Organize exports in year and month folders ​

Coordinator DB writes query results to the client response. An external export job chooses local folders, filenames and any object-storage destination. There is no SQL command that automatically creates year and month prefixes in a remote bucket.

Define a bounded export ​

Choose the date column, timezone interpretation and inclusive/exclusive boundaries for one period. Use those same boundaries for the export and its validation query. Select explicit columns and an order when the consumer needs reproducible row ordering.

The job can create a local path such as exports/2026/10/orders.parquet, run native --query with FORMAT parquet, and write the response to a temporary filename. Check the command's exit status and validate the completed file before renaming or uploading it.

Track successful periods ​

Use the storage provider's authenticated tooling for upload. Record the period, source boundary and resulting object name in the job's own manifest. A folder name alone does not prove that an export completed or includes all expected rows.

If a period is too large for one response, split it into ordered key ranges with separate artifacts. Do not concatenate independent Parquet files into one binary file. Decide explicitly whether reruns replace a period or create a new version.

See client exports, Parquet exchange and request limits.