RE: HiveContext, schemaRDD.printSchema get different dataTypes, feature or a bug? really strange and surprised...
thanks a lot, Hao, finally solved this problem, changes of CSVSerDe are here: https://github.com/chutium/csv-serde/commit/22c667c003e705613c202355a8791978d790591e btw, add jar in spark hive or hive-thriftserver always doesn't work, we build the spark with libraryDependencies += csv-serde ... or maybe should try to add it to SPARK_CLASSPATH ? -- View this message in context: http://apache-spark-developers-list.1001551.n3.nabble.com/HiveContext-schemaRDD-printSchema-get-different-dataTypes-feature-or-a-bug-really-strange-and-surpri-tp8035p8166.html Sent from the Apache Spark Developers List mailing list archive at Nabble.com. - To unsubscribe, e-mail: dev-unsubscr...@spark.apache.org For additional commands, e-mail: dev-h...@spark.apache.org
Re: HiveContext, schemaRDD.printSchema get different dataTypes, feature or a bug? really strange and surprised...
Hi Cheng, thank you very much for helping me to finally find out the secret of this magic... actually we defined this external table with SID STRING REQUEST_ID STRING TIMES_DQ TIMESTAMP TOTAL_PRICE FLOAT ... using desc table ext_fullorders it is only shown as [# col_name data_type comment ] ... [times_dq string from deserializer ] [total_pricestring from deserializer ] ... because, as you said, CSVSerde sets all field object inspectors to javaStringObjectInspector and therefore there are comments from deserializer but in StorageDescriptor, are the real user defined types, using desc extended table ext_fullorders we can see his sd:StorageDescriptor is: FieldSchema(name:times_dq, type:timestamp, comment:null), FieldSchema(name:total_price, type:float, comment:null) and Spark HiveContext reads the schema info from this StorageDescriptor https://github.com/apache/spark/blob/7e191fe29bb09a8560cd75d453c4f7f662dff406/sql/hive/src/main/scala/org/apache/spark/sql/hive/HiveMetastoreCatalog.scala#L316 so, in the SchemaRDD, the fields in Row were filled with strings (via fillObject, all of values were retrieved from CSVSerDe with javaStringObjectInspector) but Spark considers that some of them are float or timestamp (schema info were got from sd:StorageDescriptor) crazy... and sorry for update on the weekend... a little more about how i fand this problem and why it is a trouble for us. we use the new spark thrift server, to query normal managed hive table, it works fine but when we try to access the external tables with custom SerDe such as this CSVSerDe, then we will get this ClassCastException, such as: java.lang.ClassCastException: java.lang.String cannot be cast to java.lang.Float the reason is https://github.com/apache/spark/blob/d94a44d7caaf3fe7559d9ad7b10872fa16cf81ca/sql/hive-thriftserver/src/main/scala/org/apache/spark/sql/hive/thriftserver/server/SparkSQLOperationManager.scala#L104-L105 here Spark's thrift server try to get a float value from SparkRow, because in the schema info (sd:StorageDescriptor) this column is float, but actually in SparkRow, this field was filled with string value... -- View this message in context: http://apache-spark-developers-list.1001551.n3.nabble.com/HiveContext-schemaRDD-printSchema-get-different-dataTypes-feature-or-a-bug-really-strange-and-surpri-tp8035p8157.html Sent from the Apache Spark Developers List mailing list archive at Nabble.com. - To unsubscribe, e-mail: dev-unsubscr...@spark.apache.org For additional commands, e-mail: dev-h...@spark.apache.org
RE: HiveContext, schemaRDD.printSchema get different dataTypes, feature or a bug? really strange and surprised...
Yes, the root cause for that is the output ObjectInspector in SerDe implementation doesn't reflect the real typeinfo. Hive actually provides the API like TypeInfoUtils.getStandardJavaObjectInspectorFromTypeInfo(TypeInfo) for the mapping. You probably need to update the code at https://github.com/ogrodnek/csv-serde/blob/master/src/main/java/com/bizo/hive/serde/csv/CSVSerde.java#L60. -Original Message- From: chutium [mailto:teng@gmail.com] Sent: Monday, September 01, 2014 2:58 AM To: d...@spark.incubator.apache.org Subject: Re: HiveContext, schemaRDD.printSchema get different dataTypes, feature or a bug? really strange and surprised... Hi Cheng, thank you very much for helping me to finally find out the secret of this magic... actually we defined this external table with SID STRING REQUEST_ID STRING TIMES_DQ TIMESTAMP TOTAL_PRICE FLOAT ... using desc table ext_fullorders it is only shown as [# col_name data_type comment ] ... [times_dq string from deserializer ] [total_pricestring from deserializer ] ... because, as you said, CSVSerde sets all field object inspectors to javaStringObjectInspector and therefore there are comments from deserializer but in StorageDescriptor, are the real user defined types, using desc extended table ext_fullorders we can see his sd:StorageDescriptor is: FieldSchema(name:times_dq, type:timestamp, comment:null), FieldSchema(name:total_price, type:float, comment:null) and Spark HiveContext reads the schema info from this StorageDescriptor https://github.com/apache/spark/blob/7e191fe29bb09a8560cd75d453c4f7f662dff406/sql/hive/src/main/scala/org/apache/spark/sql/hive/HiveMetastoreCatalog.scala#L316 so, in the SchemaRDD, the fields in Row were filled with strings (via fillObject, all of values were retrieved from CSVSerDe with javaStringObjectInspector) but Spark considers that some of them are float or timestamp (schema info were got from sd:StorageDescriptor) crazy... and sorry for update on the weekend... a little more about how i fand this problem and why it is a trouble for us. we use the new spark thrift server, to query normal managed hive table, it works fine but when we try to access the external tables with custom SerDe such as this CSVSerDe, then we will get this ClassCastException, such as: java.lang.ClassCastException: java.lang.String cannot be cast to java.lang.Float the reason is https://github.com/apache/spark/blob/d94a44d7caaf3fe7559d9ad7b10872fa16cf81ca/sql/hive-thriftserver/src/main/scala/org/apache/spark/sql/hive/thriftserver/server/SparkSQLOperationManager.scala#L104-L105 here Spark's thrift server try to get a float value from SparkRow, because in the schema info (sd:StorageDescriptor) this column is float, but actually in SparkRow, this field was filled with string value... -- View this message in context: http://apache-spark-developers-list.1001551.n3.nabble.com/HiveContext-schemaRDD-printSchema-get-different-dataTypes-feature-or-a-bug-really-strange-and-surpri-tp8035p8157.html Sent from the Apache Spark Developers List mailing list archive at Nabble.com. - To unsubscribe, e-mail: dev-unsubscr...@spark.apache.org For additional commands, e-mail: dev-h...@spark.apache.org - To unsubscribe, e-mail: dev-unsubscr...@spark.apache.org For additional commands, e-mail: dev-h...@spark.apache.org
Re: HiveContext, schemaRDD.printSchema get different dataTypes, feature or a bug? really strange and surprised...
I believe in your case, the “magic” happens in TableReader.fillObject https://github.com/apache/spark/blob/4fa2fda88fc7beebb579ba808e400113b512533b/core/src/main/scala/org/apache/spark/storage/BlockManager.scala#L706-L712. Here we unwrap the field value according to the object inspector of that field. It seems that somehow a FloatObjectInspector is specified for the total_price field. I don’t think CSVSerde is responsible for this, since it sets all field object inspectors to javaStringObjectInspector (here https://github.com/ogrodnek/csv-serde/blob/f315c1ae4b21a8288eb939e7c10f3b29c1a854ef/src/main/java/com/bizo/hive/serde/csv/CSVSerde.java#L59-L61 ). Which version of Spark SQL are you using? If you are using a snapshot version, please provide the exact Git commit hash. Thanks! On Tue, Aug 26, 2014 at 8:29 AM, chutium teng@gmail.com wrote: oops, i tried on a managed table, column types will not be changed so it is mostly due to the serde lib CSVSerDe ( https://github.com/ogrodnek/csv-serde/blob/master/src/main/java/com/bizo/hive/serde/csv/CSVSerde.java#L123 ) or maybe CSVReader from opencsv?... but if the columns are defined as string, no matter what type returned from custom SerDe or CSVReader, they should be cast to string at the end right? why do not use the schema from hive metadata directly? -- View this message in context: http://apache-spark-developers-list.1001551.n3.nabble.com/HiveContext-schemaRDD-printSchema-get-different-dataTypes-feature-or-a-bug-really-strange-and-surpri-tp8035p8039.html Sent from the Apache Spark Developers List mailing list archive at Nabble.com. - To unsubscribe, e-mail: dev-unsubscr...@spark.apache.org For additional commands, e-mail: dev-h...@spark.apache.org
Re: HiveContext, schemaRDD.printSchema get different dataTypes, feature or a bug? really strange and surprised...
oops, i tried on a managed table, column types will not be changed so it is mostly due to the serde lib CSVSerDe (https://github.com/ogrodnek/csv-serde/blob/master/src/main/java/com/bizo/hive/serde/csv/CSVSerde.java#L123) or maybe CSVReader from opencsv?... but if the columns are defined as string, no matter what type returned from custom SerDe or CSVReader, they should be cast to string at the end right? why do not use the schema from hive metadata directly? -- View this message in context: http://apache-spark-developers-list.1001551.n3.nabble.com/HiveContext-schemaRDD-printSchema-get-different-dataTypes-feature-or-a-bug-really-strange-and-surpri-tp8035p8039.html Sent from the Apache Spark Developers List mailing list archive at Nabble.com. - To unsubscribe, e-mail: dev-unsubscr...@spark.apache.org For additional commands, e-mail: dev-h...@spark.apache.org