13 releases

Uses old Rust 2015

0.2.9 Feb 18, 2021
0.2.8 Feb 18, 2021
0.2.7 Sep 16, 2020
0.2.4 Aug 6, 2020
0.1.6 Jul 28, 2020

#2 in #queueing

42 downloads per month

GPL-2.0 license

77KB
2K SLoC

Build Status

Round queue

The rq program implements a very simple queuing system. This system favors convention over configuration, and hence requires zero configuration, but does impose a constraint on how you set up your system. In particular, all users who use rq must have home directories within a single parent directory (this is typically /home, but could be some other directory). Also, rq requires that home directories be shared across any computers involved in the cluster, and mounted at the same point on each computer.

This program is designed to be very easy to configure and use, and should be reliable. However, it is not designed to be efficient when there are large numbers of jobs or users. It is intended for a small computing cluster, where $O(N^2)$ operation is acceptable.

Installing rq

To install rq, you must have rust installed (see https://rustup.rs). Then just type

cargo install roundqueue

Using rq

Each user running rq should run rq daemon on each compute node on which they wish jobs to be able to run. On my cluster, I do this on users' behalf using systemd to start them up. This means that with $N$ users on $M$ nodes, you will have $NM$ daemons running.

To submit a job, you can run something like

rq run --job-name my-compute-job ./compute --flag-for-compute --other-flag

in the directory where you want this command to run. By default, jobs are assumed to run on a single cpu. You can see the running (and waiting to run) jobs by simply executing

rq

You can cancel a job by executing

rq cancel my-compute-job

Finally, you can see how busy the nodes are with

rq nodes

How does it work

The security design of rq involves never communicating directly over the network. Instead, communication is entirely done by creating and reading files in the home directory (which is presumably shared over the network). This simplifies the design of rq, and presumably increases its security, provided your file system is secure.

It is not necessary for all users to have the rq daemon running on all nodes, however, any nodes that a given user does not have the daemon running on will not run that user's jobs.

Each users job information is stored in the $HOME/.roundqueue/ directory. Each job is a separate file in a subdirectory there.

rq run

Submitting a job consists of creating a single JSON file in $HOME/.roundqueue/waiting/. That's all.

Dependencies

~3–12MB
~123K SLoC