Introduction

CDAP offers change data capture via three different approaches

• Golden gate for Oracle
• Log miner for Oracle
• Change tracking for SQL server

All these CDC mechanisms are supported via Realtime data pipelines and the plugins are available from Hub. The CDC solution currently runs on Spark 1.x and has experimental support for BigTable. 

Use case(s)

The scope of work involves making the CDC solution work with Spark 2.x and being able to write to BigTable. The performance numbers throughput and latency should be published for these two destinations with all the three CDC approaches.

• ETL developers should be able to set up realtime pipelines to write data to BigTable/BigQuery
• Users should get field level lineage for the source and sink that is being used
• Reference documentation should be updated to account for the changes 
• The solution should run with all versions of Spark 2.x
• Integration tests for specific plugins should be added in the test repos
• Reference document should be updated for the CDC plugins

Deliverables 

Relevant links 

• Existing CDC plugin code: https://github.com/cask-solutions/cdc
• Experimental CDC for Big Table in a branch: https://github.com/cask-solutions/cdc/tree/bigtable-cdc-sink
• Field level lineage: https://docs.cdap.io/cdap/5.1.0-SNAPSHOT/en/developer-manual/metadata/field-lineage.html

Plugin Type

• Batch Source
• Batch Sink 
• Real-time Source
• Real-time Sink
• Action
• Post-Run Action
• Aggregate
• Join
• Spark Model
• Spark Compute

Configurables

CDCBigTable Sink Properties

CDCOracleLogMiner Source Properties

Design

Approach(s)

Iteration 1: Support Spark 2, update documentation, add lineage.

• Add lineage support for real-time plugins. Requires CDAP core modules modification (co.cask.cdap.etl.spark.streaming.DefaultStreamingContext). 
• Make credentials optional for BigTable Sink.
• Add config, schema and third-party application health validation to "configurePipeline" step for all CDC plugins. To validate schema properly, source and sink plugins should use single schema for messages.
• Register datasets of source CDC plugins for lineage view.
• Register fields of all CDC plugins for lineage view.
• Create new core library for Spark 2 plugins. Analogue of "cdap-etl-api-spark".
• Migrate CDC plugins to CDAP core v6.x.x.
• Update documentation to be consistent with the other GCP plugins.
• Write documentation for CDCKudu Sink, CDCDatabase Source, CTSQLServer Source.
• Update UI widgets for all CDC plugins to match properties and schema defined in documentation.
• Try to support both Spark 1 and Spark 2 but drop Spark 1 support if necessary.

Iteration 2: Implement LogMiner Source Plugin.

• Implement LogMiner Source Plugin for Oracle (CDCOracleLogMiner Source). Plugin should use JDBC to retrieve updates from LogMiner.

Iteration 3: Integration tests.

• Implement integration tests for all CDC plugins. Tests should pass using Spark 2. Tests should use local environment if possible. Tests against paid services should run only with special maven profile. All tests should run on local and distributed environments. Provide docker-compose for local environment setup.

Iteration 4: Performance tests.

• Implement performance tests for the three approaches with BigTable.

Message Schema

{
  "type": "record",
  "name": "etlSchemaBody",
  "fields": [
    {
      "name": "op_type",
      "type": [
        "string",
        "null"
      ]
    },
    {
      "name": "table",
      "type": "string"
    },
    {
      "name": "primary_keys",
      "type": [
        {
          "type": "array",
          "items": "string"
        },
        "null"
      ]
    },
    {
      "name": "schema",
      "type": [
        "string",
        "null"
      ]
    },
    {
      "name": "change",
      "type": [
        {
          "type": "map",
          "keys": "string",
          "values": [
            "null",
            "boolean",
            "int",
            "long",
            "float",
            "double",
            "bytes",
            "string"
          ]
        },
        "null"
      ]
    }
  ]
}

Limitation(s)

• CDC plugins may not be compatible with Spark 2.

Future Work

• Error rendering macro 'jira' : Unable to locate Jira server for this macro. It may be due to Application Link configuration.
• Oracle LogMiner Source Plugin
• Integration tests for CDC plugins
• Performance tests for CDC plugins

Test Case(s)

• Track changes from Golden gate for Oracle (create table, insert row, update row, delete row).
• Track changes from Log miner for Oracle (create table, insert row, update row, delete row).
• Track changes from SQL Server (create table, insert row, update row, delete row).
• Write changes to Apache Kudu (create table, insert row, update row, delete row).
• Write changes to HBase (create table, insert row, update row, delete row).
• Write changes to BigTable (create table, insert row, update row, delete row).

Sample Pipeline

Pipeline #1 (SQL Server → BigTable)

TODO