Details
-
Bug
-
Status: Open
-
Major
-
Resolution: Unresolved
-
None
-
None
Description
Consumer metrics are emitted at consumer level, and also with topic and partition as tags. for example `records-conumed-rate` is emitted per topic as well as overall.
Using the same metric names lead to metric systems rolling up both values, imposing challenges on users creating dashboards / alerts for higher level aggregations such as "Records consumed rate" across all instances of a particular app - which use:
sum(records-consumed-rate)
which would end up summing the rate per topic and overall leading to double counting the rate.
Attached a sample chart which shows both individual time series (with topic tag and without), which was discovered while checking consumer-lag vs consumed-rate based alerts.