Fast failover is the headline reliability feature of Amazon Aurora, and it works as advertised. The question worth asking is what your application does during those seconds, because that behavior belongs to your code rather than to the database.

Seconds are long enough to matter, and Gremlin lets you watch them happen on your schedule. With Gremlin, you can:

• Rehearse a failover before AWS runs one for you. A blackhole experiment against the cluster endpoint reproduces exactly what a promotion looks like from the client side, including how long your connection pool takes to notice a dead socket.
• Find out whether you need a restart. Some applications reconnect cleanly. Others hold a stale endpoint until someone cycles the service, which turns a thirty-second event into a much longer one.
• Check where stale reads reach. A latency experiment simulates a lagging reader endpoint, which is fine for a dashboard and not fine for anything transactional.

Aurora will fail over on its own schedule eventually. It's worth having watched it happen at least once on yours.