This week at work has been pretty crazy. On Monday I coded up a tiny change to a tiny little thing that over 4 million people use. This change made the tiny little thing a little more dynamic, which required a little more processing power on our end. Multiply that by four million and you get this:
The kicker is that this tiny little change was only showing up for half the users (an A/B Test, in industry parlance).
Monday very quickly became time to scramle. I had to quickly revert the change and figure out a way to make the same change but oodles faster. Staying at work until 10:30pm paid off this Monday because the rest of that graph includes the same change for all users. Note how on the second time I didn't double our bandwith intake.