I was going through NGINX source code to implement keepalive for nginx zeromq plugin that I have developed. I have been inspired by ngx_http_upstream_keepalive_module. Was wondering why nginx uses a kind of two linkedlist based stack implementation to implement keepalive connection pool. Why not use typical linkedlist based queue implementation. Kindly refer to ngx_http_upstream_keepalive_module.c :ngx_http_upstream_get_keepalive_peer ngx_http_upstream_keepalive_module.c:ngx_http_upstream_free_keepalive_peer I used exactly like this without wondering why, now I kind of want to know the reason. Posted at Nginx Forum: http://forum.nginx.org/read.php?2,251033,251033#msg-251033
on 2014-06-19 07:25
on 2014-06-19 15:38
Hello! On Thu, Jun 19, 2014 at 01:24:53AM -0400, prkumar wrote: > I used exactly like this without wondering why, now I kind of want to know > the reason. The upstream keepalive module uses the "cache" queue as an LRU queue, to be able to drop unused connections if there isn't enough room. Hence it basically uses it as a stack while storing / retrieving connections. The "free" queue is used to avoid runtime allocations. -- Maxim Dounin http://nginx.org/
on 2014-06-19 23:33
Hello, It seems ZMQ and Nginx keep coming up in conversations lately. There seems to be differing opinions on it's performance/feasibility in production. Do you plan to open source your code? I'd be curious to look at it.