Spark Alchemy is a collection of open-source Spark tools & frameworks that have made the data engineering and data science teams at Swoop highly productive in our demanding petabyte-scale environment with rich data (thousands of columns).
While spark-alchemy, like Spark itself, is written in Scala, much of its functionality, such as interoperable HyperLogLog functions, can be used from other Spark-supported languages such as SparkSQL and Python.
Add the following to your libraryDependencies in SBT:
libraryDependencies += "com.swoop" %% "spark-alchemy" % "1.0.1"You can find all released versions here.
Some use cases such as interoperability with PySpark may require the assembly of a fat JAR of spark-alchemy. To assemble, run sbt assembly. To skip tests during assembly, run sbt 'set sbt.Keys.test in assembly := {}' assembly instead.
- Native HyperLogLog functions that offer reaggregatable fast approximate distinct counting capabilities far beyond those in OSS Spark with interoperability to Postgres and even JavaScript. Just as Spark's own native functions, once the functions are registered with Spark, they can be used from SparkSQL, Python, etc.
-
Helpers for native function registration
-
Look at
SparkSessionSpecas an example of how to reuse advanced Spark testing functionality from OSS Spark.
- See HyperLogLog functions for an example of how
spark-alchemyHLL functions can be registered for use through PySpark.
-
Configuration Addressable Production (CAP), Automatic Lifecycle Management (ALM) and Just-in-time Dependency Resolution (JDR) as outlined in our Spark+AI Summit talk Unafraid of Change: Optimizing ETL, ML, and AI in Fast-Paced Environments.
-
Utilities that make Delta Lake development substantially more productive.
-
Hundreds of productivity-enhancing extensions to the core user-level data types:
Column,Dataset,SparkSession, etc. -
Data discovery and cleansing tools we use to ingest and clean up large amounts of dirty data from third parties.
-
Cross-cluster named lock manager, which simplifies data production by removing the need for workflow servers much of the time.
-
case classcode generation from Spark schema, with easy implementation customization. -
Tools for deploying Spark ML pipelines to production.
- spark-records: bulletproof Spark jobs with fast root cause analysis in the case of failures
Contributions and feedback of any kind are welcome. Please, create an issue and/or pull request.
Spark Alchemy is maintained by the team at Swoop. If you'd like to contribute to our open-source efforts, by joining our team or from your company, let us know at spark-interest at swoop dot com.
spark-alchemy is Copyright © 2018-2020 Swoop, Inc. It is free software, and may be redistributed under the terms of the LICENSE.