Submitting a Spark Jar Job Using Livy

Updated on 2025-01-10 GMT+08:00

View PDF

Introduction to DLI Livy

DLI Livy is an Apache Livy-based client tool used to submit Spark jobs to DLI.

Preparations

Create a queue. Set Queue Usage to For general purpose, that is, the computing resources of the Spark job. For details, see Creating a Queue.
Prepare a Linux ECS for installing DLI Livy.
- Enable ports 30000 to 32767 and port 8998 on the ECS. For details, see Adding a Security Group Rule.
- Install JDK on the ECS. JDK 1.8 is recommended. Configure Java environment variable JAVA_HOME.
- View the ECS details to obtain its private IP address.
Use an enhanced datasource connection to connect the DLI queue to the VPC where the Livy instance is located. For details, see Enhanced Datasource Connection.

Downloading and Installing DLI Livy

NOTE:

The software package used in the following operations is apache-livy-0.7.2.0107-bin.tar.gz. Replace it with the latest one.

Download the DLI Livy software package.
Use WinSCP to upload the obtained software package to the prepared ECS directory.
Log in to ECS as user root and perform the following steps to install DLI Livy:
1. Run the following command to create an installation directory:
  mkdir Livy installation directory
  
  For example, to create the /opt/livy directory, run the mkdir /opt/livy command. The following operations use the /opt/livy installation directory as an example. Replace it as required.
2. Run the following command to decompress the software package to the installation directory:
  tar --extract --file apache-livy-0.7.2.0107-bin.tar.gz --directory /opt/livy --strip-components 1 --no-same-owner
3. Run the following commands to change the configuration file name:
  cd /opt/livy/conf
  
  mv livy-client.conf.template livy-client.conf
  
  mv livy.conf.template livy.conf
  
  mv livy-env.sh.template livy-env.sh
  
  mv log4j.properties.template log4j.properties
  
  mv spark-blacklist.conf.template spark-blacklist.conf
  
  touch spark-defaults.conf

Modifying the DLI Livy Configuration File

Upload the specified DLI Livy JAR package to the OBS bucket directory.
1. Log in to OBS console and create a directory for storing the DLI Livy JAR package in the specified OBS bucket, for example: obs://bucket/livy/jars/.
2. Go to the installation directory of the ECS where the DLI-Livy tool has been installed in 3.a, obtain Livy JAR packages, and upload them to the OBS bucket directory created in 1.a:
  For example, if the installation path is /opt/livy, the JAR packages you need to upload are as follows:
```
/opt/livy/rsc-jars/livy-api-0.7.2.0107.jar
/opt/livy/rsc-jars/livy-rsc-0.7.2.0107.jar
/opt/livy/repl_2.11-jars/livy-core_2.11-0.7.2.0107.jar
/opt/livy/repl_2.11-jars/livy-repl_2.11-0.7.2.0107.jar
```

Modify the DLI Livy configuration file.

Run the following command to modify the /opt/livy/conf/livy-client.conf configuration file:

vi /opt/livy/conf/livy-client.conf

Add the following content to the file and modify the configuration items as required:

# Set the private IP address of the ECS, which can be obtained by running the ifconfig command.
livy.rsc.launcher.address = X.X.X.X
# Set the ports enabled on the ECS.
livy.rsc.launcher.port.range = 30000~32767

Run the following command to modify the /opt/livy/conf/livy.conf configuration file:

vi /opt/livy/conf/livy.conf

Add the following content to the file and modify the configuration items as required:

livy.server.port = 8998
livy.spark.master = yarn
 
livy.server.contextLauncher.custom.class=org.apache.livy.rsc.DliContextLauncher
livy.server.batch.custom.class=org.apache.livy.server.batch.DliBatchSession
livy.server.interactive.custom.class=org.apache.livy.server.interactive.DliInteractiveSession
livy.server.sparkApp.custom.class=org.apache.livy.utils.SparkDliApp
 
livy.server.recovery.mode = recovery
livy.server.recovery.state-store = filesystem
# Change the following file directory of DLI Livy as needed:
livy.server.recovery.state-store.url = file:///opt/livy/store/
 
livy.server.session.timeout-check = true
livy.server.session.timeout = 1800s
livy.server.session.state-retain.sec = 1800s
 
livy.dli.spark.version = 2.3.2
livy.dli.spark.scala-version = 2.11
 
# Enter the OBS bucket path that stores the Livy JAR file.
livy.repl.jars = obs://bucket/livy/jars/livy-core_2.11-0.7.2.0107.jar, obs://bucket/livy/jars/livy-repl_2.11-0.7.2.0107.jar
livy.rsc.jars = obs://bucket/livy/jars/livy-api-0.7.2.0107.jar, obs://bucket/livy/jars/livy-rsc-0.7.2.0107.jar

Run the following command to modify the /opt/livy/conf/spark-defaults.conf configuration file:

vi /opt/livy/conf/spark-defaults.conf

Add the following content to the file. Set the parameters based on Table 1.

# The following parameters can be overwritten when a job is submitted.
spark.yarn.isPython=true
spark.pyspark.python=python3
 
# Enter the production environment URL of DLI.
spark.dli.user.uiBaseAddress=https://console.huaweicloud.com/dli/web
# Set the region where the queue is located.
spark.dli.user.regionName=XXXX
 
# Set the DLI endpoint address.
spark.dli.user.dliEndPoint=XXXX
 
# Enter the name of the created DLI queue.
spark.dli.user.queueName=XXXX
 
# Set the AK used for submitting a job.
spark.dli.user.access.key=XXXX
# Set the SK used for submitting a job.
spark.dli.user.secret.key=XXXX
 
# Set the project ID used for submitting a job.
spark.dli.user.projectId=XXXX

**Table 1** Mandatory parameters in spark-defaults.conf
Parameter	Description
spark.dli.user.regionName	Name of the region where the DLI queue is. Obtain the region name from Regions and Endpoints.
spark.dli.user.dliEndPoint	Endpoint where the DLI queue is located. Obtain the endpoint from Regions and Endpoints.
spark.dli.user.queueName	Queue name.
spark.dli.user.access.key	User's AK/SK. The user must have Spark job permissions. For details, see Permissions Management. For how to obtain the AK/SK, see Obtaining the AK/SK.
spark.dli.user.secret.key
spark.dli.user.projectId	Project ID. Obtain it by referring to Obtaining a Project ID.

The following parameters are optional. Set them based on the parameter description and site requirements. For details about these parameters, see Spark Configuration.

**Table 2** Optional parameters in spark-defaults.conf
Spark Job Parameter	Spark Batch Processing Parameter	Remarks
spark.dli.user.file	file	Not required for connecting to the notebook tool
spark.dli.user.className	class_name	Not required for connecting to the notebook tool
spark.dli.user.scType	sc_type	Same as the native Livy configuration
spark.dli.user.args	args	Same as the native Livy configuration
spark.submit.pyFiles	python_files	Same as the native Livy configuration
spark.files	files	Same as the native Livy configuration
spark.dli.user.modules	modules	-
spark.dli.user.image	image	Custom image used for submitting a job. This parameter is available for container clusters only and is not set by default.
spark.dli.user.autoRecovery	auto_recovery	-
spark.dli.user.maxRetryTimes	max_retry_times	-
spark.dli.user.catalogName	catalog_name	To access metadata, set this parameter to dli.

Starting DLI Livy

Run the following command to go to the DLI Livy installation directory:

Example: cd /opt/livy
Run the following command to start DLI Livy:

./bin/livy-server start

Submitting a Spark Job Using DLI Livy

The following demonstrates how to submit a Spark job to DLI using DLI Livy and by running the curl command.

Upload the JAR file of the developed Spark job program to the OBS directory.

For example, upload spark-examples_2.11-XXXX.jar to the obs://bucket/path directory.
Log in to the ECS server where DLI Livy is installed as user root.

Run the curl command to submit a Spark job request to DLI using DLI Livy.

NOTE:

ECS_IP indicates the private IP address of the ECS where DLI Livy is installed.

curl --location --request POST 'http://ECS_IP:8998/batches' \
--header 'Content-Type: application/json' \
--data '{
    "driverMemory": "3G",
    "driverCores": 1,
    "executorMemory": "2G",
    "executorCores": 1,
    "numExecutors": 1,
    "args": [
        "1000"
    ],
    "file": "obs://bucket/path/spark-examples_2.11-XXXX.jar",
    "className": "org.apache.spark.examples.SparkPi",
    "conf": {
        "spark.dynamicAllocation.minExecutors": 1,
        "spark.executor.instances": 1,
        "spark.dynamicAllocation.initialExecutors": 1,
        "spark.dynamicAllocation.maxExecutors": 2
    }
}'

Parent topic: Connecting to DLI Using a Client

Previous topic: Using Spark-submit to Submit a Spark Jar Job

Next topic: SQL Jobs