parallel graph management and execution in heterogeneous computing

pathos is a framework for heterogeneous computing. It provides a consistent high-level interface for configuring and launching parallel computations across heterogeneous resources. pathos provides configurable launchers for parallel and distributed computing, where each launcher contains the syntactic logic to configure and launch jobs in an execution environment. Examples of launchers that plug into pathos are: a queue-less MPI-based launcher (in pyina), a ssh-based launcher (in pathos), and a multi-process launcher (in multiprocess).

pathos provides a consistent interface for parallel and/or distributed versions of map and apply for each launcher, thus lowering the barrier for users to extend their code to parallel and/or distributed resources. The guiding design principle behind pathos is that map and apply should be drop-in replacements in otherwise serial code, and thus switching to one or more of the pathos launchers is all that is needed to enable code to leverage the selected parallel or distributed computing resource. This not only greatly reduces the time to convert a code to parallel, but it also enables a single code-base to be maintained instead of requiring parallel, serial, and distributed versions of a code. pathos maps can be nested, thus hierarchical heterogeneous computing is possible by merely selecting the desired hierarchy of map and pipe (apply) objects.

The pathos framework is composed of several interoperating packages:

• dill: serialize all of Python
• pox: utilities for filesystem exploration and automated builds
• klepto: persistent caching to memory, disk, or database
• multiprocess: better multiprocessing and multithreading in Python
• ppft: distributed and parallel Python
• pyina: MPI parallel map and cluster scheduling
• pathos: graph management and execution in heterogeneous computing

The pathos package provides a few basic tools to make parallel and distributed computing more accessible to the end user. The goal of pathos is to enable the user to extend their own code to parallel and distributed computing with minimal refactoring.

pathos provides methods for configuring, launching, monitoring, and controlling a service on a remote host. One of the most basic features of pathos is the ability to configure and launch a RPC-based service on a remote host. pathos seeds the remote host with the portpicker script, which allows the remote host to inform the localhost of a port that is available for communication.

Beyond the ability to establish a RPC service, and then post requests, is the ability to launch code in parallel. Unlike parallel computing performed at the node level (typically with MPI), pathos enables the user to launch jobs in parallel across heterogeneous distributed resources. pathos provides distributed map and pipe algorithms, where a mix of local processors and distributed workers can be selected. pathos also provides a very basic automated load balancing service, as well as the ability for the user to directly select the resources.

The high-level pool.map interface, yields a map implementation that hides the RPC internals from the user. With pool.map, the user can launch their code in parallel, and as a distributed service, using standard Python and without writing a line of server or parallel batch code.

RPC servers and communication in general is known to be insecure. However, instead of attempting to make the RPC communication itself secure, pathos provides the ability to automatically wrap any distributes service or communication in a ssh-tunnel. Ssh is a universally trusted method. Using ssh-tunnels, pathos has launched several distributed calculations on national lab clusters, and to date has performed test calculations that utilize node-to-node communication between several national lab clusters and a user's laptop. pathos allows the user to configure and launch at a very atomistic level, through raw access to ssh and scp.

pathos is the core of a Python framework for heterogeneous computing. pathos is in active development, so any user feedback, bug reports, comments, or suggestions are highly appreciated. A list of issues is located at https://github.com/uqfoundation/pathos/issues, with a legacy list maintained at https://uqfoundation.github.io/project/pathos/query.

pathos provides a configurable distributed parallel map interface to launching RPC service calls, with:

• a map interface that meets and extends the Python map standard
• the ability to submit service requests to a selection of servers
• the ability to tunnel server communications with ssh

The pathos core is built on low-level communication to remote hosts using ssh. The interface to ssh, scp, and ssh-tunneled connections can:

• configure and launch remote processes with ssh
• configure and copy file objects with scp
• establish an tear-down a ssh-tunnel

To get up and running quickly, pathos also provides infrastructure to:

• easily establish a ssh-tunneled connection to a RPC server

The latest released version of pathos is available from: https://pypi.org/project/pathos

pathos is distributed under a 3-clause BSD license.

You can get the latest development version with all the shiny new features at: https://github.com/uqfoundation

If you have a new contribution, please submit a pull request.

pathos can be installed with pip::

$ pip install pathos

pathos requires:

• python (or pypy), >=3.8
• setuptools, >=42
• pox, >=0.3.5
• dill, >=0.3.9
• ppft, >=1.7.6.9
• multiprocess, >=0.70.17

Probably the best way to get started is to look at the documentation at https://pathos.rtfd.io. Also see pathos.tests and https://github.com/uqfoundation/pathos/tree/master/examples for a set of scripts that demonstrate the configuration and launching of communications with ssh and scp, and demonstrate the configuration and execution of jobs in a hierarchical parallel workflow. You can run the test suite with python -m pathos.tests. Tunnels and other connections to remote servers can be established with the pathos_connect script (or with python -m pathos). See pathos_connect --help for more information. pathos also provides a portpicker script to select an open port (also available with python -m pathos.portpicker). The source code is generally well documented, so further questions may be resolved by inspecting the code itself. Please feel free to submit a ticket on github, or ask a question on stackoverflow (@Mike McKerns). If you would like to share how you use pathos in your work, please send an email (to mmckerns at uqfoundation dot org).

Important classes and functions are found here:

• pathos.abstract_launcher [the worker pool API definition]
• pathos.pools [all of the pathos worker pools]
• pathos.core [the high-level command interface]
• pathos.hosts [the hostname registry interface]
• pathos.serial.SerialPool [the serial Python worker pool]
• pathos.parallel.ParallelPool [the parallelpython worker pool]
• pathos.multiprocessing.ProcessPool [the multiprocessing worker pool]
• pathos.threading.ThreadPool [the multithreading worker pool]
• pathos.connection.Pipe [the launcher base class]
• pathos.secure.Pipe [the secure launcher base class]
• pathos.secure.Copier [the secure copier base class]
• pathos.secure.Tunnel [the secure tunnel base class]
• pathos.selector.Selector [the selector base class]
• pathos.server.Server [the server base class]
• pathos.profile [profiling in threads and processes]
• pathos.maps [standalone map instances]

pathos also provides two convenience scripts that are used to establish secure distributed connections. These scripts are installed to a directory on the user's $PATH, and thus can be run from anywhere:

• portpicker [get the portnumber of an open port]
• pathos_connect [establish tunnel and/or RPC server]

Typing --help as an argument to any of the above scripts will print out an instructive help message.

If you use pathos to do research that leads to publication, we ask that you acknowledge use of pathos by citing the following in your publication::

M.M. McKerns, L. Strand, T. Sullivan, A. Fang, M.A.G. Aivazis,
"Building a framework for predictive science", Proceedings of
the 10th Python in Science Conference, 2011;
https://arxiv.org/pdf/1202.1056

Michael McKerns and Michael Aivazis,
"pathos: a framework for heterogeneous computing", 2010- ;
https://uqfoundation.github.io/project/pathos

Please see https://uqfoundation.github.io/project/pathos or https://arxiv.org/pdf/1202.1056 for further information.