在阅读了Apache Flume以及它在处理客户端事件方面提供的好处之后,我决定是时候开始更详细地研究它了。另一个好处似乎是它可以处理Apache Avro对象:-)但是,我很难理解Avro架构如何用于验证收到的Flume事件。
为了更详细地了解我的问题,我在下面提供了代码段;
为了这篇文章的目的,我使用一个示例模式定义一个嵌套的Object1
记录与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"
}
]
}
}
]
}
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,我创建了一个新的Flume事件,其中“Test”被定义为将事件发送到在VM(192.168.99.101)内运行的单独Apache Flume代理的事件主体。
如上所述,我已将此代理配置为从嵌入的Flume代理接收事件。该代理的Flume配置看起来像;
# 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程序;
./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架构以确保Flume只接收和处理有效事件。有人可以帮我理解我哪里错了吗?或者,如果我误解了Flume如何将Flume事件转换为Avro事件的意图?
除了上述内容之外,我还尝试在更改Avro架构后使用Avro RPC客户端指定直接与我的远程Flume代理通信的协议,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,我能够确保应用程序填充的事件符合生成的Avro架构,以避免发布无效事件。我希望我使用嵌入式Flume代理实现这一点,但如果不可能,那么我会考虑使用Avro RPC方法直接与我的远程Flume代理进行对话。
任何帮助/指导都将是一个很大的帮助。提前致谢。
在进一步阅读后,我想知道我是否误解了Apache Flume的目的。我原本以为这可以用来根据数据/模式自动创建Avro事件,但现在想知道应用程序是否应该负责生成Avro事件,这些事件将根据通道配置存储在Flume中并通过批处理发送接收器(在我的例子中是一个Spark Streaming集群)。
如果以上是正确的,那么我想知道Flume是否需要知道架构或者我的Spark Streaming集群是否最终会处理这些数据?如果需要Flume了解架构,那么请您详细说明如何实现这一点?
提前致谢。
由于您的目标是使用Spark Streaming集群处理数据,因此您可以使用2个解决方案解决此问题
1)使用Flume客户端(使用flume-ng-sdk 1.9.0测试)和Spark Streaming(使用spark-streaming_2.11 2.4.0和spark-streaming-flume_2.11 2.3.0测试),在网络之间没有Flume服务器拓扑结构。
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)在+ Spark Streaming(作为Flume Sink)之间使用Flume客户端+ Flume服务器作为网络拓扑。
对于此选项,代码是相同的,但现在SparkStreaming必须指定完整的dns限定主机名而不是localhost,以便在同一端口41416启动SparkStreaming服务器,如果您在本地运行此测试。 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才能工作。
请参阅下面的水槽配置
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
你应该用这两个解决方案获得相同的结果,但是回到你的问题是,对于来自Json流的Spark Streaming内容是否真的需要Flume,答案是它取决于,Flume支持拦截器所以在这种情况下它可以用来清理或过滤无效您的Spark项目的数据,但由于您要向拓扑添加额外的组件,它可能会影响性能并且需要比没有Flume更多的资源(CPU /内存)。