See https://docs.google.com/document/d/16PFU3OpjMvnYjILAY00FlnabfUyIOWOPR-1tTFh_I10/edit?ts=60129a42 This corresponds with phase 3 of the linked document. The incremental engine's scope needs to be reworked so that it only compares received data to cached data, thus outputting the new, deleted, and updated data to lower layers. The lower layers then need to operate on this "diff" data. This issue may need to be divided into subtasks based on individual data types handled by the incremental engine. However, issue #1922486 needs to be completed first so that the data types are known before this can be subdivided.
Prioritizing this "low" because since this issue was opened, the idea of an ovn-controller refactor has fallen way down the list of important things. Changes in ovn-controller have made it perform light years better than it used to. While it's still in need of cleanup, we don't also have the increased performance angle to lean on. In fact, it's possible that the refactor on its own would cause performance regressions initially, since the focus would mainly be on writing error-free clean code.