No Steps to Faster Testing
Posted by Doug Mon, 10 Oct 2005 03:30:48 GMT
I was actually hoping to title this blog post “Five Easy Steps to Faster Testing.” Unfortunately not. Here’s why I’m complaining about testing performance:
$ time rake
blah blah blah blah blah
Started
.................
Finished in 81.306421 seconds.
115 tests, 353 assertions, 0 failures, 0 errors
blah blah blah blah blah blah
Started
.......................
Finished in 85.885546 seconds.
89 tests, 703 assertions, 0 failures, 0 errors
real 3m5.970s
user 1m48.390s
sys 0m8.410s
Ugh. It get’s really, really annoying to have to do that a few times in an evening.
So, I had heard that SQLite was really, really fast. I thought that most of my bottle neck in testing is loading and unloading the database. I’m still using MySQL with old MyISAM files; so not transactions for me. The directions on How to use SQLite in Ruby on Rails seemed so clear, I was sure I could get it working in no time.
My plan was to leave MySQL as my development and production databases, but switch to SQLite for testing. The first problem turned out to be what I’m calling a bug in DarwinPorts portfile for SQLite. It appears that swig must be installed before you install SQLite, but it’s not a dependency. So, install swig via ports, sqlite3 via ports, and sqlite3-ruby via gems. OK, I’m working.
Second problem: rake wants to just dump the structure from development and then load that into testing. But the schema from MySQL isn’t compatible with SQLite. OK, I modify the Rakefile to not do that.
So now I’m running my tests. Keep in mind all my tests pass with MySQL. The first thing I note is that my tests don’t seem to be running that much faster. Then I see a test fail. Then things get really slow; no wait, it’s hung. The thing is I have no idea why my test failed and then another one hung. By this point I had sunk more time into it than I had to spare. I decided to punt.
I don’t mean to trash-talk SQLite. I’m sure it’s a marvelous database and I’m the screwup for not being able to get it to work smoothly. But, I did loose probably four or five hours to this experiment. I had to write something to account for my time.
