Github user manishgupta88 commented on a diff in the pull request:
https://github.com/apache/carbondata/pull/2990#discussion_r242871412
--- Diff:
integration/spark2/src/main/scala/org/apache/spark/util/AlterTableUtil.scala ---
@@ -249,32 +249,99 @@ object AlterTableUtil {
* @param timeStamp
* @param sparkSession
*/
- def revertDataTypeChanges(dbName: String, tableName: String, timeStamp:
Long)
+ def revertColumnRenameAndDataTypeChanges(dbName: String, tableName:
String, timeStamp: Long)
(sparkSession: SparkSession): Unit = {
val metastore = CarbonEnv.getInstance(sparkSession).carbonMetaStore
val carbonTable = CarbonEnv.getCarbonTable(Some(dbName),
tableName)(sparkSession)
val thriftTable: TableInfo = metastore.getThriftTableInfo(carbonTable)
val evolutionEntryList =
thriftTable.fact_table.schema_evolution.schema_evolution_history
- val updatedTime = evolutionEntryList.get(evolutionEntryList.size() -
1).time_stamp
- if (updatedTime == timeStamp) {
- LOGGER.error(s"Reverting changes for $dbName.$tableName")
- val removedColumns =
evolutionEntryList.get(evolutionEntryList.size() - 1).removed
- thriftTable.fact_table.table_columns.asScala.foreach { columnSchema
=>
- removedColumns.asScala.foreach { removedColumn =>
- if
(columnSchema.column_id.equalsIgnoreCase(removedColumn.column_id) &&
- !columnSchema.isInvisible) {
- columnSchema.setData_type(removedColumn.data_type)
- columnSchema.setPrecision(removedColumn.precision)
- columnSchema.setScale(removedColumn.scale)
- }
+ // here, there can be maximum of two entries for schemaEvolution, when
my operation is
+ // both column rename and datatype change. So check if last two
Evolution entry timestamp is
+ // same, then it is both column rename and datatype change, so revert
two entries,else one entry
+ if (evolutionEntryList.size() > 1 &&
+ (evolutionEntryList.get(evolutionEntryList.size() - 1).time_stamp
== timeStamp) &&
+ (evolutionEntryList.get(evolutionEntryList.size() - 2).time_stamp
== timeStamp)) {
+ LOGGER.error(s"Reverting column rename and datatype changes for
$dbName.$tableName")
+ revertColumnSchemaChanges(thriftTable, evolutionEntryList, true)
+ } else {
+ if (evolutionEntryList.get(evolutionEntryList.size() - 1).time_stamp
== timeStamp) {
+ LOGGER.error(s"Reverting changes for $dbName.$tableName")
+ revertColumnSchemaChanges(thriftTable, evolutionEntryList, false)
+ }
+ }
+ metastore
+
.revertTableSchemaInAlterFailure(carbonTable.getCarbonTableIdentifier,
+ thriftTable, carbonTable.getAbsoluteTableIdentifier,
timeStamp)(sparkSession)
+ }
+
+ /**
+ * This method reverts the column schema in case of failure in alter
datatype change or col rename
+ * @param thriftTable thrift table
+ * @param evolutionEntryList SchemaEvolutionEntry List
+ * @param isBothColRenameAndDataTypeChange true if operation done is
noth rename and datatype chng
+ */
+ private def revertColumnSchemaChanges(thriftTable: TableInfo,
+ evolutionEntryList: util.List[SchemaEvolutionEntry],
+ isBothColRenameAndDataTypeChange: Boolean): Unit = {
+ var removedColumns:
mutable.Buffer[org.apache.carbondata.format.ColumnSchema] = null
+ if (isBothColRenameAndDataTypeChange) {
+ removedColumns = evolutionEntryList.get(evolutionEntryList.size() -
1).removed.asScala ++
+ evolutionEntryList.get(evolutionEntryList.size() -
2).removed.asScala
+ } else {
+ removedColumns = evolutionEntryList.get(evolutionEntryList.size() -
1).removed.asScala
--- End diff --
I can see the same code in multiple classes. Try and refine the code and
move this method at a common place
---
