WSGI Profiling middleware
linesman is a much needed profiler-for-WSGI applications. It installs as middleware, can be configured entirely from any paster config, and aims to be a jack-of-all-trades when it comes to profiling WSGI apps.
Since a picture is worth a thousand words, here are a few screenshots of the interface:
Reasoning behind this library
One of my team’s stories at work was to investigate existing Python profiling tools for use with some of our new web stacks (all in Pylons). I looked at a few–repoze.profile, kea.profile, and even dozer (still in 0.2alpha)–but couldn’t find any that suited our use case. We wanted to…
- visualize the flow of our code
- identify bottlenecks quickly and easily
- have the ability to strip out extraneous calls
Many of the tools simply outputted the pstats object from cProfile, which can be difficult to parse, and even more difficult to identify the call order. Considering that cProfile provided all the information needed, I figured it would be just as easy to write our own middleware.
linesman is a name given to people who inspect electrical Pylons, and was a meek attempt at having a relevant library name.
Setting up middleware
Now, you’ll need to tell your WSGI application how to use Linesman. Assuming you’re using Paster, you can do this very easily in your development.ini (or similar) config file. Add a new filter section:
[filter:linesman] use = egg:linesman#profiler
Then, find the section for your specific application. Typically, it will have a section header that looks like [app:main]. Add the following config option somewhere within this section:
filter-with = linesman
Wallah! Once you start your paster server, you’ll be all set. Verify that all is working correctly by accessing pages on your server. This will also create profile entries for the next step.
Accessing the profiles
This will assume that your application is mounted at the root directory, /, and that your server is running on localhost at port 5000. If not, make sure you adjust your URLs accordingly.
Access the URL at http://127.0.0.1:5000/__profiler__, which should present you with a list of profiles and times, with a link to the stats page. If you can see this (and view the profiles), then you’re all set!