text stringlengths 0 2.67k |
|---|
\[19:49\] So ultimately, it really comes down to being incredibly close to your customers who are scaling on the platform, and really understanding their needs. And AWS scaled in the early days, and built a lot of the stuff because Netflix wanted that from them. And Netflix was responsible for a lot of AWS innovation, ... |
**Autumn Nash:** Do you offer any kind of classes documentation? How do you teach developers to be able to innovate on your platform successfully? How do you help them to -- because you're saying you need to learn Kubernetes, you need to learn all these different things, and what you're doing is supposed to make it eas... |
**Anurag Goel:** You mentioned documentation just now... Obviously, there's a lot of documentation. But there's also a lot of in-product guidance; there's a lot of information architecture work, there's a lot of design, just product/UX design that goes into making it obvious to people how to build the kind of app that ... |
One of the things that I pride ourselves on is that our customers are able to grow on the platform without really getting in touch with us. And we have a support team, and they're really great, and we offer free email and chat support. At the same time, when you have over a million people on your platform, you can't re... |
Or there are things we do like, well, if something breaks in the logs, and there is something that Render can inject into your log messages to tell you exactly what to do, that is what it will do. |
Again, as an example, Render automatically detects which ports are open on your service. So if it's like a private service that is opening up tons of ports that other people can connect to, then Render detects all those ports and explains "These ports are open." But if it's a web service and you wanted to serve HTTP tr... |
**Autumn Nash:** Good logs, good documents and good error messages are key. |
**Anurag Goel:** Yes. Good error messages are so, so incredibly important, and so often overlooked... |
**Autumn Nash:** Underrated. |
**Justin Garrison:** What's at the bottom of Render? Is this Kubernetes on top of AWS? Is it some other provider? Is it just a bunch of VMs that scale an autoscaling group? Where did you start when you were building this? |
**Anurag Goel:** I started on Kubernetes, because I think Kubernetes does a really good job exposing the kind of abstractions that make complex architectures possible. And there was no need for us to reinvent that wheel. And Kubernetes does a good job with container orchestration, and there's so much open source activi... |
**Justin Garrison:** \[24:17\] But, I mean, that's the goal of a platform team at pretty much any company, right? Platform teams are that. And in this case, Render is that platform team. It's that 20% of engineers that manages infrastructure, built on top of something, and you shouldn't have to worry about what's under... |
**Anurag Goel:** No, we realized that -- obviously, there's a lot of stuff within Kubernetes that you can customize. And for Render, like as an infra company, we've taken out certain parts of Kubernetes, replaced them with our own... But at the end of the day, we're running our own fully self-managed clusters across th... |
But 80% of the companies out there that are building software don't actually need that level of control. So Render is not great if you're building an infra company and infrastructure as a service. You should not use Render, you should use Kubernetes. But if you're building sort of a b2b eCommerce SaaS app, then you rea... |
**Autumn Nash:** What was your biggest challenge in building Render and helping your customers to use it? Also, if you are building something on top of open source software, do you contribute back to that open source software? |
**Anurag Goel:** Personally, I have to reinvent my job every few months as CEO, as the company scales. So the things that I did when the company was five people are very different from the things I did when the company was 20, or 50. Or now we're over 70 people. And you have to really think about, as the CEO, what the ... |
So I think every successful founder, I would say, goes through that journey, and has to be -- if they remain CEO. Obviously, as a founder you can choose to not be CEO, and hire someone who has done this before. I chose not to. I chose to take on this more challenging role, as opposed to just like becoming CTO or stayin... |
What was your hardest engineering problem? Because you said that you first \[unintelligible 00:27:22.08\] So what was your hardest from engineering, and do you ever miss just doing the engineering part, I guess? |
**Anurag Goel:** Yeah... I think in the past, one of the biggest things we've had to deal -- so first of all, scale makes everything hard. And as you grow, you find new bottlenecks. One of the things we ran into very early on - I'm talking like 2018, 2019 - is as we became a bit more popular, we started getting hit by ... |
\[28:12\] We realized that AWS, GCP and CloudFlare, they've dealt with these problems before. And they're really good at dealing with low-level DDoS attacks. And instead of building all of that functionality out ourselves, we decided to change our architecture significantly, and focus much more on bringing Cloudflare i... |
**Justin Garrison:** I'm a little confused there, because you're using, say, Cloudflare to handle the DDoS. How does that relate to bare metal? Because you could put Cloudflare in front of your data center... I don't know, that was confusing. |
**Anurag Goel:** Oh, that was for L7. But I'm talking about L4 as well. So if you're doing bare metal, most bare metal providers, including the best ones like Equinix Metal, they don't actually have functionality that protects L4 level attacks. They will ask you to install some sort of appliance. Or they'll have like s... |
The other big challenge was with GKE. arly on, we started using GKE, and we quickly realized that we were actually one -- at least in US West 1, for GKE we were their largest cluster by certain axes. And that led to a ton of bugs that we found in GKE. And we were actually -- so this is a an interesting story... Okay, I... |
**Justin Garrison:** Yeah. |
**Anurag Goel:** So they did that, and again, obviously, it went down. And then we finally, finally managed to get a hold of someone who realized what the problem was, which was we had upgraded the control plane, and the new version had a bug that was spewing logs onto the disk for the control plane nodes themselves...... |
And that's on the sort of evening, even during -- we had some warning of like when Render \[unintelligible 00:31:28.28\] and a lot of people are looking at us and trying to sign up... And it was just a really terrible experience, because I'm sure we lost a bunch of people who couldn't spin up services because our contr... |
And you asked about contributing back... I think it depends on what the changes are. We've certainly contributed certain things, but most of the customizations we're making are kind of writing our own code to match our own scheduling needs, or using sort of our own operator pattern to do what we want to do to manage da... |
**Autumn Nash:** \[32:23\] Have you actually made any contributions to open source? |
**Anurag Goel:** I will have to go back and check with the engineers, but actually just this morning someone was asking about some Tilt and Bazel development configurations that they're working on, and they're thinking of contributing it, and they're like "Well, where do I go to contribute it?" So we definitely want to... |
**Justin Garrison:** In the vein of giving back, you are basically the platform team that a lot of people are trying to build internally at their companies. They have a lot of people that at least 20% are trying to build these platforms today, internal development platforms, whatever. Can you describe the internal stru... |
**Anurag Goel:** I think the short answer is it continues to evolve. And it depends on what our customers need, how we scale... But at the very high level, obviously namespaces are a good way to isolate things. But also, depending on a customer scale, you can imagine a whole Kubernetes cluster that is kind of dedicated... |
We have to do things in a way that is boring and stable, and we can't really go for the coolest tech that is out there, like Kubernetes in Kubernetes, because we need to make sure that our customers are not affected by whatever bugs that might occur in those newer systems. |
We have talked about some of those things, like Kubernetes in Kubernetes, virtual kubelets, and our engineers - they keep an eye on those things... But at the end of the day, I think the foundation for us is to try to make it so that each customer is completely isolated through a variety of means, but not necessarily a... |
**Justin Garrison:** That's always like one of those concerns, because there's components or resources inside of Kubernetes that aren't multi-tenantable. There's plenty of things that are globally viewed inside of a cluster, and sharing that with multiple people - you have to be very strict on how you isolate those thi... |
**Anurag Goel:** \[36:03\] Yeah, absolutely. And like I said, I think this continues to evolve, and we've had -- I mean, obviously, Docker or just container capabilities are a great way to figure out what people can or cannot do on the platform... And so we rely a lot on that. We rely a lot on other kinds of protection... |
There's also isolation by default, in that the workloads that customers have on Render, they don't run on machines that are directly accessible over the internet. So those machines don't have public IP addresses. Everything is behind NAT gateways. And that allows us to control traffic and isolation a little bit better ... |
There's also this notion of shared load balancing and shared router layers. And early on, very early on, we ran into cases where one customer went viral, and brought our load balancers down... And so we've learned to, over time, build out isolation techniques there, and things like rate limiting we've had to build ours... |
And having said that, I think there's a world in which, for specific customers, we make sure that they have their own VMs, for example, so they don't have to worry about other containers running on the sort of similar infrastructure. |
**Justin Garrison:** Yeah. I mean, there's multiple levels of noisy neighbor that you want to avoid it... And one is just a local resource, CPU, RAM, disk. The other one is networking, is a big one. And so being able to isolate a multi-tenant environment is pretty key there. |
What do you see going forward from here for Render? Because I look at maybe -- I don't know if they're your competitors, but other people in this space, like fly.io has an offering of "Give me a Kubernetes API and let me do whatever I want with that." That's a thing that I can do out there, and it's still an abstractio... |
**Anurag Goel:** \[39:49\] The latter is what we're focused on. So we're not planning to be any kind of managed Kubernetes provider. We could be pretty easily. We run all of this stuff already. But the goal is really - and this is a long-term thing, but also a short-term thing - is to really raise the level of abstract... |
To give you an example, one of the things we might want to do in the future is find ways for our customers to use feature flags and progressive rollouts. And that is the layer at which we want to innovate, as opposed to the "We're gonna give you a Kubernetes API server." I think the numbers kind of play themselves out ... |
So I think the point there is that there will always be a massive need for simplifying the cloud. And especially when you think about on the AI front, a lot of people are saying "Well, you will be able to write much more code with fewer developers, because you have Copilot, you have other things... And I think that the... |
The same thing applies to being able to build really complex architectures with really tiny infrastructure teams. So instead of hiring 20 people, you hire one or two. And that's the world we're operating in. So it's not like you'll have zero infrastructure engineers when you're a 500-person company running on Render. B... |
**Autumn Nash:** Can we go back to kind of how you were saying your learnings from bare metal to cloud and all that? I think a lot of people are in that transition, because at first everything was on-prem, then everybody was really excited about the cloud, and now everybody wants to go back to on prem, and everyone's t... |
**Anurag Goel:** I think that it's always very enticing to go to on-prem or bare metal for a company like Render, because yes, CPUs are cheaper there, bandwidth is cheaper there, you end up either having better margins, or giving people a better deal, or both. We, as a startup have a lot of constraints on how many engi... |
\[44:14\] So Render now has an engineering team that does focus on building stuff on top of AWS. But for the next stage, you can imagine us hiring a team that is already experienced with running infrastructure, let's say Kubernetes even, on prem; or running, say, a hypervisor on bare metal. And at that point, I think w... |
My learning is to just focus on what gets users value the fastest as a startup, and not try to be technologically cool, or say "Oh, we're going to do bare metal because it's going to increase our margins when we're tiny." When you're tiny, margins should not be what you focus on. When you're tiny, you should be focused... |
And then instead of focusing on "Oh, let's do bare metal" - and this is a debatable opinion, but for us it's important to focus on providing value. And I do see now a lot of people saying "Oh, you should run applications yourself on, say, Kamal, Rails" and DHH is spearheading this sort of "Let's not --" Even though I k... |
**Justin Garrison:** Well, they care about value, right? The value is that key piece there. And I love the -- Bill Gates a long time ago defined what a platform is, and I always go back to this, because I think a lot of platform engineering teams need to focus on that value piece. Bill Gates said something along the li... |
Subsets and Splits
No community queries yet
The top public SQL queries from the community will appear here once available.