Apache Flume中的Apache Avro模式验证

时间:2019-02-18 15:01:46

标签: avro flume flume-ng

在了解了Apache Flume及其在处理客户端事件方面所提供的好处之后,我决定是时候开始更详细地研究它了。另一个巨大的好处似乎是它可以处理Apache Avro对象:-)但是,我很难理解Avro模式如何用于验证收到的Flume事件。

为帮助更详细地了解我的问题,我在下面提供了代码段;

Avro模式

出于这篇文章的目的,我使用一个示例架构,该架构定义了一个嵌套的Object1记录,其中包含2个字段。

{
  "namespace": "com.example.avro",
  "name": "Example",
  "type": "record",
  "fields": [
    {
      "name": "object1",
      "type": {
        "name": "Object1",
        "type": "record",
        "fields": [
          {
            "name": "value1",
            "type": "string"
          },
          {
            "name": "value2",
            "type": "string"
          }
        ]
      }
    }
  ]
}

嵌入式Flume代理

在我的Java项目中,我目前正在使用Apache Flume嵌入式代理,如下所述;

public static void main(String[] args) {
    final Event event = EventBuilder.withBody("Test", Charset.forName("UTF-8"));

    final Map<String, String> properties = new HashMap<>();
    properties.put("channel.type", "memory");
    properties.put("channel.capacity", "100");
    properties.put("sinks", "sink1");
    properties.put("sink1.type", "avro");
    properties.put("sink1.hostname", "192.168.99.101");
    properties.put("sink1.port", "11111");
    properties.put("sink1.batch-size", "1");
    properties.put("processor.type", "failover");

    final EmbeddedAgent embeddedAgent = new EmbeddedAgent("TestAgent");
    embeddedAgent.configure(properties);
    embeddedAgent.start();

    try {
        embeddedAgent.put(event);
    } catch (EventDeliveryException e) {
        e.printStackTrace();
    }
}

在上面的示例中,我正在创建一个新的Flume事件,其“测试”定义为将事件发送到VM(192.168.99.101)中运行的单独Apache Flume代理的事件主体。

远程Flume代理

如上所述,我已将该代理配置为从嵌入式Flume代理接收事件。该代理的Flume配置如下;

# Name the components on this agent
hello.sources = avroSource
hello.channels = memoryChannel
hello.sinks = loggerSink

# Describe/configure the source
hello.sources.avroSource.type = avro
hello.sources.avroSource.bind = 0.0.0.0
hello.sources.avroSource.port = 11111
hello.sources.avroSource.channels = memoryChannel

# Describe the sink
hello.sinks.loggerSink.type = logger

# Use a channel which buffers events in memory
hello.channels.memoryChannel.type = memory
hello.channels.memoryChannel.capacity = 1000
hello.channels.memoryChannel.transactionCapacity = 1000

# Bind the source and sink to the channel
hello.sources.avroSource.channels = memoryChannel
hello.sinks.loggerSink.channel = memoryChannel

我正在执行以下命令来启动代理;

./bin/flume-ng agent --conf conf --conf-file ../sample-flume.conf --name hello -Dflume.root.logger=TRACE,console -Dorg.apache.flume.log.printconfig=true -Dorg.apache.flume.log.rawdata=true

当我执行Java项目main方法时,我看到“ Test”事件通过以下输出传递到我的记录器接收器;

2019-02-18 14:15:09,998 (SinkRunner-PollingRunner-DefaultSinkProcessor) [INFO - org.apache.flume.sink.LoggerSink.process(LoggerSink.java:95)] Event: { headers:{} body: 54 65 73 74                                     Test }

但是,我不清楚我应该在哪里配置Avro模式,以确保Flume仅接收和处理有效事件。有人可以帮我了解我要去哪里了吗?或者,如果我误解了Flume如何将Flume事件转换为Avro事件的意图?

除上述内容外,在更改Avro模式以指定直接与远程Flume代理通信的协议之后,我还尝试使用Avro RPC客户端,但是当我尝试发送事件时,我看到以下错误;

Exception in thread "main" org.apache.avro.AvroRuntimeException: Not a remote message: test
    at org.apache.avro.ipc.Requestor$Response.getResponse(Requestor.java:532)
    at org.apache.avro.ipc.Requestor$TransceiverCallback.handleResult(Requestor.java:359)
    at org.apache.avro.ipc.Requestor$TransceiverCallback.handleResult(Requestor.java:322)
    at org.apache.avro.ipc.NettyTransceiver$NettyClientAvroHandler.messageReceived(NettyTransceiver.java:613)
    at org.jboss.netty.channel.SimpleChannelUpstreamHandler.handleUpstream(SimpleChannelUpstreamHandler.java:70)
    at org.apache.avro.ipc.NettyTransceiver$NettyClientAvroHandler.handleUpstream(NettyTransceiver.java:595)
    at org.jboss.netty.channel.DefaultChannelPipeline.sendUpstream(DefaultChannelPipeline.java:558)
    at org.jboss.netty.channel.DefaultChannelPipeline$DefaultChannelHandlerContext.sendUpstream(DefaultChannelPipeline.java:786)
    at org.jboss.netty.channel.Channels.fireMessageReceived(Channels.java:296)
    at org.jboss.netty.handler.codec.frame.FrameDecoder.unfoldAndFireMessageReceived(FrameDecoder.java:458)
    at org.jboss.netty.handler.codec.frame.FrameDecoder.callDecode(FrameDecoder.java:439)
    at org.jboss.netty.handler.codec.frame.FrameDecoder.messageReceived(FrameDecoder.java:303)
    at org.jboss.netty.channel.SimpleChannelUpstreamHandler.handleUpstream(SimpleChannelUpstreamHandler.java:70)
    at org.jboss.netty.channel.DefaultChannelPipeline.sendUpstream(DefaultChannelPipeline.java:558)
    at org.jboss.netty.channel.DefaultChannelPipeline.sendUpstream(DefaultChannelPipeline.java:553)
    at org.jboss.netty.channel.Channels.fireMessageReceived(Channels.java:268)
    at org.jboss.netty.channel.Channels.fireMessageReceived(Channels.java:255)
    at org.jboss.netty.channel.socket.nio.NioWorker.read(NioWorker.java:84)
    at org.jboss.netty.channel.socket.nio.AbstractNioWorker.processSelectedKeys(AbstractNioWorker.java:471)
    at org.jboss.netty.channel.socket.nio.AbstractNioWorker.run(AbstractNioWorker.java:332)
    at org.jboss.netty.channel.socket.nio.NioWorker.run(NioWorker.java:35)
    at org.jboss.netty.util.ThreadRenamingRunnable.run(ThreadRenamingRunnable.java:102)
    at org.jboss.netty.util.internal.DeadLockProofWorker$1.run(DeadLockProofWorker.java:42)
    at java.util.concurrent.ThreadPoolExecutor.runWorker(ThreadPoolExecutor.java:1149)
    at java.util.concurrent.ThreadPoolExecutor$Worker.run(ThreadPoolExecutor.java:624)
    at java.lang.Thread.run(Thread.java:748)

我的目标是确保应用程序填充的事件符合所生成的Avro架构,从而避免发布无效事件。我希望我使用嵌入式Flume代理来实现这一目标,但是如果无法实现,那么我会考虑使用Avro RPC方法直接与远程Flume代理进行通信。

任何帮助/指导都将是一个很大的帮助。预先感谢。

更新

进一步阅读后,我想知道我是否误解了Apache Flume的目的。我本来以为可以用于根据数据/架构自动创建Avro事件,但是现在想知道应用程序是否应该负责产生Avro事件,这些事件将根据通道配置存储在Flume中,并通过接收器(在我的情况下是Spark Streaming集群)。

如果以上正确,那么我想知道是否需要Flume了解该模式,或者仅是我的Spark Streaming集群,它将最终处理此数据?如果需要Flume了解该架构,那么您能否提供详细说明该如何实现的?

谢谢。

1 个答案:

答案 0 :(得分:0)

由于您的目标是使用Spark Streaming集群处理数据,因此可以使用2种解决方案来解决此问题

1)在不使用Flume服务器的情况下使用Flume客户端(已通过flume-ng-sdk 1.9.0测试)和Spark Streaming(已通过spark-streaming_2.11 2.4.0和spark-streaming-flume_2.11 2.3.0测试)在网络拓扑之间。

客户端类在端口41416发送Flume json事件

  public class JSONFlumeClient {
    public static void main(String[] args) {
    RpcClient client = RpcClientFactory.getDefaultInstance("localhost", 41416);
    String jsonData = "{\r\n" + "  \"namespace\": \"com.example.avro\",\r\n" + "  \"name\": \"Example\",\r\n"
            + "  \"type\": \"record\",\r\n" + "  \"fields\": [\r\n" + "    {\r\n"
            + "      \"name\": \"object1\",\r\n" + "      \"type\": {\r\n" + "        \"name\": \"Object1\",\r\n"
            + "        \"type\": \"record\",\r\n" + "        \"fields\": [\r\n" + "          {\r\n"
            + "            \"name\": \"value1\",\r\n" + "            \"type\": \"string\"\r\n" + "          },\r\n"
            + "          {\r\n" + "            \"name\": \"value2\",\r\n" + "            \"type\": \"string\"\r\n"
            + "          }\r\n" + "        ]\r\n" + "      }\r\n" + "    }\r\n" + "  ]\r\n" + "}";
    Event event = EventBuilder.withBody(jsonData, Charset.forName("UTF-8"));
    try {
        client.append(event);
    } catch (Throwable t) {
        System.err.println(t.getMessage());
        t.printStackTrace();
    } finally {
        client.close();
    }
  }
}

Spark Streaming Server类侦听端口41416

public class SparkStreamingToySample {
  public static void main(String[] args) throws Exception {
    SparkConf sparkConf = new SparkConf().setMaster("local[2]")
    .setAppName("SparkStreamingToySample");
    JavaStreamingContext ssc = new JavaStreamingContext(sparkConf, Durations.seconds(30));
    JavaReceiverInputDStream<SparkFlumeEvent> lines = FlumeUtils
    .createStream(ssc, "localhost", 41416);
    lines.map(sfe -> new String(sfe.event().getBody().array(), "UTF-8"))
    .foreachRDD((data,time)->
    System.out.println("***" + new Date(time.milliseconds()) + "=" + data.collect().toString()));
    ssc.start();
    ssc.awaitTermination();
  }
}

2)使用Flume客户端+ Flume服务器之间+ Spark Streaming(作为Flume Sink)作为网络拓扑。

对于此选项,代码是相同的,但是现在SparkStreaming必须指定完整的dns合格主机名而不是localhost才能在相同的端口41416上启动SparkStreaming服务器(如果您在本地运行此端口进行测试)。 Flume客户端将连接到水槽服务器端口41415。现在,棘手的部分是如何定义水槽拓扑。您需要同时指定源和接收器。

请参阅下面的flume conf

agent1.channels.ch1.type = memory

agent1.sources.avroSource1.channels = ch1
agent1.sources.avroSource1.type = avro
agent1.sources.avroSource1.bind = 0.0.0.0
agent1.sources.avroSource1.port = 41415

agent1.sinks.avroSink.channel = ch1
agent1.sinks.avroSink.type = avro
agent1.sinks.avroSink.hostname = <full dns qualified hostname>
agent1.sinks.avroSink.port = 41416

agent1.channels = ch1
agent1.sources = avroSource1
agent1.sinks = avroSink

两种解决方案都应该获得相同的结果,但是回到您的问题,即Json流中的Spark Streaming内容是否真的需要Flume,答案取决于它,Flume支持拦截器,因此在这种情况下可以用来清理或过滤Spark项目中的无效数据,但由于要向拓扑中添加额外的组件,因此与不使用Flume相比,它可能会影响性能并需要更多的资源(CPU /内存)。