Hello Reader,

I am sending out my newsletter on Saturday, instead of usual Friday, because I am currently returning from world's biggest cloud conference, AWS Re:Invent, from Las Vegas USA. It wrapped up Friday evening, and I am literally typing this newsletter on the plane, please forgive any typos. Now, while a lot of new announcements were made, you don't need to know all of them. There are selective ones that are impactful for interviews, and most real-world projects. We will break those in two parts. So, without further ado, here we go:

AWS Lambda Managed Instances

Challenges of AWS Lambda

Some of the biggest disadvantages and complains of AWS Lambda has been:

• No option to select specialized hardware - this is the biggest superpower of EC2s. With over 800 different instance types to choose from, you can select underlying EC2 types based on your workload need. For example, if your workload need GPUs, only EC2s can provide that. Lambda, can only run on general purpose hardware..till this announcement happened. More on this in a bit
• Cold start, cold start, cold start - everyone and their grandmother know about Lambda cold start at this point. Cold start is the time Lambda takes between you invoking the function, and when the function actually starts executing. AWS takes this time to provision an underlying Lambda instance, load your code, and execute the global section of the code (before the lambda_handler() part). You might say, well AWS had Provisioned Concurrency and Snap Start for that. But read on...
• Handling steady high traffic - Lambda is pretty good from cost perspective, for handling unpredictable traffic. But when the traffic is predictable, especially high traffic, EC2s become more cost efficient

What is Lambda Managed Instances

You can now run AWS Lambda functions on your Amazon Elastic Compute Cloud (Amazon EC2) while maintaining serverless operational simplicity.

Major Changes/Advantages:

• Pick any EC2 instance, even GPU or Graviton 5, to run your Lambda function. The specialized hardware issue is gone!
• AWS manages underlying EC2 i.e. scaling, and patching
• Because it's EC2, you can use Reserved Instances for this, which saves you more money than existing Compute Savings Plan.
• Till now, one Lambda instance could only handle one concurrent transaction. But with Lambda managed instances, one EC2 instance running your Lambda can handle multiple concurrent transactions. This combined with Reserved Instances, can be perfect for steady state traffic
• Because one EC2 can handle multiple concurrent transactions, cold start is gone. You could use Provisioned Concurrency or Snap Start, but AWS wanted to bundle all the customer complains into one feature. Lambda Managed Instances solves all that

Considerations:

• As always, there are tradeoffs. Because AWS is literally running an entire EC2 for you, you'd be charged for uptime. As you (and your grandma) know, you pay for the full price of regular EC2s, irrespective of utilization. But that's not it..
• AWS charges you 15% additional cost of the EC2 as management fee.
• Estimate price in cost calculator to know if this is viable for you

Overall Trend:

There is a bigger AWS trend here. Remember last year's re:Invent, AWS announced EKS Auto (ahem..I am the one who announced it on stage in a breakout ;)). As with EKS Auto, and with Lambda Managed Instances (and more), AWS is offering to manage the underlying infrastructure, and operational overhead..for a fee. The choice is upto the customer - they can manage the components themselves if they have the expertise, or let AWS manage it, and focus on business needs. There is no perfect answer - it depends on the company.

Amazon EKS Capabilities

I know what you are thinking - why is it "AWS" Lambda, but "Amazon" EKS ;). But let's move past that. This new announcement is also following the same bigger trend - give flexibility to the customer on whether to have more control (and overhead), or let AWS do some management (for a fee). Now that you know the trend, look out for this in other announcements as well.

Challenges of Amazon EKS (or Kubernetes in general)

• Management overhead - Managing different software/controllers is tough. You need to scale it, secure it, ensure their versions are compatible with EKS versions
• Data plane bloat - Running all the controllers in the worker nodes causes you to spend compute resources (and money). isolating them from application also is a headache. For example, on a standard EKS cluster you want to use Karpenter, but the nodes running Karpenter controller and other controllers, can't be controlled by Karpenter
• Amazon EKS Auto solved some of these challenges, but customers wanted to have some of these management overhead gone in standard EKS clusters too

What is Amazon EKS Capabilities

With a click of a button, now AWS can manage Argo CD, AWS Controllers for Kubernetes, and Kube Resource Orchestrator. AWS manages the scaling, patching, and updates of these controllers.

Major Changes/Advantages:

• AWS runs these controllers as part of Control Plane, no need to worry about resource usage and app isolation in data plane
• AWS can intelligently upgrade these controllers with your cluster upgrade or keep at previous version if not compatible
• Select one or more addons (ahem..capabilities). Just need ArgoCD? Sure, you can do that. Need all three? Sure thing
• This works BOTH for EKS Auto, and Standard EKS
• Even though AWS is managing ArgoCD, it gives you access to familiar Argo UI, which is nice

Considerations:

• You guessed it..it comes with a fee. There is a flat charge, and an usage charge
• AWS is smart here. They know ArgoCD is the most popular one. Hence you pay more for both flat, and usage charge for ArgoCD compare to ACK and KRO
• Estimate price in cost calculator to know if this is viable for you
• For high compliance and regulated workloads if you need to show scanning results of each controllers, then this is not a good fit

My wish list will be AWS also delivers just the managed Karpenter in Control plane as one of the capabilitites, instead of making customers go EKS Auto Mode

One BIG Miss this re:Invent

First of all, it's no secret that I worked at AWS for almost 7 years in various roles - Sr. Enterprise Solutions Architect, then Principal SA, hence I want AWS to win because, over half of my salary in those 7 years was on Amazon stocks. If AWS wins, I win.

Being said that, I can see that other hyperscalers, and providers are getting ahead in the database offerings. MongoDB, Oracle, GCP all are providing a database option where a single database can handle vectors, AND/OR SQL, AND/OR JSON. I was expecting AWS to announce something on those lines. In AWS, you need different databases for different purpose. S3 vector is nice, but it can't match a SQL or NoSQL DB.

But Raj, what about Gen AI announcements?

These days, I don't comment or react on any Gen AI announcements, till couple weeks pass. Because one hot Gen AI feature becomes stale in a week. And I don't want you to study something that's not useful. I will cover them...perhaps in a week or so

Some Pictures

I will go over couple other major announcements in next edition. In the mean time, enjoy some of my pics from this re:Invent:

P.S - As always, SA Bootcamp will include all the necessary enhancements based on re:Invent. If you want to get personally mentored by me and crack top tech jobs in AWS, Microsoft, Google, JPMC, reddit, Databricks, CoreWeave etc., check out AWS SA Bootcamp with Live Classes, Mock Interviews, Hands-On, Resume Improvement and more: https://sabootcamp.com/. Still debating on the release date for Jan Cohort, you all will be the first to know!

Keep learning and keep rocking 🚀,

Raj