fix(ebean-dao): do not cache empty schema loads (SharedSchemaCache poisoning, #618 regression)

[NatalliaUlashchick]

Summary

PR #618 (gma 0.6.185) moved the SharedSchemaCache pre-warm into the EbeanLocalAccess constructor so it always runs at startup. When the DAO is constructed before schema migration creates the entity table — integration tests, or deploys that run migrations as a separate job (LinkedIn's prod pattern) — loadColumns() returns an empty set, and caching it pinned a stale "table has no columns" view for the full 10-minute TTL. Every later columnExists() then returned false, so EbeanLocalAccess.batchGetUnion dropped the aspect from its query and EbeanLocalRelationshipQueryDAO mis-resolved tables.

Reported by the AIM ai-metadata-gms team: bumping datahub-gma to 0.6.184–0.6.189 broke 7 MlProject integ tests — 404s on aspect reads (Projects.getDeprecation), wrong autofill defaults, silently dropped writes, and wrong snapshot / member-update counts. Their bisect: 0.6.183 PASS → 0.6.186 FAIL (0.6.184/0.6.185 were unpublished to artifactory); testMode=false, NEW_SCHEMA_ONLY. Root-caused to #618.

Fix: never cache an empty column/index load — an empty result means the table is not present yet, so leave it uncached and let a later access re-resolve against the migrated schema. Applied in both:

• SharedSchemaCache (shared path used by EbeanLocalAccess reads), and
• SchemaValidatorUtil's local cache (used by EbeanLocalRelationshipQueryDAO).

Index-expression results may be legitimately empty and are left unchanged. The cold-start pre-warm perf intent of #618 is preserved.

Testing Done

• Added 2 SharedSchemaCacheTest regression tests (pre-warm and lazy-read before table creation). They fail on the pre-fix behavior (4 failures across both nonDollarVirtualColumnsEnabled factory variants) and pass with the fix.
• Full ./gradlew :dao-impl:ebean-dao:test suite passes (BUILD SUCCESSFUL, 40m) — no collateral.
• AIM to verify end-to-end on ai-metadata-gms once this lands as a published release.

Checklist

• The PR conforms to DataHub's Contributing Guideline (particularly Commit Message Format)
• Links to related issues (if applicable) — regression introduced by perf(ebean-dao): move SharedSchemaCache pre-warm to constructor #618
• Docs related to the changes have been added/updated (if applicable) — n/a

🤖 Generated with Claude Code

[codecov-commenter]

Codecov Report

✅ All modified and coverable lines are covered by tests.
✅ Project coverage is 66.88%. Comparing base (e528785) to head (5e12922).

Additional details and impacted files

@@             Coverage Diff              @@
##             master     #625      +/-   ##
============================================
+ Coverage     66.86%   66.88%   +0.01%     
- Complexity     1886     1888       +2     
============================================
  Files           148      148              
  Lines          7274     7277       +3     
  Branches        881      885       +4     
============================================
+ Hits           4864     4867       +3     
  Misses         2025     2025              
  Partials        385      385              

☔ View full report in Codecov by Harness.
📢 Have feedback on the report? Share it here.

🚀 New features to boost your workflow:

• ❄️ Test Analytics: Detect flaky tests, report on failures, and find test suite problems.