Skip to main content

Boiling kettles, unit tests and data

·3 mins

The Launchpad test suite takes way too long to run: somewhere between three and four hours.  Incidentally, Rob has just started some work that ought to make it run a lot faster, for which I am truly thankful.  Anyway, over the last three (almost four!) years, I’ve watched the test suite run many, many times, waiting for it to finish or at least reveal a failure.

It has been easy to see that not all tests are created equal. Some are very slow, others are very fast. Zope’s terrible layer mechanism means that these are often grouped together: slow with slow, fast with fast. Watching the test suite churn away for the umpteenth time made me wonder, exactly how are the tests distributed over time.

Luckily, thanks to subunit, a bit of Python glue, and OpenOffice it’s really easy to get a picture:


Bar chart of tests finished each minute-long block


That’s a graph of the number of tests that complete within each minute of the Launchpad test suite run.  What it means is that you are more likely to see an actual test result if you watch at a random time during the first half of the test run. It also means that we have a relatively small number of very slow tests.

Sometimes these graphs are more useful at lower granularity, so I made this one:


Bar chart of tests finished each ten-minute-long block


And this one:


Bar chart of tests finished each hour-long block


The raw data is taken from a successful run of one of my recent branches. If you download it and open it up, you’ll see that the subunit output includes a large number of “time:” statements. Each of these indicates that all subunit statements following it should be considered as taking place at the time given in the original statement.

I used a small Python script to generate the CSVs, running it like this:

gzip -dc testtools-experiment-r11753.subunit.gz | python test-distribution.py 60 > /home/jml/Desktop/tests-each-minute.csv

Where “60” is the number of seconds of granularity.

It was about twenty minutes work all up, and subunit made much of it dead simple. subunit has its flaws, but it’s a really good idea.

That said, I can’t help but feel that it should have been less work. Partly, subunit should have a way to convert data to a format more amenable to analysis by third party tools. Mostly though, those third party tools ought to exist and be known to me.

I can easily imagine a tool where I wouldn’t have to run the script each time to get a different level of granularity, but rather I could use a slider in a UI and watch the graph update itself in real time. That would be cool, and relatively easy. How come nobody has done that yet?

Also, although OpenOffice’s graphing thing is fairly nice, why isn’t there a better tool? One that makes it easier to share graphs as images on the web without having to take crummy screenshots.