Examples

1. Cross-Source Join (Postgres + Parquet)

This is the "Hello World" of federation. Join hot transactional data (Postgres) with cold historical logs (S3 Parquet).

Scenario

You have: * users table in PostgreSQL (User metadata). * logs directory of Parquet files in S3 (Clickstream events).

Configuration (sources.yaml)

sources:
  - name: pg
    type: sql
    dialect: postgres
    connection: "postgres://user:pass@localhost:5432/app_db"

  - name: s3_logs
    type: s3
    bucket: "my-datalake"
    paths: ["logs/*.parquet"]
    # ... creds ...

Query

Join the s3_logs (implicitly registered as tables based on filename or folder) with pg.public.users.

query = """
    SELECT 
        u.email, 
        count(l.event_id) as event_count
    FROM strake.pg.public.users u
    JOIN strake.s3_logs.logs l ON u.id = l.user_id
    WHERE l.event_date > '2023-01-01'
    GROUP BY u.email
    ORDER BY event_count DESC
    LIMIT 10
"""

df = conn.sql(query)

2. Using trace() for Performance Tuning

If a query feels slow, use trace() to see if pushdown is working.

conn.trace("SELECT * FROM strake.pg.public.users WHERE id > 1000")

Output Analysis:

Look for TableScan.

• Good (Pushdown):

  TableScan: users projection=[id, email], filters=[id > 1000]
  

  The filter is part of the scan! It's happening inside Postgres.

• Bad (No Pushdown):

  Filter: id > 1000
    TableScan: users
  

  The filter is separate. Strake is scanning everything and filtering locally.