M3D Engine

M3D stands for Metadata Driven Development and is a cloud and platform agnostic framework for the automated creation, management and governance of metadata and data flows from multiple source to multiple target systems. The main features and design goals of M3D are:

• Cloud and platform agnostic
• Enforcement global data model including speaking names and business objects
• Governance by conventions instead of maintaining state and logic
• Lightweight and easy to use
• Flexible development of new features
• Stateless execution with minimal external dependencies
• Enable self-service
• Possibility to extend to multiple destination systems (currently AWS EMR)

M3D consists of two components. m3d-engine, which we are providing in this repo, and m3d-api which contains the api as python module.

The architecture of M3D is described in detail here.

Use cases

M3D can be used for:

• Creation of data lake environments
• Management and governance of metadata
• Data flows from multiple sources
• Data flows to multiple target systems
• Algorithms as data frame transformations

adidas is not responsible for the usage of this software for different purposes that the ones described in the use cases.

M3D Engine

M3D Engine is a framework written in Scala for distributed execution of ingestion and transformation workloads to and within data lake.

Algorithms

In M3D terminology an algorithm can be for example:

• a data transformation from a source on the data lake to a target on the data lake
• a data load from raw files on the landing layer to the parquet files on the lake layer
• decompression of compressed data
• materialization of partitioned data

M3D Engine Features

M3D Engine supports:

• Loading structured and semi-structured data in Full mode
• Loading structured and semi-structured data in Append mode
• Loading structured and semi-structured data in Delta mode (DeltaLoad - in memory, by comparing new data and target table partitions; DeltaLakeLoad - using Delta Lake IO capabilities)
• Decompression of compressed data
• Extraction from parquet file format
• Extraction from delimiter separated files (CSV,TSV,etc.)
• Extraction from fixed length string data
• Partitioned materialization of different types (full, range, query)
• Usable from jupyter notebooks (using the JavaConsumable trait)
• Extensible with new algorithms

Usage

To execute an algorithm implemented in m3d-engine, it is required to have a Spark cluster running that can access a parameters file and the compiled m3d-engine jar artifact. To execute an Algorithm use can call spark-submit with:

 spark-submit --master yarn \
 --deploy-mode cluster --class com.adidas.analytics.AlgorithmFactory \
 s3://application_bucket/m3d/test/m3d/m3d-api/m3d-engine-assembly.jar \
 FullLoad s3://application_bucket/m3d/test/apps/m3d-engine/fullload/bdp-emr_prod-test.fullload.20190815T134744.json

Input parameters for m3d-engine-assembly.jar

• appClassName class name of the algorithm to be executed
• appParamFile location of the parameters file

Specification of the parameter file

The parameter file is a json file containing algorithm specific configuration.

The parameter file for the full load algorithm for example has the following content:

{
  "current_dir": "s3://lake_bucket/test/source_system/table_name/data/", 
  "backup_dir": "s3://lake_bucket/test/source_system/table_name/data_backup/", 
  "delimiter": "|", 
  "file_format": "dsv", 
  "has_header": false, 
  "partition_column": "date_column_name", 
  "partition_column_format": "yyyyMMdd", 
  "target_partitions": [
      "year", 
      "month"
  ], 
  "source_dir": "s3://landing_bucket/test/source_system/table_name/data/", 
  "target_table": "test_lake.table_name"
}

• current_dir location of the currently stored data and where it should be written by the algorithm
• backup_dir backup location of the data before the existing data is overwritten
• source_dir location of the source data to be ingested
• file_format format of the source data, e.g. dsv or parquet
• delimiter delimiter used in the case of dsv format
• has_header flag defining whether the input files have a header
• partition_column column that contains the partitioning information
• partition_column_format format of the partitioning column in the case of time/date columns
• target_partitions partitioning columns in the target
• target_table target table where the data will be available for querying after loading

License and Software Information

© adidas AG

adidas AG publishes this software and accompanied documentation (if any) subject to the terms of the Apache 2.0 license with the aim of helping the community with our tools and libraries which we think can be also useful for other people. You will find a copy of the Apache 2.0 license in the root folder of this package. All rights not explicitly granted to you under the Apache 2.0 license remain the sole and exclusive property of adidas AG.

NOTICE: The software has been designed solely for the purpose of automated creation, management and governance of metadata and data flows. The software is NOT designed, tested or verified for productive use whatsoever, nor or for any use related to high risk environments, such as health care, highly or fully autonomous driving, power plants, or other critical infrastructures or services.

If you want to contact adidas regarding the software, you can mail us at software.engineering@adidas.com.

For further information open the adidas terms and conditions page.

License

Apache 2.0