parsl.data_provider.staging.Staging

class parsl.data_provider.staging.Staging[source]

This class defines the interface for file staging providers.

For each file to be staged in, the data manager will present the file to each configured Staging provider in turn: first, it will ask if the provider can stage this file by calling can_stage_in, and if so, it will call both stage_in and replace_task to give the provider the opportunity to perform staging.

For each file to be staged out, the data manager will follow the same pattern using the corresponding stage out methods of this class.

The default implementation of this class rejects all files, and performs no staging actions.

To implement a concrete provider, one or both of the can_stage_* methods should be overridden to match the appropriate files, and then the corresponding stage_* and/or replace_task* methods should be implemented.

__init__()[source]

Methods

__init__()

can_stage_in(file)

Given a File object, decide if this staging provider can stage the file.

can_stage_out(file)

Like can_stage_in, but for staging out.

replace_task(dm, executor, file, func)

For a file to be staged in, optionally return a replacement app function, which usually should be the original app function wrapped in staging code.

replace_task_stage_out(dm, executor, file, func)

For a file to be staged out, optionally return a replacement app function, which usually should be the original app function wrapped in staging code.

stage_in(dm, executor, file, parent_fut)

This call gives the staging provider an opportunity to prepare for stage-in and to launch arbitrary tasks which must complete as part of stage-in.

stage_out(dm, executor, file, app_fu)

This call gives the staging provider an opportunity to prepare for stage-out and to launch arbitrary tasks which must complete as part of stage-out.

can_stage_in(file: File) bool[source]

Given a File object, decide if this staging provider can stage the file. Usually this is be based on the URL scheme, but does not have to be. If this returns True, then other methods of this Staging object will be called to perform the staging.

can_stage_out(file: File) bool[source]

Like can_stage_in, but for staging out.

replace_task(dm: DataManager, executor: str, file: File, func: Callable) Callable | None[source]

For a file to be staged in, optionally return a replacement app function, which usually should be the original app function wrapped in staging code.

replace_task_stage_out(dm: DataManager, executor: str, file: File, func: Callable) Callable | None[source]

For a file to be staged out, optionally return a replacement app function, which usually should be the original app function wrapped in staging code.

stage_in(dm: DataManager, executor: str, file: File, parent_fut: Future | None) DataFuture | None[source]

This call gives the staging provider an opportunity to prepare for stage-in and to launch arbitrary tasks which must complete as part of stage-in.

This call will be made with a fresh copy of the File that may be modified for the purposes of this particular staging operation, rather than the original application-provided File. This allows staging specific information (primarily localpath) to be set on the File without interfering with other stagings of the same File.

The call can return a:
  • DataFuture: the corresponding task input parameter will be replaced by the DataFuture, and the main task will not run until that DataFuture is complete. The DataFuture result should be the file object as passed in.

  • None: the corresponding task input parameter will be replaced by a suitable automatically generated replacement that container the File fresh copy, or is the fresh copy.

stage_out(dm: DataManager, executor: str, file: File, app_fu: Future) Future | None[source]

This call gives the staging provider an opportunity to prepare for stage-out and to launch arbitrary tasks which must complete as part of stage-out.

Even though it should set up stageout, it will be invoked before the task executes. Any work which needs to happen after the main task should depend on app_fu.

For a given file, either return a Future which completes when stageout is complete, or return None to indicate that no stageout action need be waited for. When that Future completes, parsl will mark the relevant output DataFuture complete.

Note the asymmetry here between stage_in and stage_out: this can return any Future, while stage_in must return a DataFuture.