ContextQMD
Back to Connect

Message Multicast (Fan-Out)

.claude-plugin/plugins/redpanda-connect/skills/pipeline-assistant/resources/recipes/multicast.md

4.90.23.8 KB
Original Source

Message Multicast (Fan-Out)

Pattern: Routing - Multicast / Fan-Out Difficulty: Basic Components: kafka_franz, broker output, mapping Use Case: Send the same message to multiple destinations simultaneously

Overview

The multicast pattern delivers a single message to multiple recipients. This recipe shows how to fan out Kafka messages to multiple topics based on message content, enabling parallel processing by different consumers. Essential for building event-driven architectures where multiple services need the same data.

Configuration

See multicast.yaml for the complete configuration.

Key Concepts

1. Dynamic Destination List

Build a list of target topics based on message content:

bloblang
let target_topics = []

if (this.type.contains("A")) {
  let target_topics = $target_topics.append("topic_a")
}
if (this.type.contains("B")) {
  let target_topics = $target_topics.append("topic_b")
}

meta target_topics = $target_topics

The list determines which outputs receive the message.

2. Broker Output Pattern

The broker output with fan_out pattern sends to all targets:

yaml
output:
  broker:
    pattern: fan_out
    outputs:
      - kafka_franz:
          topic: topic_a
      - kafka_franz:
          topic: topic_b

All outputs receive the message simultaneously.

3. Metadata Preservation

Preserve source Kafka metadata for each destination:

  • Original partition key
  • Original timestamp
  • Custom headers

This maintains message ordering and traceability.

Important Details

  • Security: Use environment variables for broker addresses
  • Performance:
    • Messages sent in parallel to all destinations
    • fan_out pattern waits for all outputs to succeed
    • Use fan_out_sequential for ordered delivery
  • Error handling: If any destination fails, entire message fails (can be changed with drop_on)
  • Ordering: Preserved per-destination via partition key

Testing

bash
# Set environment variables
export KAFKA_BROKER=localhost:9092
export SOURCE_TOPIC=multicast_in
export CONSUMER_GROUP=multicast_cg

# Run the pipeline
rpk connect run multicast.yaml

# Send test messages
echo '{"data":"hello","type":"A"}' | rpk topic produce $SOURCE_TOPIC
echo '{"data":"world","type":"AB"}' | rpk topic produce $SOURCE_TOPIC
echo '{"data":"test","type":"ABC"}' | rpk topic produce $SOURCE_TOPIC

# Check destinations
rpk topic consume topic_a  # Should see all messages with "A"
rpk topic consume topic_b  # Should see messages with "B"
rpk topic consume topic_c  # Should see messages with "C"

Variations

Static Fan-Out (All Messages to All Topics)

yaml
output:
  broker:
    pattern: fan_out
    outputs:
      - kafka_franz:
          topic: topic_a
      - kafka_franz:
          topic: topic_b
      - kafka_franz:
          topic: topic_c

All messages go to all three topics.

Conditional with Drop on Error

yaml
output:
  broker:
    pattern: fan_out
    outputs:
      - kafka_franz:
          topic: topic_a
        drop_on:
          error: true  # Don't fail entire message if topic_a fails

Continue on partial failures.

Cross-System Multicast

yaml
output:
  broker:
    pattern: fan_out
    outputs:
      - kafka_franz:
          topic: kafka_destination
      - aws_s3:
          bucket: s3_destination
      - http_client:
          url: http://webhook

Fan out to different systems simultaneously.

References