Recent data plus working set is always in Postgres page format.

Historical data when pushed to s3 is in parquet. This happens async - not on the transaction hot path.

So older data below certain LSN is on s3 in parquet available to all analytics processing. Hot data is on page servers in page format for OLTP.

You can be smart in querying both representations for real time analytical queries