,
108 saveInitOutputs: bool = True,
109 registerDatasetTypes: bool = False,
110 saveVersions: bool = True,
111 ) -> None:
112 """Perform all initialization steps.
114 Convenience method to execute all initialization steps. Instead of
115 calling this method and providing all options it is also possible to
116 call methods individually.
118 Parameters
119 ----------
120 graph : `~lsst.pipe.base.QuantumGraph`
121 Execution graph.
122 saveInitOutputs : `bool`, optional
123 If ``True`` (default) then save "init outputs", configurations,
124 and package versions to butler.
125 registerDatasetTypes : `bool`, optional
126 If ``True`` then register dataset types in registry, otherwise
127 they must be already registered.
128 saveVersions : `bool`, optional
129 If ``False`` then do not save package versions even if
130 ``saveInitOutputs`` is set to ``True``.
131 """
132 # register dataset types or check consistency
133 self.initializeDatasetTypes(graph, registerDatasetTypes)
135 # Save task initialization data or check that saved data
136 # is consistent with what tasks would save
137 if saveInitOutputs:
138 self.saveInitOutputs(graph)
139 self.saveConfigs(graph)
140 if saveVersions:
141 self.savePackageVersions(graph)
143 @abc.abstractmethod
144 def initializeDatasetTypes(self, graph: QuantumGraph, registerDatasetTypes: bool = False) -> None:
145 """Save or check DatasetTypes output by the tasks in a graph.
147 Iterates over all DatasetTypes for all tasks in a graph and either
148 tries to add them to registry or compares them to existing ones.
150 Parameters
151 ----------
152 graph : `~lsst.pipe.base.QuantumGraph`
153 Execution graph.
154 registerDatasetTypes : `bool`, optional
155 If ``True`` then register dataset types in registry, otherwise
156 they must be already registered.
158 Raises
159 ------
160 ValueError
161 Raised if existing DatasetType is different from DatasetType
162 in a graph.
163 KeyError
164 Raised if ``registerDatasetTypes`` is ``False`` and DatasetType
165 does not exist in registry.
166 """
167 raise NotImplementedError()
169 def saveInitOutputs(self, graph: QuantumGraph) -> None:
170 """Write any datasets produced by initializing tasks in a graph.
172 Parameters
173 ----------
174 graph : `~lsst.pipe.base.QuantumGraph`
175 Execution graph.
177 Raises
178 ------
179 TypeError
180 Raised if the type of existing object in butler is different from
181 new data.
182 """
183 _LOG.debug("Will save InitOutputs for all tasks")
184 for taskDef in self._task_iter(graph):
185 init_input_refs = graph.initInputRefs(taskDef) or []
186 task = self.taskFactory.makeTask(taskDef, self.butler, init_input_refs)
187 for name in taskDef.connections.initOutputs:
188 attribute = getattr(taskDef.connections, name)
189 init_output_refs = graph.initOutputRefs(taskDef) or []
190 init_output_ref, obj_from_store = self._find_dataset(init_output_refs, attribute.name)
191 if init_output_ref is None:
192 raise ValueError(f"Cannot find dataset reference for init output {name} in a graph")
193 init_output_var = getattr(task, name)
195 if obj_from_store is not None:
196 _LOG.debug(
197 "Retrieving InitOutputs for task=%s key=%s dsTypeName=%s", task, name, attribute.name
198 )
199 obj_from_store = self.butler.get(init_output_ref)
200 # Types are supposed to be identical.
201 # TODO: Check that object contents is identical too.
202 if type(obj_from_store) is not type(init_output_var):
203 raise TypeError(
204 f"Stored initOutput object type {type(obj_from_store)} "
205 "is different from task-generated type "
206 f"{type(init_output_var)} for task {taskDef}"
207 )
208 else:
209 _LOG.debug("Saving InitOutputs for task=%s key=%s", taskDef.label, name)
210 # This can still raise if there is a concurrent write.
211 self.butler.put(init_output_var, init_output_ref)
213 def saveConfigs(self, graph: QuantumGraph) -> None:
214 """Write configurations for pipeline tasks to butler or check that
215 existing configurations are equal to the new ones.
217 Parameters
218 ----------
219 graph : `~lsst.pipe.base.QuantumGraph`
220 Execution graph.
222 Raises
223 ------
224 TypeError
225 Raised if existing object in butler is different from new data.
226 Exception
227 Raised if ``extendRun`` is `False` and datasets already exists.
228 Content of a butler collection should not be changed if exception
229 is raised.
230 """
232 def logConfigMismatch(msg: str) -> None:
233 """Log messages about configuration mismatch.
235 Parameters
236 ----------
237 msg : `str`
238 Log message to use.
239 """
240 _LOG.fatal("Comparing configuration: %s", msg)
242 _LOG.debug("Will save Configs for all tasks")
243 # start transaction to rollback any changes on exceptions
244 with self.transaction():
245 for taskDef in self._task_iter(graph):
246 # Config dataset ref is stored in task init outputs, but it
247 # may be also be missing.
248 task_output_refs = graph.initOutputRefs(taskDef)
249 if task_output_refs is None:
250 continue
252 config_ref, old_config = self._find_dataset(task_output_refs, taskDef.configDatasetName)
253 if config_ref is None:
254 continue
256 if old_config is not None:
257 if not taskDef.config.compare(old_config, shortcut=False, output=logConfigMismatch):
258 raise TypeError(
259 f"Config does not match existing task config {taskDef.configDatasetName!r} in "
260 "butler; tasks configurations must be consistent within the same run collection"
261 )
262 else:
263 _LOG.debug(
264 "Saving Config for task=%s dataset type=%s", taskDef.label, taskDef.configDatasetName
265 )
266 self.butler.put(taskDef.config, config_ref)
268 def savePackageVersions(self, graph: QuantumGraph) -> None:
269 """Write versions of software packages to butler.
271 Parameters
272 ----------
273 graph : `~lsst.pipe.base.QuantumGraph`
274 Execution graph.
276 Raises
277 ------
278 TypeError
279 Raised if existing object in butler is incompatible with new data.
280 """
281 packages = Packages.fromSystem()
282 _LOG.debug("want to save packages: %s", packages)
284 # start transaction to rollback any changes on exceptions
285 with self.transaction():
286 # Packages dataset ref is stored in graph's global init outputs,
287 # but it may be also be missing.
289 packages_ref, old_packages = self._find_dataset(
290 graph.globalInitOutputRefs(), PipelineDatasetTypes.packagesDatasetName
291 )
292 if packages_ref is None:
293 return
295 if old_packages is not None:
296 # Note that because we can only detect python modules that have
297 # been imported, the stored list of products may be more or
298 # less complete than what we have now. What's important is
299 # that the products that are in common have the same version.
300 _compare_packages(old_packages, packages)
301 # Update the old set of packages in case we have more packages
302 # that haven't been persisted.
303 extra = packages.extra(old_packages)
304 if extra:
305 _LOG.debug("extra packages: %s", extra)
306 old_packages.update(packages)
307 # have to remove existing dataset first, butler has no
308 # replace option.
309 self.butler.pruneDatasets([packages_ref], unstore=True, purge=True)
310 self.butler.put(old_packages, packages_ref)
311 else:
312 self.butler.put(packages, packages_ref)
314 def _find_dataset(
315 self, refs: Iterable[DatasetRef], dataset_type: str
316 ) -> tuple[DatasetRef | None, Any | None]:
317 """Find a ref with a given dataset type name in a list of references
318 and try to retrieve its data from butler.
320 Parameters
321 ----------
322 refs : `~collections.abc.Iterable` [ `~lsst.daf.butler.DatasetRef` ]
323 References to check for matching dataset type.
324 dataset_type : `str`
325 Name of a dataset type to look for.
327 Returns
328 -------
329 ref : `~lsst.daf.butler.DatasetRef` or `None`
330 Dataset reference or `None` if there is no matching dataset type.
331 data : `Any`
332 An existing object extracted from butler, `None` if ``ref`` is
333 `None` or if there is no existing object for that reference.
334 """
335 ref: DatasetRef | None = None
336 for ref in refs:
337 if ref.datasetType.name == dataset_type:
338 break
339 else:
340 return None, None
342 try:
343 data = self.butler.get(ref)
344 if data is not None and not self.extendRun:
345 # It must not exist unless we are extending run.
346 raise ConflictingDefinitionError(f"Dataset {ref} already exists in butler")
347 except (LookupError, FileNotFoundError):
348 data = None
349 return ref, data
351 def _task_iter(self, graph: QuantumGraph) -> Iterator[TaskDef]:
352 """Iterate over TaskDefs in a graph, return only tasks that have one or
353 more associated quanta.
354 """
355 for taskDef in graph.iterTaskGraph():
356 if graph.getNumberOfQuantaForTask(taskDef) > 0:
357 yield taskDef
359 @contextmanager
360 def transaction(self) -> Iterator[None]:
361 """Context manager for transaction.
363 Default implementation has no transaction support.
365 Yields
366 ------
367 `None`
368 No transaction support.
369 """
370 yield
373class PreExecInit(PreExecInitBase):
374 """Initialization of registry for QuantumGraph execution.
376 This class encapsulates all necessary operations that have to be performed
377 on butler and registry to prepare them for QuantumGraph execution.
379 Parameters
380 ----------
381 butler : `~lsst.daf.butler.Butler`
382 Data butler instance.
383 taskFactory : `~lsst.pipe.base.TaskFactory`
384 Task factory.
385 extendRun : `bool`, optional
386 If `True` then do not try to overwrite any datasets that might exist
387 in ``butler.run``; instead compare them when appropriate/possible. If
388 `False`, then any existing conflicting dataset will cause a butler
389 exception to be raised.
390 """
392 def __init__(self, butler: Butler, taskFactory: TaskFactory, extendRun: bool = False):
393 super().__init__(butler, taskFactory, extendRun)
394 self.full_butler = butler
395 if self.extendRun and self.full_butler.run is None:
396 raise RuntimeError(
397 "Cannot perform extendRun logic unless butler is initialized "
398 "with a default output RUN collection."
399 )
401 @contextmanager
402 def transaction(self) -> Iterator[None]:
403 # dosctring inherited
404 with self.full_butler.transaction():
405 yield
407 def initializeDatasetTypes(self, graph: QuantumGraph, registerDatasetTypes: bool = False) -> None:
408 # docstring inherited
409 pipeline = graph.taskGraph
410 pipelineDatasetTypes = PipelineDatasetTypes.fromPipeline(
411 pipeline, registry=self.full_butler.registry, include_configs=True, include_packages=True
412 )
413 # The "registry dataset types" saved with the QG have had their storage
414 # classes carefully resolved by PipelineGraph, whereas the dataset
415 # types from PipelineDatasetTypes are a mess because it uses
416 # NamedValueSet and that ignores storage classes. It will be fully
417 # removed here (and deprecated everywhere) on DM-40441.
418 # Note that these "registry dataset types" include dataset types that
419 # are not actually registered yet; they're the PipelineGraph's
420 # determination of what _should_ be registered.
421 registry_storage_classes = {
422 dataset_type.name: dataset_type.storageClass_name for dataset_type in graph.registryDatasetTypes()
423 }
424 registry_storage_classes[acc.PACKAGES_INIT_OUTPUT_NAME] = acc.PACKAGES_INIT_OUTPUT_STORAGE_CLASS
425 dataset_types: Iterable[DatasetType]
426 for dataset_types, is_input in (
427 (pipelineDatasetTypes.initIntermediates, True),
428 (pipelineDatasetTypes.initOutputs, False),
429 (pipelineDatasetTypes.intermediates, True),
430 (pipelineDatasetTypes.outputs, False),
431 ):
432 dataset_types = [
433 (
434 # The registry dataset types do not include components, but
435 # we don't support storage class overrides for those in
436 # other contexts anyway, and custom-built QGs may not have
437 # the registry dataset types field populated at all.x
438 dataset_type.overrideStorageClass(registry_storage_classes[dataset_type.name])
439 if dataset_type.name in registry_storage_classes
440 else dataset_type
441 )
442 for dataset_type in dataset_types
443 ]
444 self._register_output_dataset_types(registerDatasetTypes, dataset_types, is_input)
446 def _register_output_dataset_types(
447 self, registerDatasetTypes: bool, datasetTypes: Iterable[DatasetType], is_input: bool
448 ) -> None:
449 def _check_compatibility(datasetType: DatasetType, expected: DatasetType, is_input: bool) -> bool:
450 # These are output dataset types so check for compatibility on put.
451 is_compatible = expected.is_compatible_with(datasetType)
453 if is_input:
454 # This dataset type is also used for input so must be
455 # compatible on get as ell.
456 is_compatible = is_compatible and datasetType.is_compatible_with(expected)
458 if is_compatible:
459 _LOG.debug(
460 "The dataset type configurations differ (%s from task != %s from registry) "
461 "but the storage classes are compatible. Can continue.",
462 datasetType,
463 expected,
464 )
465 return is_compatible
467 missing_datasetTypes = set()
468 for datasetType in datasetTypes:
469 # Only composites are registered, no components, and by this point
470 # the composite should already exist.
471 if registerDatasetTypes and not datasetType.isComponent():
472 _LOG.debug("Registering DatasetType %s with registry", datasetType)
473 # this is a no-op if it already exists and is consistent,
474 # and it raises if it is inconsistent.
475 try:
476 self.full_butler.registry.registerDatasetType(datasetType)
477 except ConflictingDefinitionError:
478 if not _check_compatibility(
479 datasetType, self.full_butler.get_dataset_type(datasetType.name), is_input
480 ):
481 raise
482 else:
483 _LOG.debug("Checking DatasetType %s against registry", datasetType)
484 try:
485 expected = self.full_butler.get_dataset_type(datasetType.name)
486 except KeyError:
487 # Likely means that --register-dataset-types is forgotten.
488 missing_datasetTypes.add(datasetType.name)
489 continue
490 if expected != datasetType:
491 if not _check_compatibility(datasetType, expected, is_input):
492 raise ValueError(
493 f"DatasetType configuration does not match Registry: {datasetType} != {expected}"
494 )
496 if missing_datasetTypes:
497 plural = "s" if len(missing_datasetTypes) != 1 else ""
498 raise KeyError(
499 f"Missing dataset type definition{plural}: {', '.join(missing_datasetTypes)}. "
500 "Dataset types have to be registered with either `butler register-dataset-type` or "
501 "passing `--register-dataset-types` option to `pipetask run`."
502 )
505class PreExecInitLimited(PreExecInitBase):
506 """Initialization of registry for QuantumGraph execution.
508 This class works with LimitedButler and expects that all references in
509 QuantumGraph are resolved.
511 Parameters
512 ----------
513 butler : `~lsst.daf.butler.LimitedButler`
514 Limited data butler instance.
515 taskFactory : `~lsst.pipe.base.TaskFactory`
516 Task factory.
517 """
519 def __init__(self, butler: LimitedButler, taskFactory: TaskFactory):
520 super().__init__(butler, taskFactory, False)
522 def initializeDatasetTypes(self, graph: QuantumGraph, registerDatasetTypes: bool = False) -> None:
523 # docstring inherited
524 # With LimitedButler we never create or check dataset types.
525 pass