Actions Reference¶

Concept¶

What is an Action¶

An Action is the atomic unit of work in a Lakehouse Plumber (LHP) FlowGroup. A Pipeline contains one or more FlowGroups, and a FlowGroup contains an ordered list of Actions. Each Action declares one step — read a source, transform a view, persist a table, or assert a quality rule — and LHP compiles it into the corresponding Lakeflow Declarative Pipelines Python construct.

Action flow¶

        graph LR
    A[Load] --> B{0..N Transform}
    B --> C[Write]

Every FlowGroup starts with one Load, applies zero or more Transforms, and ends with one Write. Test actions attach data-quality expectations to any view along the way.

Minimum FlowGroup¶

This FlowGroup ingests a Delta source, renames a column, and writes the result to a streaming table:

pipelines/bronze/customer.yaml¶

pipeline: bronze_ingestion
flowgroup: customer

actions:
  - name: load_customer
    type: load
    readMode: stream
    source:
      type: delta
      database: samples.tpch
      table: customer_sample
    target: v_customer_raw

  - name: rename_customer_key
    type: transform
    transform_type: sql
    source: v_customer_raw
    target: v_customer
    sql: "SELECT c_custkey AS customer_id, c_name AS name FROM stream(v_customer_raw)"

  - name: write_customer_bronze
    type: write
    source: v_customer
    write_target:
      type: streaming_table
      database: "${catalog}.${bronze_schema}"
      table: customer

Action types reference¶

Actions come in four top-level types:

Type	Purpose
Load	Bring data into a temporary view (e.g. CloudFiles, Delta, JDBC, SQL, Python).
Transform	Manipulate data in one or more steps (SQL, Python, schema adjustments, data-quality checks, temp tables…).
Write	Persist the final dataset to a streaming_table or materialized_view.
Test	Validate data quality using DLT expectations in temporary tables (uniqueness, referential integrity…).

Where to start¶

Quickstart — build your first pipeline end-to-end in about ten minutes, including the load and write Actions used above.
Decisions — decision matrices for picking a load source, write target, streaming vs. batch mode, and write mode.