inconsistency of ngx.ctx between access_by_lua and header_filter_by_lua when limit_req is set

With this configuration

events {
    worker_connections 1024;
}


http {
    limit_req_zone $binary_remote_addr zone=req:100m rate=5r/s;
    limit_req zone=req burst=25;
	server{
		listen 80;
		location /{
			access_by_lua_block{
				ngx.log(ngx.ERR, 'access ', tostring(ngx.ctx))
			}
			proxy_pass http://www.baidu.com;
			header_filter_by_lua_block{
				ngx.log(ngx.ERR, 'content ', tostring(ngx.ctx))
			}
		}
	}
}

When test with ab or any other tools we can observe from log that some access_by_lua_block do not have a corressponding call to header_filter_by_lua_block with the same table address, and vice versa.

example for test: ab -c 100 -n 500 http://address/

The log is too long to paste here, which will be uploaded later.

It seems like responses are not associated to the request it's for. The problem goes away when set limit_req with nodelay setting.

below is env to reproduce the problem, which I don't think related:
OpenResty version: 1.17.8.2（binary release）
OS: Win 10

The problem also shows up for linux and my build with extra configuration. This enviroment is built to confirm it's not caused by build configuration.

Reproduce the issue with the provided nginx configuration and the ab -c 100 -n 500 command, comparing ngx.ctx addresses in the access and header-filter logs with and without nodelay. Start by tracing the limit_req request lifecycle and the Lua access/header-filter phases; done means determining whether the mismatch is reproducible and identifying the affected request association.