Differentiate status codes from functions vs OpenFaaS components about faas HOT 2 OPEN

I think that options 1 and 2 should both be included, and that option 2 should be opt-in for the sake of backwards compatibility.

Here's a truth table for describing the cases using only option 1:

As you can see, in this approach you don't know if the response is coming from infrastructure (non-openfaas, non-function. k8s, istio, etc) or from the function.

Here's another table using only option 2:

This option only lets you know if the response came from a function. All cases other than the function are effectively unknown as to what the source was.

Here's a final table using both:

This approach allows you know the most; it allows you to know if the response came from inside openfaas, the user's function, or some other unknown/infrastructural source.

As for whether or not option 1 should be "only when an OF component has an error", I would like to clarify that I don't think that's specifically necessary to call out, as the source of a response should effectively be invisible to the user unless there's an error. In other words, there's no specific reason why the x-openfaas-component cannot be set when the response is not an error, but in normal operating procedure, the other components are not responding, they are proxying, and therefore the response itself isn't coming from them.

If it matters, you could have a header stating every OF component that "touched" this request along the way, but that's more of a debug feature in my opinion.

from faas.

Summarised from Kevin on weekly call:

Responses coming from OpenFaaS components should be retried according to a default retry policy - i.e. defined in Helm chart.

Responses not coming from OpenFaaS components would be retried according to the configured policy on the component i.e. queue-worker / connector.

Responses coming from function would use the retry configuration set through annotations on the function. If no annotations are set the function would not be retried. This functionality should be opt-in and can be turned on via a parameter in the chart. This would need the watchdog update.

from faas.