Retries Without a Stampede

1. The Promise and the Trap

A retry means calling again after a call fails. The bet is that the failure was transient. Transient means temporary and self-healing: a dropped packet, one bad instance behind the load balancer, a two-second database failover.

When the bet is right, retries feel like magic. The user never learns that anything went wrong.

Here is the trap. A retry is a load multiplier aimed at whatever is already failing.

Say Payments starts to struggle. Every caller reacts by calling it three times instead of once