Sometimes deleting some parts of the data history is intended, e.g. if there are semantic errors in the data that you don't like to see being reproduced when restored. In order to apply corrections against the complete audit trail there are two options: Deleting or updating a key hence a column.

## Delete a key from the `row_log`

Call `pgmemento.delete_key` and pass the audit_id, column name and old value (any type). This removes the given key-value pair from the JSONB logs. The subsequent value will then also cover the life time of the old value. Keep in mind that this action can also produce emtpy logs in the `row_log` table.

```sql
SELECT pgmemento.delete_key(555, 'column_B', 'old_value'::text);
```

The audit trail in the `new_data` column of the `row_log` table is also covered by the `delete_key` function. If you delete a key-value pair from a DELETE event log, it will also be removed from the `new_data` log where the value has been introduced. If you delete a key from an UPDATE event log, it will also be removed from the `new_data` log of the same event as if the column wasn't updated during that transaction. The new value for the given key then replaces the old one (you have just removed from the `old_data` log) in the `new_data` log, where it has been introduced.

**Before:**
| old_data                  | new_data                   |
| ------------------------- | -------------------------- |
| ...                       | ...                        |
| {"column_B": "ancient"}   | {"column_B": "old_value"}  |
| ...                       | ...                        |
| {"column_B": "ancient"}   | {"column_B": "new_value"}  |

**After:**
| old_data                  | new_data                   |
| ------------------------- | -------------------------- |
| ...                       | ...                        |
| {"column_B": "ancient"}   | {"column_B": "new_value"}  |
| ...                       | ...                        |
| {}                        | {}                         |

As you can see in this example, deleting a key from the `row_log` with `pgmemento.delete_key` can produce an empty JSONB. This can have a negative side effect when reverting some events that expect the existence of that specific key, e.g. DELETE, TRUNCATE, ALTER COLUMN, DROP COLUMN. In this case the **restored value will be NULL**. Reverting an ALTER COLUMN event would then transform existing data into NULL.

## Update a key in the `row_log`

Call `pgmemento.update_key`, pass the audit_id, the path to the key in the form of an array, the old and new value which can be of any type.

```sql
SELECT pgmemento.update_key(555, '{column_B}', 'old_value'::text, 'corrected value'::text);
```

Both, the `old_data` and `new_data` columns will be updated, no matter if logging for these columns is active or not.