/cc @ncdc

Labelling this PR as size/XS

To elaborate:

• the admission controller normally works against an index.
• the flake happens when that index is not yet populated
• @deads2k added a live lookup cache to do a live lookup when the index was not populated, and hold onto that for 30s as an effort to reduce flake
• but limit ranger admission code intercepts all create operations for any resource
• so when the limit range itself is first created, admission controller intercepts that, and it did a live lookup, and said 'oh there are actually no limit ranges, let me remember that in my expiry cache`
• then we go to create a pod, and the index in admission is still not updated, so it falls back to its live lookup cache and sees no limit ranges for the past action and so no resources are applied

The fix is to not populate a live lookup cache until we see our first pod so live lookup cache is not prematurely updated when the limit range itself is created.

GCE e2e test build/test passed for commit 4858b48.

• Build Log
• Test Artifacts
• Internal Jenkins Results

@derekwaynecarr good catch. Is this already in the origin pick list?

@k8s-bot test this [submit-queue is verifying that this PR is safe to merge]

GCE e2e test build/test passed for commit 4858b48.

• Build Log
• Test Artifacts
• Internal Jenkins Results

@k8s-bot test this [submit-queue is verifying that this PR is safe to merge]

@deads2k - I added it to the issue list.

GCE e2e build/test failed for commit 4858b48.

• Build Log
• Test Artifacts
• Internal Jenkins Results

@k8s-bot e2e test this github flake: #21798

GCE e2e test build/test passed for commit 4858b48.

• Build Log
• Test Artifacts
• Internal Jenkins Results