What we learned from load testing our own API with k6 — things nobody warns you about

We ran our first real k6 load test last quarter targeting 500 concurrent users. A few things that surprised us: **The bottleneck wasn't where we thought** We expected the DB. Actual bottleneck: the image resizing service we forgot existed. One endpoint that triggered it had p99 of 8 seconds under load. **The test environment behaved completely differently** Our staging DB has 1/10th the data of production. Query plans were different, indexes weren't hit the same way. First load test that matters is on prod-like data. **Connection pool exhaustion looks like random 502s** Spent two days thinking our load balancer was broken. It was the PostgreSQL connection pool on the API server hitting max connections. **k6 reporting needs work out of the box** Default console output is hard to interpret mid-test. Set up InfluxDB + Grafana or use k6 Cloud even for a trial to get useful real-time dashboards.