Introduction

As a user of the Pipeline tool, I would like to define a single plugin that writes to multiple file sets depending on a value in the message. This is useful for building a generic pipeline that can be used to write to a different number fo sinks on each users start to make more and more generic pipelines that are configured at runtime through macros, we will need the ability to write to to multiple sinks with a single plugin since we cannot modify the DAG once a pipeline is deployed. This plugin is designed to be dynamically configured at runtime to write to multiple filesets for a single run.

Use case(s)

• A user would like to create a single pipeline that can be used to process files from different clients. The pipeline will be completely configured using macros so that each run can be customized depending on the client file being read in. The Each file contains multiple lines that start with identifiers such as 101 and 202, each of these lines should be sent to a different dataset. Each client has a different set of identifiers in the file and would like to write to a different name and number of output file sets. The user would use this plugin, in combination with conditional directives from Wrangler to dynamically configure and write the records to different file setsfilesets.
• This plugin would almost always be configured via macros

User Storie(s)

• User story #1
• User story #2
• User story #3
• User story #mat the pipeline runtime.

User Stories

• As a pipeline developer, I would like the ability to dynamically change the number of filesets being written to at runtime. Determining which record to write to each dataset should be based on the data in each record.
• As a pipeline developer, I would like to define different schemas for each fileset I am writing to, which means this plugin will allow different input schemas.
• As a pipeline developer, I would like to determine which fileset to write each record to based on one or more fields in the record.
• As a pipeline developer, I would like to be able to compress the data written to each of these datasets.
• As a pipeline developer, I would like to be able to partition the filesets using the runtime

Plugin Type

•  Batch Sink 

Configurable

This section defines properties that are configurable for this plugin. 

OutputFileSets:

Example

This example will write to a PartitionedFileSet named 'users'. It will write data in Parquet format compressed using Snappy compressions using the given schema. Every time the pipeline runs, the most recent run will be stored in a new partition in the PartitionedFileSet:

{
  "name": "Multiple File Set Sink", 
  "outputFileSets": [
    {
      "compressionCodec": "Snappy", 
      "datasetName": "fileset0", 
      "datasetTargetPaths": "", 
      "expression": "(id.startsWith(\"201\") || id.startsWith(\"202\") || id.startsWith(\"203\") )&& salary >= 5000 && salary <= 7000", 
      "filesetProperties": "", 
      "schema": {
        "fields": [
          {
            "idname": "id", 
            "type": "string"
          }, 
          {
            "first name": "first name", 
            "type": "string"
          }, 
          {
            "last name": "last name", 
            "type": "string"
          }, 
          {
            "sexname": "sex", 
            "type": "charstring"
          }, 
          {
            "addressname": "address", 
            "type": "string"
          }, 
          {
            "salaryname": "salary", 
            "type": "int"
          }
        ]
      }
    }, 
    {
      "compressionCodec": "Snappy", 
      "datasetName": "fileset1", 
      "datasetTargetPaths": "", 
      "expression": " id.startsWith(\"202\") && salary >= 5000 && salary <= 7000", 
      "filesetProperties": "", 
      "schema": {
        "fields": [
          {
            "idname": "id", 
            "type": "string"
          }, 
          {
            "first name": "first name", 
            "type": "string"
          }, 
          {
            "salaryname": "salary", 
            "type": "int"
          }
        ]
      }
    }, 
    {
      "compressionCodec": "Snappy", 
      "datasetName": "fileset2", 
      "datasetTargetPaths": "", 
      "expression": "(id.startsWith(\"202\") || id.startsWith(\"203\")) && salary >= 5000 && salary < 9000", 
      "filesetProperties": "", 
      "schema": {
        "fields": [
          {
            "idname": "id", 
            "type": "string"
          }, 
          {
            "first name": "first name", 
            "type": "string"
          }, 
          {
            "sexname": "sex", 
            "type": "charstring"
          }, 
          {
            "addressname": "address", 
            "type": "string"
          }, 
          {
            "salaryname": "salary", 
            "type": "int"
          }
        ]
      }
    }
  ], 
  "type": "batchsink"
}

Original data source is as following:

After sink process, it will be split into follow filesets:

fileset0

fileset1

fileset2

Design

• Investigate if the Wrangler core library can be used to evaluate the expression for this plugin
• Can compression codec be changed for each dataset? or should it be set on the job level?
• How can we write to multiple filesets from the same transform function?
  
  

Security

Limitation(s)

Future Work

• Some future work – HYDRATOR-99999
• Another future work – HYDRATOR-99999This can only be used to write to time partitioned filesets

Future Work

• Add the ability to write to all different types of datasets
• Add the ability to write datasets with different partitioning schemes

Test Case(s)

• Test case #1
• Test case #2

Sample Pipeline

Please attach one or more sample pipeline(s) and associated data. 

Pipeline #1

Pipeline #2