Spark delete operation fails for partitioned Hudi tables - Hudi 0.14.0

### Bug Description

## **Environment**

* **Hudi version:** `0.14.0`
* **Spark version:** Spark 3.4.x

* **Storage:** Parquet
* **Table type:** COW
* **Read type:** Incremental query

***

## **Problem Description**

I am attempting to run delete operations on Hudi tables using Spark by reading data incrementally, filtering rows based on a condition, and writing them back using `hoodie.datasource.write.operation=delete`.

This logic works correctly for:

* ✅ Non‑partitioned Hudi tables
* ✅ Tested across multiple ETLs

However, the same logic consistently **fails for partitioned Hudi tables**.

***

## **Tables Tested**

### ✅ Works

* Non‑partitioned tables

### ❌ Fails

All failing tables are **partitioned** .

***

## **Reproducible Code Snippet**

```java
HoodieTableMetaClient metaClient = HoodieTableMetaClient.builder()
.setConf(new org.apache.hadoop.conf.Configuration())
.setBasePath(tablePath)
.build();

HoodieTableConfig config = metaClient.getTableConfig();

String recordKey = config.getRecordKeyFieldProp();
String precombineField = config.getPreCombineField();
Option partitionFields = config.getPartitionFields();
String partitionPathField = String.join(",", partitionFields.orElse(new String[0]));

// Incremental read
Dataset sourceDF = spark.read()
.format("org.apache.hudi")
.option(DataSourceReadOptions.QUERY_TYPE_OPT_KEY(),
DataSourceReadOptions.QUERY_TYPE_INCREMENTAL_OPT_VAL())
.option(DataSourceReadOptions.BEGIN_INSTANTTIME_OPT_KEY(), "0")
.load(tablePath);

sourceDF.createOrReplaceTempView("incremental_view");

// Select rows to delete
Dataset deleteDF = spark.sql(
"SELECT * FROM incremental_view WHERE ")
.withColumn("_hoodie_is_deleted", functions.lit(true));

// Write delete operation
DataFrameWriter writer = deleteDF.write()
.format("hudi")
.option("hoodie.table.name", config.getTableName())
.option("hoodie.datasource.write.recordkey.field", recordKey)
.option("hoodie.datasource.write.precombine.field", precombineField)
.option("hoodie.datasource.write.operation", "delete")
.mode(SaveMode.Append);

// Partition handling
if (!partitionPathField.isEmpty()) {
writer = writer
.option("hoodie.datasource.write.partitionpath.field", partitionPathField)
.option("hoodie.datasource.write.keygenerator.class",
"org.apache.hudi.keygen.TimestampBasedKeyGenerator")
.option("hoodie.keygen.timebased.timestamp.type", "SCALAR")
.option("hoodie.keygen.timebased.input.dateformat", "yyyy-MM-dd")
.option("hoodie.keygen.timebased.output.dateformat", "yyyy-MM-dd");
}

writer.save(tablePath);
```

***

## **Suspected Root Cause**

* Failure only occurs for **partitioned tables**
* All affected tables have **INTEGER partition columns**

***

### Environment

**Hudi version:** `0.14.0`
**Query engine:** (Spark)

### Logs and Stack Trace

## **Error Observed**

```text
Lost task 2.3 in stage 4.0 (TID 18):
java.lang.ClassCastException: java.lang.Integer cannot be cast to
org.apache.spark.unsafe.types.UTF8String
at org.apache.spark.sql.catalyst.expressions.BaseGenericInternalRow.getUTF8String(rows.scala:46)
at org.apache.spark.sql.catalyst.expressions.GenericInternalRow.getUTF8String(rows.scala:195)
at org.apache.spark.sql.execution.vectorized.ColumnVectorUtils.populate(ColumnVectorUtils.java:76)
at org.apache.spark.sql.execution.datasources.parquet.VectorizedParquetRecordReader.initBatch(VectorizedParquetRecordReader.java:294)
at org.apache.spark.sql.execution.datasources.parquet.VectorizedParquetRecordReader.initBatch(VectorizedParquetRecordReader.java:307)
at org.apache.spark.sql.execution.datasources.parquet
.Spark34LegacyHoodieParquetFileFormat
.$anonfun$buildReaderWithPartitionValues$2(Spark34LegacyHoodieParquetFileFormat.scala:332)
at org.apache.spark.sql.execution.datasources.FileScanRDD$$anon$1.readCurrentFile(FileScanRDD.scala:357)
at org.apache.spark.sql.execution.datasources.FileScanRDD$$anon$1.hasNext(FileScanRDD.scala:245)
at org.apache.hudi.HoodieSparkUtils$.$anonfun$createRdd$2(HoodieSparkUtils.scala:108)
```

Start by running the supplied Spark incremental-read and delete reproduction against a partitioned Hudi 0.14.0 table with INTEGER partition columns. Trace the failure through Spark34LegacyHoodieParquetFileFormat.scala, ColumnVectorUtils.java, and HoodieSparkUtils.java; done means the delete operation completes for partitioned tables without the reported ClassCastException.

Written by the indexing model from the issue text.