Allow CastTypeAliasRewriter to recurse into nested CAST operands to fix upgrade incompatibility issue

[rsrkpatwari1234]

Fixes #18152

After normalizing the cast target (BIGINT→LONG), recurse into operand 0 so nested CASTs and CASTs inside divide/etc. are fully rewritten.

bug backward-incompat dependencies multi-stage

[codecov-commenter]

Codecov Report

✅ All modified and coverable lines are covered by tests.
✅ Project coverage is 63.14%. Comparing base (7e10a36) to head (4d2c294).

Additional details and impacted files

@@             Coverage Diff              @@
##             master   #18153      +/-   ##
============================================
- Coverage     63.18%   63.14%   -0.04%     
  Complexity     1616     1616              
============================================
  Files          3214     3214              
  Lines        195838   195840       +2     
  Branches      30251    30251              
============================================
- Hits         123734   123666      -68     
- Misses        62236    62296      +60     
- Partials       9868     9878      +10     

Flag	Coverage Δ
custom-integration1	100.00% <ø> (ø)
integration	100.00% <ø> (ø)
integration1	100.00% <ø> (ø)
integration2	0.00% <ø> (ø)
java-11	63.10% <100.00%> (-0.03%)	⬇️
java-21	63.12% <100.00%> (-0.05%)	⬇️
temurin	63.14% <100.00%> (-0.04%)	⬇️
unittests	63.14% <100.00%> (-0.04%)	⬇️
unittests1	55.37% <100.00%> (-0.02%)	⬇️
unittests2	34.77% <0.00%> (-0.02%)	⬇️

Flags with carried forward coverage won't be shown. Click here to find out more.

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

🚀 New features to boost your workflow:

• 📦 JS Bundle Analysis: Save yourself from yourself by tracking and limiting bundle sizes in JS merges.

[Copilot]

Pull request overview

This PR fixes a backward-compatibility bug in the SQL CAST type-alias rewriter by ensuring it rewrites nested CAST expressions inside the first operand of an outer CAST (e.g., CAST(CAST(x AS LONG) AS DOUBLE), or CAST((CAST(x AS LONG)/1000) AS DOUBLE)), which previously could break mixed-version clusters.

Changes:

• Update CastTypeAliasRewriter to recurse into CAST operand 0 after normalizing the cast target type (e.g., BIGINT → LONG).
• Add comprehensive unit tests covering nested CASTs in SELECT, WHERE, GROUP BY, and under arithmetic.
• Extend the compatibility verifier sample suite with nested-CAST queries and expected results.

Reviewed changes

Copilot reviewed 4 out of 4 changed files in this pull request and generated no comments.

File	Description
pinot-common/src/main/java/org/apache/pinot/sql/parsers/rewriter/CastTypeAliasRewriter.java	Recurses into the CAST expression operand so nested CAST nodes get rewritten.
pinot-common/src/test/java/org/apache/pinot/sql/parsers/rewriter/CastTypeAliasRewriterTest.java	Adds coverage for nested CAST rewrite behavior across multiple query locations/shapes.
compatibility-verifier/sample-test-suite/config/queries/feature-test-1-sql.queries	Adds sample nested-CAST queries that previously could fail during rolling upgrades.
compatibility-verifier/sample-test-suite/config/query-results/feature-test-1-rest-sql.results	Adds expected outputs for the new nested-CAST compatibility queries.

[rsrkpatwari1234]

@rohityadav1993 @Jackie-Jiang @yashmayya Requesting your review on this

[shauryachats]

LGTM

[yashmayya]

@rsrkpatwari1234 the release candidate for 1.5.0 has already been built -- this patch is only needed for upgrade compatibility right? Is it even useful once servers are able to process BIGINT directly?

During rolling upgrades (Controller+Broker -> 1.4 and Server -> 1.3), queries that nest CAST(... AS LONG) (→ BIGINT in the plan) under another CAST or under an expression that is the first operand of an outer CAST can fail on older servers because CastTypeAliasRewriter never visited those inner CASTs before dispatch.

I'm not sure I follow this - CastTypeAliasRewriter doesn't even exist in 1.4 (https://github.com/apache/pinot/commits/release-1.4.0/).

[rsrkpatwari1234]

@rsrkpatwari1234 the release candidate for 1.5.0 has already been built -- this patch is only needed for upgrade compatibility right? Is it even useful once servers are able to process BIGINT directly?

During rolling upgrades (Controller+Broker -> 1.4 and Server -> 1.3), queries that nest CAST(... AS LONG) (→ BIGINT in the plan) under another CAST or under an expression that is the first operand of an outer CAST can fail on older servers because CastTypeAliasRewriter never visited those inner CASTs before dispatch.

I'm not sure I follow this - CastTypeAliasRewriter doesn't even exist in 1.4 (https://github.com/apache/pinot/commits/release-1.4.0/).

@yashmayya

• We are upgrading pinot from 1.3 -> 1.4 and currently at stage where controllers + brokers are at 1.4 and servers at 1.3
• We cherry-picked Fix upgrade compatibility issue with type casts in SSE #16682 on top of pinot release-1.4.0 to fix issue - (IllegalArgumentException: Unable to cast expression to type - BIGINT in CastTransformFunction.init)
• Even with Fix upgrade compatibility issue with type casts in SSE #16682, we saw multiple failures for our critical queries since nested CAST is not handled as part of above.
• Hence, we need a fix for this to continue upgrade rollouts and ensure backward compatibility

[rohityadav1993]

+1, this is patch for the patch of the original bug. Anybody upgrading from 1.3 -> 1.4 will need to cherrypick #16682 and current PR.

[rsrkpatwari1234]

@yashmayya Requesting your review on this