Hide keyboard shortcuts

Hot-keys on this page

r m x p   toggle line displays

j k   next/prev highlighted chunk

0   (zero) top of page

1   (one) first highlighted chunk

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/>. 

21from __future__ import annotations 

22 

23__all__ = ("QueryBuilder",) 

24 

25from typing import AbstractSet, Any, Iterable, List, Optional 

26 

27import sqlalchemy.sql 

28 

29from ...core import ( 

30 DimensionElement, 

31 SkyPixDimension, 

32 Dimension, 

33 DatasetType, 

34 SimpleQuery, 

35) 

36 

37from ...core.named import NamedKeyDict, NamedValueAbstractSet, NamedValueSet 

38 

39from .._collectionType import CollectionType 

40from ._structs import QuerySummary, QueryColumns, DatasetQueryColumns, RegistryManagers 

41from .expressions import convertExpressionToSql 

42from ._query import DirectQuery, DirectQueryUniqueness, EmptyQuery, Query 

43from ..wildcards import CollectionSearch, CollectionQuery 

44 

45 

46class QueryBuilder: 

47 """A builder for potentially complex queries that join tables based 

48 on dimension relationships. 

49 

50 Parameters 

51 ---------- 

52 summary : `QuerySummary` 

53 Struct organizing the dimensions involved in the query. 

54 managers : `RegistryManagers` 

55 A struct containing the registry manager instances used by the query 

56 system. 

57 """ 

58 def __init__(self, summary: QuerySummary, managers: RegistryManagers): 

59 self.summary = summary 

60 self._simpleQuery = SimpleQuery() 

61 self._elements: NamedKeyDict[DimensionElement, sqlalchemy.sql.FromClause] = NamedKeyDict() 

62 self._columns = QueryColumns() 

63 self._managers = managers 

64 

65 def hasDimensionKey(self, dimension: Dimension) -> bool: 

66 """Return `True` if the given dimension's primary key column has 

67 been included in the query (possibly via a foreign key column on some 

68 other table). 

69 """ 

70 return dimension in self._columns.keys 

71 

72 def joinDimensionElement(self, element: DimensionElement) -> None: 

73 """Add the table for a `DimensionElement` to the query. 

74 

75 This automatically joins the element table to all other tables in the 

76 query with which it is related, via both dimension keys and spatial 

77 and temporal relationships. 

78 

79 External calls to this method should rarely be necessary; `finish` will 

80 automatically call it if the `DimensionElement` has been identified as 

81 one that must be included. 

82 

83 Parameters 

84 ---------- 

85 element : `DimensionElement` 

86 Element for which a table should be added. The element must be 

87 associated with a database table (see `DimensionElement.hasTable`). 

88 """ 

89 assert element not in self._elements, "Element already included in query." 

90 storage = self._managers.dimensions[element] 

91 fromClause = storage.join( 

92 self, 

93 regions=self._columns.regions if element in self.summary.spatial else None, 

94 timespans=self._columns.timespans if element in self.summary.temporal else None, 

95 ) 

96 self._elements[element] = fromClause 

97 

98 def joinDataset(self, datasetType: DatasetType, collections: Any, *, 

99 isResult: bool = True, findFirst: bool = False) -> bool: 

100 """Add a dataset search or constraint to the query. 

101 

102 Unlike other `QueryBuilder` join methods, this *must* be called 

103 directly to search for datasets of a particular type or constrain the 

104 query results based on the exists of datasets. However, all dimensions 

105 used to identify the dataset type must have already been included in 

106 `QuerySummary.requested` when initializing the `QueryBuilder`. 

107 

108 Parameters 

109 ---------- 

110 datasetType : `DatasetType` 

111 The type of datasets to search for. 

112 collections : `Any` 

113 An expression that fully or partially identifies the collections 

114 to search for datasets, such as a `str`, `re.Pattern`, or iterable 

115 thereof. `...` can be used to return all collections. See 

116 :ref:`daf_butler_collection_expressions` for more information. 

117 isResult : `bool`, optional 

118 If `True` (default), include the dataset ID column in the 

119 result columns of the query, allowing complete `DatasetRef` 

120 instances to be produced from the query results for this dataset 

121 type. If `False`, the existence of datasets of this type is used 

122 only to constrain the data IDs returned by the query. 

123 `joinDataset` may be called with ``isResult=True`` at most one time 

124 on a particular `QueryBuilder` instance. 

125 findFirst : `bool`, optional 

126 If `True` (`False` is default), only include the first match for 

127 each data ID, searching the given collections in order. Requires 

128 that all entries in ``collections`` be regular strings, so there is 

129 a clear search order. Ignored if ``isResult`` is `False`. 

130 

131 Returns 

132 ------- 

133 anyRecords : `bool` 

134 If `True`, joining the dataset table was successful and the query 

135 should proceed. If `False`, we were able to determine (from the 

136 combination of ``datasetType`` and ``collections``) that there 

137 would be no results joined in from this dataset, and hence (due to 

138 the inner join that would normally be present), the full query will 

139 return no results. 

140 """ 

141 assert datasetType.dimensions.issubset(self.summary.requested) 

142 if isResult and findFirst: 

143 collections = CollectionSearch.fromExpression(collections) 

144 else: 

145 collections = CollectionQuery.fromExpression(collections) 

146 explicitCollections = frozenset(collections.explicitNames()) 

147 # If we are searching all collections with no constraints, loop over 

148 # RUN collections only, because that will include all datasets. 

149 collectionTypes: AbstractSet[CollectionType] 

150 if collections == CollectionQuery(): 

151 collectionTypes = {CollectionType.RUN} 

152 else: 

153 collectionTypes = CollectionType.all() 

154 datasetRecordStorage = self._managers.datasets.find(datasetType.name) 

155 if datasetRecordStorage is None: 

156 # Unrecognized dataset type means no results. It might be better 

157 # to raise here, but this is consistent with previous behavior, 

158 # which is expected by QuantumGraph generation code in pipe_base. 

159 return False 

160 subsubqueries = [] 

161 runKeyName = self._managers.collections.getRunForeignKeyName() 

162 baseColumnNames = {"id", runKeyName, "ingest_date"} if isResult else set() 

163 baseColumnNames.update(datasetType.dimensions.required.names) 

164 for rank, collectionRecord in enumerate(collections.iter(self._managers.collections, 

165 collectionTypes=collectionTypes)): 

166 if collectionRecord.type is CollectionType.CALIBRATION: 

167 # If collection name was provided explicitly then say sorry, 

168 # otherwise collection is a part of chained one and we skip it. 

169 if datasetType.isCalibration() and collectionRecord.name in explicitCollections: 

170 raise NotImplementedError( 

171 f"Query for dataset type '{datasetType.name}' in CALIBRATION-type collection " 

172 f"'{collectionRecord.name}' is not yet supported." 

173 ) 

174 else: 

175 # We can never find a non-calibration dataset in a 

176 # CALIBRATION collection. 

177 continue 

178 ssq = datasetRecordStorage.select(collection=collectionRecord, 

179 dataId=SimpleQuery.Select, 

180 id=SimpleQuery.Select if isResult else None, 

181 run=SimpleQuery.Select if isResult else None, 

182 ingestDate=SimpleQuery.Select if isResult else None) 

183 if ssq is None: 

184 continue 

185 assert {c.name for c in ssq.columns} == baseColumnNames 

186 if findFirst: 

187 ssq.columns.append(sqlalchemy.sql.literal(rank).label("rank")) 

188 subsubqueries.append(ssq.combine()) 

189 if not subsubqueries: 

190 return False 

191 subquery = sqlalchemy.sql.union_all(*subsubqueries) 

192 columns: Optional[DatasetQueryColumns] = None 

193 if isResult: 

194 if findFirst: 

195 # Rewrite the subquery (currently a UNION ALL over 

196 # per-collection subsubqueries) to select the rows with the 

197 # lowest rank per data ID. The block below will set subquery 

198 # to something like this: 

199 # 

200 # WITH {dst}_search AS ( 

201 # SELECT {data-id-cols}, id, run_id, 1 AS rank 

202 # FROM <collection1> 

203 # UNION ALL 

204 # SELECT {data-id-cols}, id, run_id, 2 AS rank 

205 # FROM <collection2> 

206 # UNION ALL 

207 # ... 

208 # ) 

209 # SELECT 

210 # {dst}_window.{data-id-cols}, 

211 # {dst}_window.id, 

212 # {dst}_window.run_id 

213 # FROM ( 

214 # SELECT 

215 # {dst}_search.{data-id-cols}, 

216 # {dst}_search.id, 

217 # {dst}_search.run_id, 

218 # ROW_NUMBER() OVER ( 

219 # PARTITION BY {dst_search}.{data-id-cols} 

220 # ORDER BY rank 

221 # ) AS rownum 

222 # ) {dst}_window 

223 # WHERE 

224 # {dst}_window.rownum = 1; 

225 # 

226 search = subquery.cte(f"{datasetType.name}_search") 

227 windowDataIdCols = [ 

228 search.columns[name].label(name) for name in datasetType.dimensions.required.names 

229 ] 

230 windowSelectCols = [ 

231 search.columns["id"].label("id"), 

232 search.columns[runKeyName].label(runKeyName), 

233 search.columns["ingest_date"].label("ingest_date"), 

234 ] 

235 windowSelectCols += windowDataIdCols 

236 assert {c.name for c in windowSelectCols} == baseColumnNames 

237 windowSelectCols.append( 

238 sqlalchemy.sql.func.row_number().over( 

239 partition_by=windowDataIdCols, 

240 order_by=search.columns["rank"] 

241 ).label("rownum") 

242 ) 

243 window = sqlalchemy.sql.select( 

244 windowSelectCols 

245 ).select_from(search).alias( 

246 f"{datasetType.name}_window" 

247 ) 

248 subquery = sqlalchemy.sql.select( 

249 [window.columns[name].label(name) for name in baseColumnNames] 

250 ).select_from( 

251 window 

252 ).where( 

253 window.columns["rownum"] == 1 

254 ).alias(datasetType.name) 

255 else: 

256 subquery = subquery.alias(datasetType.name) 

257 columns = DatasetQueryColumns( 

258 datasetType=datasetType, 

259 id=subquery.columns["id"], 

260 runKey=subquery.columns[runKeyName], 

261 ingestDate=subquery.columns["ingest_date"], 

262 ) 

263 else: 

264 subquery = subquery.alias(datasetType.name) 

265 self.joinTable(subquery, datasetType.dimensions.required, datasets=columns) 

266 return True 

267 

268 def joinTable(self, table: sqlalchemy.sql.FromClause, dimensions: NamedValueAbstractSet[Dimension], *, 

269 datasets: Optional[DatasetQueryColumns] = None) -> None: 

270 """Join an arbitrary table to the query via dimension relationships. 

271 

272 External calls to this method should only be necessary for tables whose 

273 records represent neither datasets nor dimension elements. 

274 

275 Parameters 

276 ---------- 

277 table : `sqlalchemy.sql.FromClause` 

278 SQLAlchemy object representing the logical table (which may be a 

279 join or subquery expression) to be joined. 

280 dimensions : iterable of `Dimension` 

281 The dimensions that relate this table to others that may be in the 

282 query. The table must have columns with the names of the 

283 dimensions. 

284 datasets : `DatasetQueryColumns`, optional 

285 Columns that identify a dataset that is part of the query results. 

286 """ 

287 unexpectedDimensions = NamedValueSet(dimensions - self.summary.mustHaveKeysJoined.dimensions) 

288 unexpectedDimensions.discard(self.summary.universe.commonSkyPix) 

289 if unexpectedDimensions: 

290 raise NotImplementedError( 

291 f"QueryBuilder does not yet support joining in dimensions {unexpectedDimensions} that " 

292 f"were not provided originally to the QuerySummary object passed at construction." 

293 ) 

294 joinOn = self.startJoin(table, dimensions, dimensions.names) 

295 self.finishJoin(table, joinOn) 

296 if datasets is not None: 

297 assert self._columns.datasets is None, \ 

298 "At most one result dataset type can be returned by a query." 

299 self._columns.datasets = datasets 

300 

301 def startJoin(self, table: sqlalchemy.sql.FromClause, dimensions: Iterable[Dimension], 

302 columnNames: Iterable[str] 

303 ) -> List[sqlalchemy.sql.ColumnElement]: 

304 """Begin a join on dimensions. 

305 

306 Must be followed by call to `finishJoin`. 

307 

308 Parameters 

309 ---------- 

310 table : `sqlalchemy.sql.FromClause` 

311 SQLAlchemy object representing the logical table (which may be a 

312 join or subquery expression) to be joined. 

313 dimensions : iterable of `Dimension` 

314 The dimensions that relate this table to others that may be in the 

315 query. The table must have columns with the names of the 

316 dimensions. 

317 columnNames : iterable of `str` 

318 Names of the columns that correspond to dimension key values; must 

319 be `zip` iterable with ``dimensions``. 

320 

321 Returns 

322 ------- 

323 joinOn : `list` of `sqlalchemy.sql.ColumnElement` 

324 Sequence of boolean expressions that should be combined with AND 

325 to form (part of) the ON expression for this JOIN. 

326 """ 

327 joinOn = [] 

328 for dimension, columnName in zip(dimensions, columnNames): 

329 columnInTable = table.columns[columnName] 

330 columnsInQuery = self._columns.keys.setdefault(dimension, []) 

331 for columnInQuery in columnsInQuery: 

332 joinOn.append(columnInQuery == columnInTable) 

333 columnsInQuery.append(columnInTable) 

334 return joinOn 

335 

336 def finishJoin(self, table: sqlalchemy.sql.FromClause, joinOn: List[sqlalchemy.sql.ColumnElement] 

337 ) -> None: 

338 """Complete a join on dimensions. 

339 

340 Must be preceded by call to `startJoin`. 

341 

342 Parameters 

343 ---------- 

344 table : `sqlalchemy.sql.FromClause` 

345 SQLAlchemy object representing the logical table (which may be a 

346 join or subquery expression) to be joined. Must be the same object 

347 passed to `startJoin`. 

348 joinOn : `list` of `sqlalchemy.sql.ColumnElement` 

349 Sequence of boolean expressions that should be combined with AND 

350 to form (part of) the ON expression for this JOIN. Should include 

351 at least the elements of the list returned by `startJoin`. 

352 """ 

353 onclause: Optional[sqlalchemy.sql.ColumnElement] 

354 if len(joinOn) == 0: 

355 onclause = None 

356 elif len(joinOn) == 1: 

357 onclause = joinOn[0] 

358 else: 

359 onclause = sqlalchemy.sql.and_(*joinOn) 

360 self._simpleQuery.join(table, onclause=onclause) 

361 

362 def _joinMissingDimensionElements(self) -> None: 

363 """Join all dimension element tables that were identified as necessary 

364 by `QuerySummary` and have not yet been joined. 

365 

366 For internal use by `QueryBuilder` only; will be called (and should 

367 only by called) by `finish`. 

368 """ 

369 # Join all DimensionElement tables that we need for spatial/temporal 

370 # joins/filters or a nontrivial WHERE expression. 

371 # We iterate over these in *reverse* topological order to minimize the 

372 # number of tables joined. For example, the "visit" table provides 

373 # the primary key value for the "instrument" table it depends on, so we 

374 # don't need to join "instrument" as well unless we had a nontrivial 

375 # expression on it (and hence included it already above). 

376 for element in self.summary.universe.sorted(self.summary.mustHaveTableJoined, reverse=True): 

377 self.joinDimensionElement(element) 

378 # Join in any requested Dimension tables that don't already have their 

379 # primary keys identified by the query. 

380 for dimension in self.summary.universe.sorted(self.summary.mustHaveKeysJoined, reverse=True): 

381 if dimension not in self._columns.keys: 

382 self.joinDimensionElement(dimension) 

383 

384 def _addWhereClause(self) -> None: 

385 """Add a WHERE clause to the query under construction, connecting all 

386 joined dimensions to the expression and data ID dimensions from 

387 `QuerySummary`. 

388 

389 For internal use by `QueryBuilder` only; will be called (and should 

390 only by called) by `finish`. 

391 """ 

392 if self.summary.where.tree is not None: 

393 self._simpleQuery.where.append( 

394 convertExpressionToSql( 

395 self.summary.where.tree, 

396 self.summary.universe, 

397 columns=self._columns, 

398 elements=self._elements, 

399 bind=self.summary.where.bind, 

400 TimespanReprClass=self._managers.TimespanReprClass, 

401 ) 

402 ) 

403 for dimension, columnsInQuery in self._columns.keys.items(): 

404 if dimension in self.summary.where.dataId.graph: 

405 givenKey = self.summary.where.dataId[dimension] 

406 # Add a WHERE term for each column that corresponds to each 

407 # key. This is redundant with the JOIN ON clauses that make 

408 # them equal to each other, but more constraints have a chance 

409 # of making things easier on the DB's query optimizer. 

410 for columnInQuery in columnsInQuery: 

411 self._simpleQuery.where.append(columnInQuery == givenKey) 

412 else: 

413 # Dimension is not fully identified, but it might be a skypix 

414 # dimension that's constrained by a given region. 

415 if self.summary.where.region is not None and isinstance(dimension, SkyPixDimension): 

416 # We know the region now. 

417 givenSkyPixIds: List[int] = [] 

418 for begin, end in dimension.pixelization.envelope(self.summary.where.region): 

419 givenSkyPixIds.extend(range(begin, end)) 

420 for columnInQuery in columnsInQuery: 

421 self._simpleQuery.where.append(columnInQuery.in_(givenSkyPixIds)) 

422 # If we are given an dataId with a timespan, and there are one or more 

423 # timespans in the query that aren't given, add a WHERE expression for 

424 # each of them. 

425 if self.summary.where.dataId.graph.temporal and self.summary.temporal: 

426 # Timespan is known now. 

427 givenInterval = self.summary.where.dataId.timespan 

428 assert givenInterval is not None 

429 for element, intervalInQuery in self._columns.timespans.items(): 

430 assert element not in self.summary.where.dataId.graph.elements 

431 self._simpleQuery.where.append( 

432 intervalInQuery.overlaps(self._managers.TimespanReprClass.fromLiteral(givenInterval)) 

433 ) 

434 

435 def finish(self, joinMissing: bool = True) -> Query: 

436 """Finish query constructing, returning a new `Query` instance. 

437 

438 Parameters 

439 ---------- 

440 joinMissing : `bool`, optional 

441 If `True` (default), automatically join any missing dimension 

442 element tables (according to the categorization of the 

443 `QuerySummary` the builder was constructed with). `False` should 

444 only be passed if the caller can independently guarantee that all 

445 dimension relationships are already captured in non-dimension 

446 tables that have been manually included in the query. 

447 

448 Returns 

449 ------- 

450 query : `Query` 

451 A `Query` object that can be executed and used to interpret result 

452 rows. 

453 """ 

454 if joinMissing: 

455 self._joinMissingDimensionElements() 

456 self._addWhereClause() 

457 if self._columns.isEmpty(): 

458 return EmptyQuery(self.summary.requested.universe, managers=self._managers) 

459 return DirectQuery(graph=self.summary.requested, 

460 uniqueness=DirectQueryUniqueness.NOT_UNIQUE, 

461 whereRegion=self.summary.where.dataId.region, 

462 simpleQuery=self._simpleQuery, 

463 columns=self._columns, 

464 managers=self._managers)