Correct me if i'm wrong, but for "serverless" to work, does there still not need to be a "long running" daemon somewhere that runs the containers on demand? Who pays for that? Is it just extra capacity that AWS just takes the hit on for the cost of the service? And if so, can this extra capacity burst if needed when millions of requests come in at the drop of a hat?
You're right, the "long running daemon" is the serverless platform, i.e. the process that hosts the serverless applications (microflows). In case of public cloud providers like AWS, there is ample spare capacity, so hosting (and charging for) microflows in addition to other workloads makes economic sense.
With respect to the extra burst capacity -- in theory there should be excess capacity available, in practice it varies depending on your cloud provider.
The lack of the excess capacity is usually a financial, not an engineering problem. The cloud provider may offer SLAs that ensure that a customer gets credit in case if some requests haven't been serviced as expected.
That makes sense, it just seems wasteful on aws's end to have thousands of machines running "for free" waiting on requests that may or may not come. Provisioning more capacity also can't be instant. For a large scale enterprise system it seems like it could be unpredictable.
3 comments
[ 12.6 ms ] story [ 123 ms ] threadWith respect to the extra burst capacity -- in theory there should be excess capacity available, in practice it varies depending on your cloud provider.
The lack of the excess capacity is usually a financial, not an engineering problem. The cloud provider may offer SLAs that ensure that a customer gets credit in case if some requests haven't been serviced as expected.