The original idea was to made all broker-side metrics start with the same prefix to get a similar behavior than client-side metrics (all producer metrics start with kafka_producer_, same for the consumer, Streams and Connect).
However this creates some issues:
The Kafka metrics from the broker already start with kafka_server_, so these currently become kafka_server_kafka_server_<METRIC>, for example: kafka_server_kafka_server_replicamanager_leadercount.
The Yammer metrics retrieved through JMX did not have the kafka_server_ prefix. For example we had kafka_controller_controllerstats_uncleanleaderelections_total. Currently the reporter creates kafka_server_kafka_controller_controllerstats_uncleanleaderelectionspersec_total for that metric. Dropping the prefix could ease the transition.
For these 2 reasons, I propose removing the kafka_server_ prefix we add to Yammer metrics.
We currently inject the
kafka_server_
prefix to all Yammer metrics. https://github.com/strimzi/metrics-reporter/blob/main/src/main/java/io/strimzi/kafka/metrics/yammer/YammerMetricWrapper.java#L44The original idea was to made all broker-side metrics start with the same prefix to get a similar behavior than client-side metrics (all producer metrics start with
kafka_producer_
, same for the consumer, Streams and Connect).However this creates some issues:
kafka_server_
, so these currently becomekafka_server_kafka_server_<METRIC>
, for example:kafka_server_kafka_server_replicamanager_leadercount
.kafka_server_
prefix. For example we hadkafka_controller_controllerstats_uncleanleaderelections_total
. Currently the reporter createskafka_server_kafka_controller_controllerstats_uncleanleaderelectionspersec_total
for that metric. Dropping the prefix could ease the transition.For these 2 reasons, I propose removing the
kafka_server_
prefix we add to Yammer metrics.Here is a gist comparing both the current output and the output if the remove the prefix: https://gist.github.com/mimaison/6731c7f088080e4d766c87f3280e2814