monedula-metrics-reporter

monedula-metrics-reporter is a Kafka MetricsReporter plugin that runs inside the Kafka process and exports metrics over OTLP, by gRPC or HTTP, to an OpenTelemetry Collector. Kafka operations never wait for the collector, whether it is up, slow or unreachable (Fail-safe by design).

What it exports

• Kafka SPI metrics, the registry behind metric.reporters: client metrics and many broker request and network metrics.
• Yammer metrics on brokers: the legacy registry that holds broker-internal signals such as UnderReplicatedPartitions and the per-topic BrokerTopicMetrics.
• JVM runtime metrics: memory, garbage collection, threads, classes and CPU.
• KIP-714 client metrics on brokers: the telemetry that producers and consumers push to the broker, forwarded without changes to the client applications.
• Its own health: the self-monitoring metrics count successful and failed exports.

On a broker, every series carries the broker’s identity as kafka_cluster_id and kafka_node_id labels. The reporter reads Kafka’s metric registries directly and pushes OTLP, so there is no JMX rule file or scrape endpoint to maintain; How it compares covers jmx_exporter and the other options. The same JAR also runs inside Java clients.

Where to start

• New to the reporter: Run the Grafana quickstart starts Kafka, the reporter, a collector, Prometheus and Grafana on your machine.
• Installing it on a cluster: Install the reporter on brokers.
• Collecting producer and consumer metrics: Collect Kafka client metrics with KIP-714 without changing the clients, or Report metrics from Java clients from inside them.
• Understanding the series you get: Metric names and labels.
• Looking up a setting: the configuration keys.

Version and license

These docs describe monedula-metrics-reporter 0.10.0, the latest release; the changelog lists what the main branch has added since. The reporter is open source under the AGPL-3.0 license, and its source and the quickstart are on GitHub.