[jira] [Commented] (SPARK-5049) ParquetTableScan always prepends the values of partition columns in output rows irrespective of the order of the partition columns in the original SELECT query
[ https://issues.apache.org/jira/browse/SPARK-5049?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanelfocusedCommentId=14272748#comment-14272748 ] Apache Spark commented on SPARK-5049: - User 'marmbrus' has created a pull request for this issue: https://github.com/apache/spark/pull/3990 ParquetTableScan always prepends the values of partition columns in output rows irrespective of the order of the partition columns in the original SELECT query --- Key: SPARK-5049 URL: https://issues.apache.org/jira/browse/SPARK-5049 Project: Spark Issue Type: Bug Components: SQL Affects Versions: 1.1.0, 1.2.0 Reporter: Rahul Aggarwal This happens when ParquetTableScan is being used by turning on spark.sql.hive.convertMetastoreParquet For example: spark-sql set spark.sql.hive.convertMetastoreParquet=true; spark-sql create table table1(a int , b int) partitioned by (p1 string, p2 int) ROW FORMAT SERDE 'parquet.hive.serde.ParquetHiveSerDe' STORED AS INPUTFORMAT 'parquet.hive.DeprecatedParquetInputFormat' OUTPUTFORMAT 'parquet.hive.DeprecatedParquetOutputFormat'; spark-sql insert into table table1 partition(p1='January',p2=1) select key, 10 from src; spark-sql select a, b, p1, p2 from table1 limit 10; January 1 484 10 January 1 484 10 January 1 484 10 January 1 484 10 January 1 484 10 January 1 484 10 January 1 484 10 January 1 484 10 January 1 484 10 January 1 484 10 The correct output should be 484 10 January 1 484 10 January 1 484 10 January 1 484 10 January 1 484 10 January 1 484 10 January 1 484 10 January 1 484 10 January 1 484 10 January 1 484 10 January 1 This also leads to schema mismatch if the query is run using HiveContext and the result is a SchemaRDD. For example : scala import org.apache.spark.sql.hive._ scala val hc = new HiveContext(sc) scala hc.setConf(spark.sql.hive.convertMetastoreParquet, true) scala val res = hc.sql(select a, b, p1, p2 from table1 limit 10) scala res.collect res2: Array[org.apache.spark.sql.Row] = Array([January,1,238,10], [January,1,86,10], [January,1,311,10], [January,1,27,10], [January,1,165,10], [January,1,409,10], [January,1,255,10], [January,1,278,10], [January,1,98,10], [January,1,484,10]) scala res.schema res5: org.apache.spark.sql.StructType = StructType(ArrayBuffer(StructField(a,IntegerType,true), StructField(b,IntegerType,true), StructField(p1,StringType,true), StructField(p2,IntegerType,true))) -- This message was sent by Atlassian JIRA (v6.3.4#6332) - To unsubscribe, e-mail: issues-unsubscr...@spark.apache.org For additional commands, e-mail: issues-h...@spark.apache.org
[jira] [Commented] (SPARK-5049) ParquetTableScan always prepends the values of partition columns in output rows irrespective of the order of the partition columns in the original SELECT query
[ https://issues.apache.org/jira/browse/SPARK-5049?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanelfocusedCommentId=14262546#comment-14262546 ] Apache Spark commented on SPARK-5049: - User 'rahulaggarwalguavus' has created a pull request for this issue: https://github.com/apache/spark/pull/3870 ParquetTableScan always prepends the values of partition columns in output rows irrespective of the order of the partition columns in the original SELECT query --- Key: SPARK-5049 URL: https://issues.apache.org/jira/browse/SPARK-5049 Project: Spark Issue Type: Bug Components: SQL Affects Versions: 1.1.0, 1.2.0 Reporter: Rahul Aggarwal This happens when ParquetTableScan is being used by turning on spark.sql.hive.convertMetastoreParquet For example: spark-sql set spark.sql.hive.convertMetastoreParquet=true; spark-sql create table table1(a int , b int) partitioned by (p1 string, p2 int) ROW FORMAT SERDE 'parquet.hive.serde.ParquetHiveSerDe' STORED AS INPUTFORMAT 'parquet.hive.DeprecatedParquetInputFormat' OUTPUTFORMAT 'parquet.hive.DeprecatedParquetOutputFormat'; spark-sql insert into table table1 partition(p1='January',p2=1) select key, 10 from src; spark-sql select a, b, p1, p2 from table1 limit 10; January 1 484 10 January 1 484 10 January 1 484 10 January 1 484 10 January 1 484 10 January 1 484 10 January 1 484 10 January 1 484 10 January 1 484 10 January 1 484 10 The correct output should be 484 10 January 1 484 10 January 1 484 10 January 1 484 10 January 1 484 10 January 1 484 10 January 1 484 10 January 1 484 10 January 1 484 10 January 1 484 10 January 1 This also leads to schema mismatch if the query is run using HiveContext and the result is a SchemaRDD. For example : scala import org.apache.spark.sql.hive._ scala val hc = new HiveContext(sc) scala hc.setConf(spark.sql.hive.convertMetastoreParquet, true) scala val res = hc.sql(select a, b, p1, p2 from table1 limit 10) scala res.collect res2: Array[org.apache.spark.sql.Row] = Array([January,1,238,10], [January,1,86,10], [January,1,311,10], [January,1,27,10], [January,1,165,10], [January,1,409,10], [January,1,255,10], [January,1,278,10], [January,1,98,10], [January,1,484,10]) scala res.schema res5: org.apache.spark.sql.StructType = StructType(ArrayBuffer(StructField(a,IntegerType,true), StructField(b,IntegerType,true), StructField(p1,StringType,true), StructField(p2,IntegerType,true))) -- This message was sent by Atlassian JIRA (v6.3.4#6332) - To unsubscribe, e-mail: issues-unsubscr...@spark.apache.org For additional commands, e-mail: issues-h...@spark.apache.org
[jira] [Commented] (SPARK-5049) ParquetTableScan always prepends the values of partition columns in output rows irrespective of the order of the partition columns in the original SELECT query
[ https://issues.apache.org/jira/browse/SPARK-5049?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanelfocusedCommentId=14262547#comment-14262547 ] Rahul Aggarwal commented on SPARK-5049: --- https://github.com/apache/spark/pull/3870 ParquetTableScan always prepends the values of partition columns in output rows irrespective of the order of the partition columns in the original SELECT query --- Key: SPARK-5049 URL: https://issues.apache.org/jira/browse/SPARK-5049 Project: Spark Issue Type: Bug Components: SQL Affects Versions: 1.1.0, 1.2.0 Reporter: Rahul Aggarwal This happens when ParquetTableScan is being used by turning on spark.sql.hive.convertMetastoreParquet For example: spark-sql set spark.sql.hive.convertMetastoreParquet=true; spark-sql create table table1(a int , b int) partitioned by (p1 string, p2 int) ROW FORMAT SERDE 'parquet.hive.serde.ParquetHiveSerDe' STORED AS INPUTFORMAT 'parquet.hive.DeprecatedParquetInputFormat' OUTPUTFORMAT 'parquet.hive.DeprecatedParquetOutputFormat'; spark-sql insert into table table1 partition(p1='January',p2=1) select key, 10 from src; spark-sql select a, b, p1, p2 from table1 limit 10; January 1 484 10 January 1 484 10 January 1 484 10 January 1 484 10 January 1 484 10 January 1 484 10 January 1 484 10 January 1 484 10 January 1 484 10 January 1 484 10 The correct output should be 484 10 January 1 484 10 January 1 484 10 January 1 484 10 January 1 484 10 January 1 484 10 January 1 484 10 January 1 484 10 January 1 484 10 January 1 484 10 January 1 This also leads to schema mismatch if the query is run using HiveContext and the result is a SchemaRDD. For example : scala import org.apache.spark.sql.hive._ scala val hc = new HiveContext(sc) scala hc.setConf(spark.sql.hive.convertMetastoreParquet, true) scala val res = hc.sql(select a, b, p1, p2 from table1 limit 10) scala res.collect res2: Array[org.apache.spark.sql.Row] = Array([January,1,238,10], [January,1,86,10], [January,1,311,10], [January,1,27,10], [January,1,165,10], [January,1,409,10], [January,1,255,10], [January,1,278,10], [January,1,98,10], [January,1,484,10]) scala res.schema res5: org.apache.spark.sql.StructType = StructType(ArrayBuffer(StructField(a,IntegerType,true), StructField(b,IntegerType,true), StructField(p1,StringType,true), StructField(p2,IntegerType,true))) -- This message was sent by Atlassian JIRA (v6.3.4#6332) - To unsubscribe, e-mail: issues-unsubscr...@spark.apache.org For additional commands, e-mail: issues-h...@spark.apache.org