Question

我之前使用的是暴风雨，我需要更多的批处理功能，所以我在风暴中搜索批处理。我发现三叉戟实时进行微批处理。

但不知何故，我无法弄清楚Trident如何处理微批处理（流量，批量大小，批处理间隔）以确定它真的有我需要的东西。

我想要做的是收集/保存一个间隔中的喷口发出的元组，并用另一个时间间隔将它们重新发射到下游组件/螺栓/功能。（例如，spout每秒发出一个元组，下一个三叉戟函数将收集/保存元组，每分钟发出50个元组到下一个函数。）

在这种情况下，有人可以指导我如何申请Trident吗？或使用风暴功能的任何其他适用方式？

Answer 1

很棒的问题！但遗憾的是，三叉戟盒子不支持这种微型批处理。

但您可以尝试实施自己的频率驱动微批处理。像这个骨架的例子：

import java.util.ArrayList;
import java.util.List;
import java.util.Map;
import java.util.concurrent.LinkedBlockingQueue;

import org.apache.storm.task.OutputCollector;
import org.apache.storm.task.TopologyContext;
import org.apache.storm.topology.OutputFieldsDeclarer;
import org.apache.storm.topology.base.BaseRichBolt;
import org.apache.storm.tuple.Tuple;
import org.slf4j.Logger;
import org.slf4j.LoggerFactory;

public class MicroBatchingBolt extends BaseRichBolt {

    private static final long serialVersionUID = 8500984730263268589L;
    private static final Logger LOG = LoggerFactory.getLogger(MicroBatchingBolt.class);

    protected LinkedBlockingQueue<Tuple> queue = new LinkedBlockingQueue<Tuple>();

    /** The threshold after which the batch should be flushed out. */
    int batchSize = 100;

    /**
     * The batch interval in sec. Minimum time between flushes if the batch sizes
     * are not met. This should typically be equal to
     * topology.tick.tuple.freq.secs and half of topology.message.timeout.secs
     */
    int batchIntervalInSec = 45;

    /** The last batch process time seconds. Used for tracking purpose */
    long lastBatchProcessTimeSeconds = 0;

    private OutputCollector collector;

    @Override
    @SuppressWarnings("rawtypes")
    public void prepare(Map stormConf, TopologyContext context, OutputCollector collector) {
        this.collector = collector;
    }

    @Override
    public void execute(Tuple tuple) {
      // Check if the tuple is of type Tick Tuple
      if (isTickTuple(tuple)) {
         // If so, it is indication for batch flush. But don't flush if previous
         // flush was done very recently (either due to batch size threshold was
         // crossed or because of another tick tuple

        if ((System.currentTimeMillis() / 1000 - lastBatchProcessTimeSeconds) >= batchIntervalInSec) {
          LOG.debug("Current queue size is " + this.queue.size()
              + ". But received tick tuple so executing the batch");

          finishBatch();
        } else {
          LOG.debug("Current queue size is " + this.queue.size()
              + ". Received tick tuple but last batch was executed "
              + (System.currentTimeMillis() / 1000 - lastBatchProcessTimeSeconds)
              + " seconds back that is less than " + batchIntervalInSec
              + " so ignoring the tick tuple");
        }
      } else {
        // Add the tuple to queue. But don't ack it yet.
        this.queue.add(tuple);
        int queueSize = this.queue.size();
        LOG.debug("current queue size is " + queueSize);
        if (queueSize >= batchSize) {
          LOG.debug("Current queue size is >= " + batchSize
              + " executing the batch");

          finishBatch();
        }
      }
    }

    private boolean isTickTuple(Tuple tuple) {
        // Check if it is tick tuple here
        return false;
    }

    /**
     * Finish batch.
     */
    public void finishBatch() {

      LOG.debug("Finishing batch of size " + queue.size());
      lastBatchProcessTimeSeconds = System.currentTimeMillis() / 1000;
      List<Tuple> tuples = new ArrayList<Tuple>();
      queue.drainTo(tuples);

      for (Tuple tuple : tuples) {
        // Prepare your batch here (may it be JDBC, HBase, ElasticSearch, Solr or
        // anything else.
        // List<Response> responses = externalApi.get("...");
      }

      try {
        // Execute your batch here and ack or fail the tuples
        LOG.debug("Executed the batch. Processing responses.");
        //        for (int counter = 0; counter < responses.length; counter++) {
        //          if (response.isFailed()) {
        //            LOG.error("Failed to process tuple # " + counter);
        //            this.collector.fail(tuples.get(counter));
        //          } else {
        //            LOG.debug("Successfully processed tuple # " + counter);
        //            this.collector.ack(tuples.get(counter));
        //          }
        //        }
      } catch (Exception e) {
        LOG.error("Unable to process " + tuples.size() + " tuples", e);
        // Fail entire batch
        for (Tuple tuple : tuples) {
          this.collector.fail(tuple);
        }
      }
    }

    @Override
    public void declareOutputFields(OutputFieldsDeclarer declarer) {
        // ... 
    }

}

来源：http://hortonworks.com/blog/apache-storm-design-pattern-micro-batching/和Using tick tuples with trident in storm

如何使用风暴Trident进行批处理元组？

1 个答案: