PyMargo (Margo for Python)
PyMargo is a python binding for Margo, with an object-oriented interface similar to that of thallium.
This section will walk you through a series of tutorials on how to use PyMargo. We highly recommend that you read the Margo tutorials first.
Important
While working with Mochi in Python can be convenient, please keep in mind that Python has a major limitation: its global interpreter lock (GIL). The GIL ensures that only one thread at a time executes Python code, which drastically limits parallelization opportunities in Mochi. For instance, no more than one RPC written in Python will be executed at any time, even if you specified multiple execution streams in your Margo configuration.
More importantly, the GIL is not Argobots-aware, which means that a ULT blocking on acquiring it will block the ES it runs on, preventing other ULTs from running. This can cause deadlocks in certain situations, in particular if it prevents the Mercury progress loop from running.
All in all, we recommend that you restrict your use of PyMargo to the following scenarios.
You want to write a Python client interface for a Mochi service written in C/C++ (in this case we recommend that you look at how Yokan, Warabi, or Flock do this, for instance);
You want to write a Mochi service based on a Python library that does not have a C or C++ equivalent, and you are ready to work with Python’s constraints.
Note
The Margo and Thallium microservice templates include examples of Python bindings too!
PyMargo can be installed using spack, just like any other Mochi library,
using spack install py-mochi-margo.
Note
PyMargo’s namespace has been updated to mochi.margo. If you have existing code
using import pymargo, you should update it to from mochi.margo import ...
or import mochi.margo as pymargo for backward compatibility.