Zabbix with EKS and Karpenter

[rfrozza]

Hi everyone!
I'm currently deploying Zabbix inside a cluster (AWS EKS and with Karpenter on top of that).
Since nodes will be drained and new nodes will join the cluster while Karpenter orchestrates the resources, how can I properly manage such a case dynamically?
Is creating a host with the Kubernetes nodes by HTTP template the best approach? If yes, is it possible to create the {$KUBE.API.TOKEN} through the current helm chart? I'm unable to find something related to it in the values or documentation.

Anyone already had such an experience with Zabbix?
Any tips or help will be appreciated 👍
Thanks in advance!

[aeciopires]

Hi @rfrozza!

I confess I haven't tried this use case.

I did test installing the Zabbix components (server, proxy, frontend, agent) on an EKS cluster with Karpenter. The helm chart supports adding any YAML file to manage the nodepool, nodeclass, provisioner, etc... using the key extraManifests in values.yaml file.

Sorry I couldn't help you with your use case. Let's wait for someone else to respond.

[rfrozza]

Thanks anyway @aeciopires!
I'm currently using the approach I've mentioned

So far, using three hosts:

• Kubernetes Nodes (template Kubernetes nodes by HTTP)
• Kubernetes API (template Kubernetes API server by HTTP)
• Kubernetes Cluster State (template Kubernetes cluster state by HTTP)

I've created them manually, and made a token through the already created SA zabbix.
Since I'm not familiar with Zabbix, do you have any suggestions on how to create those hosts automatically (link the template, create the interface, macros, and so on)?
I think a POST request to the Zabbix API is the right approach, but is it possible through the extraManifests feature you mentioned?
I'm considering a Python script, and then containerizing it to run as a job haha

[fibbs]

This helm chart is to run Zabbix in Kubernetes, not to monitor Kubernetes with the monitoring templates delivered with Zabbix. For these to work, you would use the official Helm Chart provided by Zabbix to install Zabbix Proxy and Zabbix agents (as daemonsets, so they dynamically run on each node) onto the K8S cluster you want to monitor (can be the same where Zabbix runs on, but doesn't have to). In the linked documentation it is described how to get the token needed to use the templates.

You then would add a "dummy host" in your Zabbix, assign the Kubernetes nodes by HTTP template to it, set the host macros approprietely, and Low Level Discovery will auto-discover the K8S nodes as Hosts and start monitoring them.

As mentioned, as the helmchart provided with this repository is for the usecase of running Zabbix, this topic is a bit "off topic" here...

[fibbs]

And indeed, for automating Zabbix configurations such as "after plain installation, please add these hosts, assign those templates to them" etc. I have a couple of things in my mind, but nothing of this ready yet: I am thinking of a kind of a "Zabbix entity operator" bridging CRD objects into Zabbix API calls. This would Kubernetize the Zabbix configuration and would be very helpful for mass-installations.

Beside of that, it's doable with Ansible, as there is a rather good Ansible library for Zabbix configuration. As far as I know, it's not too difficult to wrap an ansible role into an operator by using Red Hat's operator framework, but I have not yet had the chance to build such one myself. Let me know if you get anything running....

[rfrozza]

Thanks @fibbs !
It was indeed a mix of the charts you provided and the one to install components for the Kubernetes cluster monitoring (from Zabbix repo).
I'm using your helm-zabbix chart to deploy everything, except the proxy and agents.
Actually, it is possible to use your helm chart too, but the kube-state-metrics is mandatory for the "Kubernetes cluster state by HTTP" template.
Maybe, in a future release, add the zabbix-agent-kube-state-metrics deployment in your helm chart.