JoyJoyJo commented on issue #12561:
URL: https://github.com/apache/hudi/issues/12561#issuecomment-2570634112
I think I found the main cause. When Spark executed clustering using row
writer (`hoodie.datasource.write.row.writer.enable = true`), it read records
through `HadoopFSRelation` and passed a nullable schema. The code is shown as
below:
```Java
// Spark-sql_2.12-3.5.3
// DataSource(line: 414）
HadoopFsRelation(
fileCatalog,
partitionSchema = partitionSchema,
dataSchema = dataSchema.asNullable,
bucketSpec = bucketSpec,
format,
caseInsensitiveOptions)(sparkSession)
```
And Spark save these records into parquet file using the same schema.
```Java
// Hudi: master
// HoodieDatasetBulkInsertHelper(lines: 150)
def bulkInsert(dataset: Dataset[Row],
instantTime: String,
table: HoodieTable[_, _, _, _],
writeConfig: HoodieWriteConfig,
arePartitionRecordsSorted: Boolean,
shouldPreserveHoodieMetadata: Boolean):
HoodieData[WriteStatus] = {
val schema = dataset.schema
HoodieJavaRDD.of(
injectSQLConf(dataset.queryExecution.toRdd.mapPartitions(iter => {
val taskContextSupplier: TaskContextSupplier =
table.getTaskContextSupplier
val taskPartitionId = taskContextSupplier.getPartitionIdSupplier.get
val taskId = taskContextSupplier.getStageIdSupplier.get.toLong
val taskEpochId = taskContextSupplier.getAttemptIdSupplier.get
...
}
```
The row writer changed the schema when read records which lead to the schema
confilct later.
--
This is an automated message from the Apache Git Service.
To respond to the message, please log on to GitHub and use the
URL above to go to the specific comment.
To unsubscribe, e-mail: [email protected]
For queries about this service, please contact Infrastructure at:
[email protected]
