Failover (primary -> replica promotion)
When the main database server dies, a standby copy steps in and takes over its job. There's a bumpy few minutes while the handover happens -- some requests fail -- and then service is back to normal. The crucial part: this only works if the customer paid for the standby in the first place.
Skip the replica and a dead database server means the customer is simply down until you get the original host back -- which could be hours. The replica is insurance: it costs the price of a second machine kept in sync, and it converts "we lost the database" into "we had a five-minute wobble." For a customer whose product is their database, that's an easy yes.
Detailed explanation
A managed DB with with_replica=true runs a primary VM + a replica VM on different hosts. On primary-host failure: a 5-minute failover window with ~50% request failure (DB_FAILOVER_FAIL_RATIO) while the replica promotes, then full recovery. No replica = full unavailability until the original host recovers -- no automatic path back.
Failover is the database-specific instance of a general pattern: keep a standby on an independent failure domain and promote it on loss. The redundancy tier the customer buys determines whether there's a replica at all, and multi-AZ placement determines whether the replica survives a whole-building outage rather than just a single host failure.