Tech Lounge
00:10:44 · Chris Chinchilla, Will Stewart
[00:10:44] S0: there were any I haven't had any massive experience with it, I wonder if there were any compatibility issues. But I'm guessing if there are, it's probably showing it's a bad sign in some cases. [00:10:55] S1: So you can only get virtualization on AWS with the metal nodes, so they're very expensive. So for example, if you wanted to add you would then have to spend $21 just having three nodes. Not many people are going to want to do that for that. So gVisor is very credible for that. And then on GCP, GCP only supports nested virtualization on Intel. And they've also got some fantastic AMD nodes, you have to run gVisor on AMD for that reason. So Azure supports nested virtualization on all CPU types. This is like the matrix, the feature matrix of secure run time, performance, price, availability. That's something that we do for our customers. We sort of provide them the best secure performance. Again, exactly a high level of abstraction. And then people who deploy on their own cloud account get the choice if they want that. [00:11:45] S0: And some of the other opinions you kind of make, I see things like Istio Linkerd for service meshes. Is that kind of your preferences? Or is it a force dependency? Or can people bring their own as well?
00:11:45 · Chris Chinchilla, Will Stewart
[00:11:45] S0: And some of the other opinions you kind of make, I see things like Istio Linkerd for service meshes. Is that kind of your preferences? Or is it a force dependency? Or can people bring their own as well? [00:11:59] S1: So we have a golden path at Northflank. The old golden path production is same defaults. Northflank has made same default decisions at every single level of the stack. And we have this pick and mix approach. Use Northflank for everything post commit, or just use us for build, or just use us for deploy, and that goes the service mesh. By default, we expose Istio service mesh. We try and expose as many features as possible in a high level of abstraction. Again, it's consistency and workflow abstraction across all these features. But some of our customers say, well, we don't like the performance of Istio. We don't like Envoy. We happen to love Envoy. It's very memory intensive and very CPU intensive, and we do experience Istio attacks. 30% of our cluster resources are dedicated to mesh, but so be it. You have to live with it until Waypoint and some of the ambient mesh. These technologies that have sidecar less when they go to production, I think they have recently gone to production, it becomes a serious option. But we have to offer Linkerd for some customers because they need Linkerd. [00:13:01] S0: Okay. Because then the interesting one, you get to the bottom of the stack, it's kind of at the top of the stack, you have the BYO logs and DNS and secrets and things [00:13:14] S1: like that.
00:13:14 · Will Stewart, Chris Chinchilla
[00:13:14] S1: like that. [00:13:15] S0: So is there no golden path on those? Or is it just assumed that, you know, 95% of people probably use the same So options anyway in this [00:13:24] S1: if you just use Northflank by default, you add a card, you get started. You don't need to worry about all those features. And Northflank can scale and be a very successful startup if we do that, get to hundreds of thousands of users just using that model. But then you see at Heroku, there's that sort of capped outcome. So the reason why we're building bring your own logs, bring your own XYZ, think we even put on the marketing site BYOX or something like that as a joke. [00:13:48] S0: It's above a code field, so it's not quite so marketing. [00:13:51] S1: It's like bring... When you need to sort of have more control, you can do that. For example, if you need to have the logging back end on your infrastructure because you don't want personal information going to the Northflank logging back end, we have that feature available to you. If you want to drive the DNS and secrets in your ECR or your AWS Vault, have at it because we appreciate that people are trying to build serious platforms, and they need a serious platform to deliver that to. So it's this pick and mix approach again. But it's not the default. That's for sort of our enterprise customers.
00:13:51 · Will Stewart, Chris Chinchilla
[00:13:51] S1: It's like bring... When you need to sort of have more control, you can do that. For example, if you need to have the logging back end on your infrastructure because you don't want personal information going to the Northflank logging back end, we have that feature available to you. If you want to drive the DNS and secrets in your ECR or your AWS Vault, have at it because we appreciate that people are trying to build serious platforms, and they need a serious platform to deliver that to. So it's this pick and mix approach again. But it's not the default. That's for sort of our enterprise customers. [00:14:27] S0: I appreciate this is probably some of the kind of the magic source that a company like Northflank adds. But how hard is it to make all these different services and different opinions of the implementers of these services from AWS to GCP to open source products to etc, etc, to actually communicate and work together in a seemingly seamless way for your customers? Is it really like that all the time?
00:14:27 · Chris Chinchilla, Will Stewart
[00:14:27] S0: I appreciate this is probably some of the kind of the magic source that a company like Northflank adds. But how hard is it to make all these different services and different opinions of the implementers of these services from AWS to GCP to open source products to etc, etc, to actually communicate and work together in a seemingly seamless way for your customers? Is it really like that all the time? [00:14:54] S1: Obviously a long time and a lot of hard work. Kubernetes has gone a long way to help us here. Lot of the cloud providers have to move to a sort of standard, and that's helped us with networking, storage, and compute. But as you mentioned, these open source tools, certainly hasn't been solved yet. So it's a lot of engineering work and then speaking with customers, understanding use cases, then trying to consistently expose that to them through and again, we make it more complex. We have to do it consistently through UI, API, CLI, GitOps, and templates. And I'm not sure if you've taken a look at our template builder and our preview environments, but that's sort of a step change in developer experience for CICD. It's not just templates for... We would call it infrastructure as data. So people are focused on Terraform and defining static environments, like create this cluster, create this auto scaling group. What we're saying is, let's start defining your workloads, defining the CICD workflows for those repositories, for those databases, and then the infrastructure is then provisioned from that.
00:15:55 · Chris Chinchilla, Will Stewart
[00:15:55] S0: Yeah. I'm not sure if this ties into what you're talking about, but there's a couple of companies emerging, some using AI, some not, that are twisting that on its head. You you actually make the infrastructure from code instead. Polygamy is one of the earlier ones, but there's another one I've spoken to whose name escapes me right now. But you know, instead of having it kind of this separate thing, you can either, you know, explicitly mention it in code, which is similar, or if it's the AI based ones, it kind of infers from the application. Is that anything that doing? [00:16:28] S1: I'd say that I understand the principle that just write your code and we're gonna magically create all the gap projections. I'm not proven [00:16:36] S0: it is. It sounded amazing, but yeah. [00:16:38] S1: I think it's... I'm not sure the... Is it worth the squeeze? The effort to do that? I'm not sure if it's worthwhile. If you can just build high level objections to infrastructure problems, and a Northflank config looks as simple as it does and is as sort of easy to onboard as it is. Obviously I'd be biased saying that, but if it is that simple, maybe we don't need to have that even more insane sort of high level abstraction, which means you lose all control. And I would say that engineers always love more control. And abstraction. [00:17:14] S0: It's just interesting you mentioned that deploying to glucose monitors. So, that what we would consider an edge device? It sounds like something even less than that.