Looks pretty interesting. I presume the boom in cloudflare workers and other hosting platforms has been a forcing function to make running a web app more “normal” looking with their lambda offering.
Lucasoato 5 hours ago [-]
I had the same impression that this is an answer to Cloudflare workers success, but are AWS Lamba web functions hosted in containers or WASM?
chrismustcode 5 hours ago [-]
MicroVM which I am presuming are firecracker
squiwardStan5 5 hours ago [-]
I believe standard firecracker microvms.
julianwood 2 hours ago [-]
yes it is. hardware virtualization for isolation.
nemothekid 5 hours ago [-]
Rust + Axum has sort of supported this with a middleware (https://github.com/lazear/axum-aws-lambda). Essentially you write your API as normal, and then when you build for lambda you enable a feature that does the special handler for you.
This + Function URL + Cloudfront, and you end up with an API entirely hosted in lambda. I have a backend that does is hosted like this that does pretty complex work (image/video uploads, image conversions, on top of the standard auth, database connections, etc). It even renders a frontend with HTMX/Alpine.
Even with all the bot traffic I receive, at almost 15 million requests per month the cost is under $10/mo (median latency is roughly 90ms). Cold starts are a non-issue with the Rust runtime.
remify 5 hours ago [-]
I've built 2 applications using Lambdas as a backend and I love how simple and it is. It cost literally 0$ (the lambda part).
The frontend is hosted on cloudfront, the backend is a lambda and you basically just pay for the database and certificate.
I don't want to manage a server or even a docker image. I just want my code to exist there so I like the idea
holler 3 hours ago [-]
> I don't want to manage a server or even a docker image. I just want my code to exist there so I like the idea
agree, unfortunately in practice it's more common to throw the api behind k8s with containers
torginus 5 hours ago [-]
Something very similar to this was already possible via serverless-express for node:
This wraps Lambda's 'handler' function and allows express to run on top of it, essentially you get a node web server with no persistence between requests which usually isn't a good thing to rely on, but you can also kinda have that with persistent tcp connections.
Of course since this is not real HTTP, things like WebSocket aren't available, but it's good enough to build a CRUD backend or two.
Similar libs exist for other runtimes as well.
bob1029 12 hours ago [-]
I strongly recommend never using functions as a service unless you genuinely have some client edge compute requirement that absolutely cannot be solved via some other means.
Everything about running software is made more difficult this way. You might think it would make certain infrastructure aspects easier, but it just changes the semantics of the same problem. It's the same quantity of mess to clean up, but now it's in these really weird piles that are often much harder to access.
> up to 128 concurrent requests per environment
I am fairly confident I could support more capacity with a single t3a.nano instance.
badatnames 12 hours ago [-]
These one-sided arguments come up every time there's a Lambda thread. They're useful and easily cheaper almost anywhere you have sparse compute or RAM requirements, even for heavyweight tasks like transcoding. A t3.nano meanwhile can simultaneously be overpriced ($3.50 for an idle task) and a bottleneck (30Mbit/s sustained network) in just as many places as a Lambda could be applied incorrectly.
ryandvm 11 hours ago [-]
They're also good for extremely bursty loads. I use Lambdas extensively for a radar imagery API that delivers the images for an Android live wallpaper that shows weather radar.
That particular use case can go from hardly any traffic at all to a hurricane of traffic during something like, well, a hurricane.
For consistent processing loads, I think I could save money by just running on dedicated EC2 instances, but you cannot beat the instant scale of Lambdas.
There's nothing particularly special about Lambda's scaling. You can do the same with GCP Cloud Run[0] or Azure Container Apps[1]
Specify the maximum number of allowed instances and maximum concurrency per instance. Optionally configure more targeted scale rules. It will scale it from 0 - n as needed.
So long as you don't count >2 orders of magnitude difference in scaleup time as particularly special..
CharlieDigital 10 hours ago [-]
That's not how it works since each instance can handle multiple requests (up to 1000 per instance). So you may set this to scale at say 75% load level.
Indeed if you configured each instance as a single threaded handler, you will not have fun.
badatnames 9 hours ago [-]
That's only how it works when misapplying mean() to characterise application performance. In your scenario, every request is hitting either an overloaded instance, or when scaled down, a 20-30 second wait for spin up, better hope they're neither interactive nor some jobs configured with a short timeout. The mechanisms are so completely different they're hard to even make meaningful comparisons between
CharlieDigital 8 hours ago [-]
> In your scenario, every request is hitting either an overloaded instance,
Why would this be the case? Cloud Run containers and an app runtime like .NET, Java, Go, Rust can inherently be multi-threaded, multi-connection. It can handle up to 1000 concurrent connections per container using just threads and async (assuming sufficient size). Any workload that requires hitting the DB or backend service (S3, etc) is going to I/O bound and sitting in a future most of the time anyways.
The container instances auto-scale by default to a configured n (0 - n) with the option to define specific scale points. If I'm sensitive to high bursts, this can be a matter of scaling at 50% capacity and a warm instance is available while the application still has headway.
Cases where this falls apart: DDoS in which case, I argue that 1) it is better in any case that it falls apart rather than absorbing that type of traffic spike, 2) such traffic is better filtered upstream in any case in an API Gateway + WAF.
I have worked with both types of solutions for different use cases (was invited presenter at Azure Serverless Conf[0]), but my take nowadays is that serverless containers are almost always a better approach since it provides much more flexibility.
At which point we're paying double just to work around cold start latency, which was the original point.
CharlieDigital 6 hours ago [-]
Cloud Run doesn't bill when idle; it only bills when active.
So if you're at 50% capacity with lots of idle time, it can scale to 0.
Scale to 0 doesn't mean "cold". One easy way to keep both Cloud Run and Container Apps "warm" but functionally scaled to 0 for costs is to just ping it every few minutes with Cloud Scheduler set up on a cron pointing to a health endpoint.
rahimnathwani 5 hours ago [-]
If you scale Cloud Run to zero and then send a request, won't you immediately wake it up and be charged until it powers down in at least 15 minutes?
CharlieDigital 4 hours ago [-]
When scale to 0, it only bills when actively serving a request. BUT, the instance doesn't go cold immediately; you are only billed while it is serving even if the instance is active.
If there's a 10s gap between requests, you are not billed for the 10s gap, but it is still effectively hot; instance does not stop. Request that comes back in 10s later does not have a startup penalty because the instance is still up.
How do you leverage this to build a high throughput system with no cold starts? Just prime it with a short single request (e.g. `/health`) before instance is stopped.
rahimnathwani 4 hours ago [-]
I assumed you were talking about instance-based billing (what I've always used), but I think you're talking about request-based billing?
CharlieDigital 3 hours ago [-]
Request based billing indeed.
greatgib 9 hours ago [-]
I often hear this argument that it is good for bursty load, but for real, there are a lot of inconvenient limitations.
For exemple, you don't really control the number of concurrent instances or the scaling except if you pay for very expensive provisioning.
So you might have to wait for 10s or more to have a new lambda instance spawning up and replying to your requests. And you might have to have your function finish in 15 minutes, deals with the strange concurrency and scaling rules lambda is based on.
joshstrange 11 hours ago [-]
I agree and disagree with this. I run multiple software platforms that are incredibly bursty, to the level of "no traffic for 10 months out of the year, small traffic of 1-1.5 months, medium for <1 week, and high for 1-2 days" (event-based, like a physical in-person event, not events in the software sense). While I agree lambdas are fiddly and I have some reasons to move off them, they also are kind of perfect for my use case from a billing perspective. Even a t3a.nano would have cost more and not scaled quite as easily as I needed.
All that said, there are times I regret my lambda decision and would prefer just a EC2 host even if it cost more. Deployment would be easier/simpler. I've spent considerable time to improve my deploy process and lower the deploy times but it's still ~2-3min from tag -> deployed which I know I could do faster on just a VM. Also everything gets harder in a lambda, packages/binaries, logging, monitoring, etc.
vermilingua 12 hours ago [-]
I seem to have had the opposite experience. Using Lambda (I can't speak for many of the other FaaS platforms) is so straightforward to me that there need to be some strong arguments for hosting code/containers in my projects.
My guess is that the IaC is the special sauce that's missing; you do need to go all in with some IaC solution to make using Lambda easy, but it doesn't take a lot of it to make it slick.
Cost/compute is obviously strictly worse for FaaS, but most web workloads don't _really_ need to be optimal.
CharlieDigital 10 hours ago [-]
If you haven't tried Azure Container Apps or GCP Cloud Run, I would strongly recommend trying them at least once deploying any standard web API app in your language of choice, just for your edification.
No adapters needed; just write a web API and ship it. Public URL gets assigned and its ready for use. Code works locally with no special adapters, emulators, or tooling needed. Need to support another cloud plat tomorrow? Ship the same container. Moving it to self-managed K8s? Same container.
ekorondy 10 hours ago [-]
[flagged]
hmokiguess 11 hours ago [-]
> unless you genuinely have some client edge compute requirement that absolutely cannot be solved via some other means
You make it sound like these cases are rare, not sure what kind of work you do but there are markets of all sizes in many different use cases out there.
I feel like what you wanted to say was "I haven't found a need for them" as opposed to "Stay away from it"
Maybe you're trying to say something else we don't know because we don't have your backstory either, did folks at your work tried pushing it for scenarios in which they became a problem? Share that instead as it brings a lot more value to the readers, otherwise it sounds like you are trying to go for a weird flex.
bbg2401 10 hours ago [-]
They said what they wanted to say, and it was quite clear. The person who commented suggests we don't use client-edge compute unless we have a specific need for it, i.e., don't treat it as a default due to its apparent simplicity up front.
You've taken it in a rather odd direction with the request for a framing more palatable to you.
hmokiguess 3 hours ago [-]
We don't have to agree, and I guess me understanding it differently could also be like how you understood me in a way I didn't intend to
CharlieDigital 10 hours ago [-]
I would agree with OP with "stay away".
The Google paper Towards Modern Development of Cloud Applications[0] does a great job explaining why.
TL;DR the case they make against micro-services architectures (which Lambda encourages):
1. It hurts performance
2. It hurts correctness
3. It is hard to manage
4. It freezes APIs
5. It slows down application development
For even a moderately complex system, spreading logic out over multiple discrete runtimes doesn't make sense; instantly run into each of the 5 points. The only times I've found functions useful is truly single-point behavior where we were running some logic connected to CloudFront on each request for some header parsing.
It is certainly possible to write a monolith in Lambda as well, but then that would be bad architecture and a bad fit and I suppose why they are now introducing this new model that is more like GCP Cloud Run or Azure Container Apps.
This is only a single environment for multi-concurrency (compared to previous Lambda Event model which was a single environment per request). Lambda can scale to huge numbers of environments. See https://aws.amazon.com/blogs/compute/exploring-aws-lambda-we... for more information
yipinwong 5 hours ago [-]
I take it as you rarely ever consider tradeoffs when you build your stuff. No good. You are the kinda person who builds, and waste my or companies money and time.
demilicious 5 hours ago [-]
Sometimes things are easily and objectively worse than the alternatives.
yipinwong 4 hours ago [-]
that "objectively" portion stems from trade-off discussion.
> Sometimes things are easily and objectively worse than the alternatives.
That's the most subjective thing one can say.
demilicious 3 hours ago [-]
Am I in a postmodernism 101 class or something? No, reality begs to differ.
yipinwong 2 hours ago [-]
[dead]
otterley 12 hours ago [-]
> I strongly recommend never using functions as a service unless you genuinely have some client edge compute requirement that absolutely cannot be solved via some other means.
Why? What’s the downside? Are you taking the benefits into account in your calculus? This sort of general, input-free recommendation is junior level thinking at best.
> I am fairly confident I could support more capacity with a single t3a.nano instance.
It depends on what resources those connections are using (memory, CPU, I/O, network). What actual work is being done?
nater5000 10 hours ago [-]
This is like the Normal Distribution meme: naively, using Lambdas for these services seems like an obvious choice. Once you start getting into the weeds of it, though, you start to realize that there are real trade-offs which make Lambda not as obvious a choice as they appear at first glance. But once you've worked through those pains and complexity, you end up realizing that there's no free lunch and Lambda is actually a great service for these tasks, all things considered.
>> up to 128 concurrent requests per environment
> I am fairly confident I could support more capacity with a single t3a.nano instance.
Don't forget that this is referring to 128 concurrent requests per environment. By default, an AWS account can run up to 1,000 concurrent Lambdas, and that number can be increased with support requests.
Of course, the REAL win is that you only pay for what you use. Your t3a.nano is costing you 24/7, and if you want it to scale on demand, your stack will get very complicated very quickly. Lambdas just scale "for free".
holler 9 hours ago [-]
Strong disagree, we use lambda/sfn/other services to process hundreds of millions of pages of data annually and it scales beautifully while being mostly hands-free and easy to maintain with a small team.
julianwood 2 hours ago [-]
true that!
hbcondo714 24 hours ago [-]
> a burst of incoming traffic can share warm environments instead of mapping each request to a new one
But cold starts could still occur, especially in a low-traffic environment where requests come in seldomly?
Reads like absolute tail, after idle period will see a coldstart but due to multi-concurrency the subsequent requests as a burst will be amortized and automatic/proactive scale-up should avoid further coldstarts.
nater5000 10 hours ago [-]
Jeeze, there's a lot of comments in this thread written by people who have no clue how any of this works lol
dang 5 hours ago [-]
I don't mean to pile on (given the reply by awithrow) but maybe it's worth repeating the principle: if you know more than others do, that's great, but in that case please share some of what you know, so the rest of us can learn!
why not reply and engage with them then rather than posting a vague dismissal?
tombert 11 hours ago [-]
This sort of just feels like those “free PHP hosting” sites you would see in ~2004 now.
Not to say it’s “bad”, just interesting how things loop around.
hk1337 11 hours ago [-]
Feels like they’re placating to those that keep bundling their entire application, php or node, into a single lambda.
HatchedLake721 14 hours ago [-]
looks like an answer to Cloudflare Workers
ZiiS 13 hours ago [-]
Isn't this a move in the other direction. Workers and traditional Lambda need to you write short lived event handlers; this is about running a standard server.
CharlieDigital 11 hours ago [-]
Google Cloud Run and Azure Container Apps feels like the closest analogues.
Both allow deployment of web apps without an adapter layer. Both scale to zero (really useful for hobby apps; small internal apps that do not get sustained traffic, but can get bursty). Both allow concurrency control.
Cloud Run "family" is quite a nice way of managing serverless ("orchestrationless"?) containers. Cloud Run worker pools for background workers that connect to Pub/Sub. Cloud Run jobs for long running tasks like extracts, ETL. Cloud Run instances (newer offering) for always on agents.
mikeocool 11 hours ago [-]
The real advantage of cloudflare workers is you don’t pay for the processing time while you’re waiting for upstream network responses.
I believe Cloudflare is the only FaaS that offers that. It makes it fantastic for putting a light weight layer in front of a few upstream API, like say inference providers.
syrusakbary 11 hours ago [-]
> I believe Cloudflare is the only FaaS that offers that
Wasmer also offers the same (charging only for active CPU usage), with the advantage that you can run full Node.js and Python workloads
Absolutely, and the offer is better in terms of DX (no need for an adapter) and it’s not limited to 128MB of memory.
Exciting times, I can’t wait for Cloudflares response.
dzonga 11 hours ago [-]
but this is built on firecracker vm. while Cloudflare workers are built on v8 isolates. at the moment Cloudflare workers have no equivalent. Deno was the only one trying to do the same.
what will be interesting to see is startup times.
AznHisoka 11 hours ago [-]
Why doesnt Amazon or any hyperscale replicate Workers? After all they all have the scale of infrastructure
kentonv 10 hours ago [-]
My biased take as the architect of Workers: It's a classic Clay Christensen innovator's dilemma.
V8 isolates seem "worse" on their face (e.g. mostly incompatible with non-JS apps, more difficult to secure), while aiming down-market. The hyperscalers' biggest existing customers aren't asking for them; they are asking for more and better micro VMs to run their existing codebases. It's only the little guys looking to deploy greenfield apps with low ops overhead and on a tight budget that really appreciate the Workers model, and to AWS these seem like small potatoes. Meanwhile there's a huge amount of investment needed to build out this entirely new and different way of doing things (we've been working on it almost a decade now!). In the worst case, this investment is wasted, and in the best case, cannibalizes the existing product line.
There likely many people at AWS who recognize the dilemma, but still cannot push the organization in the right direction. Too much inertia, too many stakeholders who don't get it, because it's unintuitive. When orgs get big they just cannot take disruptive steps, even when they were built on disruption in the past. They only finally come around after the new disruptor is eating their lunch, but by then it's too late.
felixgallo 11 hours ago [-]
Probably because their solution is technically superior from a safety perspective?
beastman82 13 hours ago [-]
And appengine
cowsandmilk 12 hours ago [-]
A response to App Engine 18 years after its release?
beastman82 12 hours ago [-]
It's the same idea however you characterize it.
Imustaskforhelp 14 hours ago [-]
Also vercel, railway, render etc
Cloudflare workers wasn’t always an 1:1 option for simple hosting (iirc) and people used to think it was more complicated than vercel etc but it has changed in recent years and right now they have made things smoother and simpler to compete with vercel and others.
Now that AWS has entered into the field as well, given that vercel and others already build on top of AWS. It will be interesting to see how this all goes.
I wonder if more og hyper scalers like say hetzner/ovh and others will also target this audience
solatic 5 hours ago [-]
Disappointed by the NodeJS-only requirement and the 128 req/env/sec ceiling.
NodeJS-only feels very pre-AI. Let me tell my agent to write my web service in Rust/C++/C/Go/Zig, and ship a statically-linked single binary that fits within a maximum binary size ceiling. I'd get much better performance, and make much more of the premium CPU/RAM pricing that I'm being asked to pay in exchange for scale-to-zero and fast autoscaling.
The rest is amazing. First-class support as a CloudFront origin. Easy multi-region and gradual deployment. Clearly, finally, a great response to Cloudflare. Just.... so close, and yet, so far. No way am I going back to Node as a web backend for greenfield projects in 2026.
It should just accept a docker container (which normal lambdas can) or a WASM blob that exposes HTTP over a standard port.
Could use optional headers for extra integration points (eg. better per-request resource usage tracking for scaling).
ryandvm 10 hours ago [-]
I feel like we're in an age where a very small team of devs armed with enterprise LLM subscriptions could absolutely build and maintain an open source, reverse engineered, self-hosted implementation of all of Amazon's most high profile AWS offerings (S3, SQS, Lambdas, IAM, EC2, DynamoDB, etc.). All as a single, cohesive open source project.
Just feed it the API docs and loop.
gregfurman 5 hours ago [-]
You’re pretty much describing LocalStack [1].
IMO much of the product’s success was derived from the extensive snapshot + behavioural tests they created to validate parity with AWS.
The basic engineering loop was: 1. Write tests against AWS
2. Record all results
3. Run tests from (1) against emulator and diff with results from (2)
Like, theoretically, you could use their preexisting suite of tests to construct a pretty decent emulator, since the engineering + cloud costs of recording behavior has already been frontloaded, and resides within their snapshot files [2].
> feel like we're in an age where a very small team of devs armed with enterprise LLM subscriptions
Ok, but who is paying them?
gman2093 10 hours ago [-]
AWS also tends to have hardware that is available for you to use.
nater5000 10 hours ago [-]
Sure... but the point with using AWS is that I don't have to self-host these services lol
Like, most of these services already have proper open-source alternatives you're free to use on your own hardware or even on AWS.
TheJCDenton 10 hours ago [-]
Yay ! Active CPU billing mean the same thing as Cloudflare ! But you still pay memory for as long as the function is up. Data transfer is not included, you have to take that into account. There is also the number of requests...
AWS should have understood by now that Cloudflare workers pricng is so much more legible, it's one of the reason of their success.
mattkenefick 4 hours ago [-]
In this thread: "If my grandma had wheels, she'd be a bike."
chaz6 12 hours ago [-]
Being able to handle concurrent requests with one worker is nice. I hope this becomes available for Python too.
jack_sunsetless 12 hours ago [-]
Finally Lambda without the adapter layer.. Now please add .NET and Java with those concurrent requests
CharlieDigital 12 hours ago [-]
Cloud Run says "Hi"
Edit: even better for .NET and Java as Cloud Run supports startup CPU boost[0] to improve initial cold start JIT performance.
What this looks like deploying a .NET web API to Cloud Run:
# Auth first: gcloud auth configure-docker us-central1-docker.pkg.dev
# Package and push using .NET tools (built in)
dotnet publish --os linux --arch x64 /t:PublishContainer \
-p ContainerRegistry=us-central1-docker.pkg.dev \
-p ContainerRepository=ORG/REG/IMG \
-c Release
# Deploy web API to Cloud Run
gcloud run deploy SVC \
--image=us-central1-docker.pkg.dev/ORG/REG/IMG:latest \
--project=PROJ \
--region=us-central1 \
--cpu-boost \
--max-instances=2 \
--min-instances=0
# Get back a URL to the ingress
I run a bunch of workloads "for free" because most apps are going to fall into the monthly credit pool (180k vCPU seconds; 50h @ 1 vCPU) if it scales to 0.
The scaling part. You don't need to do anything. It just runs as many instances of the Lambda as needed for you.
cowlevel 12 hours ago [-]
Does Elastic Container Service not do that?
kichik 12 hours ago [-]
No, not really. You can attach auto scaling to it. But it doesn't scale as fast. And it's not very straightforward to scale to zero. And you have to maintain all of it yourself.
oblio 12 hours ago [-]
Fargate does, though. And it's been doing it for at least 5 years, if not more.
kichik 12 hours ago [-]
Fargate, just like ECS, doesn't natively scale to zero. You can set the minimum number of instances to zero for both, but there is no native mechanism to scale it back to one when traffic arrives. And especially not one that would hold the connection until the container is ready. Not saying it's impossible... Just not native. And it would require extra resources. So the price won't scale to zero without something like Lambda in front of it.
Callumk77 13 hours ago [-]
Skip the docker part I guess? More approachable for new vibe-orientated developers
13 hours ago [-]
paduc 12 hours ago [-]
It’s one isolated process per request.
In a container the memory/cpu is shared for all incoming requests.
ZiiS 12 hours ago [-]
It supports 128 simultaneous requests web function; and many more if they are serial. It also supports multiple processes per function.
vends 13 hours ago [-]
[flagged]
joshstrange 11 hours ago [-]
I guess the event translation step is not needed but I don't see what this gives you over what many of us do today: shove Hono/Express/Fastify into a lambda with some shim/adapter.
It annoys me that when I started with Lambda (2019-ish?) it (at least seemed to) advised against this. I went down the "one function per route/method" path but always disliked it and it added a decent amount of complexity for packaging/deploying. A year or so ago I abandoned that and switched to Fastify (one function instead of 100+) and never looked back. It also makes it easier to test in dev/local and gives me more flexibility to move to a different host in the future.
Weird to see them come around on this but again, what does this give me over what I currently do? It looks like they are just making migrating an existing stack easier (it wasn't that difficult before IMHO).
ecshafer 9 hours ago [-]
Lambda can have its place in some architectures. I am not a fan of them in general, I've seen enough giant webs of lambdas that become impossible to truly understand the full architecture as they are split across 100 repos, just a web of slop lambdas. This feels like a canary in a coal mine, if you need this, you really need to start migrating to a standalone service on a machine instead of a lambda, where you are using it wrong.
viccis 5 hours ago [-]
I agree they can get bad but it's so nice to be able to take event driven workflows, like "pick up this file just written to S3, do some stuff to it, and write results out somewhere", and just have it work with basic features like autoretry, DLQs, built-in logging, etc.
Stringing together a bunch is an anti-pattern IMO but when it's not being abused, it's wonderful tech.
Onavo 5 hours ago [-]
This looks suspiciously like Vercel's fluid compute..
This + Function URL + Cloudfront, and you end up with an API entirely hosted in lambda. I have a backend that does is hosted like this that does pretty complex work (image/video uploads, image conversions, on top of the standard auth, database connections, etc). It even renders a frontend with HTMX/Alpine.
Even with all the bot traffic I receive, at almost 15 million requests per month the cost is under $10/mo (median latency is roughly 90ms). Cold starts are a non-issue with the Rust runtime.
The frontend is hosted on cloudfront, the backend is a lambda and you basically just pay for the database and certificate.
I don't want to manage a server or even a docker image. I just want my code to exist there so I like the idea
agree, unfortunately in practice it's more common to throw the api behind k8s with containers
https://github.com/CodeGenieApp/serverless-express
This wraps Lambda's 'handler' function and allows express to run on top of it, essentially you get a node web server with no persistence between requests which usually isn't a good thing to rely on, but you can also kinda have that with persistent tcp connections.
Of course since this is not real HTTP, things like WebSocket aren't available, but it's good enough to build a CRUD backend or two.
Similar libs exist for other runtimes as well.
Everything about running software is made more difficult this way. You might think it would make certain infrastructure aspects easier, but it just changes the semantics of the same problem. It's the same quantity of mess to clean up, but now it's in these really weird piles that are often much harder to access.
> up to 128 concurrent requests per environment
I am fairly confident I could support more capacity with a single t3a.nano instance.
That particular use case can go from hardly any traffic at all to a hurricane of traffic during something like, well, a hurricane.
For consistent processing loads, I think I could save money by just running on dedicated EC2 instances, but you cannot beat the instant scale of Lambdas.
Shameless plug: https://play.google.com/store/apps/details?id=app.radarlove....
Specify the maximum number of allowed instances and maximum concurrency per instance. Optionally configure more targeted scale rules. It will scale it from 0 - n as needed.
[0] https://docs.cloud.google.com/run/docs/about-instance-autosc...
[1] https://learn.microsoft.com/en-us/azure/container-apps/scale...
Indeed if you configured each instance as a single threaded handler, you will not have fun.
The container instances auto-scale by default to a configured n (0 - n) with the option to define specific scale points. If I'm sensitive to high bursts, this can be a matter of scaling at 50% capacity and a warm instance is available while the application still has headway.
Cases where this falls apart: DDoS in which case, I argue that 1) it is better in any case that it falls apart rather than absorbing that type of traffic spike, 2) such traffic is better filtered upstream in any case in an API Gateway + WAF.
I have worked with both types of solutions for different use cases (was invited presenter at Azure Serverless Conf[0]), but my take nowadays is that serverless containers are almost always a better approach since it provides much more flexibility.
[0] https://learn.microsoft.com/en-us/shows/azure-serverless-con...
At which point we're paying double just to work around cold start latency, which was the original point.
So if you're at 50% capacity with lots of idle time, it can scale to 0.
Scale to 0 doesn't mean "cold". One easy way to keep both Cloud Run and Container Apps "warm" but functionally scaled to 0 for costs is to just ping it every few minutes with Cloud Scheduler set up on a cron pointing to a health endpoint.
This diagram helps visualize: https://www.gstatic.com/bricks/image/5eb3c8e9-6159-4e2f-ba98... (gray bar at the top shows when the instance is started and when it ends; blue bars show actual request, and green bars show the billing)
If there's a 10s gap between requests, you are not billed for the 10s gap, but it is still effectively hot; instance does not stop. Request that comes back in 10s later does not have a startup penalty because the instance is still up.
How do you leverage this to build a high throughput system with no cold starts? Just prime it with a short single request (e.g. `/health`) before instance is stopped.
So you might have to wait for 10s or more to have a new lambda instance spawning up and replying to your requests. And you might have to have your function finish in 15 minutes, deals with the strange concurrency and scaling rules lambda is based on.
All that said, there are times I regret my lambda decision and would prefer just a EC2 host even if it cost more. Deployment would be easier/simpler. I've spent considerable time to improve my deploy process and lower the deploy times but it's still ~2-3min from tag -> deployed which I know I could do faster on just a VM. Also everything gets harder in a lambda, packages/binaries, logging, monitoring, etc.
My guess is that the IaC is the special sauce that's missing; you do need to go all in with some IaC solution to make using Lambda easy, but it doesn't take a lot of it to make it slick.
Cost/compute is obviously strictly worse for FaaS, but most web workloads don't _really_ need to be optimal.
No adapters needed; just write a web API and ship it. Public URL gets assigned and its ready for use. Code works locally with no special adapters, emulators, or tooling needed. Need to support another cloud plat tomorrow? Ship the same container. Moving it to self-managed K8s? Same container.
You make it sound like these cases are rare, not sure what kind of work you do but there are markets of all sizes in many different use cases out there.
I feel like what you wanted to say was "I haven't found a need for them" as opposed to "Stay away from it"
Maybe you're trying to say something else we don't know because we don't have your backstory either, did folks at your work tried pushing it for scenarios in which they became a problem? Share that instead as it brings a lot more value to the readers, otherwise it sounds like you are trying to go for a weird flex.
You've taken it in a rather odd direction with the request for a framing more palatable to you.
The Google paper Towards Modern Development of Cloud Applications[0] does a great job explaining why.
TL;DR the case they make against micro-services architectures (which Lambda encourages):
For even a moderately complex system, spreading logic out over multiple discrete runtimes doesn't make sense; instantly run into each of the 5 points. The only times I've found functions useful is truly single-point behavior where we were running some logic connected to CloudFront on each request for some header parsing.It is certainly possible to write a monolith in Lambda as well, but then that would be bad architecture and a bad fit and I suppose why they are now introducing this new model that is more like GCP Cloud Run or Azure Container Apps.
[0] https://sigops.org/s/conferences/hotos/2023/papers/ghemawat....
> Sometimes things are easily and objectively worse than the alternatives.
That's the most subjective thing one can say.
Why? What’s the downside? Are you taking the benefits into account in your calculus? This sort of general, input-free recommendation is junior level thinking at best.
> I am fairly confident I could support more capacity with a single t3a.nano instance.
It depends on what resources those connections are using (memory, CPU, I/O, network). What actual work is being done?
>> up to 128 concurrent requests per environment > I am fairly confident I could support more capacity with a single t3a.nano instance.
Don't forget that this is referring to 128 concurrent requests per environment. By default, an AWS account can run up to 1,000 concurrent Lambdas, and that number can be increased with support requests.
Of course, the REAL win is that you only pay for what you use. Your t3a.nano is costing you 24/7, and if you want it to scale on demand, your stack will get very complicated very quickly. Lambdas just scale "for free".
But cold starts could still occur, especially in a low-traffic environment where requests come in seldomly?
There was also this a week ago, maybe you can combine this with containers? https://aws.amazon.com/blogs/compute/get-started-with-aws-la...
> Lambda scales down to zero environments when there is no traffic. The next request after that starts a new environment (a cold start).
https://docs.aws.amazon.com/lambda/latest/dg/web-functions-s...
https://hn.algolia.com/?dateRange=all&page=0&prefix=true&sor...
Not to say it’s “bad”, just interesting how things loop around.
Both allow deployment of web apps without an adapter layer. Both scale to zero (really useful for hobby apps; small internal apps that do not get sustained traffic, but can get bursty). Both allow concurrency control.
Cloud Run "family" is quite a nice way of managing serverless ("orchestrationless"?) containers. Cloud Run worker pools for background workers that connect to Pub/Sub. Cloud Run jobs for long running tasks like extracts, ETL. Cloud Run instances (newer offering) for always on agents.
I believe Cloudflare is the only FaaS that offers that. It makes it fantastic for putting a light weight layer in front of a few upstream API, like say inference providers.
Wasmer also offers the same (charging only for active CPU usage), with the advantage that you can run full Node.js and Python workloads
https://wasmer.io/
And the maximum worker time is based on processing time, not wall time.
https://vercel.com/fluid
Exciting times, I can’t wait for Cloudflares response.
what will be interesting to see is startup times.
V8 isolates seem "worse" on their face (e.g. mostly incompatible with non-JS apps, more difficult to secure), while aiming down-market. The hyperscalers' biggest existing customers aren't asking for them; they are asking for more and better micro VMs to run their existing codebases. It's only the little guys looking to deploy greenfield apps with low ops overhead and on a tight budget that really appreciate the Workers model, and to AWS these seem like small potatoes. Meanwhile there's a huge amount of investment needed to build out this entirely new and different way of doing things (we've been working on it almost a decade now!). In the worst case, this investment is wasted, and in the best case, cannibalizes the existing product line.
There likely many people at AWS who recognize the dilemma, but still cannot push the organization in the right direction. Too much inertia, too many stakeholders who don't get it, because it's unintuitive. When orgs get big they just cannot take disruptive steps, even when they were built on disruption in the past. They only finally come around after the new disruptor is eating their lunch, but by then it's too late.
Cloudflare workers wasn’t always an 1:1 option for simple hosting (iirc) and people used to think it was more complicated than vercel etc but it has changed in recent years and right now they have made things smoother and simpler to compete with vercel and others.
Now that AWS has entered into the field as well, given that vercel and others already build on top of AWS. It will be interesting to see how this all goes.
I wonder if more og hyper scalers like say hetzner/ovh and others will also target this audience
NodeJS-only feels very pre-AI. Let me tell my agent to write my web service in Rust/C++/C/Go/Zig, and ship a statically-linked single binary that fits within a maximum binary size ceiling. I'd get much better performance, and make much more of the premium CPU/RAM pricing that I'm being asked to pay in exchange for scale-to-zero and fast autoscaling.
The rest is amazing. First-class support as a CloudFront origin. Easy multi-region and gradual deployment. Clearly, finally, a great response to Cloudflare. Just.... so close, and yet, so far. No way am I going back to Node as a web backend for greenfield projects in 2026.
Could use optional headers for extra integration points (eg. better per-request resource usage tracking for scaling).
Just feed it the API docs and loop.
IMO much of the product’s success was derived from the extensive snapshot + behavioural tests they created to validate parity with AWS.
The basic engineering loop was: 1. Write tests against AWS 2. Record all results 3. Run tests from (1) against emulator and diff with results from (2)
Like, theoretically, you could use their preexisting suite of tests to construct a pretty decent emulator, since the engineering + cloud costs of recording behavior has already been frontloaded, and resides within their snapshot files [2].
[1] https://github.com/localstack/localstack
[2] https://github.com/localstack/localstack/tree/main/tests/aws...
Ok, but who is paying them?
Like, most of these services already have proper open-source alternatives you're free to use on your own hardware or even on AWS.
AWS should have understood by now that Cloudflare workers pricng is so much more legible, it's one of the reason of their success.
Edit: even better for .NET and Java as Cloud Run supports startup CPU boost[0] to improve initial cold start JIT performance.
What this looks like deploying a .NET web API to Cloud Run:
I run a bunch of workloads "for free" because most apps are going to fall into the monthly credit pool (180k vCPU seconds; 50h @ 1 vCPU) if it scales to 0.[0] https://cloud.google.com/blog/products/serverless/announcing...
In a container the memory/cpu is shared for all incoming requests.
It annoys me that when I started with Lambda (2019-ish?) it (at least seemed to) advised against this. I went down the "one function per route/method" path but always disliked it and it added a decent amount of complexity for packaging/deploying. A year or so ago I abandoned that and switched to Fastify (one function instead of 100+) and never looked back. It also makes it easier to test in dev/local and gives me more flexibility to move to a different host in the future.
Weird to see them come around on this but again, what does this give me over what I currently do? It looks like they are just making migrating an existing stack easier (it wasn't that difficult before IMHO).
Stringing together a bunch is an anti-pattern IMO but when it's not being abused, it's wonderful tech.