mimirtool rules prepare: do not aggregation label to on() clause if already present in group_left() or group_right()

[pracucci]

What this PR does

During the weekend, while on-call, I supported an engineer at Grafana Labs which hit an edge case in mimirtool rules prepare command. The tool is used to add an aggregation label (defaults to cluster) to any aggregation and binary operator vector matcher of each rule.

The edge case we found is that when the aggregation label is already present in group_left/right() it shouldn't be added to on() too, because incorrect. A PromQL query with the same label both specified in on() and group_left/right() doesn't even parse (parsing it returns the error: label "cluster" must not occur in ON and GROUP clause at once).

Example

As an example, consider the following original query:

count by(cluster, namespace) (label_replace(cortex_build_info, "cluster", "", "cluster", "(.*)"))
* on(namespace) group_left(cluster) group by(cluster, namespace) (cortex_build_info)

With the code in main, mimirtool rules prepare rewrites it as follow:

count by(cluster, namespace) (label_replace(cortex_build_info, "cluster", "", "cluster", "(.*)"))
* on(namespace, cluster) group_left(cluster) group by(cluster, namespace) (cortex_build_info)

But it's invalid, and a subsequent parsing will return label "cluster" must not occur in ON and GROUP clause at once. With the modified code in this PR, mimirtool rules prepare would keep the original query as is, because cluster label is already part of group_left() and so shouldn't be added to on() too.

Which issue(s) this PR fixes or relates to

N/A

Checklist

• Tests updated.
• Documentation added.
• CHANGELOG.md updated - the order of entries should be [CHANGE], [FEATURE], [ENHANCEMENT], [BUGFIX].
• about-versioning.md updated with experimental features.

[rafa-munoz]

Thanks for taking the time to fix it