Monitoring Data Prepper with metrics

You can monitor Data Prepper with metrics using Micrometer. There are two types of metrics: JVM/system metrics and plugin metrics. Prometheus is used as the default metrics backend.

JVM and system metrics

JVM and system metrics are runtime metrics that are used to monitor Data Prepper instances. They include metrics for classloaders, memory, garbage collection, threads, and others. For more information, see JVM and system metrics.

Naming

JVM and system metrics follow predefined names in Micrometer. For example, the Micrometer metrics name for memory usage is jvm.memory.used. Micrometer changes the name to match the metrics system. Following the same example, jvm.memory.used is reported to Prometheus as jvm_memory_used, and is reported to Amazon CloudWatch as jvm.memory.used.value.

Serving

By default, metrics are served from the /metrics/sys endpoint on the Data Prepper server in Prometheus scrape format. You can configure Prometheus to scrape from the Data Prepper URL. Prometheus then polls Data Prepper for metrics and stores them in its database. To visualize the data, you can set up any frontend that accepts Prometheus metrics, such as Grafana. You can update the configuration to serve metrics to other registries like Amazon CloudWatch, which does not require or host the endpoint but publishes the metrics directly to CloudWatch.

Plugin metrics

Plugins report their own metrics. Data Prepper uses a naming convention to help with consistency in the metrics. Plugin metrics do not use dimensions.

1. AbstractBuffer
   • Counter
     • recordsWritten: The number of records written into a buffer
     • recordsRead: The number of records read from a buffer
     • recordsProcessed: The number of records read from a buffer and marked as processed
     • writeTimeouts: The count of write timeouts in a buffer
   • Gaugefir
     • recordsInBuffer: The number of records in a buffer
     • recordsInFlight: The number of records read from a buffer and being processed by data-prepper downstreams (for example, processor, sink)
   • Timer
     • readTimeElapsed: The time elapsed while reading from a buffer
     • checkpointTimeElapsed: The time elapsed while checkpointing
2. AbstractProcessor
   • Counter
     • recordsIn: The number of records ingressed into a processor
     • recordsOut: The number of records egressed from a processor
   • Timer
     • timeElapsed: The time elapsed during initiation of a processor
3. AbstractSink
   • Counter
     • recordsIn: The number of records ingressed into a sink
   • Timer
     • timeElapsed: The time elapsed during execution of a sink

Naming

Metrics follow a naming convention of PIPELINE_NAME_PLUGIN_NAME_METRIC_NAME. For example, a recordsIn metric for the opensearch-sink plugin in a pipeline named output-pipeline has a qualified name of output-pipeline_opensearch_sink_recordsIn.

Serving

By default, metrics are served from the /metrics/sys endpoint on the Data Prepper server in a Prometheus scrape format. You can configure Prometheus to scrape from the Data Prepper URL. The Data Prepper server port has a default value of 4900 that you can modify, and this port can be used for any frontend that accepts Prometheus metrics, such as Grafana. You can update the configuration to serve metrics to other registries like CloudWatch, that does not require or host the endpoint, but publishes the metrics directly to CloudWatch.