Documentation Grafana Cloud Instrument and send data Grafana Alloy Reference Components otelcol otelcol.processor.batch

otelcol.processor.batch

otelcol.processor.batch accepts telemetry data from other otelcol components and places them into batches. Batching improves the compression of data and reduces the number of outgoing network requests required to transmit data. This processor supports both size and time based batching.

Grafana Labs strongly recommends that you configure the batch processor on every Alloy that uses OpenTelemetry (otelcol) Alloy components. Define the batch processor in the pipeline after the otelcol.processor.memory_limiter as well as any sampling processors. Batching should happen after any processing that drops data such as sampling.

You can specify multiple otelcol.processor.batch components by giving them different labels.

Usage

alloy Copy

otelcol.processor.batch "<LABEL>" {
  output {
    metrics = [...]
    logs    = [...]
    traces  = [...]
  }
}

Arguments

You can use the following arguments with otelcol.processor.batch:

otelcol.processor.batch accumulates data into a batch until one of the following events happens:

• The duration specified by timeout elapses since the time the last batch was sent.
• The number of spans, log records, or metric data points processed reaches or exceeds the number specified by send_batch_size.

send_batch_size acts as a trigger threshold, not the exact batch size. When data arrives in large chunks, the actual batch size may exceed send_batch_size unless you configure send_batch_max_size to enforce an upper limit.

Logs, traces, and metrics are processed independently. For example, if send_batch_size is set to 1000:

• The processor may, at the same time, buffer 1,000 spans, 1,000 log records, and 1,000 metric data points before flushing them.
• If there are enough spans for a batch of spans, for example 1,000 or more, but not enough for a batch of metric data points, for example less than 1,000, then only the spans are flushed.

Use send_batch_max_size to limit the amount of data contained in a single batch:

• When set to 0, batches can be any size.
• When set to a non-zero value, send_batch_max_size must be greater than or equal to send_batch_size. Every batch contains up to the send_batch_max_size number of spans, log records, or metric data points. The excess spans, log records, or metric data points aren’t lost - instead, they’re added to the next batch.

For example, assume you set send_batch_size to the default 8192 and there are 8,000 batched spans. If the batch processor receives 8,000 more spans at once, its behavior depends on how you configure send_batch_max_size:

• If you set send_batch_max_size to 0, the total batch size would be 16,000 spans which are then flushed as a single batch.
• If you set send_batch_max_size to 10000, then the batch has a limit of 10,000 spans and the processor adds the remaining 6,000 spans to the next batch.

This demonstrates how send_batch_size acts as a trigger while send_batch_max_size enforces the actual maximum batch size.

metadata_cardinality_limit applies for the lifetime of the process.

Configure receivers with include_metadata = true so that metadata keys are available to the processor.

Each distinct combination of metadata triggers the allocation of a background task in the Alloy process that runs for the lifetime of the process, and each background task holds one pending batch of up to send_batch_size telemetry items such as spans, log records, or metric data points. Batching by metadata can therefore substantially increase the amount of memory dedicated to batching.

The maximum number of distinct combinations is limited to the configured metadata_cardinality_limit, which defaults to 1000 to limit memory impact.

Blocks

You can use the following blocks with otelcol.processor.batch:

output

debug_metrics

Exported fields

The following fields are exported and can be referenced by other components:

input accepts otelcol.Consumer data for any telemetry signal including metrics, logs, or traces.

Component health

otelcol.processor.batch is only reported as unhealthy if given an invalid configuration.

Debug information

otelcol.processor.batch doesn’t expose any component-specific debug information.

Debug metrics

• otelcol_processor_batch_batch_send_size_bytes (histogram): Number of bytes in each sent batch.
• otelcol_processor_batch_batch_send_size (histogram): Number of units in the batch.
• otelcol_processor_batch_batch_size_trigger_send_total (counter): Number of times a batch was sent due to a size trigger.
• otelcol_processor_batch_metadata_cardinality (gauge): Number of distinct metadata value combinations processed.
• otelcol_processor_batch_timeout_trigger_send_total (counter): Number of times a batch was sent due to a timeout trigger.

Examples

Basic usage

This example batches telemetry data before sending it to otelcol.exporter.otlp for further processing:

alloy Copy

otelcol.processor.batch "default" {
  output {
    metrics = [otelcol.exporter.otlp.production.input]
    logs    = [otelcol.exporter.otlp.production.input]
    traces  = [otelcol.exporter.otlp.production.input]
  }
}

otelcol.exporter.otlp "production" {
  client {
    endpoint = sys.env("OTLP_SERVER_ENDPOINT")
  }
}

Batching with a timeout

This example buffers up to 10,000 spans, log records, or metric data points for up to 10 seconds. Because send_batch_max_size isn’t set and defaults to 0, the actual batch size may exceed 10,000 if large amounts of data arrive simultaneously.

alloy Copy

otelcol.processor.batch "default" {
  timeout = "10s"
  send_batch_size = 10000

  output {
    metrics = [otelcol.exporter.otlp.production.input]
    logs    = [otelcol.exporter.otlp.production.input]
    traces  = [otelcol.exporter.otlp.production.input]
  }
}

otelcol.exporter.otlp "production" {
  client {
    endpoint = sys.env("OTLP_SERVER_ENDPOINT")
  }
}

Metadata-based batching

Batching by metadata enables support for multi-tenant OpenTelemetry pipelines with batching over groups of data having the same authorization metadata.

alloy Copy

otelcol.receiver.jaeger "default" {
  protocols {
    grpc {
      include_metadata = true
    }
    thrift_http {}
    thrift_binary {}
    thrift_compact {}
  }

  output {
    traces = [otelcol.processor.batch.default.input]
  }
}

otelcol.processor.batch "default" {
  // batch data by tenant id
  metadata_keys = ["tenant_id"]
  // limit to 10 batch processor instances before raising errors
  metadata_cardinality_limit = 123

  output {
    traces  = [otelcol.exporter.otlp.production.input]
  }
}

otelcol.exporter.otlp "production" {
  client {
    endpoint = sys.env("OTLP_SERVER_ENDPOINT")
  }
}

Compatible components

otelcol.processor.batch can accept arguments from the following components:

• Components that export OpenTelemetry otelcol.Consumer

otelcol.processor.batch has exports that can be consumed by the following components:

• Components that consume OpenTelemetry otelcol.Consumer