Agent Skills: Data Pipeline Patterns Skill

Data Pipeline Patterns Skill

Use this skill when implementing:

• Data processing pipelines
• Streaming data transformations
• Backpressure handling
• ETL operations
• Real-time event processing
• High-throughput data ingestion

When to Use

• Processing streams of data with backpressure
• Building ETL pipelines
• Real-time event streaming
• Multi-stage data transformations
• Rate-limiting external API calls

GenStage

Basic Producer

defmodule MyApp.Producer do
  use GenStage

  def start_link(_opts) do
    GenStage.start_link(__MODULE__, :normal)
  end

  @impl true
  def init(:normal) do
    {:producer, state}
  end

  @impl true
  def handle_demand(:start, _from, state) do
    # Emit data events
    {:noreply, [], state}
  end
end

ProducerConsumer

defmodule MyApp.ProducerConsumer do
  use GenStage

  def start_link(_opts) do
    GenStage.start_link(__MODULE__, :normal)
  end

  @impl true
  def init(:normal) do
    {:producer_consumer, state}
  end

  @impl true
  def handle_events(events, _from, state) do
    # Process each event
    {:noreply, state}
  end

  @impl true
  def handle_demand(:start, _from, state) do
    # Request more events from producer
    {:noreply, 100, state}
  end
end

Broadway

Data Processing Pipeline

defmodule MyApp.Pipeline do
  use Broadway

  def start_link(opts) do
    Broadway.start_link(__MODULE__, opts)
  end

  @impl true
  def init(opts) do
    Broadway.Topic.producer_name(opts)
  end

  @impl true
  def handle_message(message, metadata) do
    # Process each message
    message
    |> Broadway.Message.update_data(&transform/1)
    |> ack()
  end

  defp transform_1(data) do
    # Transformation logic
    data
  end
end

Backpressure Handling

# GenStage automatic backpressure
@max_demand 10  # Consumer asks for at most 10 events

# Broadway automatic backpressure with :concurrency
@concurrency 4  # Process up to 4 messages concurrently

Patterns

1. Fan-Out

# One producer, multiple consumers
defmodule MyApp.FanOut do
  use GenStage

  def start_link do
    {:producer, [consumer1: :consumer1, consumer2: :consumer2]}
  end
end

2. Fan-In

# Multiple producers, one consumer
defmodule MyApp.FanIn do
  use GenStage

  def start_link do
    {:consumer, [producer1: :producer1, producer2: :producer2]}
  end
end

3. ETL Pipeline

defmodule MyApp.ETL do
  use GenStage

  # Extract
  defmodule Extractor do
    use GenStage
    # Extract data from source
  end

  # Transform
  defmodule Transformer do
    use GenStage
    # Transform data
  end

  # Load
  defmodule Loader do
    use GenStage
    # Load data to destination
  end
end

Best Practices

• Always handle backpressure - Never ignore demand
• Use supervisors - Ensure proper supervision trees
• Handle failures - Implement proper error handling
• Test with real data - Test with production-like data volumes
• Monitor metrics - Use Telemetry to monitor pipeline performance

Token Efficiency

Use for:

• Streaming data processing
• ETL operations
• Real-time event handling
• Backpressure-controlled pipelines

Savings: ~40% vs manual pipeline implementation

Tools to Use

• GenStage: Built-in Elixir/OTP
• Broadway: Elixir/Erlang data processing
• Flow: Parallel data transformation
• Telemetry: Metrics and monitoring