limit_req_zone: per‑location burst limits and interaction with limit_conn

Goal

To understand how nginx applies burst limits on a per‑location basis when using limit_req_zone and how these limits interact with limit_conn when both directives are enabled on the same location.

Constraints & Uncertainty

• The limit_req_zone directive defines a shared memory zone and a global rate; it does not expose a burst parameter inside the zone definition, forcing burst control via limit_req applied to each location.
• When limit_req and limit_conn are both active on a location, the documentation states the order of processing but does not clarify whether a request rejected by limit_req is counted against the limit_conn counter.
• It is unclear how the nodelay flag behaves when the burst is exceeded, especially if the configured rate is zero.

Unresolved Questions

1. Can a per‑location burst limit be enforced without creating separate limit_req_zone definitions, or is the only viable approach to duplicate zones with different burst values?
2. When both limit_req and limit_conn reject a request, does the limit_conn counter increment for the same request, potentially causing double‑counting?
3. How does the nodelay setting affect the handling of requests that exceed the burst when the rate is set to zero in the zone?