Skip to content

RelationConversions PostHoc Logical Evaluation Rule

RelationConversions is a HiveSessionStateBuilder.md#postHocResolutionRules[posthoc logical resolution rule] that the HiveSessionStateBuilder.md#analyzer[Hive-specific logical analyzer] uses to <> with Parquet and ORC storage formats.

CAUTION: FIXME Show example of a hive table, e.g. spark.table(...)

RelationConversions is <> when the HiveSessionStateBuilder.md#analyzer[Hive-specific logical analyzer] is created.

=== [[creating-instance]] Creating RelationConversions Instance

RelationConversions takes the following when created:

=== [[apply]] Executing Rule -- apply Method

[source, scala]

apply( plan: LogicalPlan): LogicalPlan


NOTE: apply is part of the ../catalyst/Rule.md#apply[Rule] contract to execute (apply) a rule on a ../spark-sql-LogicalPlan.md[LogicalPlan].

apply traverses the input ../spark-sql-LogicalPlan.md[logical plan] looking for ../InsertIntoTable.md[InsertIntoTables] (over a HiveTableRelation.md[HiveTableRelation]) or HiveTableRelation.md[HiveTableRelation] logical operators:

[[apply-InsertIntoTable]] * For an ../InsertIntoTable.md[InsertIntoTable] over a HiveTableRelation.md[HiveTableRelation] that is HiveTableRelation.md#isPartitioned[non-partitioned] and <>, apply creates a new InsertIntoTable with the HiveTableRelation <>.

[[apply-HiveTableRelation]] * For a HiveTableRelation logical operator alone apply...FIXME

=== [[isConvertible]] Does Table Use Parquet or ORC SerDe? -- isConvertible Internal Method

[source, scala]

isConvertible( relation: HiveTableRelation): Boolean


isConvertible is positive when the input HiveTableRelation.md#tableMeta[HiveTableRelation] is a parquet or ORC table (and corresponding SQL properties are enabled).

Internally, isConvertible takes the Hive SerDe of the table (from HiveTableRelation.md#tableMeta[table metadata]) if available or assumes no SerDe.

isConvertible is turned on when either condition holds:

NOTE: isConvertible is used when RelationConversions is <>.

=== [[convert]] Converting HiveTableRelation to HadoopFsRelation -- convert Internal Method

[source, scala]

convert( relation: HiveTableRelation): LogicalRelation


convert branches based on the SerDe of (the storage format of) the input HiveTableRelation logical operator.

For Hive tables in parquet format, convert creates options with one extra mergeSchema per spark.sql.hive.convertMetastoreParquet.mergeSchema configuration property and requests the HiveMetastoreCatalog to convert a HiveTableRelation to a LogicalRelation (with ParquetFileFormat).

For non-parquet Hive tables, convert assumes ORC format:

  • When spark.sql.orc.impl configuration property is native (default) convert requests HiveMetastoreCatalog to HiveMetastoreCatalog.md#convertToLogicalRelation[convert a HiveTableRelation to a LogicalRelation over a HadoopFsRelation] (with org.apache.spark.sql.execution.datasources.orc.OrcFileFormat as fileFormatClass).

  • Otherwise, convert requests HiveMetastoreCatalog to HiveMetastoreCatalog.md#convertToLogicalRelation[convert a HiveTableRelation to a LogicalRelation over a HadoopFsRelation] (with org.apache.spark.sql.hive.orc.OrcFileFormat as fileFormatClass).

NOTE: convert uses the <> to access the HiveMetastoreCatalog.

[NOTE]

convert is used when RelationConversions logical evaluation rule is <> and does the following transformations:

  • Transforms an ../InsertIntoTable.md[InsertIntoTable] over a HiveTableRelation with a Hive table (i.e. with hive provider) that is not partitioned and uses parquet or orc data storage format

* Transforms an HiveTableRelation.md[HiveTableRelation] with a Hive table (i.e. with hive provider) that uses parquet or orc data storage format


Last update: 2020-11-08