pro.m0te

It was super late at night. We're in this dark little office down a hall from Mission Control. It's not really clear why we all have the lights of, but that's the vibe, cool. We've deployed the live-streaming arch on AWS, load tested with Soasta, and can handle some 7x the traffic expected during the launch... whew.

I'm looking at one of the event manager's screen and there's this little border around the video, with a note about "the livestream will start shortly" crawling along the bottom like some myspace effect. I ask "hey, what serves that border, what web page is that?" the event manager responds "oh, it's a wordpress site we embed the video on, it's all good it's on AWS too"

Well, that's not good. See, even if the video stream can handle the load, what if the wordpress website can't? I rush over to the other table, and yes, there's CloudFront, and yes it's on a huge (for the time) 16-node m2.4xlarge cluster. I ask, "where's the database backend?" the site operations lead: "oh it's all good, it's on RDS, on AWS, that means it autoscales, right?"

Well, that's not good. The AWS console shows that the db is running a single m1.small, and now 16, 8-core machines are gonna come stampeding to get an update to that little crawly text. We stand up what may be the only time a single core RDS instance has had 3, m2.4xl instances as read replicas in history, in the span of about 3 minutes of swearing and typing, and well, it didn't crash. Got the sticker!

Look at the dude down inside the guts of the rover... Goals.