[kubernetes] add more metrics related to k8s deployments

[onbjerg]

Right now we can only filter based on replication controllers using kube_replication_controller. It is my understanding that ReplicaSet are also included in this.

It would be a lot better if you could filter based on kube_replica_set and/or kube_deployment. Specific metrics for these could prove valuable for monitoring, e.g. the number of pods running for a specific ReplicaSet/Deployment vs. the desired count. All of these values can be extracted from the k8s API.

Another nice addition would be the number of available and unavailable pods. A pod can be running, but not be available. A pod is only available when it's readiness probe succeeds, otherwise it is unavailable. These metrics could be called kubernetes.pods.available and kubernetes.pods.unavailable respectively.

My proposal is to add the following:

• kubernetes.pods.desired (for ReplicaSet, Deployments and ReplicationControllers)
• kubernetes.pods.available
• kubernetes.pods.unavailable
• kube_replica_set or/and kube_deployment tags

The addition of these metrics and tags would improve the overall experience with the k8s integration.

[onbjerg]

I might work on this if I have time later this week and if the proposal is 👍

[onbjerg]

The deployment statuses can be found in /apis/extensions/v1beta1/deployments.

[jonmoter]

+1 on kube_deployment tags

[onbjerg]

Waiting for merge #2728