需要帮助在 Apache Airflow 中运行 spark-submit

Question

I am a relatively new user to Python and Airflow and am having a very difficult time getting spark-submit to run in an Airflow task. My goal is to get the following DAG task to run successfully

from datetime import datetime, timedelta
from airflow import DAG
from airflow.contrib.operators.spark_submit_operator import SparkSubmitOperator
from airflow.operators.bash_operator import BashOperator

default_args = {
    'owner': 'matthew',
    'start_date': datetime(2019, 7, 8)
}

dag = DAG('CustomCreate_test2',
          default_args=default_args,
          schedule_interval=timedelta(days=1))

t3 = BashOperator(
    task_id='run_test',
    bash_command='spark-submit --class CLASSPATH.CustomCreate ~/IdeaProjects/custom-create-job/build/libs/custom-create.jar',
    dag=dag
)

I know the problem lies with Airflow and not with the bash because when I run the command spark-submit --class CLASSPATH.CustomCreate ~/IdeaProjects/custom-create-job/build/libs/custom-create.jar in the terminal it runs successfully.

I have been getting the following error from the Airflow logs

...
[2019-08-28 15:55:34,750] {bash_operator.py:132} INFO - Command exited with return code 1
[2019-08-28 15:55:34,764] {taskinstance.py:1047} ERROR - Bash command failed
Traceback (most recent call last):
  File "/Users/matcordo2/.virtualenv/airflow/lib/python3.7/site-packages/airflow/models/taskinstance.py", line 922, in _run_raw_task
    result = task_copy.execute(context=context)
  File "/Users/matcordo2/.virtualenv/airflow/lib/python3.7/site-packages/airflow/operators/bash_operator.py", line 136, in execute
    raise AirflowException("Bash command failed")
airflow.exceptions.AirflowException: Bash command failed
...

I have also tried working with the SparkSubmitOperator(...) but have had no successful runs using it, I have only ever ended up with error logs like the following

...
[2019-08-28 15:54:49,749] {logging_mixin.py:95} INFO - [[34m2019-08-28 15:54:49,749[0m] {[34mspark_submit_hook.py:[0m427} INFO[0m - at org.apache.spark.deploy.SparkSubmit.main(SparkSubmit.scala)[0m
[2019-08-28 15:54:49,803] {taskinstance.py:1047} ERROR - Cannot execute: ['spark-submit', '--master', 'yarn', '--num-executors', '2', '--total-executor-cores', '1', '--executor-cores', '1', '--executor-memory', '2g', '--driver-memory', '1g', '--name', 'CustomCreate', '--class', 'CLASSPATH.CustomCreate', '--verbose', '--queue', 'root.default', '--deploy-mode', 'cluster', '~/IdeaProjects/custom-create-job/build/libs/custom-create.jar']. Error code is: 1.
...

Is there something I have to do using SparkSubmitOperator(...) before I can run the spark-submit ... command in a BashOperator(...) task?

Is there a way to run my spark-submit command directly from the SparkSubmitOperator(...) task?

Is there anything that I have to do to spark_default in the Admin->Connections page of Airflow?

Is there anything that must be set in the Admin->Users page of Airflow? Is there anything that must be set to allow Airflow to run spark or run a jar file created by a specific user? If so, what/how?

Answer 1

I found a workaround that solved this problem.

Create a new ssh connection (or edit the default) like the one below in the Airflow Admin->Connection page Airflow SSH Connection Example

Below is a text version if you cannot see the image
Conn ID : ssh_connection
Conn Type : SSH
Host : HOST IP ADDRESS
Username : HOST USERNAME
Password : HOST PASSWORD
Port :
Extra : {"key_file": "/PATH TO HOME DIR/airflow/.ssh/id_rsa", "allow_host_key_change": "true", "no_host_key_check": "true"}

Then make the proper adjustments to your python script

from airflow import DAG
from airflow.contrib.operators.ssh_operator import SSHOperator
from airflow.operators.bash_operator import BashOperator
from datetime import datetime, timedelta

default_args = {
    'owner': 'matthew',
    'start_date': datetime(2019, 8, 28)
}

dag = DAG('custom-create',
          default_args=default_args,
          schedule_interval=timedelta(days=1),
          params={'project_source': '~/IdeaProjects/custom-create-job',
                  'spark_submit': '/usr/local/bin/spark-submit',
                  'classpath': 'CLASSPATH.CustomCreate',
                  'jar_file': 'build/libs/custom-create.jar'}
          )

templated_bash_command = """
    echo 'HOSTNAME: $HOSTNAME' #To check that you are properly connected to the host
    cd {{ params.project_source }}
    {{ params.spark_submit }} --class {{ classpath }} {{ jar_file }}
"""

t1 = SSHOperator(
    task_id="SSH_task",
    ssh_conn_id='ssh_connection',
    command=templated_bash_command,
    dag=dag
)

I hope this solution helps other people who may be running into a similar problem like I was.

Answer 2

A similar question has already been answered - StackOverFlow Link

I guess the above link will help you.

In the future if you would like to implement the same thing on AWS EMR Or AZURE , then there you have an a beautiful way to schedule spark jobs - Airflow Documentation

An Example for the above - (AWS EMR)

 <airflow_EMR_task> =cover_open(json.load(open(airflow_home+'/<tasks_json_containing_all_spark_configurations>')))
 <airflow_EMR_task>['Job']['Name'] =  <airflow_EMR_task>['Job']['Name'] + <'optional_postfix'>
airflow_swperformance_cpu_creator = EmrRunJobFlowOperator(
    task_id='<task_id>',
    job_flow_overrides= <airflow_EMR_task>['Job'],
    aws_conn_id='aws_default',
    emr_conn_id='emr_default',
    retries=1,
    dag=dag
)

And A Simple JSON would be - (the same json file as mentioned above )

{
    "Job": {
        "Name": "<task_name>",
        "LogUri": "<task_log_uri>",
        "ReleaseLabel": "emr-5.6.0",
        "Applications": [
            {
                "Name": "Spark"
            },
            {
                "Name": "Hive"
            }
        ],
        "Tags": [
            {
                "Key" : "<any_tag>",
                "Value" : "<any_tag>"
            },
            {
                "Key" : "<any tag>",
                "Value": "<any_tag>"
            },
            {
                "Key" : "<any_tag>",
                "Value": "<any_tag value>"
            }
        ],
        "JobFlowRole": "EMR_EC2_DefaultRole_Stable",
        "ServiceRole": "EMR_DefaultRole",
        "VisibleToAllUsers": true,
        "Configurations": [
            {
                "Classification": "spark-defaults",
                "Properties": {
                    "spark.driver.extraJavaOptions":"-XX:+UseParallelGC -XX:+UseParallelOldGC -XX:CMSInitiatingOccupancyFraction=70 -XX:MaxHeapFreeRatio=70 -XX:+CMSClassUnloadingEnabled -XX:+ExitOnOutOfMemoryError -Dlog4j.configuration=log4j-custom.properties",
                    "spark.executor.extraJavaOptions":"-verbose:gc -XX:+PrintGCDetails -XX:+PrintGCDateStamps -XX:+UseParallelGC -XX:+UseParallelOldGC -XX:CMSInitiatingOccupancyFraction=70 -XX:MaxHeapFreeRatio=70 -XX:+CMSClassUnloadingEnabled -XX:+ExitOnOutOfMemoryError -Dlog4j.configuration=log4j-custom.properties",
                    "spark.scheduler.mode": "FAIR",
                    "spark.eventLog.enabled": "true",
                    "spark.serializer": "org.apache.spark.serializer.KryoSerializer",
                    "spark.sql.orc.filterPushdown": "true",
                    "spark.dynamicAllocation.enabled": "false"
                },
                "Configurations": []
            },
            {
                "Classification": "spark",
                "Properties": {
                    "maximizeResourceAllocation": "true"
                },
                "Configurations": []
            },
            {
                "Classification": "hive-site",
                "Properties": {
                    "javax.jdo.option.ConnectionUserName": "<HIVE USERNAME IF ANY>",
                    "javax.jdo.option.ConnectionPassword": "<<hive_connection_password>>",
                    "javax.jdo.option.ConnectionURL": "<Hive_URL_IF_ANY"
                },
                "Configurations": []
            },
            {
                "Classification": "emrfs-site",
                "Properties": {
                    "fs.s3.serverSideEncryption.kms.keyId": "<<encryption_key>>",
                    "fs.s3.enableServerSideEncryption": "true"
                },
                "Configurations": []
            },
            {
                "Classification":"spark-env",
                "Configurations":[{
                    "Classification":"export",
                    "Configurations":[],
                    "Properties": {
                        "ANY_ENV_VARIABLE_REQUIRED_FOR_SPECIFIC_JOB",
                        "ANY_ENV_VARIABLE_REQUIRED_FOR_SPECIFIC_JOB",
                        "ANY_ENV_VARIABLE_REQUIRED_FOR_SPECIFIC_JOB",
                        "ANY_ENV_VARIABLE_REQUIRED_FOR_SPECIFIC_JOB",
                        "ANY_ENV_VARIABLE_REQUIRED_FOR_SPECIFIC_JOB"
            "S3_BUCKET_NAME":"<S3_bucekt_naem_if_Required>"
                    }
                }
                ]}
        ],
        "Instances": {
            "Ec2KeyName": "<ssh_key>",
            "KeepJobFlowAliveWhenNoSteps": false,
            "Ec2SubnetId": "<subnet>",
            "EmrManagedSlaveSecurityGroup": "<security_group>",
            "EmrManagedMasterSecurityGroup": "<security_group_parameter>",
            "AdditionalSlaveSecurityGroups": [
                "<self_explanatory>"
            ],
            "AdditionalMasterSecurityGroups": [
                "<self_explanatory>"
            ],
            "InstanceGroups": [
                {
                    "InstanceCount": 4,
                    "InstanceRole": "CORE",
                    "InstanceType": "r3.xlarge",
                    "Name": "Core instance group - 2"
                },
                {
                    "InstanceCount": 1,
                    "InstanceRole": "MASTER",
                    "InstanceType": "r3.xlarge",
                    "Name": "Master instance group - 1"
                }
            ]
        },
        "BootstrapActions": [],
        "Steps": [
            {
                "Name": "download-dependencies",
                "HadoopJarStep": {
                    "Jar": "command-runner.jar",
                    "Args": [
                        "aws",
                        "s3",
                        "cp",
                        "<appropriate_s3_location>",
                        "/home/hadoop",
                        "--recursive"
                    ],
                    "Properties": []
                },
                "ActionOnFailure": "TERMINATE_CLUSTER"
            },
            {
                "Name": "run-script",
                "HadoopJarStep": {
                    "Jar": "command-runner.jar",
                    "Args": [
                        "sudo",
                        "/bin/sh",
                        "/home/hadoop/pre-executor.sh"
                    ],
                    "Properties": []
                },
                "ActionOnFailure": "TERMINATE_CLUSTER"
            },
            {
                "Name": "spark-submit",
                "HadoopJarStep": {
                    "Jar": "command-runner.jar",
                    "Args": [
                        "spark-submit",
                        "/home/hadoop/analytics-job.jar",
            "--run-gold-job-only"
                    ],
                    "Properties": []
                },
                "ActionOnFailure": "TERMINATE_CLUSTER"
            }
        ]
    }
}

And that will be all.

Answer 3

Not sure if still relevant, but I solved a similar problem simply by using the absolute path to the file (full path without the '~'). Not sure why, maybe airflow consider 'home' as another place.