A tool to easily orchestrate general computational workflows both locally and on supercomputers.
Maestro Workflow Conductor (maestrowf)
Maestro can be installed via pip:
pip install maestrowf
Getting Started is Quick and Easy
YAML file named
study.yaml and paste the following content into the file:
description: name: hello_world description: A simple 'Hello World' study. study: - name: say-hello description: Say hello to the world! run: cmd: | echo "Hello, World!" > hello_world.txt
PHILOSOPHY: Maestro believes in the principle of a clearly defined process, specified as a list of tasks, that are self-documenting and clear in their intent.
hello_world study is as simple as...
maestro run study.yaml
Creating a Parameter Study is just as Easy
With the addition of the
global.parameters block, and a few simple tweaks to your
study block, the complete specification should look like this:
description: name: hello_planet description: A simple study to say hello to planets (and Pluto) study: - name: say-hello description: Say hello to a planet! run: cmd: | echo "Hello, $(PLANET)!" > hello_$(PLANET).txt global.parameters: PLANET: values: [Mercury, Venus, Earth, Mars, Jupiter, Saturn, Uranus, Neptune, Pluto] label: PLANET.%%
PHILOSOPHY: Maestro believes that a workflow should be easily parameterized with minimal modifications to the core process.
Maestro will automatically expand each parameter into its own isolated workspace, generate a script for each parameter, and automatically monitor execution of each task.
And, running the study is still as simple as:
maestro run study.yaml
Scheduling Made Simple
But wait there's more! If you want to schedule a study, it's just as simple. With some minor modifications, you are able to run on an HPC system.
description: name: hello_planet description: A simple study to say hello to planets (and Pluto) batch: type: slurm queue: pbatch host: quartz bank: science study: - name: say-hello description: Say hello to a planet! run: cmd: | echo "Hello, $(PLANET)!" > hello_$(PLANET).txt nodes: 1 procs: 1 walltime: "00:02:00" global.parameters: PLANET: values: [Mercury, Venus, Earth, Mars, Jupiter, Saturn, Uranus, Neptune, Pluto] label: PLANET.%%
NOTE: This specification is configured to run on LLNL's quartz cluster. Under the
batchheader, you will need to make the necessary changes to schedule onto other HPC resources.
PHILOSOPHY: Maestro believes that how a workflow is defined should be decoupled from how it's run. We achieve this capability by providing a seamless interface to multiple schedulers that allows Maestro to readily port workflows to multiple platforms.
An Example Study using LULESH
What is Maestro?
Maestro is an open-source HPC software tool that defines a YAML-based study specification for defining multistep workflows and automates execution of software flows on HPC resources. The core design tenants of Maestro focus on encouraging clear workflow communication and documentation, while making consistent execution easier to allow users to focus on science. Maestro's study specification helps users think about complex workflows in a step-wise, intent-oriented, manner that encourages modularity and tool reuse. These principles are becoming increasingly important as computational science is continuously more present in scientific fields and has started to require a similar rigor to physical experiment. Maestro is currently in use for multiple projects at Lawrence Livermore National Laboratory and has been used to run existing codes including MFEM, and other simulation codes. It has also been used in other areas including in the training of machine-learned models and more.
Maestro's Foundation and Core Concepts
There are many definitions of workflow, so we try to keep it simple and define the term as follows:
A set of high level tasks to be executed in some order, with or without dependencies on each other.
We have designed Maestro around the core concept of what we call a "study". A study is defined as a set of steps that are executed (a workflow) over a set of parameters. A study in Maestro's context is analogous to an actual tangible scientific experiment, which has a set of clearly defined and repeatable steps which are repeated over multiple specimen.
Maestro's core tenets are defined as follows:
A study should be easily repeatable. Like any well-planned and implemented science experiment, the steps themselves should be executed the exact same way each time a study is run over each set of parameters or over different runs of the study itself.
Studies should be consistently documented and able to be run in a consistent fashion. The removal of variation in the process means less mistakes when executing studies, ease of picking up studies created by others, and uniformity in defining new studies.
Documentation is important in computational studies as much as it is in physical science. The YAML specification defined by Maestro provides a few required key encouraging human-readable documentation. Even further, the specification itself is a documentation of a complete workflow.
Setting up your Python Environment
To get started, we recommend using virtual environments. If you do not have the
virtualenv package installed, take a look at their official documentation to get started.
To create a new virtual environment:
python -m virtualenv maestro_venv source maestro_venv/bin/activate
Getting Started for Contributors
If you plan to develop on Maestro, install the repository directly using:
pip install -r requirements.txt pip install -e .
Once set up, test the environment. The paths should point to a virtual environment folder.
which python which pip
Many thanks go to MaestroWF's contributors.
If you have any questions or to submit feature requests please open a ticket.
MaestroWF is released under an MIT license. For more details see the NOTICE and LICENSE files.
Release history Release notifications | RSS feed
Download the file for your platform. If you're not sure which to choose, learn more about installing packages.
|Filename, size||File type||Python version||Upload date||Hashes|
|Filename, size maestrowf-1.1.8-py2.py3-none-any.whl (126.0 kB)||File type Wheel||Python version py2.py3||Upload date||Hashes View|
|Filename, size maestrowf-1.1.8.tar.gz (79.0 kB)||File type Source||Python version None||Upload date||Hashes View|
Hashes for maestrowf-1.1.8-py2.py3-none-any.whl