Since the deployment of RC2 the CPU usage on the secondaries has increased from around 20% to around 70%:  Looking at the tuple statistics one particular secondary shows this pattern:  The big increase here is the number of sequentially scanned rows in the `users` table. This pattern occurs on both secondaries. This increase in load suggests we're doing something bad, something that might even affect availability if we don't solve it. cc @selfup @timothyandrew @smcgivern @DouweM @rspeicher