Spot instance is not a guaranteed instance type

[Haridev Ngangbam]

Question in the picture. Spot instances are not guaranteed instance they can be reclaimed by AWS anytime. How do you know the application can handle consistency in such a case? How do you know spot instances will be available (at your price point) when you want to scale up?

[Tutorials-Dojo]

Hi Haridev,

Take note that there are two types of load in the scenario, which are:

– Steady-state Load

– Peak Load

The scenario says:

A web application is composed of an Application Load Balancer and EC2 instances across three Availability Zones. During peak load, the web servers operate at 95% utilization. The system is set up to use Reserved Instances to handle steady-state load and On-Demand Instances to handle the peak load. Your manager instructed you to review the current architecture and do the necessary changes to improve the system.

Which of the following provides the most cost-effective architecture to allow the application to recover quickly in the event that an Availability Zone is unavailable during peak load?

The provided answer is:

Launch a Spot Fleet using a diversified allocation strategy, with Auto Scaling enabled on each AZ to handle the peak load instead of On-Demand instances. Retain the current setup for handling the steady-state load.

Throughout the scenario, the steady-state load of the application is still handled by Reserved EC2 instances which means that the application’s availability is not compromised. The only thing that’s changed here is the use of a Spot Fleet for handling the peak load.

Furthermore, we are also using a Spot Fleet with a diversified allocation strategy. This is an advanced compute setup in AWS that balances out the price and computing capability/availability. We have to fully understand why a Spot instance is getting interrupted so we can implement a cost-effective, yet highly available, solution.

AWS usually interrupts a specific Instance type, based on the demand it’s getting so for example, if there’s a surge of On-Demand t3.large instance type, then your Spot t3.large instance would most probably be interrupted. Hence, a Spot Fleet with a variety of EC2 instance types (t3.large, m6.medium, a4.small, etc), provides more durability. If the demand for t3.large instance type goes up and it affects/terminates the t3.large instances on your end, then you still have m6.medium and a4.small EC2 instance types available.

More information here:

https://docs.amazonaws.cn/en_us/AWSEC2/latest/UserGuide/ec2-fleet-configuration-strategies.html#ec2-fleet-allocation-strategy

I’ve worked with various clients that also implement the same thing in their architecture. Different applications have different SLAs and compliance requirements. It depends on the various factors but the most important thing to note here is that you shouldn’t use a Spot Instance solely for your steady-state load.

Cheers,

Jon Bonso

[Haridev Ngangbam]

Sorry Jon, I don’t think the gap is filled.

Although I understand your point but I wonder would the exam consider such an answer with a technical gap correct.

Does it mean in peak load you can just abandoned a process (I wouldn’t unless the application has a built in capability to gracefully continue the process in another instance or the usecase is such that it can tolerate such a situation – eg. a Hadoop task node) ?

Maybe you right many organization use this strategy that’s doesn’t mean that this automatically become the right answer to this question.

• This reply was modified 2 years, 8 months ago by  Haridev Ngangbam.

[Tutorials-Dojo]

Hi,

I understand your point but it is clearly mentioned in the AWS documentation that Spot Instances can be used for handling the peak load or peak traffic of your web application.

“Maintain optimal web performance during peak traffic at low costs using Spot Instances…”

https://aws.amazon.com/ec2/spot/?cards.sort-by=item.additionalFields.startDateTime&cards.sort-order=asc

Spot Fleets are also used to provide a certain level of availability to your applications: https://aws.amazon.com/blogs/compute/running-web-applications-on-amazon-ec2-spot-instances/

Moreover, a Spot Fleet can deploy different EC2 instances across different Availability Zones, and not just a single AZ. This strengthens its compatibility for the explicit requirement in the scenario that says:

“…to recover quickly in the event that an Availability Zone is unavailable during peak load? “

Take note that the scenario explicitly asks for the cost-effective option. That’s the primary keyword here, and again, the overall availability of the application isn’t compromised.

You have a point that there would be several requests that would be affected when the Spot Instances got interrupted by AWS. That’s an edge case scenario and if you chose On-Demand/Reserved Instance simply based on this “edge-case”, then you are ignoring the primary keyword that focuses on cost. Doing so will make you fail the exam as AWS won’t add an important keyword at the bottom of the scenario that isn’t related to the answer it expects.

The SA Pro exam is a difficult test because you have to weigh in these different considerations on your architecture. We can’t be too obsessed with application availability if the scenario explicitly mentioned focusing on cost, especially if there’s already a provision for availability in terms of a Spot Fleet with a diversified allocation strategy solution.

If the web application is stateless and can handle the request failures through retries, then it shouldn’t be a problem. Spot Instances have 2-minute notification too before it gets terminated.

I’ve passed 10 AWS Certifications and in my opinion, if there are 2 explicit key phrases in the scenario that points to a particular AWS Service, then AWS considers that as the correct answer and not the minor details in the scenario.

Regards,

Jon Bonso