feat: As a user, I want to have custom send_timeout config per route

[parsa97]

Description

As a platform provider, we need to have send_timeout customizable per host/path/location. Unlike proxy_send_timeout (which controls the timeout between APISIX and the upstream), send_timeout controls the timeout between APISIX and the downstream client.

Currently in APISIX this can only be configured globally, but since increasing it globally could impact all services and lead to connection exhaustion, we need per-route control via annotations or a plugin.

[Baoyuantop]

The send_timeout directive in nginx is a static directive that can only be configured at the http/server/location block level and cannot be dynamically modified at runtime in Lua. APISIX routing is runtime-driven, and its plugin ecosystem does not provide an API to modify nginx send_timeout. Therefore, without modifying the APISIX kernel (which requires a separate nginx location block for each route, or calling ngx.var.send_timeout via Lua—a variable not present in the nginx core variable set), this functionality is architecturally impossible to implement within the current APISIX framework.

[parsa97]

@Baoyuantop, Thank you for the detailed explanation.

Given that dynamically having this configuration is architecturally not possible, could you suggest any workaround to control send_timeout at a finer granularity than global? For example, is there a way to configure it per server block, even if it requires a restart or reload of the ingress instances to apply the change?

If no workaround exists, our fallback would be deploying a separate APISIX instance with a higher send_timeout only for the routes that require it, to mitigate the connection exhaustion risk from our main instance.

[Baoyuantop]

I agree with deploying a separate APISIX instance.