Kubernetes Cluster Monitoring

[jeremy.hall]

Howdy folks -

Just wondering what you have tried for monitoring kubernetes clusters.  I have an upcoming project to intigrate with kubernetes and since there is no community extension then I was curious as to what I can possibly monitor without making our own extension to plug into AppD.  Any guideance is greately appriciated. 

It's almost Friday :) We can make it!

[Arun.Dasetty]

Hi,

Unfortunately as per internal request PLAT_-596 (feature request) and as per internal request this is currently not supported as of now, However we see using docket extension will monitor apps in docket if i interpret right, can you see if using docket you will get some visibility https://www.appdynamics.com/community/exchange/extension/docker-monitoring-extension/?

[mia.avery99]

Hello, Jeremy.hall

Please tell me, I have a doubt

what is a difference between Apache's Mesos and Google's Kubernetes?

 

[Arun.Dasetty]

Hi,

Can you see if below links helps?
https://mesosphere.com/blog/docker-vs-kubernetes-vs-apache-mesos/

https://stackoverflow.com/questions/37766277/apache-mesos-vs-google-kubernetes

[ben.coughlan]

Hey, any luck with this? 

I've been trying to monitor K8's too, but it seems to be a bit of a mess. 

Cheers, 

B

[jeremy.hall]

Nope, nothing yet. The best you can currently do is to deploy the AppD agent in EACH contrainer you deploy, which is not the best solution to the issue. To have K8 supported would make perfect sense for AppD to add into the product and it is going to be horrible until they do. 

Many fortune 100's that use the product use K8s nowadays and AppD needs to step it up and realize it. 

I have had to revert to DataDog to monitor my infrastrutre correctly and reserve AppD strictly for APM only which is disappointing. 

[ben.coughlan]

Cheers Jeremy, 

I've been told that k8's integration will be here in April or so, but obviously, that isn't a hard date. 

On the surface, it doesn't look like it's doing anything particularly intuitive, just adds another layer to label pods, which is a bit disappointing. 

As for this: "The best you can currently do is to deploy the AppD agent in EACH contrainer you deploy, which is not the best solution to the issue. ", I raised the same concerns, if I wanted to monitor Kafka or an nginx container, I'd have to build appd into the container, which is a crazy business model, not at all practical. 

Are you also using Coreos by any chance? I can't see host level processes or volumes from the machine agent running in a docker container, not sure if it's exclusively a coreos bug or not. 

[jeremy.hall]

We're actually in the middle of moving to a CoreOS model - we're moving out of ACS and the insane workarounds needed to get something to work right in Azure K8s deployments. 

If you can't see the processes - do you mean the host level on the container itself?  From my experiance you should be able to see them.  I would submit a report to support and see if they can verify this as a known issue or not. 

[ben.coughlan]

Yeah, I can't see them at the host level, it show's the proc's for the agent itself, and that's it. 

I've submitted a report but I'm not sure the agent understood it. I'll follow it up I guess. 

Good luck with Azure!

Cheers, 

Ben

[mia.avery99]

. A Kubernetes Service for a given Pod will son let you expose an array of ports. A Marathon group of apps does let you expose only a single port.

3. Kubernetes Services is a much more elegant solution to haproxy but this can be a subjective opinion.

4. I prefer Kubernetes Replica Controller concept instead Marathon Constrains concept. Is also more elegant in my opinion. However, I think Constrain is a much flexible solution.

Kubernetes is an open source project that introduces Google's cluster management tools into virtual machines and bare metal scenes. It runs perfectly in modern operating system environments such as CoreOS and Red Hat Atomic and provides lightweight computing nodes that can be controlled by you. Kubernetes is developed using Golang, with lightweight, modular, portable and scalable features. We (Kubernetes development team) are working with a number of different technology companies, including the mesosphere that maintains the Mesos project, to upgrade Kubernetes to a standard way to interact with computing clusters. Kubernetes reworked Google's experience in building cluster applications.