Retry & backoff simulator

Critical

2 of 5 attempts happen after the caller has given up

The caller's deadline is 30s, but attempt 4 does not even start until 37s. Those attempts cannot help anyone — the response has nowhere to go — while still consuming a connection, a thread and a database handle on a system that is already failing. This is the configuration that turns a slow dependency into an outage.

Critical

One request at the top becomes up to 125 at the bottom

5 attempts at each of 3 layers multiply rather than add. At 50 req/s entering the chain, the deepest service can see 6,250 req/s — 99% of which is retries. Each layer's policy looks reasonable on its own; nobody owns the product. This is why a small dependency blip becomes a full outage.

Critical

No jitter — every client retries at the same instant

Clients that fail together back off by identical amounts and therefore return together. The recovering service is hit by the entire population at 1s, fails again, and the herd re-forms. Full jitter — a random wait between zero and the computed backoff — spreads the same load across the window and costs nothing.