Customers can't run gRPC services on compute #490

Customers can't run a gRPC service on compute today. Our edge always talks HTTP/1.1 to the workload, and gRPC needs HTTP/2 all the way through. On top of that, the edge doesn't offer HTTP/2 to clients on most of our hostnames, so gRPC clients refuse to connect before a request is even sent.

(this is specifically about workload to workload gRPC via an ALB)

Why this matters

A customer asked us for this because they want to deploy a gRPC server on compute, and right now they have no workaround. gRPC is a common way to build APIs and services that talk to each other, so other customers will run into the same wall.

Desired outcome

• A customer can mark their workload's port as HTTP/2 when they deploy it, and the setting sticks across later deploys
• A standard gRPC client can call that workload through its URL in production

Related work

• feat: add network service port -> member HTTP/2 network-services-operator#653 lets the edge talk HTTP/2 to a workload's port
• chore: bump network-services-operator to v0.32.24 #483 picks up that change in compute
• feat: serve gRPC workloads with --http2 option #484 adds the HTTP/2 option to datumctl compute deploy
• feat: allow h2 in ALPN network-services-operator#660 lets clients negotiate HTTP/2 with the edge on every hostname