Arc has been packaged as a Docker image to simplify deployment as a stateless process on cloud infrastructure. As there are multiple versions of Arc, Spark, Scala and Hadoop see the for the relevant version.

Running a Job

An example command to start a job is:

docker run \
-e "ETL_CONF_ENV=production" \
-e "ETL_CONF_JOB_PATH=/opt/tutorial/basic/job/0" \
-it -p 4040:4040 triplai/arc:arc_2.9.0_spark_2.4.5_scala_2.12_hadoop_2.9.2_1.1.0 \
bin/spark-submit \
--master local[*] \
--class ai.tripl.arc.ARC \
/opt/spark/jars/arc.jar \

This job executes the following job file which is included in the docker image:

    "type": "SQLValidate",
    "name": "a simple stage which prints a message",
    "environments": [
    "inputURI": ${ETL_CONF_JOB_PATH}"/print_message.sql",
    "sqlParams": {
      "message0": "Hello",
      "message1": "World!"
    "authentication": {},
    "params": {}

This example is included to demonstrate:

  • ETL_CONF_ENV is a reserved environment variable which determines which stages to execute in the current mode. For each of the stages the job designer can specify an array of stages under which that stage will be executed (in the case above production and test are specified).

    The purpose of this stage is so that it is possible to add or remove stages for execution modes like test or integration which are executed by a CI/CD tool prior to deployment and that you do not want to run in production mode - so maybe a comparison against a known ‘good’ test dataset could be executed in only test mode.

  • ETL_CONF_JOB_PATH is an environment variable that is parsed and included by string interpolation when the job file is executed. So when then job starts Arc will attempt to resolve all environment variables set in the basic.json job file. In this case "inputURI": ${ETL_CONF_JOB_PATH}"/print_message.sql", becomes "inputURI": "/opt/tutorial/basic/job/0/print_message.sql", after resolution. This is included so that potentially different paths would be set for running in test vs production mode.

  • In this sample job the spark master is local[*] indicating that this is a single instance ‘cluster’ where Arc relies on vertical not horizonal scaling. Depending on the constrains of the job (i.e. CPU vs disk IO) it is often better to execute with vertical scaling on cloud compute rather than pay the cost of network shuffling.

  • etl.config.uri is a reserved JVM property which describes to Arc which job to execute. See below for all the properties that can be passed to Arc.

Configuration Parameters

Variable Property Description
ETL_CONF_ENABLE_STACKTRACE etl.config.enableStackTrace Whether to enable stacktraces in the event of exception which can be useful for debugging but is not very intuitive for many users. Boolean. Default false.
ETL_CONF_ENV_ID An environment identifier to be added to all logging messages. Could be something like a UUID which allows joining to logs produced by ephemeral compute started by something like Terraform.
ETL_CONF_ENV etl.config.environment The environment to run under.

E.g. if ETL_CONF_ENV is set to production then a stage with "environments": ["production", "test"] would be executed and one with "environments": ["test"] would not be executed.
ETL_CONF_IGNORE_ENVIRONMENTS etl.config.ignoreEnvironments Allows skipping the environments tests and execute all stages/plugins.
ETL_CONF_JOB_ID A job identifier added to all the logging messages.
ETL_CONF_JOB_NAME A job name added to all logging messages and Spark history server.
ETL_CONF_STORAGE_LEVEL etl.config.storageLevel The StorageLevel used when persisting datasets. String. Default MEMORY_AND_DISK_SER.
ETL_CONF_STREAMING etl.config.streaming Run in Structured Streaming mode or not. Boolean. Default false.
ETL_CONF_TAGS etl.config.tags Custom key/value tags separated by space to add to all logging messages.

E.g. ETL_CONF_TAGS=cost_center=123456 owner=jovyan.
ETL_CONF_URI etl.config.uri The URI of the job file to execute.

Additionally there are permissions arguments that can be used to retrieve the job file from cloud storage:

Variable Property Description
ETL_CONF_ADL_OAUTH2_CLIENT_ID The OAuth client identifier for connecting to Azure Data Lake.
ETL_CONF_ADL_OAUTH2_REFRESH_TOKEN etl.config.fs.adl.oauth2.refresh.token The OAuth refresh token for connecting to Azure Data Lake.
ETL_CONF_AZURE_ACCOUNT_KEY The account key for connecting to Azure Blob Storage.
ETL_CONF_AZURE_ACCOUNT_NAME The account name for connecting to Azure Blob Storage.
ETL_CONF_GOOGLE_CLOUD_AUTH_SERVICE_ACCOUNT_JSON_KEYFILE The service account json keyfile path for connecting to Google Cloud Storage.
ETL_CONF_GOOGLE_CLOUD_PROJECT_ID The project identifier for connecting to Google Cloud Storage.
ETL_CONF_S3A_ACCESS_KEY etl.config.fs.s3a.access.key The access key for connecting to Amazon S3.
ETL_CONF_S3A_CONNECTION_SSL_ENABLED etl.config.fs.s3a.connection.ssl.enabled Whether to enable SSL connection to Amazon S3.
ETL_CONF_S3A_ENDPOINT etl.config.fs.s3a.endpoint The endpoint for connecting to Amazon S3.
ETL_CONF_S3A_SECRET_KEY etl.config.fs.s3a.secret.key The secret for connecting to Amazon S3.
ETL_CONF_S3A_ANONYMOUS etl.config.fs.s3a.anonymous Whether to connect to Amazon S3 in anonymous mode. e.g. ETL_CONF_S3A_ANONYMOUS=true.
ETL_CONF_S3A_ENCRYPTION_ALGORITHM etl.config.fs.s3a.encryption.algorithm The bucket encrpytion algorithm: SSE-S3, SSE-KMS, SSE-C.
ETL_CONF_S3A_KMS_ARN The Key Management Service Amazon Resource Name when using SSE-KMS encryptionAlgorithm e.g. arn:aws:kms:us-west-2:111122223333:key/1234abcd-12ab-34cd-56ef-1234567890ab.
ETL_CONF_S3A_CUSTOM_KEY etl.config.fs.s3a.custom.key The key to use when using Customer-Provided Encryption Keys (SSE-C).



This is an example of a streaming job source. This job is intended to be executed after the integration test envornment has been started:

Start integration test environments:

docker-compose -f src/it/resources/docker-compose.yml up --build -d

Start the streaming job:

docker run \
--net "arc-integration" \
-e "ETL_CONF_ENV=test" \
-it -p 4040:4040 triplai/arc:arc_2.9.0_spark_2.4.5_scala_2.12_hadoop_2.9.2_1.1.0 \
bin/spark-submit \
--master local[*] \
--class ai.tripl.arc.ARC \
/opt/spark/jars/arc.jar \

Spark and ulimit

On larger instances with many cores per machine it is possible to exceed the default (1024) max open files (ulimit). This should be verified on your instances if you are receiving too many open files type errors.