[BatchUpdate 2/3] feat: add batch SQL layer for multi-aspect upsert and batch deletion

[jphui]

Overview

Extracted from #598 for easier review (part 2 of 4, depends on #606). This PR contains only the SQL-level batch operations — no orchestration logic in BaseLocalDAO or EbeanLocalDAO.

Note that upon inspection, the comparison between this PR and the relevant sections of 598 are identical 😎

Changes

1. EbeanLocalAccess.batchUpsert() — multi-aspect upsert in a single SQL call

• Generates one INSERT ... ON DUPLICATE KEY UPDATE that writes all aspects for a URN at once
• Uses UPSERT semantics: always updates if exists, clears deleted_ts
• Contrast with create() which throws DuplicateKeyException on non-soft-deleted duplicates

2. prepareMultiColumnInsert() — shared SQL builder

• Extracted common logic from create() into a reusable helper
• Handles: INSERT INTO + VALUES clause building, audit stamp extraction, aspect parameter binding, URN extraction
• Caller provides only the ON DUPLICATE KEY clause, eliminating ~80 lines of duplication
• Note: create() now calls this helper instead of inline SQL construction. Output SQL is identical — pure extract-method refactor.

3. buildOnDuplicateKeyForCreate() / buildOnDuplicateKeyForUpsert() — split clause generation

• create: sets aspects + conditionally throws via CAST('DuplicateKeyException' AS UNSIGNED)
• upsert: sets aspects + lastmodifiedon + deleted_ts = NULL
• DELETED_TS_DUPLICATE_KEY_CHECK renamed to ON_DUPLICATE_KEY_UPDATE for clarity

4. IEbeanLocalAccess + InstrumentedEbeanLocalAccess — interface and instrumentation

• Added batchUpsert() to the interface
• InstrumentedEbeanLocalAccess delegates with latency/error tracking

5. EbeanLocalAccessTest — integration tests against embedded MariaDB

• testBatchUpsertMultipleAspects: inserts two aspects in one call, verifies both persisted
• testBatchUpsertOverwritesExistingValues: upserts over existing data, verifies update
• testBatchUpsertClearsDeletedTs: soft-delete then upsert, verifies deleted_ts cleared
• testBatchUpsertSingleAspect: single-aspect degenerate case
• testBatchUpsertPreservesIngestionTrackingContext: verifies emitter/emitTime round-trip

Testing

• All EbeanLocalAccessTest and InstrumentedEbeanLocalAccessTest tests pass
• Existing tests unaffected — create() produces identical SQL output via the shared helper
• *** The FULL E2E Testing suite from the original mega PR still applies here too

Checklist

• Compiles with ./gradlew :dao-api:compileJava :dao-impl:ebean-dao:compileJava
• Tests pass with ./gradlew :dao-impl:ebean-dao:test --tests EbeanLocalAccessTest --tests InstrumentedEbeanLocalAccessTest
• No breaking changes to public interfaces
• Files are byte-identical to corresponding code in (feat) Transactional Multi-Aspect Update #598

[codecov-commenter]

Codecov Report

❌ Patch coverage is 82.02247% with 16 lines in your changes missing coverage. Please review.
✅ Project coverage is 65.56%. Comparing base (fb9fa85) to head (f26ca81).

Files with missing lines	Patch %	Lines
...va/com/linkedin/metadata/dao/EbeanLocalAccess.java	83.90%	6 Missing and 8 partials ⚠️
...din/metadata/dao/InstrumentedEbeanLocalAccess.java	0.00%	2 Missing ⚠️

Additional details and impacted files

@@             Coverage Diff              @@
##             master     #607      +/-   ##
============================================
+ Coverage     65.43%   65.56%   +0.12%     
- Complexity     1749     1759      +10     
============================================
  Files           144      144              
  Lines          6813     6847      +34     
  Branches        826      829       +3     
============================================
+ Hits           4458     4489      +31     
- Misses         1993     1996       +3     
  Partials        362      362              

☔ View full report in Codecov by Sentry.
📢 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.

[chmeng0]

nit: it's not used in this pr, but i see it's used in the large pr containing all the changes (#598) so it should be good

[chmeng0]

@jphui could you help confirm the behavior of batchUpsert() on duplicate key (row exists)?
in that case, this buildOnDuplicateKeyForUpsert() appends ON DUPLICATE KEY UPDATE <aspect columns>, lastmodifiedon = :lastmodifiedon, deleted_ts = NULL; and lastmodifiedby is not updated? should we update lastmodifiedby as well?

[jphui]

ON DUPLICATE KEY UPDATE %s = :metadata, lastmodifiedon = :lastmodifiedon, deleted_ts = NULL;

Same pattern — lastmodifiedby is not updated on duplicate key in the existing single-aspect path either. So the batch upsert is consistent with the existing behavior. It's not a
new bug we introduced — it's pre-existing.

The lastmodifiedby is baked into each aspect's JSON blob (via AuditedAspect.setLastmodifiedby), so the actual actor info is preserved inside the aspect column. The top-level
lastmodifiedby column just doesn't get refreshed on update, which has been the case all along.

This is basically just keeping behavior consistent with existing behavior.

[chmeng0]

LGTM