These would boot faster since you don’t have to boot a whole OS, just a V8 VM. Downside is compatibility, your lambdas would have to be largely rewritten (especially if they rely on any other AWS services). It’s also probably more expensive since you’d have to keep a EC2 or something up running this.
Kinda wish Amazon would have an alternative runtime that was more like this. They kinda do with lambda@edge, but that’s only for CloudFront stuff, mostly.
Lambda@Edge is a great solution for the locality problem (CloudFront can just route to the nearest edge function, and despite certain headaches, it works well), but what it doesn’t help to solve is the cold start problem. In a sense Lambda@Edge works against it, as provisioned concurrency is only supported for ‘regular’ Lambda functions.
At a certain level of scale, the cost of spinning of an ALB Fargate task (which could run these servers relatively easily) ends up being relatively negligible. Although that’s certainly not true for every org.
Kinda wish Amazon would have an alternative runtime that was more like this. They kinda do with lambda@edge, but that’s only for CloudFront stuff, mostly.