Revisit endpoint logic

[shikhar]

Goals

• We want to be able to target a cell URI that can handle all requests (also basin-level requests, with the S2-Basin header) – this captures local testing and our sandbox/staging environments, as well as potentially in future PrivateLink where a direct cell-level endpoint will be required also in prod.

• In general, we want to allow talking to a basin without knowing the cell ID and rely on DNS for that mapping. We also want to be able to test this in our sandbox & staging environments – so we need to avoid setting the S2-Basin header when including basin in the URI, to make sure we are exercising all the right bits.

Proposal

AccountService requests: look for S2_CELL_ENDPOINT

• if present, use that
• otherwise, use {cloud}.s2.dev

(Basin|Stream)Service requests: look for S2_BASIN_ZONE

• if present, use {basin}.{S2_BASIN_ZONE} without a S2-Basin header
• otherwise, look for S2_CELL_ENDPOINT
  • if present, use that with S2-Basin header
  • otherwise, use {basin}.b.{cloud}.s2.dev (effectively, S2_BASIN_ZONE="b.{cloud}.s2.dev")

These env vars should not contain the https:// scheme, that should be assumed as plaintext is not supported. They will correspond to the 'authority' portion of a URI.

[shikhar]

Instead of environment variables, I think better to have optional args, and let app do the plumbing (I think this is the current pattern).