What is the purpose of running Kubernetes nodes in full stack host monitoring mode

[Hillman]

We have installed Dynatrace Operator to our Kubernetes clusters, monitor the infra metrics on clusters and the application within the pods.

However, the default monitoring configured for Kubernetes nodes are in full-stack monitoring, which is very, very costly. We have changed our Kubernetes architecture recently and the nodes are all running in full load, which cause the license consumption acclerating.

So what it the use of full-stack monitoring? If we change the Kubernetes nodes monitoring from full-stack to infra-only, will it impact the application injection?

Regards,
Hillman

[AntonPineiro]

Hi,

I understand POD injection would be as before but no information in infra & operations app about those:

But lets wait anothers answers, I am just guessing.

Best regards

[Julius_Loman]

If you want deep monitoring (full stack, tracing, bizevent capturing, ... all the features of full stack agent), you have to run the cloudNativeFullStack (or legacy classicFullStack). If you are interested only in k8s platform monitoring, you can have either just k8s observability (without visibility into the applications) or k8s observability + hostmonitoring only (oneagents in infrastructure mode, see https://docs.dynatrace.com/docs/shortlink/k8s-host-monitoring ).

Without knowing your requirements, it's difficult to advise the optimal deployment mode.

[Hillman]

Thank you @AntonPineiro for the support feature list table. It is very useful to us.

@Julius_Loman The Kubernetes clusters are currently running on Azure AKS. The CPU, memory and disk of AKS nodes are automatically maintained by Azure themselves, we do not have much control about them. It is of no use for us to monitor them. I believe the only thing we care is the pods have reached the maximum limit and it can no longer scale up new pods into the AKS nodes.

What we want to monitor are the followings:

1. Alerts and Events of k8s clusters and pods
2. Resources usage for k8s clusters and pods
3. Status of processes / process groups running on k8s clusters
4. Applications status, distributed trace deployed on pods

[gopher]

Hi @Hillman 

Fullstack is going to give you a little more coverage when it comes to nodes and injection capabilities,  however, 
unless you are using non standard tech stacks, you should be fine with: 

Kubernetes Monitoring and Automatic Application Only injection (CSI). 

You will get cost benefits from this as well. 

Full stack is - Monitor Everything, exclude by rules.
Application is - Monitor Nothing, Include or exclude by rules or labels.

This gives you options to deploy with labels that are matched and only those pods or namespaces will get automatic injection. the Kubernetes monitoring will then give you the basic node utilisation details and other Kubernetes metrics and events.    If you want Logs, you can then also do the logging module (rule based). 
 
The new Kubernetes dashboards and apps don't need full stack for the topology (unlike the classic screens). 
Application injection is pretty much on par with full stack, excluding a few scenarios. 

Suggest you take a look at this approach, we run on EKS & GKE like this, no issues. 

[Hillman]

Thank you for your suggestion, @gopher.

I have worked with our DevOps team to enable application-only on our Dev environment. It is able to perform injection into the pods, discover service, display resource usage on pods or namespaces. The only drawback is the k8s clusters sort of disappear from the cluster view and we are not able to check on the resource usage on the k8s clusters.

That is why I want to check whether changing the AKS nodes to infra-only mode can achieve application + cluster resource usage.