在级联中构建自定义连接逻辑,仅确保MAP_SIDE

Question

我有3个级联管道(一个连接其他两个)如下所述,

• LHSPipe - (更大尺寸)

• RHSPipes - (可能适合内存的较小尺寸)

Psuedocode如下,这个例子涉及两个连接

如果 F1DecidingFactor = YES,则 使用RHS Lookup#1 BY加入LHSPipe(LHSPipe.F1Input = RHS Lookup#1.Join #F1)并设置查找结果(SET LHSPipe.F1Output = Result#F1) 否则 SET LHSPipe.F1Output = N /一个

相同的逻辑适用于F2计算.

预期产量,

这种情况迫使我采用自定义连接操作,因为IF-ELSE决定是否加入.

考虑到上面的场景,我想进行MAP-SIDE连接(将RHSPipe保留在MAP任务节点的内存中),我想到了以下可能的解决方案,每个都有其优缺点.需要你对这些建议.

选项1:

CoGroup - 我们可以使用带有BufferJoiner的CoGroup和自定义连接(操作)构建自定义连接逻辑,但这不会确保MAP-SIDE连接.

选项2:

HashJoin - 它确保了MAP-SIDE连接,但据我所知,自定义连接无法使用它构建.

请纠正我的理解并建议您的意见以满足此要求.

提前致谢.

Answer 1

解决这个问题的最好方法（我能想到的）是修改较小的数据集。您可以向较小的数据集添加新字段 ( F1DecidingFactor)。can的值F1Result应该是：

须藤代码

if F1DecidingFactor == "Yes" then
    F1Result = ACTUAL_VALUE
else
    F1Result = "N/A"

结果表

|F1#Join|F1#Result|F1#DecidingFactor|
|    Yes|        0|             True|
|    Yes|        1|            False|
|     No|        0|              N/A|
|     No|        1|              N/A|

您也可以通过级联执行上述操作。

之后，您可以进行地图侧连接。

如果无法修改较小的数据集，那么我有两种选择来解决问题。

选项1

将新字段添加到您的小管道中，这相当于您的决定因素（即F1DecidingFactor_RHS = Yes）。然后将其包含到您的加入标准中。连接完成后，您将仅获得与此条件匹配的那些行的值。否则它将为空/空白。示例代码：

主班

import cascading.operation.Insert;
import cascading.pipe.Each;
import cascading.pipe.HashJoin;
import cascading.pipe.Pipe;
import cascading.pipe.assembly.Discard;
import cascading.pipe.joiner.LeftJoin;
import cascading.tuple.Fields;

public class StackHashJoinTestOption2 {
    public StackHashJoinTestOption2() {
        Fields f1Input = new Fields("F1Input");
        Fields f2Input = new Fields("F2Input");
        Fields f1Join = new Fields("F1Join");
        Fields f2Join = new Fields("F2Join");

        Fields f1DecidingFactor = new Fields("F1DecidingFactor");
        Fields f2DecidingFactor = new Fields("F2DecidingFactor");
        Fields f1DecidingFactorRhs = new Fields("F1DecidingFactor_RHS");
        Fields f2DecidingFactorRhs = new Fields("F2DecidingFactor_RHS");

        Fields lhsJoinerOne = f1DecidingFactor.append(f1Input);
        Fields lhsJoinerTwo = f2DecidingFactor.append(f2Input);

        Fields rhsJoinerOne = f1DecidingFactorRhs.append(f1Join);
        Fields rhsJoinerTwo = f2DecidingFactorRhs.append(f2Join);

        Fields functionFields = new Fields("F1DecidingFactor", "F1Output", "F2DecidingFactor", "F2Output");

        // Large Pipe fields : 
        // F1DecidingFactor F1Input F2DecidingFactor F2Input
        Pipe largePipe = new Pipe("large-pipe");

        // Small Pipe 1 Fields : 
        // F1Join F1Result
        Pipe rhsOne = new Pipe("small-pipe-1");

        // New field to small pipe. Expected Fields:
        // F1Join F1Result F1DecidingFactor_RHS
        rhsOne = new Each(rhsOne, new Insert(f1DecidingFactorRhs, "Yes"), Fields.ALL);

        // Small Pipe 2 Fields : 
        // F2Join F2Result
        Pipe rhsTwo = new Pipe("small-pipe-2");

        // New field to small pipe. Expected Fields:
        // F2Join F2Result F2DecidingFactor_RHS
        rhsTwo = new Each(rhsTwo, new Insert(f1DecidingFactorRhs, "Yes"), Fields.ALL);

        // Joining first small pipe. Expected fields after join: 
        // F1DecidingFactor F1Input F2DecidingFactor F2Input F1Join F1Result F1DecidingFactor_RHS
        Pipe resultsOne = new HashJoin(largePipe, lhsJoinerOne, rhsOne, rhsJoinerOne, new LeftJoin());

        // Joining second small pipe. Expected fields after join: 
        // F1DecidingFactor F1Input F2DecidingFactor F2Input F1Join F1Result F1DecidingFactor_RHS F2Join F2Result F2DecidingFactor_RHS
        Pipe resultsTwo = new HashJoin(resultsOne, lhsJoinerTwo, rhsTwo, rhsJoinerTwo, new LeftJoin());

        Pipe result = new Each(resultsTwo, functionFields, new TestFunction(), Fields.REPLACE);

        result = new Discard(result, f1DecidingFactorRhs);
        result = new Discard(result, f2DecidingFactorRhs);

        // result Pipe should have expected result
    }
}

选项2

如果您想使用默认值而不是 null/blank，那么我建议您首先使用HashJoin默认的 Joiners，然后使用一个函数来使用适当的值更新元组。就像是：

主班

import cascading.pipe.Each;
import cascading.pipe.HashJoin;
import cascading.pipe.Pipe;
import cascading.pipe.joiner.LeftJoin;
import cascading.tuple.Fields;

public class StackHashJoinTest {
    public StackHashJoinTest() {
        Fields f1Input = new Fields("F1Input");
        Fields f2Input = new Fields("F2Input");
        Fields f1Join = new Fields("F1Join");
        Fields f2Join = new Fields("F2Join");

        Fields functionFields = new Fields("F1DecidingFactor", "F1Output", "F2DecidingFactor", "F2Output");

        // Large Pipe fields : 
        // F1DecidingFactor F1Input F2DecidingFactor F2Input
        Pipe largePipe = new Pipe("large-pipe");

        // Small Pipe 1 Fields : 
        // F1Join F1Result
        Pipe rhsOne = new Pipe("small-pipe-1");

        // Small Pipe 2 Fields : 
        // F2Join F2Result
        Pipe rhsTwo = new Pipe("small-pipe-2");

        // Joining first small pipe. 
        // Expected fields after join: 
        // F1DecidingFactor F1Input F2DecidingFactor F2Input F1Join F1Result
        Pipe resultsOne = new HashJoin(largePipe, f1Input, rhsOne, f1Join, new LeftJoin());

        // Joining second small pipe. 
        // Expected fields after join: 
        // F1DecidingFactor F1Input F2DecidingFactor F2Input F1Join F1Result F2Join F2Result
        Pipe resultsTwo = new HashJoin(resultsOne, f2Input, rhsTwo, f2Join, new LeftJoin());

        Pipe result = new Each(resultsTwo, functionFields, new TestFunction(), Fields.REPLACE);

        // result Pipe should have expected result
    }
}

更新功能

import cascading.flow.FlowProcess;
import cascading.operation.BaseOperation;
import cascading.operation.Function;
import cascading.operation.FunctionCall;
import cascading.tuple.Fields;
import cascading.tuple.TupleEntry;

public class TestFunction extends BaseOperation<Void> implements Function<Void> {

    private static final long serialVersionUID = 1L;

    private static final String DECIDING_FACTOR = "No";
    private static final String DEFAULT_VALUE = "N/A";

    // Expected Fields: "F1DecidingFactor", "F1Output", "F2DecidingFactor", "F2Output"
    public TestFunction() {
        super(Fields.ARGS);
    }

    @Override
    public void operate(@SuppressWarnings("rawtypes") FlowProcess process, FunctionCall<Void> call) {
        TupleEntry arguments = call.getArguments();

        TupleEntry result = new TupleEntry(arguments);

        if (result.getString("F1DecidingFactor").equalsIgnoreCase(DECIDING_FACTOR)) {
            result.setString("F1Output", DEFAULT_VALUE);
        }

        if (result.getString("F2DecidingFactor").equalsIgnoreCase(DECIDING_FACTOR)) {
            result.setString("F2Output", DEFAULT_VALUE);
        }

        call.getOutputCollector().add(result);
    }

}

参考

• Insert Function
• Custom Function
• HashJoin

这应该可以解决你的问题。让我知道这是否有帮助。