Parquet files

Parquet is columnar and compressed, which makes it the best of the file formats to actually query. AddisDB opens a .parquet file and reads only what your query needs from it.

Who it is for

Parquet stores data by column rather than by row, so an aggregate over one column never reads the other forty. Combined with strong compression, that typically makes it five to ten times smaller than the equivalent CSV and far faster to scan.

There is more in the file than the data: each row group carries min and max statistics per column, so a reader can skip whole groups that cannot match a predicate. That is why a filtered query over a large Parquet file often touches a fraction of it, and why the format is the storage layer under most lakehouses.

It is the standard storage format for data lakes and the export format every warehouse offers. If you are handling the same large dataset repeatedly, converting it to Parquet once pays for itself immediately.

Getting a file

1. Warehouse exports — Snowflake, BigQuery, Databricks, Redshift — all offer Parquet output.
2. Data lakes on S3, GCS or ADLS are usually Parquet already.
3. Convert a CSV yourself with the DuckDB CLI in one statement.
4. Partitioned exports arrive as many files; open the one you need, or combine them first.
5. Compression is chosen at write time — zstd is the sensible default now, snappy the older one.

-- Convert a CSV to Parquet with the DuckDB CLI
COPY (SELECT * FROM read_csv_auto('big.csv'))
  TO 'big.parquet' (FORMAT PARQUET);

-- Combine a partitioned export into one file first
COPY (SELECT * FROM read_parquet('exports/part-*.parquet'))
  TO 'combined.parquet' (FORMAT PARQUET, COMPRESSION ZSTD);

Open it in AddisDB

1. New Connection → Files → Other → Parquet.
2. Choose the .parquet file, name the connection, and Save.

What AddisDB gives you

• Full analytical SQL, with the schema and real types read from the file’s own metadata.
• The Chart view — Parquet files are usually opened to aggregate something, and this is where the answer goes.
• Notebooks for analysis worth keeping, and AI chat grounded in the file’s columns.
• ⌘K search across every connection.
• A source file that is never written to.