Extension/possible to respond with a status code directly from HTTPRoute, such as 503 Service Unavailable during maintenance?

[meastp]

Is there a way to point to no service and get 503 Service Unavailable[1] instead?

That would be perfect during maintenance/(db schema) upgrade/scheduled downtime when the service is not available.

[1] https://developer.mozilla.org/en-US/docs/Web/HTTP/Status/503

[pleshakov]

Hi @meastp

I see a few options:

• Deploy a pod that will return 503s and redirect traffic in the HTTPRoute to the corresponding Service.
• Configure an redirect filter that redirects traffic to the URI that can return 503. https://gateway-api.sigs.k8s.io/reference/spec/#gateway.networking.k8s.io/v1.HTTPRequestRedirectFilter

There are a few related things:

(1)
If you reference a non-existing service as a backend ref in an HTTPRoute, NGF (NGINX Gateway Fabric) will return 500. This is prescribed by the Gateway API here https://gateway-api.sigs.k8s.io/reference/spec/#gateway.networking.k8s.io%2fv1.HTTPRouteRule , field backendRefs

(2)
If the referenced Service has 0 Endpoints, NGF will return 502. The Gateway API doesn't prescribe what to do in this case, so in NGF, we use the default NGINX behavior in this case.

(3) There is no filter in the Gateway API that can return some static response like 503. https://gateway-api.sigs.k8s.io/reference/spec/#gateway.networking.k8s.io/v1.HTTPRouteFilter

[meastp]

Thank you for providing the options. Reading the HTTPRequestRedirectFilter documentation in the link you provided, it does look like the default is a Location-header that points to the request endpoint. If I set statusCode to 503, the response should be "like" the 503 I wanted (except, perhaps, the ability to provide a body text) with an additional Location header that can be ignored (or, if followed, points to the same endpoint).... is this correct? :)

[pleshakov]

the supported status codes in HTTPRequestRedirectFilter are 301 and 302, so it won't work for 503 https://github.com/kubernetes-sigs/gateway-api/blob/0f040de255887d80a7df2a9fdc7278a08b38de9c/apis/v1/httproute_types.go#L1058

[meastp]

I see, that's a shame - seems (to me) like simple functionality :)

I'll have to go with deploying a pod, then. Thanks for your thorough answers and explanation :)

[brianehlert]

This reads like a desire to support custom response codes, and possibly custom responses.
Would either / both of those capabilities satisfy your needs @meastp ?

[meastp]

@brianehlert Yes, just having the ability to provide a custom response code (503 Service Uavailable) would save me from having to create a whole new container/service that just responds with a custom response code when the url/route target is down for maintenance/upgrade/deploy.

Having the ability to also provide a custom response body (e.g. "The dataset is down for maintenance, retry your request later") and/or custom header(s) (e.g. Retry-After 30) would be great as well, but not critical :)