21 from __future__
import annotations
23 """Module defining GraphBuilder class and related methods.
26 __all__ = [
'GraphBuilder']
32 from collections
import ChainMap
33 from contextlib
import contextmanager
34 from dataclasses
import dataclass
35 from typing
import Dict, Iterable, Iterator, List
42 from .connections
import iterConnections
43 from .pipeline
import PipelineDatasetTypes, TaskDatasetTypes, TaskDef, Pipeline
44 from .graph
import QuantumGraph, QuantumGraphTaskNodes
45 from lsst.daf.butler
import (
54 from lsst.daf.butler.registry.queries.exprParser
import ParseError, ParserYacc, TreeVisitor
61 _LOG = logging.getLogger(__name__.partition(
".")[2])
64 class _DatasetDict(NamedKeyDict[DatasetType, Dict[DataCoordinate, DatasetRef]]):
65 """A custom dictionary that maps `DatasetType` to a nested dictionary of
66 the known `DatasetRef` instances of that type.
71 Positional arguments are forwarded to the `dict` constructor.
72 universe : `DimensionUniverse`
73 Universe of all possible dimensions.
75 def __init__(self, *args, universe: DimensionGraph):
81 universe: DimensionUniverse) -> _DatasetDict:
82 """Construct a dictionary from a flat iterable of `DatasetType` keys.
86 datasetTypes : `iterable` of `DatasetType`
87 DatasetTypes to use as keys for the dict. Values will be empty
89 universe : `DimensionUniverse`
90 Universe of all possible dimensions.
94 dictionary : `_DatasetDict`
95 A new `_DatasetDict` instance.
97 return cls({datasetType: {}
for datasetType
in datasetTypes}, universe=universe)
100 def fromSubset(cls, datasetTypes: Iterable[DatasetType], first: _DatasetDict, *rest: _DatasetDict
102 """Return a new dictionary by extracting items corresponding to the
103 given keys from one or more existing dictionaries.
107 datasetTypes : `iterable` of `DatasetType`
108 DatasetTypes to use as keys for the dict. Values will be obtained
109 by lookups against ``first`` and ``rest``.
110 first : `_DatasetDict`
111 Another dictionary from which to extract values.
113 Additional dictionaries from which to extract values.
117 dictionary : `_DatasetDict`
118 A new dictionary instance.
120 combined = ChainMap(first, *rest)
121 return cls({datasetType: combined[datasetType]
for datasetType
in datasetTypes},
122 universe=first.universe)
126 """The union of all dimensions used by all dataset types in this
127 dictionary, including implied dependencies (`DimensionGraph`).
132 return base.union(*[datasetType.dimensions
for datasetType
in self.keys()])
135 """Unpack nested single-element `DatasetRef` dicts into a new
136 mapping with `DatasetType` keys and `DatasetRef` values.
138 This method assumes that each nest contains exactly one item, as is the
139 case for all "init" datasets.
143 dictionary : `NamedKeyDict`
144 Dictionary mapping `DatasetType` to `DatasetRef`, with both
145 `DatasetType` instances and string names usable as keys.
147 def getOne(refs: Dict[DataCoordinate, DatasetRef]) -> DatasetRef:
150 return NamedKeyDict({datasetType: getOne(refs)
for datasetType, refs
in self.items()})
153 """Unpack nested multi-element `DatasetRef` dicts into a new
154 mapping with `DatasetType` keys and `set` of `DatasetRef` values.
158 dictionary : `NamedKeyDict`
159 Dictionary mapping `DatasetType` to `DatasetRef`, with both
160 `DatasetType` instances and string names usable as keys.
162 return NamedKeyDict({datasetType: list(refs.values())
for datasetType, refs
in self.items()})
164 def extract(self, datasetType: DatasetType, dataIds: Iterable[DataCoordinate]
165 ) -> Iterator[DatasetRef]:
166 """Iterate over the contained `DatasetRef` instances that match the
167 given `DatasetType` and data IDs.
171 datasetType : `DatasetType`
172 Dataset type to match.
173 dataIds : `Iterable` [ `DataCoordinate` ]
178 refs : `Iterator` [ `DatasetRef` ]
179 DatasetRef instances for which ``ref.datasetType == datasetType``
180 and ``ref.dataId`` is in ``dataIds``.
182 refs = self[datasetType]
183 return (refs[dataId]
for dataId
in dataIds)
187 """Helper class aggregating information about a `Quantum`, used when
188 constructing a `QuantumGraph`.
190 See `_PipelineScaffolding` for a top-down description of the full
191 scaffolding data structure.
195 task : _TaskScaffolding
196 Back-reference to the helper object for the `PipelineTask` this quantum
197 represents an execution of.
198 dataId : `DataCoordinate`
199 Data ID for this quantum.
201 def __init__(self, task: _TaskScaffolding, dataId: DataCoordinate):
204 self.
inputs = _DatasetDict.fromDatasetTypes(task.inputs.keys(), universe=dataId.universe)
205 self.
outputs = _DatasetDict.fromDatasetTypes(task.outputs.keys(), universe=dataId.universe)
206 self.
prerequisites = _DatasetDict.fromDatasetTypes(task.prerequisites.keys(),
207 universe=dataId.universe)
209 __slots__ = (
"task",
"dataId",
"inputs",
"outputs",
"prerequisites")
212 return f
"_QuantumScaffolding(taskDef={self.taskDef}, dataId={self.dataId}, ...)"
214 task: _TaskScaffolding
215 """Back-reference to the helper object for the `PipelineTask` this quantum
216 represents an execution of.
219 dataId: DataCoordinate
220 """Data ID for this quantum.
224 """Nested dictionary containing `DatasetRef` inputs to this quantum.
226 This is initialized to map each `DatasetType` to an empty dictionary at
227 construction. Those nested dictionaries are populated (with data IDs as
228 keys) with unresolved `DatasetRef` instances in
229 `_PipelineScaffolding.connectDataIds`.
232 outputs: _DatasetDict
233 """Nested dictionary containing `DatasetRef` outputs this quantum.
236 prerequisites: _DatasetDict
237 """Nested dictionary containing `DatasetRef` prerequisite inputs to this
242 """Transform the scaffolding object into a true `Quantum` instance.
247 An actual `Quantum` instance.
249 allInputs = self.
inputs.unpackMultiRefs()
255 allInputs = self.
task.taskDef.connections.adjustQuantum(allInputs)
257 taskName=self.
task.taskDef.taskName,
258 taskClass=self.
task.taskDef.taskClass,
260 initInputs=self.
task.initInputs.unpackSingleRefs(),
261 predictedInputs=allInputs,
262 outputs=self.
outputs.unpackMultiRefs(),
268 """Helper class aggregating information about a `PipelineTask`, used when
269 constructing a `QuantumGraph`.
271 See `_PipelineScaffolding` for a top-down description of the full
272 scaffolding data structure.
277 Data structure that identifies the task class and its config.
278 parent : `_PipelineScaffolding`
279 The parent data structure that will hold the instance being
281 datasetTypes : `TaskDatasetTypes`
282 Data structure that categorizes the dataset types used by this task.
284 def __init__(self, taskDef: TaskDef, parent: _PipelineScaffolding, datasetTypes: TaskDatasetTypes):
285 universe = parent.dimensions.universe
287 self.
dimensions = DimensionGraph(universe, names=taskDef.connections.dimensions)
288 assert self.
dimensions.issubset(parent.dimensions)
291 self.
initInputs = _DatasetDict.fromSubset(datasetTypes.initInputs, parent.initInputs,
292 parent.initIntermediates)
293 self.
initOutputs = _DatasetDict.fromSubset(datasetTypes.initOutputs, parent.initIntermediates,
295 self.
inputs = _DatasetDict.fromSubset(datasetTypes.inputs, parent.inputs, parent.intermediates)
296 self.
outputs = _DatasetDict.fromSubset(datasetTypes.outputs, parent.intermediates, parent.outputs)
297 self.
prerequisites = _DatasetDict.fromSubset(datasetTypes.prerequisites, parent.prerequisites)
304 return f
"_TaskScaffolding(taskDef={self.taskDef}, ...)"
307 """Data structure that identifies the task class and its config
311 dimensions: DimensionGraph
312 """The dimensions of a single `Quantum` of this task (`DimensionGraph`).
315 initInputs: _DatasetDict
316 """Dictionary containing information about datasets used to construct this
317 task (`_DatasetDict`).
320 initOutputs: _DatasetDict
321 """Dictionary containing information about datasets produced as a
322 side-effect of constructing this task (`_DatasetDict`).
326 """Dictionary containing information about datasets used as regular,
327 graph-constraining inputs to this task (`_DatasetDict`).
330 outputs: _DatasetDict
331 """Dictionary containing information about datasets produced by this task
335 prerequisites: _DatasetDict
336 """Dictionary containing information about input datasets that must be
337 present in the repository before any Pipeline containing this task is run
341 quanta: Dict[DataCoordinate, _QuantumScaffolding]
342 """Dictionary mapping data ID to a scaffolding object for the Quantum of
343 this task with that data ID.
347 """Create a `QuantumGraphTaskNodes` instance from the information in
352 nodes : `QuantumGraphTaskNodes`
353 The `QuantumGraph` elements corresponding to this task.
357 quanta=[q.makeQuantum()
for q
in self.
quanta.values()],
358 initInputs=self.
initInputs.unpackSingleRefs(),
365 """A helper data structure that organizes the information involved in
366 constructing a `QuantumGraph` for a `Pipeline`.
370 pipeline : `Pipeline`
371 Sequence of tasks from which a graph is to be constructed. Must
372 have nested task classes already imported.
373 universe : `DimensionUniverse`
374 Universe of all possible dimensions.
378 The scaffolding data structure contains nested data structures for both
379 tasks (`_TaskScaffolding`) and datasets (`_DatasetDict`). The dataset
380 data structures are shared between the pipeline-level structure (which
381 aggregates all datasets and categorizes them from the perspective of the
382 complete pipeline) and the individual tasks that use them as inputs and
385 `QuantumGraph` construction proceeds in four steps, with each corresponding
386 to a different `_PipelineScaffolding` method:
388 1. When `_PipelineScaffolding` is constructed, we extract and categorize
389 the DatasetTypes used by the pipeline (delegating to
390 `PipelineDatasetTypes.fromPipeline`), then use these to construct the
391 nested `_TaskScaffolding` and `_DatasetDict` objects.
393 2. In `connectDataIds`, we construct and run the "Big Join Query", which
394 returns related tuples of all dimensions used to identify any regular
395 input, output, and intermediate datasets (not prerequisites). We then
396 iterate over these tuples of related dimensions, identifying the subsets
397 that correspond to distinct data IDs for each task and dataset type,
398 and then create `_QuantumScaffolding` objects.
400 3. In `resolveDatasetRefs`, we run follow-up queries against all of the
401 dataset data IDs previously identified, transforming unresolved
402 DatasetRefs into resolved DatasetRefs where appropriate. We then look
403 up prerequisite datasets for all quanta.
405 4. In `makeQuantumGraph`, we construct a `QuantumGraph` from the lists of
406 per-task `_QuantumScaffolding` objects.
409 _LOG.debug(
"Initializing data structures for QuantumGraph generation.")
412 datasetTypes = PipelineDatasetTypes.fromPipeline(pipeline, registry=registry)
415 for attr
in (
"initInputs",
"initIntermediates",
"initOutputs",
416 "inputs",
"intermediates",
"outputs",
"prerequisites"):
417 setattr(self, attr, _DatasetDict.fromDatasetTypes(getattr(datasetTypes, attr),
418 universe=registry.dimensions))
421 self.
dimensions = self.inputs.dimensions.union(self.intermediates.dimensions,
422 self.outputs.dimensions)
427 if isinstance(pipeline, Pipeline):
428 pipeline = pipeline.toExpandedPipeline()
430 for taskDef, taskDatasetTypes
in zip(pipeline,
431 datasetTypes.byTask.values())]
436 return f
"_PipelineScaffolding(tasks={self.tasks}, ...)"
438 tasks: List[_TaskScaffolding]
439 """Scaffolding data structures for each task in the pipeline
440 (`list` of `_TaskScaffolding`).
443 initInputs: _DatasetDict
444 """Datasets consumed but not produced when constructing the tasks in this
445 pipeline (`_DatasetDict`).
448 initIntermediates: _DatasetDict
449 """Datasets that are both consumed and produced when constructing the tasks
450 in this pipeline (`_DatasetDict`).
453 initOutputs: _DatasetDict
454 """Datasets produced but not consumed when constructing the tasks in this
455 pipeline (`_DatasetDict`).
459 """Datasets that are consumed but not produced when running this pipeline
463 intermediates: _DatasetDict
464 """Datasets that are both produced and consumed when running this pipeline
468 outputs: _DatasetDict
469 """Datasets produced but not consumed when when running this pipeline
473 prerequisites: _DatasetDict
474 """Datasets that are consumed when running this pipeline and looked up
475 per-Quantum when generating the graph (`_DatasetDict`).
478 dimensions: DimensionGraph
479 """All dimensions used by any regular input, intermediate, or output
480 (not prerequisite) dataset; the set of dimension used in the "Big Join
481 Query" (`DimensionGraph`).
483 This is required to be a superset of all task quantum dimensions.
488 """Query for the data IDs that connect nodes in the `QuantumGraph`.
490 This method populates `_TaskScaffolding.dataIds` and
491 `_DatasetScaffolding.dataIds` (except for those in `prerequisites`).
495 registry : `lsst.daf.butler.Registry`
496 Registry for the data repository; used for all data ID queries.
497 collections : `lsst.daf.butler.CollectionSearch`
498 Object representing the collections to search for input datasets.
499 userQuery : `str`, optional
500 User-provided expression to limit the data IDs processed.
505 `lsst.daf.butler.registry.queries.DataCoordinateQueryResults`
506 An interface to a database temporary table containing all data IDs
507 that will appear in this `QuantumGraph`. Returned inside a
508 context manager, which will drop the temporary table at the end of
509 the `with` block in which this method is called.
511 _LOG.debug(
"Building query for data IDs.")
513 emptyDataId = DataCoordinate.makeEmpty(registry.dimensions)
514 for datasetType, refs
in itertools.chain(self.initInputs.items(),
515 self.initIntermediates.items(),
516 self.initOutputs.items()):
517 refs[emptyDataId] = DatasetRef(datasetType, emptyDataId)
522 _LOG.debug(
"Submitting data ID query and materializing results.")
524 datasets=list(self.inputs),
525 collections=collections,
527 ).materialize()
as commonDataIds:
528 _LOG.debug(
"Expanding data IDs.")
529 commonDataIds = commonDataIds.expanded()
530 _LOG.debug(
"Iterating over query results to associate quanta with datasets.")
534 for n, commonDataId
in enumerate(commonDataIds):
540 for datasetType, refs
in itertools.chain(self.inputs.items(), self.intermediates.items(),
541 self.outputs.items()):
542 datasetDataId = commonDataId.subset(datasetType.dimensions)
543 ref = refs.get(datasetDataId)
545 ref = DatasetRef(datasetType, datasetDataId)
546 refs[datasetDataId] = ref
547 refsForRow[datasetType.name] = ref
550 for task
in self.
tasks:
551 quantumDataId = commonDataId.subset(task.dimensions)
552 quantum = task.quanta.get(quantumDataId)
555 task.quanta[quantumDataId] = quantum
564 for datasetType
in task.inputs:
565 ref = refsForRow[datasetType.name]
566 quantum.inputs[datasetType.name][ref.dataId] = ref
567 for datasetType
in task.outputs:
568 ref = refsForRow[datasetType.name]
569 quantum.outputs[datasetType.name][ref.dataId] = ref
570 _LOG.debug(
"Finished processing %d rows from data ID query.", n)
574 """Perform follow up queries for each dataset data ID produced in
577 This method populates `_DatasetScaffolding.refs` (except for those in
582 registry : `lsst.daf.butler.Registry`
583 Registry for the data repository; used for all data ID queries.
584 collections : `lsst.daf.butler.CollectionSearch`
585 Object representing the collections to search for input datasets.
586 run : `str`, optional
587 Name of the `~lsst.daf.butler.CollectionType.RUN` collection for
588 output datasets, if it already exists.
590 `lsst.daf.butler.registry.queries.DataCoordinateQueryResults`
591 Result of a previous call to `connectDataIds`.
592 skipExisting : `bool`, optional
593 If `True` (default), a Quantum is not created if all its outputs
594 already exist in ``run``. Ignored if ``run`` is `None`.
599 Raised if an output dataset already exists in the output run
600 and ``skipExisting`` is `False`. The case where some but not all
601 of a quantum's outputs are present and ``skipExisting`` is `True`
602 cannot be identified at this stage, and is handled by `fillQuanta`
608 for datasetType, refs
in itertools.chain(self.initIntermediates.items(),
609 self.initOutputs.items(),
610 self.intermediates.items(),
611 self.outputs.items()):
612 _LOG.debug(
"Resolving %d datasets for intermediate and/or output dataset %s.",
613 len(refs), datasetType.name)
614 isInit = datasetType
in self.initIntermediates
or datasetType
in self.initOutputs
615 resolvedRefQueryResults = commonDataIds.subset(
616 datasetType.dimensions,
623 for resolvedRef
in resolvedRefQueryResults:
628 assert resolvedRef.dataId
in refs
629 if skipExisting
or isInit:
630 refs[resolvedRef.dataId] = resolvedRef
633 f
"output RUN collection '{run}' with data ID"
634 f
" {resolvedRef.dataId}.")
636 for datasetType, refs
in itertools.chain(self.initInputs.items(), self.inputs.items()):
637 _LOG.debug(
"Resolving %d datasets for input dataset %s.", len(refs), datasetType.name)
638 resolvedRefQueryResults = commonDataIds.subset(
639 datasetType.dimensions,
643 collections=collections,
646 dataIdsNotFoundYet = set(refs.keys())
647 for resolvedRef
in resolvedRefQueryResults:
648 dataIdsNotFoundYet.discard(resolvedRef.dataId)
649 refs[resolvedRef.dataId] = resolvedRef
650 if dataIdsNotFoundYet:
652 f
"{len(dataIdsNotFoundYet)} dataset(s) of type "
653 f
"'{datasetType.name}' was/were present in a previous "
654 f
"query, but could not be found now."
655 f
"This is either a logic bug in QuantumGraph generation "
656 f
"or the input collections have been modified since "
657 f
"QuantumGraph generation began."
661 for task
in self.
tasks:
663 "Applying resolutions and finding prerequisites for %d quanta of task with label '%s'.",
668 c.name: c.lookupFunction
669 for c
in iterConnections(task.taskDef.connections,
"prerequisiteInputs")
670 if c.lookupFunction
is not None
673 for quantum
in task.quanta.values():
680 if run
is not None and skipExisting:
683 for datasetType, originalRefs
in quantum.outputs.items():
684 for ref
in task.outputs.extract(datasetType, originalRefs.keys()):
685 if ref.id
is not None:
686 resolvedRefs.append(ref)
688 unresolvedRefs.append(ref)
692 f
"Quantum {quantum.dataId} of task with label "
693 f
"'{quantum.taskDef.label}' has some outputs that exist ({resolvedRefs}) "
694 f
"and others that don't ({unresolvedRefs})."
699 dataIdsToSkip.append(quantum.dataId)
703 for datasetType, refs
in quantum.inputs.items():
704 for ref
in task.inputs.extract(datasetType, refs.keys()):
705 refs[ref.dataId] = ref
714 for datasetType
in task.prerequisites:
715 lookupFunction = lookupFunctions.get(datasetType.name)
716 if lookupFunction
is not None:
718 lookupFunction(datasetType, registry, quantum.dataId, collections)
721 refs = list(registry.queryDatasets(datasetType,
722 collections=collections,
723 dataId=quantum.dataId,
724 deduplicate=
True).expanded())
725 quantum.prerequisites[datasetType].update({ref.dataId: ref
for ref
in refs})
728 _LOG.debug(
"Pruning %d quanta for task with label '%s' because all of their outputs exist.",
729 len(dataIdsToSkip), task.taskDef.label)
730 for dataId
in dataIdsToSkip:
731 del task.quanta[dataId]
734 """Create a `QuantumGraph` from the quanta already present in
735 the scaffolding data structure.
739 graph : `QuantumGraph`
740 The full `QuantumGraph`.
743 graph.initInputs = self.initInputs.unpackSingleRefs()
744 graph.initOutputs = self.initOutputs.unpackSingleRefs()
745 graph.initIntermediates = self.initIntermediates.unpackSingleRefs()
750 """Implementation of TreeVisitor which looks for instrument name
752 Instrument should be specified as a boolean expression
754 instrument = 'string'
755 'string' = instrument
757 so we only need to find a binary operator where operator is "=",
758 one side is a string literal and other side is an identifier.
759 All visit methods return tuple of (type, value), non-useful nodes
760 return None for both type and value.
771 return (
"str", value)
782 if name.lower() ==
"instrument":
783 return (
"id",
"instrument")
792 if lhs == (
"id",
"instrument")
and rhs[0] ==
"str":
794 elif rhs == (
"id",
"instrument")
and lhs[0] ==
"str":
807 def _findInstruments(queryStr):
808 """Get the names of any instrument named in the query string by searching
809 for "instrument = <value>" and similar patterns.
813 queryStr : `str` or None
814 The query string to search, or None if there is no query.
818 instruments : `list` [`str`]
819 The list of instrument names found in the query.
824 If the query expression can not be parsed.
828 parser = ParserYacc()
831 tree = parser.parse(queryStr)
832 except ParseError
as exc:
833 raise ValueError(f
"failed to parse query expression: {queryStr}")
from exc
835 return finder.instruments
844 """Base class for exceptions generated by graph builder.
849 class OutputExistsError(GraphBuilderError):
850 """Exception generated when output datasets already exist.
856 """Exception generated when a prerequisite dataset does not exist.
862 """GraphBuilder class is responsible for building task execution graph from
867 registry : `~lsst.daf.butler.Registry`
868 Data butler instance.
869 skipExisting : `bool`, optional
870 If `True` (default), a Quantum is not created if all its outputs
879 def makeGraph(self, pipeline, collections, run, userQuery):
880 """Create execution graph for a pipeline.
884 pipeline : `Pipeline`
885 Pipeline definition, task names/classes and their configs.
886 collections : `lsst.daf.butler.CollectionSearch`
887 Object representing the collections to search for input datasets.
888 run : `str`, optional
889 Name of the `~lsst.daf.butler.CollectionType.RUN` collection for
890 output datasets, if it already exists.
892 String which defines user-defined selection for registry, should be
893 empty or `None` if there is no restrictions on data selection.
897 graph : `QuantumGraph`
902 Raised when user expression cannot be parsed.
904 Raised when output datasets already exist.
906 Other exceptions types may be raised by underlying registry
911 instrument = pipeline.getInstrument()
912 if isinstance(instrument, str):
913 instrument = doImport(instrument)
914 instrumentName = instrument.getName()
if instrument
else None
917 with scaffolding.connectDataIds(self.
registry, collections, userQuery)
as commonDataIds:
918 scaffolding.resolveDatasetRefs(self.
registry, collections, run, commonDataIds,
920 return scaffolding.makeQuantumGraph()
923 def _verifyInstrumentRestriction(instrumentName, query):
924 """Add an instrument restriction to the query if it does not have one,
925 and verify that if given an instrument name that there are no other
926 instrument restrictions in the query.
930 instrumentName : `str`
931 The name of the instrument that should appear in the query.
938 The query string with the instrument added to it if needed.
943 If the pipeline names an instrument and the query contains more
944 than one instrument or the name of the instrument in the query does
945 not match the instrument named by the pipeline.
947 if not instrumentName:
949 queryInstruments = _findInstruments(query)
950 if len(queryInstruments) > 1:
951 raise RuntimeError(f
"When the pipeline has an instrument (\"{instrumentName}\") the query must "
952 "have zero instruments or one instrument that matches the pipeline. "
953 f
"Found these instruments in the query: {queryInstruments}.")
954 if not queryInstruments:
956 restriction = f
"instrument = '{instrumentName}'"
957 _LOG.debug(f
"Adding restriction \"{restriction}\" to query.")
958 query = f
"{restriction} AND ({query})" if query
else restriction
959 elif queryInstruments[0] != instrumentName:
962 raise RuntimeError(f
"The instrument named in the query (\"{queryInstruments[0]}\") does not "
963 f
"match the instrument named by the pipeline (\"{instrumentName}\")")