API Gateway XDS Primitives, endpoints and clusters

[sarahalsmiller]

Description

Work to move APIGateway away from using prexisting ingress gateway functionality.

Testing & Reproduction steps

Existing gateway behavior should be unchanged with this change, so existing CI tests continuing to pass should be sufficient.

Links

Relevant RFC: https://docs.google.com/document/d/15kzB5wqhzzPOMWCD8SphpcYUAigUmJvV6R6y2MZUnJU/edit?usp=sharing

PR Checklist

• updated test coverage
• external facing docs updated
• appropriate backport labels added
• not a security concern

[andrewstucki]

Hey Sarah, so I'll need to review some of the RFC to see the full approach suggested, but part of our desire is to get fully rid of the translation that happens from the APIGateway portion of a proxycfg snapshot into the IngressGateway portion, which means that ideally none of our code would use any of the IngressGateway fields. That would allow us to drop all of the transformation logic found here:

consul/agent/proxycfg/snapshot.go

Line 749 in 0dd4ea2

func (c *configSnapshotAPIGateway) ToIngress(datacenter string) (configSnapshotIngressGateway, error) {

and in the corresponding discovery chain "synthesizing" code that exists in the gateway*.go files in this subpackage.

But it does mean that we'll need to come up with the direct HTTPRoute --> xDS route synthesis code (which may or may not also leak to listeners, and seemingly here into clusters as well).

Maybe we can set up some time next week to chat about this synchronously to get on the same page?

[sarahalsmiller]

Hi Andrew, I would be happy to hop on a call with you sometime this week, but I just accidentally missed that IngressGateway field in my initial pass. Part of the trouble I'm having with this is how much of the preexisting Xds generation/helper codes seem to rely heavily on the preexisting IngressGateway struct, so I do think it'd be helpful to talk through how our architecture is set up and what we want this to look like in the end.