Per today's 1x1 with @RKemper , @EBernhardson and myself, we talked about what a new outage recovery process might look like.

Today's process uses a mediawiki extension script, ForceSearchIndex.php to read from the database and re-queue the writes.

We could potentially use another instance of the Cirrus Streaming Updater itself to do this, but it would be a one-off job...the type of operation we typically do with cookbooks. That type of job doesn't seem to fit the helm paradigm.

Per IRC conversation with @RLazarus , he's done a lot of work towards this exact use case (see T341553 ).