Details
-
New Feature
-
Status: Open
-
Major
-
Resolution: Unresolved
-
None
-
None
-
None
Description
The current
{records/bytes}-consumed-rate metrics are useful to determine the actual rate of consumption, but I've seen these be misinterpreted to mean the number/size of records actually processed – in general these should be the same but if a consumer drops out of the group, a consumed record will not be committed and upon rejoin this same record will be re-consumed. This means the actual rate of processing will be lower than the X-consumed-rate metric.
We should consider adding a records-committed-rate and bytes-committed-rate metric for users who are interested in the actual processing rate, eg as an indication of "real" throughput