After resolving the immediate issue of T312747 there remains the small list of entities that may not have been edited since the fix and therefore not propagated to the queryservice.
A similar tactic to that used before may be possible: Marking some batches as no longer done (https://phabricator.wikimedia.org/T309692#7975936) but this time we need to be more careful since most of the "heavy" wikis have had lost of edits in this time and were not impacted by this issue.
This probably needs a reworking of the query updater for wbstack and how it interacts with the platform API to get batches etc.
At the same time it might be worth putting thought into how this works with multiple query backends per wiki, or with wikis split between multiple backends (or both)
For one off cases, and immediate need, the batches that were generated in the platform API during the outage can just be deleted, and the query updater will update all entities that changed during the outage.
This can now be resolved; we've now totally rebuilt all QS long since this happened and we have tooling for complete rebuilds now