BLUF

Bottom Line Up Front

Understand your IOP and Throughput peak needs before you create the cluster. Let the data dictate your requirements.

1. VM IOPs are not the same as Disk IOPs, be sure you choose a VM and disk capable of your workloads peak.
2. Think about potential growth. Is your design idea going to easily allow you to scale as you grow organically?
3. Understand that Virtual Machine Scale Sets (VMSS) allow you superior growth flexibility over Availability Sets. If you’re thinking of Availability Sets over Scale Sets: Why?
4. Use the Diagnose and Solve blade in the portal to help you…

Understanding an applications resource consumption is vital to ensuring proper performance in a Kubernetes cluster. This is part 1 of 3 and we will focus on looking at our application at rest to understand the resource impact the application has without any load

Optimizing cluster performance isn’t something that is unique to the Azure Kubernetes Service. All the concepts and ideas in this article stretch not only to every cloud provider and K8 application but all applications in general.

There are three other articles I wrote that will help you understand why these concepts are important in Kubernetes. …

Ramblings from a veteran

I have gone through many interviews in my nearly 20 years in the IT field. Many of them I had little interest in before even accepting an interview. I know some of you might be thinking “why waste everyone's time if you’re not interested?”. Well simply put, its an interview and it goes two ways. I am just as unsure if I am interested in working with them as they are with me.

An interview isn’t just about the candidate selling themselves. The company must also sell to the candidate. It’s a fine balance of yes…

Be like water, my friend.

It was bound to happen eventually

Most of you who consume Kubernetes have likely already heard the news that Kubernetes is going to deprecate Docker support in an upcoming release. Surprisingly I have seen quite a bit of panic from both consumers who host their own clusters and from those who use hosted solutions like AKS and GKE. Specifically because some how its getting around the deprecation is happening on version 1.2.

The only thing that consumers should notice in 1.2 is a kubelet warning that gets logged if Docker is the current runtime, the actual deprecation is not expected to…

Knowing where you’re going by knowing where you’ve been

So bonus points for anyone who knows that reference. Not like…cool points or anything…just…points. ;) One of the most water cooler talked about things I tend to hear is around monitoring VM performance. Microsoft has all kinds of crazy awesome tools to look at how a VM is performing but what about the consumers?

Believe it or not, most of the resources Microsoft evaluates comes from the same source that is available to all customers directly from within the portal. Now depending on the type of compute resources you have (IE…

• Limit Ranges
• Preventing Creation of Large Pods
• Preventing Creation of Tiny Pods
• Creating a Limit Range
• Applying a Limit Range
• Verifying the Range
• Deploying Pods
• Summary

How much vs how little

In my previous article I discussed Resource Quotas and gave an example of how and why someone may want to consider using them on their cluster. You might have be thinking to yourself in the last article “Well that's nice you can limit a namespaces resources but what happens if someone creates a pod that just takes it all up at once?”. …

Limiting resource consumption for multi-team clusters

Always remember to evaluate your needs and requirements against what resources you have available to determine your best approach.

Preventing Resource Exhaustion

When limits are not enough

In my previous article I discussed Resource Request and Limits to help people understand the importance of setting appropriate pod specs to avoid over saturating nodes. What if the cluster is shared across multiple teams or developers? How do you prevent one single person from taking up all the cluster resources?

Many teams have limited resources and specific resource needs that must be shared

In clusters like this setting pod…

How much do I need vs how much can I use

One of the most important things Kubernetes can do for you is to help ensure workloads are running on nodes capable of handling their resource requirements. Setting proper requests and limits can help prevent you from over saturating your nodes which would inevitably result in poor performance.

Resource Requests

Definition of the amount of resources K8s will guarantee to a pod hosted on a node

Resource Requests define the amount of resources that K8s will guarantee to a pod. To some degree you could think of this definition as the minimum…