Can a Short-Link Service Really Reach 100,000 QPS in One Weekend by Just Adding Redis?
The reader’s brief
Before sharing a large QPS number, ask what passed: status, destination, latency and workload. Preserve the test window, hot/cold key distribution, dropped iterations and error counts. This article is useful as a measurement checklist, with an explicit unmeasured target. For our full review, open the dated developer briefing from Guides.
Bug Jun's October 16 Juejin article presents 100,000 QPS as a target, not a measured achievement. The author says k6 was unavailable in the validation environment. Its lab starts with one Go process, in-memory mappings and a local append log; PostgreSQL and Redis are a proposed later evolution. A correct redirect requires 302 and the matching Location header. Disable automatic redirect following for endpoint-only tests, and separate planned arrivals, requests sent and correct responses. Grafana's constant-arrival-rate executor starts iterations independently of response time while virtual users are available; its configured rate alone is not achieved correct-response throughput. We read both sources; we did not execute the code, reproduce throughput or audit the implementation.
Read original on Juejin via UAPI ↗ · Daily record · Monthly report