Hi,

I have a requirement to de-select or exclude one or more servers from an upstream group, if that server is either just newborn or exhausted (overloaded or showing reduced performance).

So, consider there are servers A, B and C in the upstream group. An external program periodically checks A, B and C and keeps track of different parameters (uptime, average response time etc), and assigns a "good" or "bad" status flag to each of them.

I would like to implement a mechanism in NGINX, in which it can periodically (say every 5 minutes) communicate with this external program and collect the status flag for A,B and C. This would be an HTTP communication with a response in a parse-able text format. Assume if B is reported as "bad", I would like to exclude B temporarily, till I get a "good" value back for it.

I am aware of the native upstream healthcheck mechanism in NGINX.

However, the requirement here is to bring some additional parameters into consideration for the selection of upstream.

I would like to know whether this is feasible to realize in NGINX. Could someone please explain some design insights for this problem?

Going a bit deeper, I could identify the following potential requirements or questions?

a) Should I depend on both (a) the passive upstream check native to NGINX and (b) the status value to be obtained from the external process? I think since the native health check is passive in nature (like when NGINX tries to connect to an upstream when it wants to proxy the request), it is beyond my programmable-control. So, I needn't worry about that. What I should be worried is about enabling NGINX to dynamically exclude a particular server, based on a "bad" feedback from the external process.

b) How to communicate periodically with the external process over HTTP? How can I get an HTTP client functionality in NGINX to send an external request and process the response? Can I build this functionality as an NGINX module?

c) Last but not the least, I would need some mechanism by which I should be able to map the server(s) reported as "bad" to the configured upstream servers, and then temporarily exclude them (mark down), and later include them when the status changes to "good".

Thanks
Rajesh