k8s/controller: always map "default" Consul namespace to empty string

[mikemorris]

Changes proposed in this PR:

Always map the "default" Consul namespace to the empty string in the memory backend store, for compatibility with Consul OSS and the logic to parse the namespace of a Gateway from the SPIFFE URI, used when reading a Gateway from a store backend.

This Consul namespace mapper function is already called when reconciling a Gateway to a backend store, but the corresponding lookup fails with Consul Enterprise when --consul-destination-namespace is set to "default" because of the special case handling in the parseURI helper.

No migration handling should be needed because this is all within the ephemeral memory store for now, but this does need an e2e test for a service in the "default" namespace on Consul Enterprise.

As a temporary workaround, this edge case can be avoided by configuring consulNamespaces.mirroringK8S: true or specfiying "" or any value other than "default" for consulNamespaces.consulDestinationNamespace in the Helm chart.

How I've tested this PR:

How I expect reviewers to test this PR:

Checklist:

• Tests added
• CHANGELOG entry added

  Run make changelog-entry for guidance in authoring a changelog entry, and
  commit the resulting file, which should have a name matching your PR number.
  Entries should use imperative present tense (e.g. Add support for...)

[mikemorris]

I was surprised this would be necessary, as I expected SetUpStack to create a new context for each test, but it appears to be reused - I'm not sure if that's intentional/common, or if that should be changed and this cleanup could be removed?

Without this cleanup, the subsequent TestGatewayBasic e2e fails when running against Consul Enterprise, as it checks for the presence of a Consul service registered in an expected namespace.

If keeping this cleanup logic makes more sense, I'm wondering if e2e.SetConsulNamespace(nil) should also be called here to randomize the Consul namespace again.

[sarahalsmiller]

I've run into this issue before when touching the e2e tests as well. I'm not sure what the right solution is because I definitely think spinning up a new kind cluster for each test would make these tests take way longer, but the current set up make it difficult to do any kind of namespace manipulation.

[sarahalsmiller]

Thanks for hopping on this

[andrewstucki]

The patch itself looks fine to me