lookml-explore（Lookml 探索）

Instructions

1. Core Standards

1. Naming Convention: snake_case for the Explore name.
2. Required Parameters:

- description: 100% Coverage. Every Explore MUST have a description. - label: A user-friendly name for the Explore in the UI. - view_name: Defaults to explore name, but explicit definition is safer.

1. Joins:

- relationship: Required (one_to_one, many_to_one). - sql_on: Required. Use ${left.id} = ${right.id} syntax. - type: defaults to left_outer. Use inner or full_outer explicitly if needed.

1. Formatting:

- Do NOT use from to rename views just for aesthetics. Use view_label instead. - Exception: Polymorphic joins, Self-joins, Rescoping extensions.

2. Advanced Configuration

• always_filter: specific filters that users can change but cannot remove.
• sql_always_where: specific restrictions that users *cannot* change.
• persist_with: Link explore cache to datagroups (e.g., default_datagroup).
• fields: Use inclusive lists to strictly control content when necessary (ALL_FIELDS*, -view.field).

3. Performance Optimization (Aggregate Tables)

Aggregate Tables (Aggregate Awareness) allow Looker to query smaller, pre-aggregated tables instead of the raw granular data, drastically improving query performance.

Anatomy of an Aggregate Table

explore: orders {
  aggregate_table: rollup_name {
    query: {
      dimensions: [created_date, status]
      measures: [total_revenue, count]
      filters: [orders.created_date: "6 months"]
    }
    materialization: {
      datagroup_trigger: ecommerce_etl
      # partition_keys: ["created_date"] # BigQuery/Presto optimization
      # increment_key: "created_date"    # Incremental builds
      # increment_offset: 3              # Rebuild last 3 periods
    }
  }
}

Key Parameters

1. Query: Defines the "shape" of the rollup.

- Dimensions: Include all dimensions commonly used in dashboards (including filters). - Measures: Include base measures (sum, count). Looker can derive averages from sum+count. - Filters: Optional. Restricts the rollup to a subset of data (e.g., "last 6 months").

1. Materialization:

- datagroup_trigger: (Recommended) Rebuilds when the ETL job completes. - sql_trigger_value: Rebuilds when a SQL query returns a new value. - increment_key: (Advanced) Appends new data instead of full rebuilds. Best for massive tables. - indexes / partition_keys / cluster_keys: Dialect-specific optimizations.

1. Best Practices:

- Timeframes: Include the finest grain needed (e.g., date). Looker can roll up date to month or year automatically. - Exact Match: The user's query must be a *strict subset* of the aggregate table's fields to satisfy the awareness logic. - Filter Awareness: If a user filters on a field *not* in the aggregate table, Looker cannot use it (unless it's an "exact match" special case). Add common filter fields to the dimensions list.

4. Extending Explores

• Extends: Use extends: [base_explore] to inherit joins, fields, and descriptions from another explore.

- Use Case: Create a "Base" explore with common joins, then "Extended" explores for specific analysis (e.g., orders -> marketing_orders).

Examples

Basic Explore

explore: orders {
  label: "Orders"
  description: "Analyze order data, including user and product details."
  view_name: orders

  join: users {
    relationship: many_to_one
    sql_on: ${orders.user_id} = ${users.id} ;;
  }
}

Explore with Filters & Caching

explore: events {
  label: "Web Events"
  description: "User interaction events."
  persist_with: default_datagroup

  # Users can change this filter, but it defaults to '7 days'
  always_filter: {
    filters: [events.created_date: "7 days"]
  }

  # Users CANNOT change this filter.
  sql_always_where: ${events.is_test_data} = false ;;

  join: sessions {
    relationship: many_to_one
    sql_on: ${events.session_id} = ${sessions.session_id} ;;
  }
}

## Aggregate Table (Advanced)

explore: orders { aggregate_table: monthly_sales_summary { query: { dimensions: [created_month, status, products.category] measures: [total_revenue, count] filters: [orders.created_date: "2 years"] } materialization: { datagroup_trigger: ecommerce_etl partition_keys: ["created_month"] increment_key: "created_month" increment_offset: 1 # Rebuild current and previous month } } }

## Extended Explore

explore: orders_extended { extends: [orders] label: "Orders (Marketing View)" view_name: orders

# Add new joins specific to this view join: marketing_channels { sql_on: ${orders.channel_id} = ${marketing_channels.id} ;; relationship: many_to_one } }

Reference Skills

For more complex scenarios, refer to these specialized skills:

• Advanced Explore Configuration: UNNESTing, lateral flattens, and row-level security.
• Joins Deep Dive: Detailed join types, relationships, and aliasing.