The rename writes to Postgres and enqueues a reindex, and the queue is where the minute goes. Two options, and they are not the same trade.

Reindex inline on rename: correct immediately, but the rename request now waits on the search cluster and fails when it is slow. Or keep the queue and patch the local result set optimistically, so the person who renamed it sees the new title at once and everyone else catches up on the next tick.

Agreed. That makes it a client change only, which is the smaller diff and the one that cannot take the rename endpoint down with it.