Summary

Queries against wide-schema indices (100+ fields) fail with script size exceeded even when the query references only a few fields. The Calcite engine appears to serialize all index fields into the execution script, causing the generated script to exceed OpenSearch’s 64 KB inline script limit (script.max_size_in_bytes).

Error (example):

java.lang.IllegalArgumentException: exceeded max allowed inline script size in bytes [65535] with size [130362] for script

Steps to Reproduce

1. Create a wide-schema index (example with 600+ fields):

PUT /wide_schema_index
{
  "mappings": {
    "properties": {
      "@timestamp": { "type": "date" },
      "message":    { "type": "text" }
      // ... 600+ additional fields (attributes.*, metrics.*, tags.*, resource.*)
    }
  }
}

2. Run a simple PPL query referencing 1–2 fields:

POST /_plugins/_ppl
{
  "query": "source=wide_schema_index | bin @timestamp span=1minute | stats count() by `@timestamp`"
}

3. Observe the failure:

{
  "error": {
    "reason": "Error occurred in OpenSearch engine: all shards failed",
    "details": "java.lang.IllegalArgumentException: exceeded max allowed inline script size in bytes [65535] with size [130362]"
  }
}

Expected Result

• Query succeeds, since it references only a small subset of fields.

Actual Result

• Query fails with script size exceeded because the generated script includes all fields from the index mapping.

Suspected Root Cause

In CalciteLogicalIndexScan.java:

// Current behavior (collects ALL fields)
private Map<String, Map<String, ExprType>> getAllFieldTypes() {
  return table.getRowType().getFieldList().stream()  // returns ALL fields
    .collect(Collectors.toMap(
      field -> field.getName(),
      field -> extractNestedTypes(field.getType())
    ));
}

• getFieldList() returns the entire schema; all fields are serialized into the execution script, making script size scale with total schema width rather than referenced fields.