Coverage for python/lsst/daf/butler/registry/collections/_base.py: 90%
180 statements
« prev ^ index » next coverage.py v6.5.0, created at 2022-11-29 02:00 -0800
« prev ^ index » next coverage.py v6.5.0, created at 2022-11-29 02:00 -0800
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
23__all__ = ()
25import itertools
26from abc import abstractmethod
27from collections import namedtuple
28from collections.abc import Iterable, Iterator, Set
29from typing import TYPE_CHECKING, Any, Generic, TypeVar, cast
31import sqlalchemy
32from lsst.utils.ellipsis import Ellipsis
34from ...core import DimensionUniverse, Timespan, TimespanDatabaseRepresentation, ddl
35from .._collectionType import CollectionType
36from .._exceptions import MissingCollectionError
37from ..interfaces import ChainedCollectionRecord, CollectionManager, CollectionRecord, RunRecord
38from ..wildcards import CollectionWildcard
40if TYPE_CHECKING: 40 ↛ 41line 40 didn't jump to line 41, because the condition on line 40 was never true
41 from ..interfaces import Database, DimensionRecordStorageManager
44def _makeCollectionForeignKey(
45 sourceColumnName: str, collectionIdName: str, **kwargs: Any
46) -> ddl.ForeignKeySpec:
47 """Define foreign key specification that refers to collections table.
49 Parameters
50 ----------
51 sourceColumnName : `str`
52 Name of the column in the referring table.
53 collectionIdName : `str`
54 Name of the column in collections table that identifies it (PK).
55 **kwargs
56 Additional keyword arguments passed directly to `ddl.ForeignKeySpec`.
58 Returns
59 -------
60 spec : `ddl.ForeignKeySpec`
61 Foreign key specification.
63 Notes
64 -----
65 This method assumes fixed name ("collection") of a collections table.
66 There is also a general assumption that collection primary key consists
67 of a single column.
68 """
69 return ddl.ForeignKeySpec("collection", source=(sourceColumnName,), target=(collectionIdName,), **kwargs)
72CollectionTablesTuple = namedtuple("CollectionTablesTuple", ["collection", "run", "collection_chain"])
75def makeRunTableSpec(
76 collectionIdName: str, collectionIdType: type, TimespanReprClass: type[TimespanDatabaseRepresentation]
77) -> ddl.TableSpec:
78 """Define specification for "run" table.
80 Parameters
81 ----------
82 collectionIdName : `str`
83 Name of the column in collections table that identifies it (PK).
84 collectionIdType
85 Type of the PK column in the collections table, one of the
86 `sqlalchemy` types.
87 TimespanReprClass : `type` [ `TimespanDatabaseRepresentation` ]
88 Subclass of `TimespanDatabaseRepresentation` that encapsulates how
89 timespans are stored in this database.
92 Returns
93 -------
94 spec : `ddl.TableSpec`
95 Specification for run table.
97 Notes
98 -----
99 Assumption here and in the code below is that the name of the identifying
100 column is the same in both collections and run tables. The names of
101 non-identifying columns containing run metadata are fixed.
102 """
103 result = ddl.TableSpec(
104 fields=[
105 ddl.FieldSpec(collectionIdName, dtype=collectionIdType, primaryKey=True),
106 ddl.FieldSpec("host", dtype=sqlalchemy.String, length=128),
107 ],
108 foreignKeys=[
109 _makeCollectionForeignKey(collectionIdName, collectionIdName, onDelete="CASCADE"),
110 ],
111 )
112 for fieldSpec in TimespanReprClass.makeFieldSpecs(nullable=True):
113 result.fields.add(fieldSpec)
114 return result
117def makeCollectionChainTableSpec(collectionIdName: str, collectionIdType: type) -> ddl.TableSpec:
118 """Define specification for "collection_chain" table.
120 Parameters
121 ----------
122 collectionIdName : `str`
123 Name of the column in collections table that identifies it (PK).
124 collectionIdType
125 Type of the PK column in the collections table, one of the
126 `sqlalchemy` types.
128 Returns
129 -------
130 spec : `ddl.TableSpec`
131 Specification for collection chain table.
133 Notes
134 -----
135 Collection chain is simply an ordered one-to-many relation between
136 collections. The names of the columns in the table are fixed and
137 also hardcoded in the code below.
138 """
139 return ddl.TableSpec(
140 fields=[
141 ddl.FieldSpec("parent", dtype=collectionIdType, primaryKey=True),
142 ddl.FieldSpec("position", dtype=sqlalchemy.SmallInteger, primaryKey=True),
143 ddl.FieldSpec("child", dtype=collectionIdType, nullable=False),
144 ],
145 foreignKeys=[
146 _makeCollectionForeignKey("parent", collectionIdName, onDelete="CASCADE"),
147 _makeCollectionForeignKey("child", collectionIdName),
148 ],
149 )
152class DefaultRunRecord(RunRecord):
153 """Default `RunRecord` implementation.
155 This method assumes the same run table definition as produced by
156 `makeRunTableSpec` method. The only non-fixed name in the schema
157 is the PK column name, this needs to be passed in a constructor.
159 Parameters
160 ----------
161 db : `Database`
162 Registry database.
163 key
164 Unique collection ID, can be the same as ``name`` if ``name`` is used
165 for identification. Usually this is an integer or string, but can be
166 other database-specific type.
167 name : `str`
168 Run collection name.
169 table : `sqlalchemy.schema.Table`
170 Table for run records.
171 idColumnName : `str`
172 Name of the identifying column in run table.
173 host : `str`, optional
174 Name of the host where run was produced.
175 timespan : `Timespan`, optional
176 Timespan for this run.
177 """
179 def __init__(
180 self,
181 db: Database,
182 key: Any,
183 name: str,
184 *,
185 table: sqlalchemy.schema.Table,
186 idColumnName: str,
187 host: str | None = None,
188 timespan: Timespan | None = None,
189 ):
190 super().__init__(key=key, name=name, type=CollectionType.RUN)
191 self._db = db
192 self._table = table
193 self._host = host
194 if timespan is None: 194 ↛ 196line 194 didn't jump to line 196, because the condition on line 194 was never false
195 timespan = Timespan(begin=None, end=None)
196 self._timespan = timespan
197 self._idName = idColumnName
199 def update(self, host: str | None = None, timespan: Timespan | None = None) -> None:
200 # Docstring inherited from RunRecord.
201 if timespan is None:
202 timespan = Timespan(begin=None, end=None)
203 row = {
204 self._idName: self.key,
205 "host": host,
206 }
207 self._db.getTimespanRepresentation().update(timespan, result=row)
208 count = self._db.update(self._table, {self._idName: self.key}, row)
209 if count != 1:
210 raise RuntimeError(f"Run update affected {count} records; expected exactly one.")
211 self._host = host
212 self._timespan = timespan
214 @property
215 def host(self) -> str | None:
216 # Docstring inherited from RunRecord.
217 return self._host
219 @property
220 def timespan(self) -> Timespan:
221 # Docstring inherited from RunRecord.
222 return self._timespan
225class DefaultChainedCollectionRecord(ChainedCollectionRecord):
226 """Default `ChainedCollectionRecord` implementation.
228 This method assumes the same chain table definition as produced by
229 `makeCollectionChainTableSpec` method. All column names in the table are
230 fixed and hard-coded in the methods.
232 Parameters
233 ----------
234 db : `Database`
235 Registry database.
236 key
237 Unique collection ID, can be the same as ``name`` if ``name`` is used
238 for identification. Usually this is an integer or string, but can be
239 other database-specific type.
240 name : `str`
241 Collection name.
242 table : `sqlalchemy.schema.Table`
243 Table for chain relationship records.
244 universe : `DimensionUniverse`
245 Object managing all known dimensions.
246 """
248 def __init__(
249 self,
250 db: Database,
251 key: Any,
252 name: str,
253 *,
254 table: sqlalchemy.schema.Table,
255 universe: DimensionUniverse,
256 ):
257 super().__init__(key=key, name=name, universe=universe)
258 self._db = db
259 self._table = table
260 self._universe = universe
262 def _update(self, manager: CollectionManager, children: tuple[str, ...]) -> None:
263 # Docstring inherited from ChainedCollectionRecord.
264 rows = []
265 position = itertools.count()
266 for child in manager.resolve_wildcard(CollectionWildcard.from_names(children), flatten_chains=False):
267 rows.append(
268 {
269 "parent": self.key,
270 "child": child.key,
271 "position": next(position),
272 }
273 )
274 with self._db.transaction():
275 self._db.delete(self._table, ["parent"], {"parent": self.key})
276 self._db.insert(self._table, *rows)
278 def _load(self, manager: CollectionManager) -> tuple[str, ...]:
279 # Docstring inherited from ChainedCollectionRecord.
280 sql = (
281 sqlalchemy.sql.select(
282 self._table.columns.child,
283 )
284 .select_from(self._table)
285 .where(self._table.columns.parent == self.key)
286 .order_by(self._table.columns.position)
287 )
288 return tuple(manager[row._mapping[self._table.columns.child]].name for row in self._db.query(sql))
291K = TypeVar("K")
294class DefaultCollectionManager(Generic[K], CollectionManager):
295 """Default `CollectionManager` implementation.
297 This implementation uses record classes defined in this module and is
298 based on the same assumptions about schema outlined in the record classes.
300 Parameters
301 ----------
302 db : `Database`
303 Interface to the underlying database engine and namespace.
304 tables : `CollectionTablesTuple`
305 Named tuple of SQLAlchemy table objects.
306 collectionIdName : `str`
307 Name of the column in collections table that identifies it (PK).
308 dimensions : `DimensionRecordStorageManager`
309 Manager object for the dimensions in this `Registry`.
311 Notes
312 -----
313 Implementation uses "aggressive" pre-fetching and caching of the records
314 in memory. Memory cache is synchronized from database when `refresh`
315 method is called.
316 """
318 def __init__(
319 self,
320 db: Database,
321 tables: CollectionTablesTuple,
322 collectionIdName: str,
323 *,
324 dimensions: DimensionRecordStorageManager,
325 ):
326 super().__init__()
327 self._db = db
328 self._tables = tables
329 self._collectionIdName = collectionIdName
330 self._records: dict[K, CollectionRecord] = {} # indexed by record ID
331 self._dimensions = dimensions
333 def refresh(self) -> None:
334 # Docstring inherited from CollectionManager.
335 sql = sqlalchemy.sql.select(
336 *(list(self._tables.collection.columns) + list(self._tables.run.columns))
337 ).select_from(self._tables.collection.join(self._tables.run, isouter=True))
338 # Put found records into a temporary instead of updating self._records
339 # in place, for exception safety.
340 records = []
341 chains = []
342 TimespanReprClass = self._db.getTimespanRepresentation()
343 for row in self._db.query(sql).mappings():
344 collection_id = row[self._tables.collection.columns[self._collectionIdName]]
345 name = row[self._tables.collection.columns.name]
346 type = CollectionType(row["type"])
347 record: CollectionRecord
348 if type is CollectionType.RUN:
349 record = DefaultRunRecord(
350 key=collection_id,
351 name=name,
352 db=self._db,
353 table=self._tables.run,
354 idColumnName=self._collectionIdName,
355 host=row[self._tables.run.columns.host],
356 timespan=TimespanReprClass.extract(row),
357 )
358 elif type is CollectionType.CHAINED:
359 record = DefaultChainedCollectionRecord(
360 db=self._db,
361 key=collection_id,
362 table=self._tables.collection_chain,
363 name=name,
364 universe=self._dimensions.universe,
365 )
366 chains.append(record)
367 else:
368 record = CollectionRecord(key=collection_id, name=name, type=type)
369 records.append(record)
370 self._setRecordCache(records)
371 for chain in chains:
372 try:
373 chain.refresh(self)
374 except MissingCollectionError:
375 # This indicates a race condition in which some other client
376 # created a new collection and added it as a child of this
377 # (pre-existing) chain between the time we fetched all
378 # collections and the time we queried for parent-child
379 # relationships.
380 # Because that's some other unrelated client, we shouldn't care
381 # about that parent collection anyway, so we just drop it on
382 # the floor (a manual refresh can be used to get it back).
383 self._removeCachedRecord(chain)
385 def register(
386 self, name: str, type: CollectionType, doc: str | None = None
387 ) -> tuple[CollectionRecord, bool]:
388 # Docstring inherited from CollectionManager.
389 registered = False
390 record = self._getByName(name)
391 if record is None:
392 row, inserted_or_updated = self._db.sync(
393 self._tables.collection,
394 keys={"name": name},
395 compared={"type": int(type)},
396 extra={"doc": doc},
397 returning=[self._collectionIdName],
398 )
399 assert isinstance(inserted_or_updated, bool)
400 registered = inserted_or_updated
401 assert row is not None
402 collection_id = row[self._collectionIdName]
403 if type is CollectionType.RUN:
404 TimespanReprClass = self._db.getTimespanRepresentation()
405 row, _ = self._db.sync(
406 self._tables.run,
407 keys={self._collectionIdName: collection_id},
408 returning=("host",) + TimespanReprClass.getFieldNames(),
409 )
410 assert row is not None
411 record = DefaultRunRecord(
412 db=self._db,
413 key=collection_id,
414 name=name,
415 table=self._tables.run,
416 idColumnName=self._collectionIdName,
417 host=row["host"],
418 timespan=TimespanReprClass.extract(row),
419 )
420 elif type is CollectionType.CHAINED:
421 record = DefaultChainedCollectionRecord(
422 db=self._db,
423 key=collection_id,
424 name=name,
425 table=self._tables.collection_chain,
426 universe=self._dimensions.universe,
427 )
428 else:
429 record = CollectionRecord(key=collection_id, name=name, type=type)
430 self._addCachedRecord(record)
431 return record, registered
433 def remove(self, name: str) -> None:
434 # Docstring inherited from CollectionManager.
435 record = self._getByName(name)
436 if record is None: 436 ↛ 437line 436 didn't jump to line 437, because the condition on line 436 was never true
437 raise MissingCollectionError(f"No collection with name '{name}' found.")
438 # This may raise
439 self._db.delete(
440 self._tables.collection, [self._collectionIdName], {self._collectionIdName: record.key}
441 )
442 self._removeCachedRecord(record)
444 def find(self, name: str) -> CollectionRecord:
445 # Docstring inherited from CollectionManager.
446 result = self._getByName(name)
447 if result is None:
448 raise MissingCollectionError(f"No collection with name '{name}' found.")
449 return result
451 def __getitem__(self, key: Any) -> CollectionRecord:
452 # Docstring inherited from CollectionManager.
453 try:
454 return self._records[key]
455 except KeyError as err:
456 raise MissingCollectionError(f"Collection with key '{key}' not found.") from err
458 def resolve_wildcard(
459 self,
460 wildcard: CollectionWildcard,
461 *,
462 collection_types: Set[CollectionType] = CollectionType.all(),
463 done: set[str] | None = None,
464 flatten_chains: bool = True,
465 include_chains: bool | None = None,
466 ) -> list[CollectionRecord]:
467 # Docstring inherited
468 if done is None: 468 ↛ 470line 468 didn't jump to line 470, because the condition on line 468 was never false
469 done = set()
470 include_chains = include_chains if include_chains is not None else not flatten_chains
472 def resolve_nested(record: CollectionRecord, done: set[str]) -> Iterator[CollectionRecord]:
473 if record.name in done:
474 return
475 if record.type in collection_types:
476 done.add(record.name)
477 if record.type is not CollectionType.CHAINED or include_chains:
478 yield record
479 if flatten_chains and record.type is CollectionType.CHAINED:
480 done.add(record.name)
481 for name in cast(ChainedCollectionRecord, record).children:
482 # flake8 can't tell that we only delete this closure when
483 # we're totally done with it.
484 yield from resolve_nested(self.find(name), done) # noqa: F821
486 result: list[CollectionRecord] = []
488 if wildcard.patterns is Ellipsis:
489 for record in self._records.values():
490 result.extend(resolve_nested(record, done))
491 del resolve_nested
492 return result
493 for name in wildcard.strings:
494 result.extend(resolve_nested(self.find(name), done))
495 if wildcard.patterns:
496 for record in self._records.values():
497 if any(p.fullmatch(record.name) for p in wildcard.patterns):
498 result.extend(resolve_nested(record, done))
499 del resolve_nested
500 return result
502 def getDocumentation(self, key: Any) -> str | None:
503 # Docstring inherited from CollectionManager.
504 sql = (
505 sqlalchemy.sql.select(self._tables.collection.columns.doc)
506 .select_from(self._tables.collection)
507 .where(self._tables.collection.columns[self._collectionIdName] == key)
508 )
509 return self._db.query(sql).scalar()
511 def setDocumentation(self, key: Any, doc: str | None) -> None:
512 # Docstring inherited from CollectionManager.
513 self._db.update(self._tables.collection, {self._collectionIdName: "key"}, {"key": key, "doc": doc})
515 def _setRecordCache(self, records: Iterable[CollectionRecord]) -> None:
516 """Set internal record cache to contain given records,
517 old cached records will be removed.
518 """
519 self._records = {}
520 for record in records:
521 self._records[record.key] = record
523 def _addCachedRecord(self, record: CollectionRecord) -> None:
524 """Add single record to cache."""
525 self._records[record.key] = record
527 def _removeCachedRecord(self, record: CollectionRecord) -> None:
528 """Remove single record from cache."""
529 del self._records[record.key]
531 @abstractmethod
532 def _getByName(self, name: str) -> CollectionRecord | None:
533 """Find collection record given collection name."""
534 raise NotImplementedError()