Release v2.0.0-b9 (Getting Started)

License PyPI Version Python Versions Documentation Downloads

HyperShell is an elegant, cross-platform, high-performance computing utility for processing shell commands over a distributed, asynchronous queue. It is a highly scalable workflow automation tool for many-task scenarios.

The software is pure Python and has been tested on Linux, macOS, and Windows 10. The server and clients don’t even need to use the same platform.


Simple, Scalable

Take a listing of shell commands and process them in parallel. In this example, we use the -t option to specify a template for the input arguments which are not fully formed shell commands. Larger workloads will want to use a database for managing tasks and scheduling. In this case, we can run this small example with --no-db to disable the database and submit tasks directly to the shared queue.

Hello World

seq 4 | hyper-shell cluster -N2 -t 'echo {}' --no-db

Scale out to remote servers with SSH and even define groups in your configuration file. By default, all command stdout and stderr are joined and written out directly. Capture individual task stdout and stderr with --capture. Set the logging level to INFO to see each task start or DEBUG to see additional detail about what is running, where, and when.

Distributed Cluster over SSH

hyper-shell cluster tasks.in -N4 --ssh-group=xyz --capture
2022-03-14 12:29:19.659 a00.cluster.xyz   INFO [hypershell.client] Running task (5fb74a31-fc38-4535-8b45-c19bc3dbedee)
2022-03-14 12:29:19.665 a01.cluster.xyz   INFO [hypershell.client] Running task (c1d32c32-3e76-48e0-b2c3-9420ea20b41b)
2022-03-14 12:29:19.668 a02.cluster.xyz   INFO [hypershell.client] Running task (4a6e19ec-d325-468f-a55b-03a797eb51d5)
2022-03-14 12:29:19.671 a03.cluster.xyz   INFO [hypershell.client] Running task (09587f55-4b50-4e2b-a528-55c60667b62a)
2022-03-14 12:29:19.674 a04.cluster.xyz   INFO [hypershell.client] Running task (1336f778-c9ab-4111-810e-229d572be62e)


One of several novel features of hyper-shell, however, is the ability to independently stand up the server on one machine and then connect to that server using a client from a different environment.

Start the hyper-shell server and set the bind address to to allow remote connections. The server schedules tasks on a distributed queue. It is recommended that you protect your instance with a private key (-k/--auth).


hyper-shell server -H '' -k '<AUTHKEY>' --print < tasks.in > tasks.failed

Connect to the running server from a different host (even from a different platform, e.g., Windows). You can connect with any number of clients from any number of hosts. The separate client connections will each pull tasks off the queue asynchronously, balancing the load.


hyper-shell client -H '<HOSTNAME>' -k '<AUTHKEY>'


Special variables are automatically defined for each individual task. For example, TASK_ID gives a unique UUID for each task (regardless of which client executes the task).

Further, any environment variable defined with the HYPERSHELL_EXPORT_ prefix will be injected into the environment of each task, sans prefix.

Use -t (short for --template) to expand a template, {} can be used to insert the incoming task arguments (alternatively, use TASK_ARGS). Be sure to use single quotes to delay the variable expansion. Many meta-patterns are supported (see full overview of templates):

  • File operations (e.g., the basename '{/}')

  • Slicing on whitespace (e.g., first '{[0]}', first three '{[:3]}', every other '{[::2]}')

  • Sub-commands (e.g., '{% dirname @ %}')

  • Lambda expressions in x (e.g., '{= x + 1 =}')


hyper-shell cluster tasks.in -N12 -t './some_program.py {} >outputs/{/-}.out'

Capturing stdout and stderr is supported directly in fact with the --capture option. See the full documentation for environment variables under configuration.