Shuffle Sharding
Assigning each customer a random combination of workers rather than a fixed shard, so that any two customers rarely share their whole set.
Ordinary sharding puts a customer on one shard, so a customer causing damage takes down every other customer on that shard. Shuffle sharding assigns each customer a random subset — say two workers from a pool of eight — and the combinatorics do the work: there are 28 such pairs, so two given customers have roughly a 1-in-28 chance of overlapping completely.
The effect is that a single abusive or pathological tenant degrades only the small fraction of customers who share both of its workers, and with retries against the customer's other assigned worker, most of them see no impact at all. Scaling the pool or the subset size improves the odds sharply.
AWS has published this as one of the techniques behind Route 53 and other services. It is particularly effective for multi-tenant systems where one tenant's workload is unpredictable, and it composes with cell-based architecture rather than competing with it.