test(integ-test): stabilize eval max/min schema assertions across shards

[mengweieric]

Description

CalcitePPLEvalMaxMinFunctionIT had three tests that passed on a single shard and failed on five.

eval max()/min() carries no plan-time result type, so the reported type of the computed column is taken from the first row of the result set. OpenSearchExecutionEngine does this deliberately as a documented workaround for ANY-typed columns. With one shard the first row is always the same document and the reported type is stable. With more shards the arriving row varies, and because the winning operand differs per row, so does the type.

source=opensearch-sql_test_index_dog | eval new = max(1, 3, age) | fields age, new

age=2  ->  new=3  from the int literal
age=4  ->  new=4  from the bigint field

Both rows are correct. Only the column label moves, between int and bigint, depending on which row arrives first. The failures were always in verifySchema, never in verifyDataRows.

The three affected tests now filter to a single known document, which makes the sampled row deterministic under any shard count. Each one gained a companion test for the other selection direction, so the pair still asserts everything the original two-row query asserted.

Test	Query scope	Selected value	Reported type
testEvalMaxNumeric	where age = 2	literal 3	int
testEvalMaxNumericWhenFieldSelected (new)	where age = 4	field 4	bigint
testEvalMinNumeric	where age = 2	field 2	bigint
testEvalMinNumericWhenLiteralSelected (new)	where age = 4	literal 3	int
testEvalMinIgnoresNulls	where key = 'values'	field 3.1415	double
testEvalMinIgnoresNullsWhenLiteralSelected (new)	where key != 'values'	literal 5	int

The three tests that assert int are gated on EVAL_MAX_MIN_INT_WIDENING, since the analytics-engine route reports a wider type for the same value.

Scope notes:

• One test file. No shared fixture, mapping, enum or constant is touched, and no new index is added.
• The original method names are kept and no test is removed.
• testEvalMaxIgnoresNulls and the four string and mixed-type tests are unchanged. max(int, 3) is an int on every row, so that test was already shard-stable.
• The null case still covers both explicit-null and missing-field documents, matching the original row coverage.
• No production code changes. This makes the assertions shard-independent and does not alter what max() or min() return.

Validation

Run on an external cluster, comparing the same class before and after the change.

Cell	Before	After
1 shard	8 pass	11 pass
5 shards	5 pass, 3 fail	11 pass, 0 skipped

The five-shard run before the change reproduced exactly the three reported failures, and the cluster log confirms the fixtures were created with five primary shards.

spotlessJavaCheck, compileTestJava and git diff --check pass.

This class is not a member of the CalciteNoPushdownIT suite, so that mode is unaffected. The analytics-engine lane was reasoned from the capability gate rather than executed locally, because that plugin is not available in this tree.

Related Issues

The underlying limitation is that the result type is recovered from data rather than declared at plan time. That is a separate product discussion and is not addressed here.

Check List

• New functionality includes testing.
• Commits are signed per the DCO using --signoff or -s.

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