law 0.0.36

Build large-scale task workflows using luigi, remote job submission, remote targets, and environment

Note: This project is currently under development. Version 0.1.0 will be the first, fully documented beta release, targetted for early 2020.

Use law to build complex and large-scale task workflows. It is build on top of luigi and adds abstractions for run locations, storage locations and software environments. Law strictly disentangles these building blocks and ensures they remain interchangeable and resource-opportunistic.

Key features:

• CLI with auto-completion and interactive status and dependency inspection.

• Remote targets with automatic retries and local caching

  • WebDAV, HTTP, Dropbox, SFTP, all WLCG protocols (srm, xrootd, rfio, dcap, gsiftp, …)

• Automatic submission to batch systems from within tasks

  • HTCondor, LSF, gLite, ARC

• Environment sandboxing, configurable on task level

  • Docker, Singularity, Sub-Shells

First steps

Installation and Dependencies

Install via pip:

pip install law

This command also installs luigi and six.

Remote targets also require gfal2 and gfal2-python (optional, also via pip) to be installed.

Docker Images

To run and test law, there are three docker images available on the DockerHub, corresponding to Python versions 2.7, 3.7 and 3.8. They are based on CentOS 7 and ship with the dependencies listed above, including gfal2.

docker run -ti riga/law:latest

Tags:

• latest, py3, py38: Python 3.8.1

• py37: Python 3.7.6

• py2, py27: Python 2.7.5

Usage at CERN

Overcomplete example config

Examples

• loremipsum: The hello world of law.

• workflows: Law workflows.

• dropbox_targets: Working with targets that are stored on Dropbox.

• wlcg_targets: Working with targets that are stored on WLCG storage elements (dCache, EOS, …). TODO.

• htcondor_at_vispa: HTCondor workflows at the VISPA service.

• htcondor_at_cern: HTCondor workflows at the CERN batch infrastructure.

• grid_at_cern: Workflows that run jobs and store data on the WLCG.

• lsf_at_cern: LSF workflows at the CERN batch infrastructure.

• docker_sandboxes: Environment sandboxing using Docker. TODO.

• singularity_sandboxes: Environment sandboxing using Singularity. TODO.

• subshell_sandboxes: Environment sandboxing using Subshells. TODO.

• parallel_optimization: Parallel optimization using scikit optimize.

• notifications: Demonstration of slack and telegram task status notifications..

• CMS Single Top Analysis: Simple physics analysis using law.

Auto completion on the command-line

bash

source "$( law completion )"

zsh

zsh is able to load and evaluate bash completion scripts via bashcompinit. In order for bashcompinit to work, you should run compinstall to enable completion scripts:

autoload -Uz compinstall && compinstall

After following the instructions, these lines should be present in your ~/.zshrc:

# The following lines were added by compinstall
zstyle :compinstall filename '~/.zshrc'

autoload -Uz compinit
compinit
# End of lines added by compinstall

If this is the case, just source the law completion script (which internally enables bashcompinit) and you’re good to go:

source "$( law completion )"

Development

• Source hosted at GitHub

• Report issues, questions, feature requests on GitHub Issues