Retention and archive

Recent logs remain in the primary database. A scheduled job processes older complete days.

Plan behavior

Free projects prune aged rows after the configured grace period. Pro projects export eligible days to Parquet in S3-compatible object storage, verify the object, then delete hot rows when configured to do so.

Required configuration

Set ARCHIVE_ENABLED=true and provide the S3 endpoint, bucket, access key, and secret key. Optional settings control region, URL style, prefix, TLS, hot-window days, deletion after verification, timeouts, and maximum files per query.

An incomplete archive configuration stops archive work and leaves database rows in place.

Querying cold data

Archive search uses DuckDB for a bounded date range. Only completed ledger partitions count as archived ownership, preventing a staged copy from being counted with the same hot rows.

./buddy archive:run