cProfile and Profile Setup and Justification

cProfile Description

cProfile, like profile is a deterministic profiler. Unlike the Python-backed Profile, cProfile is backed by the lsprof , and is a C extension. This profiler is the one that comes recommended for most use-cases as it has reasonable overhead and works really well for some long running programs.

With this in mind cProfile is quite efficient and is able to run quickly allowing for quick and easy profiling of plugin code. However, it may not be available in all distributions as it requires platform-specific compiled c binaries.

The official Python Profiling Documentation can be found here.

Getting Started with cProfile

To get started using the cProfile profiler you will need to import the profile into a new or existing plugin.

import cProfile

To enable this profiler you can add the following code to your plugin to profile against any performance or other issues that may be suspected:

# This is with using the `cProfile` profiler to test the version plugin.

import cProfile
import pstats
from pstats import SortKey
from aac.plugins.version.version_impl import version

return_value = None
with cProfile.Profile() as pr:
    return_value = version()
ps = pstats.Stats(pr).sort_stats(SortKey.CUMULATIVE) # Keep this outside the context manager scope to prevent pollution in the profiler.
ps.print_stats(10) # Print the top 10 calls
return return_value

When you don’t sort your results you can get a lot of noise from some functions and methods that are being called which can add to the confusion of what the plugin is doing.

Profile Description

profile is another deterministic profiler. There are a lot of similarities to the cProfile however, it has a considerable overhead when compared to cProfile and it is written in Python. Therefore, if you need to extend the profiler, you can accomplish that with this package more easily.

It is recommended to use cProfile for the majority of use-cases, and to use profile for extensibility and when there is less consideration for overhead or performance.

Getting Started with Profile

To get started using the profile package you will need to import this into the plugin that is being developed, new or existing.

import profile

To enable the profile package is a little more complicated than the cProfile conterpart. profile does not support using a Context Manager so the wrapping around the activity will be a little more work:

# Profile an AaC command within the plugin code
profile.run(command, filename=None, sort=- 1)

# This command is similar to run() except that it accepts globals and locals definitions that are supplied and passed # through the command being executed.
profile.runctx(command, globals, locals, filename=None, sort=- 1)

# The Profile() class. This is only needed for more precise control over the profiling being done
class profile.Profile(timer=None, timeunit=0.0, subcalls=True, builtins=True)