[ https://issues.apache.org/jira/browse/SPARK-21994?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=16247403#comment-16247403 ]
Guillaume Van Delsen commented on SPARK-21994: ---------------------------------------------- [~vsr] Good news, could share Cloudera logchange about this fix ? Did not find it anywhere. Thanks! > Spark 2.2 can not read Parquet table created by itself > ------------------------------------------------------ > > Key: SPARK-21994 > URL: https://issues.apache.org/jira/browse/SPARK-21994 > Project: Spark > Issue Type: Bug > Components: SQL > Affects Versions: 2.2.0 > Environment: Spark 2.2 on Cloudera CDH 5.10.1, Hive 1.1 > Reporter: Jurgis Pods > > This seems to be a new bug introduced in Spark 2.2, since it did not occur > under Spark 2.1. > When writing a dataframe to a table in Parquet format, Spark SQL does not > write the 'path' of the table to the Hive metastore, unlike in previous > versions. > As a consequence, Spark 2.2 is not able to read the table it just created. It > just outputs the table header without any row content. > A parallel installation of Spark 1.6 at least produces an appropriate error > trace: > {code:java} > 17/09/13 10:22:12 WARN metastore.ObjectStore: Version information not found > in metastore. hive.metastore.schema.verification is not enabled so recording > the schema version 1.1.0 > 17/09/13 10:22:12 WARN metastore.ObjectStore: Failed to get database default, > returning NoSuchObjectException > org.spark-project.guava.util.concurrent.UncheckedExecutionException: > java.util.NoSuchElementException: key not found: path > [...] > {code} > h3. Steps to reproduce: > Run the following in spark2-shell: > {code:java} > scala> val df = spark.sql("show databases") > scala> df.show() > +--------------------+ > | databaseName| > +--------------------+ > | mydb1| > | mydb2| > | default| > | test| > +--------------------+ > scala> df.write.format("parquet").saveAsTable("test.spark22_test") > scala> spark.sql("select * from test.spark22_test").show() > +------------+ > |databaseName| > +------------+ > +------------+{code} > When manually setting the path (causing the data to be saved as external > table), it works: > {code:java} > scala> df.write.option("path", > "/hadoop/eco/hive/warehouse/test.db/spark22_parquet_with_path").format("parquet").saveAsTable("test.spark22_parquet_with_path") > scala> spark.sql("select * from test.spark22_parquet_with_path").show() > +--------------------+ > | databaseName| > +--------------------+ > | mydb1| > | mydb2| > | default| > | test| > +--------------------+ > {code} > A second workaround is to update the metadata of the managed table created by > Spark 2.2: > {code} > spark.sql("alter table test.spark22_test set SERDEPROPERTIES > ('path'='hdfs://my-cluster-name:8020/hadoop/eco/hive/warehouse/test.db/spark22_test')") > spark.catalog.refreshTable("test.spark22_test") > spark.sql("select * from test.spark22_test").show() > +--------------------+ > | databaseName| > +--------------------+ > | mydb1| > | mydb2| > | default| > | test| > +--------------------+ > {code} > It is kind of a disaster that we are not able to read tables created by the > very same Spark version and have to manually specify the path as an explicit > option. -- This message was sent by Atlassian JIRA (v6.4.14#64029) --------------------------------------------------------------------- To unsubscribe, e-mail: issues-unsubscr...@spark.apache.org For additional commands, e-mail: issues-h...@spark.apache.org