ajmazurie/spate

Lightweight Python API for workflow composition and generation

Spate workflow composition toolkit

Spate is a lightweight Python-based file processing workflow composition and visualization toolkit. It offers an API to create and export workflows to various popular execution engines, as well as shells and job schedulers.

The purpose of Spate is to offer a simple way to design a dataflow instance (i.e., file-based data processing steps, and how they depend on each others) and export it to one or more execution resources (i.e., hardware and software environments) where it will run. Add jobs, remove jobs, merge workflows; Spate will take care of the details.

Spate can export the whole workflow, or only the steps that need to be executed. Spate checks input and output files for modification time, and iteratively identify all steps that need to be run or updated. No more worries about re-running a whole time-consuming workflow when you update few files; Spate will ensure that only the affected steps are executed.

Example

This workflow will take two input files, file_A and file_B, and process them to create an output file file_C by running the Unix cat command:

import spate

# create a new empty workflow
workflow = spate.new_workflow("my_workflow")

# add a simple job; here a Unix shell
# command to merge two files into one
workflow.add_job(
	inputs = ("file_A", "file_B"),  # two input files
	outputs = "file_C",  # one output file
	content = """
	   cat {{INPUT0}} {{INPUT1}} > {{OUTPUT}}
	   """)

# export this workflow as a shell script; to run
# it, type `./my_workflow.sh` the the shell
spate.to_shell_script(workflow, "my_workflow.sh")

# export this workflow as a SLURM sbatch script; to run it,
# type `sbatch my_workflow.slurm` on a SLURM-enabled cluster
spate.to_slurm(workflow, "my_workflow.slurm")

see doc/ for additional examples

Installation

Spate is hosted both on PyPI and GitHub, thus offering two options to install it:

• If using PyPI (recommended to get the latest release) and the pip package manager, type pip install spate on the command line.
• If using GitHub (in case you want one of the development version) you can download the archive of one of the releases or commits and run pip install <archive>.

Supported execution environments

Environment	Status	Comment
Unix shell scripts	Stable	No parallel job execution; jobs are processed sequencially.
Make files	Stable
Makeflow scripts	Experimental	Not fully tested.
Drake scripts	Experimental	Not fully tested.
SLURM job scheduler	Stable	Produces sbatch scripts with directed acyclic graph of jobs. Tested with SLURM 14.11.11.
TORQUE/PBS job scheduler	Experimental	Produces job arrays; does not handle job dependencies. Currently untested.