User stories documented (Albert/Alvin)
User stories reviewed (Nitin)
Design documented (Albert/Alvin)
Design reviewed (Andreas/Terence)
Feature merged (Albert/Alvin)
Examples and guides (Albert/Alvin)
Integration tests (Albert/Alvin)
Documentation for feature (Albert/Alvin)
Blog post

The developer to load webpage click and view data (customer id, timestamp, action, url) into a partitioned fileset. After loading the data, the developer wants to de-duplicate records and calculate how many times each customer clicked and viewed over the past hour, past day, and past month.

User Stories:

(3.4) A developer should be able to create pipelines that contain aggregations (GROUP BY -> count/sum/unique)
(3.5) A developer should be able to create a pipeline with multiple sources, with one happening after the otherA control some parts of the pipeline running before others. For example, one source -> sink branch running before another source -> sink branch.
(3.4) A developer should be able to use a Spark ML job as a pipeline stage
A (3.4) A developer should be able to rerun failed pipeline runs without reconfiguring the pipeline
A (3.4) A developer should be able to de-duplicate records in a pipeline
A (3.5) A developer should be able to join multiple branches of a pipeline
A (3.5) A developer should be able to use an Explore action as a pipeline stage
A (3.5) A developer should be able to create pipelines that contain Spark Streaming jobs
A (3.5) A developer should be able to create pipelines that run based on various conditions, including input data availability and Kafka events

Design:

Story 1: Group By -> Aggregations

Option 1:

Introduce a new plugin type "aggregation"aggregator". In general, to support more and more plugin types in a generic way, we want to refactor the config:

Code Block

Option 1:
{
  "stages": [
    {
      "name": "inputTable",
      "plugin": {
        "name": "Table",
        "type": "batchsource",  // new field
        "properties": {
        }
      }
    },
    {
      "name": "aggStage",
      "plugin": {
        "name": "RecordAggregatorGroupByAggregate",
        "type": "aggregationaggregator",  // new plugin type
        "properties": {
          "groupBy": "id",
          "functions": "{[
            "totalPrice": {
              "namecolumnName": "sumtotalPrice",
              "propertiesplugin": {
                "columnname": "pricesum",
              }  "properties": {
         },         "column": "price"
  "numTransactions":  {            }
  "name": "count"           }
 }           }",
         }   {
   }     }   ],   "connectionscolumnName": ["numTransactions",
    { "from": "inputTable", "to": "aggStage" }    ] }   Option 2"plugin": {
   "sources": [     {       "name": "inputTablecount",
      "plugin": {       }
 "name": "Table",         "type": "batchsource", }
 // new field       ]"
 "properties": {         }
      }
    }
  ],
  "aggregationsconnections": [
    {
      "namefrom": "aggStageinputTable",       "groupBy"to": "idaggStage", } 
     "aggregations": [
        {
          "columnName": "totalPrice",
          "plugin": {
 ]
}

Some problems with this is that the plugin property "functions" is itself a json describing plugins to use. This is not easy for somebody to configure, but maybe it could be simplified by a UI widget type.

Java APIs for plugin developers. It is basically mapreduce, 'Aggregation' is probably a bad name for this. Need to see if this fits into Spark. Would we have to remove the emitters?

Code Block

public abstract class Aggregation<GROUP_BY, RECORD_TYPE, OUTPUT_TYPE> {
 
  public abstract void groupBy(RECORD_TYPE input, Emitter<GROUP_BY> emitter);
 
  public abstract void aggregate(GROUP_BY groupKey, Iterable<RECORD_TYPE> groupRecords, Emitter<OUTPUT_TYPE> emitter);
 
}

@Plugin(type = "aggregation")
@Name("GroupByAggregate")
public GroupByAggregate extends Aggregation<StructuredRecord, StructuredRecord, StructuredRecord> {
  private static final AggConfig config;
  
  public static class AggConfig extends PluginConfig {
    private String groupBy;
 
    // ideally this would be Map<String, FunctionConfig> functions
    private String functions;
  }
 
  public void configurePipeline(PipelineConfigurer configurer) {
    Map<String, FunctionConfig> functions = gson.fromJson(config.functions, MAP_TYPE);
    for each function:
      usePlugin(id, type, name, properties);
  }
 
  public groupBy(StructuredRecord input, Emitter<StructuredRecord> emitter) {
    // key = new record from input with only fields in config.groupBy
    Set<String> fields = config.groupBy.split(",");
    emitter.emit(recordSubset(input, fields));
  }
 
  public void initialize() {
    Map<String, FunctionConfig> functions = gson.fromJson(config.functions, MAP_TYPE);
    for each function:
      val = function.aggregate(groupRecords);
  }
 
  public void aggregate(StructuredRecord groupKey, Iterable<StructuredRecord> groupRecords, Emitter<StructuredRecord> emitter) {
    // reset all functions
    for (StructuredRecord record : groupRecords) {
      foreach function:
        function.update(record);
    }
    // build record from group key and function values
    for each function:
      val = function.aggregate();
    // emit record
  }
 
}
 
public abstract class AggregationFunction<RECORD_TYPE, OUTPUT_TYPE> {
  public abstract void reset();
  public abstract void update(RECORD_TYPE record);
  public abstract OUTPUT_TYPE aggregate();
}
 
@Plugin(type = "aggregationFunction")
@Name("sum")
public SumAggregation extends AggregationFunction<StructuredRecord, Number> {
  private final SumConfig config;
  private Number sum;
  
  public static class SumConfig extends PluginConfig {
    private String column;
  }
 
  public void update(StructuredRecord record) {
    // get type of config.column, initialize sum to right type based on that
    sum += (casted to correct thing) record.get(config.column);
  }
 
  public Number aggregate() {
    return sum;
  }
}

Note: This would also satisfy user story 5, where a unique can be implemented as a Aggregation plugin, where you group by the fields you want to unique, and ignore the Iterable<> in aggregate and just emit the group key.

Story 2: Control Flow (Not Reviewed, WIP)

Option 1: Introduce different types of connections. One for data flow, one for control flow

Code Block

{
  "stages": [
    {
      "name": "customersTable",
      "plugin": {
        "name": "Database",
        "type": "batchsource", ...
      }
    },    
    {
      "name": "customersFiles",
      "plugin": {
        "name": "TPFSParquet",
        "type": "batchsink", ...
      }
    },
    {
      "name": "purchasesTable",
      "plugin": {
        "name": "Database",
        "type": "batchsource"
      }
    },
    {
      "name": "purchasesFiles",
      "plugin": {
        "name": "TPFSParquet",
        "type": "batchsink", ...
      }
    },
  ],
  "connections": [
    { "from": "customersTable", "to": "customersFiles", "type": "data" },
    { "from": "customersFiles", "to": "purchasesTable", "type": "control" },
    { "from": "purchasesTable", "to": "purchasesFiles", "type": "data" }
  ]
}

An alternative could be to introduce the concept of "phases", with connections between phases always control flow, and connections within phases as data flow

Code Block

{
  "phases": [
    {
      "name": "phase1",
      "stages": [
        {
          "name": "sumcustomersTable",
            "propertiesplugin": {
 
            "columnname": "priceDatabase",
            "type":  }"batchsource", ...
          }
        },   
        {
          "columnNamename": "numTransactionscustomersFiles",
          "plugin": {
            "name": "count"
          }: {
         }     "name": "TPFSParquet",
 ]     }   ],   "connectionstype": [
"batchsink", ...
   { "from": "inputTable", "to": "aggStage" }
      ],
  }   public abstract class Aggregation<INPUT_TYPE, GROUP_BY, RECORD_TYPE, OUTPUT_TYPE> {
 "connections": [
       public abstract groupBy(INPUT_TYPE input, Emitter<KeyValue<GROUP_BY, RECORD_TYPE>> emitter);
 
  public abstract aggregate(GROUP_BY groupKey, Iterable<RECORD_TYPE> groupRecords, Emitter<OUTPUT_TYPE> emitter);
 
}

@Plugin(type = "aggregation")
@Name("record")
public RecordAggregation extends Aggregation<StructuredRecord, StructuredRecord, StructuredRecord, StructuredRecord> {
  private static final AggConfig config;{ "from": "customersTable", "to": "customersFiles" }
      ]
    },
    {
      "name": "phase2",
      "stages": [
      public static class{
AggConfig extends PluginConfig {     private String groupBy;
  "name": "purchasesTable",
    // ideally this would be Map<String, FunctionConfig> functions "plugin": {
    private String functions;   }     public void configurePipeline(PipelineConfigurer configurer) {"name": "Database",
        Map<String, FunctionConfig> functions = gson.fromJson(config.functions, MAP_TYPE); "type": "batchsource"
    for each function:    }
  usePlugin(id, type, name, properties);   },
     public groupBy(StructuredRecord input, Emitter<KeyValue<StructuredRecord,{
StructuredRecord>> emitter) {     // key = new record from input with only fields in config.groupBy"name": "purchasesFiles",
           // emitter.emit(new KeyValue<>(key, input));"plugin": {
      }     public aggregate(StructuredRecord groupKey, Iterable<StructuredRecord> groupRecords, Emitter<StructuredRecord> emitter) { "name": "TPFSParquet",
          Map<String, FunctionConfig> functions = gson.fromJson(config.functions, MAP_TYPE);"type": "batchsink", ...
     for each function:     }
 val = function.aggregate(groupRecords);     for}
(StructuredRecord record : groupRecords) {  ],
    function.update(record);  "connections": [
 }     // build record{ "from group key and function values": "purchasesTable", "to": "purchasesFiles" }
      ]
for each function:  }
  ]
 val = function.aggregate();"connections": [
    // emit record
 { "from": "phase1", "to": "phase2" }
  ]
}
 
public abstract class AggregationFunction<RECORD_TYPE, OUTPUT_TYPE> {
 
  public abstract void update(RECORD_TYPE record);
  public abstract OUTPUT_TYPE aggregate();
}
 
@Plugin(type = "aggregationFunction")
@Name("sum")
public SumAggregation extends AggregationFunction<StructuredRecord, Number> {
  private final SumConfig config;
  private Number sum;

Option2: Make it so that connections into certain plugin types imply control flow rather than data flow. For example, introduce "condition" plugin type. Connections into a condition imply control flow rather than data flow. Similarly, connections into an "action" plugin type would imply control flow

Code Block

{
  "stages": [
    {
      "name": "customersTable",
      "plugin": {
        "name": "Database",
      public static class SumConfig extends PluginConfig {"type": "batchsource", ...
      private}
String column;   },     
public void update(StructuredRecord record) {
    // get type of config.column, initialize sum to right type based on that  "name": "customersFiles",
      "plugin": {
      sum += record;
 "name": "TPFSParquet",
 }     public Number aggregate() {
  "type": "batchsink", ...
  return sum;   }
}

Story 2: Multiple Sources

Option 1: Introduce the concept of "phases". Each phase has its own dag. Phases can be connected, with connections denoting control flow rather than data flow.

Code Block

{
  "phases": [   },
    {
      "name": "phase1afterDump",
      "stagesplugin": [{
        {
 "name": "AlwaysRun",
        "nametype": "customersTablecondition",
      }
   "plugin": { },
     {
      "name": "DatabasepurchasesTable",
            "type"plugin": "batchsource", ...{
        "name": "Database",
 }       "type": "batchsource"
},      }
    },
{    {
      "name": "customersFilespurchasesFiles",

         "plugin": {
   
        "name": "TPFSParquet",
  
         "type": "batchsink", ...
      }
 }    },
  ],

     "connections": [
   
    { "from": "customersTable", "to": "customersFiles" }
      ]
    },
    {
      "namefrom": "phase2customersFiles",
      "stagesto": [
   : "afterDump" },
    { "from": "afterDump", "to": "purchasesTable" },
    { "namefrom": "purchasesTable", "to": "purchasesFiles" }
       "plugin": {
            "name": "Database",
            "type": "batchsource"
          }
        },
        {
 ]
}

You could also say that connections into a source imply control flow, or connections into an action imply control flow.

Story 3: Spark ML in a pipeline

Add a plugin type "sparksink" that is treated like a sink. When present, a spark program will be used to read data, transform it, then send all transformed results to the sparksink plugin.

Code Block

{
  "stages": [
    {
        "name": "purchasesFilescustomersTable",
   
      "plugin": {
   
        "name": "TPFSParquetDatabase",
   
        "type": "batchsinkbatchsource", ...
      }
    },    
    }{
      ]"name": "categorizer",
      "connectionsplugin": [{
        { "fromname": "purchasesTableSVM",
        "totype": "purchasesFilessparksink", }...
      ]}
    },
  ]  {
"connections": [     { "from": "phase1", "toname": "phase2models",
}     ] }

Option2: Introduce "condition" plugin type. Connections into a condition imply control flow rather than data flow.

Code Block

{
  "stages": [
    {"plugin": {
        "name": "Table",
        "nametype": "customersTablebatchsink", ...
      }
"plugin": {   },
  ],
  "nameconnections": "Database",[
    {    "typefrom": "batchsourcecustomersTable", ...
 "to": "categorizer" }
  ]
 }}

Story 6: Join (Not Reviewed, WIP)

Add a join plugin type. Different implementations could be inner join, left outer join, etc.

Code Block

{
    },    "stages": [
    {
      "name": "customersFilescustomers",
      "plugin": {
        "name": "TPFSParquetTable",
        "type": "batchsinkbatchsource", ...
      }
    },
    {
      "name": "afterDumppurchases",
      "plugin": {
        "name": "AlwaysRunTable",
        "type": "conditionbatchsource", ...
      }
    },
    {
      "name": "purchasesTablecustomerPurchaseJoin",
      "plugin": {
        "name": "Databaseinner",
        "type": "batchsourcejoin",
      }  "properties": {
 },     {       "nameleft": "purchasesFiles",
      "plugin": {customers.id",
          "nameright": "TPFSParquetpurchases.id",
          "typerename": "batchsink", ...customers.name:customername,purchases.name:itemname"
        }
      },
  ],  },
"connections": [   ...
 { "from": "customersTable", "to": "customersFiles" }, ],
  "connections": [
    { "from": "customersFilescustomers", "to": "afterDumpcustomerPurchaseJoin" },
    { "from": "afterDumppurchases", "to": "purchasesTablecustomerPurchaseJoin" },
    { "from": "purchasesTablecustomerPurchaseJoin", "to": "purchasesFilessink" },
  ]
}

You could also say that certain plugin types imply control flow (runcondition being one of them), whereas other plugin types imply data flow.

Java API for join plugin type: these might just be built into the app. Otherwise the interface is basically MapReduce.

Versions Compared

Old Version 32

New Version Current

Key

User Stories:

Design:

Story 1: Group By -> Aggregations

Story 2: Control Flow (Not Reviewed, WIP)

Story 2: Multiple Sources

Story 3: Spark ML in a pipeline

Story 6: Join (Not Reviewed, WIP)

Page Comparison

Versions Compared

Old Version 32

New Version Current

Key

User Stories:

Design:

Story 1: Group By -> Aggregations

Story 2: Control Flow (Not Reviewed, WIP)

Story 2: Multiple Sources

Story 3: Spark ML in a pipeline

Story 6: Join (Not Reviewed, WIP)