SF Scala: Mark McBride Interview
Recording: SF Scala: Mark McBride Interview
poor body I'm electric rubber of the organizer of excess color here are on location at fauna and look at just for the panel infrastructure after dinner and who was a mark mcbride from turbine hello what is the current battle I'm currently CEO and co-founder of turbine labs and what does our lives so turbine labs is a product that helps you balance the tension between iterating your infrastructure quickly and maintaining a stable customer experience so we built a reverse proxy and a modern management plane to allow you to do a lot of sort of modern release practices testing and production Bluegreen deploys Canaries etc in a way that's seamless and fits into whatever stack you have mm-hmm so it's just the where of several kind of infrastructure start up in the service yes our house of skyliner and lower rent works and I think people want from different angles I want to give you yes sure the whole focus on structure and so as I think that is that developers will just you know pushed up to github rate as a magical happen so what is kind of your specific taken last year so I think skyliner is great we followed their trajectory pretty closely and I think they're really good at getting you from like a raw github repository to something deployed in production as a v1 and I think you know they're they're offering a lot of just ops out-of-the-box best practices honed at stripe and Yammer and other places what we're looking at is really for people who already have it be one out in production when they reach the end of the natural life of that architecture how do they move to what's next and so you know microservices are a big thing now a lot of people are looking to move from what they have to Cooper Nettie's or tomatoes but none of those companies really have a story for how you move from what you have today into that system and so by putting the proxy is an indirection layer between your customers and their experience and your infrastructure you allow yourself more flexibility to switch that up hmm so you know talking over with some of the Skyliner folks there's this kind of get you to be one ours is get you from v1 to v2 interesting I want to come the thing to be compiled I think right now Skyliners opinionated enough it would be difficult and doesn't fit like seamlessly into their into their stack because they're they're using docker as a unit of deployment but I believe they're fairly opinionated with the sort of ancillary stuff they do around ec2 to get a nice stable or is AWS a nice stable operational platform I think in the future there are things we can do that probably work complimentary Lee in a complementary fashion with them but right now like our focus is people who are kind of later in life engineering wives and skyliner is well call is this one form of goondas doing so the buoyant is really focusing on our PC and making those deep service stacks work effectively and a Twitter that was a huge deal you know as I think William in the buoyant literature shows once you have a chain where there's 10 RPC calls failures on that become much much more likely time out to harder to manage deadlines are harder to manage circuit breakers retries that's all that's all really difficult we're focused more at the edge and allowing you to mutate infrastructure and modulate you know the deployment topology of your application in a way that's flexible and is really low risk to customers and so while they're dealing more with kind of this core you know fairly complicated service topology and helping you manage that effectively which is still very important we're focused more at the edge and managing that customer experience more effectively mm-hmm and so this is kind of similar to saying like how how does boy and compared to cloud player like those are very different things at very different parts of the network stack and and I think we fit in between those two things okay so it makes me super curious to see like an example Roger might become common give us a folk I'm sure maybe yeah without you for example one person I wonder right so you you were basically leading the build team a Twitter and obviously it's all very dramatic and so usually usually release cycle is very dramatic tied to the culture company may be kind of all the product eyes best practices right right so I wonder how do you see that you know work in the real world where everybody has opinions has already some kind of huge processor and entrenched kind of habit so the world so I think there's been a lot of work put in to see icd recently if you look at like Jenkins was a fairly novel thing five years ago and now there's a bunch of Jenkins clones and people have pushed more from CI to CD mm-hmm I think the question we asked was what does the D and C D mean that people have deployed or delivery or what have you but even there it's like a really sort of fuzzy notion is that meaning i push an image to docker hub does it mean I actually deploy a thing on to infrastructure does i mean i deploy and it's visible to customers yeah we're really hoping that people sort of separate deploy from release mm-hmm getting something running on compute automate the hell out of that mhm and I think that can be fairly isolated right now things like namespaces and Cooper Nettie's what's harder for people to do or risky for people to do is that step of releasing new code to customers and that's something for you if you release something that's faulty that customers impacted you're going to get called and that's bad yes and so to the extent that you can observe the differences in old code in new code so with our releases you deploy just a trickle of customers to new code you can compare the difference in experience between those two versions of code is this one slower it's a throwing errors more mmm does it have a different behavior than the one that you had deployed before and then if you're separating deploy from release and using routing as your release mechanism if you notice it's bad fixing that becomes a fairly trivial step of just we won't route to that new version anymore mm-hmm so to the extent that you can make that sort of less risky by releasing it to fewer customers quick detection of faults because you can observe old versus new system in a quick reversion if you do notice a fault those releases become less risky and you can get true sort of the dream is I check in to get and it goes to customers if it's good mm-hmm like that's everything else is so it would be fine as a follicle bicycle right everything else is sort of attacks in between and I think we're in pretty good shape there I think strict deployed based approach is going to be sort of problematic to get working right I'm treating that getting it running on computer versus managing the release to customers is a good separation and something we're working on making better right so I I don't like it you know like you know I wish you know like in my previous how much as I could about this so how easy is it to plug into existing companies sure so most people already have something like this in place is just crude right now a little Jenkins was also like some so it's good ones who think is yeah they have a deployment tool they also probably have a load balancer a reverse proxy they're using now so amazon has the OBS they just released a lb which is a more sophisticated version WOB and so we see ourselves as a better back half of the elb mm-hmm so when we go in for a deployed so i still use your lb to be that armor sort of on the edge protecting your infrastructure from scary internet traffic mm-hmm but have this companion that the software based more flexible routing engine that sits right behind it to really direct customer traffic to the right places and if you have that and the right management plane is a point to observe customer experience the point to affect control and it's also a place to keep a record of all the changes made to customer experience and so that's sort of where we slot in it's not that we're adding a whole new piece of infrastructure we're really just adding sort of an extra super power to a piece of infrastructure you already have okay so let's all super exciting you know i can't wait to see you guys present all this one for me tops all right maybe like you know it's kind of a live demo sure and that will do really cool because i think a lot of people can actually use it yeah i hope so cool it was great to have you back in the community okay yeah yeah and we're neighbors you know i'm going to visit you guys know me too nice all right thanks thank you rush