upsert performance improvement

[lucagrulla]

The current approach for upsert is to delete existing rows that are going to be reimported using a join between the staging stable and the target table to identify which rows have to be deleted:

DELETE
FROM
    my_table USING my_table_staging
WHERE
    my_table.id=my_table_staging.id

This is not performing very well.
With a destination table of ~4GB and a ~60MB of import data the query takes ~255 seconds to run.

When we changed the delete query from a join to a sub-select:

DELETE
FROM
    my_table
WHERE
    id in (select id from my_table_staging)

with the same volume of data(~4GB destination table and ~60MB file) we reduced the query time to ~1.20s.
When trying to delete from a ~4GB staging table to a 4GB target table we observed a ~1.53s execution time.