Runnable companion to How to convert CSV to Parquet.

One command, no server, no upload. Types are inferred from the CSV and carried into the Parquet schema; you choose the compression codec and row-group size.

./generate.sh   # writes orders.csv (20 rows) and orders_large.csv (~132 MB)
./run.sh        # runs every command from the article; compare with expected_output.txt

The one-liner:

clickhouse local -q "SELECT * FROM file('orders.csv') INTO OUTFILE 'orders.parquet' TRUNCATE FORMAT Parquet"

Covered in run.sh: the one-line conversion, the inferred Parquet schema, picking the codec (output_format_parquet_compression_method) and row-group size, overriding types before writing, a zstd-vs-lz4 size comparison on the ~132 MB file, and a best-of-3 conversion-throughput number.

Requirements: clickhouse (install with clickhousectl: curl https://clickhouse.com/cli | sh then clickhousectl local use latest), invoked as clickhouse local.

Prefer Python? run.py / run.ipynb do the same conversion in-process with chDB (pip install chdb).