tree: fde9eb7e77f34df872f5b11657daa1c97865bb37
  1. gen_data/
  2. run_tpcds/
  3. README.md
tools/workload/tpcds-delta/README.md

Test on Velox backend with Delta Lake and TPC-DS workload

This workload example is verified with JDK 8, Spark 3.4.4 and Delta 2.4.0.

Test dataset

Use bash script tpcds-datagen-delta.sh to generate the data. The script relies on a already-built gluten-it executable. To build it, following the steps:

cd ${GLUTEN_HOME}/tools/gluten-it/
mvn clean install -P spark-3.4,delta

Then call the data generator script:

cd ${GLUTEN_HOME}/tools/workload/tpcds-delta/gen_data
./tpcds-datagen-delta.sh

Meanings of the commands that are used in the script are explained as follows:

  • --benchmark-type=ds: “ds” for TPC-DS, “h” for TPC-H.
  • --threads=112: The parallelism. Ideal to set to the core number.
  • -s=100: The scale factor.
  • --data-dir=/tmp/my-data: The target table folder. If it doesn't exist, it will then be created.

When the command is finished, check the data folder:

ls -l /tmp/my-data/

You should see a generated table folder in it:

drwxr-xr-x 20 root root 4096 Sep  1 15:13 tpcds-generated-100.0-delta-partitioned

The folder tpcds-generated-100.0-delta-partitioned is the generated Delta TPC-DS table. As shown by the folder name, it's partitioned, and with scale factor 100.0.

Test Queries

We provide the test queries in TPC-DS Queries. We provide a Scala script in Run TPC-DS directory about how to run TPC-DS queries on the generated Delta tables.