Coverage for python/lsst/daf/butler/core/ddl.py: 55%

230 statements  

« prev     ^ index     » next       coverage.py v7.3.2, created at 2023-10-25 15:14 +0000

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/>. 

21"""Classes for representing SQL data-definition language (DDL) in Python. 

22 

23This include "CREATE TABLE" etc. 

24 

25This provides an extra layer on top of SQLAlchemy's classes for these concepts, 

26because we need a level of indirection between logical tables and the actual 

27SQL, and SQLAlchemy's DDL classes always map 1-1 to SQL. 

28 

29We've opted for the rather more obscure "ddl" as the name of this module 

30instead of "schema" because the latter is too overloaded; in most SQL 

31databases, a "schema" is also another term for a namespace. 

32""" 

33from __future__ import annotations 

34 

35from lsst import sphgeom 

36 

37__all__ = ( 

38 "TableSpec", 

39 "FieldSpec", 

40 "ForeignKeySpec", 

41 "IndexSpec", 

42 "Base64Bytes", 

43 "Base64Region", 

44 "AstropyTimeNsecTai", 

45 "GUID", 

46) 

47 

48import logging 

49import uuid 

50from base64 import b64decode, b64encode 

51from collections.abc import Callable, Iterable 

52from dataclasses import dataclass 

53from math import ceil 

54from typing import TYPE_CHECKING, Any 

55 

56import astropy.time 

57import sqlalchemy 

58from lsst.sphgeom import Region 

59from lsst.utils.iteration import ensure_iterable 

60from sqlalchemy.dialects.postgresql import UUID 

61 

62from . import time_utils 

63from .config import Config 

64from .exceptions import ValidationError 

65from .named import NamedValueSet 

66from .utils import stripIfNotNone 

67 

68if TYPE_CHECKING: 

69 from .timespan import TimespanDatabaseRepresentation 

70 

71 

72_LOG = logging.getLogger(__name__) 

73 

74 

75class SchemaValidationError(ValidationError): 

76 """Exceptions that indicate problems in Registry schema configuration.""" 

77 

78 @classmethod 

79 def translate(cls, caught: type[Exception], message: str) -> Callable: 

80 """Return decorator to re-raise exceptions as `SchemaValidationError`. 

81 

82 Decorated functions must be class or instance methods, with a 

83 ``config`` parameter as their first argument. This will be passed 

84 to ``message.format()`` as a keyword argument, along with ``err``, 

85 the original exception. 

86 

87 Parameters 

88 ---------- 

89 caught : `type` (`Exception` subclass) 

90 The type of exception to catch. 

91 message : `str` 

92 A `str.format` string that may contain named placeholders for 

93 ``config``, ``err``, or any keyword-only argument accepted by 

94 the decorated function. 

95 """ 

96 

97 def decorate(func: Callable) -> Callable: 

98 def decorated(self: Any, config: Config, *args: Any, **kwargs: Any) -> Any: 

99 try: 

100 return func(self, config, *args, **kwargs) 

101 except caught as err: 

102 raise cls(message.format(config=str(config), err=err)) from err 

103 

104 return decorated 

105 

106 return decorate 

107 

108 

109class Base64Bytes(sqlalchemy.TypeDecorator): 

110 """A SQLAlchemy custom type for Python `bytes`. 

111 

112 Maps Python `bytes` to a base64-encoded `sqlalchemy.Text` field. 

113 """ 

114 

115 impl = sqlalchemy.Text 

116 

117 cache_ok = True 

118 

119 def __init__(self, nbytes: int | None = None, *args: Any, **kwargs: Any): 

120 if nbytes is not None: 

121 length = 4 * ceil(nbytes / 3) if self.impl is sqlalchemy.String else None 

122 else: 

123 length = None 

124 super().__init__(*args, length=length, **kwargs) 

125 self.nbytes = nbytes 

126 

127 def process_bind_param(self, value: bytes | None, dialect: sqlalchemy.engine.Dialect) -> str | None: 

128 # 'value' is native `bytes`. We want to encode that to base64 `bytes` 

129 # and then ASCII `str`, because `str` is what SQLAlchemy expects for 

130 # String fields. 

131 if value is None: 

132 return None 

133 if not isinstance(value, bytes): 

134 raise TypeError( 

135 f"Base64Bytes fields require 'bytes' values; got '{value}' with type {type(value)}." 

136 ) 

137 return b64encode(value).decode("ascii") 

138 

139 def process_result_value(self, value: str | None, dialect: sqlalchemy.engine.Dialect) -> bytes | None: 

140 # 'value' is a `str` that must be ASCII because it's base64-encoded. 

141 # We want to transform that to base64-encoded `bytes` and then 

142 # native `bytes`. 

143 return b64decode(value.encode("ascii")) if value is not None else None 

144 

145 @property 

146 def python_type(self) -> type[bytes]: 

147 return bytes 

148 

149 

150# create an alias, for use below to disambiguate between the built in 

151# sqlachemy type 

152LocalBase64Bytes = Base64Bytes 

153 

154 

155class Base64Region(Base64Bytes): 

156 """A SQLAlchemy custom type for Python `sphgeom.Region`. 

157 

158 Maps Python `sphgeom.Region` to a base64-encoded `sqlalchemy.String`. 

159 """ 

160 

161 cache_ok = True # have to be set explicitly in each class 

162 

163 def process_bind_param(self, value: Region | None, dialect: sqlalchemy.engine.Dialect) -> str | None: 

164 if value is None: 

165 return None 

166 return super().process_bind_param(value.encode(), dialect) 

167 

168 def process_result_value(self, value: str | None, dialect: sqlalchemy.engine.Dialect) -> Region | None: 

169 if value is None: 

170 return None 

171 return Region.decode(super().process_result_value(value, dialect)) 

172 

173 @property 

174 def python_type(self) -> type[sphgeom.Region]: 

175 return sphgeom.Region 

176 

177 

178class AstropyTimeNsecTai(sqlalchemy.TypeDecorator): 

179 """A SQLAlchemy custom type for Python `astropy.time.Time`. 

180 

181 Maps Python `astropy.time.Time` to a number of nanoseconds since Unix 

182 epoch in TAI scale. 

183 """ 

184 

185 impl = sqlalchemy.BigInteger 

186 

187 cache_ok = True 

188 

189 def process_bind_param( 

190 self, value: astropy.time.Time | None, dialect: sqlalchemy.engine.Dialect 

191 ) -> int | None: 

192 if value is None: 

193 return None 

194 if not isinstance(value, astropy.time.Time): 

195 raise TypeError(f"Unsupported type: {type(value)}, expected astropy.time.Time") 

196 value = time_utils.TimeConverter().astropy_to_nsec(value) 

197 return value 

198 

199 def process_result_value( 

200 self, value: int | None, dialect: sqlalchemy.engine.Dialect 

201 ) -> astropy.time.Time | None: 

202 # value is nanoseconds since epoch, or None 

203 if value is None: 

204 return None 

205 value = time_utils.TimeConverter().nsec_to_astropy(value) 

206 return value 

207 

208 

209# TODO: sqlalchemy 2 has internal support for UUID: 

210# https://docs.sqlalchemy.org/en/20/core/type_basics.html#sqlalchemy.types.Uuid 

211class GUID(sqlalchemy.TypeDecorator): 

212 """Platform-independent GUID type. 

213 

214 Uses PostgreSQL's UUID type, otherwise uses CHAR(32), storing as 

215 stringified hex values. 

216 """ 

217 

218 impl = sqlalchemy.CHAR 

219 

220 cache_ok = True 

221 

222 def load_dialect_impl(self, dialect: sqlalchemy.Dialect) -> sqlalchemy.types.TypeEngine: 

223 if dialect.name == "postgresql": 

224 return dialect.type_descriptor(UUID()) 

225 else: 

226 return dialect.type_descriptor(sqlalchemy.CHAR(32)) 

227 

228 def process_bind_param(self, value: Any, dialect: sqlalchemy.Dialect) -> str | None: 

229 if value is None: 

230 return value 

231 

232 # Coerce input to UUID type, in general having UUID on input is the 

233 # only thing that we want but there is code right now that uses ints. 

234 if isinstance(value, int): 

235 value = uuid.UUID(int=value) 

236 elif isinstance(value, bytes): 

237 value = uuid.UUID(bytes=value) 

238 elif isinstance(value, str): 

239 # hexstring 

240 value = uuid.UUID(hex=value) 

241 elif not isinstance(value, uuid.UUID): 

242 raise TypeError(f"Unexpected type of a bind value: {type(value)}") 

243 

244 if dialect.name == "postgresql": 

245 return str(value) 

246 else: 

247 return "%.32x" % value.int 

248 

249 def process_result_value( 

250 self, value: str | uuid.UUID | None, dialect: sqlalchemy.Dialect 

251 ) -> uuid.UUID | None: 

252 if value is None or isinstance(value, uuid.UUID): 

253 # sqlalchemy 2 converts to UUID internally 

254 return value 

255 else: 

256 return uuid.UUID(hex=value) 

257 

258 

259VALID_CONFIG_COLUMN_TYPES = { 

260 "string": sqlalchemy.String, 

261 "int": sqlalchemy.BigInteger, 

262 "float": sqlalchemy.Float, 

263 "region": Base64Region, 

264 "bool": sqlalchemy.Boolean, 

265 "blob": sqlalchemy.LargeBinary, 

266 "datetime": AstropyTimeNsecTai, 

267 "hash": Base64Bytes, 

268 "uuid": GUID, 

269} 

270 

271 

272@dataclass 

273class FieldSpec: 

274 """A data class for defining a column in a logical `Registry` table.""" 

275 

276 name: str 

277 """Name of the column.""" 

278 

279 dtype: type 

280 """Type of the column; usually a `type` subclass provided by SQLAlchemy 

281 that defines both a Python type and a corresponding precise SQL type. 

282 """ 

283 

284 length: int | None = None 

285 """Length of the type in the database, for variable-length types.""" 

286 

287 nbytes: int | None = None 

288 """Natural length used for hash and encoded-region columns, to be converted 

289 into the post-encoding length. 

290 """ 

291 

292 primaryKey: bool = False 

293 """Whether this field is (part of) its table's primary key.""" 

294 

295 autoincrement: bool = False 

296 """Whether the database should insert automatically incremented values when 

297 no value is provided in an INSERT. 

298 """ 

299 

300 nullable: bool = True 

301 """Whether this field is allowed to be NULL. If ``primaryKey`` is 

302 `True`, during construction this value will be forced to `False`.""" 

303 

304 default: Any = None 

305 """A server-side default value for this field. 

306 

307 This is passed directly as the ``server_default`` argument to 

308 `sqlalchemy.schema.Column`. It does _not_ go through SQLAlchemy's usual 

309 type conversion or quoting for Python literals, and should hence be used 

310 with care. See the SQLAlchemy documentation for more information. 

311 """ 

312 

313 doc: str | None = None 

314 """Documentation for this field.""" 

315 

316 def __post_init__(self) -> None: 

317 if self.primaryKey: 

318 # Change the default to match primaryKey. 

319 self.nullable = False 

320 

321 def __eq__(self, other: Any) -> bool: 

322 if isinstance(other, FieldSpec): 

323 return self.name == other.name 

324 else: 

325 return NotImplemented 

326 

327 def __hash__(self) -> int: 

328 return hash(self.name) 

329 

330 @classmethod 

331 @SchemaValidationError.translate(KeyError, "Missing key {err} in column config '{config}'.") 

332 def fromConfig(cls, config: Config, **kwargs: Any) -> FieldSpec: 

333 """Create a `FieldSpec` from a subset of a `SchemaConfig`. 

334 

335 Parameters 

336 ---------- 

337 config: `Config` 

338 Configuration describing the column. Nested configuration keys 

339 correspond to `FieldSpec` attributes. 

340 **kwargs 

341 Additional keyword arguments that provide defaults for values 

342 not present in config. 

343 

344 Returns 

345 ------- 

346 spec: `FieldSpec` 

347 Specification structure for the column. 

348 

349 Raises 

350 ------ 

351 SchemaValidationError 

352 Raised if configuration keys are missing or have invalid values. 

353 """ 

354 dtype = VALID_CONFIG_COLUMN_TYPES.get(config["type"]) 

355 if dtype is None: 

356 raise SchemaValidationError(f"Invalid field type string: '{config['type']}'.") 

357 if not config["name"].islower(): 

358 raise SchemaValidationError(f"Column name '{config['name']}' is not all lowercase.") 

359 self = cls(name=config["name"], dtype=dtype, **kwargs) 

360 self.length = config.get("length", self.length) 

361 self.nbytes = config.get("nbytes", self.nbytes) 

362 if self.length is not None and self.nbytes is not None: 

363 raise SchemaValidationError(f"Both length and nbytes provided for field '{self.name}'.") 

364 self.primaryKey = config.get("primaryKey", self.primaryKey) 

365 self.autoincrement = config.get("autoincrement", self.autoincrement) 

366 self.nullable = config.get("nullable", False if self.primaryKey else self.nullable) 

367 self.doc = stripIfNotNone(config.get("doc", None)) 

368 return self 

369 

370 @classmethod 

371 def for_region(cls, name: str = "region", nullable: bool = True, nbytes: int = 2048) -> FieldSpec: 

372 """Create a `FieldSpec` for a spatial region column. 

373 

374 Parameters 

375 ---------- 

376 name : `str`, optional 

377 Name for the field. 

378 nullable : `bool`, optional 

379 Whether NULL values are permitted. 

380 nbytes : `int`, optional 

381 Maximum number of bytes for serialized regions. The actual column 

382 size will be larger to allow for base-64 encoding. 

383 

384 Returns 

385 ------- 

386 spec : `FieldSpec` 

387 Specification structure for a region column. 

388 """ 

389 return cls(name, nullable=nullable, dtype=Base64Region, nbytes=nbytes) 

390 

391 def isStringType(self) -> bool: 

392 """Indicate that this is a sqlalchemy.String field spec. 

393 

394 Returns 

395 ------- 

396 isString : `bool` 

397 The field refers to a `sqlalchemy.String` and not any other type. 

398 This can return `False` even if the object was created with a 

399 string type if it has been decided that it should be implemented 

400 as a `sqlalchemy.Text` type. 

401 """ 

402 # For short strings retain them as strings 

403 if self.dtype is sqlalchemy.String and self.length and self.length <= 32: 

404 return True 

405 return False 

406 

407 def getSizedColumnType(self) -> sqlalchemy.types.TypeEngine | type: 

408 """Return a sized version of the column type. 

409 

410 Utilizes either (or neither) of ``self.length`` and ``self.nbytes``. 

411 

412 Returns 

413 ------- 

414 dtype : `sqlalchemy.types.TypeEngine` 

415 A SQLAlchemy column type object. 

416 """ 

417 if self.length is not None: 

418 # Last chance check that we are only looking at possible String 

419 if self.dtype is sqlalchemy.String and not self.isStringType(): 

420 return sqlalchemy.Text 

421 return self.dtype(length=self.length) 

422 if self.nbytes is not None: 

423 return self.dtype(nbytes=self.nbytes) 

424 return self.dtype 

425 

426 def getPythonType(self) -> type: 

427 """Return the Python type associated with this field's (SQL) dtype. 

428 

429 Returns 

430 ------- 

431 type : `type` 

432 Python type associated with this field's (SQL) `dtype`. 

433 """ 

434 # to construct these objects, nbytes keyword is needed 

435 if issubclass(self.dtype, LocalBase64Bytes): 

436 # satisfy mypy for something that must be true 

437 assert self.nbytes is not None 

438 return self.dtype(nbytes=self.nbytes).python_type 

439 else: 

440 return self.dtype().python_type # type: ignore 

441 

442 

443@dataclass 

444class ForeignKeySpec: 

445 """Definition of a foreign key constraint in a logical `Registry` table.""" 

446 

447 table: str 

448 """Name of the target table.""" 

449 

450 source: tuple[str, ...] 

451 """Tuple of source table column names.""" 

452 

453 target: tuple[str, ...] 

454 """Tuple of target table column names.""" 

455 

456 onDelete: str | None = None 

457 """SQL clause indicating how to handle deletes to the target table. 

458 

459 If not `None` (which indicates that a constraint violation exception should 

460 be raised), should be either "SET NULL" or "CASCADE". 

461 """ 

462 

463 addIndex: bool = True 

464 """If `True`, create an index on the columns of this foreign key in the 

465 source table. 

466 """ 

467 

468 @classmethod 

469 @SchemaValidationError.translate(KeyError, "Missing key {err} in foreignKey config '{config}'.") 

470 def fromConfig(cls, config: Config) -> ForeignKeySpec: 

471 """Create a `ForeignKeySpec` from a subset of a `SchemaConfig`. 

472 

473 Parameters 

474 ---------- 

475 config: `Config` 

476 Configuration describing the constraint. Nested configuration keys 

477 correspond to `ForeignKeySpec` attributes. 

478 

479 Returns 

480 ------- 

481 spec: `ForeignKeySpec` 

482 Specification structure for the constraint. 

483 

484 Raises 

485 ------ 

486 SchemaValidationError 

487 Raised if configuration keys are missing or have invalid values. 

488 """ 

489 return cls( 

490 table=config["table"], 

491 source=tuple(ensure_iterable(config["source"])), 

492 target=tuple(ensure_iterable(config["target"])), 

493 onDelete=config.get("onDelete", None), 

494 ) 

495 

496 

497@dataclass(frozen=True) 

498class IndexSpec: 

499 """Specification of an index on table columns. 

500 

501 Parameters 

502 ---------- 

503 *columns : `str` 

504 Names of the columns to index. 

505 **kwargs: `Any` 

506 Additional keyword arguments to pass directly to 

507 `sqlalchemy.schema.Index` constructor. This could be used to provide 

508 backend-specific options, e.g. to create a ``GIST`` index in PostgreSQL 

509 one can pass ``postgresql_using="gist"``. 

510 """ 

511 

512 def __init__(self, *columns: str, **kwargs: Any): 

513 object.__setattr__(self, "columns", tuple(columns)) 

514 object.__setattr__(self, "kwargs", kwargs) 

515 

516 def __hash__(self) -> int: 

517 return hash(self.columns) 

518 

519 columns: tuple[str, ...] 

520 """Column names to include in the index (`Tuple` [ `str` ]).""" 

521 

522 kwargs: dict[str, Any] 

523 """Additional keyword arguments passed directly to 

524 `sqlalchemy.schema.Index` constructor (`dict` [ `str`, `Any` ]). 

525 """ 

526 

527 

528@dataclass 

529class TableSpec: 

530 """A data class used to define a table or table-like query interface. 

531 

532 Parameters 

533 ---------- 

534 fields : `~collections.abc.Iterable` [ `FieldSpec` ] 

535 Specifications for the columns in this table. 

536 unique : `~collections.abc.Iterable` [ `tuple` [ `str` ] ], optional 

537 Non-primary-key unique constraints for the table. 

538 indexes: `~collections.abc.Iterable` [ `IndexSpec` ], optional 

539 Indexes for the table. 

540 foreignKeys : `~collections.abc.Iterable` [ `ForeignKeySpec` ], optional 

541 Foreign key constraints for the table. 

542 exclusion : `~collections.abc.Iterable` [ `tuple` [ `str` or `type` ] ] 

543 Special constraints that prohibit overlaps between timespans over rows 

544 where other columns are equal. These take the same form as unique 

545 constraints, but each tuple may contain a single 

546 `TimespanDatabaseRepresentation` subclass representing a timespan 

547 column. 

548 recycleIds : `bool`, optional 

549 If `True`, allow databases that might normally recycle autoincrement 

550 IDs to do so (usually better for performance) on any autoincrement 

551 field in this table. 

552 doc : `str`, optional 

553 Documentation for the table. 

554 """ 

555 

556 def __init__( 

557 self, 

558 fields: Iterable[FieldSpec], 

559 *, 

560 unique: Iterable[tuple[str, ...]] = (), 

561 indexes: Iterable[IndexSpec] = (), 

562 foreignKeys: Iterable[ForeignKeySpec] = (), 

563 exclusion: Iterable[tuple[str | type[TimespanDatabaseRepresentation], ...]] = (), 

564 recycleIds: bool = True, 

565 doc: str | None = None, 

566 ): 

567 self.fields = NamedValueSet(fields) 

568 self.unique = set(unique) 

569 self.indexes = set(indexes) 

570 self.foreignKeys = list(foreignKeys) 

571 self.exclusion = set(exclusion) 

572 self.recycleIds = recycleIds 

573 self.doc = doc 

574 

575 fields: NamedValueSet[FieldSpec] 

576 """Specifications for the columns in this table.""" 

577 

578 unique: set[tuple[str, ...]] 

579 """Non-primary-key unique constraints for the table.""" 

580 

581 indexes: set[IndexSpec] 

582 """Indexes for the table.""" 

583 

584 foreignKeys: list[ForeignKeySpec] 

585 """Foreign key constraints for the table.""" 

586 

587 exclusion: set[tuple[str | type[TimespanDatabaseRepresentation], ...]] 

588 """Exclusion constraints for the table. 

589 

590 Exclusion constraints behave mostly like unique constraints, but may 

591 contain a database-native Timespan column that is restricted to not overlap 

592 across rows (for identical combinations of any non-Timespan columns in the 

593 constraint). 

594 """ 

595 

596 recycleIds: bool = True 

597 """If `True`, allow databases that might normally recycle autoincrement IDs 

598 to do so (usually better for performance) on any autoincrement field in 

599 this table. 

600 """ 

601 

602 doc: str | None = None 

603 """Documentation for the table.""" 

604 

605 @classmethod 

606 @SchemaValidationError.translate(KeyError, "Missing key {err} in table config '{config}'.") 

607 def fromConfig(cls, config: Config) -> TableSpec: 

608 """Create a `ForeignKeySpec` from a subset of a `SchemaConfig`. 

609 

610 Parameters 

611 ---------- 

612 config: `Config` 

613 Configuration describing the constraint. Nested configuration keys 

614 correspond to `TableSpec` attributes. 

615 

616 Returns 

617 ------- 

618 spec: `TableSpec` 

619 Specification structure for the table. 

620 

621 Raises 

622 ------ 

623 SchemaValidationError 

624 Raised if configuration keys are missing or have invalid values. 

625 """ 

626 return cls( 

627 fields=NamedValueSet(FieldSpec.fromConfig(c) for c in config["columns"]), 

628 unique={tuple(u) for u in config.get("unique", ())}, 

629 foreignKeys=[ForeignKeySpec.fromConfig(c) for c in config.get("foreignKeys", ())], 

630 doc=stripIfNotNone(config.get("doc")), 

631 )