22 from __future__
import absolute_import, division
28 from builtins
import str
29 from builtins
import object
32 from lsst.base
import disableImplicitThreading
33 import lsst.afw.table
as afwTable
34 from .task
import Task, TaskError
35 from .struct
import Struct
36 from .argumentParser
import ArgumentParser
37 from lsst.base
import Packages
38 from lsst.log
import Log
40 __all__ = [
"CmdLineTask",
"TaskRunner",
"ButlerInitializedTaskRunner"]
43 def _poolFunctionWrapper(function, arg):
44 """Wrapper around function to catch exceptions that don't inherit from `Exception`. 46 Such exceptions aren't caught by multiprocessing, which causes the slave process to crash and you end up 55 cls, exc, tb = sys.exc_info()
56 log = Log.getDefaultLogger()
57 log.warn(
"Unhandled exception %s (%s):\n%s" % (cls.__name__, exc, traceback.format_exc()))
58 raise Exception(
"Unhandled exception: %s (%s)" % (cls.__name__, exc))
61 def _runPool(pool, timeout, function, iterable):
62 """Wrapper around ``pool.map_async``, to handle timeout 64 This is required so as to trigger an immediate interrupt on the KeyboardInterrupt (Ctrl-C); see 65 http://stackoverflow.com/questions/1408356/keyboard-interrupts-with-pythons-multiprocessing-pool 67 Further wraps the function in ``_poolFunctionWrapper`` to catch exceptions 68 that don't inherit from `Exception`. 70 return pool.map_async(functools.partial(_poolFunctionWrapper, function), iterable).get(timeout)
73 @contextlib.contextmanager
75 """Context manager for profiling with cProfile. 81 Filename to which to write profile (profiling disabled if `None` or empty). 82 log : `lsst.log.Log`, optional 83 Log object for logging the profile operations. 85 If profiling is enabled, the context manager returns the cProfile.Profile object (otherwise 86 it returns None), which allows additional control over profiling. You can obtain this using 87 the "as" clause, e.g.: 89 with profile(filename) as prof: 92 The output cumulative profile can be printed with a command-line like:: 94 python -c 'import pstats; pstats.Stats("<filename>").sort_stats("cumtime").print_stats(30)' 100 from cProfile
import Profile
103 log.info(
"Enabling cProfile profiling")
107 profile.dump_stats(filename)
109 log.info(
"cProfile stats written to %s" % filename)
113 """Run a command-line task, using `multiprocessing` if requested. 117 TaskClass : `lsst.pipe.base.Task` subclass 118 The class of the task to run. 119 parsedCmd : `argparse.Namespace` 120 The parsed command-line arguments, as returned by the task's argument parser's 121 `~lsst.pipe.base.ArgumentParser.parse_args` method. 125 Do not store ``parsedCmd``, as this instance is pickled (if multiprocessing) and parsedCmd may 126 contain non-picklable elements. It certainly contains more data than we need to send to each 127 instance of the task. 128 doReturnResults : `bool`, optional 129 Should run return the collected result from each invocation of the task? This is only intended for 130 unit tests and similar use. It can easily exhaust memory (if the task returns enough data and you 131 call it enough times) and it will fail when using multiprocessing if the returned data cannot be 134 Note that even if ``doReturnResults`` is False a struct with a single member "exitStatus" is returned, 135 with value 0 or 1 to be returned to the unix shell. 140 If multiprocessing is requested (and the task supports it) but the multiprocessing library cannot be 145 Each command-line task (subclass of `lsst.pipe.base.CmdLineTask`) has a task runner. By default it is this 146 class, but some tasks require a subclass. See the manual :ref:`creating-a-command-line-task` for more 147 information. See `CmdLineTask.parseAndRun` to see how a task runner is used. 149 You may use this task runner for your command-line task if your task has a run method that takes exactly 150 one argument: a butler data reference. Otherwise you must provide a task-specific subclass of this runner 151 for your task's ``RunnerClass`` that overrides `TaskRunner.getTargetList` and possibly 152 `TaskRunner.__call__`. See `TaskRunner.getTargetList` for details. 154 This design matches the common pattern for command-line tasks: the run method takes a single data 155 reference, of some suitable name. Additional arguments are rare, and if present, require a subclass of 156 `TaskRunner` that calls these additional arguments by name. 158 Instances of this class must be picklable in order to be compatible with multiprocessing. If 159 multiprocessing is requested (``parsedCmd.numProcesses > 1``) then `run` calls `prepareForMultiProcessing` 160 to jettison optional non-picklable elements. If your task runner is not compatible with multiprocessing 161 then indicate this in your task by setting class variable ``canMultiprocess=False``. 163 Due to a `python bug`__, handling a `KeyboardInterrupt` properly `requires specifying a timeout`__. This 164 timeout (in sec) can be specified as the ``timeout`` element in the output from 165 `~lsst.pipe.base.ArgumentParser` (the ``parsedCmd``), if available, otherwise we use `TaskRunner.TIMEOUT`. 167 .. __: http://bugs.python.org/issue8296 168 .. __: http://stackoverflow.com/questions/1408356/keyboard-interrupts-with-pythons-multiprocessing-pool 172 """Default timeout (seconds) for multiprocessing.""" 174 def __init__(self, TaskClass, parsedCmd, doReturnResults=False):
184 self.
timeout = getattr(parsedCmd,
'timeout',
None)
189 if not TaskClass.canMultiprocess:
190 self.
log.warn(
"This task does not support multiprocessing; using one process")
194 """Prepare this instance for multiprocessing 196 Optional non-picklable elements are removed. 198 This is only called if the task is run under multiprocessing. 203 """Run the task on all targets. 207 parsedCmd : `argparse.Namespace` 208 Parsed command `argparse.Namespace`. 213 A list of results returned by `TaskRunner.__call__`, or an empty list if `TaskRunner.__call__` 214 is not called (e.g. if `TaskRunner.precall` returns `False`). See `TaskRunner.__call__` 219 The task is run under multiprocessing if `TaskRunner.numProcesses` is more than 1; otherwise 220 processing is serial. 224 disableImplicitThreading()
225 import multiprocessing
227 pool = multiprocessing.Pool(processes=self.
numProcesses, maxtasksperchild=1)
228 mapFunc = functools.partial(_runPool, pool, self.
timeout)
234 profileName = parsedCmd.profile
if hasattr(parsedCmd,
"profile")
else None 237 if len(targetList) > 0:
238 with
profile(profileName, log):
240 resultList = list(mapFunc(self, targetList))
242 log.warn(
"Not running the task because there is no data to process; " 243 "you may preview data using \"--show data\"")
253 """Get a list of (dataRef, kwargs) for `TaskRunner.__call__`. 257 parsedCmd : `argparse.Namespace` 258 The parsed command object returned by `lsst.pipe.base.argumentParser.ArgumentParser.parse_args`. 260 Any additional keyword arguments. In the default `TaskRunner` this is an empty dict, but having 261 it simplifies overriding `TaskRunner` for tasks whose run method takes additional arguments 262 (see case (1) below). 266 The default implementation of `TaskRunner.getTargetList` and `TaskRunner.__call__` works for any 267 command-line task whose run method takes exactly one argument: a data reference. Otherwise you 268 must provide a variant of TaskRunner that overrides `TaskRunner.getTargetList` and possibly 269 `TaskRunner.__call__`. There are two cases. 273 If your command-line task has a ``run`` method that takes one data reference followed by additional 274 arguments, then you need only override `TaskRunner.getTargetList` to return the additional arguments 275 as an argument dict. To make this easier, your overridden version of `~TaskRunner.getTargetList` may 276 call `TaskRunner.getTargetList` with the extra arguments as keyword arguments. For example, the 277 following adds an argument dict containing a single key: "calExpList", whose value is the list of data 278 IDs for the calexp ID argument:: 280 def getTargetList(parsedCmd): 281 return TaskRunner.getTargetList( 283 calExpList=parsedCmd.calexp.idList 286 It is equivalent to this slightly longer version:: 289 def getTargetList(parsedCmd): 290 argDict = dict(calExpList=parsedCmd.calexp.idList) 291 return [(dataId, argDict) for dataId in parsedCmd.id.idList] 295 If your task does not meet condition (1) then you must override both TaskRunner.getTargetList and 296 `TaskRunner.__call__`. You may do this however you see fit, so long as `TaskRunner.getTargetList` 297 returns a list, each of whose elements is sent to `TaskRunner.__call__`, which runs your task. 299 return [(ref, kwargs)
for ref
in parsedCmd.id.refList]
302 """Create a Task instance. 307 Parsed command-line options (used for extra task args by some task runners). 309 Args tuple passed to `TaskRunner.__call__` (used for extra task arguments by some task runners). 313 ``makeTask`` can be called with either the ``parsedCmd`` argument or ``args`` argument set to None, 314 but it must construct identical Task instances in either case. 316 Subclasses may ignore this method entirely if they reimplement both `TaskRunner.precall` and 317 `TaskRunner.__call__`. 321 def _precallImpl(self, task, parsedCmd):
322 """The main work of `precall`. 324 We write package versions, schemas and configs, or compare these to existing files on disk if present. 326 if not parsedCmd.noVersions:
327 task.writePackageVersions(parsedCmd.butler, clobber=parsedCmd.clobberVersions)
332 """Hook for code that should run exactly once, before multiprocessing. 336 Must return True if `TaskRunner.__call__` should subsequently be called. 340 Implementations must take care to ensure that no unpicklable 341 attributes are added to the TaskRunner itself, for compatibility 342 with multiprocessing. 344 The default implementation writes package versions, schemas and configs, or compares them to existing 345 files on disk if present. 347 task = self.
makeTask(parsedCmd=parsedCmd)
354 except Exception
as e:
355 task.log.fatal(
"Failed in task initialization: %s", e)
356 if not isinstance(e, TaskError):
357 traceback.print_exc(file=sys.stderr)
362 """Run the Task on a single target. 367 Arguments for Task.run() 371 struct : `lsst.pipe.base.Struct` 372 Contains these fields if ``doReturnResults`` is `True`: 374 - ``dataRef``: the provided data reference. 375 - ``metadata``: task metadata after execution of run. 376 - ``result``: result returned by task run, or `None` if the task fails. 378 If ``doReturnResults`` is `False` the struct contains: 384 This default implementation assumes that the ``args`` is a tuple 385 containing a data reference and a dict of keyword arguments. 389 If you override this method and wish to return something when ``doReturnResults`` is `False`, 390 then it must be picklable to support multiprocessing and it should be small enough that pickling 391 and unpickling do not add excessive overhead. 393 dataRef, kwargs = args
395 self.
log = Log.getDefaultLogger()
396 if hasattr(dataRef,
"dataId"):
397 self.
log.MDC(
"LABEL", str(dataRef.dataId))
398 elif isinstance(dataRef, (list, tuple)):
399 self.
log.MDC(
"LABEL", str([ref.dataId
for ref
in dataRef
if hasattr(ref,
"dataId")]))
404 result = task.run(dataRef, **kwargs)
407 result = task.run(dataRef, **kwargs)
408 except Exception
as e:
413 if hasattr(dataRef,
"dataId"):
414 task.log.fatal(
"Failed on dataId=%s: %s", dataRef.dataId, e)
415 elif isinstance(dataRef, (list, tuple)):
416 task.log.fatal(
"Failed on dataId=[%s]: %s",
417 ", ".join(str(ref.dataId)
for ref
in dataRef), e)
419 task.log.fatal(
"Failed on dataRef=%s: %s", dataRef, e)
421 if not isinstance(e, TaskError):
422 traceback.print_exc(file=sys.stderr)
423 task.writeMetadata(dataRef)
426 self.
log.MDCRemove(
"LABEL")
430 exitStatus=exitStatus,
432 metadata=task.metadata,
437 exitStatus=exitStatus,
442 """A TaskRunner for `CmdLineTask`\ s that require a ``butler`` keyword argument to be passed to 447 """A variant of the base version that passes a butler argument to the task's constructor. 451 parsedCmd : `argparse.Namespace` 452 Parsed command-line options, as returned by the `~lsst.pipe.base.ArgumentParser`; if specified 453 then args is ignored. 455 Other arguments; if ``parsedCmd`` is `None` then this must be specified. 460 Raised if ``parsedCmd`` and ``args`` are both `None`. 462 if parsedCmd
is not None:
463 butler = parsedCmd.butler
464 elif args
is not None:
465 dataRef, kwargs = args
466 butler = dataRef.butlerSubset.butler
468 raise RuntimeError(
"parsedCmd or args must be specified")
473 """Base class for command-line tasks: tasks that may be executed from the command-line. 477 See :ref:`task-framework-overview` to learn what tasks are and :ref:`creating-a-command-line-task` for 478 more information about writing command-line tasks. 480 Subclasses must specify the following class variables: 482 - ``ConfigClass``: configuration class for your task (a subclass of `lsst.pex.config.Config`, or if your 483 task needs no configuration, then `lsst.pex.config.Config` itself). 484 - ``_DefaultName``: default name used for this task (a str). 486 Subclasses may also specify the following class variables: 488 - ``RunnerClass``: a task runner class. The default is ``TaskRunner``, which works for any task 489 with a run method that takes exactly one argument: a data reference. If your task does 490 not meet this requirement then you must supply a variant of ``TaskRunner``; see ``TaskRunner`` 491 for more information. 492 - ``canMultiprocess``: the default is `True`; set `False` if your task does not support multiprocessing. 494 Subclasses must specify a method named ``run``: 496 - By default ``run`` accepts a single butler data reference, but you can specify an alternate task runner 497 (subclass of ``TaskRunner``) as the value of class variable ``RunnerClass`` if your run method needs 499 - ``run`` is expected to return its data in a `lsst.pipe.base.Struct`. This provides safety for evolution 500 of the task since new values may be added without harming existing code. 501 - The data returned by ``run`` must be picklable if your task is to support multiprocessing. 503 RunnerClass = TaskRunner
504 canMultiprocess =
True 508 """A hook to allow a task to change the values of its config *after* the camera-specific 509 overrides are loaded but before any command-line overrides are applied. 513 config : instance of task's ``ConfigClass`` 518 This is necessary in some cases because the camera-specific overrides may retarget subtasks, 519 wiping out changes made in ConfigClass.setDefaults. See LSST Trac ticket #2282 for more discussion. 523 This is called by CmdLineTask.parseAndRun; other ways of constructing a config will not apply 529 def parseAndRun(cls, args=None, config=None, log=None, doReturnResults=False):
530 """Parse an argument list and run the command. 534 args : `list`, optional 535 List of command-line arguments; if `None` use `sys.argv`. 536 config : `lsst.pex.config.Config`-type, optional 537 Config for task. If `None` use `Task.ConfigClass`. 538 log : `lsst.log.Log`-type, optional 539 Log. If `None` use the default log. 540 doReturnResults : `bool`, optional 541 If `True`, return the results of this task. Default is `False`. This is only intended for 542 unit tests and similar use. It can easily exhaust memory (if the task returns enough data and you 543 call it enough times) and it will fail when using multiprocessing if the returned data cannot be 548 struct : `lsst.pipe.base.Struct` 551 - ``argumentParser``: the argument parser. 552 - ``parsedCmd``: the parsed command returned by the argument parser's 553 `lsst.pipe.base.ArgumentParser.parse_args` method. 554 - ``taskRunner``: the task runner used to run the task (an instance of `Task.RunnerClass`). 555 - ``resultList``: results returned by the task runner's ``run`` method, one entry per invocation. 556 This will typically be a list of `None` unless ``doReturnResults`` is `True`; 557 see `Task.RunnerClass` (`TaskRunner` by default) for more information. 561 Calling this method with no arguments specified is the standard way to run a command-line task 562 from the command-line. For an example see ``pipe_tasks`` ``bin/makeSkyMap.py`` or almost any other 563 file in that directory. 565 If one or more of the dataIds fails then this routine will exit (with a status giving the 566 number of failed dataIds) rather than returning this struct; this behaviour can be 567 overridden by specifying the ``--noExit`` command-line option. 570 commandAsStr =
" ".join(sys.argv)
573 commandAsStr =
"{}{}".format(lsst.utils.get_caller_name(skip=1), tuple(args))
577 config = cls.ConfigClass()
578 parsedCmd = argumentParser.parse_args(config=config, args=args, log=log, override=cls.
applyOverrides)
580 parsedCmd.log.info(
"Running: %s", commandAsStr)
582 taskRunner = cls.
RunnerClass(TaskClass=cls, parsedCmd=parsedCmd, doReturnResults=doReturnResults)
583 resultList = taskRunner.run(parsedCmd)
586 nFailed = sum(((res.exitStatus != 0)
for res
in resultList))
587 except Exception
as e:
588 parsedCmd.log.warn(
"Unable to retrieve exit status (%s); assuming success", e)
593 parsedCmd.log.warn(
"%d dataRefs failed; not exiting as --noExit was set", nFailed)
598 argumentParser=argumentParser,
600 taskRunner=taskRunner,
601 resultList=resultList,
605 def _makeArgumentParser(cls):
606 """Create and return an argument parser. 610 parser : `lsst.pipe.base.ArgumentParser` 611 The argument parser for this task. 615 By default this returns an `~lsst.pipe.base.ArgumentParser` with one ID argument named `--id` of 616 dataset type ``raw``. 618 Your task subclass may need to override this method to change the dataset type or data ref level, 619 or to add additional data ID arguments. If you add additional data ID arguments or your task's 620 run method takes more than a single data reference then you will also have to provide a task-specific 621 task runner (see TaskRunner for more information). 624 parser.add_id_argument(name=
"--id", datasetType=
"raw",
625 help=
"data IDs, e.g. --id visit=12345 ccd=1,2^0,3")
629 """Write the configuration used for processing the data, or check that an existing 630 one is equal to the new one if present. 634 butler : `lsst.daf.persistence.Butler` 635 Data butler used to write the config. The config is written to dataset type 636 `CmdLineTask._getConfigName`. 637 clobber : `bool`, optional 638 A boolean flag that controls what happens if a config already has been saved: 639 - `True`: overwrite or rename the existing config, depending on ``doBackup``. 640 - `False`: raise `TaskError` if this config does not match the existing config. 641 doBackup : bool, optional 642 Set to `True` to backup the config files if clobbering. 645 if configName
is None:
648 butler.put(self.
config, configName, doBackup=doBackup)
649 elif butler.datasetExists(configName):
652 oldConfig = butler.get(configName, immediate=
True)
653 except Exception
as exc:
654 raise type(exc)(
"Unable to read stored config file %s (%s); consider using --clobber-config" %
657 def logConfigMismatch(msg):
658 self.
log.fatal(
"Comparing configuration: %s", msg)
660 if not self.
config.compare(oldConfig, shortcut=
False, output=logConfigMismatch):
662 (
"Config does not match existing task config %r on disk; tasks configurations " +
663 "must be consistent within the same output repo (override with --clobber-config)") %
666 butler.put(self.
config, configName)
669 """Write the schemas returned by `lsst.pipe.base.Task.getAllSchemaCatalogs`. 673 butler : `lsst.daf.persistence.Butler` 674 Data butler used to write the schema. Each schema is written to the dataset type specified as the 675 key in the dict returned by `~lsst.pipe.base.Task.getAllSchemaCatalogs`. 676 clobber : `bool`, optional 677 A boolean flag that controls what happens if a schema already has been saved: 678 - `True`: overwrite or rename the existing schema, depending on ``doBackup``. 679 - `False`: raise `TaskError` if this schema does not match the existing schema. 680 doBackup : `bool`, optional 681 Set to `True` to backup the schema files if clobbering. 685 If ``clobber`` is `False` and an existing schema does not match a current schema, 686 then some schemas may have been saved successfully and others may not, and there is no easy way to 690 schemaDataset = dataset +
"_schema" 692 butler.put(catalog, schemaDataset, doBackup=doBackup)
693 elif butler.datasetExists(schemaDataset):
694 oldSchema = butler.get(schemaDataset, immediate=
True).getSchema()
695 if not oldSchema.compare(catalog.getSchema(), afwTable.Schema.IDENTICAL):
697 (
"New schema does not match schema %r on disk; schemas must be " +
698 " consistent within the same output repo (override with --clobber-config)") %
701 butler.put(catalog, schemaDataset)
704 """Write the metadata produced from processing the data. 709 Butler data reference used to write the metadata. 710 The metadata is written to dataset type `CmdLineTask._getMetadataName`. 714 if metadataName
is not None:
716 except Exception
as e:
717 self.
log.warn(
"Could not persist metadata for dataId=%s: %s", dataRef.dataId, e)
720 """Compare and write package versions. 724 butler : `lsst.daf.persistence.Butler` 725 Data butler used to read/write the package versions. 726 clobber : `bool`, optional 727 A boolean flag that controls what happens if versions already have been saved: 728 - `True`: overwrite or rename the existing version info, depending on ``doBackup``. 729 - `False`: raise `TaskError` if this version info does not match the existing. 730 doBackup : `bool`, optional 731 If `True` and clobbering, old package version files are backed up. 732 dataset : `str`, optional 733 Name of dataset to read/write. 738 Raised if there is a version mismatch with current and persisted lists of package versions. 742 Note that this operation is subject to a race condition. 744 packages = Packages.fromSystem()
747 return butler.put(packages, dataset, doBackup=doBackup)
748 if not butler.datasetExists(dataset):
749 return butler.put(packages, dataset)
752 old = butler.get(dataset, immediate=
True)
753 except Exception
as exc:
754 raise type(exc)(
"Unable to read stored version dataset %s (%s); " 755 "consider using --clobber-versions or --no-versions" %
760 diff = packages.difference(old)
763 "Version mismatch (" +
764 "; ".join(
"%s: %s vs %s" % (pkg, diff[pkg][1], diff[pkg][0])
for pkg
in diff) +
765 "); consider using --clobber-versions or --no-versions")
767 extra = packages.extra(old)
770 butler.put(old, dataset, doBackup=doBackup)
772 def _getConfigName(self):
773 """Get the name of the config dataset type, or `None` if config is not to be persisted. 777 The name may depend on the config; that is why this is not a class method. 779 return self._DefaultName +
"_config" 781 def _getMetadataName(self):
782 """Get the name of the metadata dataset type, or `None` if metadata is not to be persisted. 786 The name may depend on the config; that is why this is not a class method. 788 return self._DefaultName +
"_metadata"
def _makeArgumentParser(cls)
def parseAndRun(cls, args=None, config=None, log=None, doReturnResults=False)
def _precallImpl(self, task, parsedCmd)
def getFullMetadata(self)
def writePackageVersions(self, butler, clobber=False, doBackup=True, dataset="packages")
def getAllSchemaCatalogs(self)
def writeSchemas(self, butler, clobber=False, doBackup=True)
def prepareForMultiProcessing(self)
def _getMetadataName(self)
def makeTask(self, parsedCmd=None, args=None)
def writeMetadata(self, dataRef)
def precall(self, parsedCmd)
def __init__(self, TaskClass, parsedCmd, doReturnResults=False)
def profile(filename, log=None)
def makeTask(self, parsedCmd=None, args=None)
def getTargetList(parsedCmd, kwargs)
def writeConfig(self, butler, clobber=False, doBackup=True)
def applyOverrides(cls, config)