Consumer Lag
How far behind the newest record a consumer is, expressed in time rather than in message count, and its rate of change.
Lag in messages is the number most systems report and the less useful of the two. Ten thousand messages behind means nothing without the rate: at a hundred thousand per second it is a tenth of a second, and at ten per second it is a quarter of an hour.
Lag in time — the age of the oldest unprocessed record — is what maps to a service level objective, because it is what a downstream user experiences.
More important than the level is the derivative. Constant lag means the consumer is keeping up with a steady offset. Growing lag means throughput is below arrival rate and the situation will not recover on its own — that is the alert worth paging on, and it is often visible long before the absolute number looks alarming.
Two behaviours to design for. A consumer restart produces a lag spike that is normal, so alerting must tolerate it or it will be muted. And per-partition lag matters more than the aggregate: a single partition falling behind because its key is hot is invisible in the total and is the common shape of a real incident.