Wrote a (rough) clone of this design internally based off a slide deck[1] Netflix put out a while back. Kind of (gladly) surprised they released their tools.
I really like the approach of bringing up a second set of systems during an upgrade, so rollback is easy: makes the most of virtual machines. I'll have to add that to PlatformLayer!
Asgard looks to be very heavily tied to AWS though, which I guess explains why NetFlix are always pleading with everyone else to stick with the AWS APIs. With the cloud wars getting really interesting (Google's cloud this week, we hope!), locking yourself in to the obscure features of Amazon's cloud seems like a mis-step I wouldn't want to emulate.
Speaking as a company providing cloud streaming since before the term cloud was popular (2001, and our name signifies a rolling cloud/fog bank), the so-called "obscure features" tend to be a sign of customer maturity.
While "minimum viable product" applies to getting off the ground, as your customers grow, eventually none of them will need only 80% type features. They will need that last 20% of features that handle edge cases, unique capabilities not offered elsewhere, and the sorts of things one can do in one's own IT department but cannot usually do outsourced.
AWS "obscure" features, like ours, tend to be these capabilities: those things that leading customers (the category driving quarter to quarter organic growth independent of sales/marketing overhead) need to complete their own offerings.
If you want those customers, and to help your overhead you do, then you will eventually have to develop and support those obscure features that through your customer communication processes with your top 5% you've found keep you competitive with "in house" IT.
Interesting. Not sure how it compares to Wealthfront's continuous deployment[1] one though (which also includes automated rollback based on business metrics).
