Prometheus remote write receiver

The Prometheus remote write receiver gets metrics from Prometheus instances using the Prometheus remote write 2.0 protocol. This receiver allows you to forward Prometheus metrics to the OpenTelemetry Collector for processing and exporting to Elasticsearch or other destinations.

To use the Prometheus remote write receiver, include it in the receiver definitions of the Collector configuration:

receivers:
  prometheusremotewrite:
    endpoint: 0.0.0.0:9090
		

You must also configure Prometheus to send metrics using the remote write 2.0 protocol. Refer to Configure your Prometheus instance for details.

The receiver configuration is based on the standard OpenTelemetry HTTP server configuration. The main option is:

For the full list of HTTP server options, refer to the confighttp documentation.

To send metrics from Prometheus to this receiver, configure Prometheus with the following settings.

Prometheus remote write relies on write-ahead log (WAL) records. By default, metadata information, such as metric type, unit, and help description, are not appended to the WAL. Because this information is required to translate remote write data into OTLP, turn on the metadata-wal-records feature flag when starting Prometheus:

				
					./prometheus --config.file config.yml --enable-feature=metadata-wal-records
		

This receiver only supports the Prometheus remote write v2 protocol. Configure your Prometheus remote_write block to use the v2 protocol:

remote_write:
  - url: "http://<collector-host>:9090/api/v1/write"
    protobuf_message: io.prometheus.write.v2.Request
		

Replace <collector-host> with the address where your EDOT Collector is running.

For more details on configuring Prometheus, refer to the upstream receiver documentation.

The remote write 2.0 protocol specification is still evolving, and compatibility depends on matching versions:

The Prometheus remote write receiver has some behaviors to be aware of:

• Summaries and Classic Histograms are unsupported. Classic Histograms and Summaries are composed of multiple separate time series, which can be sent in separate Remote Write requests. This makes it impossible to determine if all parts have been received. Use Prometheus Native Histograms instead.

• Resource metrics cache. The receiver uses an internal LRU cache to store target_info metrics across requests. The cache has a hardcoded limit of 1,000 resource metrics. If the process restarts, the cache is lost, which might cause some inconsistencies.

• Remote write v1 is not supported. The receiver only supports the remote write 2.0 protocol. Remote write v1 lacks support for metadata, created timestamps, and atomic histogram ingestion.

• Upstream component: prometheusremotewritereceiver
• Prometheus Remote Write 2.0 specification
• Configure metrics collection in EDOT