[Bug]: Node memory should not include terminated containers

[bboreham]

What happened?

Prometheus, by default, will echo the last value of cAdvisor metrics for 5 minutes after they disappear.
This leads to artefacts where a pod restarts. To illustrate:

So this query for example will sum them and double-count:

kubernetes-mixin/dashboards/resources/node.libsonnet

Line 189 in eb631d8

'sum(node_namespace_pod_container:container_memory_working_set_bytes{%(clusterLabel)s="$cluster", node=~"$node", container!=""}) by (pod)' % $._config,

This can be fixed by turning on track_timestamps_staleness, added to Prometheus in v2.48, but you could also amend the queries. That example could change to:

sum(max by (cluster, namespace, pod, container)(node_namespace_pod_container:container_memory_working_set_bytes{%(clusterLabel)s="$cluster", node=~"$node", container!=""})) by (pod)

Please provide any helpful snippets.

What parts of the codebase are affected?

Dashboards

I agree to the following terms:

• I agree to follow this project's Code of Conduct.
• I have filled out all the required information above to the best of my ability.
• I have searched the issues of this repository and believe that this is not a duplicate.
• I have confirmed this bug exists in the default branch of the repository, as of the latest commit at the time of submission.

[skl]

@bboreham I added a bunch of deduplication statements to the dashboard container queries in #1130.