Coverage for python/lsst/daf/butler/registry/dimensions/static.py: 93%

144 statements  

« prev     ^ index     » next       coverage.py v6.5.0, created at 2023-04-04 02:05 -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/>. 

21from __future__ import annotations 

22 

23import itertools 

24from collections import defaultdict 

25from collections.abc import Mapping, Set 

26from typing import TYPE_CHECKING 

27 

28import sqlalchemy 

29from lsst.daf.relation import Relation 

30 

31from ...core import ( 

32 DatabaseDimensionElement, 

33 DatabaseTopologicalFamily, 

34 DimensionElement, 

35 DimensionGraph, 

36 DimensionUniverse, 

37 GovernorDimension, 

38 NamedKeyDict, 

39 SkyPixDimension, 

40 ddl, 

41) 

42from .._exceptions import MissingSpatialOverlapError 

43from ..interfaces import ( 

44 Database, 

45 DatabaseDimensionOverlapStorage, 

46 DatabaseDimensionRecordStorage, 

47 DimensionRecordStorage, 

48 DimensionRecordStorageManager, 

49 GovernorDimensionRecordStorage, 

50 StaticTablesContext, 

51 VersionTuple, 

52) 

53 

54if TYPE_CHECKING: 

55 from .. import queries 

56 

57 

58# This has to be updated on every schema change 

59_VERSION = VersionTuple(6, 0, 2) 

60 

61 

62class StaticDimensionRecordStorageManager(DimensionRecordStorageManager): 

63 """An implementation of `DimensionRecordStorageManager` for single-layer 

64 `Registry` and the base layers of multi-layer `Registry`. 

65 

66 This manager creates `DimensionRecordStorage` instances for all elements 

67 in the `DimensionUniverse` in its own `initialize` method, as part of 

68 static table creation, so it never needs to manage any dynamic registry 

69 tables. 

70 

71 Parameters 

72 ---------- 

73 db : `Database` 

74 Interface to the underlying database engine and namespace. 

75 records : `NamedKeyDict` 

76 Mapping from `DimensionElement` to `DimensionRecordStorage` for that 

77 element. 

78 overlaps : `list` [ `DatabaseDimensionOverlapStorage` ] 

79 Objects that manage materialized overlaps between database-backed 

80 dimensions. 

81 dimensionGraphStorage : `_DimensionGraphStorage` 

82 Object that manages saved `DimensionGraph` definitions. 

83 universe : `DimensionUniverse` 

84 All known dimensions. 

85 """ 

86 

87 def __init__( 

88 self, 

89 db: Database, 

90 *, 

91 records: NamedKeyDict[DimensionElement, DimensionRecordStorage], 

92 overlaps: dict[ 

93 tuple[DatabaseDimensionElement, DatabaseDimensionElement], DatabaseDimensionOverlapStorage 

94 ], 

95 dimensionGraphStorage: _DimensionGraphStorage, 

96 universe: DimensionUniverse, 

97 registry_schema_version: VersionTuple | None = None, 

98 ): 

99 super().__init__(universe=universe, registry_schema_version=registry_schema_version) 

100 self._db = db 

101 self._records = records 

102 self._overlaps = overlaps 

103 self._dimensionGraphStorage = dimensionGraphStorage 

104 

105 @classmethod 

106 def initialize( 

107 cls, 

108 db: Database, 

109 context: StaticTablesContext, 

110 *, 

111 universe: DimensionUniverse, 

112 registry_schema_version: VersionTuple | None = None, 

113 ) -> DimensionRecordStorageManager: 

114 # Docstring inherited from DimensionRecordStorageManager. 

115 # Start by initializing governor dimensions; those go both in the main 

116 # 'records' mapping we'll pass to init, and a local dictionary that we 

117 # can pass in when initializing storage for DatabaseDimensionElements. 

118 governors = NamedKeyDict[GovernorDimension, GovernorDimensionRecordStorage]() 

119 records = NamedKeyDict[DimensionElement, DimensionRecordStorage]() 

120 for dimension in universe.getGovernorDimensions(): 

121 governorStorage = dimension.makeStorage(db, context=context) 

122 governors[dimension] = governorStorage 

123 records[dimension] = governorStorage 

124 # Next we initialize storage for DatabaseDimensionElements. Some 

125 # elements' storage may be views into anothers; we'll do a first pass 

126 # to gather a mapping from the names of those targets back to their 

127 # views. 

128 view_targets = { 

129 element.viewOf: element 

130 for element in universe.getDatabaseElements() 

131 if element.viewOf is not None 

132 } 

133 # We remember the spatial ones (grouped by family) so we can go back 

134 # and initialize overlap storage for them later. 

135 spatial = NamedKeyDict[DatabaseTopologicalFamily, list[DatabaseDimensionRecordStorage]]() 

136 for element in universe.getDatabaseElements(): 

137 if element.viewOf is not None: 

138 # We'll initialize this storage when the view's target is 

139 # initialized. 

140 continue 

141 elementStorage = element.makeStorage(db, context=context, governors=governors) 

142 records[element] = elementStorage 

143 if element.spatial is not None: 

144 spatial.setdefault(element.spatial, []).append(elementStorage) 

145 if (view_element := view_targets.get(element.name)) is not None: 

146 view_element_storage = view_element.makeStorage( 

147 db, 

148 context=context, 

149 governors=governors, 

150 view_target=elementStorage, 

151 ) 

152 records[view_element] = view_element_storage 

153 if view_element.spatial is not None: 153 ↛ 154line 153 didn't jump to line 154, because the condition on line 153 was never true

154 spatial.setdefault(view_element.spatial, []).append(view_element_storage) 

155 

156 # Finally we initialize overlap storage. The implementation class for 

157 # this is currently hard-coded (it's not obvious there will ever be 

158 # others). Note that overlaps between database-backed dimensions and 

159 # skypix dimensions is internal to `DatabaseDimensionRecordStorage`, 

160 # and hence is not included here. 

161 from ..dimensions.overlaps import CrossFamilyDimensionOverlapStorage 

162 

163 overlaps: dict[ 

164 tuple[DatabaseDimensionElement, DatabaseDimensionElement], DatabaseDimensionOverlapStorage 

165 ] = {} 

166 for (family1, storages1), (family2, storages2) in itertools.combinations(spatial.items(), 2): 

167 for elementStoragePair in itertools.product(storages1, storages2): 

168 governorStoragePair = (governors[family1.governor], governors[family2.governor]) 

169 if elementStoragePair[0].element > elementStoragePair[1].element: 169 ↛ 170line 169 didn't jump to line 170, because the condition on line 169 was never true

170 elementStoragePair = (elementStoragePair[1], elementStoragePair[0]) 

171 governorStoragePair = (governorStoragePair[1], governorStoragePair[1]) 

172 overlapStorage = CrossFamilyDimensionOverlapStorage.initialize( 

173 db, 

174 elementStoragePair, 

175 governorStoragePair, 

176 context=context, 

177 ) 

178 elementStoragePair[0].connect(overlapStorage) 

179 elementStoragePair[1].connect(overlapStorage) 

180 overlaps[overlapStorage.elements] = overlapStorage 

181 # Create table that stores DimensionGraph definitions. 

182 dimensionGraphStorage = _DimensionGraphStorage.initialize(db, context, universe=universe) 

183 return cls( 

184 db=db, 

185 records=records, 

186 universe=universe, 

187 overlaps=overlaps, 

188 dimensionGraphStorage=dimensionGraphStorage, 

189 registry_schema_version=registry_schema_version, 

190 ) 

191 

192 def get(self, element: DimensionElement | str) -> DimensionRecordStorage | None: 

193 # Docstring inherited from DimensionRecordStorageManager. 

194 r = self._records.get(element) 

195 if r is None: 

196 if isinstance(element, str): 

197 element = self.universe[element] 

198 if isinstance(element, SkyPixDimension): 198 ↛ 200line 198 didn't jump to line 200, because the condition on line 198 was never false

199 return self.universe.skypix[element.system][element.level].makeStorage() 

200 return r 

201 

202 def register(self, element: DimensionElement) -> DimensionRecordStorage: 

203 # Docstring inherited from DimensionRecordStorageManager. 

204 result = self.get(element) 

205 assert result, "All records instances should be created in initialize()." 

206 return result 

207 

208 def saveDimensionGraph(self, graph: DimensionGraph) -> int: 

209 # Docstring inherited from DimensionRecordStorageManager. 

210 return self._dimensionGraphStorage.save(graph) 

211 

212 def loadDimensionGraph(self, key: int) -> DimensionGraph: 

213 # Docstring inherited from DimensionRecordStorageManager. 

214 return self._dimensionGraphStorage.load(key) 

215 

216 def clearCaches(self) -> None: 

217 # Docstring inherited from DimensionRecordStorageManager. 

218 for storage in self._records.values(): 

219 storage.clearCaches() 

220 

221 def make_spatial_join_relation( 

222 self, 

223 element1: str, 

224 element2: str, 

225 context: queries.SqlQueryContext, 

226 governor_constraints: Mapping[str, Set[str]], 

227 ) -> tuple[Relation, bool]: 

228 # Docstring inherited. 

229 storage1 = self[element1] 

230 storage2 = self[element2] 

231 overlaps: Relation | None = None 

232 needs_refinement: bool = False 

233 match (storage1, storage2): 

234 case [ 

235 DatabaseDimensionRecordStorage() as db_storage1, 

236 DatabaseDimensionRecordStorage() as db_storage2, 

237 ]: 

238 # Construction guarantees that we only need to try this in one 

239 # direction; either both storage objects know about the other 

240 # or neither do. 

241 overlaps = db_storage1.make_spatial_join_relation( 

242 db_storage2.element, context, governor_constraints 

243 ) 

244 if overlaps is None: 244 ↛ 262line 244 didn't jump to line 262, because the condition on line 244 was never false

245 # No direct materialized overlaps; use commonSkyPix as an 

246 # intermediary. 

247 common_skypix_overlap1 = db_storage1.make_spatial_join_relation( 

248 self.universe.commonSkyPix, context, governor_constraints 

249 ) 

250 common_skypix_overlap2 = db_storage2.make_spatial_join_relation( 

251 self.universe.commonSkyPix, context, governor_constraints 

252 ) 

253 assert ( 

254 common_skypix_overlap1 is not None and common_skypix_overlap2 is not None 

255 ), "Overlaps with the common skypix dimension should always be available," 

256 overlaps = common_skypix_overlap1.join(common_skypix_overlap2) 

257 needs_refinement = True 

258 case [DatabaseDimensionRecordStorage() as db_storage, other]: 258 ↛ 259line 258 didn't jump to line 259, because the pattern on line 258 never matched

259 overlaps = db_storage.make_spatial_join_relation(other.element, context, governor_constraints) 

260 case [other, DatabaseDimensionRecordStorage() as db_storage]: 260 ↛ 262line 260 didn't jump to line 262, because the pattern on line 260 never matched

261 overlaps = db_storage.make_spatial_join_relation(other.element, context, governor_constraints) 

262 if overlaps is None: 

263 # In the future, there's a lot more we could try here: 

264 # 

265 # - for skypix dimensions, looking for materialized overlaps at 

266 # smaller spatial scales (higher-levels) and using bit-shifting; 

267 # 

268 # - for non-skypix dimensions, looking for materialized overlaps 

269 # for more finer-grained members of the same family, and then 

270 # doing SELECT DISTINCT (or even tolerating duplicates) on the 

271 # columns we care about (e.g. use patch overlaps to satisfy a 

272 # request for tract overlaps). 

273 # 

274 # It's not obvious that's better than just telling the user to 

275 # materialize more overlaps, though. 

276 raise MissingSpatialOverlapError( 

277 f"No materialized overlaps for spatial join between {element1!r} and {element2!r}." 

278 ) 

279 return overlaps, needs_refinement 

280 

281 @classmethod 

282 def currentVersions(cls) -> list[VersionTuple]: 

283 # Docstring inherited from VersionedExtension. 

284 return [_VERSION] 

285 

286 

287class _DimensionGraphStorage: 

288 """Helper object that manages saved DimensionGraph definitions. 

289 

290 Should generally be constructed by calling `initialize` instead of invoking 

291 the constructor directly. 

292 

293 Parameters 

294 ---------- 

295 db : `Database` 

296 Interface to the underlying database engine and namespace. 

297 idTable : `sqlalchemy.schema.Table` 

298 Table that just holds unique IDs for dimension graphs. 

299 definitionTable : `sqlalchemy.schema.Table` 

300 Table that maps dimension names to the IDs of the dimension graphs to 

301 which they belong. 

302 universe : `DimensionUniverse` 

303 All known dimensions. 

304 """ 

305 

306 def __init__( 

307 self, 

308 db: Database, 

309 idTable: sqlalchemy.schema.Table, 

310 definitionTable: sqlalchemy.schema.Table, 

311 universe: DimensionUniverse, 

312 ): 

313 self._db = db 

314 self._idTable = idTable 

315 self._definitionTable = definitionTable 

316 self._universe = universe 

317 self._keysByGraph: dict[DimensionGraph, int] = {universe.empty: 0} 

318 self._graphsByKey: dict[int, DimensionGraph] = {0: universe.empty} 

319 

320 @classmethod 

321 def initialize( 

322 cls, 

323 db: Database, 

324 context: StaticTablesContext, 

325 *, 

326 universe: DimensionUniverse, 

327 ) -> _DimensionGraphStorage: 

328 """Construct a new instance, including creating tables if necessary. 

329 

330 Parameters 

331 ---------- 

332 db : `Database` 

333 Interface to the underlying database engine and namespace. 

334 context : `StaticTablesContext` 

335 Context object obtained from `Database.declareStaticTables`; used 

336 to declare any tables that should always be present. 

337 universe : `DimensionUniverse` 

338 All known dimensions. 

339 

340 Returns 

341 ------- 

342 storage : `_DimensionGraphStorage` 

343 New instance of this class. 

344 """ 

345 # We need two tables just so we have one where the autoincrement key is 

346 # the only primary key column, as is required by (at least) SQLite. In 

347 # other databases, we might be able to use a Sequence directly. 

348 idTable = context.addTable( 

349 "dimension_graph_key", 

350 ddl.TableSpec( 

351 fields=[ 

352 ddl.FieldSpec( 

353 name="id", 

354 dtype=sqlalchemy.BigInteger, 

355 autoincrement=True, 

356 primaryKey=True, 

357 ), 

358 ], 

359 ), 

360 ) 

361 definitionTable = context.addTable( 

362 "dimension_graph_definition", 

363 ddl.TableSpec( 

364 fields=[ 

365 ddl.FieldSpec(name="dimension_graph_id", dtype=sqlalchemy.BigInteger, primaryKey=True), 

366 ddl.FieldSpec(name="dimension_name", dtype=sqlalchemy.Text, primaryKey=True), 

367 ], 

368 foreignKeys=[ 

369 ddl.ForeignKeySpec( 

370 "dimension_graph_key", 

371 source=("dimension_graph_id",), 

372 target=("id",), 

373 onDelete="CASCADE", 

374 ), 

375 ], 

376 ), 

377 ) 

378 return cls(db, idTable, definitionTable, universe=universe) 

379 

380 def refresh(self) -> None: 

381 """Refresh the in-memory cache of saved DimensionGraph definitions. 

382 

383 This should be done automatically whenever needed, but it can also 

384 be called explicitly. 

385 """ 

386 dimensionNamesByKey: dict[int, set[str]] = defaultdict(set) 

387 with self._db.query(self._definitionTable.select()) as sql_result: 

388 sql_rows = sql_result.mappings().fetchall() 

389 for row in sql_rows: 

390 key = row[self._definitionTable.columns.dimension_graph_id] 

391 dimensionNamesByKey[key].add(row[self._definitionTable.columns.dimension_name]) 

392 keysByGraph: dict[DimensionGraph, int] = {self._universe.empty: 0} 

393 graphsByKey: dict[int, DimensionGraph] = {0: self._universe.empty} 

394 for key, dimensionNames in dimensionNamesByKey.items(): 

395 graph = DimensionGraph(self._universe, names=dimensionNames) 

396 keysByGraph[graph] = key 

397 graphsByKey[key] = graph 

398 self._graphsByKey = graphsByKey 

399 self._keysByGraph = keysByGraph 

400 

401 def save(self, graph: DimensionGraph) -> int: 

402 """Save a `DimensionGraph` definition to the database, allowing it to 

403 be retrieved later via the returned key. 

404 

405 Parameters 

406 ---------- 

407 graph : `DimensionGraph` 

408 Set of dimensions to save. 

409 

410 Returns 

411 ------- 

412 key : `int` 

413 Integer used as the unique key for this `DimensionGraph` in the 

414 database. 

415 """ 

416 key = self._keysByGraph.get(graph) 

417 if key is not None: 

418 return key 

419 # Lock tables and then refresh to guard against races where some other 

420 # process is trying to register the exact same dimension graph. This 

421 # is probably not the most efficient way to do it, but it should be a 

422 # rare operation, especially since the short-circuit above will usually 

423 # work in long-lived data repositories. 

424 with self._db.transaction(lock=[self._idTable, self._definitionTable]): 

425 self.refresh() 

426 key = self._keysByGraph.get(graph) 

427 if key is None: 427 ↛ 433line 427 didn't jump to line 433, because the condition on line 427 was never false

428 (key,) = self._db.insert(self._idTable, {}, returnIds=True) # type: ignore 

429 self._db.insert( 

430 self._definitionTable, 

431 *[{"dimension_graph_id": key, "dimension_name": name} for name in graph.required.names], 

432 ) 

433 self._keysByGraph[graph] = key 

434 self._graphsByKey[key] = graph 

435 return key 

436 

437 def load(self, key: int) -> DimensionGraph: 

438 """Retrieve a `DimensionGraph` that was previously saved in the 

439 database. 

440 

441 Parameters 

442 ---------- 

443 key : `int` 

444 Integer used as the unique key for this `DimensionGraph` in the 

445 database. 

446 

447 Returns 

448 ------- 

449 graph : `DimensionGraph` 

450 Retrieved graph. 

451 """ 

452 graph = self._graphsByKey.get(key) 

453 if graph is None: 

454 self.refresh() 

455 graph = self._graphsByKey[key] 

456 return graph