Retries Multiply

A request passes through gateway, then service A, then service B, then a database. Each of the three upper hops independently attempts a failed call up to 3 times. During a brownout where the database is slow but alive, what does the database see, and what is the usual mitigation?

Unchanged load, because upper layers time out before their retries land; the fix is raising them

Up to 27 attempts, because retries multiply down the stack; cap retries at one layer or use a budget

Up to 9 attempts, because retries add across the three layers; the fix is a shorter timeout per hop

Up to 3 attempts, because only the innermost caller sees the failure; the fix is a longer backoff