Mapreduce atop Apache Phoenix (ScanPlan 初探)

Posted 2020-10-30 BigData and Machine Learning

利用Mapreduce/hive查询Phoenix数据时如何划分partition？
PhoenixInputFormat的源码一看便知：

    public List<InputSplit> getSplits(JobContext context) throws IOException, InterruptedException {
        Configuration configuration = context.getConfiguration();
        QueryPlan queryPlan = this.getQueryPlan(context, configuration);
        List allSplits = queryPlan.getSplits();
        List splits = this.generateSplits(queryPlan, allSplits);
        return splits;
    }

根据select查询语句创建查询计划，QueryPlan，实际是子类ScanPlan。getQueryPlan函数有一个特殊操作：
queryPlan.iterator(MapReduceParallelScanGrouper.getInstance());
如果HBase表有多个Region，则会将一个Scan划分为多个，每个Region对应一个Split。这个逻辑跟MR on HBase类似。只是这边的实现过程不同，这边调用的是Phoenix的QueryPlan，而不是HBase API。

以下是一个示例，加深这一过程的理解。

Phoenix 建表

将表presplit为4个region：[-∞,CS), [CS, EU), [EU, NA), [NA, +∞)

CREATE TABLE TEST (HOST VARCHAR NOT NULL PRIMARY KEY, DESCRIPTION VARCHAR) SPLIT ON ('CS','EU','NA');
upsert into test(host, description) values ('CS11', 'cccccccc');
upsert into test(host, description) values ('EU11', 'eeeddddddddd')
upsert into test(host, description) values ('NA11', 'nnnnneeeddddddddd');

0: jdbc:phoenix:localhost> select * from test;
+-------+--------------------+
| HOST  |    DESCRIPTION     |
+-------+--------------------+
| CS11  | cccccccc           |
| EU11  | eeeddddddddd       |
| NA11  | nnnnneeeddddddddd  |
+-------+--------------------+

窥探ScanPlan

import org.apache.hadoop.hbase.client.Scan;
import org.apache.log4j.BasicConfigurator;
import org.apache.phoenix.compile.QueryPlan;
import org.apache.phoenix.iterate.MapReduceParallelScanGrouper;
import org.apache.phoenix.jdbc.PhoenixStatement;

import java.io.IOException;
import java.sql.*;
import java.util.List;


public class LocalPhoenix {
    public static void main(String[] args) throws SQLException, IOException {
        BasicConfigurator.configure();

        Statement stmt = null;
        ResultSet rs = null;

        Connection con = DriverManager.getConnection("jdbc:phoenix:localhost:2181:/hbase");
        stmt = con.createStatement();
        PhoenixStatement pstmt = (PhoenixStatement)stmt;
        QueryPlan queryPlan = pstmt.optimizeQuery("select * from TEST");
        queryPlan.iterator(MapReduceParallelScanGrouper.getInstance());

        Scan scan = queryPlan.getContext().getScan();
        List<List<Scan>> scans = queryPlan.getScans();

        for (List<Scan> sl : scans) {
            System.out.println();
            for (Scan s : sl) {
                System.out.print(s);
            }
        }

        con.close();

    }
}

4个scan如下：

{"loadColumnFamiliesOnDemand":null,"startRow":"","stopRow":"CS","batch":-1,"cacheBlocks":true,"totalColumns":1,"maxResultSize":-1,"families":{"0":["ALL"]},"caching":100,"maxVersions":1,"timeRange":[0,1523338217847]}
{"loadColumnFamiliesOnDemand":null,"startRow":"CS","stopRow":"EU","batch":-1,"cacheBlocks":true,"totalColumns":1,"maxResultSize":-1,"families":{"0":["ALL"]},"caching":100,"maxVersions":1,"timeRange":[0,1523338217847]}
{"loadColumnFamiliesOnDemand":null,"startRow":"EU","stopRow":"NA","batch":-1,"cacheBlocks":true,"totalColumns":1,"maxResultSize":-1,"families":{"0":["ALL"]},"caching":100,"maxVersions":1,"timeRange":[0,1523338217847]}
{"loadColumnFamiliesOnDemand":null,"startRow":"NA","stopRow":"","batch":-1,"cacheBlocks":true,"totalColumns":1,"maxResultSize":-1,"families":{"0":["ALL"]},"caching":100,"maxVersions":1,"timeRange":[0,1523338217847]}Disconnected from the target VM, address: '127.0.0.1:63406', transport: 'socket'