lsst-pipe-base v26.0.0 (2023-09-22)

New Features

  • Added system for obtaining data ID packer objects from the combination of an Instrument class and configuration. (DM-31924)

  • Added a PipelineGraph class that represents a Pipeline with all configuration overrides applied as a graph. (DM-33027)

  • Added new command butler transfer-from-graph to transfer results of execution with Quantum-backed butler. (DM-33497)

  • buildExecutionButler method now supports input graph with all dataset references resolved. (DM-37582)

  • Added convince methods to the Python api for Pipelines. These methods allow merging pipelines, adding labels to / removing labels from subsets, and finding subsets containing a specified label. (DM-37655)

  • An Instrument can now specify the dataset type definition that it would like to use for raw data. This can be done by setting the raw_definition class property to a tuple of the dataset type name, the dimensions to use for this dataset type, and the storage class name. (DM-37950)

  • Modified InMemoryDatasetHandle to allow it to be constructed with keyword arguments that will be converted to the relevant DataId. (DM-38091)

  • Modified InMemoryDatasetHandle to allow it to be configured to always deep copy the Python object on get(). (DM-38694)

  • Revived bit-rotted support for “mocked” PipelineTask execution and moved it here (from ctrl_mpexec). (DM-38952)

  • Formalized support for modifying connections in PipelineTaskConnections.__init__ implementations.

    Connections can now be added, removed, or replaced with normal attribute syntax. Removing entries from e.g. self.inputs in __init__ still works for backwards compatibility, but deleting attributes is generally preferred. The task dimensions can also be replaced or modified in place in __init__. (DM-38953)

  • Added a method on PipelineTaskConfig objects named applyConfigOverrides. This method is called by the system executing PipelineTasks within a pipeline, and is passed the instrument and config overrides defined within the pipeline for that task. (DM-39100)

  • Add Instrument.make_default_dimension_packer to restore simple access to the default data ID packer for an instrument. (DM-39453)

  • The back-end to quantum graph loading has been optimized such that duplicate objects are not created in memory, but create shared references. This results in a large decrease in memory usage, and decrease in load times. (DM-39582)

    • A new class ExecutionResources has been created to record the number of cores and memory that has been allocated for the execution of a quantum.

    • QuantumContext (newly renamed from ButlerQuantumContext) now has a resources property that can be queried by a task in runQuantum. This can be used to tell the task that it can use multiple cores or possibly should make a more efficient use of the available memory resources. (DM-39661)

  • Made it possible to deprecate PipelineTask connections. (DM-39902)

  • Parameters defined in a Pipeline can now be used within a config Python block as well as within config files loaded by a Pipeline. (DM-40198)

  • When looking up prerequisite inputs with skypix data IDs (e.g., reference catalogs) for a quantum whose data ID is not spatial, use the union of the spatial regions of the input and output datasets as a constraint.

    This keeps global sequence-point tasks from being given all such datasets in the input collections. (DM-40243)

  • Added support for init-input/output datasets in PipelineTask mocking. (DM-40381)

API Changes

  • Several changes to API to add support for QuantumBackedButler:

    • Added a globalInitOutputRefs method to the QuantumGraph class which returns global per-graph output dataset references (e.g. for “packages” dataset type).

    • ButlerQuantumContext can work with either Butler or LimitedButler. Its __init__ method should not be used directly, instead one of the two new class methods should be used - from_full or from_limited.

    • The ButlerQuantumContext.registry attribute was removed, and ButlerQuantumContext.dimensions has been added to hold the DimensionUniverse.

    • The abstract method TaskFactory.makeTask was updated and simplified to accept TaskDef and LimitedButler. (DM-33497)

    • ButlerQuantumContext was updated to only need a LimitedButler.

    • Factory methods from_full and from_limited were dropped, a constructor accepting a LimitedButler instance is now used to make instances. (DM-37704)

    • Added method QuantumGraph.updateRun. This new method updates run collection name and dataset IDs for all output and intermediate datasets in a graph, allowing the graph to be reused.

    • GraphBuilder.makeGraph method dropped the resolveRefs argument, the builder now always makes resolved references. The run argument is now required to be non-empty string. (DM-38780)

Bug Fixes

  • Fixed a bug that led to valid storage class conversions being rejected when using execution butler. (DM-38614)

  • Fixed a bug related to checking component datasets in execution butler creation, introduced in DM-38614. (DM-38888)

  • Fixed handling of storage classes in QuantumGraph generation.

    This could lead to a failure downstream in execution butler creation, and would likely have led to problems with Quantum-Backed Butler usage as well. (DM-39198)

  • Fixed a bug in QuantumGraph generation that could result in datasets from skip_existing_in collections being used as outputs, and another that prevented QuantumGraph generation when a skip_existing_in collection has some outputs from a failed quantum. (DM-39672)

  • Fixed a bug in quantum graph builder which resulted in missing datastore records for calibration datasets. This bug was causing failures for pipetask execution with quantum-backed butler. (DM-40254)

  • Ensured QuantumGraphs are built with datastore records for init-input datasets that might have been produced by another task in the pipeline, but will not be because all quanta for that task were skipped due to existing outputs. (DM-40381)

  • QuantumGraph.updateRun() method was fixed to update dataset ID in references which have their run collection changed. (DM-40392)

Other Changes and Additions

  • Modified the calling signature for the Task constructor such that only the config parameter can be positional. All other parameters must now be keyword parameters. (DM-15325)

  • The Struct class is now a subclass of SimpleNamespace. (DM-36649)

  • The DuplicateOutputError logger now produces a more helpful error message. (DM-38234)

    • Execution butler creation has been changed to use the DatasetRefs from the graph rather than creating new registry entries from the dataIDs. This is possible now that the graph is always created with resolved refs and ensures that provenance is consistent between the graph and the outputs.

    • This change to execution butler required that ButlerQuantumContext.put() no longer unresolves the graph DatasetRef (otherwise there would be a dataset ID mismatch). This results in the dataset always using the output run defined in the graph even if the Butler was created with a different default run. (DM-38779)

  • Stopped sorting Pipeline elements on read.

    Ordering specified in pipeline files is now preserved instead. (DM-38953)

  • Loosened documentation of QuantumGraph.inputQuanta and outputQuanta. They are not guaranteed to be (and currently are not) lists, so the new documentation describes them as iterables.

    Documented universe constructor parameter to QuantumGraph.

    Brought QuantumGraph property docs in line with DM standards.

An API Removal or Deprecation

    • Removed deprecated kwargs parameter from in-memory equivalent dataset handle.

    • Removed deprecated pipe_base timer module (it was moved to utils).

    • Removed the warning from deprecated PipelineIR._read_imports and replaced with a raise.

    • Removed the warning from deprecated Pipeline._parse_file_specifier and replaced with a raise.

    • Removed deprecated methods from TaskMetadata. (DM-37534)

    • The PipelineTaskConfig.saveMetadata field is now deprecated and will be removed after v26. Its value is ignored and task metadata is always saved.

    • The ResourceConfig class has been removed; it was never used. (DM-39377)

  • Deprecated the reconstituteDimensions argument from QuantumNode.from_simple (DM-39582)

  • ButlerQuantumContext has been renamed to QuantumContext. This reflects the additional functionality it now has. (DM-39661)

  • Removed support for reading quantum graphs in pickle format. (DM-40032)

lsst-pipe-base v25.0.0 (2023-02-28)

This is the first release without any support for the Generation 2 middleware.

New Features

  • Added PipelineStepTester class, to enable testing that multi-step pipelines are able to run without error. (DM-33779)

  • QuantumGraph now saves the DimensionUniverse it was created with when it is persisted. This removes the need to explicitly pass the DimensionUniverse when loading a saved graph. (DM-35082)

    • Added support for transferring files into execution butler. (DM-35494)

  • A new class InMemoryDatasetHandle is now available. This class provides a variant of lsst.daf.butler.DeferredDatasetHandle that does not require a butler and lets you store your in-memory objects in something that looks like one and so can be passed to Task.run() methods that expect to be able to do deferred loading. (DM-35741)

    • Add unit test to cover the new getNumberOfQuantaForTask method.

    • Add graph interface, getNumberOfQuantaForTask, to determine number of quanta associated with a given taskDef.

    • Modifications to getQuantaForTask to support showing added additional quanta information in the logger. (DM-36145)

  • Allow PipelineTasks to provide defaults for the --dataset-query-constraints option for the pipetask tool. (DM-37786)

API Changes

  • ButlerQuantumContext.get method can accept None as a reference and returns None as a result object. (DM-35752)

  • GraphBuilder.makeGraph method adds bind parameter for bind values to use with the user expression. (DM-36487)

  • InMemoryDatasetHandle now supports storage class conversion on get(). (DM-4551)

Bug Fixes

  • lsst.pipe.base.testUtils.makeQuantum no longer crashes if given a connection that is set to a dataset component. (DM-35721)

  • Ensure QuantumGraphs are given a DimensionUniverse at construction.

    This fixes a mostly-spurious dimension universe inconsistency warning when reading QuantumGraphs, introduced on DM-35082. (DM-35681)

  • Fixed an error message that says that repository state has changed during QuantumGraph generation when init input datasets are just missing. (DM-37786)

Other Changes and Additions

  • Make diagnostic logging for empty QuantumGraphs harder to ignore.

    Log messages have been upgraded from WARNING to FATAL, and an exception traceback that tends to hide them has been removed. (DM-36360)

An API Removal or Deprecation

  • Removed the Task.getSchemaCatalogs and Task.getAllSchemaCatalogs APIs. These were used by CmdLineTask but are no longer used in the current middleware. (DM-2850)

  • Relocated lsst.pipe.base.cmdLineTask.profile to lsst.utils.timer.profile. This was relocated as part of the Gen2 removal that includes the removal of CmdLineTask. (DM-35697)

    • ArgumentParser, CmdLineTask, and TaskRunner classes have been removed and associated gen2 documentation.

    • The PipelineIR.from_file() method has been removed.

    • The getTaskLogger function has been removed. (DM-35917)

  • Replaced CmdLineTask and ArgumentParser with non-functioning stubs, disabling all Gen2 functionality. A deprecation message is now issued but the classes do nothing. (DM-35675)

lsst-pipe-base v24.0.0 (2022-08-26)

New Features

  • Add the ability for user control over dataset constraints in QuantumGraph creation. (DM-31769)

  • Builds using setuptools now calculate versions from the Git repository, including the use of alpha releases for those associated with weekly tags. (DM-32408)

  • Improve diagnostics for empty QuantumGraph. (DM-32459)

  • A new class has been written for handling Task metadata. lsst.pipe.base.TaskMetadata will in future become the default metadata class for Task, replacing lsst.daf.base.PropertySet. The new metadata class is not yet enabled by default. (DM-32682)

    • Add TaskMetadata.to_dict() method (this is now used by the lsst.daf.base.PropertySet.from_mapping() method and triggered by the Butler if type conversion is needed).

    • Use the existing metadata storage class definition if one already exists in a repository.

    • Switch Task to use TaskMetadata for storing task metadata, rather than lsst.daf.base.PropertySet. This removes a C++ dependency from the middleware. (DM-33155)

    • Added lsst.pipe.base.Instrument to represent an instrument in Butler registry.

    • Added butler register-instrument command (relocated from obs_base).

    • Added a formatter for pex_config Config objects. (DM-34105)

Bug Fixes

  • Fixed a bug where imported pipeline parameters were taking preference over “top-level” preferences (DM-32080)

Other Changes and Additions

  • If a PipelineTask has connections that have a different storage class for a dataset type than the one defined in registry, this will now be allowed if the storage classes are compatible. The Task run() method will be given the Python type it expects and can return the Python type it has declared it returns. The Butler will do the type conversion automatically. (DM-33303)

  • Topological sorting of pipelines on write has been disabled; the order in which the pipeline tasks were read/added is preserved instead. This makes it unnecessary to import all tasks referenced by the pipeline in order to write it. (DM-34155)

lsst-pipe-base v23.0.1 (2022-02-02)

Miscellaneous Changes of Minor Interest

  • Execution butler creation time has been reduced significantly by avoiding unnecessary checks for existence of files in the datastore. (DM-33345)

lsst-pipe-base v23.0.0 (2021-12-10)

New Features

  • Added a new facility for creating “lightweight” (execution) butlers that pre-fills a local SQLite registry. This can allow a pipeline to be executed without talking to the main registry. (DM-28646)

  • Allow PipelineTasks inputs and outputs to be optional under certain conditions, so tasks with no work to do can be skipped without blocking downstream tasks from running. (DM-30649)

  • Log diagnostic information when QuantumGraphs are empty because the initial query yielded no results.

    At present, these diagnostics only cover missing input datasets, which is a common way to get an empty QuantumGraph, but not the only way. (DM-31583)

API Changes

  • GraphBuilder constructor boolean argument skipExisting is replaced with skipExistingIn which accepts collections to check for existing quantum outputs. (DM-27492)

Other Changes and Additions

  • The logger associated with Task is now derived from a Python logging.Logger and not lsst.log.Log. This logger includes a new verbose() log method as an intermediate between INFO and DEBUG. (DM-30301)

  • Added metadata to QuantumGraphs. This changed the on disk save format, but is backwards compatible with graphs saved with previous versions of the QuantumGraph code. (DM-30702)

  • All Doxygen documentation has been removed and replaced by Sphinx. (DM-23330)

  • New documentation on writing pipelines has been added. (DM-27416)

lsst-pipe-base v22.0 (2021-04-01)

New Features

  • Add ways to test a PipelineTask’s init inputs/outputs [DM-23156]

  • Pipelines can now support URIs [DM-28036]

  • Graph files can now be loaded and saved via URIs [DM-27682]

  • A new format for saving graphs has been developed (with a .qgraph extension). This format supports the ability to read a subset of a graph from an object store. [DM-27784]

  • Graph building with a pipeline that specifies an instrument no longer needs an explicit instrument to be given. [DM-27985]

  • A parameters section has been added to pipeline definitions. [DM-27633]