If the replica is down and durable writes are blocking, disarm synchronous replication so the primary keeps serving (async). It auto-re-arms when the replica returns. The primary also auto-disarms after ~2 min — this is the "do it now". Safe to run twice (no-ops if already async).