A working ELB would be nice.
The existing CIA inspired torture technique that you guys call an ELB is a really poor joke.
Let's maybe design an ELB that :
1) Isn't frontended by the worst UX to ever be unleashed upon the earth.
2) doesn't require ingress setup for 120k+ externally routable addresses.
3) can port map
4) can protocol map
5) doesn't randomly 502 sporadically for seemingly no reason
6) isn't designed to harm the mental health of users.
Thanks for your comments, they’ve been passed to the networking team. If you have more specific feedback on the UI or on how to preserve mental health, please let us know in the comments so we can address your exact feedback.
I would tend to agree that creation of a load balancer does seem unnecessarily complex in GCP.
The whole setup is (I think) something like the following:
Global Forwarding Rule -> Target HTTPS Proxy (SSL offload) -> URL Map -> Backend Service -> Instance Group
The problems with this are I think:
* There is no load balancer construct, and instead we have a series of obscure entities that must be cobbled together to create one
* Things that I would think of as attributes of a load balancer are entities in their own right, and I don't think it's clear why that needs to be the case, as it seems quite counter-intuitive, which makes it frustrating and difficult to comprehend - not just for people coming from AWS backgrounds
Perhaps a simpler construct called a "load balancer" as a front-end, that links to an instance group as a back-end. And all the other constructs such as healthchecks, url maps, target http proxies and global forwarding rules become mere properties of the load balancer rather than standalone entities.