Fix flaky DerivedSourceLeafReaderTests

[andrross]

testWithRandomDocuments was flaky due to two issues:

1. NoMergePolicy blocked forceMerge(1), so randomized IndexWriterConfig settings that caused multiple flushes resulted in multiple segments instead of the expected one.

2. The test stored _source in the index, but DerivedSourceStoredFields injects _source via the source provider then delegates to the underlying stored fields. This caused the visitor to receive _source twice, with the second call using Lucene's internal buffer which could differ from the original bytes.

Fix by removing NoMergePolicy so forceMerge(1) works, and not storing _source in the index, which matches the production use case where derived source synthesizes _source from other data.

Related Issues

Resolves #20812

Check List

• Functionality includes testing.

By submitting this pull request, I confirm that my contribution is made under the terms of the Apache 2.0 license.
For more information on following Developer Certificate of Origin and signing off your commits, please check here.

[github-actions]

PR Reviewer Guide 🔍

Here are some key observations to aid the review process:

🧪 PR contains tests

🔒 No security concerns identified

✅ No TODO sections

🔀 No multiple PR themes

⚡ Recommended focus areas for review Unused Variable The source byte array is still generated via randomByteArrayOfLength and stored in docIdToSource, but it is no longer added to the document as a stored field. The map is presumably used later to verify that the derived source provider returns the correct bytes. It should be verified that the test still correctly validates the source content, since the document no longer stores _source and the map values are random bytes unrelated to any stored field. byte[] source = randomByteArrayOfLength(randomIntBetween(10, 50)); docIdToSource.put(i, source); Document doc = new Document(); doc.add(new StoredField("field", i)); randomWriter.addDocument(doc);

[github-actions]

PR Code Suggestions ✨

Explore these optional code suggestions:

Category	Suggestion	Impact
Possible issue	Restore correct stored field for source verification The original test stored _source bytes and verified them via docIdToSource, but now the field name is changed to "field" with an integer value. This likely breaks the test's ability to verify source data retrieval through DerivedSourceLeafReader, since the map docIdToSource still maps doc IDs to byte arrays but the stored field no longer contains those bytes. The field name and type should remain consistent with what DerivedSourceLeafReader expects to read. server/src/test/java/org/opensearch/common/lucene/index/DerivedSourceLeafReaderTests.java [159-160] -doc.add(new StoredField("field", i)); +doc.add(new StoredField("_source", source)); randomWriter.addDocument(doc); Suggestion importance[1-10]: 5 __ Why: The suggestion raises a valid concern about changing the stored field from "_source" with byte array to "field" with integer value, which could affect the test's ability to verify source data retrieval. However, this appears to be an intentional change in the PR (possibly to test a different aspect of DerivedSourceLeafReader), and without seeing the full test context (especially how docIdToSource is used later), it's unclear if this is actually a bug.	Low

[github-actions]

✅ Gradle check result for 4766166: SUCCESS

[codecov]

Codecov Report

✅ All modified and coverable lines are covered by tests.
✅ Project coverage is 73.39%. Comparing base (28fa177) to head (4766166).
⚠️ Report is 12 commits behind head on main.

Additional details and impacted files

@@             Coverage Diff              @@
##               main   #20904      +/-   ##
============================================
+ Coverage     73.32%   73.39%   +0.06%     
- Complexity    72272    72387     +115     
============================================
  Files          5797     5802       +5     
  Lines        330323   330405      +82     
  Branches      47676    47686      +10     
============================================
+ Hits         242215   242497     +282     
+ Misses        68663    68514     -149     
+ Partials      19445    19394      -51     

☔ 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.