8 ms·
Why Netflix Rolled Its Own Node.js Functions-As-a-Service Runtime
- paulgrant999 8y agoThere are legitimate reasons for serverless; particularly in the startup space. Specifically relating to: * cost containment * uncertain demand * promotion-based hotspots
- christophilus 8y agoTL;DR: It sounds like they wanted their UI teams to be able to easily deploy UI specific API endpoints, and existing offerings had too much latency. I’ve often built UI-specific API endpoints, but have often been told such endpoints are a bad practice. In reality, I think they’re great. It allows you to serve up optimal payloads and move much more quickly than writing the equivalent general APIs. That said, wasn’t Falcor the Netflix solution to this very problem?
- stephengillie 8y agoWhile the article focuses on the Why, the article isn't thick with implementation details about the How. From the image, it looks like they wrote a Node.js program and have it running inside Docker. The article mentions their custom IDE (NeWT - Netflix Workflow Toolkit) that connects to the Docker instance. The lower half of the article covers their development toolchain: - Repo: Github? - CI/CD: Spinnaker - Testing: ??? - Container orchestration: Titus - Dashboards and metrics: Atlas - Alerts and notifications: Pagerduty
- alexandercrohde 8y agoThis is a garbage article. It reads as a commercial and not an informed technical piece. As I understand it, according to this article, netflix is 90% complex services that have nothing to do with node. All this article is describing is adding an additional layer in front of those services, which uses a hosted solution that abstracts away servers. So it sounds like by-and-large netflix still uses traditional solutions. So what's not clear from the article: 1. Why node? 2. Why add a layer at all? I don't see why UI A/B testing couldn't be handled with truly front-end changes. Does this somehow enable A/B tests on TV apps, for example? 3. How does this play with traditional version-control? (e.g. in a post-mortem if you need to see who changed something, and who approved it, is there a written log? And merge conflicts?) 4. Does this add a new single-point-of-failure?
- ianseyer 8y agoInteresting that they show a usecase involving Vizceral (their traffic visualizer, "intuition engineering"), despite them publicly abandoning the project. Hope this implies they are working on releasing it formally! https://github.com/Netflix/vizceral https://github.com/Netflix/vizceral
- nicoburns 8y agoFunction As A Service (FAAS) is so much of a better name than "serverless". I wish everyone would adopt it, and stop confusing everyone.
- Waterluvian 8y agoYes! I've been advocating among my colleagues to perceive "serverless" as "functional programming -> web services" I get that the allure of not focusing on infrastructure, scaling, etc. is strong. But that's really a side effect of what you're actually doing: applying FP principles to web service design. And it really fits the paradigm well. You've got HTTP services (short lived, therefore comfortably stateless) that are ideally infinitely scaleable (easy to do when your HTTP services are basically pure functions). Do you need caching? That's like memoization of your functions!
- scarface74 8y agoNo one who has any technical competence is “confused” by the term “serverless”. No one thinks that thier code is being run by unicorns and fairies. It’s more like people yelling at the wind or thinking they are intellectually superior because they are in on the little secret that servers are involved.
- manigandham 8y agoIt does cause confusion because it is absolutely non-descriptive, especially when the literal definition is completely the opposite of anything it has ever described. Pick a better term.
- scarface74 8y agoConfusion by who? I’m sure the same people who are “confused” by Amazon’s use of “serverless” would also be “confused” by terms like lambda, cloud formation, cloud trail, Redshift, Route 53, Elastic Beanstalk, Aurora, OpsWorks, Mechanical Turk, etc. Just like everything else in technology, if someone is confused by tech jargon, they should research it.
- cheriot 8y agoCan we not debate the term "serverless" on every FaaS and PaaS article? Edit: remove false claim
- invaliduser 8y agoActually it's categorized as "Serverless" on their blog, and it appears 4 times in the article.
- stephengillie 8y agoReally it should be called Serverless as a Service.
- mattrjacobs 8y agoEvery single comment here is about the name “serverless”. Can’t actually think of a better bikeshed example than this.
- EngineerBetter 8y agoI find it easier to live with if you consider it to mean "daemonless". So not about hardware, but what software you don't have in your project.
- rasteau 8y ago"[T]he English language... becomes ugly and inaccurate because our thoughts are foolish, but the slovenliness of our language makes it easier for us to have foolish thoughts."
- bryanrasmussen 8y agoConsidering that all we have in that shed is a bunch of tools for the county workers maybe it shouldn't be called a bikeshed.
- kbenson 8y agoJust coming to this article, I find my self agreeing with you and the gist of the thread about serverless being a horrible descriptor at the same time. It was just one thread though, and I could see how it sucked people into its vortex, I haven't even had a chance to read the article yet, and here I am doing the same as everyone else (including you...). Well, worse actually, as we've contributed to a new thread.