Aurora read replicas can be set up with different failover priorities (Tier0 – Tier9 ...).
Can the failover logic in the driver be modified to detect the read replica with higher priority (Tier0 over Tier 1 over Tier 2 ....) and switch writes to that read node.
In our initial tests we are seeing that to be up faster than when the failover is reported as complete on the cluster (seconds instead of 10s of seconds) - so there might be some value there but we need more tests to confirm.
That way the effective failover time might be faster than what is reported by Aurora.
If there are multiple read nodes with the same priority we might have to cycle through them - so it might be slightly more complex.
This might be a custom code on our side and not a generic fix.
However, it will be good to know if someone else have tried anything along these lines.