---
title: Sam Newman - Confusion In The Land Of The Serverless
disqus: pierodibello
---

###### tags: `serverless`

# Sam Newman - Confusion In The Land Of The Serverless

There are many takes from Sam Newman about this talk.
A couple are the following:

* https://www.infoq.com/presentations/serverless-issues/ (one of the first take, took at CraftConf)
* https://www.youtube.com/watch?v=aZlrv-0PE_c (one of the most recent takes)

## A definition of Serverless

Serverless - five things a platform need to have to be serverless (see https://blog.symphonia.io/defining-serverless-part-1-704d72bc8a32)

1. [Requires no management of Server hosts or Server processes ](https://blog.symphonia.io/defining-serverless-part-1-704d72bc8a32)
2. [Self auto-scale and auto-provision, based on load](https://blog.symphonia.io/defining-serverless-part-2-8cc67bc0bdd8)
3. [The costs are based on precise usage](https://blog.symphonia.io/defining-serverless-part-3-f65856be4ade)
4. [Performance capabilities defined in terms other than host size / count](https://blog.symphonia.io/defining-serverless-part-4-587be1e64294)
5. [Implicit High Availability](https://blog.symphonia.io/defining-serverless-part-5-5452308de55c)

Be warned on "high-availability" though: high availability of what? You don't have high-availability of all your solution/platform just because you use some components that provide high-availability out-of-the-box.

Serverless is a form of PAAS, a most prescriptive abstraction of a PAAS

### Resiliency

When one component is massively scalable and the other is not, you have a potential issue: **cascading failures**

Typically, you would use protection patterns like **connection pools** or **circuit breakers**.

**Connection pools**: throttles the number of clients going to talk to the db => load shedding

How could I throttle functions? you can't, or better, you've to use api gateway

How would you mix a sql db (e.g. mysql) into a serverless architecture?
If you want all the solution to be highly scalable, you should use a BAAS (e.g. dynamodb, google big data), but mysql and dynamo are not semantic equivalent!

So, what about **hybrid apps**?

Can I build a highly resilient serverless app with an hybrid approach? probably not.

**Circuit breaker**

To integrate FAAS components with other non-serverless components you have to provide some sort of middleware to load-shedding the components that cannot transparently scale

lambda => circuit breaker: can I call downstream component?
circuit breaker => lambda: ok, go
lambda => calls the component

#### Mixing vendors?

Is mixing vendors a good strategy to "spreading your eggs" => not so much => you got latency (es. BAAS google, frontend in lambda)

#### Closing
Serverless definition... Ignore the specific definition: the goal of a platform is making easy for you to focus on the job you're doing

Serverless is a much better developer-friendly abstraction that allow to focus on the problems at hand