Scenario:
Store Hudi Spark dataframe using saveAsTable(data frame writer) method, such that Hudi supported table with org.apache.hudi.hadoop.HoodieParquetInputFormat Input format schema is automaticaly generated.
Currently, saveAsTable works fine with normal (non Hudi table), Which generates default input format.
I want to automate the Hudi table creation with the supported input file format, either with some overridden version saveAsTable or other way staying in the premise of spark.
Hudi DOES NOT support saveAsTable yet.
You have two options to sync hudi tables with a hive metastore:
Sync inside spark
val hudiOptions = Map[String,String](
...
DataSourceWriteOptions.HIVE_URL_OPT_KEY -> "jdbc:hive2://<thrift server host>:<port>",
DataSourceWriteOptions.HIVE_SYNC_ENABLED_OPT_KEY -> "true",
DataSourceWriteOptions.HIVE_DATABASE_OPT_KEY -> "<the database>",
DataSourceWriteOptions.HIVE_TABLE_OPT_KEY -> "<the table>",
DataSourceWriteOptions.HIVE_PARTITION_FIELDS_OPT_KEY -> "<the partition field>",
DataSourceWriteOptions.HIVE_PARTITION_EXTRACTOR_CLASS_OPT_KEY -> classOf[MultiPartKeysValueExtractor].getName
...
)
// Write the DataFrame as a Hudi dataset
// it will appear in hive (similar to saveAsTable..)
test_parquet_partition.write
.format("org.apache.hudi")
.option(DataSourceWriteOptions.OPERATION_OPT_KEY, DataSourceWriteOptions.INSERT_OPERATION_OPT_VAL)
.options(hudiOptions)
.mode(SaveMode.Overwrite)
.save(hudiTablePath)
Sync outside spark
use the bash script after running your hudi spark transformations hudi documentation
cd hudi-hive
./run_sync_tool.sh --jdbc-url jdbc:hive2:\/\/hiveserver:10000 --user hive --pass hive --partitioned-by partition --base-path <basePath> --database default --table <tableName>```)
```bash
cd hudi-hive
./run_sync_tool.sh --jdbc-url jdbc:hive2:\/\/hiveserver:10000 --user hive --pass hive --partitioned-by partition --base-path <basePath> --database default --table <tableName>```
Related
Conf
spark.conf.set('spark.sql.hive.convertMetastoreParquet', "true")
Hive table
spark.sql("create table table_name (ip string, user string) PARTITIONED BY (date date) STORED AS PARQUET")
InsertInto
df.write.insertInto("table_name", overwrite=True)
Error
Caused by: java.lang.ClassNotFoundException: org.apache.spark.sql.hive.execution.HiveFileFormat$$anon$1
Btw insert into ORC table is good. Running on cluster with client mode.
Is your hive-site.xml file present in the Spark config folder?
Edit:
Can you try with:
df.write.mode("overwrite").partitionBy("date").saveAsTable("db.table_name")
It should not be necessary to set any configuration beforehand and to run the SQL create statement.
We have a custom file system class which is an extension of hadoop.fs.FileSystem. This file system has a uri scheme of abfs:///. External hive tables have been created over this data.
CREATE EXTERNAL TABLE testingCustomFileSystem (a string, b int, c double) PARTITIONED BY dt
STORED AS PARQUET
LOCATION 'abfs://<host>:<port>/user/name/path/to/data/'
Using loginbeeline, I'm able to query the table and it would fetch the results.
Now I'm trying to load the same table into a spark dataframe using spark.table('testingCustomFileSystem') and it would throw the following exception
java.io.IOException: No FileSystem for scheme: abfs
at org.apache.hadoop.fs.FileSystem.getFileSystemClass(FileSystem.java:2586)
at org.apache.hadoop.fs.FileSystem.createFileSystem(FileSystem.java:2593)
at org.apache.hadoop.fs.FileSystem.access$200(FileSystem.java:91)
at org.apache.hadoop.fs.FileSystem$Cache.getInternal(FileSystem.java:2632)
at org.apache.hadoop.fs.FileSystem$Cache.get(FileSystem.java:2614)
at org.apache.hadoop.fs.FileSystem.get(FileSystem.java:370)
at org.apache.hadoop.fs.Path.getFileSystem(Path.java:296)
at org.apache.spark.sql.execution.datasources.CatalogFileIndex$$anonfun$2.apply(CatalogFileIndex.scala:77)
at org.apache.spark.sql.execution.datasources.CatalogFileIndex$$anonfun$2.apply(CatalogFileIndex.scala:75)
at scala.collection.immutable.Stream.map(Stream.scala:418)
The jar containing the CustomFileSystem (defining the abfs:// scheme) was loaded into the classpath and was also available.
How does the spark.table parse a hive table definition in a metastore and resolve the uri?.
After looking into the configurations in spark, I happened to notice by setting the following hadoop configuration, I was able to resolve.
hadoopConfiguration.set("fs.abfs.impl",<fqcn of the FileSystemImplementation>)
In Spark, this setting is done during the sparkSession creation (just used only the appName and
like
val spark = SparkSession
.builder()
.setAppName("Name")
.setMaster("yarn")
.getOrCreate()
spark.sparkContext
.hadoopConfiguration.set("fs.abfs.impl",<fqcn of the FileSystemImplementation>)
and it worked !
I have an orc hive table that is created using Hive command
create table orc1(line string) stored as orcfile
I want to write some data to this table using spark sql, I use following code and want the data to be snappy compressed on HDFS
test("test spark orc file format with compression") {
import SESSION.implicits._
Seq("Hello Spark", "Hello Hadoop").toDF("a").createOrReplaceTempView("tmp")
SESSION.sql("set hive.exec.compress.output=true")
SESSION.sql("set mapred.output.compress=true")
SESSION.sql("set mapred.output.compression.codec=org.apache.hadoop.io.compress.SnappyCodec")
SESSION.sql("set io.compression.codecs=org.apache.hadoop.io.compress.SnappyCodec")
SESSION.sql("set mapred.output.compression.type=BLOCK")
SESSION.sql("insert overwrite table orc1 select a from tmp ")
}
The data is written, but it is NOT compressed with snnapy.
If I run the insert overwrite in Hive Beeline/Hive to write the data and use the above set command , then I could see that the table's files are compressed with snappy.
So, I would ask how to write data with snappy compression in Spark SQL 2.1 to orc tables that are created by Hive
You can set the compression to snappy on the create table command like so
create table orc1(line string) stored as orc tblproperties ("orc.compress"="SNAPPY");
Then any inserts to the table will be snappy compressed (I corrected orcfile to orc in the command also).
This is a followup to Save Spark dataframe as dynamic partitioned table in Hive . I tried to use suggestions in the answers but couldn't make it to work in Spark 1.6.1
I am trying to create partitions programmatically from `DataFrame. Here is the relevant code (adapted from a Spark test):
hc.setConf("hive.metastore.warehouse.dir", "tmp/tests")
// hc.setConf("hive.exec.dynamic.partition", "true")
// hc.setConf("hive.exec.dynamic.partition.mode", "nonstrict")
hc.sql("create database if not exists tmp")
hc.sql("drop table if exists tmp.partitiontest1")
Seq(2012 -> "a").toDF("year", "val")
.write
.partitionBy("year")
.mode(SaveMode.Append)
.saveAsTable("tmp.partitiontest1")
hc.sql("show partitions tmp.partitiontest1").show
Full file is here: https://gist.github.com/SashaOv/7c65f03a51c7e8f9c9e018cd42aa4c4a
Partitioned files are created fine on the file system but Hive complains that the table is not partitioned:
======================
HIVE FAILURE OUTPUT
======================
SET hive.support.sql11.reserved.keywords=false
SET hive.metastore.warehouse.dir=tmp/tests
OK
OK
FAILED: Execution Error, return code 1 from org.apache.hadoop.hive.ql.exec.DDLTask. Table tmp.partitiontest1 is not a partitioned table
======================
It looks like the root cause is that org.apache.spark.sql.hive.HiveMetastoreCatalog.newSparkSQLSpecificMetastoreTable always creates table with empty partitions.
Any help to move this forward is appreciated.
EDIT: also created SPARK-14927
I found a workaround: if you pre-create the table then saveAsTable() won't mess with it. So the following works:
hc.setConf("hive.metastore.warehouse.dir", "tmp/tests")
// hc.setConf("hive.exec.dynamic.partition", "true")
// hc.setConf("hive.exec.dynamic.partition.mode", "nonstrict")
hc.sql("create database if not exists tmp")
hc.sql("drop table if exists tmp.partitiontest1")
// Added line:
hc.sql("create table tmp.partitiontest1(val string) partitioned by (year int)")
Seq(2012 -> "a").toDF("year", "val")
.write
.partitionBy("year")
.mode(SaveMode.Append)
.saveAsTable("tmp.partitiontest1")
hc.sql("show partitions tmp.partitiontest1").show
This workaround works in 1.6.1 but not in 1.5.1
I'd like to save data in a Spark (v 1.3.0) dataframe to a Hive table using PySpark.
The documentation states:
"spark.sql.hive.convertMetastoreParquet: When set to false, Spark SQL will use the Hive SerDe for parquet tables instead of the built in support."
Looking at the Spark tutorial, is seems that this property can be set:
from pyspark.sql import HiveContext
sqlContext = HiveContext(sc)
sqlContext.sql("SET spark.sql.hive.convertMetastoreParquet=false")
# code to create dataframe
my_dataframe.saveAsTable("my_dataframe")
However, when I try to query the saved table in Hive it returns:
hive> select * from my_dataframe;
OK
Failed with exception java.io.IOException:java.io.IOException:
hdfs://hadoop01.woolford.io:8020/user/hive/warehouse/my_dataframe/part-r-00001.parquet
not a SequenceFile
How do I save the table so that it's immediately readable in Hive?
I've been there...
The API is kinda misleading on this one.
DataFrame.saveAsTable does not create a Hive table, but an internal Spark table source.
It also stores something into Hive metastore, but not what you intend.
This remark was made by spark-user mailing list regarding Spark 1.3.
If you wish to create a Hive table from Spark, you can use this approach:
1. Use Create Table ... via SparkSQL for Hive metastore.
2. Use DataFrame.insertInto(tableName, overwriteMode) for the actual data (Spark 1.3)
I hit this issue last week and was able to find a workaround
Here's the story:
I can see the table in Hive if I created the table without partitionBy:
spark-shell>someDF.write.mode(SaveMode.Overwrite)
.format("parquet")
.saveAsTable("TBL_HIVE_IS_HAPPY")
hive> desc TBL_HIVE_IS_HAPPY;
OK
user_id string
email string
ts string
But Hive can't understand the table schema(schema is empty...) if I do this:
spark-shell>someDF.write.mode(SaveMode.Overwrite)
.format("parquet")
.saveAsTable("TBL_HIVE_IS_NOT_HAPPY")
hive> desc TBL_HIVE_IS_NOT_HAPPY;
# col_name data_type from_deserializer
[Solution]:
spark-shell>sqlContext.sql("SET spark.sql.hive.convertMetastoreParquet=false")
spark-shell>df.write
.partitionBy("ts")
.mode(SaveMode.Overwrite)
.saveAsTable("Happy_HIVE")//Suppose this table is saved at /apps/hive/warehouse/Happy_HIVE
hive> DROP TABLE IF EXISTS Happy_HIVE;
hive> CREATE EXTERNAL TABLE Happy_HIVE (user_id string,email string,ts string)
PARTITIONED BY(day STRING)
STORED AS PARQUET
LOCATION '/apps/hive/warehouse/Happy_HIVE';
hive> MSCK REPAIR TABLE Happy_HIVE;
The problem is that the datasource table created through Dataframe API(partitionBy+saveAsTable) is not compatible with Hive.(see this link). By setting spark.sql.hive.convertMetastoreParquet to false as suggested in the doc, Spark only puts data onto HDFS,but won't create table on Hive. And then you can manually go into hive shell to create an external table with proper schema&partition definition pointing to the data location.
I've tested this in Spark 1.6.1 and it worked for me. I hope this helps!
I have done in pyspark, spark version 2.3.0 :
create empty table where we need to save/overwrite data like:
create table databaseName.NewTableName like databaseName.OldTableName;
then run below command:
df1.write.mode("overwrite").partitionBy("year","month","day").format("parquet").saveAsTable("databaseName.NewTableName");
The issue is you can't read this table with hive but you can read with spark.
metadata doesn't already exist. In other words, it will add any partitions that exist on HDFS but not in metastore, to the hive metastore.