Spark的那些事(二)Structured streaming中Foreach sink的用法

Structured streaming默认支持的sink类型有File sink，Foreach sink，Console sink，Memory sink。
特别的说明一下Foreach sink的用法（ps：以通过Foreach sink写入外部redis为例）。

lastEtlData.writeStream().foreach(new TestForeachWriter()).outputMode("update").start();

foreach方法的参数为ForeachWriter对象，看下api说明：

  datasetOfString.writeStream().foreach(new ForeachWriter<String>() {    @Override
    public boolean open(long partitionId, long version) {      // open connection
      //此处用于打开连接，以redis为例，此处从redis连接池中获取连接
    }    @Override
    public void process(String value) {      // write string to connection
      //此处用于数据写入redis。value为GenericRowWithSchema对象
    }    @Override
    public void close(Throwable errorOrNull) {      // close the connection
      //此处用于关闭连接
    }
  });

看一下三个方法在的调用过程，因为是每个Partition的一批数据调用一次，还是需要关注下open，close的频率。一批数据open，close各一次。如下所示：

data.queryExecution.toRdd.foreachPartition { iter =>  if (writer.open(TaskContext.getPartitionId(), batchId)) {    try {      while (iter.hasNext) {
        writer.process(encoder.fromRow(iter.next()))
      }
    } catch {      case e: Throwable =>
        writer.close(e)        throw e
    }
    writer.close(null)
  } else {
    writer.close(null)
  }
}

附写redis的foreachwriter实现：

public static class TestForeachWriter extends ForeachWriter implements Serializable{       public static JedisPool jedisPool;        public Jedis jedis;        static {
                JedisPoolConfig config = new JedisPoolConfig();
                config.setMaxTotal(20);
                config.setMaxIdle(5);
                config.setMaxWaitMillis(1000);
                config.setMinIdle(2);
                config.setTestOnBorrow(false);
                jedisPool = new JedisPool(config, "127.0.0.1", 6379);
        }        public static synchronized Jedis getJedis() {            return jedisPool.getResource();
        }        @Override
        public boolean open(long partitionId, long version) {
           jedis = getJedis();            return true;
        }        @Override
        public void process(Object value) {
            GenericRowWithSchema genericRowWithSchema = (GenericRowWithSchema) value;

           System.out.println(((GenericRowWithSchema) value).get(0).toString()+"-----------"+ ((GenericRowWithSchema) value).get(2).toString());

            jedis.set(((GenericRowWithSchema) value).get(Integer.parseInt(genericRowWithSchema.schema().getFieldIndex("ID").get().toString())).toString(),((GenericRowWithSchema) value).get(Integer.parseInt(genericRowWithSchema.schema().getFieldIndex("COUNT(ADA1)").get().toString())).toString());
            System.out.println("++++++++++"+((GenericRowWithSchema) value).get(Integer.parseInt(genericRowWithSchema.schema().getFieldIndex("ID").get().toString())).toString());

        }        @Override
        public void close(Throwable errorOrNull) {

            jedis.close();
        }
    }

特别提醒：需要显示实现Serializable接口。

作者：假文艺的真码农
链接：https://www.jianshu.com/p/fa802faf747a