promql/impossible

This check will report PromQL queries that can never return anything.

Example query:

foo{job="bar"} unless sum(foo)

The right hand side (unless sum(foo)) cannot ever match any time series returned by the left hand side (foo{job="bar"}) because sum() strips away all labels. This means that we end up with left hand side having at least the job label, while the right hand side has no labels, so we end up with:

foo{job="bar"} unless {}

Both sides can only match if they have the same label set, see Prometheus docs. A result with {job="bar"} will never be matched with empty label set {}.

This check will also detect aggregations on labels that are already removed. Example:

group by(cluster) (
  sum(errors)
)

In the above sum(errors) removes all labels from the results, so there will be no cluster label to aggregate in the group by(cluster) outer query.

Similarly this check will warn about joins using labels that are already removed, example:

sum(foo) / on(cluster) count(bar)

Since both sum(...) calls remove all labels there won’t be any cluster label to join on.

Another problem detected by this check is binary aggregations that block labels from propagating to the results. Consider this query:

service_ready * on (cluster) group_left(env, location) cluster_info

What happens above is that we are joining our results for the service_ready series with an extra time series cluster_info and add two labels (env and location) to query results. But what happens if someone wanted to further filter down the results of service_ready with a filter like:

service_ready and on (instance) (service_errors == 0)

We can simply add it to the query:

service_ready and on (instance) (service_errors == 0) * on (cluster) group_left(env, location) cluster_info

But doing it this way splits our query in two parts:

• A: service_ready
• B: (service_errors == 0) * on (cluster) group_left(env, location) cluster_info

When we the evaluate A and on(instance) B we no longer propagate env and location to the results, because these labels are only added to the results on the B side. One way to fix this is to ensure that we join extra labels with the entire service_ready and on (instance) (service_errors == 0) expression by wrapping it in parentheses.

(service_ready and on (instance) (service_errors == 0)) * on (cluster) group_left(env, location) cluster_info

Configuration

This check doesn’t have any configuration options.

How to enable it

This check is enabled by default.

How to disable it

You can disable this check globally by adding this config block:

checks {
  disabled = ["promql/impossible"]
}

You can also disable it for all rules inside given file by adding a comment anywhere in that file. Example:

# pint file/disable promql/impossible

Or you can disable it per rule by adding a comment to it. Example:

# pint disable promql/impossible

How to snooze it

You can disable this check until given time by adding a comment to it. Example:

# pint snooze $TIMESTAMP promql/impossible

Where $TIMESTAMP is either use RFC3339 formatted or YYYY-MM-DD. Adding this comment will disable promql/impossible until $TIMESTAMP, after that check will be re-enabled.