I had an interesting experience just now. Setup: nginx proxies a portal server, which uses an underlying app server to render some contents. Problem: When displaying a portal page containing content from an app server, it rendered, and rendered, and rendered again, repeatedly. No error codes. The error log said "client closed prematurely connection ...", and the access log said "got page, got images, got page again, got images again, ...". We currently have only a single worker, and due to misconfiguration, the underlying app server failed miserably. The request was aborted prematurely, probably due to the app server failing. I guess nginx considered the worker broken and tried another, which was the same worker. And it kept repeating this. Question: Can I tell nginx to detect this cycle and instead display an error?
on 2008-12-03 14:49