[SPARK-10434] [SQL] Fixes Parquet schema of arrays that may contain null

To keep full compatibility of Parquet write path with Spark 1.4, we should rename the innermost field name of arrays that may contain null from "array_element" to "array".

Please refer to [SPARK-10434] [1] for more details.

[1]: https://issues.apache.org/jira/browse/SPARK-10434

Author: Cheng Lian <lian@databricks.com>

Closes #8586 from liancheng/spark-10434/fix-parquet-array-type.
This commit is contained in:
Cheng Lian 2015-09-05 17:50:12 +08:00
parent 7a4f326c00
commit bca8c072bd
2 changed files with 10 additions and 9 deletions

View file

@ -426,13 +426,14 @@ private[parquet] class CatalystSchemaConverter(
// ArrayType and MapType (for Spark versions <= 1.4.x) // ArrayType and MapType (for Spark versions <= 1.4.x)
// =================================================== // ===================================================
// Spark 1.4.x and prior versions convert ArrayType with nullable elements into a 3-level // Spark 1.4.x and prior versions convert `ArrayType` with nullable elements into a 3-level
// LIST structure. This behavior mimics parquet-hive (1.6.0rc3). Note that this case is // `LIST` structure. This behavior is somewhat a hybrid of parquet-hive and parquet-avro
// covered by the backwards-compatibility rules implemented in `isElementType()`. // (1.6.0rc3): the 3-level structure is similar to parquet-hive while the 3rd level element
// field name "array" is borrowed from parquet-avro.
case ArrayType(elementType, nullable @ true) if !followParquetFormatSpec => case ArrayType(elementType, nullable @ true) if !followParquetFormatSpec =>
// <list-repetition> group <name> (LIST) { // <list-repetition> group <name> (LIST) {
// optional group bag { // optional group bag {
// repeated <element-type> element; // repeated <element-type> array;
// } // }
// } // }
ConversionPatterns.listType( ConversionPatterns.listType(
@ -441,8 +442,8 @@ private[parquet] class CatalystSchemaConverter(
Types Types
.buildGroup(REPEATED) .buildGroup(REPEATED)
// "array_element" is the name chosen by parquet-hive (1.7.0 and prior version) // "array_element" is the name chosen by parquet-hive (1.7.0 and prior version)
.addField(convertField(StructField("array_element", elementType, nullable))) .addField(convertField(StructField("array", elementType, nullable)))
.named(CatalystConverter.ARRAY_CONTAINS_NULL_BAG_SCHEMA_NAME)) .named("bag"))
// Spark 1.4.x and prior versions convert ArrayType with non-nullable elements into a 2-level // Spark 1.4.x and prior versions convert ArrayType with non-nullable elements into a 2-level
// LIST structure. This behavior mimics parquet-avro (1.6.0rc3). Note that this case is // LIST structure. This behavior mimics parquet-avro (1.6.0rc3). Note that this case is

View file

@ -197,7 +197,7 @@ class ParquetSchemaInferenceSuite extends ParquetSchemaTest {
|message root { |message root {
| optional group _1 (LIST) { | optional group _1 (LIST) {
| repeated group bag { | repeated group bag {
| optional int32 array_element; | optional int32 array;
| } | }
| } | }
|} |}
@ -266,7 +266,7 @@ class ParquetSchemaInferenceSuite extends ParquetSchemaTest {
| optional binary _1 (UTF8); | optional binary _1 (UTF8);
| optional group _2 (LIST) { | optional group _2 (LIST) {
| repeated group bag { | repeated group bag {
| optional group array_element { | optional group array {
| required int32 _1; | required int32 _1;
| required double _2; | required double _2;
| } | }
@ -645,7 +645,7 @@ class ParquetSchemaSuite extends ParquetSchemaTest {
"""message root { """message root {
| optional group f1 (LIST) { | optional group f1 (LIST) {
| repeated group bag { | repeated group bag {
| optional int32 array_element; | optional int32 array;
| } | }
| } | }
|} |}