[Infra Monitoring UI] Verify that pods and container detail page works

[miltonhultgren]

While developing the metrics tables (which link to the node detail pages via the link-to routes) I noticed that the page for container details was empty and the page for pod details was throwing errors.

Since it was out of scope for my work and I suspected it could have been due to the data I had in my cluster I left it alone but it would be great to verify that those links/pages actually work.

[elasticmachine]

Pinging @elastic/infra-monitoring-ui (Team:Infra Monitoring UI)

[miltonhultgren]

Just had this happen to me on a Host page as well, but it's not persistent.

[crespocarlos]

What happens with hosts is most likely linked with #131308 problem. When it comes to pods and containers there are different problems going on.

For pods, link-to needs to receive the uid, but we are passing the pod name.

Now for containers the problem seems to be more complex, mostly because I don't really know how it is supposed to work. I have an oblt cluster setup and I see empty pages when I go to the details page from Metrics UI and also from the Metrics Table on APM.

What I have found is that we want to get kubernetes containers metric info, but we only have inventory models for docker containers, so the existing logic will try to get the data from docker object, but we need to look at kubernetes.container object, which is where the metrics data actually exists for this.

doc:

what's configured for node type container for the memory metric:

[miltonhultgren]

@crespocarlos That makes a lot of sense, it's similar in theme to #133125 (but in reverse).

@jasonrhodes This is a bit of an odd place to find ourselves in. For APM I think the main use case is containers in Kubernetes, that's why the original issue for this called out the Kubernetes fields for containers.
But our page assumes it's standalone Docker.

So should we do #133125 as well as create a new issue to extend the Node details page with a Kubernetes mode?
For #133125 we'll need to somehow flag which mode it is or keep two tables.

[jasonrhodes]

Good questions. I'm nervous about trying to solve this on its own when it's so intricately tied into the relationships in the entity model discussions. But I also don't want to bottleneck everything around that effort, either.

Can one of you create a ticket of type: Epic that documents "Relationship problems between infrastructure entities" or something like that, and link to the various tickets and conversations that we've been uncovering lately? It doesn't need to be in progress yet but I'll use that as a tracking ground. I don't think there's a ton we can do yet on it but at least we won't lose track of these things and can circle back to them asap once my entity model work moves forward a bit more.

If anyone has ideas for quick wins, those are absolutely welcome, as well.

[crespocarlos]

Here is the epic: https://github.com/orgs/elastic/projects/664/views/23 (first time creating an epic, please let me know if it's ok).

I was thinking about configuring the fields in inventory_models to consider the kubernetes.container fields, but it seems that we don't have disk nor network metrics for kubernetes.containers, which is considered for docker containers. So, they seem like different entities, I'm not sure if they should be handled differently

[crespocarlos]

I've created a follow up ticket to address the problem with k8s container details page: #134467