This site is now powered by Django instead of Plone. Yes, I’ve finally made the switch!
The migration also means my RSS feeds have changed. For more information, look at the end of this entry.
I’ve been planning this migration since late 2009, but initially I lost myself in the details and had other things on my mind. But now, I’ve finally finished it.
To be honest, it wasn’t even complicated. I created a few apps for the blog, a couple of templates and some view logic. And by adding the Django packages (including my project and apps) to the Python path, I could write a quick and dirty browser view which iterated over all Plone weblog entries and saved them as Django weblog entries.
There are a couple of reasons for switching to Django. The first is quite simple: it’s fun! Writing the blog engine can even be more rewarding than writing the weblog entries themselves: tinkering with the code to get it just the way I want it, putting the pieces together, building the application step-by-step… I just love it.
Another reason is a bit more pragmatic. Plone is a complete content management system with all the bells and whistles. But with my weblog I was only using a fraction of it. I don’t need events, news items, groups, roles, et cetera. Sure, the Plone search functionality is great, but all I need is ‘good enough.’
What did I gain?
I personally think the Django site requires less resources and is faster. But I am biased and might just see what I want to see. Time to gather data…
Let’s start with the resources:
|Files (under source control)||160||87|
|Lines of code (under source control)||4336||3390|
|Disk space (MB, complete running site)||383||67|
Measuring the speed is less straightforward. Although I’ve got the feeling that the site is more responsive, I used ApacheBench to get some numbers. But before going to the results, a quick disclaimer. I do not pretend this is representative at all. I do not claim this test is in any way scientific. It’s merely a way to get some objective data.
The setup of the Plone site: Plone with CacheSetup installed (and enabled) behind Varnish behind Apache. The Django site is deployed with Apache and mod_wsgi and I haven’t setup any caching yet.
The command line for the test:
ab -n 100 -c <x> http://vlent.nl/weblog/
I used 1, 2, 4, 8, 16 and 32 for the concurrency parameter. I tested Plone both directly (by adding the ZEO client port to the URL, so even bypassing Apache) and via the normal Apache/Varnish/Plone route.
As you can see in the graph above, Plone behind Apache and Varnish is quite consistent. The fastest response takes 850 to 1050 milliseconds with 32 concurrent requests. Plone also has a minimum of 850 but climbs up to well over 4000 ms. Django sits in the middle with a range from 510 ms to 2400 in the last test.
The second number I looked at was the median request time. This means that 50% of the requests were served within that time. With a concurrency of 1, 2 and 4 Varnish doesn’t seem to make that much of a difference. Django seems to perform pretty nice compared to Plone, even at a concurrency of 16 (and again: I did not setup any form of caching here).
The maximum time requests took actually surprised me. Especially since the Apache/Varnish/Plone route seems to perform worse than just Plone when the concurrency increases. Django seems to handle requests pretty consistently: the maximum request time is about 1.3 times the median.
So although the Apache/mod_wsgi/Django combination is not necessarily the fastest, it does behave rather predictable: twice the concurrency means that the minimum, median and maximal time a request takes also is about two times as large.
(The graphs also might indicate that I’ve messed up my Plone (or Varnish) configuration somewhere. However, since I’ve moved to Django I don’t want to investigate that any further now…)
To make a long story short: I’m happy with the end result and we’ll see how long I’ll stick with Django before moving to the next (new) framework. ;-)
One last thing: by switching to Django, I also could no longer benefit from the feeds generated by one of the Plone products I used. I now had to create the feeds myself and thus had to make my own decisions. After reading RSS and Atom: Understanding and Implementing Content Feeds and Syndication by Heinz Wittenbrink I decided I liked the Atom format better and choose to only implement Atom feeds and drop support for RSS.
As a result, the weblog now has a single feed:
only get items on a specific topic, e.g django, subscribe to