Add dashboard for Kubernetes pods

[jeffkreeftmeijer]

Much like #64, but for pods. Part of appsignal/appsignal-kubernetes#2.

[matsimitsu]

Any idea how many additional metrics this would generate, asking for a friend..

[unflxw]

As mentioned in #64, it would be nice to have human readable descriptions.

[jeffkreeftmeijer]

Any idea how many additional metrics this would generate, asking for a friend..

Hm, tell your friend we're not sure as of right now. In theory, a Kubernetes cluster could have a lot of pods, much like a non-Kubernetes setup could have a lot of hosts. Would not knowing this be a blocker to push this out?

[matsimitsu]

Hm, tell your friend we're not sure as of right now. In theory, a Kubernetes cluster could have a lot of pods, much like a non-Kubernetes setup could have a lot of hosts. Would not knowing this be a blocker to push this out?

Not nessicarily, it's mainly that we know lots of short-lived hosts(pods) could be an issue in terms of fetching the metrics. It's a weak point in the way the current custom metrics system works, that if you have one metric name (e.g. pod_load_avg and it has a huge number of individual tags (pod-xxx) that it could time out.

Also in the ingestion side we have a 4-megabyte document limit for MongoDB, and we know that customers with a ton of short-lived unique hosts hit that limit regularly. (The way it currently works is that the entire 4+ megabytes will be discarded, creating a gap in all the charts).

This doesn't block releasing this dashboard, but it's good to know the implications :)

[backlog-helper]

• This Pull Request needs more reviews. @tombruijn @luismiramirez - (More info)

---

This is a message from the daily scheduled checks.

New issue guide | Backlog management | Rules | Feedback

[jeffkreeftmeijer]

@matsimitsu thanks for the explanation here. Let's keep an eye on this.

I'm merging this now. Could you ping me when this is deployed? :)

[matsimitsu]

@jeffkreeftmeijer it's been deployed!