text stringlengths 0 2.67k |
|---|
**John Watson:** I think for me it's like "Yeah, we're hosting a SaaS with our own product", which is certainly unique. There's not many companies doing something like that. And that was awesome; what a milestone. And now we're in like the management phase, and we're still kind of learning how to effectively manage wha... |
**Justin Garrison:** What was something surprising while you were building that out, that you're like "Oh, the tool can't do this thing that I need, or we should be able to do, or we were able to do with other tools"? |
**John Watson:** Traditionally, you write some AIC, and it represents your infrastructure. And it's like "Cool, archive that." Right? "Now I'm going to write a pipeline to rotate all my EC2 nodes" or "I'm going to write a pipeline to \[unintelligible 00:23:43.07\] do something", various other things. You know the deal.... |
So imagine what you wanted to do was run a health check, and then when the health check passes, execute something. So we can orchestrate that. It's not really like a pipeline management tool, but you can do some of the pipeline orchestration directly against the infrastructure in context, if that makes sense. |
**Justin Garrison:** You bring up an interesting point that I think has kind of been repeated with tools over the years, at least from my experience, where - like, early on config management days, it was like "Oh, everything's going to be Puppet, or Chef. That's the source of truth, and everything needs to be declared ... |
**John Watson:** Of course, yeah. |
**Justin Garrison:** So what makes System Initiative a better source of truth? |
**John Watson:** Here's an example. You've got a security group, and the security group is "Allow all traffic on \[unintelligible 00:25:29.18\] |
**Justin Garrison:** But for some reason it still doesn't work. I still cannot hit 80. Why? |
**John Watson:** Yeah, yeah. Security groups are the pain of my life. It's always the security group; that's a rule. |
**Autumn Nash:** Or DNS. Don't forget DNS. |
**John Watson:** Yeah, yeah. Okay, fair. Fair, fair. And it's like, we've got these things internally called refresh functions. So inside the application holds a state of what that resource actually looks like inside AWS. So it hits AWS, EC2, I think, the security group \[unintelligible 00:25:58.02\] pretty much. And t... |
\[26:07\] So imagine you have an incident, all the houses are on fire, and it's like, it's always a security group, or I don't know... And somebody changes it in the incident. You could see that in the resource definition in that panel. |
There's still some work in the future for us to -- we've got this thing called the attribute panel, which is basically like user inputs. And when you created the security group, you would say \[unintelligible 00:26:28.26\] And then the resource representation in JSON would have that additional security group that was a... |
We're still working on how we would sync that back into user managed attributes, but you still would be able to see the representation that "Hang on a minute... We had an incident. We changed something. Should we put it back?" So in System Initiative it kind of exposes that delta at some level, and you could rerun the ... |
**Justin Garrison:** And that's the thing with all of these tools where something changes outside of something that I originally had intent to change. If I'm running Puppet on a host and I say "Put this file on disk." Like, I'm not managing Etsy hosts, but I'm doing resolve conf. So once someone changes Etsy hosts, it ... |
**John Watson:** So if you think about the Terraform version of this, when you do an apply, it's like a three-way diff. It's what's in the code, what's in the state, and what's reality, and you have to smash everything together and go "If there are changes that aren't in the code, we need to get rid of them", et cetera... |
**Justin Garrison:** And that's the Kubernetes control loop model, right? ...where it's like "Oh, I own this deployment \[unintelligible 00:29:11.20\] and that's pods, and I will make it look exactly like this back over and over again." And then that's been extended; so cross-plane exists as a thing that creates infras... |
I think Kubernetes did a good thing here of like it tries not to by default manage things outside of itself too much, or at least not in reconcile loops. Load balancers are one of those things where it's like it will constantly check those, but it tries to do it in a way that's like "I'm not going to hit API limits for... |
**John Watson:** \[30:36\] Sure. Yeah. |
**Justin Garrison:** And I don't think that the Terraform "shove a JSON file in S3" is necessarily the right model either, but I also don't think that it's Etcd. And there has to be some other middle ground to make it so lik "Oh, this is truth enough that this isn't going to fail for me." Or it should fail a fewer amou... |
**John Watson:** I think what's very different about what we do is that ours is not really a three-way diff, it's like a two-way diff, because we don't have a code representation of the infrastructure. Like, what the UI is serving is the representation, and in the backend it's just a giant graph. And the graph has node... |
**Justin Garrison:** But the UI representation is like some form of code. I'm not like writing HCL, but it does need to diff that into -- it's Postgres under the hood, right? |
**John Watson:** We store the actual data of -- like, a node on the graph gets stored in Postgres, but the actual graph itself, the whole thing, it's stored like piecemeal, basically. And addresses to nodes are stored in Postgres. But it's not like a relational database; it's like a math, like a -- I don't know how to ... |
**Autumn Nash:** Is it a graph database that's putting information -- like, but the actual information from those nodes are being stored in Postgres? |
**John Watson:** Yeah. |
**Autumn Nash:** Wiz is built on a graph database. Is it the same kind of concept, I guess? |
**John Watson:** It sounds similar. The actual like information from the nodes on the graph -- on the graph, they're just an address, and the actual hashed \[unintelligible 00:32:20.04\] and hashed bits are the address, and the rest of it's just shipped off into Postgres, or in memory, or on disk. We have layers of whe... |
**Autumn Nash:** Your actual graph, is it a database? I guess - I don't know - it's a data structure, but like they actually have graph databases? Is it like -- |
**John Watson:** No... Probably not in the way that you're describing it, or I'm misunderstanding and I don't know. |
**Justin Garrison:** And that was one of the things I was curious, because Postgres is storing the data, and then the application is putting the links between them. Because it's not like -- not Node.js. What's the -- Neo4j. Neo4j has a graph database -- or the graph database layers on top of Postgres. It isn't what you... |
**John Watson:** Kind of, yeah. |
**Autumn Nash:** Yeah, because some applications will actually have a graph database, like Neptune or something, and then I guess it could either just store it in Neptune, or I could store it somewhere else after. But it holds all the relations between the different entities. |
**Justin Garrison:** How is that different than storing that in a JSON? Like, you said, you can store it on a file system. So I could store this on a file system, and then if I squint, it looks a lot like Terraform, at some point. I'm like "Oh, it's a big JSON file", and then I can pull out the nodes out of it. Like, t... |
**Autumn Nash:** The graph would be more accurate with you changing it on UI, I'm guessing, and keeping all those relationships straight and accurate, I would think. |
**John Watson:** Yeah, you could make arguments about "Are you transforming HCL into some other representation on the backend?" Like, it's not human-readable. It's like a data structure. I mean, I guess it's human-readable if you try hard enough, but... You know. Yeah, so it just gives us access to be able to like walk... |
\[34:17\] We have a model very close to Git, where you have like a head, and when you want to make changes in the infrastructure, you open a change set, make changes, and then basically push the changes onto the graph, and it reconciles and does what it needs to make that the new reality, and push it up to whatever clo... |
**Autumn Nash:** How do you make sure it's all secure without kind of shooting yourself in the foot at the same time? Because I feel like it's either -- like every automated security or automated anything, it's like, you want to automate things to make your life easier, and then at some point it makes it harder. So how... |
**Scott Prutton:** It's very hard to keep it secure. That would be my honest answer. |
**Autumn Nash:** You're giving something -- the permissions to make things and delete things, and interact with each other... So how do you -- |
**Scott Prutton:** You mean between the components in the graph? |
**Autumn Nash:** Yeah. Well, I mean, not just that, but it can make an EC2, right? It can make all these different infrastructure pieces. |
So how do you then keep them secure that something can't go in and, I don't know, make a hundred EC2s on one of your accounts, or...? You know what I mean? Like a bad actor isn't doing -- |
**Scott Prutton:** So the hardest thing probably for us as a company to protect from is user-written code; unprotected, just raw TypeScript or similar, that we execute on behalf of users... And that's why we went down the road of Firecracker. Because that's basically the Lambda model. So we could have went for somethin... |
So security on that front, I think like all of the networking rules are written -- like, we've done it basically from absolute fundamentals. There's no networking controller on a node for us. There are specific IP routes that are permitted for each VM, through to public access, through to the OTel collector... And ther... |
**Autumn Nash:** Well, I mean, also a human can be like "I'm going to go and --" I feel like it's weird how sometimes customers will use your product, and you think they're going to use it in a straightforward way, and then all of a sudden they're like "We want a thousand VMs", and like you don't want them to take down... |
**John Watson:** I'm very fortunate, because Scott is great, and I've never looked at what he's done in our workspace and been like "Dude, what is wrong with you?" So that's amazing. But each -- so we've got this concept of workspaces and change sets. So I hope this answers your question... Basically, that if you're co... |
**Scott Prutton:** \[38:27\] And let's say somebody starts using the product and they try to create 10,000 EC2 instances all at once... I mean, we have pretty rudimentary resource limits and things on the size of the VMs that we create, and like throttling, and how many we can run at a time... So ideally, all that happ... |
**Justin Garrison:** One of the things, Scott, you mentioned earlier about making the diagram, the infrastructure or the more truth infrastructure representation is I love drawing and diagramming, and every job that I've had, when someone's like "I'm describing the thing to you", I'm like "Can you draw me a picture?" B... |
And one thing we can't do easily in a static representation of a diagram is like "I need to go from architecture A to architecture B. How do I get there?" And in our heads, as administrators and architects and as developers, we know the steps that we need, and I'm assuming you would have to build some of that logic int... |
Subsets and Splits
No community queries yet
The top public SQL queries from the community will appear here once available.