Why Do People Hate Aurora Serverless?
They've been around for a long time.
Speaker 2:Can I tell you why I don't use it? Why? I just don't like the name. I really just don't like the name. That's it.
Speaker 1:That's funny.
Speaker 2:It's hilarious. You make your own LUTs?
Speaker 1:Well, what did you do?
Speaker 2:I just used the the Sony official LUT for my camera.
Speaker 1:But does does that like just work with your lighting and your background and like it it's all fine?
Speaker 2:Yeah. It it's made so you you basically like, you take a gray card and you expose to 44% or whatever and it's based off that. It's s log three.
Speaker 1:I write that's basically roughly what I did when I made it myself.
Speaker 2:Yeah. But you just, like did you use, like, Da Vinci or something?
Speaker 1:Yeah. Exactly.
Speaker 2:And you saved a lot. Yeah. Interesting. I am still shaking from an hour ago getting in the cold plunge hour and a half ago. I've never gotten warm today.
Speaker 2:It's like I've just been cold all day.
Speaker 1:Is it because it's cold outside?
Speaker 2:It's below freezing. It's 31 degrees outside. That doesn't help. It was the worst experience I've had in a cold punch yet. But by that, I mean the best experience because I feel better than ever.
Speaker 2:I'm just shaking.
Speaker 1:Yeah. I can kinda hear it.
Speaker 2:It's so bad. It's like I started cold. Like, I've been up a long time. We're doing some stuff, maintenance production stuff in the middle of the night. And when you're like, do you if you work early in the morning, you ever work?
Speaker 2:You work pretty early, don't you?
Speaker 1:Yeah. My favorite
Speaker 2:thing Yeah. Do your hands get cold? Like, I get everything gets really cold after
Speaker 1:So five I used to always have this problem of, my hands were just always cold all the time. Yeah. And you solved it? But then I just moved somewhere where I don't get cold ever.
Speaker 2:I was like, what's the trick? Oh, move to Miami.
Speaker 1:Got it. Yeah. Well, the other thing is I've never been colder anywhere than indoors in Miami because people love to just crank up the AC to a point where it's like indoors here, it gets like crazy cold. So when we were, like, staying at Liz's parents house while we're figuring out where to live, somehow no matter where I was, there was like an AC blowing like onto me. So yeah, my hands would freeze like working in the the office there because the AC would be blowing right on my hands and, yeah, I don't like
Speaker 2:it. I appreciate that about Miami and South Florida in general that you guys embrace the air conditioning. There's there's tropical climates where people don't, like Hawaii.
Speaker 1:I'm more like that. I don't know. I feel like the AC is overdone here.
Speaker 2:Oh, really? Oh, man. Yeah. I don't I don't enjoy a hot climate without air conditioning.
Speaker 1:Yeah. I mean, that's that's most people I think but I think like a like a fan and like a breeze, that's like enough for me for the most part but yeah. I mean, don't get me wrong, I appreciate the AC especially when I'm like coming back inside after being hot. But Yeah. It just like, it's freezing here inside sometimes.
Speaker 2:I might need to get a hoodie. I'm so cold.
Speaker 1:Go go do that. I do it. I think
Speaker 2:I literally have my arms crossed.
Speaker 1:I feel like you might like go into shock or something.
Speaker 2:Something. I actually this morning after I got out, my my hand had, like, little bumps on it and they were my veins, but they were, like, in balls, like little BBs. It was weird looking. Like, my blood froze or something. I'm gonna get a hoodie.
Speaker 1:Yeah. Go do that.
Speaker 2:Hey, I have a question.
Speaker 1:Yeah.
Speaker 2:Why do people hate Aurora serverless v two?
Speaker 1:Are you sure this cold plunge is good for you? Because you do not sound good in any way.
Speaker 2:That was just, like, food that got caught in my throat
Speaker 1:I see. As I started speaking. Okay. Your question.
Speaker 2:Yeah. So I don't I don't remember why, but I felt like I was supposed to not like Aurora serverless. And I looked at it and for our needs, it made sense and we switched to it and it's good. Don't
Speaker 1:two. Think possible. Okay. Yeah. So there's like a whole cascading list of reasons, I think.
Speaker 1:One is that when you think Aurora serverless, like they threw this word serverless in there, it's just like more aggressive auto scaling Postgres. Like, you can't really do it in a serverless way so it's not like the same as Dynamo.
Speaker 2:It's snappy. Like, it it is snappy with the way it scales up and down with the ACUs. Like, it's it's not like a, oh, you started getting a bunch of traffic. In a minute and a half, you'll have another instance deployed. It's like it's pretty opaque or abstract.
Speaker 2:Like, you just Yeah. Set them in max and it just kind of glides around.
Speaker 1:So v one was a lot slower. It would like it would scale up but then it wouldn't scale down for like an an hour. So if you got like a spike for like twenty minutes, still paying for the whole hour. So it's it wasn't like the most responsive thing especially for people coming from Dynamo where it's like completely opaque. The but then the other issue is the stupid v one v two split.
Speaker 1:Like they launched the v one version of Aurora and now there's a v two and in your head you're like, oh, that means v two is what we should all be moving to because it's the next version. They're gonna like kill v one.
Speaker 2:Yeah. That's not how v one and v two works at
Speaker 1:AWS, is it? No. It's it's like completely two different approaches to the same problem and they just named them v one and v two. And I would love to use v two.
Speaker 2:API gateway. Yeah.
Speaker 1:API gateway is another classic example. I would love to use v two except that, it doesn't support data API, which is one of the best things they did with, serverless v one. So you don't need to have CP connection, it's all HTTP based. This isn't like super obvious why this is good. So one, obviously it gets rid of the need for like connection pullers, RDS proxy, all that, like it's it's very simple.
Speaker 1:The other case that, you have to be under certain conditions to run into, so you might not have run into this but when your function is done executing, it gets frozen. Right? It gets frozen and from its point of view, like, no time is
Speaker 2:When function?
Speaker 1:Like, when your lambda function is done executing. Oh, It's gonna get frozen for a while and kept around so it can handle the next invocation pretty quickly. Right. But if you have an open TCP connection that's like frozen, the other thing at the end other end of that connection, the database might look at it and be like, this connection is not being used, I'm gonna terminate it. But then when your container gets unfrozen, it thinks it's still connected and you might have like weird like connection reset error.
Speaker 2:Okay. So I have not run into that but it's been, like, a few hours.
Speaker 1:You're not gonna you probably aren't gonna run into it because your functions are probably always busy. So the next request comes in, like, immediately after it's done. But if you're in this like middle zone where like you get enough traffic to keep the function warm but not enough so that like there's so like these like long gaps of being it being frozen like several minutes whatever. Yeah. You're gonna get these weird you can't just resume a TCP connection as though nothing has changed because something on the other end can can decide to stop communicating.
Speaker 1:So it's not it's fine in a lot of cases but like the data API is kind of necessary. That's why you see all these other database providers like PlanetScale has their own HTTP based API, like Neon came out with their own HTTP based one. Yeah. I think theirs might be WebSocket based but, yeah, they just support these more stateless ways of communicating. And I want if v two had data API, we're all good.
Speaker 1:We're all gonna move to v two.
Speaker 2:I'm using RDS proxy in front of it.
Speaker 1:Yeah. But I mean, it's it's gonna have that same problem.
Speaker 2:Oh, okay.
Speaker 1:Yeah. You might like I said, you might not because your functions might be like super busy. Yeah. But you're you're gonna probably once in a while see this, like, connection reset error. It's not the biggest deal, but it's just indicative of why the underlying system, like, doesn't line up well.
Speaker 2:Okay. Well, I'm gonna be checking invocation logs because I don't know if we're handling that, like, if that happens.
Speaker 1:Yeah. I mean, the request was just error and then I don't know what it's serving. So it might Okay. The user might just retry or whatever.
Speaker 2:Yeah. Well, we were going from classic RDS Postgres.
Speaker 1:That's a big jump. Yeah.
Speaker 2:Yeah. And so it's our first jump into Aurora. Obviously, like, Dynamo is great. I love building stuff with Dynamo. This was built nine years ago, and we weren't gonna port it to Dynamo.
Speaker 2:But we did move to Aurora, and we decided to do serverless. And so far, it just seems a lot better. I think we're gonna save money Yeah. Because our traffic is so, like, time of day dependent. Like, the evenings, we get these crazy amounts
Speaker 1:of and stuff like that. Yeah.
Speaker 2:Yeah. Exactly. So I I do think, like, it's gonna scale up quite a bit then. And and we were having issues. The reason we moved to Aurora is that we were pegging out our current instance size.
Speaker 2:Right. And we just had been previously scaling up. And I just thought, like, maybe it's time to make the leap. So far, so good.
Speaker 1:Yeah. I mean, it it makes sense to me. It almost feels like that should just be the default. Like, is it it just is the same as normal Aurora but can scale up and down fast. Like, why wouldn't you want that?
Speaker 2:Yeah. Why choose a cluster size? Like, we were not using most of it all the time and then we were sized such that we could handle that peak at the end of the day and it lasts for like an hour. And then we're just using we just have this idle box. That's the case for serverless, isn't it?
Speaker 1:Yeah. The over provisioning. Yeah. You don't you don't wanna be over provisioned. Yeah.
Speaker 1:Like, I think my issue really is overall there isn't like a killer serverless Lambda function plus relational database setup on on AWS. Yeah. I think, like you guys are already in a VPC so it's not like you've started to adopt a VPC for it so that was not a big deal but, yeah, now for the average person, they now need to like put their lambdas in a VPC and there's there's a lot of stuff like that. So yeah, I would like if the data API came to v two, then I would like, no problem, no hesitation, recommend. Here's a good Yeah.
Speaker 1:RDS setup. Right now whenever people ask me like what should I I wanna use a relational database, like what should I do? I actually usually tell them like consider going outside of AWS.
Speaker 2:Yeah. PlanetScale's great. I mean, I love PlanetScale.
Speaker 1:But if they added the data API, then I can, like, kinda comfortably say that you're still gonna have a pretty pretty good experience. But, yeah, there's no like, I don't understand why they lost it and why there's no signal on whether they're gonna add it back. Like, I've seen some discussions and it's always some vague thing like, oh, yeah, like, we're looking into this. But,
Speaker 2:Maybe maybe pre invent, maybe re invent. Who knows?
Speaker 1:What is up with PreInvent? I haven't heard anything.
Speaker 2:Is this There has not been any mumbling. Are they just holding it all for Reinvent this year?
Speaker 1:Maybe There isn't as much.
Speaker 2:Well, I was gonna say maybe fifteen years in or whatever it's been for AWS, maybe there's so little big announcements that they're just like, you know what? Let's save it all for the conference. That that's gonna happen eventually. Right? They can't continually just have, like, overloads of information every year.
Speaker 1:Yeah. But at least last year, got a bunch of leaks. Like, I I haven't had any leaks this year.
Speaker 2:I guess I have some leaks.
Speaker 1:Look at
Speaker 2:me not sharing them with I
Speaker 1:see. I see. Well, I don't need you to tell me the details. Is it is there anything that I would be excited by in your leaks?
Speaker 2:To be honest, no.
Speaker 1:Because I'm not excited about it. Yeah. Here we go. I There's some
Speaker 2:mean, there's some stuff that I once I see more than just I haven't joined any of the calls. Like, the heroes have these NDA calls where we get briefed by the product teams ahead of re:Invent. I haven't joined any of them. I've just read the, like, announcement for the call. And in those little announcements, there'll be, like, bullet points.
Speaker 2:So I've gotten very little information from that. There are a couple of them that depending on how it all really turns out, it could be an interesting. But, yeah, nothing like major. This is there's no, like, Lambda going to one millisecond billing or, you know, that kind of stuff.
Speaker 1:Yeah. I
Speaker 2:don't see anything like that.
Speaker 1:Well, okay. I I do feel like I've heard murmurings of big changes in Lambda. So I'm still kind of hoping for that. Like I don't imagine that would be a pre invent thing. I think
Speaker 2:that would
Speaker 1:That'd be a
Speaker 2:yeah, that'd be the conference thing.
Speaker 1:Yeah yeah. We'll see. Yeah. But I've I haven't heard anything. That said, I don't even know what I want.
Speaker 1:Like, I don't even know what I'm imagining. Like, the there's just a classic things of like, search serverless search. Like, where is that? But besides that
Speaker 2:Yeah. So is OpenSearch so I'm gonna be looking at this pretty soon. OpenSearch serverless is bad. I just remember all the services that came out as serverless and they weren't very serverless.
Speaker 1:It's a classic thing where if you have to put a label on it, it's not the thing. Yeah. Like social science and this is same case. And anything that has the word serverless in it, it's it's not serverless. Like, they don't call it Lambda serverless, you know.
Speaker 2:No, right. That's true.
Speaker 1:Or like Dynamo serverless or SQL serverless. Yeah. Yeah. OpenSearch is it's the same thing. It's the same thing as that's actually the exact same thing as RDS situation.
Speaker 1:Okay. Except the floor is crazy high. The floor is like $800 a month or Okay. I get it with search, like search is anything that isn't like designed from day one to be serverless is hard to make serverless. That's why it's hard to make a relational database serverless.
Speaker 1:Even plant scale, like, you very quickly switch back to having, like, instance sizes there. Yeah. So and search is like, especially that way because it's so Yeah. Memory heavy and Yeah. Know, that.
Speaker 1:So I I get it but at the same time, Algolia exists, TypeSense exists. And I I think TypeSense is actually instance based, but Algolia is not. Amazon's scale, I feel like they can effectively make anything seem serverless even if it's hard just through
Speaker 2:like Yeah.
Speaker 1:Pricing tricks. Yep. So yeah. I don't know. That's that's the one big thing that's missing.
Speaker 1:So you guys are looking for search stuff?
Speaker 2:We already use we use Elasticsearch. I think we use like a hosted elastic.co, like the actual company. But we're gonna move it in house, I'm pretty sure, to AWS and
Speaker 1:To open source?
Speaker 2:Probably use probably, yeah. I think so. And we may not do the serverless thing, but I don't know. Just there's an option. It seems better than the traditional.
Speaker 1:Yeah.
Speaker 2:I'm I'm remembering the Chris Moon's tweet where he was like, serverless is dead. He was like that that was like the catchy phrase. But what he said was that
Speaker 1:You keep calling Chris Moons. That's like, you're never gonna doing that, are you?
Speaker 2:Well, I forgot what it was. We haven't said his name in so long.
Speaker 1:It's m u n n s,
Speaker 2:which Munns. Is Munns. That's what I said, isn't it?
Speaker 1:Oh.
Speaker 2:I feel like I said Munns and you were like, you're gonna keep calling him Chris Moons, aren't you? No.
Speaker 1:We can go to the playback. We can go to the playback. Let's not
Speaker 2:do that. Let's not do that. I believe you. Let me get my little recording device here. It says oh, yeah.
Speaker 2:No. No. What he said was that, like, every company actually has a VPC. Anybody building with serverless, it's like this mishmash of some of it serverless, some of it but they all have VPCs. And at the time, I remember scoffing.
Speaker 2:And, like, when I look at my situation, it's like, oh, yeah. No. We have this mix mash of all kinds of stuff, and not all of it's purely serverless stuff.
Speaker 1:But you have a VPC because AWS requires you to have a VPC. Yeah. It's it's like a circular thing.
Speaker 2:Well, yeah. I mean, like, everybody has a VPC. Sure.
Speaker 1:We don't have a VPC. We
Speaker 2:have a default VPC. Did you delete it? Yeah. Oh, for real? That's awesome.
Speaker 1:We have nothing deployed in VPCs.
Speaker 2:Well, I figured you didn't have anything deployed in it, but everyone gets one in their account. Then you got to explicitly delete it. And I deleted it.
Speaker 1:You hated it. You got rid of
Speaker 2:that thing. But oh, okay. So you have nothing deployed in a VPC. I guess, like, we we started with just, like, e c two applications and a database nine years ago, and it made sense. Like, we just it's all in a VPC.
Speaker 2:Yeah. Once you have a VPC, you're never getting rid of that VPC. Like, I can't imagine that we ever just get rid of everything that's running in our VPC. I guess we could.
Speaker 1:It just comes down to if people had the option to use one and to not use one, everyone's gonna pick to not use one.
Speaker 2:And it's to me, it's
Speaker 1:that simple. So yeah, of course, a lot of people use VPCs and will continue to do so because they have stuff that was forced into a VPC and they're not gonna be able to move it or change it. But Yeah. Moving forward, the more stuff that is not VPC required, the better.
Speaker 2:Yeah. I'm trying to think of like what what we have to have a VPC for. I guess, like, if we were using well, if the data API existed and we didn't have to connect to our database through a VPC That's not
Speaker 1:what wanted. Data API doesn't need a VPC. That's a big Yeah. Big reason I wanted.
Speaker 2:That that'd be great. If we had that, then we could continue on with Aurora. And then presumably well, no. I mean, like, our our some of our services are just, like, long living, like I mean, one's a dot net app that runs an ECS. It's gonna have to have a VPC.
Speaker 2:Right? You can't run, like, Fargate services without a VPC.
Speaker 1:Yeah. Fargate has to be in a VPC.
Speaker 2:Yeah. So, like, those apps I guess we don't all of our workloads can't just run as, like, Lambda. Well, they could. I mean, architecturally, there's no reason they couldn't. Just like we can't we're not gonna
Speaker 1:be right. You're switch it. Yeah. That makes sense. Yeah, I mean, it makes sense for existing applications that are like hybrid of the two.
Speaker 1:Oh, yeah. The other thing is like, the reality of why VPCs are required, I don't know why they're not just straightforward about this because if you look at something like, what's their memory, their Redis equivalent? Mem
Speaker 2:ElastiCache.
Speaker 1:ElastiCache, okay. So to deploy ElastiCache, it needs to be in a VPC, which is why I never use it. Like I
Speaker 2:I I
Speaker 1:like Yeah. Just make do without, like, Redis equivalent. Dynamo's put it in Dynamo and, yeah, that's usually good enough. If I need more, I go to Upstash. Mhmm.
Speaker 1:The question is why is it not like, why does it require VPC? It's because with services like that, you need to build centralized authentication somehow, right? Like how do you do IAM auth Yeah. Over Redis protocol? You have to, like, invent a way to do that over Redis protocol.
Speaker 1:They don't wanna do that or maybe it's too hard or maybe it's impossible, so they shove in a VPC and say, okay, your VPC itself does a centralized auth Yeah. Then everything in there can just not have that. So there's only some services they've been able to bring that to. Right? They've been able to bring it to Dynamo, they've been able to bring it to DIT API.
Speaker 1:I think they can do it for everything.
Speaker 2:S three, right?
Speaker 1:S three, yeah, the ones that that was like built from scratch.
Speaker 2:Yeah, secrets manager.
Speaker 1:I think built from scratch at at AWS will have it. Yeah. But same with the open search. Open search is just like needs to work over the whatever Elasticsearch. I mean, that's just HTTP so I feel like that one could be done.
Speaker 1:But the other side is, other companies have done it like Upstash gives you like an HTTP based API for Redis.
Speaker 2:And Yeah.
Speaker 1:So they they managed to move the auth somewhere else. Yeah. But, yeah, VPC was just a nice way to be like, we don't have to worry about it, throw it in there and centralized auth just works.
Speaker 2:So if I were building a new app, let's say I could build something from scratch today and it would immediately have lots of users. And I use, like, API Gateway, Lambda functions, and Dynamo. All of that stuff is just communicating over the Internet, but it's over like, Amazon's smart enough to know it's all within Amazon data centers. But is there, like, a billing implication? I should know the answer
Speaker 1:to this. No. So if it's within the same availability zone, it is there is no billing implication.
Speaker 2:What availability zone? The if I'm using all these public services?
Speaker 1:I'm saying the general rule is if it's within an availability zone, it's, not gonna cost you. So when you use these services abstract away the constant availability zone, that's handled for you.
Speaker 2:Oh, they put it in the same availability zone. Like, if I'm calling Dynamo from a Lambda function, yeah, they just Yeah. They it does it from within the same availability zone. So there's no
Speaker 1:Yeah. Exactly.
Speaker 2:Cross region bandwidth charges.
Speaker 1:Just cross availability Yeah. Yeah. Yeah. So that's like a little I think people end up paying a lot when they, don't use the managed services because they're like they don't think about those things and they have traffic going across availability zones by accident. Yeah.
Speaker 1:Yeah. So yeah, the public internet thing also confused me for a bit. But yeah, it like gets routed correctly or, like, routed efficiently.
Speaker 2:It all stays it it's not like it literally goes out to some outside the data center DNS resolver and, like,
Speaker 1:oh, oops, come back. Yeah. Come back. Yeah. Exactly.
Speaker 2:Okay. So they do smart things. But but I guess, like, is there any security implication? Like, is there any is it less secure to build that than to build a bunch of stuff living in a private subnet?
Speaker 1:I think that I think people get confused with this. So I think and it it shows up a lot when you're, like, going through compliance when when they hear that you don't have a VPC, they're like, woah, like, that's not safe. Woah. Yeah. But remember the point of a VPC is to, like, do central centralized auth, which a VPC does with like in like a really crude way.
Speaker 1:Whereas, these these more purpose built services, they do centralized auth through IAM and that's no more or no less secure than doing it through a VPC.
Speaker 2:Hang on. I wanna dig at that a little bit. So when you say the purpose of a VPC is to handle centralized auth, you just mean like mean, because the purpose is to create, like, a virtual network that can be private and isolated at the network level. Right? Like
Speaker 1:But it's all done in software. It's not any
Speaker 2:different. Software, but it's like software network components. Like, they've literally created, like, software routers, right, that are that were revolutionary. But, like, it's not I don't know. Now I'm just selling it.
Speaker 2:Like, it's it's ultimately like you have a private physical network. You have network isolation whereas with IAM, it's still it's not like, you don't have network isolation. Right?
Speaker 1:Yeah. No. It's implemented at a different layer of the the protocol. But if you think about what the software router is doing, there's just some rule somewhere saying, if you receive a packet meant for this destination from this source Mhmm. Make sure it's allowed to go through.
Speaker 1:If it's
Speaker 2:not Yeah, I guess that's what you're saying.
Speaker 1:So it's just a rule implemented somewhere. So yes, it's like, it might have a lower level of the the TCP IP, whatever that stack is.
Speaker 2:Whatever the OSI model Yeah. Or
Speaker 1:Yeah. Just butchered all of that just Alright. To the it's funny. Yeah. So it's it's it is the same thing.
Speaker 1:The same rules are implemented just at a different part of the stack. And I would argue in some ways, it's more precise because a network rule can't be super precise. It can't say like, this is allowed but only like read read commands from Dynamo. But you can't implement that at that low Right. Level.
Speaker 1:So to me at worst, it's equivalent. At worst, it's equivalent. And I think it's better because in practice, you're more likely to get IAM right than to get, like, VPC routing rules right.
Speaker 2:Yeah. No. That makes sense. Just managing a VPC is so hard. Like, it's because there's not, like, one unified view of it in AWS.
Speaker 2:It's like, I'm gonna click on the subnets page. Now I'm gonna go to the route tables page. And, like, everything in isolation means nothing. Like, I don't know the ID of my route table. That doesn't help me when I'm trying to find it in a list.
Speaker 2:Like, it's just the whole navigation that's in the VPC console is is weird.
Speaker 1:I have a tough time holding the concepts in my head because there's like so many places where you secure things. There's like the routing, but there's also security group. And I'm always like, wait, what does a security do for the group? Because it's like, one is for the destination, one is for the source, like it's Yeah, it is very hard and I'm glad that I haven't had to like really think about this in a long time. Because back in the day, that this was like the first thing you did whenever you started a new project, you went and like
Speaker 2:Oh, yeah.
Speaker 1:Set up all
Speaker 2:Configured your v VPC.
Speaker 1:Yeah. I mean, now it's it's like a one liner CDK thing but
Speaker 2:Right. Yeah. But, yeah. It it used to be, like, thinking about IP address allocation. Yeah.
Speaker 2:Know. That's funny. Like, what's your CIDR mask gonna
Speaker 1:be? Yeah. I never learned that, by the way. Yeah. Yeah.
Speaker 1:Did you ever learn how they worked?
Speaker 2:I did because I took all the certs. So, you know, there's, like, tools. Cidr.xyz actually helps conceptualize it. You don't need to learn it at this point. Like, you you made it.
Speaker 1:I don't yeah. I survived. I somehow configured all my networks all these years without knowing what how to do a SiderMask. I just know that slash eight means a lot. Slash 16 means less.
Speaker 1:Oh, that's funny. Slash 32. Yeah, exactly. I don't understand why it's not well, I'm sure it's a good reason, like an efficiency reason, but just give me like a like a regex or like a pattern, like a wildcard. Like 10 dot zero dot 0 dot star.
Speaker 2:Yeah. Dot star. Yeah. That makes sense. Dot star dot star dot star.
Speaker 2:There you go.
Speaker 1:Yeah. That much easier.
Speaker 2:So I I realized I was, like, making the case for the, like, old way of thinking. And you've convinced me, which is not hard to do, that it really is just all better. Like, we didn't need to know all that stuff. And people who think it's more secure to to set up their VPCs and to run these old things on Docker containers or EC two images. Like, imagine right now there are people, lots of them, that are managing their Docker installations on raw EC two and running containers in their little EC two instances inside of some VPC and manually, like, configuring their load balancers and doing all this stuff.
Speaker 2:And, like, that's just a lot of stuff you don't have to do anymore. And people are gonna keep building that way for another decade. It saddens me.
Speaker 1:You know who was doing that up until very recently? Frank at our company because. Oh, really? Okay. Well, there you go.
Speaker 1:Actually, don't remember exactly the setup but like, for
Speaker 2:There's no hope for any of our
Speaker 1:our CI tool. I know if if if even we're we're still doing that stuff, our c like, our CI process was running on containers in EC two, I think. You just moved it all to EC two.
Speaker 2:You weren't even using, like, one of the managed container services. You're using, like like, you were installing Docker on EC two, like a custom AMI and you're, like, building up your EC two instances. Yeah? Wow.
Speaker 1:Here's the thing though, it's, at the limits, the problems show up, which is why we were doing this. So for a CI tool, it actually really matters a lot how fast your container starts up because Oh, yeah. It's a CI service, like, you're gonna start up containers whenever people need to build. Yeah. So ECS for a while was not a good option because ECS took forever to boot up.
Speaker 1:But they have this new thing called, some kind of in rum index, no, not rum index, some kind of index thing. You you can like do this thing to your docker image that makes it boot up way faster. So we ended up moving to ECS and it actually shaved like 2 or $3,000 off of our monthly bill. Wow. And we get almost the same performance and like a lot more flexibility.
Speaker 1:So, yeah, it was a good move but again, like, you talked about it the other day, like, are some exceptions to this stuff. Like a CI servers is a great example of an exception. I don't think anyone should build a CI servers anymore, like, GitHub Actors exists. Like, everyone just wants to use that.
Speaker 2:Wait, aren't you building one right now?
Speaker 1:With with Asterisk, like, we
Speaker 2:don't really want existing business into a new UI, like a So new Our
Speaker 1:our new approach So historically, there's been a lot of CI services and at some point GitHub actions came out and it was like, duh, you want your CI servers right next to your code, like, obviously that's a much better setup. And there's like a million undifferentiated parts of CI, environment variables, like custom scripts, like setting up steps, like installing a version of p n p m you want.
Speaker 2:There's like a there's like an
Speaker 1:infinite list of just stuff that is just not fun about running a CI service. Yep. So our thinking now is we probably still will have like an extremely minimal, if you just want a standalone git connect deploy, very minimal and not very customizable. But we're also gonna make it so our CI servers can just be a step inside your actual CI servers. So if you're using GitHub Actions, you can do like, ssc cloud deploy or whatever.
Speaker 1:And that'll deploy using like a really optimized builder on our end. So that way, we don't to like rebuild all of CI and all the boring parts like the one little differentiator that
Speaker 2:It's just like an adapter for GitHub actions that
Speaker 1:Yeah. It's like a step. It's like a step in GitHub actions.
Speaker 2:Yeah. I really I want the CI stuff built into console. I just want everything you guys do in seed in the new modern hot tool because I mean, I I don't mind having both of them open. It's fine. But the new one looks nice.
Speaker 1:It's nice to have it on one place. Yeah. Yeah. No. We we definitely need CI.
Speaker 1:It it's it's this funny thing where, like, you know, we had Seed for a while and we were like, the CI business is, not really a great business. And we were like, okay, we probably won't need to be in it anymore. Then I posted that question, where I was like, everyone says Vercel has great sorry. Everyone says Triangle Company has great, great great DX. And I was like, are you guys referring to when you when you say that?
Speaker 1:And every single response. The only response that I got really was the CI, how easy it was to connect to GitHub and and get into production. So I
Speaker 2:like, oh. That's every single tool today. Like, I don't understand how that's, like, the level the bar for I know. Developer experience. Like, I can connect my GitHub and it just deploys.
Speaker 2:Yeah. You can do that to, like, 15 different services. That's, like, pretty table stakes at this point, isn't it?
Speaker 1:But that's what we learned.
Speaker 2:We're like, with table stakes, we're gonna have it. And and that's what I I wanna know. Like, when people are like, oh, well, I can't give up the Vercel developer experience to, like, run my own stuff. Like, what is the thing you can't give up? I don't understand.
Speaker 2:I haven't thought about CI in I don't remember. Five years?
Speaker 1:Yeah. I mean, they might just be comparing it to, like, the worst version of it. Because, like, imagine, like
Speaker 2:People use CodePipeline?
Speaker 1:Or, like, imagine even just, like, getting an SSL certificate working.
Speaker 2:So that's another thing. I saw Theo tweeted this the other day about how things he hasn't had to think about since he started using Vercel. I haven't had to think about SSL certificate in, like, ten years. I don't even remember the last time. And I used to configure them manually, but we haven't had to do that for so long and I don't use Vercel.
Speaker 2:Well, I do.
Speaker 1:But Yeah. The point is it's easy. Like, a lot of different people have made all of those things pretty invisible at this point. But, like, in your head, if you're not aware of that, you're probably comparing it to, like, going to some, like, SSL website and, like, buying SSL.
Speaker 2:Yeah. Like Digicert. I mean, I did
Speaker 1:the same
Speaker 2:back And in then, like, oh, it's gonna expire. Does anybody have the credentials to Digicert? Oh, shoot. We're screwed. Yeah.
Speaker 2:Those days. Yeah. That's been a long time ago.
Speaker 1:And then even then, like, setting up the was it OpenSSL? What was it called? What's the thing we all use now?
Speaker 2:See, The thing we use now?
Speaker 1:I've been so detached from this, don't even know.
Speaker 2:I mean, I just use ACM. I just use I'm an AWS
Speaker 1:There was a step in between that made SLS certificates free and widely available, and I'm totally blanking
Speaker 2:on a name. Yeah. I'm blanking on it too.
Speaker 1:I forgot forgot what it was. No.
Speaker 2:SSL. No. It's the name of it. It's gonna drive me crazy. It's like a free no.
Speaker 2:Open was it? Hang on. Let's just go to a website and it's gonna be like what they
Speaker 1:used. Let's Encrypt. Let's There it is.
Speaker 2:We go. Let's go.
Speaker 1:Alright. So we we used to run, like, Let's Encrypt services in our Kubernetes cluster and that handled it. And that was like much better than the previous step but also it still kinda sucked. So I can see how if you don't know, one, if you don't even know how this stuff works or if your like perception of the alternative is like one of the options we just talked about.
Speaker 2:Yeah, one of old ways to
Speaker 1:do Yeah, I can see that. But yeah, the git push deploy, that is whether or not it's like truly useful or like truly like a killer feature, it is something that sticks in people's heads. And I guess it's because it's like their first your first interaction with the service is usually getting deployed for the first time, so I guess that's why it sticks in people's heads. Yeah. So, yeah, we realized, okay, we can't skip CI that we need to we need to include it.
Speaker 1:Yeah. Gotta have that. It is an important part.
Speaker 2:It's table stakes. Want you to figure out how to use the CloudFront blue green deployments because it feels bad to make big changes to our distro when at the levels of traffic we have constantly flowing in. I would love to be able to put them in a blue green deployment in CloudFront, which is like a feature in CloudFront now. I don't know if you've seen this. You basically create
Speaker 1:like a Yeah.
Speaker 2:It's a feature in CloudFront. And like you Yeah. It's basically it creates like a second, like a shadow distribution that you can configure however you want and you can you can transition traffic to it, like 10%, 15%. You can have it do it all the time. Yeah.
Speaker 2:Yeah. It's it's pretty great.
Speaker 1:That's really great. Yeah. Don't know how long it works exactly. Figure it out. Yeah.
Speaker 2:Yeah. Figure it out because I would love to build, like, AB test. Or not AB test, but just, send slivers of traffic to because for instance, this the latest SSC stuff, what's now been weeks ago that you guys that was merged in, it's pretty big sweeping changes to the distro, like, to the the behaviors and everything. And it's scary to just, like, flip the switch and send that to everybody.
Speaker 1:Yeah. It'd be
Speaker 2:nice to be like, here's 10% of the traffic, and it's like a feature that's built in. I don't know how it would work from like a infrastructure as code standpoint, like a custom resource that just like I
Speaker 1:don't know.
Speaker 2:You could configure it to, like, slowly fade it or something. I don't know. That's for you smart people to figure out.
Speaker 1:Yeah. We gotta look into that. Yeah. It's funny because we've been, yeah, we were talking with the team yesterday and I feel like we've have so much clarity now that we just haven't had because whenever you do when you're like starting a company, it's just all lack of clarity. You're like
Speaker 2:Yeah.
Speaker 1:Here's a thousand things we can do, a thousand directions we can go in, thousand unsolved things, like what the hell do we focus on? Mhmm. And for a while, we were stuck on this problem of, like, what is a default project? Like, what is a default SSE project? What does it look like?
Speaker 1:Like, what are good practices? What's like the framework part of it? Like, how does it work? It was so hard for us to like draw an opinion on this because so many so many different setups, so many like, everyone had different ways they wanted to do things and it was so hard to like, do one setup that worked for everyone. Yep.
Speaker 1:But now that we have this deployment story around these frameworks, like, Astro, Remix, NextJS, whatever. Mhmm. We like realizing that it's that's no longer our problem. Like, we just support the framework itself really well, everything you need to do in production with the framework. Yeah.
Speaker 1:We don't have to worry about like the design of the framework. You someone's already decided that they like this framework and they're use it, we just gotta make sure it deploys really well. Yeah. And initially, I was like, really thinking narrowly of like, oh, that's just these quote unquote front end frameworks but now they deploy so well that like, I can actually see myself using all the back end features of it as well. Yeah.
Speaker 1:And there's all these other frameworks that exists outside of this like front end framework zone like, even just Express or like Fastify or like Hono.
Speaker 2:Yeah. What's like Nest? People always talk about Nest.
Speaker 1:I don't like that. Like it's Okay. Yeah. That's another example.
Speaker 2:I just I've seen people say it.
Speaker 1:Oh, yeah. Exactly.
Speaker 2:Because it was like, you mean And
Speaker 1:it's like, we can just focus on figuring out going step by step with to each of those frameworks and figuring out here's the right way to deploy it. And that makes our job so much more narrow and clear. And it also expands our we don't have to convince anyone of anything. Like, whatever you choose to already use, we we can just help you deploy it.
Speaker 2:Can I throw more names out there and just see if you viscerally viscerally react to them as well?
Speaker 1:Yeah.
Speaker 2:Redwood JS?
Speaker 1:Yeah. We talked about Redwood as well. Okay. That's another one we can support. Sounds good.
Speaker 1:I haven't like really used it so I don't Okay. I don't know.
Speaker 2:Adonis? Adonis?
Speaker 1:Yeah. I think the problem with Adonis is it's very geared towards I don't know how well it can be deployed as a serverless thing. Like, you can use our container option.
Speaker 2:That's the thing. That's the the big thing. Right? People wanna keep using their fifteen year old frameworks, which I get. It's mature.
Speaker 2:But they just, like, accept that their infrastructure is also gonna be 15 years old forever.
Speaker 1:Yeah. Exactly. They're never
Speaker 2:gonna be taking advantage. It's all worth it. Right? I I always ask this question to you. But, like, I just start questioning my own conviction when everybody's like, why do I care if it's serverless?
Speaker 2:Well, like, there it's all worth it. Right? All this stuff I think so. I mean, we're using it at scale and it's great.
Speaker 1:Yeah. I don't know. I don't at this point, I don't I know it's worth it.
Speaker 2:I just need to stop trying to convince people. I don't need to. Like, it's not my job.
Speaker 1:Yeah. I mean, we literally see every single day thousands of companies, like real companies making money.
Speaker 2:Yeah.
Speaker 1:We lit like, the SST team literally sees that every single day. Even then, it's it's like not enough to not have like a doubt in my head, you know.
Speaker 2:Yeah. It's just funny like if you get on Twitter, you'd think like AWS is this very obscure arcane technology that nobody uses. But I remember at TwitchCon, of all places, TwitchCon, sitting around with all these people that you see on Twitter, and they're all talking about AWS. Like, they're literally all like, oh, yeah. Our s our SQS queue got backed up and then we had a da da da da da It's like, I forget that all these people at their jobs at these FAANG companies, they use AWS.
Speaker 2:Like, everybody uses AWS. It's so dumb.
Speaker 1:Yeah. They have a million customers. That's a lot of customers.
Speaker 2:That's a lot of customers.
Speaker 1:Yeah.
Speaker 2:Okay. I wanna stop trying to convince people of it. What about like oh, no. That's the same problem. Never mind.
Speaker 2:I'm gonna go into I was gonna say Laravel.
Speaker 1:Laravel deploys in serverless way.
Speaker 2:Yeah. Do you guys think about Laravel? Do you guys think about, like or do they just have their own products?
Speaker 1:Okay. So our our dream is that there is an SST config in every single project no matter what it is. So today Mhmm. It's reasonable to see an SST config inside an Astra project, inside a Next. Js project, Remix project.
Speaker 1:We wanna The next step is to then like, support like the more back end y frameworks of JavaScript, like Hono, Express, Fastify. Yeah. Like, get it That's like obviously very adjacent. And the next step is to say, oh, that Go framework that you're all using Ah. We deploy that too.
Speaker 1:Or like Mhmm. This, Python framework you're using, deploy it too. Rust framework is exactly. And then eventually that gets to things like Laravel and and all of that. So I think the root idea is managed services are just gonna continue to grow in use even if your framework is very capable and you always need to glue that together somewhere.
Speaker 1:And I think that s s t config file could be that place.
Speaker 2:That just made me realize I haven't ever asked you. And maybe I don't know if you don't wanna share it, don't share it. But are most of your customers are most of the s s t users like doing the drop in thing within an existing project or are they building their project from start with SST like the SST mono repo style?
Speaker 1:I have no idea.
Speaker 2:Yeah. Guess how would you know? Well, don't
Speaker 1:know. This this is like a weird blind spot for us because we don't track, our create SST
Speaker 2:Yeah. And we So it'd only be off like just anecdotally seeing the support stuff and looking at people's projects.
Speaker 1:Yeah. I I think support wise we're seeing a lot of drop in stuff and I think realistically that is the future. Like, we need to be as light as possible.
Speaker 2:Actual companies have actual projects already.
Speaker 1:Yeah, exactly.
Speaker 2:They need to be able to drop it in. It's not all greenfield projects.
Speaker 1:Yeah. And especially the next version of SST will like, not be scope will not be built on CDK. So that just opens up.
Speaker 2:Oh, for real, that's actually happening?
Speaker 1:Oh, yeah. For sure.
Speaker 2:Like, what does that look like? It's gonna be Terraform? No. Or something like it? No.
Speaker 2:What?
Speaker 1:It's gonna be Terraform.
Speaker 2:Are you doing your own thing? Like No.
Speaker 1:Not the same. No.
Speaker 2:What do you what is it gonna be? Do you not wanna say right now?
Speaker 1:Well, I mean, I feel there's only one I've eliminated all the
Speaker 2:Wait. You have Pulumi? Yeah. For real?
Speaker 1:Don't know
Speaker 2:anything about the inner I can't. I mean, that's the next thing I would have guessed.
Speaker 1:Okay. So we have we haven't we haven't made any firm decisions yet. But if I have to like say where I'm leaning, it's definitely definitely that way. Yeah. Reason being, we've all heard of Pulumi but no one's actually tried it.
Speaker 2:I've never used it. Yeah. Or if
Speaker 1:you've tried it, we've like barely tried it. And do you know why this is the case? Here's my theory.
Speaker 2:Not enough devrels.
Speaker 1:No. Weirdly, that's usually the problem. No. Sorry. That's never the problem.
Speaker 1:That's yeah.
Speaker 2:The opposite of Why has nobody tried it? Yeah. Go ahead.
Speaker 1:Because the from the moment you try to use it, it feels like a paid product.
Speaker 2:Oh, interesting.
Speaker 1:Like every step of the way. You go to their site, you go to their docs, you run your first command, the CLI, they ask you to put it in your email and sign up and make a password.
Speaker 2:Ah.
Speaker 1:So I think they have I mean, they're they've been around from before CDK. Yeah. They've been around for a long time.
Speaker 2:Can I tell you why I don't use it? Why? I just don't like the name. I really just don't like the name. That's it.
Speaker 1:That's funny.
Speaker 2:It's the only reason. It so it does can I you just tell me enough about the inner workings? It's like Terraform in that they just make direct SDK calls. Right?
Speaker 1:It's a it's like yes. It's like Terraform in that it can coordinate anything. But it's like CDK in that it's like a programming language base. You're defining stuff using a programming language.
Speaker 2:Yeah. Does it does it compile the types or to not compile, but does it turn into CloudFormation?
Speaker 1:No. It doesn't. It doesn't. No.
Speaker 2:Interesting. So we bypass the CloudFormation, change that hell.
Speaker 1:Yeah. Exactly. So there were a lot So with my experience of using CDK, there were a whole set of issues that I bring up all the time. There was a subset of those issues where I'm like, okay, this sucks, but I also see why CDK did it this way. Yeah.
Speaker 1:I was like, okay, I'm I'm I can be reasonable there. Then I went and tried Pulumi and I was like, oh, never mind. There was a good solution to literally all of these things. It is such a good product. It is insane.
Speaker 1:It's so I mean, I think the level that I appreciate it is because I like know some of these inner abilities that were like, so yeah. There were things that I perceived as really hard to solve that they solved elegantly. There were things that I wish that someone would do a certain way. It turns out they did it that way. It like it allows for some like really crazy things.
Speaker 1:And this is a really stupid example, but this example demonstrates why it's like the best infrastructure as code tool.
Speaker 2:But before you give that example
Speaker 1:Yeah.
Speaker 2:We're gonna take a break because I have to pee.
Speaker 1:Okay. Good.
Speaker 2:Come right back. Have to pee so
Speaker 1:bad. Nice.
Speaker 2:And that just seemed like a good cliffhanger.
Speaker 1:Be That's a very good cliffhanger. Good job.
Speaker 2:Okay. I'm very curious. Continue.
Speaker 1:So, here's a really stupid example. Let's say you want to create, an s three bucket And then based off of the generated ARN, you wanna create a lambda function for each letter in the ARN.
Speaker 2:Lambda function for each letter in the ARN.
Speaker 1:Okay. So really, contrived stupid example but Yeah yeah yeah. You could never do this in any of the traditional IAC solutions because all of them are built on this idea of you need to be able to synthesize a plan ahead of time and the plan gets executed. So in the in the CK case is confirmation. But you you can't know the plan ahead of time because that ARN might be generated by like during the process of Yep.
Speaker 1:Of synthesis of of actually deploying.
Speaker 2:So then you've got custom resources and all this hell.
Speaker 1:All this hell, exactly. Yeah. To like make up for the fact that you need to do something based off of what might happen.
Speaker 2:Uh-huh.
Speaker 1:But, Pulumi's approach is like, we're gonna give up the idea of a plan. So the downside is so they'd still have the concept of a plan.
Speaker 2:Can you roll back?
Speaker 1:Yeah, Pulumi's like, you can always fix things but Yeah. They saw the concept of a plan but if you're doing something like this, it's not gonna be able to tell you exactly what's gonna happen. It's gonna only only be able to tell you like as much as it knows. Mhmm. But it effectively creates like this I mean, the technical term is a DAG.
Speaker 1:Like there's a there's a graph of actions that need to happen and based off of an action completing, other actions happen. Mhmm. And it can execute that that whole thing. And at the end, you'll get the right result, but you don't know determinist I don't wanna say deterministically, like you don't know Yeah. Exact details of what's gonna happen.
Speaker 1:So there's all kinds of scenarios this unlocks like deploy this, if that ends up creating this other thing, then build your front end. And then once you've built your front end, analyze the metadata to figure out what routes you need to deploy and update your CloudFront behavior. Like you can do these really complex Yeah. Series of actions and it's really beautiful and elegant and it solves so like every single hack that we have, that fundamental problem, if you solve it, it gets rid of. So I was pretty excited when I saw that.
Speaker 2:This is it's yeah. It's high praise. I mean, like, if anybody knows what would make IAC frameworks good, it would be you. And I feel like that's like, you being impressed by something, That's that's a good what's the word I'm looking for? No.
Speaker 2:Hang on. That's a good, testimonial Signal? Signal. Thank you. That's the word.
Speaker 1:It's a good signal, but it's funny because they don't need my signal. Because who's the CTO of Pulumi? Do you know?
Speaker 2:I have no idea.
Speaker 1:The co founder of TypeScript. So it's not What? It's not surprise yeah, it's like
Speaker 2:How did I not know this? I didn't know this.
Speaker 1:So it's not surprising at all, right?
Speaker 2:Yeah. So are
Speaker 1:they are they
Speaker 2:doing well
Speaker 1:as a business? Because like, you're not surprised. Okay. So as a business, I think what they're doing makes sense for them which is they are head to head focused on Terraform. Any place in the world that's deciding on Terraform, they want to get in there and they want to be an alternative to Terraform.
Speaker 1:So they are just in that world of DevOps, people that like are really into infrastructure as code, people like big companies that that like, you know, are making decisions around this thing. But they're not in the business of bringing infrastructure to code to people that don't already know about it. Right? They're like just trying to Terraform establish a market, they're trying to now like, fight with them for Feed
Speaker 2:that market share? Yeah. Yeah, yeah.
Speaker 1:But the average person that's not already using infrastructure as code, they're never they're not being marketed to by Pulumi. There's like no on ramp for Pulumi. You have to like already be an infrastructure as code person. Mhmm. But that's where I'm like a little bit interested because I'm like, we definitely focus more on that kind of person where we trick them into using infrastructure as code because we promise these other things.
Speaker 1:But we think we can like take their stuff and hide it a little bit and like, kind of build on top of it and use it as a primitive and get to like a really high quality product. So that's like in like rallying around in my head as as a thing and I think realistically we'll support that as like part of what SSE is today so you can like have Pulumi stuff side by side of your existing stuff.
Speaker 2:Okay.
Speaker 1:Yeah. And that'll enable us to do so many things. Like, we love working with other projects. I love working with the Astro team and the Remix team and all of that and this lets us now work with the planet scale with like, planet scale or with Clerk or with whatever.
Speaker 2:Oh, yeah.
Speaker 1:Managed services so we can coordinate that stuff as well.
Speaker 2:So I was gonna ask, like, the migration path because it sounded like this would be a huge lift if it's like a big rewrite to use Pulumi, but it's gonna be like an incremental thing. It could
Speaker 1:Yeah. It could be. Like I said, we haven't even decided whether we're gonna use Pulumi. It's my initial research has me pretty excited. And yeah, so I think we we don't know what that'll look like, realistically, I think the simplest thing we can do is start to let you spin up non AWS stuff and have all that be powered by Pulumi.
Speaker 1:Yeah. And then we can kind of take it from there. But there's a lot of stuff that simplify like the import stuff, importing existing resources, like that is so much easier. So a migration path for our existing resources could could be done. Yeah.
Speaker 1:So it's a fantastic tool, just it has less downloads in CDK. Yeah, it's a shame. It's like, it's definitely what deserves more attention than it gets.
Speaker 2:CDK definitely markets to people who aren't infrastructure as code people, right? Like, they wanna be like, this is your first experience with infrastructure as code.
Speaker 1:Building stuff on AWS. Yeah, exactly. Yeah, so it just was even myself, had a bias against them for a while just like for no reason because it just seems
Speaker 2:The name.
Speaker 1:It just seems to me.
Speaker 2:Just I don't like the name. I don't know.
Speaker 1:But I like
Speaker 2:it more now that I know it's a good That's Yeah. That's cool.
Speaker 1:But I I like I looked at it and something about it felt so like not open source. Like Terraform feels so much and ironically, it ended up being not true. Yeah. Terraform feels like such an open source primitive thing. Yeah.
Speaker 1:Until they like changed that but
Speaker 2:Until they weren't.
Speaker 1:Yeah. Whereas, Pulumi from the beginning, they're like, try to get you into their like, use our cloud service to store your state and like kind of all that stuff so Yeah. It doesn't feel like you're using an open source thing and later you're optionally upgrading to their service. So I think that that's a good positioning thing that I think prevents them from getting much much bigger.
Speaker 2:Interesting. Well, I'm very excited to see all that shake out.
Speaker 1:I just I love SST. So if
Speaker 2:it gets better and faster and I can incorporate with outside services
Speaker 1:Oh, man. It's so fast. Got Gloomy is so fast. Yeah.
Speaker 2:I mean, you're getting rid of CloudFormation, like, there you Yeah.
Speaker 1:Like that I think that example, the bucket example I gave, I think I did I think I was doing create one bucket and create another bucket for each letter in the ARN. Yeah. I think it deployed in like under ten seconds or something. It was something in ten. Because SDK AWS SDK calls are fast for a lot of things.
Speaker 2:Oh, yeah. They are. Yeah. It's always amazing when you use, the CLI and you do something, you're like, wait a minute.
Speaker 1:Did did it do it? Or is
Speaker 2:that just like it started the the the process and now it's gonna take a while.
Speaker 1:Yeah. It's funny because some things in AWS are, like, stupid fast. I I can't think of an example right now but,
Speaker 2:like CloudFront. Yeah. Distribution. Yeah.
Speaker 1:No, not that. That's a joke. The opposite of that. Well, like, spending a new SQS queue or tearing one down Yeah. Takes, like, less than a second or something.
Speaker 1:But then other trivial things, like updating an API route, API gateway takes Takes forever.
Speaker 2:Yeah. I did that thing. Was That fun because I did the thing where I said something with confidence and it kind of fooled you for a second. That was fun. I wanna do that.
Speaker 2:Yeah.
Speaker 1:Well, at first, I wasn't sure if you misunderstood me and you meant that.
Speaker 2:CloudFront is fast.
Speaker 1:Yeah. I thought you were talking about the service being fast.
Speaker 2:Yeah. No. But, like, updating a CloudFront
Speaker 1:The thirty minute invalidations. Yeah.
Speaker 2:So I don't wanna get too in the weeds here because people don't care, but I care. CloudFormation claims, they're slow because they give you all this safety, this level of, like what's the word? Atomicity or something? I
Speaker 1:don't know. Yeah.
Speaker 2:Yeah. They give you this, like, ability to roll back safely. You're saying in Pulumi, you don't lose the ability to roll back safely.
Speaker 1:So that claim has already been like You can already be proved that it's not as intense as it seems because Terraform has been around forever and Terraform is bigger than CDK.
Speaker 2:That's true. It might even be No, they don't.
Speaker 1:It's bigger than CloudFormation. And they don't Yeah,
Speaker 2:they don't do all that. It's fine. Yeah, everyone's Okay.
Speaker 1:I I think this stuff is more like, let's say you update an IAM policy. IAM is global. The CloudFormation under the hood, they might ping every single region and wait for it to have replicated to every single region before they consider it done.
Speaker 2:Oh, I see.
Speaker 1:So imagine they're pinging every single region, they're only pinging once every thirty seconds and they're getting replicated. Yeah. So this like can this can might end up, you know, cascading into something So Yeah. Really Yeah. I feel like it just doesn't matter for most people.
Speaker 1:I don't know why they optimize for that that level of consistency but
Speaker 2:And they can never change, right? CloudFormation is what it is. It's never gonna like
Speaker 1:They're gonna change.
Speaker 2:Yeah. Suddenly twice as fast. Yeah. It's not gonna happen.
Speaker 1:Unless they give you a toggle to be like, you can deploy in fast mode or safe mode. Mhmm. And then like, do something speed. Wildly
Speaker 2:Yeah. How did we just spend the whole time? We've been, like, talking for an hour about all this AWS stuff. That's not normal for us.
Speaker 1:It's not normal. But
Speaker 2:I mean, I guess it used to be.
Speaker 1:It used to. I see.
Speaker 2:Yeah. Well,
Speaker 1:talk okay. Shifting It's topics. Your tweet about Please do. Are you gonna try not being a jerk anymore on Twitter?
Speaker 2:Oh, yeah.
Speaker 1:Yeah. Someone replied being like AWS FM Adam hit different. And I'm like, that is really funny.
Speaker 2:Oh, yeah. Well, it's very different. I feel like my Twitter, I've always sounded like a jerk in writing versus, like, if you meet me in person, like, I'm a Midwesterner. I'm pretty, like, low conflict.
Speaker 1:IQ.
Speaker 2:I wasn't gonna say IQ, but I'm I'm not, like, an intimidating person. I'm not like I'm not a jerk. I'm really not. Yeah. But in writing, for some reason, I always kinda sound like a jerk.
Speaker 2:But I feel like over the last year, I've definitely turned into more of a jerk on the Internet. And I blame you. I really do. I started I started lowercasing my tweets just like you. So I stopped using punctuation.
Speaker 2:I just looked like a jerk now.
Speaker 1:Yeah. Someone replied being like, I don't know, thought Adam was nice but I only have Dax to compare so You're nice by comparison. Okay, here's the thing. I'm not absolving us at all but Mhmm. I gotta be honest, like, it is kinda hard to be public and, like, just constantly get the kind of replies that we get, like, every single day.
Speaker 1:Yeah. It definitely has, like, a long term impact to some degree.
Speaker 2:It probably does. And I feel like I care less about what people think than the average person.
Speaker 1:Yeah. Even then, it's just annoying.
Speaker 2:But it's just annoying. It just, like, annoys me how many people can have the same bad take. Not that they're I don't know. They could say I have the bad take. But, like, so uninformed.
Speaker 1:It's not even the bad take. It's it's like a lot of times they the thing that is annoying for me is they read the worst possible interpretation or the stupidest possible interpretation of what I said Mhmm. And reply to that. And I'm just like, that's just it's it's annoying, it makes that's what makes me think that everyone's stupid, like, anytime I see that, I'm like, you're an idiot. If that's your way of perceiving the world, like, try to find the dumbest angle of what everyone's saying, like, that's yeah.
Speaker 1:You know, over time that just impacts you to some degree. And I feel like I've dealt with it better at different times in my life.
Speaker 2:You know you know who doesn't get these kind of responses though? It's people who just post like JavaScript tips and stuff.
Speaker 1:That's not true.
Speaker 2:Should we just do that? That's not true. Is it because we're making claims? I'm gonna stop making claims, I think. I think it just gets in people's they they get, like, really irritated when you make a claim of any kind, like, that you know something.
Speaker 2:If you're like, I know a thing and you should know it too, people just get real pissed real fast. So should we stop doing that? You're not gonna stop doing that. You're good at it and you actually have good things to say. I'm probably gonna
Speaker 1:stop. Think I mean, it's unfortunately, it's just the small minority ruining it for everyone else because let's say I post a tweet, let's say he gets like 300 likes, it probably will have like one or two annoying comments. That's one or two annoying people of 300 people that saw the purpose of it. Oh, that's
Speaker 2:a good point. I didn't think about the actual likes meaning anything. Yeah.
Speaker 1:That's a
Speaker 2:good point.
Speaker 1:So it's just a small group of annoying people.
Speaker 2:But Okay.
Speaker 1:Everyone. Try to be less annoying.
Speaker 2:I mean, me too. I need to be less annoying. I didn't but I wrote a tweet just the other day that I wasn't trying to be abrasive or anything. I was literally, I was just stating what I feel is a fact. And the amount of people that think I was doing it for for bait I have never written a tweet to, like, bait people.
Speaker 2:I don't even know what that means. Like, why why would I write a tweet to, like, make people angry? I don't do that. I write tweets that express what I think or feel, and I generally think I'm right as one does. That doesn't mean I want you to get angry and reply to it.
Speaker 2:Like, what is that? Why would I do that? I don't understand.
Speaker 1:People confuse the idea of something being provocative and making you think with something that they kind of classify as bait, especially if they don't like it. You cannot agree, but it's so provocative.
Speaker 2:They cannot agree. Yeah. And I don't mind the debates when they're intelligent and not just like, well, you're just a shitfluencer or whatever they wanna call you.
Speaker 1:And, like, you're just trying
Speaker 2:to farm that x money. Yeah. Because we make a lot of money on x. Could we stop that narrative too?
Speaker 1:Like That's so funny.
Speaker 2:Nobody is making a lot of money on x. That's just so dumb.
Speaker 1:And less and less every every time. Every time. My payments have only
Speaker 2:gone down. It's not nobody is nobody is writing, like, engagement bait so they can get a check from Twitter. I promise you. Well, maybe like the people who have, 20,000,000 followers. I don't Maybe they do, but I promise you, nobody here is able to buy their coffee with their ex whatever.
Speaker 1:I know. Yeah. It just doesn't it's it's like it just it's nice. It's like a nice little thing that shows up, but it really hasn't had zero impact on anything that I do. Exactly.
Speaker 1:But if anything, if it had an impact, it's like gone. I think it had an impact for like two weeks and then people were like
Speaker 2:Yeah.
Speaker 1:Okay. My favorite thing is, I definitely love in terms of there are some annoying replies that I love. I will say that.
Speaker 2:Oh, yeah?
Speaker 1:Yeah. I love when someone, explains something to me that is like my main area of expertise. Like if someone explained to me how like infrastructure is code worked. Like that happens all the time, which which I love because I'm just like, oh, this is so good. Yeah.
Speaker 1:Because like, it's such a it's an example of someone being so in the wrong and so indisputable. Yeah yeah yeah. And and whenever I see this, I love That okay. Yeah. You see what I have with other people too.
Speaker 2:Yeah. When it's when I know how wrong they are, it is kind of fun.
Speaker 1:Yeah. Like the other day, I forgot who. Someone from the React team posted on Hacker News, like in a comment and someone was like, trying to explain how React worked to them. Yeah. And I was like, I love I love when this happened.
Speaker 1:That's It's like it's like the best thing ever because like that person, it doesn't matter what the details are, like, they realize, they just know they fucked up. Uh-huh. Yeah.
Speaker 2:No. That's great. There's lots of those like people will screenshot when somebody, like, tries to explain something to, like, the creator of a thing. I love it.
Speaker 1:That's the best.
Speaker 2:I may have read my DMs. I'm sorry. I just I started responding to a DM in middle of our podcast. What has this become? Need to become a I need to get professional around here.
Speaker 2:This is I'm just getting sloppy.
Speaker 1:I don't think it's about professionalism. I think we're all of our brains are just eroding Distracted. Eroding. Stay focused.
Speaker 2:It's really true. But it's gonna get better. This podcast is about tomorrow and about how things are we got we got 30 more episodes until we hit a 100.
Speaker 1:And then what? Then we give up?
Speaker 2:Will it be like my Duolingo streak? Did I tell you at a hundred day streak, I stopped?
Speaker 1:So and it's so you. You, like, just get to So a place that is impressive and then it just turns off overnight.
Speaker 2:But not even impressive. Like, people have, like, fifteen hundred day streaks on Duolingo. I get to a place that's, like, the first milestone you could possibly hit. And then I'm like, well, I did that thing.
Speaker 1:Okay. Here's the thing, for most people, I think it's like because it's not like you stopped three days in. You stopped a hundred days in.
Speaker 2:Oh, sure.
Speaker 1:And it's kind of like for most people, it's like, well, if I put in all this work, I'm just gonna keep continuing it. But then you just you just abandon. You just don't care.
Speaker 2:I do. I just abandon. You put in more
Speaker 1:work than the average abandoner, but then you still abandon. You know?
Speaker 2:Yeah. No. I do I do things in three months stints. It gets three months of my attention. And then but look, the podcast, we've been going for a year.
Speaker 2:So this is not this is not dying at a 100 episodes.
Speaker 1:That's true.
Speaker 2:But what I
Speaker 1:was trying to
Speaker 2:say is we've got 30 more episodes to get, you know, good at this stuff. Like, professional.
Speaker 1:So we have to be good by a 100.
Speaker 2:Yeah. We really need to do, like, some segments or something. I had some ideas. We could just use chat GBT to give us ideas. Right?
Speaker 2:I don't know. I had some like, I I thought of a segment, what's more likely? Like, two different like, we have to come with some of these and we we do them to each other. Like, which of these two events and they can be totally off the wall and crazy and neither is gonna ever happen or they can be realistic scenarios, but which is more likely to happen? That'd be fun.
Speaker 1:That's cool. Where'd you get that idea from?
Speaker 2:Another podcast. That stole
Speaker 1:was gonna say I was the reason I was asking was, is this a validated idea? And it sounds like it is.
Speaker 2:Oh, yeah. No, it is. They they've been doing it for years on this podcast. It's not related to tech at all.
Speaker 1:Nice. That's a good that's a good one. Yeah. Have you tried the, the speech mode in, ChatGPT, the app on your phone? Yes.
Speaker 1:It's so good. I I feel like I'm Like in the the voice is so good and because like JCP is so good at speaking, it's Yeah. I really feel like like it felt magical.
Speaker 2:I know. We're getting there. This the AI stuff is is more impressive all the time. As far as, like, hype cycles go, I don't this is the first thing that I feel like is actually transformative. I mean, like, people didn't call the mobile revolution a hype cycle.
Speaker 2:Right? Like, everybody switched to phones. That wasn't a hype cycle. Right? It didn't get puffed up too early.
Speaker 2:Is it possible AI is just the next big thing?
Speaker 1:Well, the I don't remember the the phone thing. I think the phone thing was funny because like, nobody thought it was gonna be big. Like, nobody thought it was gonna be big and it kept getting bigger and bigger and bigger.
Speaker 2:And I guess it it was it before all these networks where no one was able to talk about it?
Speaker 1:Yeah. Maybe. Yeah. Maybe, like, it was hard to get a hype cycle going. Yeah.
Speaker 1:Had to put in the newspaper. Like, Twitter didn't exist.
Speaker 2:Yeah. No. Well, the there's something else that g p OpenAI did the combined GPT four model. Did you see that?
Speaker 1:Yeah. I've been using that.
Speaker 2:Yeah. Like it's got all of it built in. You just like you can just use the one thing. Pretty cool.
Speaker 1:One of my favorite things I like doing is, like jailbreaking chat GBT. Like, it's so fun when I, like, coerce it into doing something it doesn't it doesn't want to. Like, it's a you have like, find different angles to, like, trick it.
Speaker 2:I've seen lots of these two examples on the Internet. That's so something you would do.
Speaker 1:I keep trying to, like, give it pictures of me and Liz and asking it to do, like, funny things with them and it just refuses. But you can always get around it. You spend, like, five minutes, you can do it. You can always
Speaker 2:trigger Like, fine. It's never
Speaker 1:like that. It's always like, it like realizes that it should be doing it. It's always like It's always like, sorry, can't do this, sorry, can't do this. And it's like, oh, okay. You know, once you break Yeah.
Speaker 1:Of course. Of course. In that case, yeah, very exciting. I gotta
Speaker 2:use the voice thing more. I'm gonna play with that more. I used it like twice.
Speaker 1:Mee and Liz tried that prompt where, you ask it to help you flesh out a story, don't give you any suggestions, just help you like work through creating a plot and a
Speaker 2:Oh, I've not heard of this.
Speaker 1:I'll send you the product. Yeah. So it I mean, you you go for a walk so that's actually what they recommended. They said turn on voice mode, put in your AirPods and then ask it to help you flesh out a story and you can just talk back and forth with it and it'll like, you know, you'll do like this whole narrative with it. And it's it's pretty fun.
Speaker 2:That sounds entertaining. Yeah. Your video is off again. So like, I can't see your facial expressions
Speaker 1:with your words. Sounds like a you problem.
Speaker 2:It probably is
Speaker 1:a me problem.
Speaker 2:What is a me problem is it's over an hour and I do need to get off here. Yeah. That made me sound like I'm the busy pretentious one. I'm not any more busy than you.
Speaker 1:I think your approach is saying you need to pee whether or not it's true. I think that's a good
Speaker 2:Oh, no. I don't
Speaker 1:That's I know you're saying.
Speaker 2:I should yeah. Yeah. Oh, I gotta pee again. Yeah. No.
Speaker 2:People will know it's only been fifteen minutes. They'll, like, call my doctor.
Speaker 1:I feel like I've seen you pee, you need to pee twice within an hour, It's at definitely happened.
Speaker 2:If there's caffeine involved, like, at Reinvent, yeah, you would have seen that
Speaker 1:for sure. Okay.
Speaker 2:Alright. Alright. It's been good. Thanks, Dex.
Speaker 1:See you. See you.
