Examples - NVIDIA Docs

Please visit spark-rapids-examples repo for ETL, ML/DL, UDF related examples using the RAPIDS Accelerator for Apache Spark. It includes Scala/Python source code and related notebooks for different examples.

Benchmarks

Please visit spark-rapids-benchmarks repo for Spark related benchmark sets and utilities using the RAPIDS Accelerator for Apache Spark.

Qualification Tool

CLI Samples

This section shows samples of Qualification CLI cmd assuming the following inputs:

CLUTER_NAME: The cluster name on CSP (dataproc, Databricks, or EMR)
PROP_FILE: Path to a cluster property file. The path can be a local filesystem, HDFS, S3, ABFS, or GCS. The file can be formatted according to gcloud specs (DATAPROC_PROP), EMR (EMR_PROP)
EVENTLOG: Path to Spark event logs without the scheme part. The scheme can be a local filesystem, HDFS, S3, ABFS, or GCS.

The following table shows samples of CLI cmds along with the expected functionalities and the platform based on which the analysis is performed.

*List of arguments and options for `qualification` CLI cmd*
CMD	Platform	Cost Savings	Speedups	Comments
Copy Copied! `spark_rapids qualification \ --cluster $CLUSTER_NAME \ --eventlogs gs://$EVENTLOG`	Dataproc	☑️	☑️	cost savings are calculated based on Dataproc cluster because EVENTLOG is stored on GCS
Copy Copied! `spark_rapids qualification \ --cluster $DATAPROC_PROP \ --eventlogs file://$EVENTLOG`	Dataproc	☑️	☑️	The `cluster` arguments to a property file matching the Dataproc specs
Copy Copied! `spark_rapids qualification \ --eventlogs file://$EVENTLOG`	On-prem		☑️	The cost-savings cannot be generated without cluster argument while EVENTLOG is stored on a local filesystem

Java CMD Samples

Process the 10 newest logs, and only output the top 3 in the output:

Copy
Copied!

            
            java ${QUALIFICATION_HEAP} \
  -cp ~/rapids-4-spark-tools_2.12-<version>.jar:$SPARK_HOME/jars/*:$HADOOP_CONF_DIR/ \
  com.nvidia.spark.rapids.tool.qualification.QualificationMain -f 10-newest -n 3 /eventlogDir

Process last 100 days’ logs:

Copy
Copied!

            
            java ${QUALIFICATION_HEAP} \
  -cp ~/rapids-4-spark-tools_2.12-<version>.jar:$SPARK_HOME/jars/*:$HADOOP_CONF_DIR/ \
  com.nvidia.spark.rapids.tool.qualification.QualificationMain -s 100d /eventlogDir

Process only the newest log with the same application name:

Copy
Copied!

            
            java ${QUALIFICATION_HEAP} \
  -cp ~/rapids-4-spark-tools_2.12-<version>.jar:$SPARK_HOME/jars/*:$HADOOP_CONF_DIR/ \
  com.nvidia.spark.rapids.tool.qualification.QualificationMain -f 1-newest-per-app-name /eventlogDir

Parse ML functions from the eventlog:

Copy
Copied!

            
            java ${QUALIFICATION_HEAP} \
  -cp ~/rapids-4-spark-tools_2.12-<version>.jar:$SPARK_HOME/jars/*:$HADOOP_CONF_DIR/ \
  com.nvidia.spark.rapids.tool.qualification.QualificationMain --ml-functions /eventlogDir

Profiling Tool

CLI Samples

This section shows samples of Profiling CLI cmd assuming the following inputs:

CLUTER_NAME: The GPU cluster name on CSP (dataproc, Databricks, or EMR)
PROP_FILE: Path to a GPU cluster property file. The path can be a local filesystem, HDFS, S3, ABFS, or GCS. The file can be formatted according to gcloud specs (DATAPROC_PROP), EMR (EMR_PROP)
EVENTLOG: Path to Spark event logs without the scheme part. The scheme can be a local filesystem, HDFS, S3, ABFS, or GCS.

The following table shows samples of CLI cmds along with the expected functionalities and the platform based on which the analysis is performed.

*List of arguments and options for `profiling` CLI cmd*
CMD	Platform	Auto-Tuner	Comments
Copy Copied! `spark_rapids profiling \ --cluster $CLUSTER_NAME \ --eventlogs gs://$EVENTLOG`	Dataproc	☑️	Auto-Tuner recommendations are based on accelerated Dataproc cluster because EVENTLOG is stored on GCS
Copy Copied! `spark_rapids profiling \ --cluster $DATAPROC_PROP \ --eventlogs file://$EVENTLOG`	Dataproc	☑️	Auto-Tuner recommendations are based on accelerated Dataproc cluster because `cluster` arguments to a property file matching the Dataproc specs
Copy Copied! `spark_rapids profiling \ --eventlogs file://$EVENTLOG`	On-prem		The recommendations cannot be generated without cluster argument while EVENTLOG is stored on a local filesystem

Java CMD Samples

Collection Modes

Example running Profiling tool with different collections modes:

Collection Mode
Combined Mode
Compare Mode

Copy
Copied!

            
            java -cp rapids-4-spark-tools_2.12-<version>.jar:$SPARK_HOME/jars/* \
     com.nvidia.spark.rapids.tool.profiling.ProfileMain [options] \
     <eventlogs | eventlog directories ...>

Copy
Copied!

            
            java -cp rapids-4-spark-tools_2.12-<version>.jar:$SPARK_HOME/jars/* \
     com.nvidia.spark.rapids.tool.profiling.ProfileMain --combined \
     <eventlogs | eventlog directories ...>

Copy
Copied!

            
            java -cp rapids-4-spark-tools_2.12-<version>.jar:$SPARK_HOME/jars/* \
     com.nvidia.spark.rapids.tool.profiling.ProfileMain --compare \
     <eventlogs | eventlog directories ...>