Read and write hive-partitioned Parquet datasets in a DuckDB pipeline

Steps

1. Read partitioned data with SELECT * FROM read_parquet('orders/*/*/*.parquet', hive_partitioning = true) — DuckDB auto-detects hive-style 'key=value' folder naming and enables partitioning automatically, but set the flag explicitly in scripts.
2. Set hive_partitioning = false explicitly if partition columns should be excluded from the result set rather than surfaced as regular columns.
3. Rely on automatic filter pushdown on partition keys — a WHERE clause on a partition column prunes files at scan time with no extra config.
4. Write partitioned output with COPY <table> TO 'orders' (FORMAT parquet, PARTITION_BY (year, month)).
5. Since PARTITION_BY can't take expressions directly, pre-compute any derived partition columns in a subquery first, e.g. COPY (SELECT *, year(timestamp) AS year FROM services) TO 'test' (PARTITION_BY (year, month)).