bug: promql/fragile should not warn when using without and joining with explicit labelset

[skl]

There is a PromQL alert rule in the kubernetes-mixin for CPUThrottlingHigh which looks like this (abbreviated for readability):

sum without (id, metrics_path, name, image, endpoint, job, node) (
  increase(container_cpu_cfs_throttled_periods_total[5m])
)
/ on (cluster, namespace, pod, container) group_left
sum without (id, metrics_path, name, image, endpoint, job, node) (
  increase(container_cpu_cfs_periods_total[5m])
)
> ( 25 / 100 )

Source:

• fix: lint failure on CPUThrottlingHigh alert kubernetes-monitoring/kubernetes-mixin#986

The warning from pint look like this:

Warning: Aggregation using without() can be fragile when used inside binary expression because both sides must have identical sets of labels to produce any results, adding or removing labels to metrics used here can easily break the query, consider aggregating using by() to ensure consistent labels. (promql/fragile)

I believe that the explicit usage of labels in the join on (cluster, namespace, pod, container) means that this query should not result in a fragile warning.

[prymitive]

Thanks, yes this is wrong.
I might go and remove the entire promql/fragile as it's pretty old, naive and can get a bit too spammy.

[skl]

Sounds fine to me, we have this check disabled at the moment anyway 👍

[prymitive]

promql/fragile does other checks as well, so just removing the aggregation logic from it #1299

[skl]

Thank you!