Comparison
Consumer lagvsIngestion lag
Consumer lag
the producers are writing faster than the consumers are reading and the backlog has been growing all afternoon.
How far behind the head of the stream a consumer group is, or how many messages are waiting in a queue. It is the single best health metric for an async system, because it captures capacity, failure and slowdown in one number. Alert on the trend rather than the absolute value: a constant lag is a working system, a rising one is an outage in about an hour.
Full entry →Ingestion lag
the row was committed in the source at 09:00 and appeared in the warehouse at 09:41, and nobody could say which hop ate the forty minutes.
The delay between a change happening in the source and it being queryable at the destination. It is the number the business actually feels, and it is a sum over every hop, so it can only be diagnosed if each stage stamps its own arrival time. Measuring it end to end rather than per component is the difference between 'the pipeline is slow' and 'the connector is fine, the merge step waits for a lock'.
Full entry →