• You have built one or more successful application
• You need to scale it to reach a larger audience

Performance Testing

• Debug Toolbar
• Debug Logging
• Continuous testing

Load Testing

• Jmeter
• Command-line tools

• How to optimize django apps

  • Optimizing queries
  • Caching
  • etc...

• How scale your infrastructure

Where are we coming from?

• 2008 - Building reusable apps
• 2009, 2010 - Testing and continous integration

Where are we (hopefully) going?

• Larger scale
• Better performance

We built few utilities on top of django-cms to:

• Create a dummy web site on demand
• Get the list of urls for the django-cms pages

"You can't manage what you don't measure"

What are we looking for?

• Time to load a page
• Number of queries per page
• Time taken by the queries
• Number of cache hits/misses
• Time spent waiting for external services

Pros

• No setup.
• Closest to the user experience.

Cons

• It doesn't measure the internals of your application.
• It's a simple good/not good evaluation, with no clues about why.

Pros

• Provides a variety of information about what is going on inside.

Cons

• Requires a third party application and some configuration.
• It's on demand.
• No history.

Pros

• Automated test covering a set of URLs
• Optimised for history tracking
• Repeatable process
• Provides aggregated data

Cons

• Young project
• Requires another third party application and more configuration.

At this stage it is a proof-of-concept. There are few outstanding issues and pending features.

• Better integration with django-debug-toolbar
• Support configurable log handlers, instead of just the database handler
• Optionally use unit tests instead of just url lists

Why?

• Helps narrow down the origin of performance problems
• Performance issues can't always be easily predicted
• Identify issues before they impact production

How?

• Manually run performance test throughout the day
• Run periodically with cron
• Integrate with a tool like Jenkins

• Taking your entire application infrastructure into account.
• Understanding how your performance changes under stress and over time.
• Creating realistic test plans that imitate real life traffic.

Database

• Handling numerous queries in parallel
• Types of queries running at the same time

Caching

• Cache key expiration
• Cache size and key eviction

Servers

• Memory utilization and swapping
• CPU load averages
• Balancing load effectively between database servers, app servers, utility servers, etc.

• Use a GUI to plan out a load test with a huge variety of highly configurable components.
• Set up samplers for each view your users will access.
• Create text files with sample data that can be read by each sampler.
• Use post-processors to retrieve data from your responses, such as CSRF keys.
• Authenticate a Django user by taking the CSRF key from a GET request, then POSTing it along with your credentials.
• Use throughput controllers to define the percentage of traffic on each request type.

jmeter -n -p user.properties -t windmill_plan.jmx -l results.jtl

• htop - An amplified version of top to review load averages, memory utilization, swapping, etc.
• dstat - Shows a variety of information about cpu, memory, disk, and network usage.
• pg_top, mtop - Displays databases statistics and the process list, allowing you to view details about individual queries.
• memcache-top - Retrieves statistics from multiple memcached instances, such as space used/free, hit percentage, evictions.

• Firebug
• Django Debug Toolbar
• Django Debug Logging
• Funkload
• Bees with Machine Guns
• Webunit
• Grinder
• Siege
• Django Performance Patterns

• Building Scalable Webapps
• Switching Mozilla Addons to Django
• Scaling Django Web Applications
• Django Performance Tips PostgreSQL Performance Tuning
• Queryset Transform
• Johnny Cache
• Cachebot
• Cache Machine

• User Manual
• Performance Testing with Jmeter
• Jenkins Performance Plugin