I'm currently running PySpark via local mode. I want to be able to efficiently output parquet files to S3 via the S3 Directory Committer. This PySpark instance is using the local disk, not HDFS, as it is being submitted via spark-submit --master local[*].
I can successfully write to my S3 Instance without enabling the directory committer. However, this involves writing staging files to S3 and renaming them, which is slow and unreliable. I would like for Spark to write to my local filesystem as a temporary store, and then copy to S3.
I have the following configuration in my PySpark conf:
self.spark.conf.set("spark.sql.execution.arrow.pyspark.enabled", "true")
self.spark.conf.set("spark.hadoop.fs.s3a.committer.name", "directory")
self.spark.conf.set("spark.sql.sources.commitProtocolClass", "org.apache.spark.internal.io.cloud.PathOutputCommitProtocol")
self.spark.conf.set("spark.sql.parquet.output.committer.class", "org.apache.spark.internal.io.cloud.BindingParquetOutputCommitter")
My spark-submit command looks like this:
spark-submit --master local[*] --py-files files.zip --packages com.amazonaws:aws-java-sdk:1.11.375,org.apache.hadoop:hadoop-aws:3.2.0,org.apache.spark.internal.io.cloud.PathOutputCommitProtocol --driver-memory 4G --name clean-raw-recording_data main.py
spark-submit gives me the following error, due to the requisite JAR not being in place:
java.lang.ClassNotFoundException: org.apache.spark.internal.io.cloud.PathOutputCommitProtocol
My questions are:
Which JAR (specifically, the maven coordinates) do I need to include in spark-submit --packages in order to be able to reference PathOutputCommitProtocol?
Once I have (1) working, will I be able to use PySpark's local mode to stage temporary files on the local filesystem? Or is HDFS a strict requirement?
I need this to be running in local mode, not cluster mode.
EDIT:
I got this to work with the following configuration:
Using pyspark version 3.1.2 and the package
org.apache.spark:spark-hadoop-cloud_2.12:3.1.1.3.1.7270.0-253.
I needed to add the cloudera repository using the --repositories option for spark-submit:
spark-submit --repositories https://repository.cloudera.com/artifactory/cloudera-repos/ --packages com.amazonaws:aws-java-sdk:1.11.375,org.apache.hadoop:hadoop-aws:3.2.0,org.apache.spark:spark-hadoop-cloud_2.12:3.1.1.3.1.7270.0-253
you need the spark-hadoop-cloud module for the release of spark you are using
the committer is happy using the local fs (it's now the public integration test suites work https://github.com/hortonworks-spark/cloud-integration. all that's needed is a "real" filesystem shared across all workers and the spark driver, so the driver gets the manifests of each pending commit.
print the _SUCCESS file after a job to see what the committer did: 0 byte file == old committer, JSON with diagnostics == new one
Related
I have existing EMR cluster running and wish to create DF from Postgresql DB source.
To do this, it seems you need to modify the spark-defaults.conf with the updated spark.driver.extraClassPath and point to the relevant PostgreSQL JAR that has been already downloaded on master & slave nodes, or you can add these as arguments to a spark-submit job.
Since I want to use existing Jupyter notebook to wrangle the data, and not really looking to relaunch cluster, what is the most efficient way to resolve this?
I tried the following:
Create new directory (/usr/lib/postgresql/ on master and slaves and copied PostgreSQL jar to it. (postgresql-9.41207.jre6.jar)
Edited spark-default.conf to include wildcard location
spark.driver.extraClassPath :/usr/lib/postgresql/*:/usr/lib/hadoop/hadoop-aws.jar:/usr/share/aws/aws-java-sdk/*:/usr/share/aws/emr/emrfs/conf:/$
Tried to create dataframe in Jupyter cell using the following code:
SQL_CONN = "jdbc:postgresql://some_postgresql_db:5432/dbname?user=user&password=password"
spark.read.jdbc(SQL_CONN, table="someTable", properties={"driver":'com.postgresql.jdbc.Driver'})
I get a Java error as per below:
Py4JJavaError: An error occurred while calling o396.jdbc.
: java.lang.ClassNotFoundException: com.postgresql.jdbc.Driver
Help appreciated.
I think you don't need to copy postgres jar in slaves as the driver programme and cluster manager take care everything. I've created dataframe from Postgres external source by the following way:
Download postgres driver jar:
cd $HOME && wget https://jdbc.postgresql.org/download/postgresql-42.2.5.jar
Create dataframe:
atrribute = {'url' : 'jdbc:postgresql://{host}:{port}/{db}?user={user}&password={password}' \
.format(host=<host>, port=<port>, db=<db>, user=<user>, password=<password>),
'database' : <db>,
'dbtable' : <select * from table>}
df=spark.read.format('jdbc').options(**attribute).load()
Submit to spark job:
Add the the downloaded jar to driver class path while submitting the spark job.
--properties spark.driver.extraClassPath=$HOME/postgresql-42.2.5.jar,spark.jars.packages=org.postgresql:postgresql:42.2.5
Check the github repo of the Driver. The class path seems to be something like this org.postgresql.Driver. Try using the same.
I'm trying to access a remote cloudera HDFS cluster from my local PC (win7). As cricket_007 suggested in my last question I did the following things:
(1) I created the next Spark session
val spark = SparkSession
.builder()
.appName("API")
.config("spark.sql.warehouse.dir", "/user/hive/warehouse")
.master("local")
.enableHiveSupport()
.getOrCreate()
(2) I copied the next files from the cluster :
core-site.xml
hdfs-site.xml
hive-site.xml
mapred-site.xml
yarn-site.xml
and configured the variable HADOOP_CONF_DIR to the directory that contains them
(3) I downloaded Spark and configured the variables SPARK_HOME and SPARK_CONF_DIR
(4) I downloaded winutils and set it in the path variable. I changed the permissions of /tmp/hive to 777.
When the master set to local I see only the default database which means it doesn't identify the XML files. When it is set to yarn the screen is stuck and it looks like my pc is thinking but it is taking to much time and doesn't end. When I use local and I also use the line .config("hive.metastore.uris","thrift://MyMaster:9083") everything works well.
Why might this be happening? Why locally I see only the default database? Why when the master set to yarn I can't connect and it is stuck? And why when I add the config line it solved my problem only locally?
I'm trying to read/write data to an Azure SQL Data Warehouse from a spark on demand HDInsight cluster.
I can do this from a normal HDInsight spark cluster by using a script action to install the jdbc driver but I don't think it's possible to run script actions on the on demand clusters.
I've tried
Copying the files from %user%.m2\repository\com\microsoft\sqlserver\mssql-jdbc\6.2.2.jre8 up to blob storage in a folder called jars next to where the built spark code is.
including the driver dependency in the built jar file
Both of these led to a java.lang.NoClassDefFoundError
I'm not too familiar with scala/maven/JVM/etc so not sure what else to try or include in this SO question.
Scala code i'm trying to run is
val sqlContext = SparkSession.builder().appName("GenerateEventsSql").getOrCreate()
val jdbcSqlConnStr = "jdbc:sqlserver://someserver.database.windows.net:1433;databaseName=myDW;user=admin;password=XXXX;"
val tableName = "dbo.SomeTable"
val allTableData = sqlContext.read.format("jdbc")
.options(Map(
"driver" -> "com.microsoft.sqlserver.jdbc.SQLServerDriver",
"url" -> jdbcSqlConnStr, "dbtable" -> tableName)
)
.load()
Jars on Blob storage folder are not accessible to the Class path of HDinsight spark job. You need to copy the jar files to the local host for example /tmp/jars/xyz.jar and mention the same in Spark-submit command.
For e.g.
nohup spark-submit --jars /tmp/jars/xyz.jar
I have a spark cluster I created via google dataproc. I want to be able to use the csv library from databricks (see https://github.com/databricks/spark-csv). So I first tested it like this:
I started a ssh session with the master node of my cluster, then I input:
pyspark --packages com.databricks:spark-csv_2.11:1.2.0
Then it launched a pyspark shell in which I input:
df = sqlContext.read.format('com.databricks.spark.csv').options(header='true', inferschema='true').load('gs:/xxxx/foo.csv')
df.show()
And it worked.
My next step is to launch this job from my main machine using the command:
gcloud beta dataproc jobs submit pyspark --cluster <my-dataproc-cluster> my_job.py
But here It does not work and I get an error. I think because I did not gave the --packages com.databricks:spark-csv_2.11:1.2.0 as an argument, but I tried 10 different ways to give it and I did not manage.
My question are:
was the databricks csv library installed after I typed pyspark --packages com.databricks:spark-csv_2.11:1.2.0
can I write a line in my job.py in order to import it?
or what params should I give to my gcloud command to import it or install it?
Short Answer
There are quirks in ordering of arguments where --packages isn't accepted by spark-submit if it comes after the my_job.py argument. To workaround this, you can do the following when submitting from Dataproc's CLI:
gcloud beta dataproc jobs submit pyspark --cluster <my-dataproc-cluster> \
--properties spark.jars.packages=com.databricks:spark-csv_2.11:1.2.0 my_job.py
Basically, just add --properties spark.jars.packages=com.databricks:spark-csv_2.11:1.2.0 before the .py file in your command.
Long Answer
So, this is actually a different issue than the known lack of support for --jars in gcloud beta dataproc jobs submit pyspark; it appears that without Dataproc explicitly recognizing --packages as a special spark-submit-level flag, it tries to pass it after the application arguments so that spark-submit lets the --packages fall through as an application argument rather than properly parsing it as a submission-level option. Indeed, in an SSH session, the following does not work:
# Doesn't work if job.py depends on that package.
spark-submit job.py --packages com.databricks:spark-csv_2.11:1.2.0
But switching the order of the arguments does work again, even though in the pyspark case, both orderings work:
# Works with dependencies on that package.
spark-submit --packages com.databricks:spark-csv_2.11:1.2.0 job.py
pyspark job.py --packages com.databricks:spark-csv_2.11:1.2.0
pyspark --packages com.databricks:spark-csv_2.11:1.2.0 job.py
So even though spark-submit job.py is supposed to be a drop-in replacement for everything that previously called pyspark job.py, the difference in parse ordering for things like --packages means it's not actually a 100% compatible migration. This might be something to follow up with on the Spark side.
Anyhow, fortunately there's a workaround, since --packages is just another alias for the Spark property spark.jars.packages, and Dataproc's CLI supports properties just fine. So you can just do the following:
gcloud beta dataproc jobs submit pyspark --cluster <my-dataproc-cluster> \
--properties spark.jars.packages=com.databricks:spark-csv_2.11:1.2.0 my_job.py
Note that the --properties must come before the my_job.py, otherwise it gets sent as an application argument rather than as a configuration flag. Hope that works for you! Note that the equivalent in an SSH session would be spark-submit --packages com.databricks:spark-csv_2.11:1.2.0 job.py.
Additionally to #Dennis.
Note that if you need to load multiple external packages, you need to specify a custom escape character like so:
--properties ^#^spark.jars.packages=org.elasticsearch:elasticsearch-spark_2.10:2.3.2,com.databricks:spark-avro_2.10:2.0.1
Note the ^#^ right before the package list.
See gcloud topic escaping for more details.
Given below is the version of Spark & Hive I have installed in my system
Spark : spark-1.4.0-bin-hadoop2.6
Hive : apache-hive-1.0.0-bin
I have configured the Hive installation to use MySQL as Metastore. The goal is to access the MySQL Metastore & execute HiveQL queries inside spark-shell(using HiveContext)
So far I am able to execute the HiveQL queries by accessing the Derby Metastore(As described here, believe Spark-1.4 comes bundled with Hive 0.13.1 which in turn uses the internal Derby database as Metastore)
Then I tried to point spark-shell to my external Metastore(MySQL in this case) by setting the property(as suggested here) given below in $SPARK_HOME/conf/spark-defaults.conf,
spark.sql.hive.metastore.jars /home/mountain/hv/lib:/home/mountain/hp/lib
I have also copied $HIVE_HOME/conf/hive-site.xml into $SPARK_HOME/conf. But I am getting the following exception when I start the spark-shell
mountain#mountain:~/del$ spark-shell
Spark context available as sc.
java.lang.ClassNotFoundException: java.lang.NoClassDefFoundError:
org/apache/hadoop/hive/ql/session/SessionState when creating Hive client
using classpath: file:/home/mountain/hv/lib/, file:/home/mountain/hp/lib/
Please make sure that jars for your version of hive and hadoop are
included in the paths passed to spark.sql.hive.metastore.jars.
Am I missing something (or) not setting the property spark.sql.hive.metastore.jars correctly?
Note: In Linux Mint verified.
If you are setting properties in spark-defaults.conf, spark will take those settings only when you submit your job using spark-submit.
file: spark-defaults.conf
spark.driver.extraJavaOptions -Dlog4j.configuration=file:log4j.properties -Dspark.yarn.app.container.log.dir=app-logs -Dlogfile.name=hello-spark
spark.jars.packages org.apache.spark:spark-sql-kafka-0-10_2.12:3.0.1,org.apache.spark:spark-avro_2.12:3.0.1
In the terminal run your job say wordcount.py
spark-submit /path-to-file/wordcount.py
If you want to run your job in development mode from an IDE then you should use config() method. Here we will set Kafka jar packages
spark = SparkSession.builder \
.appName('Hello Spark') \
.master('local[3]') \
.config("spark.streaming.stopGracefullyOnShutdown", "true") \
.config("spark.jars.packages", "org.apache.spark:spark-sql-kafka-0-10_2.12:3.0.1") \
.getOrCreate()
Corrupted version of hive-site.xml will cause this... please copy the correct hive-site.xml