Guy on my team said my rate limit was backwards and I've been doing it wrong for 3 years
I've always set rate limits per API key, like 500 requests a minute each, and felt pretty good about it. Then a new dev on our team, Marcus, pulled me aside after a review and said the limits protect the wrong thing, because a client with ten keys just gets ten times the traffic and nobody notices until the gateway falls over. I pushed back at first, honestly pretty hard, because it meant rewriting config on about 40 endpoints. He was right though, so now I set limits per user and per IP too, not just per key, and we catch the aggressive scrapers way faster. He also pointed out that my 429s had no retry-after header, which meant clients were hammering us on backoff, and I fixed that across the board last month. It stung to admit it but our p95 latency is down around 30 percent on the public route. Curious if anyone here still swears by key-only limits, or if I'm the last one off that ship.