Amazon RDS is the dependency with the widest blast radius in most architectures. Dozens of services share it, so a few seconds of unavailability degrades everything at once. Connection pools exhaust, requests queue, and the pressure travels upstream faster than your team can respond to it.

Managed failover helps, but it's not instant. What happens to in-flight transactions, how quickly your pool notices a dead socket, and whether your application reconnects without a restart are all properties of your code. With Gremlin, you can:

• Rehearse a failover before AWS schedules one. Blackhole experiments make the RDS endpoint unreachable from your application, reproducing exactly what a failover looks like from the client side. You learn how long recovery takes and whether it needs human intervention.
• Find the timeouts that nobody tuned. Latency experiments slow the connection so you can watch requests back up. Most teams discover their database timeout is longer than the timeout of the service calling them, which turns one slow query into a cascade.
• Score every service that touches the database. Gremlin maps your dependencies automatically, so RDS shows up wherever it's used. Reliability Tests then score each dependent service, giving you a ranked list of what to fix first.

Nearly everything you run sits downstream of this database. Finding and fixing its failure modes is the best step you can take for your applications.