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 ctrl_mpexec. 

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 

22__all__ = ['PreExecInit'] 

23 

24# ------------------------------- 

25# Imports of standard modules -- 

26# ------------------------------- 

27import logging 

28import itertools 

29 

30# ----------------------------- 

31# Imports for other modules -- 

32# ----------------------------- 

33from lsst.base import Packages 

34from lsst.daf.butler import DatasetType 

35from lsst.pipe.base import PipelineDatasetTypes 

36 

37_LOG = logging.getLogger(__name__.partition(".")[2]) 

38 

39 

40class PreExecInit: 

41 """Initialization of registry for QuantumGraph execution. 

42 

43 This class encapsulates all necessary operations that have to be performed 

44 on butler and registry to prepare them for QuantumGraph execution. 

45 

46 Parameters 

47 ---------- 

48 butler : `~lsst.daf.butler.Butler` 

49 Data butler instance. 

50 taskFactory : `~lsst.pipe.base.TaskFactory` 

51 Task factory. 

52 skipExisting : `bool`, optional 

53 If `True` then do not try to overwrite any datasets that might exist 

54 in ``butler.run``; instead compare them when appropriate/possible. If 

55 `False`, then any existing conflicting dataset will cause a butler 

56 exception to be raised. 

57 """ 

58 def __init__(self, butler, taskFactory, skipExisting=False): 

59 self.butler = butler 

60 self.taskFactory = taskFactory 

61 self.skipExisting = skipExisting 

62 if self.skipExisting and self.butler.run is None: 

63 raise RuntimeError( 

64 "Cannot perform skipExisting logic unless butler is initialized " 

65 "with a default output RUN collection." 

66 ) 

67 

68 def initialize(self, graph, saveInitOutputs=True, registerDatasetTypes=False, saveVersions=True): 

69 """Perform all initialization steps. 

70 

71 Convenience method to execute all initialization steps. Instead of 

72 calling this method and providing all options it is also possible to 

73 call methods individually. 

74 

75 Parameters 

76 ---------- 

77 graph : `~lsst.pipe.base.QuantumGraph` 

78 Execution graph. 

79 saveInitOutputs : `bool`, optional 

80 If ``True`` (default) then save "init outputs", configurations, 

81 and package versions to butler. 

82 registerDatasetTypes : `bool`, optional 

83 If ``True`` then register dataset types in registry, otherwise 

84 they must be already registered. 

85 saveVersions : `bool`, optional 

86 If ``False`` then do not save package versions even if 

87 ``saveInitOutputs`` is set to ``True``. 

88 """ 

89 # register dataset types or check consistency 

90 self.initializeDatasetTypes(graph, registerDatasetTypes) 

91 

92 # Save task initialization data or check that saved data 

93 # is consistent with what tasks would save 

94 if saveInitOutputs: 

95 self.saveInitOutputs(graph) 

96 self.saveConfigs(graph) 

97 if saveVersions: 

98 self.savePackageVersions(graph) 

99 

100 def initializeDatasetTypes(self, graph, registerDatasetTypes=False): 

101 """Save or check DatasetTypes output by the tasks in a graph. 

102 

103 Iterates over all DatasetTypes for all tasks in a graph and either 

104 tries to add them to registry or compares them to exising ones. 

105 

106 Parameters 

107 ---------- 

108 graph : `~lsst.pipe.base.QuantumGraph` 

109 Execution graph. 

110 registerDatasetTypes : `bool`, optional 

111 If ``True`` then register dataset types in registry, otherwise 

112 they must be already registered. 

113 

114 Raises 

115 ------ 

116 ValueError 

117 Raised if existing DatasetType is different from DatasetType 

118 in a graph. 

119 KeyError 

120 Raised if ``registerDatasetTypes`` is ``False`` and DatasetType 

121 does not exist in registry. 

122 """ 

123 pipeline = graph.taskGraph 

124 

125 # Make dataset types for configurations 

126 configDatasetTypes = [DatasetType(taskDef.configDatasetName, {}, 

127 storageClass="Config", 

128 universe=self.butler.registry.dimensions) 

129 for taskDef in pipeline] 

130 

131 # And one dataset type for package versions 

132 packagesDatasetType = DatasetType("packages", {}, 

133 storageClass="Packages", 

134 universe=self.butler.registry.dimensions) 

135 

136 datasetTypes = PipelineDatasetTypes.fromPipeline(pipeline, registry=self.butler.registry) 

137 for datasetType in itertools.chain(datasetTypes.initIntermediates, datasetTypes.initOutputs, 

138 datasetTypes.intermediates, datasetTypes.outputs, 

139 configDatasetTypes, [packagesDatasetType]): 

140 # Only composites are registered, no components, and by this point 

141 # the composite should already exist. 

142 if registerDatasetTypes and not datasetType.isComponent(): 

143 _LOG.debug("Registering DatasetType %s with registry", datasetType) 

144 # this is a no-op if it already exists and is consistent, 

145 # and it raises if it is inconsistent. 

146 self.butler.registry.registerDatasetType(datasetType) 

147 else: 

148 _LOG.debug("Checking DatasetType %s against registry", datasetType) 

149 expected = self.butler.registry.getDatasetType(datasetType.name) 

150 if datasetType.isComponent() \ 

151 and datasetType.parentStorageClass == DatasetType.PlaceholderParentStorageClass: 

152 # Force the parent storage classes to match since we 

153 # are using a placeholder 

154 datasetType.finalizeParentStorageClass(expected.parentStorageClass) 

155 if expected != datasetType: 

156 raise ValueError(f"DatasetType configuration does not match Registry: " 

157 f"{datasetType} != {expected}") 

158 

159 def saveInitOutputs(self, graph): 

160 """Write any datasets produced by initializing tasks in a graph. 

161 

162 Parameters 

163 ---------- 

164 graph : `~lsst.pipe.base.QuantumGraph` 

165 Execution graph. 

166 

167 Raises 

168 ------ 

169 Exception 

170 Raised if ``skipExisting`` is `False` and datasets already 

171 exists. Content of a butler collection may be changed if 

172 exception is raised. 

173 

174 Notes 

175 ----- 

176 If ``skipExisting`` is `True` then existing datasets are not 

177 overwritten, instead we should check that their stored object is 

178 exactly the same as what we would save at this time. Comparing 

179 arbitrary types of object is, of course, non-trivial. Current 

180 implementation only checks the existence of the datasets and their 

181 types against the types of objects produced by tasks. Ideally we 

182 would like to check that object data is identical too but presently 

183 there is no generic way to compare objects. In the future we can 

184 potentially introduce some extensible mechanism for that. 

185 """ 

186 _LOG.debug("Will save InitOutputs for all tasks") 

187 for taskDef in graph.iterTaskGraph(): 

188 task = self.taskFactory.makeTask(taskDef.taskClass, taskDef.config, None, self.butler) 

189 for name in taskDef.connections.initOutputs: 

190 attribute = getattr(taskDef.connections, name) 

191 initOutputVar = getattr(task, name) 

192 objFromStore = None 

193 if self.skipExisting: 

194 # check if it is there already 

195 _LOG.debug("Retrieving InitOutputs for task=%s key=%s dsTypeName=%s", 

196 task, name, attribute.name) 

197 try: 

198 objFromStore = self.butler.get(attribute.name, {}, collections=[self.butler.run]) 

199 # Types are supposed to be identical. 

200 # TODO: Check that object contents is identical too. 

201 if type(objFromStore) is not type(initOutputVar): 

202 raise TypeError(f"Stored initOutput object type {type(objFromStore)} " 

203 f"is different from task-generated type " 

204 f"{type(initOutputVar)} for task {taskDef}") 

205 except LookupError: 

206 pass 

207 if objFromStore is None: 

208 # butler will raise exception if dataset is already there 

209 _LOG.debug("Saving InitOutputs for task=%s key=%s", task, name) 

210 self.butler.put(initOutputVar, attribute.name, {}) 

211 

212 def saveConfigs(self, graph): 

213 """Write configurations for pipeline tasks to butler or check that 

214 existing configurations are equal to the new ones. 

215 

216 Parameters 

217 ---------- 

218 graph : `~lsst.pipe.base.QuantumGraph` 

219 Execution graph. 

220 

221 Raises 

222 ------ 

223 Exception 

224 Raised if ``skipExisting`` is `False` and datasets already exists. 

225 Content of a butler collection should not be changed if exception 

226 is raised. 

227 """ 

228 def logConfigMismatch(msg): 

229 """Log messages about configuration mismatch. 

230 """ 

231 _LOG.fatal("Comparing configuration: %s", msg) 

232 

233 _LOG.debug("Will save Configs for all tasks") 

234 # start transaction to rollback any changes on exceptions 

235 with self.butler.transaction(): 

236 for taskDef in graph.taskGraph: 

237 configName = taskDef.configDatasetName 

238 

239 oldConfig = None 

240 if self.skipExisting: 

241 try: 

242 oldConfig = self.butler.get(configName, {}, collections=[self.butler.run]) 

243 if not taskDef.config.compare(oldConfig, shortcut=False, output=logConfigMismatch): 

244 raise TypeError( 

245 f"Config does not match existing task config {configName!r} in butler; " 

246 "tasks configurations must be consistent within the same run collection") 

247 except LookupError: 

248 pass 

249 if oldConfig is None: 

250 # butler will raise exception if dataset is already there 

251 _LOG.debug("Saving Config for task=%s dataset type=%s", taskDef.label, configName) 

252 self.butler.put(taskDef.config, configName, {}) 

253 

254 def savePackageVersions(self, graph): 

255 """Write versions of software packages to butler. 

256 

257 Parameters 

258 ---------- 

259 graph : `~lsst.pipe.base.QuantumGraph` 

260 Execution graph. 

261 

262 Raises 

263 ------ 

264 Exception 

265 Raised if ``checkExisting`` is ``True`` but versions are not 

266 compatible. 

267 """ 

268 packages = Packages.fromSystem() 

269 _LOG.debug("want to save packages: %s", packages) 

270 datasetType = "packages" 

271 dataId = {} 

272 oldPackages = None 

273 # start transaction to rollback any changes on exceptions 

274 with self.butler.transaction(): 

275 if self.skipExisting: 

276 try: 

277 oldPackages = self.butler.get(datasetType, dataId, collections=[self.butler.run]) 

278 _LOG.debug("old packages: %s", oldPackages) 

279 except LookupError: 

280 pass 

281 if oldPackages is not None: 

282 # Note that because we can only detect python modules that have been imported, the stored 

283 # list of products may be more or less complete than what we have now. What's important is 

284 # that the products that are in common have the same version. 

285 diff = packages.difference(oldPackages) 

286 if diff: 

287 versions_str = "; ".join(f"{pkg}: {diff[pkg][1]} vs {diff[pkg][0]}" for pkg in diff) 

288 raise TypeError(f"Package versions mismatch: ({versions_str})") 

289 else: 

290 _LOG.debug("new packages are consistent with old") 

291 # Update the old set of packages in case we have more packages that haven't been persisted. 

292 extra = packages.extra(oldPackages) 

293 if extra: 

294 _LOG.debug("extra packages: %s", extra) 

295 oldPackages.update(packages) 

296 # have to remove existing dataset first, butler nas no replace option 

297 ref = self.butler.registry.findDataset(datasetType, dataId, collections=[self.butler.run]) 

298 self.butler.pruneDatasets([ref], unstore=True, purge=True) 

299 self.butler.put(oldPackages, datasetType, dataId) 

300 else: 

301 self.butler.put(packages, datasetType, dataId)