Serverless is Dead
Hey, Dax. Did you see the Chris Moon's tweet? Serverless is dead? Yeah. I saw the tweet.
Speaker 1:I saw the whole thread. I spent a lot of time thinking about it. I felt hurt. I felt confused. I felt angry.
Speaker 1:I felt sleepy. Like, I thought about it a lot. So, yeah, it really got in my head. And the whole weekend, I was like, is serverless dead?
Speaker 2:Yeah. I've kinda felt like everything I thought was true is now false. It just shattered my entire world view in technology. And it so for some context, Chris is, one, a great guy. So, Chris, if we say anything that sounds like we think you're wrong on this podcast, because we might say something that we I don't know.
Speaker 2:We'll see. You're a great guy. We love you, Chris. But Chris is, like, involved with startups at AWS. Right?
Speaker 2:He, like, he does a lot. He works with a lot of startups from the AWS side. So he sees what people actually use in startup land, And, like, his whole thing is basically, like, startups don't really use serverless or at least not pure serverless. Maybe we should define pure serverless because that is an important note in this whole conversation.
Speaker 1:Yeah. You're right. He did say pure serverless. So the idea behind pure serverless is you're using managed services everywhere, you don't have something that's running all the time, like, you know, a database that's running all the time that you're being charged for or a VPC that has like a managed managed NAT in it that's running all the time. So you're only paying for when your system is doing something.
Speaker 1:And, you know, it's obviously a spectrum, but we're the idea is you push more and more towards not having stuff running that is being built.
Speaker 2:Yeah. I just the moment I have to create a VPC or I guess use the default VPC, I'm just pretty upset. So PureServerless is sort of like, I thought, just a default, a way of operating when you're building any new products, especially for startups. I know that that's not the case. Like, I know having worked with a lot of startups, they're mostly still building Rails apps.
Speaker 2:Like, I I get that. But I thought, like, if you were in serverless, you were pretty well defaulting to pure serverless. That was just, like, assumed in my mind. So do you think there's a lot of, like, this hybrid stuff out there? Is that what he's kind of implying?
Speaker 1:Yeah. I I think I they're deaf we see the same thing. We see container workloads. We see RDS clusters in a VPC with just, you know, everything else is serverless, but database isn't. It's definitely common and maybe even the norm because it's hard to go pure serverless.
Speaker 1:At least today, you have to jump through a lot of hoops. And we try to like, we spend a lot time thinking about that and trying to make that easier. But I I think the data is probably the data he's referring to is is correct. I think today, it probably is pretty hybrid.
Speaker 2:So is there a future for Pure Serverless? Is he is he right? Is it dead and it's not coming back? Because he implies that it's just not ever gonna be a thing.
Speaker 1:Yeah. I think the reason I was so disappointed is I agree with him on the current state of things, but I thought everyone was aligned with where we were going. And the, you know, the ideal state being okay, today, it's like a mix of stuff, but eventually it's gonna be pure serverless. And this was actually a vision that AWS sold me, you know, I was not doing things this way at all. AWS came in and started talking about the advantages of serverless, like a serverless mindset, and making things more and more serverless.
Speaker 1:So I thought we were all kind of going the same direction of, hey, like, this is just gonna get better and better, and things are gonna be more like this. So it was a little weird for me to hear, like, that dream is dead. We're kinda settling on, okay. It's gonna be a mix mix of a few things because that's not where I wanna be, and it's not, like
Speaker 2:Right.
Speaker 1:You know, where I was hoping we'd go.
Speaker 2:Yeah. And Chris also I mean, he's he's, like, very involved with startups at AWS now. But before that, he was, like, the big face of serverless at AWS. Right? What was his title then?
Speaker 2:He was, a developer something relations. Like, that's all he did. Right? Was serverless.
Speaker 1:Yeah. So I'm, like, I'm really curious what happened or the vibe I got from that thread was it seemed like he had been patient for a long time, and the patient kinda gave out. And now it's, you know, a little bit of frustration. So I'm kinda, like, wondering what he was hit over and over with. Yeah.
Speaker 1:I think some of the things he said, he I think just to kinda summarize what the thread was, Alex debrief pointed out that to use DynamoDB DAX yes, DAX. You need a VPC, and that's a problem. And then Chris responded saying, can we stop bringing up this VPC issue? You're gonna end up with a VPC realistically, so give up on not ever having one. And he and his his argument was, well, the data shows that most people have VPCs.
Speaker 1:And I'm not questioning the data, but I think whenever you're trying to make a new paradigm or something better, the data is gonna show the opposite because that's the status quo. That's what everyone's doing. Yeah. Most of those companies are probably not fun to work at. So, yes, that's what the data shows.
Speaker 1:But to me, what the data is actually showing is serverless is still too hard, which is why people are still using these VPCs. So it's kind of like, you know, AWS has all this great stuff. It's not great to it's not, like, super easy to use or discoverable or people don't really know about it, so they're not using it. And then it's kinda now it's kind of like, well, what's the point of improving it if no one's using it? So it feels a little circular.
Speaker 1:Think we just need to keep pushing on getting people to use it. And eventually, you know, we'll get we'll get there.
Speaker 2:Yeah. And that I think that's the biggest bummer is the feeling that if it's coming from AWS, that, like, that impacts product decisions roadmap. I mean, they're so driven off of what customers are asking for and what customers are needing and using. So to think that that trend I I remember it was Chris even. I remember an interaction on Twitter, like, a year ago.
Speaker 2:I think it was leading up to re:Invent last year where some sort of similar thread was going on, and I sort of asked Chris on Twitter. I'm have to find it. Like, is it feels like serverless is slow moving. Like, we're we're just having a hard time making progress as a community. Like, it's just not that it's not moving in the direction and the speed I would think.
Speaker 2:Like, is there any chance that AWS doesn't prioritize it? It was something exactly along these lines. And he said that AWS was, like, the biggest customer. Like, internally, they use so many serverless offerings for things that they build. So that it was, like, just on that front alone, a huge priority.
Speaker 2:And and to your point, like, we've had at every re:Invent, Sheen Bristles is up there, like, talking about LEGOs transformation, and you just kinda felt like that was, you know, the future and that every every re:Invent would be seeing more and more progress towards a more serverless future. So it's all very depressing. If if now AWS is like, internally, people who are big champions of serverless are frustrated that the adoption's just not there and they're sort of giving up and saying it's gonna be this weird mishmash of using services that aren't. Because I just hope, like, we're just a few things away, like search and, like, they're gonna hit a couple more of these buckets, and we're gonna have everything we need to build anything you can imagine with serverless first stuff. But, yeah, maybe it's the tides are turning.
Speaker 2:I don't like it. I don't like it.
Speaker 1:Yeah. And to be fair, it has been a while. You know, serverless has been around the corner for seven years maybe. I don't know what the exact timeline is, but I know Jay and Frank have been doing it for a very long time. And Yeah.
Speaker 1:They've always been hearing, oh, this is, like, just a matter of time before this is a thing. So I can see why, you know, as time has gone, the energy behind it has slowed down. Simultaneously, I think there's a few interesting things that are happening. Now within AWS, it does feel like they've slowed down launching, like, really good serverless offerings. Yeah.
Speaker 1:I think RDS is a big one. Serverless v one came out. Serverless v two came out. Both of those are not really serverless. They're kind of just auto scaling.
Speaker 1:Serverless? So, yeah, it feels like AWS is kind of polluting that word a little bit. And some this new Neptune thing also is not really serverless. Redshift serverless is, like, kinda serverless, not really.
Speaker 2:Yeah. Oh, that big batch that they launched last re:Invent, there were like four of those analytics offerings that got serverless options, and they all were kind of like a mixed bag of, is that serverless? Like, how managed is it?
Speaker 1:Definitely feel like they're they're diluting it. So from the AWS side, it feels like, for whatever reason, it's not there's no, like, new Lambda type thing coming out where it definitely a 100% is serverless. But then if you look at the market outside of AWS, there are so many companies coming out with Yeah. Truly serverless offerings. I think I would trace this all the way back to Snowflake.
Speaker 1:Snowflake recognized, yes, Redshift exists, but it's hard to deploy because it's a managed, you know, you're managing nodes. So they made a serverless data warehouse, and that was a long time ago. Now they're like a $100,000,000,000 company. And since then, there's just been, you know, PlanetScale, serverless MySQL, Upstash, serverless Redis, serverless Kafka, serverless messaging. Like, just a ton of companies that are really they kinda picked up the torch, they're running with it now.
Speaker 1:Yeah. So it feels odd that AWS is slowing down when the interest from, you know, new companies is higher than ever.
Speaker 2:So maybe the yeah, maybe the third party ecosystem is enough. And maybe maybe this is just part of like, AWS isn't great at building the vertical, like, developer experiences. Is serverless just sort of a flavor of that? Like, is this really going to take the the community building on top of the more traditional primitives that AWS builds and launching startups that survive? Like, the ones that survive will give us all that we need.
Speaker 2:Because, yeah, PlanetScale is looking really good these days. And I think from a serverless perspective, I'd choose it any day over Aurora serverless, v one or v two. If you need Postgres, I guess, then there's Neon and I don't know. Like, there's some companies sprouting up for all these things. Different degrees of serverless, I guess.
Speaker 2:But but that is all that serverless is, I guess, is just sort of, like, managed services. It doesn't have to be all in your AWS account. Right?
Speaker 1:Yeah. And I I think at this point, any new dev tool type company that launches, if they don't have serverless pricing, it's almost to me, it feels like dead on arrival. So I think the expectations are there.
Speaker 2:I I'm building I'm building something with serverless, and I wanna test like, this is a dumb idea I had just kinda over the weekend. But, like, thinking about the limits of what you can build with serverless in terms of like, a lot of people wanna on the container side oh, that's another topic I wanna hit. We'll do that after we take a break. Like, on the containerization side, people, like, wanna throw rocks at serverless and say, like, cold starts and does it really scale? Like, that's kind of the shtick of serverless is that it's on demand.
Speaker 2:It's, like, all elastic. You're never, like, provisioning anything. It's auto scaling. But we all know that there are actual limits in your AWS accounts for how many, concurrent Lambda functions can be executing, that sort of thing. So I'm trying to build, like, what is the biggest scale thing?
Speaker 2:I'm using, like, a Twitter clone. Like, with serverless only things, how how many tweets per second? How many transactions per second can we push with serverless? I'm doing it this week, so it's kind of in line with what we're doing what we're talking about here. Like, what what is serverless good for?
Speaker 2:What is it not? My thought was that it was good for everything and that it's just the future. We're still building the developer experience. That part's not there. But, like, from a runtime perspective, from, like, your app actually living in the cloud, felt like there are no limits to, you know, what you can build with serverless.
Speaker 2:I don't Do you have thoughts on any of that?
Speaker 1:No. I was gonna ask, what kind of constraints are you putting on? Is it all just gonna be in a single AWS account?
Speaker 2:It's all gonna be in a single AWS account. It'll be in separate regions, which I think is fine because Twitter is global. So, like, if I'm trying to mimic a Twitter API, having it spread across six regions, it it sort of, like, represents a global API that Twitter is. But that gets around some of the limits, like the per region, concurrency limits and stuff.
Speaker 1:Oh, are the Lambda function limits per region?
Speaker 2:I don't think the Lambda function ones are, but I know, like, the app sync. So I'm using app sync.
Speaker 1:Oh, right.
Speaker 2:And, like, app sync operations, so mutations Mhmm. And queries. Those are per region. So I'm I'm going through the process trying to, like, increase all the limits and get all the quotas upped for this experiment.
Speaker 1:But yeah.
Speaker 2:I don't know. I think you can build anything with serverless. And I think, like, Simon Wardley, we've mentioned on a on a past podcast, like, that the the people who can't build with serverless should be the edge case, and I hope that's still true. I hope we're not gonna be super disappointed at re:Invent because AWS doesn't care anymore.
Speaker 1:Yeah. I mean, it's a cool idea what you're doing because I think we all talk about those benefits, but very few of us actually, me myself, like, this is a space I operate in. I've never actually pushed AWS to its limits to see what you actually run into. Because like you said, it probably won't be that DynamoDB fails. It'll probably be this, like, weird bottleneck has, like, some rate limit that you're gonna
Speaker 2:hit. Yeah.
Speaker 1:So that'll be interesting to find out, like, what actually stops you. Yeah. I'm assuming you are using Dynamo for this.
Speaker 2:I am. Yeah. I I wasn't gonna try this with a SQL based database. I just don't think it makes sense from, like, a 100,000 tweets per second perspective. But if you think about even just 10,000 tweets per second, Dynamo has like an 80,000 on with on demand pricing, an 80,000 write WCUs.
Speaker 2:What does that stand for? Write capacity units. So per second, 80,000. That's a, like, a soft limit. It can be raised.
Speaker 2:But, like, one tweet being added to a database ultimately has, like, a lot of downstream things also being written to the database. So you have to write those to everyone, timelines and all that. I mean, I think that's the best architecture. So I don't know. It's gonna be really interesting to see, like, what falls over fastest, what what can't scale, because some of these sources has hard limits.
Speaker 2:Like, you can't do more than 5,000 secret value reads per second in Secrets Manager. So can I even use Secrets Manager? Those types of questions, it's gonna be fun to see.
Speaker 1:Yeah. Alright. So you're actually gonna do the feed part too. So not just tweeting, you're actually gonna be
Speaker 2:posting gonna like, the whole premise is gonna be, can it sustain 10,000, a 100,000 tweets per second with people still fetching their their timeline in a performant Like, basically mimic as much of the core Twitter functionality as possible. I'm not thinking about DMs or anything like that, but I don't know. Just this whole, like, Elon bought Twitter and everyone's upset, and then you you just there's the fever pitch around Twitter clones is even higher than ever. I just started thinking about, like, could you build a could a single engineer build a Twitter clone on serverless tech in AWS? What would it cost to run it?
Speaker 2:Like, just kinda getting into the cost and pricing conversation with serverless. Yeah. I don't know. I'm mind is going in a million miles an hour right now.
Speaker 1:Yeah. I mean, this is also funny because I think it was Tomas that tweeted this. He was like, there's so many people on Hacker News that could build Twitter in a weekend, but we don't have any but we don't actually have any alternatives. Yeah. So you're the guy that's actually gonna try to build it in a weekend instead of just just talking about it.
Speaker 2:Yeah. No. I'm gonna actually try and build it because I'm gonna I'm gonna try to make a YouTube video this week it. I mean, just the process and what I learned. But it's really more like testing, I think, the serverless limits.
Speaker 2:Like, what what can you do with serverless? Because obviously, if you start running containers and things like that, it gets easier to manage. More expensive, probably, but
Speaker 1:Yeah. Yeah. It is good to test this. I think just because I'm constantly just thinking about the people that are cynical to serverless, I know what reaction you're already gonna get. You're gonna get Yeah.
Speaker 1:So even if it all works, people are gonna say, well, like, no one has that scale. So who cares about that? I feel like one of the things that get missed that gets missed about serverless is, yeah, of course, top end. Well, we'll see. Maybe not, of course, but seemingly top end.
Speaker 2:Might all fail. Yeah.
Speaker 1:High scale, it'll do it. But that's not really the point. Because yeah, like, none of us in practice have really, really hit that. So it's fair to say that's not that important. It's more that usually when you're building something, every month or so you usually run into some kind of minor scaling limit, it might just take you look at data solve may take two days to solve.
Speaker 1:But it's pretty disruptive and messes up your roadmap, messes up your day, mess up your momentum. And for me, serverless is more about the fact that you're not going to run into those bumps, even if you're not ever going to hit like some crazy top end scale. Even at small scales, you run into those bumps with the typical architecture. So, yeah, that's why I I've never really cared about the top end. I've never had to had to do that.
Speaker 1:But, yeah, I am still curious to see if it actually works. Maybe AWS banks on nobody actually need to.
Speaker 2:Yeah. Yeah. It's kinda like testing some of these quotas that are published. Like, I'm I'm just curious what what starts falling over. But I know, like, I know the services that will be involved.
Speaker 2:So it is just a handful of services that you have to think about and what quotas will be hit. I don't know. This is good time. Let's take a break. I wanna read some chat.
Speaker 2:Jay Station. Hey, Jay. Having looked at hundreds or perhaps thousands of job posting over the last few months, I can say that pure serverless does not show up at all. However however, feel like it's still just emerging. Engineering leadership is old.
Speaker 2:People calling in shots aren't there yet. Could be a decade
Speaker 1:or more. Oof. Knife to the chest. Decade.
Speaker 2:Decade. Ouch.
Speaker 1:I don't have that much focus for anything.
Speaker 2:What's that? To oh, yeah. To, like, keep pushing the boulder for a decade.
Speaker 1:Yeah. The the irony there is with every new emergent technology, typically startups adopt it first. But Yeah. It's weird that, like, Lego and these, like, huge companies are the faces of serverless versus, you know, like, some unicorn that was founded two years ago.
Speaker 2:That is something that's interesting to me is, is serverless more adopted from an enterprise standpoint than it is in the serverless space? Chris is in the serverless space or I'm sorry, in the startup space. Is this just like it's gonna take enterprise pushing this boulder forward? Are they gonna see the benefits? It seems unlikely, but
Speaker 1:I'm wondering if that's kinda what's messing it up because I think you're right. There has been a lot of adoption enterprises. That's been the face at least when AWS showcases stuff, I typically see these big companies showcase their how they're using serverless. So, you know, AWS being very customer centric, as they say, they are probably very focused on these companies. And of course, these enterprise companies are not pure serverless.
Speaker 1:They can't be. They
Speaker 2:have Right.
Speaker 1:Potentially decades of systems that they're still dealing with. So their product energy might be going to supporting the needs of enterprise. And it makes sense just from, like, paying attention to dollars thing. Yeah. So I'm kinda wondering, do they just suck all do a big enterprise just suck all, like, the product development energy and startups are less interesting for AWS because it's way more of an investment and way more of a long term.
Speaker 1:And like like Jay just said, like a like a ten year thing.
Speaker 2:Yeah, that's an interesting thought that, like, what is the breakdown at AWS in terms of people focused on enterprise versus startups? Like, mean, obviously, the majority of their business is enterprise, and we've heard it before, even last Reinvent. I remember some of the commentary after Reinvent last year was about how it's very clear AWS just cares about enterprise, and the smaller use cases aren't as interesting to them. I just it's it's weird to me because I think of AWS as being this best place to build your startup. Is that just me looking through AWS glasses?
Speaker 2:Like, I just everything is a nail because I know there's a hammer.
Speaker 1:And I feel like every company I know eventually ends up there. Maybe that becomes less true over time, but I still think it is true today. So you might, at least today, you might as well start there. But like I said, with a lot of these services being unbundled and these other options that are serverless and and better than AWS offering, I think it's more and more likely and more and more possible to glue together a bunch of non AWS services and get a better outcome than with an AWS. Yeah.
Speaker 1:So I don't know. I think that feels that the direction it's going, I don't love it. But I think the worst part about all this is, at least for me, it's it always always wins no matter what. Right? So I'm talking about all the ways that they're disappointing my expectations for them.
Speaker 1:But it doesn't matter because the companies that are serving my needs are paying AWS. So it's like, yeah, I get why they're focused on enterprise because those are the deals that only they can win. And then the other deals, other people will win, but, you know, AWS will get their will get their cut.
Speaker 2:Yeah. And it just makes our situation building things more fractured and disparate. Like, we're having to use more and more pieces. Like, it it makes it feel less likely that we're gonna get a really good serverless search offering from AWS. We're gonna have to use the thing that the community builds that wins out, which is there I guess, is it Algolia?
Speaker 2:Or what is the, like, winning search?
Speaker 1:I mean, last I checked, the two options are Algolia and I'm blanking on the other one. I don't know if you remember. TypeSense, which is open source. They have a hosted version.
Speaker 2:Okay.
Speaker 1:Algolia is extremely expensive. TypeSense is not serverless. So I don't feel like there's a clear winner there yet. But yeah, I've heard nothing from AWS about the search thing and people constantly bring it up. I know.
Speaker 1:I'm just happy
Speaker 2:with Elasticsearch or whatever it's called now? Clouds what is it? Cloud search? Open search.
Speaker 1:Open search. It's open. But it's in a VPC, so it's not very open.
Speaker 2:No. Right. Yeah. Yeah. I don't know.
Speaker 2:It's a bummer. I I just I feel like the more you can leverage manage stuff, the more leverage an individual developer gets. And I guess that's still true if I'm using planet scale versus something built into AWS, but it does just add more and more, like, complexity and risk, I guess. Like, I always feel bad building on any startup that's like a series a startup. Mhmm.
Speaker 2:You build on their thing, and you know the success rate of it lasting the next decade is, like, whatever it is, 50 less. I don't know.
Speaker 1:Yeah. I mean, I've been through that cycle enough times where I've been really excited about something and I've gotten a company to use it and it just died. So at this point, yeah, I I agree. Like, there's a level of responsibility you have when making these decisions.
Speaker 2:Okay. So serverless is dead. Is that the conclusion?
Speaker 1:Feels that way. But maybe the the hope is in, like, these new companies. They're all doing pretty well. I feel like, like, PlanetScale is less than two years old. They're shipping accuracy.
Speaker 2:What they've raised, but they seem to be doing they're good at things.
Speaker 1:Yeah. Upstash is also Upstash seems tiny, but they're shipping so much as well. They have four services now. So, yeah, that might just be the future. And I think that lines up a little bit.
Speaker 1:I think we've talked about this before where, you know, a lot of people build their app on Next. J s deployed to Vercel full stack everything. Then eventually, they hit a pain point, whether it's something asynchronous, whether it's a background job, whatever it is. And at that point, I think we try to get them to come to AWS. But that is, like, a big step, especially when there's always perceptions around it, whether they're true or not.
Speaker 1:It might be a much more incremental step for them to use, oh, here's like this dedicated jobs API that can let you run background jobs. So that might help people step by step make their application more complex than going all like admitting, okay, we should have been on AWS and going all in.
Speaker 2:Yeah. It just it does feel painful though when I didn't realize PlanetScale's only two years old. Like, that's that kinda blows my mind. They're just they're shipping some really cool stuff. And, like, they everything they do from a marketing perspective is very polished.
Speaker 2:They seem older than they are.
Speaker 1:Yeah. I think just a lot of experienced people at the company that have done this before. And a plus are building on top of a technology that's much older than two years. Right? A decade plus.
Speaker 1:So I think for them, it's they're experts in that technology and their task is more exposing that to the rest of us who aren't. So I see why they're shipping fast, but it's still impressive all around.
Speaker 2:Interesting. Okay. Well, I think that's everything I have to say about Chris Moon's kicking this off with that tweet. Yeah. Did you have anything else?
Speaker 1:No. Cool. See you, Dax. See you.
Creators and Guests
