back project cache with local authorizer

[deads2k]

Fixes #5737

[deads2k]

I did not do a very controlled check, but creating a bunch of projects by hand after this change looks like

as opposed to the picture in the issue which I reproduced before.

[deads2k]

@liggitt This eliminates half of all server requests in @ironcladlou's runs

[ironcladlou]

LGTM, and confirmed as being enough to resolve #5737

[liggitt]

[test]

[liggitt]

adding to the milestone while we consider it

[smarterclayton]

Do we know what types don't have conversions?

[smarterclayton]

This is LGTM, and looks as simple as possible. @liggitt agree this is low risk high payoff?

[smarterclayton]

[test]

[liggitt]

LGTM. This backs the project cache with the same authorizer that is used to satisfy the localresourceaccessreview request submitted previously.

The only difference between the localresourceaccessreview call and this impl is that an empty namespace was explicitly disallowed when processing a localresourceaccessreview, but I verified that

1. all callers of Review() set a namespace
2. this implementation of Review() does the right thing when presented with an empty namespace (returns the list of users and groups that can get all namespaces)

[smarterclayton]

[merge], although this may result in test flakes due to improved performance, so I'll be watching for them today (this does not result in goroutine scheduling like previous flow did)

[openshift-bot]

continuous-integration/openshift-jenkins/merge SUCCESS (https://ci.openshift.redhat.com/jenkins/job/merge_pull_requests_origin/3965/) (Image: devenv-rhel7_2657)

[openshift-bot]

Evaluated for origin merge up to 16d6c23

[openshift-bot]

continuous-integration/openshift-jenkins/test Running (https://ci.openshift.redhat.com/jenkins/job/test_pull_requests_origin/6993/) (Extended Tests: core)

[smarterclayton]

Error from server: unable to find api field in struct BuildConfig for the json field "metadata"
!!! Error in test/cmd/builds.sh:32
    'oc patch bc/ruby-sample-build -p "{\"spec\":{\"output\":{\"to\":{\"name\":\"${REAL_OUTPUT_TO}\"}}}}"' exited with status 1
Call stack:
    1: test/cmd/builds.sh:32 main(...)
Exiting with status 1
!!! Error in hack/test-cmd.sh:281
    '${test}' exited with status 1
Call stack:
    1: hack/test-cmd.sh:281 main(...)
Exiting with status 1

I have trouble seeing how this is related - an ACL race wouldn't look like this.

[smarterclayton]

[test]

[smarterclayton]

Worried that we're seeing a lot of flakes here, but none of them look attributable so far. Testing more

[openshift-bot]

Evaluated for origin test up to 16d6c23