[BUG] Caching does not consider mode parameter for matrix-stats aggregation

[jainankitk]

Describe the bug

While working on the documentation improvements, @dwelsch-esi reported discrepancy opensearch-project/documentation-website#9434 (comment).

After spending sometime, I realized changing just the mode from avg to min or vice-versa causes same response to be returned, which is incorrect.

Related component

Search:Aggregations

To Reproduce

1. Index few documents with multi-value fields

POST _bulk
{ "create": { "_index": "students", "_id": "1" } }
{ "name": "John Doe", "gpa": 3.89, "class_grades": [3.0, 3.9, 4.0], "grad_year": 2022}
{ "create": { "_index": "students", "_id": "2" } }
{ "name": "Jonathan Powers", "gpa": 3.85, "class_grades": [4.1, 3.0, 4.0], "grad_year": 2025 }
{ "create": { "_index": "students", "_id": "3" } }
{ "name": "Jane Doe", "gpa": 3.52, "class_grades": [3.2, 2.1, 3.8], "grad_year": 2024 }

2. Run matrix_stats aggregation with mode as avg

GET students/_search
{
  "size": 0,
  "aggs": {
    "matrix_stats_taxful_total_price": {
      "matrix_stats": {
        "fields": ["gpa", "class_grades"],
        "mode": "avg"
      }
    }
  }
}

3. Change mode to min, same response is returned

GET students/_search
{
  "size": 0,
  "aggs": {
    "matrix_stats_taxful_total_price": {
      "matrix_stats": {
        "fields": ["gpa", "class_grades"],
        "mode": "min"
      }
    }
  }
}

Expected behavior

The cached response should not be returned

Additional Details

Plugins
Please list all plugins currently enabled.

Screenshots
If applicable, add screenshots to help explain your problem.

Host/Environment (please complete the following information):

• OS: [e.g. iOS]
• Version [e.g. 22]

Additional context
Add any other context about the problem here.

[jainankitk]

@sgup432 @peteralfonsi - Looks like an interesting issue with how cache keys are maintained. Hopefully, not a regression after introduction of Tiered Caching

[jinlee1703]

✅ Bug Reproduced

I've successfully reproduced the issue as described.

Using the following test data:

{ "gpa": 3.89, "class_grades": [3.0, 3.9, 4.0] }
{ "gpa": 3.85, "class_grades": [4.1, 3.0, 4.0] }
{ "gpa": 3.52, "class_grades": [3.2, 2.1, 3.8] }

I ran two matrix_stats aggregations:

• One with mode: "avg"
• One with mode: "min"

Both returned identical responses, with mean values instead of the expected min.
This confirms that the mode parameter is being ignored, or cached responses are being reused improperly across modes.

This matches the originally reported behavior:

"Changing just the mode from avg to min or vice-versa causes same response to be returned."

Let me know if you’d like help investigating the root cause or preparing a fix.

[jinlee1703]

I've opened PR #18254 to address this issue.
The fix ensures that the multiValueMode is properly considered in cache logic to prevent incorrect aggregator reuse when switching modes (e.g., from AVG to MIN).

Please take a look when you get a chance. Feedback is welcome!