Branch percentile and sum-null IT expectations for the analytics-engine route

[ahkcs]

Description

CalcitePPLAggregationIT (and its bucketSize=2 subclass CalcitePPLAggregationPaginatingIT) hard-coded three expectations from the Calcite DSL-pushdown path. When the same suite is routed through the analytics-engine (DataFusion) backend with -Dtests.analytics.parquet_indices=true, those expectations are wrong even though the backend behaves correctly:

• testPercentile — percentile() is approximate. DataFusion's t-digest interpolation returns 46576 for percentile(balance, 90), where the OpenSearch/Calcite percentile_approx returns 48086 (p50 agrees on both at 32838). Both are valid approximations of the same percentile; the value was confirmed deterministic across repeated runs.
• testSumNull / testSumGroupByNullValue — SUM over an all-null bucket is null per the SQL spec. The DSL-pushdown path returns 0 instead — a known pushdown quirk tracked in [BUG] Sum multiple null values should return null instead of 0 #3408. DataFusion follows the spec like the Calcite-no-pushdown path and returns null.

The fix branches the expected values on the existing isAnalyticsParquetIndicesEnabled() helper, exactly the pattern already used in StatsCommandIT.testSumWithNull. No production code changes — these are test-expectation corrections only, and the assertions remain unchanged for the default Calcite path.

Pass rate (these test methods)

Test method	analytics route before	analytics route after	Calcite/v2 route
testPercentile	❌	✅	✅ (unchanged)
testSumNull	❌	✅	✅ (unchanged)
testSumGroupByNullValue	❌	✅	✅ (unchanged)
same 3 via CalcitePPLAggregationPaginatingIT	❌	✅	✅ (unchanged)

Verified by running :integ-test:integTestRemote against a 9-plugin analytics cluster both with and without -Dtests.analytics.parquet_indices=true.

Check List

• New functionality includes testing.
• New functionality has been documented.
• Commits are signed per the DCO using --signoff.

By submitting this pull request, I confirm that my contribution is made under the terms of the Apache 2.0 license.

[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

⚡ No major issues detected

[github-actions]

PR Code Suggestions ✨

Explore these optional code suggestions:

Category	Suggestion	Impact
General	Extract magic numbers to constants Consider using a constant or configuration value for the expected percentile values instead of hardcoding them. This makes the test more maintainable and easier to update if the approximation algorithm changes in the future. integ-test/src/test/java/org/opensearch/sql/calcite/remote/CalcitePPLAggregationIT.java [986] -int expectedP90 = isAnalyticsParquetIndicesEnabled() ? 46576 : 48086; +private static final int ANALYTICS_ENGINE_P90 = 46576; +private static final int CALCITE_P90 = 48086; +int expectedP90 = isAnalyticsParquetIndicesEnabled() ? ANALYTICS_ENGINE_P90 : CALCITE_P90; Suggestion importance[1-10]: 5 __ Why: The suggestion correctly identifies the line and proposes extracting magic numbers to constants, which improves code maintainability. However, this is a minor code style improvement with limited impact on functionality, and the current inline approach with a clear comment is already reasonably maintainable for test code.	Low