Thread by @PaulDJohnston

I have heard serverless transition stories where the cost of compute (just servers, no db) is over $10k a month, and transitioning the compute to lambda, just the compute figure has gone down to less than $500 for the year.

Transitioning to a Lambda based compute system is not easy though. It requires a reworking and a rebuild of the entire system.

Most are unaware of this though, and think that it's relatively straight forward.

The problems of a transition don't come in the transition of code from servers/containers to the Lambda solution.

The problem comes when the system hyperscales at some point.

And. It. Will.

Lambda is a dream in terms of being able to cope with scaling up and down from zero to 100. But most solutions actually are not built to cope with that sort of scaling and most code has definitely not been built to cope with that.

And the majority of the data solutions that have been put into practice up until now are not built to handle scale.

*cough* RDBMS *cough*

I love RDBMS for non-scaling scenarios. But when you put them into a Lambda system, they become the largest *prospective* headache.

You can absolutely use RDBMS with Lambda *if* you know what the issues are beforehand, and *if* you mitigate the issues with configuring your systems and code well.

The thing is that if you're *transitioning* a system from EC2/containers to Lambda, then you haven't done that.

So transitioning your systems is doable but non-trivial. What actually tends to happen is that a completely new system is built, and an API and data layer wrapper gets deployed in some form, and the Strangler Pattern gets deployed in some form (google it)

And the biggest hurdle in a transition to a serverless system on Lambda is actually not the code. That tends to be the easiest bit.

The difficult bit to transition?

The data.

Data is the hardest part of any transition. It's why most legacy systems are hard to move. It's why most people don't like working on legacy projects. You can often upgrade hardware and software, but the problem tends to always be the data and it's associated storage.

So, if you have the chance to do a transition, remember that the easiest bit is the code.

The hardest bit is the data.

Figure out how to move the data, because the code is trivial after that.

Oh and if you need someone that knows the kind of things you're going to need to think about before you start?

I'm available. DM me. Fixed fee. Especially if you have a $10k/month fleet of EC2 instances running something. It'll be worth your while..