Add data to a Parquet export safely ​

A Parquet export is a complete file with a schema and footer. Appending the bytes from another SELECT ... FORMAT parquet does not produce a single combined file. Treat each export as an independent artifact or create a new file containing the intended combined rows.

Rebuild a combined export ​

Load the existing file into a table with an explicit compatible schema. Insert the additional rows, validate the result, then export a new file using the native client's exact-byte output:

./coordinator client --database data \
  --query 'SELECT id, name FROM people ORDER BY id FORMAT parquet' \
  > people-new.parquet

Only replace the previous artifact after the command succeeds and the new file passes a read-back check. A shell redirect can create a partial file before a query reports an error; an existing filename is therefore not proof of success.

Keep files separate when appropriate ​

For periodic exports, choose a unique filename per batch and maintain a manifest outside SQL. Consumers can process those files as separate datasets. Coordinator DB's server file() source reads one confined regular file; it does not expand path patterns or append files.

See Parquet format, native client output and file source.