This package provides a WSGI middleware component which aggregates profiling data across all requests to the WSGI application. It provides a web GUI for viewing profiling data.
Install using setuptools, e.g. (within a virtualenv):
$ easy_install repoze.profile
Configuration via Python¶
Wire up the middleware in your application:
from repoze.profile import ProfileMiddleware middleware = ProfileMiddleware( app, log_filename='/foo/bar.log', cachegrind_filename='/foo/cachegrind.out.bar', discard_first_request=True, flush_at_shutdown=True, path='/__profile__', unwind=False, )
The configuration options are as follows:
- ``log_filename`` is the name of the file to which the accumulated profiler statistics are logged. - ``cachegrind_filename`` is the optional name of the file to which the accumulated profiler statistics are logged in the KCachegrind format. - If ``discard_first_request`` to true (the default), then the middleware discards the statistics for the first request: the rationale is that there are a bunch of lazy / "first time" initializations which distort measurement of the application's normal performance. - If ``flush_at_shutdown`` is true (the default), profiling data will be deleted when the middleware instance disappears (via its __del__). If it's false, profiling data will not be deleted. - ``path`` is the URL path to the profiler UI. It defaults to ``/__profile__``. - ``unwind`` is a configuration flag which indicates whether the app_iter returned by the downstream application should unwound and its results read into memory. Setting this to true is useful for applications which use generators or other iterables to do "real work" that you'd like to profile, at the expense of consuming a lot of memory if you hit a URL which returns a lot of data. It defaults to false.
Configuration via Paste¶
Wire the middleware into a pipeline in your Paste configuration, for example:
[filter:profile] use = egg:repoze.profile log_filename = myapp.profile cachegrind_filename = cachegrind.out.myapp discard_first_request = true path = /__profile__ flush_at_shutdown = true unwind = false ... [pipeline:main] pipeline = egg:Paste#cgitb egg:Paste#httpexceptions profile myapp
Viewing the Profile Statistics¶
As you exercise your application, the profiler collects statistics about the functions or methods which are called, including timings. Please see the Python profilers documentation for an explanation of the data which the profiler gathers.
Once you have some profiling data, you can visit the configured
in your browser to see a user interface displaying profiling statistics
Profiling individual functions¶
Sometimes it might be needed to profile a specific function, be it for analyzing a bottleneck found with the full profiling, or to compare different approaches to the same problem. This package provides a decorator for this case. To use it, simply decorate the desired function like this:
.. code-block:: python
from repoze.profile.decorator import profile
@profile(‘Descriptive title’, sort_columns=(‘time’, ‘cumtime’), lines=30) my_bottleneck()
# some really time consuming code ...
The results of the profiling will be sent to standard out. The
appear at the top of the results, for guidance. All other arguments are
sort_columns allows specifying the columns to sort the timing
results. See the Python profilers documentation for available options.
is the number of lines of results to print. Default is 20. Zero means no limit.