21 from __future__
import annotations
23 """Module defining GraphBuilder class and related methods.
26 __all__ = [
'GraphBuilder']
32 from collections
import ChainMap
33 from dataclasses
import dataclass
34 from typing
import Set, List, Dict, Optional, Iterable
40 from .pipeline
import PipelineDatasetTypes, TaskDatasetTypes, TaskDef, Pipeline
41 from .graph
import QuantumGraph, QuantumGraphTaskNodes
42 from lsst.daf.butler
import (
47 ExpandedDataCoordinate,
50 from lsst.daf.butler.core.utils
import NamedKeyDict
56 _LOG = logging.getLogger(__name__.partition(
".")[2])
61 """Helper class aggregating information about a `DatasetType`, used when
62 constructing a `QuantumGraph`.
64 `_DatasetScaffolding` does not hold the `DatasetType` instance itself
65 because it is usually used as the value type in `_DatasetScaffoldingDict`,
66 which uses `DatasetType` instances as keys.
68 See `_PipelineScaffolding` for a top-down description of the full
69 scaffolding data structure.
73 dimensions : `DimensionGraph`
74 Dimensions of the `DatasetType`.
76 def __init__(self, dimensions: DimensionGraph):
83 __slots__ = (
"dimensions",
"producer",
"consumers",
"dataIds",
"refs")
85 dimensions: DimensionGraph
86 """The dimensions of the dataset type (`DimensionGraph`).
88 Set during `_PipelineScaffolding` construction.
91 producer: Optional[_TaskScaffolding]
92 """The scaffolding objects for the Task that produces this dataset.
94 Set during `_PipelineScaffolding` construction.
97 consumers: Dict[str, _TaskScaffolding]
98 """The scaffolding objects for the Tasks that consume this dataset,
99 keyed by their label in the `Pipeline`.
101 Set during `_PipelineScaffolding` construction.
104 dataIds: Set[ExpandedDataCoordinate]
105 """Data IDs for all instances of this dataset type in the graph.
107 Populated after construction by `_PipelineScaffolding.fillDataIds`.
110 refs: List[DatasetRef]
111 """References for all instances of this dataset type in the graph.
113 Populated after construction by `_PipelineScaffolding.fillDatasetRefs`.
118 """Custom dictionary that maps `DatasetType` to `_DatasetScaffolding`.
120 See `_PipelineScaffolding` for a top-down description of the full
121 scaffolding data structure.
126 Positional arguments are forwarded to the `dict` constructor.
127 universe : `DimensionUniverse`
128 Universe of all possible dimensions.
130 def __init__(self, *args, universe: DimensionGraph):
136 universe: DimensionUniverse) -> _DatasetScaffoldingDict:
137 """Construct a a dictionary from a flat iterable of `DatasetType` keys.
141 datasetTypes : `iterable` of `DatasetType`
142 DatasetTypes to use as keys for the dict. Values will be
143 constructed from the dimensions of the keys.
144 universe : `DimensionUniverse`
145 Universe of all possible dimensions.
149 dictionary : `_DatasetScaffoldingDict`
150 A new dictionary instance.
153 for datasetType
in datasetTypes),
157 def fromSubset(cls, datasetTypes: Iterable[DatasetType], first: _DatasetScaffoldingDict,
158 *rest) -> _DatasetScaffoldingDict:
159 """Return a new dictionary by extracting items corresponding to the
160 given keys from one or more existing dictionaries.
164 datasetTypes : `iterable` of `DatasetType`
165 DatasetTypes to use as keys for the dict. Values will be obtained
166 by lookups against ``first`` and ``rest``.
167 first : `_DatasetScaffoldingDict`
168 Another dictionary from which to extract values.
170 Additional dictionaries from which to extract values.
174 dictionary : `_DatasetScaffoldingDict`
175 A new dictionary instance.
177 combined = ChainMap(first, *rest)
178 return cls(((datasetType, combined[datasetType])
for datasetType
in datasetTypes),
179 universe=first.universe)
183 """The union of all dimensions used by all dataset types in this
184 dictionary, including implied dependencies (`DimensionGraph`).
189 return base.union(*[scaffolding.dimensions
for scaffolding
in self.values()])
192 """Unpack nested single-element `DatasetRef` lists into a new
195 This method assumes that each `_DatasetScaffolding.refs` list contains
196 exactly one `DatasetRef`, as is the case for all "init" datasets.
200 dictionary : `NamedKeyDict`
201 Dictionary mapping `DatasetType` to `DatasetRef`, with both
202 `DatasetType` instances and string names usable as keys.
204 return NamedKeyDict((datasetType, scaffolding.refs[0])
for datasetType, scaffolding
in self.items())
209 """Helper class aggregating information about a `PipelineTask`, used when
210 constructing a `QuantumGraph`.
212 See `_PipelineScaffolding` for a top-down description of the full
213 scaffolding data structure.
218 Data structure that identifies the task class and its config.
219 parent : `_PipelineScaffolding`
220 The parent data structure that will hold the instance being
222 datasetTypes : `TaskDatasetTypes`
223 Data structure that categorizes the dataset types used by this task.
228 Raised if the task's dimensions are not a subset of the union of the
229 pipeline's dataset dimensions.
231 def __init__(self, taskDef: TaskDef, parent: _PipelineScaffolding, datasetTypes: TaskDatasetTypes):
232 universe = parent.dimensions.universe
234 self.
dimensions = DimensionGraph(universe, names=taskDef.connections.dimensions)
235 if not self.
dimensions.issubset(parent.dimensions):
237 f
"{self.dimensions} that are not a subset of "
238 f
"the pipeline dimensions {parent.dimensions}.")
242 self.
initInputs = _DatasetScaffoldingDict.fromSubset(datasetTypes.initInputs,
243 parent.initInputs, parent.initIntermediates)
244 self.
initOutputs = _DatasetScaffoldingDict.fromSubset(datasetTypes.initOutputs,
245 parent.initIntermediates, parent.initOutputs)
246 self.
inputs = _DatasetScaffoldingDict.fromSubset(datasetTypes.inputs,
247 parent.inputs, parent.intermediates)
248 self.
outputs = _DatasetScaffoldingDict.fromSubset(datasetTypes.outputs,
249 parent.intermediates, parent.outputs)
250 self.
prerequisites = _DatasetScaffoldingDict.fromSubset(datasetTypes.prerequisites,
251 parent.prerequisites)
254 for dataset
in itertools.chain(self.
initInputs.values(), self.
inputs.values(),
256 dataset.consumers[self.
taskDef.label] = self
258 assert dataset.producer
is None
259 dataset.producer = self
264 """Data structure that identifies the task class and its config
268 dimensions: DimensionGraph
269 """The dimensions of a single `Quantum` of this task (`DimensionGraph`).
272 initInputs: _DatasetScaffoldingDict
273 """Dictionary containing information about datasets used to construct this
274 task (`_DatasetScaffoldingDict`).
277 initOutputs: _DatasetScaffoldingDict
278 """Dictionary containing information about datasets produced as a
279 side-effect of constructing this task (`_DatasetScaffoldingDict`).
282 inputs: _DatasetScaffoldingDict
283 """Dictionary containing information about datasets used as regular,
284 graph-constraining inputs to this task (`_DatasetScaffoldingDict`).
287 outputs: _DatasetScaffoldingDict
288 """Dictionary containing information about datasets produced by this task
289 (`_DatasetScaffoldingDict`).
292 prerequisites: _DatasetScaffoldingDict
293 """Dictionary containing information about input datasets that must be
294 present in the repository before any Pipeline containing this task is run
295 (`_DatasetScaffoldingDict`).
298 dataIds: Set[ExpandedDataCoordinate]
299 """Data IDs for all quanta for this task in the graph (`set` of
300 `ExpandedDataCoordinate`).
302 Populated after construction by `_PipelineScaffolding.fillDataIds`.
305 quanta: List[Quantum]
306 """All quanta for this task in the graph (`list` of `Quantum`).
308 Populated after construction by `_PipelineScaffolding.fillQuanta`.
313 connectionClass = config.connections.ConnectionsClass
314 connectionInstance = connectionClass(config=config)
317 result = connectionInstance.adjustQuantum(quantum.predictedInputs)
318 quantum._predictedInputs = NamedKeyDict(result)
321 self.
quanta.append(quantum)
324 """Create a `QuantumGraphTaskNodes` instance from the information in
329 nodes : `QuantumGraphTaskNodes`
330 The `QuantumGraph` elements corresponding to this task.
342 """A helper data structure that organizes the information involved in
343 constructing a `QuantumGraph` for a `Pipeline`.
347 pipeline : `Pipeline`
348 Sequence of tasks from which a graph is to be constructed. Must
349 have nested task classes already imported.
350 universe : `DimensionUniverse`
351 Universe of all possible dimensions.
356 Raised if the task's dimensions are not a subset of the union of the
357 pipeline's dataset dimensions.
361 The scaffolding data structure contains nested data structures for both
362 tasks (`_TaskScaffolding`) and datasets (`_DatasetScaffolding`), with the
363 latter held by `_DatasetScaffoldingDict`. The dataset data structures are
364 shared between the pipeline-level structure (which aggregates all datasets
365 and categorizes them from the perspective of the complete pipeline) and the
366 individual tasks that use them as inputs and outputs.
368 `QuantumGraph` construction proceeds in five steps, with each corresponding
369 to a different `_PipelineScaffolding` method:
371 1. When `_PipelineScaffolding` is constructed, we extract and categorize
372 the DatasetTypes used by the pipeline (delegating to
373 `PipelineDatasetTypes.fromPipeline`), then use these to construct the
374 nested `_TaskScaffolding` and `_DatasetScaffolding` objects.
376 2. In `fillDataIds`, we construct and run the "Big Join Query", which
377 returns related tuples of all dimensions used to identify any regular
378 input, output, and intermediate datasets (not prerequisites). We then
379 iterate over these tuples of related dimensions, identifying the subsets
380 that correspond to distinct data IDs for each task and dataset type.
382 3. In `fillDatasetRefs`, we run follow-up queries against all of the
383 dataset data IDs previously identified, populating the
384 `_DatasetScaffolding.refs` lists - except for those for prerequisite
385 datasets, which cannot be resolved until distinct quanta are
388 4. In `fillQuanta`, we extract subsets from the lists of `DatasetRef` into
389 the inputs and outputs for each `Quantum` and search for prerequisite
390 datasets, populating `_TaskScaffolding.quanta`.
392 5. In `makeQuantumGraph`, we construct a `QuantumGraph` from the lists of
393 per-task quanta identified in the previous step.
398 datasetTypes = PipelineDatasetTypes.fromPipeline(pipeline, registry=registry)
401 for attr
in (
"initInputs",
"initIntermediates",
"initOutputs",
402 "inputs",
"intermediates",
"outputs",
"prerequisites"):
403 setattr(self, attr, _DatasetScaffoldingDict.fromDatasetTypes(getattr(datasetTypes, attr),
404 universe=registry.dimensions))
407 self.
dimensions = self.inputs.dimensions.union(self.intermediates.dimensions,
408 self.outputs.dimensions)
413 if isinstance(pipeline, Pipeline):
414 pipeline = pipeline.toExpandedPipeline()
416 for taskDef, taskDatasetTypes
in zip(pipeline,
417 datasetTypes.byTask.values())]
419 tasks: List[_TaskScaffolding]
420 """Scaffolding data structures for each task in the pipeline
421 (`list` of `_TaskScaffolding`).
424 initInputs: _DatasetScaffoldingDict
425 """Datasets consumed but not produced when constructing the tasks in this
426 pipeline (`_DatasetScaffoldingDict`).
429 initIntermediates: _DatasetScaffoldingDict
430 """Datasets that are both consumed and produced when constructing the tasks
431 in this pipeline (`_DatasetScaffoldingDict`).
434 initOutputs: _DatasetScaffoldingDict
435 """Datasets produced but not consumed when constructing the tasks in this
436 pipeline (`_DatasetScaffoldingDict`).
439 inputs: _DatasetScaffoldingDict
440 """Datasets that are consumed but not produced when running this pipeline
441 (`_DatasetScaffoldingDict`).
444 intermediates: _DatasetScaffoldingDict
445 """Datasets that are both produced and consumed when running this pipeline
446 (`_DatasetScaffoldingDict`).
449 outputs: _DatasetScaffoldingDict
450 """Datasets produced but not consumed when when running this pipeline
451 (`_DatasetScaffoldingDict`).
454 prerequisites: _DatasetScaffoldingDict
455 """Datasets that are consumed when running this pipeline and looked up
456 per-Quantum when generating the graph (`_DatasetScaffoldingDict`).
459 dimensions: DimensionGraph
460 """All dimensions used by any regular input, intermediate, or output
461 (not prerequisite) dataset; the set of dimension used in the "Big Join
462 Query" (`DimensionGraph`).
464 This is required to be a superset of all task quantum dimensions.
468 """Query for the data IDs that connect nodes in the `QuantumGraph`.
470 This method populates `_TaskScaffolding.dataIds` and
471 `_DatasetScaffolding.dataIds` (except for those in `prerequisites`).
475 registry : `lsst.daf.butler.Registry`
476 Registry for the data repository; used for all data ID queries.
477 inputCollections : `~collections.abc.Mapping`
478 Mapping from dataset type name to an ordered sequence of
479 collections to search for that dataset. A `defaultdict` is
480 recommended for the case where the same collections should be
481 used for most datasets.
482 userQuery : `str`, optional
483 User-provided expression to limit the data IDs processed.
486 emptyDataId = ExpandedDataCoordinate(registry.dimensions.empty, (), records={})
487 for scaffolding
in itertools.chain(self.initInputs.values(),
488 self.initIntermediates.values(),
489 self.initOutputs.values()):
490 scaffolding.dataIds.add(emptyDataId)
495 resultIter = registry.queryDimensions(
498 datasetType: inputCollections[datasetType.name]
499 for datasetType
in self.inputs
516 for commonDataId
in resultIter:
517 for taskScaffolding
in self.
tasks:
518 taskScaffolding.dataIds.add(commonDataId.subset(taskScaffolding.dimensions))
519 for datasetType, scaffolding
in itertools.chain(self.inputs.items(),
520 self.intermediates.items(),
521 self.outputs.items()):
522 scaffolding.dataIds.add(commonDataId.subset(scaffolding.dimensions))
525 skipExisting=True, clobberExisting=False):
526 """Perform follow up queries for each dataset data ID produced in
529 This method populates `_DatasetScaffolding.refs` (except for those in
534 registry : `lsst.daf.butler.Registry`
535 Registry for the data repository; used for all data ID queries.
536 inputCollections : `~collections.abc.Mapping`
537 Mapping from dataset type name to an ordered sequence of
538 collections to search for that dataset. A `defaultdict` is
539 recommended for the case where the same collections should be
540 used for most datasets.
541 outputCollection : `str`
542 Collection for all output datasets.
543 skipExisting : `bool`, optional
544 If `True` (default), a Quantum is not created if all its outputs
546 clobberExisting : `bool`, optional
547 If `True`, overwrite any outputs that already exist. Cannot be
548 `True` if ``skipExisting`` is.
553 Raised if both `skipExisting` and `clobberExisting` are `True`.
555 Raised if an output dataset already exists in the output collection
556 and both ``skipExisting`` and ``clobberExisting`` are `False`. The
557 case where some but not all of a quantum's outputs are present and
558 ``skipExisting`` is `True` cannot be identified at this stage, and
559 is handled by `fillQuanta` instead.
561 if clobberExisting
and skipExisting:
562 raise ValueError(
"clobberExisting and skipExisting cannot both be true.")
564 for datasetType, scaffolding
in itertools.chain(self.initInputs.items(), self.inputs.items()):
565 for dataId
in scaffolding.dataIds:
567 registry.queryDatasets(
569 collections=inputCollections[datasetType.name],
575 assert len(refs) == 1,
"BJQ guarantees exactly one input for each data ID."
576 scaffolding.refs.extend(refs)
580 for datasetType, scaffolding
in itertools.chain(self.initIntermediates.items(),
581 self.initOutputs.items(),
582 self.intermediates.items(),
583 self.outputs.items()):
584 for dataId
in scaffolding.dataIds:
593 ref = registry.find(collection=outputCollection, datasetType=datasetType, dataId=dataId)
595 ref = DatasetRef(datasetType, dataId)
596 elif not skipExisting:
598 f
"output collection {outputCollection} with data ID {dataId}.")
599 scaffolding.refs.append(ref)
602 def fillQuanta(self, registry, inputCollections, *, skipExisting=True):
603 """Define quanta for each task by splitting up the datasets associated
604 with each task data ID.
606 This method populates `_TaskScaffolding.quanta`.
610 registry : `lsst.daf.butler.Registry`
611 Registry for the data repository; used for all data ID queries.
612 inputCollections : `~collections.abc.Mapping`
613 Mapping from dataset type name to an ordered sequence of
614 collections to search for that dataset. A `defaultdict` is
615 recommended for the case where the same collections should be
616 used for most datasets.
617 skipExisting : `bool`, optional
618 If `True` (default), a Quantum is not created if all its outputs
621 for task
in self.
tasks:
622 for quantumDataId
in task.dataIds:
629 inputs = NamedKeyDict()
630 for datasetType, scaffolding
in task.inputs.items():
631 inputs[datasetType] = [ref
for ref, dataId
in zip(scaffolding.refs, scaffolding.dataIds)
632 if quantumDataId.matches(dataId)]
634 outputs = NamedKeyDict()
635 allOutputsPresent =
True
636 for datasetType, scaffolding
in task.outputs.items():
637 outputs[datasetType] = []
638 for ref, dataId
in zip(scaffolding.refs, scaffolding.dataIds):
639 if quantumDataId.matches(dataId):
641 allOutputsPresent =
False
643 assert skipExisting,
"Existing outputs should have already been identified."
644 if not allOutputsPresent:
646 f
"{dataId} already exists, but other outputs "
647 f
"for task with label {task.taskDef.label} "
648 f
"and data ID {quantumDataId} do not.")
649 outputs[datasetType].append(ref)
650 if allOutputsPresent
and skipExisting:
661 connections = task.taskDef.connections
662 for con_name
in connections.prerequisiteInputs:
663 con = getattr(connections, con_name)
664 for datasetType
in task.prerequisites:
665 if datasetType.name == con.name:
667 if con.lookupFunction
is not None:
668 refs = list(con.lookupFunction(datasetType, registry,
669 quantumDataId, inputCollections))
672 registry.queryDatasets(
674 collections=inputCollections[con.name],
675 dataId=quantumDataId,
680 inputs[datasetType] = refs
684 taskName=task.taskDef.taskName,
685 taskClass=task.taskDef.taskClass,
686 dataId=quantumDataId,
687 initInputs=task.initInputs.unpackRefs(),
688 predictedInputs=inputs,
694 """Create a `QuantumGraph` from the quanta already present in
695 the scaffolding data structure.
698 graph.initInputs = self.initInputs.unpackRefs()
699 graph.initOutputs = self.initOutputs.unpackRefs()
700 graph.initIntermediates = self.initIntermediates.unpackRefs()
710 """Base class for exceptions generated by graph builder.
715 class OutputExistsError(GraphBuilderError):
716 """Exception generated when output datasets already exist.
722 """Exception generated when a prerequisite dataset does not exist.
728 """GraphBuilder class is responsible for building task execution graph from
733 registry : `~lsst.daf.butler.Registry`
734 Data butler instance.
735 skipExisting : `bool`, optional
736 If `True` (default), a Quantum is not created if all its outputs
738 clobberExisting : `bool`, optional
739 If `True`, overwrite any outputs that already exist. Cannot be
740 `True` if ``skipExisting`` is.
743 def __init__(self, registry, skipExisting=True, clobberExisting=False):
749 def makeGraph(self, pipeline, inputCollections, outputCollection, userQuery):
750 """Create execution graph for a pipeline.
754 pipeline : `Pipeline`
755 Pipeline definition, task names/classes and their configs.
756 inputCollections : `~collections.abc.Mapping`
757 Mapping from dataset type name to an ordered sequence of
758 collections to search for that dataset. A `defaultdict` is
759 recommended for the case where the same collections should be
760 used for most datasets.
761 outputCollection : `str`
762 Collection for all output datasets.
764 String which defunes user-defined selection for registry, should be
765 empty or `None` if there is no restrictions on data selection.
769 graph : `QuantumGraph`
774 Raised when user expression cannot be parsed.
776 Raised when output datasets already exist.
778 Other exceptions types may be raised by underlying registry
783 scaffolding.fillDataIds(self.
registry, inputCollections, userQuery)
784 scaffolding.fillDatasetRefs(self.
registry, inputCollections, outputCollection,
787 scaffolding.fillQuanta(self.
registry, inputCollections,
790 return scaffolding.makeQuantumGraph()