a Deferred profiler
Project description
A Deferred-aware profiler for Python code.
While cProfile is a very useful utility, it is limited to recording synchronous execution time. A function that returns a Deferred will typically return very quickly, while the Deferred it returns might not fire for seconds or even minutes. This is where theseus comes in: any function that returns a Deferred will be tracked by theseus. The time from when the Deferred was returned to when it fired will be measured, and recorded along with the function’s call stack.
Usage
The public interface of theseus is a class called Tracer. To get started:
from theseus import Tracer t = Tracer() t.install()
This is enough to start tracing execution. Eventually, the statistics will have to be written to disk:
with open('callgrind.theseus', 'wb') as outfile:
t.write_data(outfile)
The output is written in callgrind format, which means that standard tools can be used to interpret the results, such as kcachegrind.
Additionally, theseus is aware of inlineCallbacks, and will rewrite call stacks to make them look “correct”. For example, given this code:
from twisted.internet import defer, task @defer.inlineCallbacks def func(reactor): yield task.deferLater(reactor, 1, lambda: None) task.react(func)
The call stack according to theseus will look like this (most recent call last):
__main__ in <module> twisted.internet.task in react __main__ in func
Because theseus uses a trace hook instead of a profile hook, it’s possible to have both cProfile and theseus enabled at the same time.
