在MaxCompute中利用bitmap进行数据处理

很多数据开发者使用bitmap技术对用户数据进行编码和压缩，然后利用bitmap的与/或/非的极速处理速度，实现类似用户画像标签的人群筛选、运营分析的7日活跃等分析。
本文给出了一个使用MaxCompute MapReduce开发一个对不同日期活跃用户ID进行bitmap编码和计算的样例。供感兴趣的用户进一步了解、分析，并应用在自己的场景下。

import com.aliyun.odps.OdpsException;import com.aliyun.odps.data.Record;import com.aliyun.odps.data.TableInfo;import com.aliyun.odps.mapred.JobClient;import com.aliyun.odps.mapred.MapperBase;import com.aliyun.odps.mapred.ReducerBase;import com.aliyun.odps.mapred.conf.JobConf;import com.aliyun.odps.mapred.utils.InputUtils;import com.aliyun.odps.mapred.utils.OutputUtils;import com.aliyun.odps.mapred.utils.SchemaUtils;import org.roaringbitmap.RoaringBitmap;import org.roaringbitmap.buffer.ImmutableRoaringBitmap;import java.io.DataOutputStream;import java.io.IOException;import java.io.OutputStream;import java.nio.ByteBuffer;import java.util.Base64;import java.util.Iterator;public class bitmapDemo2{    public static class BitMapper extends MapperBase {

        Record key;
        Record value;        @Override
        public void setup(TaskContext context) throws IOException {
            key = context.createMapOutputKeyRecord();
            value = context.createMapOutputValueRecord();
        }        @Override
        public void map(long recordNum, Record record, TaskContext context)
                throws IOException        {
            RoaringBitmap mrb=new RoaringBitmap();            long AID=0;
            {
                {
                    {
                        {
                            AID=record.getBigint("id");
                            mrb.add((int) AID);                            //获取key
                            key.set(new Object[] {record.getString("active_date")});

                        }
                    }
                }
            }
            ByteBuffer outbb = ByteBuffer.allocate(mrb.serializedSizeInBytes());
            mrb.serialize(new DataOutputStream(new OutputStream(){
                ByteBuffer mBB;                OutputStream init(ByteBuffer mbb) {mBB=mbb; return this;}                public void close() {}                public void flush() {}                public void write(int b) {
                    mBB.put((byte) b);}                public void write(byte[] b) {mBB.put(b);}                public void write(byte[] b, int off, int l) {mBB.put(b,off,l);}
            }.init(outbb)));
            String serializedstring = Base64.getEncoder().encodeToString(outbb.array());
            value.set(new Object[] {serializedstring});
            context.write(key, value);
        }
    }    public static class BitReducer extends ReducerBase {        private Record result = null;        public void setup(TaskContext context) throws IOException {
            result = context.createOutputRecord();
        }        public void reduce(Record key, Iterator<Record> values, TaskContext context) throws IOException {            long fcount = 0;
            RoaringBitmap rbm=new RoaringBitmap();            while (values.hasNext())
            {
                Record val = values.next();
                ByteBuffer newbb = ByteBuffer.wrap(Base64.getDecoder().decode((String)val.get(0)));
                ImmutableRoaringBitmap irb = new ImmutableRoaringBitmap(newbb);
                RoaringBitmap p= new RoaringBitmap(irb);
                rbm.or(p);
            }
            ByteBuffer outbb = ByteBuffer.allocate(rbm.serializedSizeInBytes());
            rbm.serialize(new DataOutputStream(new OutputStream(){
                ByteBuffer mBB;                OutputStream init(ByteBuffer mbb) {mBB=mbb; return this;}                public void close() {}                public void flush() {}                public void write(int b) {
                    mBB.put((byte) b);}                public void write(byte[] b) {mBB.put(b);}                public void write(byte[] b, int off, int l) {mBB.put(b,off,l);}
            }.init(outbb)));
            String serializedstring = Base64.getEncoder().encodeToString(outbb.array());
            result.set(0, key.get(0));
            result.set(1, serializedstring);
            context.write(result);
        }
    }    public static void main( String[] args ) throws OdpsException    {

        System.out.println("begin.........");
        JobConf job = new JobConf();
        
        job.setMapperClass(BitMapper.class);
        job.setReducerClass(BitReducer.class);

        job.setMapOutputKeySchema(SchemaUtils.fromString("active_date:string"));
        job.setMapOutputValueSchema(SchemaUtils.fromString("id:string"));

        InputUtils.addTable(TableInfo.builder().tableName("bitmap_source").cols(new String[] {"id","active_date"}).build(), job);//        +------------+-------------+//        | id         | active_date |//        +------------+-------------+//        | 1          | 20190729    |//        | 2          | 20190729    |//        | 3          | 20190730    |//        | 4          | 20190801    |//        | 5          | 20190801    |//        +------------+-------------+
        OutputUtils.addTable(TableInfo.builder().tableName("bitmap_target").build(), job);//        +-------------+------------+//        | active_date | bit_map    |//        +-------------+------------+//        20190729,OjAAAAEAAAAAAAEAEAAAAAEAAgA=3D//        20190730,OjAAAAEAAAAAAAAAEAAAAAMA//        20190801,OjAAAAEAAAAAAAEAEAAAAAQABQA=3D

        JobClient.runJob(job);
    }
}

对Java应用打包后，上传到MaxCompute项目中，即可在MaxCompute中调用该MR作业，对输入表的数据按日期作为key进行用户id的编码，同时按照相同日期对bitmap后的用户id取OR操作（根据需要可以取AND，例如存留场景），并将处理后的数据写入目标结构表当中供后续处理使用。