ASCII hydra

showcase

The ASCII Hydra project demonstrates a cost-efficient alternative to being locked into specific cloud platforms like Databricks. This examples provide out of the box way of creating assets using either local pyspark, Databricks or AWS EMR.

preprequisites

• pixi curl -fsSL https://pixi.sh/install.sh | bash
• credentials for:
  • AWS EMR
    • ASCII_AWS_ACCESS_KEY_ID: Your AWS Access Key ID for EMR.
    • ASCII_AWS_SECRET_ACCESS_KEY: Your AWS Secret Access Key for EMR.
  • Databricks
    • DATABRICKS_HOST: The Databricks host URL.
    • DATABRICKS_CLIENT_ID: Your Databricks Client ID.
    • DATABRICKS_CLIENT_SECRET: Your Databricks Client Secret.
• set up a couple of environment variables:
  • SPARK_PIPES_ENGINE: Specifies the engine used for Spark Pipes (valid options: databricks, emr, or pyspark).
  • SPARK_EXECUTION_MODE: Defines the mode of execution for data (valid options: small_dev_sample_local, small_dev_sample_s3, or full).
  • DAGSTER_HOME: Path to the Dagster home directory where Dagster-related configurations and metadata are stored.

Explanation of SPARK_EXECUTION_MODE

The SPARK_EXECUTION_MODE environment variable controls the scope and source of the data used during the pipeline execution. It will be transformed on the class ExecutionMode and it's thought to be use as a flag at the external script level:

`small_dev_sample_local`:
This mode should use a small, locally stored sample dataset, ideal for fast development and testing on your local machine.

`small_dev_sample_s3`:
This mode should use a small sample dataset stored on Amazon S3, allowing you to test the pipeline in a cloud environment with minimal data.

`full`:
This mode should processes the full dataset stored on Amazon S3, intended for complete runs and production-level processing.

Creation of environment

To create the environment execute the following commands:

pixi run start

## testing
# check formatting
pixi run -e ci fmt
# check typing
pixi run -e ci lint
# run tests
pixi run -e testing test

alterantively use the makefile via:

make start
make test
make fmt
make lint

The package can be installed using pixi via:

pixi install

automatically an isolated python environment is created.

Execute make start and then go to http://localhost:3000 for the dagster UI

explanation

See https://georgheiler.com/2024/06/21/cost-efficient-alternative-to-databricks-lock-in/.