Coverage for python/lsst/daf/butler/_quantum_backed.py: 25%
164 statements
« prev ^ index » next coverage.py v6.5.0, created at 2022-10-29 02:20 -0700
« prev ^ index » next coverage.py v6.5.0, created at 2022-10-29 02:20 -0700
1# This file is part of daf_butler.
2#
3# Developed for the LSST Data Management System.
4# This product includes software developed by the LSST Project
5# (http://www.lsst.org).
6# See the COPYRIGHT file at the top-level directory of this distribution
7# for details of code ownership.
8#
9# This program is free software: you can redistribute it and/or modify
10# it under the terms of the GNU General Public License as published by
11# the Free Software Foundation, either version 3 of the License, or
12# (at your option) any later version.
13#
14# This program is distributed in the hope that it will be useful,
15# but WITHOUT ANY WARRANTY; without even the implied warranty of
16# MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
17# GNU General Public License for more details.
18#
19# You should have received a copy of the GNU General Public License
20# along with this program. If not, see <http://www.gnu.org/licenses/>.
22from __future__ import annotations
24__all__ = ("QuantumBackedButler", "QuantumProvenanceData")
26import itertools
27import logging
28import uuid
29from collections import defaultdict
30from typing import TYPE_CHECKING, Any, Dict, Iterable, List, Mapping, Optional, Set, Type, Union
32from pydantic import BaseModel
34from ._butlerConfig import ButlerConfig
35from ._deferredDatasetHandle import DeferredDatasetHandle
36from ._limited_butler import LimitedButler
37from .core import (
38 Config,
39 DatasetId,
40 DatasetRef,
41 Datastore,
42 DatastoreRecordData,
43 DimensionUniverse,
44 Quantum,
45 SerializedDatastoreRecordData,
46 StorageClass,
47 StorageClassFactory,
48 ddl,
49)
50from .registry.bridge.monolithic import MonolithicDatastoreRegistryBridgeManager
51from .registry.databases.sqlite import SqliteDatabase
52from .registry.interfaces import DatastoreRegistryBridgeManager, OpaqueTableStorageManager
53from .registry.opaque import ByNameOpaqueTableStorageManager
55if TYPE_CHECKING: 55 ↛ 56line 55 didn't jump to line 56, because the condition on line 55 was never true
56 from ._butler import Butler
58_LOG = logging.getLogger(__name__)
61class _DatasetRecordStorageManagerDatastoreConstructionMimic:
62 """A partial implementation of `DatasetRecordStorageManager` that exists
63 only to allow a `DatastoreRegistryBridgeManager` (and hence a `Datastore`)
64 to be constructed without a full `Registry`.
66 Notes
67 -----
68 The interface implemented by this class should probably be its own ABC,
69 and that ABC should probably be used in the definition of
70 `DatastoreRegistryBridgeManager`, but while prototyping I'm trying to keep
71 changes minimal.
72 """
74 @classmethod
75 def getIdColumnType(cls) -> type:
76 # Docstring inherited.
77 return ddl.GUID
79 @classmethod
80 def addDatasetForeignKey(
81 cls,
82 tableSpec: ddl.TableSpec,
83 *,
84 name: str = "dataset",
85 constraint: bool = True,
86 onDelete: Optional[str] = None,
87 **kwargs: Any,
88 ) -> ddl.FieldSpec:
89 # Docstring inherited.
90 idFieldSpec = ddl.FieldSpec(f"{name}_id", dtype=ddl.GUID, **kwargs)
91 tableSpec.fields.add(idFieldSpec)
92 return idFieldSpec
95class QuantumBackedButler(LimitedButler):
96 """An implementation of `LimitedButler` intended to back execution of a
97 single `Quantum`.
99 Parameters
100 ----------
101 quantum : `Quantum`
102 Object describing the predicted input and output dataset relevant to
103 this butler. This must have resolved `DatasetRef` instances for all
104 inputs and outputs.
105 dimensions : `DimensionUniverse`
106 Object managing all dimension definitions.
107 datastore : `Datastore`
108 Datastore to use for all dataset I/O and existence checks.
109 storageClasses : `StorageClassFactory`
110 Object managing all storage class definitions.
112 Notes
113 -----
114 Most callers should use the `initialize` `classmethod` to construct new
115 instances instead of calling the constructor directly.
117 `QuantumBackedButler` uses a SQLite database internally, in order to reuse
118 existing `DatastoreRegistryBridge` and `OpaqueTableStorage`
119 implementations that rely SQLAlchemy. If implementations are added in the
120 future that don't rely on SQLAlchemy, it should be possible to swap them
121 in by overriding the type arguments to `initialize` (though at present,
122 `QuantumBackedButler` would still create at least an in-memory SQLite
123 database that would then go unused).`
125 We imagine `QuantumBackedButler` being used during (at least) batch
126 execution to capture `Datastore` records and save them to per-quantum
127 files, which are also a convenient place to store provenance for eventual
128 upload to a SQL-backed `Registry` (once `Registry` has tables to store
129 provenance, that is).
130 These per-quantum files can be written in two ways:
132 - The SQLite file used internally by `QuantumBackedButler` can be used
133 directly but customizing the ``filename`` argument to ``initialize``, and
134 then transferring that file to the object store after execution completes
135 (or fails; a ``try/finally`` pattern probably makes sense here).
137 - A JSON or YAML file can be written by calling `extract_provenance_data`,
138 and using ``pydantic`` methods to write the returned
139 `QuantumProvenanceData` to a file.
141 Note that at present, the SQLite file only contains datastore records, not
142 provenance, but that should be easy to address (if desired) after we
143 actually design a `Registry` schema for provenance. I also suspect that
144 we'll want to explicitly close the SQLite file somehow before trying to
145 transfer it. But I'm guessing we'd prefer to write the per-quantum files
146 as JSON anyway.
147 """
149 def __init__(
150 self,
151 quantum: Quantum,
152 dimensions: DimensionUniverse,
153 datastore: Datastore,
154 storageClasses: StorageClassFactory,
155 ):
156 self._quantum = quantum
157 self._dimensions = dimensions
158 self._predicted_inputs: Set[DatasetId] = {
159 ref.getCheckedId() for ref in itertools.chain.from_iterable(quantum.inputs.values())
160 }
161 self._predicted_outputs: Set[DatasetId] = {
162 ref.getCheckedId() for ref in itertools.chain.from_iterable(quantum.outputs.values())
163 }
164 self._available_inputs: Set[DatasetId] = set()
165 self._unavailable_inputs: Set[DatasetId] = set()
166 self._actual_inputs: Set[DatasetId] = set()
167 self._actual_output_refs: Set[DatasetRef] = set()
168 self.datastore = datastore
169 self.storageClasses = storageClasses
171 @classmethod
172 def initialize(
173 cls,
174 config: Union[Config, str],
175 quantum: Quantum,
176 dimensions: DimensionUniverse,
177 filename: str = ":memory:",
178 OpaqueManagerClass: Type[OpaqueTableStorageManager] = ByNameOpaqueTableStorageManager,
179 BridgeManagerClass: Type[DatastoreRegistryBridgeManager] = MonolithicDatastoreRegistryBridgeManager,
180 search_paths: Optional[List[str]] = None,
181 ) -> QuantumBackedButler:
182 """Construct a new `QuantumBackedButler` from repository configuration
183 and helper types.
185 Parameters
186 ----------
187 config : `Config` or `str`
188 A butler repository root, configuration filename, or configuration
189 instance.
190 quantum : `Quantum`
191 Object describing the predicted input and output dataset relevant
192 to this butler. This must have resolved `DatasetRef` instances for
193 all inputs and outputs.
194 dimensions : `DimensionUniverse`
195 Object managing all dimension definitions.
196 filename : `str`, optional
197 Name for the SQLite database that will back this butler; defaults
198 to an in-memory database.
199 OpaqueManagerClass : `type`, optional
200 A subclass of `OpaqueTableStorageManager` to use for datastore
201 opaque records. Default is a SQL-backed implementation.
202 BridgeManagerClass : `type`, optional
203 A subclass of `DatastoreRegistryBridgeManager` to use for datastore
204 location records. Default is a SQL-backed implementation.
205 search_paths : `list` of `str`, optional
206 Additional search paths for butler configuration.
207 """
208 butler_config = ButlerConfig(config, searchPaths=search_paths)
209 if "root" in butler_config:
210 butler_root = butler_config["root"]
211 else:
212 butler_root = butler_config.configDir
213 db = SqliteDatabase.fromUri(f"sqlite:///{filename}", origin=0)
214 with db.declareStaticTables(create=True) as context:
215 opaque_manager = OpaqueManagerClass.initialize(db, context)
216 bridge_manager = BridgeManagerClass.initialize(
217 db,
218 context,
219 opaque=opaque_manager,
220 # MyPy can tell it's a fake, but we know it shouldn't care.
221 datasets=_DatasetRecordStorageManagerDatastoreConstructionMimic, # type: ignore
222 universe=dimensions,
223 )
224 # TODO: We need to inform `Datastore` here that it needs to support
225 # predictive reads; right now that's a configuration option, but after
226 # execution butler is retired it could just be a kwarg we pass here.
227 # For now just force this option as we cannot work without it.
228 butler_config["datastore", "trust_get_request"] = True
229 datastore = Datastore.fromConfig(butler_config, bridge_manager, butler_root)
230 datastore.import_records(quantum.datastore_records)
231 storageClasses = StorageClassFactory()
232 storageClasses.addFromConfig(butler_config)
233 return cls(quantum, dimensions, datastore, storageClasses=storageClasses)
235 def isWriteable(self) -> bool:
236 # Docstring inherited.
237 return True
239 def getDirect(
240 self,
241 ref: DatasetRef,
242 *,
243 parameters: Optional[Dict[str, Any]] = None,
244 storageClass: str | StorageClass | None = None,
245 ) -> Any:
246 # Docstring inherited.
247 try:
248 obj = super().getDirect(ref, parameters=parameters, storageClass=storageClass)
249 except (LookupError, FileNotFoundError, IOError):
250 self._unavailable_inputs.add(ref.getCheckedId())
251 raise
252 if ref.id in self._predicted_inputs:
253 # do this after delegating to super in case that raises.
254 self._actual_inputs.add(ref.id)
255 self._available_inputs.add(ref.id)
256 return obj
258 def getDirectDeferred(
259 self,
260 ref: DatasetRef,
261 *,
262 parameters: Union[dict, None] = None,
263 storageClass: str | StorageClass | None = None,
264 ) -> DeferredDatasetHandle:
265 # Docstring inherited.
266 if ref.id in self._predicted_inputs:
267 # Unfortunately, we can't do this after the handle succeeds in
268 # loading, so it's conceivable here that we're marking an input
269 # as "actual" even when it's not even available.
270 self._actual_inputs.add(ref.id)
271 return super().getDirectDeferred(ref, parameters=parameters, storageClass=storageClass)
273 def datasetExistsDirect(self, ref: DatasetRef) -> bool:
274 # Docstring inherited.
275 exists = super().datasetExistsDirect(ref)
276 if ref.id in self._predicted_inputs:
277 if exists:
278 self._available_inputs.add(ref.id)
279 else:
280 self._unavailable_inputs.add(ref.id)
281 return exists
283 def markInputUnused(self, ref: DatasetRef) -> None:
284 # Docstring inherited.
285 self._actual_inputs.discard(ref.getCheckedId())
287 @property
288 def dimensions(self) -> DimensionUniverse:
289 # Docstring inherited.
290 return self._dimensions
292 def putDirect(self, obj: Any, ref: DatasetRef) -> DatasetRef:
293 # Docstring inherited.
294 if ref.id not in self._predicted_outputs:
295 raise RuntimeError("Cannot `put` dataset that was not predicted as an output.")
296 self.datastore.put(obj, ref)
297 self._actual_output_refs.add(ref)
298 return ref
300 def pruneDatasets(
301 self,
302 refs: Iterable[DatasetRef],
303 *,
304 disassociate: bool = True,
305 unstore: bool = False,
306 tags: Iterable[str] = (),
307 purge: bool = False,
308 ) -> None:
309 # docstring inherited from LimitedButler
311 if purge:
312 if not disassociate:
313 raise TypeError("Cannot pass purge=True without disassociate=True.")
314 if not unstore:
315 raise TypeError("Cannot pass purge=True without unstore=True.")
316 elif disassociate:
317 # No tagged collections for this butler.
318 raise TypeError("Cannot pass disassociate=True without purge=True.")
320 refs = list(refs)
322 # Pruning a component of a DatasetRef makes no sense.
323 for ref in refs:
324 if ref.datasetType.component():
325 raise ValueError(f"Can not prune a component of a dataset (ref={ref})")
327 if unstore:
328 self.datastore.trash(refs)
329 if purge:
330 for ref in refs:
331 # We only care about removing them from actual output refs,
332 self._actual_output_refs.discard(ref)
334 if unstore:
335 # Point of no return for removing artifacts
336 self.datastore.emptyTrash()
338 def extract_provenance_data(self) -> QuantumProvenanceData:
339 """Extract provenance information and datastore records from this
340 butler.
342 Returns
343 -------
344 provenance : `QuantumProvenanceData`
345 A serializable struct containing input/output dataset IDs and
346 datastore records. This assumes all dataset IDs are UUIDs (just to
347 make it easier for `pydantic` to reason about the struct's types);
348 the rest of this class makes no such assumption, but the approach
349 to processing in which it's useful effectively requires UUIDs
350 anyway.
352 Notes
353 -----
354 `QuantumBackedButler` records this provenance information when its
355 methods are used, which mostly saves `~lsst.pipe.base.PipelineTask`
356 authors from having to worry about while still recording very
357 detailed information. But it has two small weaknesses:
359 - Calling `getDirectDeferred` or `getDirect` is enough to mark a
360 dataset as an "actual input", which may mark some datasets that
361 aren't actually used. We rely on task authors to use
362 `markInputUnused` to address this.
364 - We assume that the execution system will call ``datasetExistsDirect``
365 on all predicted inputs prior to execution, in order to populate the
366 "available inputs" set. This is what I envision
367 '`~lsst.ctrl.mpexec.SingleQuantumExecutor` doing after we update it
368 to use this class, but it feels fragile for this class to make such
369 a strong assumption about how it will be used, even if I can't think
370 of any other executor behavior that would make sense.
371 """
372 if not self._actual_inputs.isdisjoint(self._unavailable_inputs):
373 _LOG.warning(
374 "Inputs %s were marked as actually used (probably because a DeferredDatasetHandle) "
375 "was obtained, but did not actually exist. This task should be be using markInputUnused "
376 "directly to clarify its provenance.",
377 self._actual_inputs & self._unavailable_inputs,
378 )
379 self._actual_inputs -= self._unavailable_inputs
380 checked_inputs = self._available_inputs | self._unavailable_inputs
381 if not self._predicted_inputs == checked_inputs:
382 _LOG.warning(
383 "Execution harness did not check predicted inputs %s for existence; available inputs "
384 "recorded in provenance may be incomplete.",
385 self._predicted_inputs - checked_inputs,
386 )
387 datastore_records = self.datastore.export_records(self._actual_output_refs)
388 provenance_records = {
389 datastore_name: records.to_simple() for datastore_name, records in datastore_records.items()
390 }
392 return QuantumProvenanceData(
393 predicted_inputs=self._predicted_inputs,
394 available_inputs=self._available_inputs,
395 actual_inputs=self._actual_inputs,
396 predicted_outputs=self._predicted_outputs,
397 actual_outputs={ref.getCheckedId() for ref in self._actual_output_refs},
398 datastore_records=provenance_records,
399 )
402class QuantumProvenanceData(BaseModel):
403 """A serializable struct for per-quantum provenance information and
404 datastore records.
406 Notes
407 -----
408 This class slightly duplicates information from the `Quantum` class itself
409 (the `predicted_inputs` and `predicted_outputs` sets should have the same
410 IDs present in `Quantum.inputs` and `Quantum.outputs`), but overall it
411 assumes the original `Quantum` is also available to reconstruct the
412 complete provenance (e.g. by associating dataset IDs with data IDs,
413 dataset types, and `~CollectionType.RUN` names.
415 Note that ``pydantic`` method ``parse_raw()`` is not going to work
416 correctly for this class, use `direct` method instead.
417 """
419 # This class probably should have information about its execution
420 # environment (anything not controlled and recorded at the
421 # `~CollectionType.RUN` level, such as the compute node ID). but adding it
422 # now is out of scope for this prototype.
424 predicted_inputs: Set[uuid.UUID]
425 """Unique IDs of datasets that were predicted as inputs to this quantum
426 when the QuantumGraph was built.
427 """
429 available_inputs: Set[uuid.UUID]
430 """Unique IDs of input datasets that were actually present in the datastore
431 when this quantum was executed.
433 This is a subset of `predicted_inputs`, with the difference generally being
434 datasets were `predicted_outputs` but not `actual_outputs` of some upstream
435 task.
436 """
438 actual_inputs: Set[uuid.UUID]
439 """Unique IDs of datasets that were actually used as inputs by this task.
441 This is a subset of `available_inputs`.
443 Notes
444 -----
445 The criteria for marking an input as used is that rerunning the quantum
446 with only these `actual_inputs` available must yield identical outputs.
447 This means that (for example) even just using an input to help determine
448 an output rejection criteria and then rejecting it as an outlier qualifies
449 that input as actually used.
450 """
452 predicted_outputs: Set[uuid.UUID]
453 """Unique IDs of datasets that were predicted as outputs of this quantum
454 when the QuantumGraph was built.
455 """
457 actual_outputs: Set[uuid.UUID]
458 """Unique IDs of datasets that were actually written when this quantum
459 was executed.
460 """
462 datastore_records: Dict[str, SerializedDatastoreRecordData]
463 """Datastore records indexed by datastore name."""
465 @staticmethod
466 def collect_and_transfer(
467 butler: Butler, quanta: Iterable[Quantum], provenance: Iterable[QuantumProvenanceData]
468 ) -> None:
469 """Transfer output datasets from multiple quanta to a more permantent
470 `Butler` repository.
472 Parameters
473 ----------
474 butler : `Butler`
475 Full butler representing the data repository to transfer datasets
476 to.
477 quanta : `Iterable` [ `Quantum` ]
478 Iterable of `Quantum` objects that carry information about
479 predicted outputs. May be a single-pass iterator.
480 provenance : `Iterable` [ `QuantumProvenanceData` ]
481 Provenance and datastore data for each of the given quanta, in the
482 same order. May be a single-pass iterator.
484 Notes
485 -----
486 Input-output provenance data is not actually transferred yet, because
487 `Registry` has no place to store it.
489 This method probably works most efficiently if run on all quanta for a
490 single task label at once, because this will gather all datasets of
491 a particular type together into a single vectorized `Registry` import.
492 It should still behave correctly if run on smaller groups of quanta
493 or even quanta from multiple tasks.
495 Currently this method transfers datastore record data unchanged, with
496 no possibility of actually moving (e.g.) files. Datastores that are
497 present only in execution or only in the more permanent butler are
498 ignored.
499 """
500 grouped_refs = defaultdict(list)
501 summary_records: Dict[str, DatastoreRecordData] = {}
502 for quantum, provenance_for_quantum in zip(quanta, provenance):
503 quantum_refs_by_id = {
504 ref.getCheckedId(): ref
505 for ref in itertools.chain.from_iterable(quantum.outputs.values())
506 if ref.getCheckedId() in provenance_for_quantum.actual_outputs
507 }
508 for ref in quantum_refs_by_id.values():
509 grouped_refs[ref.datasetType, ref.run].append(ref)
511 # merge datastore records into a summary structure
512 for datastore_name, serialized_records in provenance_for_quantum.datastore_records.items():
513 quantum_records = DatastoreRecordData.from_simple(serialized_records)
514 if (records := summary_records.get(datastore_name)) is not None:
515 records.update(quantum_records)
516 else:
517 summary_records[datastore_name] = quantum_records
519 for refs in grouped_refs.values():
520 butler.registry._importDatasets(refs)
521 butler.datastore.import_records(summary_records)
523 @classmethod
524 def parse_raw(cls, *args: Any, **kwargs: Any) -> QuantumProvenanceData:
525 raise NotImplementedError("parse_raw() is not usable for this class, use direct() instead.")
527 @classmethod
528 def direct(
529 cls,
530 *,
531 predicted_inputs: Iterable[Union[str, uuid.UUID]],
532 available_inputs: Iterable[Union[str, uuid.UUID]],
533 actual_inputs: Iterable[Union[str, uuid.UUID]],
534 predicted_outputs: Iterable[Union[str, uuid.UUID]],
535 actual_outputs: Iterable[Union[str, uuid.UUID]],
536 datastore_records: Mapping[str, Mapping],
537 ) -> QuantumProvenanceData:
538 """Construct an instance directly without validators.
540 This differs from the pydantic "construct" method in that the
541 arguments are explicitly what the model requires, and it will recurse
542 through members, constructing them from their corresponding `direct`
543 methods.
545 This method should only be called when the inputs are trusted.
546 """
548 def _to_uuid_set(uuids: Iterable[Union[str, uuid.UUID]]) -> Set[uuid.UUID]:
549 """Convert input UUIDs, which could be in string representation to
550 a set of `UUID` instances.
551 """
552 return set(uuid.UUID(id) if isinstance(id, str) else id for id in uuids)
554 data = QuantumProvenanceData.__new__(cls)
555 setter = object.__setattr__
556 setter(data, "predicted_inputs", _to_uuid_set(predicted_inputs))
557 setter(data, "available_inputs", _to_uuid_set(available_inputs))
558 setter(data, "actual_inputs", _to_uuid_set(actual_inputs))
559 setter(data, "predicted_outputs", _to_uuid_set(predicted_outputs))
560 setter(data, "actual_outputs", _to_uuid_set(actual_outputs))
561 setter(
562 data,
563 "datastore_records",
564 {
565 key: SerializedDatastoreRecordData.direct(**records)
566 for key, records in datastore_records.items()
567 },
568 )
569 return data