That's true for the base S3 product, but there are a lot more storage tiers than there used to be with cheaper pricing. All the way to $0.99/TB for Deep Archive.
On price, you could definitely do better than $8/million steps, $0.25/GB written and $0.15/GB-month for state storage, but if you were designing something generic on S3/DynamoDB (state + status) to support all use cases at all scales, you'd probably end up spending something around the same order or magnitude.
But if you did that, you'd also have to implement it all yourself. This is a relatively simple checkpointing workflow orchestrator across standard Lambda functions, but with some really nice touch surfaces in the Lambda API itself.
What's only a footnote in the announcement is that this is only us-east-2 (Ohio) and TypeScript/JS + Python at the moment. Basically a public preview release. I look forward to seeing where they take this.
Sounds like a step in the right direction. I would like to see an all-up dashboard of everything in the shared state, and good control over upgrades (maybe a mode where in-progress functions can complete on version 1, even if new functions are getting kicked off on version 2, etc.)
Still $0.20/million requests, but all Lambda functions run on provisioned EC2 instances (taking into account savings plans & reservations) with a 15% premium.
You can dial up/down the vCPU:RAM ratio so that if you have a lot of functions that just, for example, wait on IO, you can use a very high ratio to run many more functions in parallel on a single instance.
This looks like it will provide an interesting middle-cost option for services at scale with a more predictable load or usage pattern better suited to higher (or lower I guess) ratios without having to sacrifice any effort already put into Lambda and lets you still use it with other AWS services (Cognito auth, IoT Events, simplified Kinesis/DynamoDB Streams client that doesn't require Java, etc).
Running your own AuthN/AuthZ with an off-the-shelf OSS is very straight-forward (as a SaaS product at least) and isn't any more burdensome from a security perspective than what you're already doing for your core service.
One feature failing like this should probably log the error and fail closed. It shouldn't take down everything else in your big proxy that sits in front of your entire business.