Aggregating with by() and without()

By default, aggregation collapses all labels into one result — a single number. Usually that's not what you want. You need to preserve some dimensions to keep the result meaningful.

Imagine rate(http_requests_total{job="demo"}[5m]) returns these series:

A plain sum(...) would collapse everything into 135. Not very useful.

by() — keep ONLY these labels

by() tells Prometheus: "Group by these labels, throw away everything else."

sum by(instance) (rate(http_requests_total{job="demo"}[5m]))

Result — one value per instance:

All other labels (method, path, status) are discarded. Think of it like SQL's GROUP BY: SELECT instance, SUM(rate) GROUP BY instance.

without() — remove ONLY these labels

without() is the opposite: "Keep all labels EXCEPT these."

sum without(method, status) (rate(http_requests_total{job="demo"}[5m]))

Result — removes method and status, keeps instance and path:

When to use which?

• by() — when you care about a few specific labels and want to discard the rest. Example: "give me totals per instance".
• without() — when you want to remove a few noisy labels but keep everything else. Example: "I don't care about method or status, keep the rest".

Both produce the same grouping when they specify complementary label sets. Choose whichever is more concise and readable for your use case.