Recovery by rolling back unwanted writes in-place
To keep the original table in service, you can roll back unwanted writes directly rather than restoring a full table. You can roll back all the writes or only those writes that you identify as mistakes.
This approach uses the incremental export to Amazon Simple Storage Service (Amazon S3) feature. An incremental export writes the changed items to Amazon S3. The export includes each item's old and new images. The time period can be as short as 15 minutes or as long as 24 hours. The time period must be contained within the PITR recovery window.
To roll back unwanted writes, follow these steps:
-
Perform an incremental export to Amazon S3. Specify the time period during which the erroneous writes were happening to the table. Select an export view of "new and old images". If the time period exceeds 24 hours, do multiple incremental export jobs that together cover the full time period exactly.
-
Drive a bulk job off the Amazon S3 data to reverse the changes listed in the export by writing the old item value back. Optionally apply a transformation to control exactly what gets undone and how (such as to correct minor mistakes on the items).
This approach provides the following advantages:
-
More cost effective, especially for large tables. Instead of restoring a full table to undo a small number of mistaken writes, you focus on just the mistakes.
-
Faster, especially for large tables. Only a subset of table data must be processed.
-
Live, in-place correction. The original table remains active during the correction. No need to adjust a new table's metadata, settings, and external references.
