Graceful Degradation and Load Shedding

1. The Choice You Make Either Way

When a system cannot serve all of its load, something has to give. The only question is whether you decided what, or whether the system decided for you.

The undecided version has a predictable shape. Load rises past capacity, queues grow, latency climbs, clients time out and retry, the retries add load, and the system serves nothing while working at full capacity. Everybody gets a slow failure instead of most people getting a fast success.

That outcome is worse than the alternative in a way worth stating precisely.