Quantiles names are now serialized without trailing zeros. So we now have labels like quantile="0.5". Previously the label was quantile="0.50".
I've added sum series to Summary metrics. They now have quantiles, count and sum. Sum series have the _sum suffix.
Removed _count suffix from Counter metrics. As per the Counter documentation, Counter metrics should use the _total suffix. Previously we were injecting _count and Prometheus was adding _total. So now we have kafka_server_kafka_server_brokertopicmetrics_bytesinpersec_total while before we had kafka_server_kafka_server_brokertopicmetrics_bytesinpersec_count_total.
The new client has a dedicated metric type of non-numeric metrics, InfoSnapshot. Under the covers this still emits a Gauge metric with a value set to 1, like we do today. This is because Prometheus does not support non-numeric metrics. It's preferable to use the dedicated type, however this new type uses the _info suffix. So now we have kafka_server_kafka_server_kafkaserver_clusterid_info{ClusterId="mHGu-1meQOSC_RpQHdCpOw"} while before we had kafka_server_kafka_server_kafkaserver_clusterid{ClusterId="mHGu-1meQOSC_RpQHdCpOw"}. The suffix helps identify metrics whose values should not be graphed (as they are constant) and where the useful information is in the labels instead.
Gauge metrics don't have a suffix anymore, as per the OpenMetrics specification. The previous Prometheus client was injecting a _total suffix. We now have kafka_server_txn_marker_channel_metrics_response, while previously we had kafka_server_txn_marker_channel_metrics_request_total.
The new client does not insert a trailing comma after the last label. We now have kafka_server_alter_partition_metrics_connection_close_rate{BrokerId="0"}, while we previously had kafka_server_alter_partition_metrics_connection_close_rate{BrokerId="0",}.
The new client allows exposing metrics in the OpenMetrics format and serialized with Protobuf. When Prometheus scrapes the endpoint, it negotiates the format using the Accept HTTP header. You can check different formats using the debug query parameter on the GET /metrics endpoint. For example:
/metrics?debug=openmetrics: View OpenMetrics text format.
/metrics?debug=text: View Prometheus text format.
/metrics?debug=prometheus-protobuf: View a text representation of the Prometheus protobuf format.
This introduces some changes:
quantile="0.5"
. Previously the label wasquantile="0.50"
.sum
series to Summary metrics. They now have quantiles, count and sum. Sum series have the_sum
suffix._count
suffix from Counter metrics. As per the Counter documentation, Counter metrics should use the_total
suffix. Previously we were injecting_count
and Prometheus was adding_total
. So now we havekafka_server_kafka_server_brokertopicmetrics_bytesinpersec_total
while before we hadkafka_server_kafka_server_brokertopicmetrics_bytesinpersec_count_total
.InfoSnapshot
. Under the covers this still emits a Gauge metric with a value set to 1, like we do today. This is because Prometheus does not support non-numeric metrics. It's preferable to use the dedicated type, however this new type uses the_info
suffix. So now we havekafka_server_kafka_server_kafkaserver_clusterid_info{ClusterId="mHGu-1meQOSC_RpQHdCpOw"}
while before we hadkafka_server_kafka_server_kafkaserver_clusterid{ClusterId="mHGu-1meQOSC_RpQHdCpOw"}
. The suffix helps identify metrics whose values should not be graphed (as they are constant) and where the useful information is in the labels instead._total
suffix. We now havekafka_server_txn_marker_channel_metrics_response
, while previously we hadkafka_server_txn_marker_channel_metrics_request_total
.kafka_server_alter_partition_metrics_connection_close_rate{BrokerId="0"}
, while we previously hadkafka_server_alter_partition_metrics_connection_close_rate{BrokerId="0",}
.The new client allows exposing metrics in the OpenMetrics format and serialized with Protobuf. When Prometheus scrapes the endpoint, it negotiates the format using the
Accept
HTTP header. You can check different formats using thedebug
query parameter on theGET /metrics
endpoint. For example:/metrics?debug=openmetrics
: View OpenMetrics text format./metrics?debug=text
: View Prometheus text format./metrics?debug=prometheus-protobuf
: View a text representation of the Prometheus protobuf format.Related resources: