flume学习（四）：Flume Channel Selectors使用

问题导读
1、怎样将不同项目的的日志输出到不同的channel？
2、如何理解一个sink为hdfs,一个sink为logger的拓扑结构？
3、怎样在Log4jExtAppender.java类里扩展一个参数？

前几篇文章只有一个项目的日志，现在我们考虑多个项目的日志的收集，我拷贝了一份flumedemo项目,重命名为flumedemo2,添加了一个WriteLog2.java类，稍微改动了一下JSON字符串的输出，将以前requestUrl中的"reporter-api"改为了"image-api",以便和WriteLog类的输出稍微区分开来，如下：

package com.besttone.flume;  
  
import java.util.Date;  
  
import org.apache.commons.logging.Log;  
import org.apache.commons.logging.LogFactory;  
  
public class WriteLog2 {  
    protected static final Log logger = LogFactory.getLog(WriteLog2.class);  
  
    /** 
     * @param args 
     * @throws InterruptedException 
     */  
    public static void main(String[] args) throws InterruptedException {  
        // TODO Auto-generated method stub  
        while (true) {  
            logger.info(new Date().getTime());  
            logger.info("{"requestTime":"  
                    + System.currentTimeMillis()  
                    + ","requestParams":{"timestamp":1405499314238,"phone":"02038824941","cardName":"测试商家名称","provinceCode":"440000","cityCode":"440106"},"requestUrl":"/image-api/reporter/reporter12/init.do"}");  
            Thread.sleep(2000);  
  
        }  
    }  
}
复制代码

现在有这么一个需求描述：要求flumedemo的项目的log4j日志输出到hdfs,而flumedemo2项目的log4j日志输出到agent的log日志中。

我们还是采用log4jappender来配置log4j输出给flume的souce,现在的需求明显是有两个sink了，一个sink为hdfs,一个sink为logger。于是现在的拓扑结构应该是这样的：

需要实现这么一个拓扑接口，就需要使用到channel selectors，让不同的项目日志通过不同的channel到不同的sink中去。

官方文档上channel selectors 有两种类型:

Replicating Channel Selector (default)

Multiplexing Channel Selector

这两种selector的区别是:Replicating 会将source过来的events发往所有channel,而Multiplexing 可以选择该发往哪些channel。对于上面的例子来说，如果采用Replicating ,那么demo和demo2的日志会同时发往channel1和channel2,这显然是和需求不符的，需求只是让demo的日志发往channel1,而demo2的日志发往channel2。

综上所述，我们选择Multiplexing Channel Selector。这里我们有遇到一个棘手的问题，Multiplexing 需要判断header里指定key的值来决定分发到某个具体的channel，我们现在demo和demo2同时运行在同一个服务器上，如果在不同的服务器上运行，我们可以在 source1上加上一个 host 拦截器（上一篇有介绍过），这样可以通过header中的host来判断event该分发给哪个channel，而这里是在同一个服务器上，由host是区分不出来日志的来源的，我们必须想办法在header中添加一个key来区分日志的来源。

设想一下，如果header中有一个key:flume.client.log4j.logger.source,我们通过设置这个key的值，demo设为app1,demo2设为app2,这样我们就能通过设置：

tier1.sources.source1.channels=channel1 channel2
tier1.sources.source1.selector.type=multiplexing
tier1.sources.source1.selector.header=flume.client.log4j.logger.source
tier1.sources.source1.selector.mapping.app1=channel1
tier1.sources.source1.selector.mapping.app2=channel2
复制代码

来将不同项目的的日志输出到不同的channel了。

我们按照这个思路继续下去，遇到了困难，log4jappender没有这样的参数来让你设置。怎么办？翻看了一下log4jappender的源码，发现可以很容易的实现扩展参数，于是我复制了一份log4jappender代码，新加了一个类叫Log4jExtAppender.java,里面扩展了一个参数叫：source,代码如下：

package com.besttone.flume;  
  
import java.io.ByteArrayOutputStream;  
import java.io.IOException;  
import java.nio.charset.Charset;  
import java.util.HashMap;  
import java.util.Map;  
import java.util.Properties;  
  
import org.apache.avro.Schema;  
import org.apache.avro.generic.GenericRecord;  
import org.apache.avro.io.BinaryEncoder;  
import org.apache.avro.io.DatumWriter;  
import org.apache.avro.io.EncoderFactory;  
import org.apache.avro.reflect.ReflectData;  
import org.apache.avro.reflect.ReflectDatumWriter;  
import org.apache.avro.specific.SpecificRecord;  
import org.apache.flume.Event;  
import org.apache.flume.EventDeliveryException;  
import org.apache.flume.FlumeException;  
import org.apache.flume.api.RpcClient;  
import org.apache.flume.api.RpcClientConfigurationConstants;  
import org.apache.flume.api.RpcClientFactory;  
import org.apache.flume.clients.log4jappender.Log4jAvroHeaders;  
import org.apache.flume.event.EventBuilder;  
import org.apache.log4j.AppenderSkeleton;  
import org.apache.log4j.helpers.LogLog;  
import org.apache.log4j.spi.LoggingEvent;  
  
/** 
 *  
 * Appends Log4j Events to an external Flume client which is decribed by the 
 * Log4j configuration file. The appender takes two required parameters: 
 * <p> 
 * <strong>Hostname</strong> : This is the hostname of the first hop at which 
 * Flume (through an AvroSource) is listening for events. 
 * </p> 
 * <p> 
 * <strong>Port</strong> : This the port on the above host where the Flume 
 * Source is listening for events. 
 * </p> 
 * A sample log4j properties file which appends to a source would look like: 
 *  
 * <pre> 
 * <p> 
 * log4j.appender.out2 = org.apache.flume.clients.log4jappender.Log4jAppender 
 * log4j.appender.out2.Port = 25430 
 * log4j.appender.out2.Hostname = foobarflumesource.com 
 * log4j.logger.org.apache.flume.clients.log4jappender = DEBUG,out2</p> 
 * </pre> 
 * <p> 
 * <i>Note: Change the last line to the package of the class(es), that will do 
 * the appending.For example if classes from the package com.bar.foo are 
 * appending, the last line would be:</i> 
 * </p> 
 *  
 * <pre> 
 * <p>log4j.logger.com.bar.foo = DEBUG,out2</p> 
 * </pre> 
 *  
 *  
 */  
public class Log4jExtAppender extends AppenderSkeleton {  
  
    private String hostname;  
    private int port;  
    private String source;  
  
    public String getSource() {  
        return source;  
    }  
  
    public void setSource(String source) {  
        this.source = source;  
    }  
  
    private boolean unsafeMode = false;  
    private long timeout = RpcClientConfigurationConstants.DEFAULT_REQUEST_TIMEOUT_MILLIS;  
    private boolean avroReflectionEnabled;  
    private String avroSchemaUrl;  
  
    RpcClient rpcClient = null;  
  
    /** 
     * If this constructor is used programmatically rather than from a log4j 
     * conf you must set the <tt>port</tt> and <tt>hostname</tt> and then call 
     * <tt>activateOptions()</tt> before calling <tt>append()</tt>. 
     */  
    public Log4jExtAppender() {  
    }  
  
    /** 
     * Sets the hostname and port. Even if these are passed the 
     * <tt>activateOptions()</tt> function must be called before calling 
     * <tt>append()</tt>, else <tt>append()</tt> will throw an Exception. 
     *  
     * @param hostname 
     *            The first hop where the client should connect to. 
     * @param port 
     *            The port to connect on the host. 
     *  
     */  
    public Log4jExtAppender(String hostname, int port, String source) {  
        this.hostname = hostname;  
        this.port = port;  
        this.source = source;  
    }  
  
    /** 
     * Append the LoggingEvent, to send to the first Flume hop. 
     *  
     * @param event 
     *            The LoggingEvent to be appended to the flume. 
     * @throws FlumeException 
     *             if the appender was closed, or the hostname and port were not 
     *             setup, there was a timeout, or there was a connection error. 
     */  
    @Override  
    public synchronized void append(LoggingEvent event) throws FlumeException {  
        // If rpcClient is null, it means either this appender object was never  
        // setup by setting hostname and port and then calling activateOptions  
        // or this appender object was closed by calling close(), so we throw an  
        // exception to show the appender is no longer accessible.  
        if (rpcClient == null) {  
            String errorMsg = "Cannot Append to Appender! Appender either closed or"  
                    + " not setup correctly!";  
            LogLog.error(errorMsg);  
            if (unsafeMode) {  
                return;  
            }  
            throw new FlumeException(errorMsg);  
        }  
  
        if (!rpcClient.isActive()) {  
            reconnect();  
        }  
  
        // Client created first time append is called.  
        Map<String, String> hdrs = new HashMap<String, String>();  
        hdrs.put(Log4jAvroHeaders.LOGGER_NAME.toString(), event.getLoggerName());  
        hdrs.put(Log4jAvroHeaders.TIMESTAMP.toString(),  
                String.valueOf(event.timeStamp));  
  
        // 添加日志来源  
        if (this.source == null || this.source.equals("")) {  
            this.source = "unknown";  
        }  
        hdrs.put("flume.client.log4j.logger.source", this.source);  
        // To get the level back simply use  
        // LoggerEvent.toLevel(hdrs.get(Integer.parseInt(  
        // Log4jAvroHeaders.LOG_LEVEL.toString()))  
        hdrs.put(Log4jAvroHeaders.LOG_LEVEL.toString(),  
                String.valueOf(event.getLevel().toInt()));  
  
        Event flumeEvent;  
        Object message = event.getMessage();  
        if (message instanceof GenericRecord) {  
            GenericRecord record = (GenericRecord) message;  
            populateAvroHeaders(hdrs, record.getSchema(), message);  
            flumeEvent = EventBuilder.withBody(  
                    serialize(record, record.getSchema()), hdrs);  
        } else if (message instanceof SpecificRecord || avroReflectionEnabled) {  
            Schema schema = ReflectData.get().getSchema(message.getClass());  
            populateAvroHeaders(hdrs, schema, message);  
            flumeEvent = EventBuilder  
                    .withBody(serialize(message, schema), hdrs);  
        } else {  
            hdrs.put(Log4jAvroHeaders.MESSAGE_ENCODING.toString(), "UTF8");  
            String msg = layout != null ? layout.format(event) : message  
                    .toString();  
            flumeEvent = EventBuilder.withBody(msg, Charset.forName("UTF8"),  
                    hdrs);  
        }  
  
        try {  
            rpcClient.append(flumeEvent);  
        } catch (EventDeliveryException e) {  
            String msg = "Flume append() failed.";  
            LogLog.error(msg);  
            if (unsafeMode) {  
                return;  
            }  
            throw new FlumeException(msg + " Exception follows.", e);  
        }  
    }  
  
    private Schema schema;  
    private ByteArrayOutputStream out;  
    private DatumWriter<Object> writer;  
    private BinaryEncoder encoder;  
  
    protected void populateAvroHeaders(Map<String, String> hdrs, Schema schema,  
            Object message) {  
        if (avroSchemaUrl != null) {  
            hdrs.put(Log4jAvroHeaders.AVRO_SCHEMA_URL.toString(), avroSchemaUrl);  
            return;  
        }  
        LogLog.warn("Cannot find ID for schema. Adding header for schema, "  
                + "which may be inefficient. Consider setting up an Avro Schema Cache.");  
        hdrs.put(Log4jAvroHeaders.AVRO_SCHEMA_LITERAL.toString(),  
                schema.toString());  
    }  
  
    private byte[] serialize(Object datum, Schema datumSchema)  
            throws FlumeException {  
        if (schema == null || !datumSchema.equals(schema)) {  
            schema = datumSchema;  
            out = new ByteArrayOutputStream();  
            writer = new ReflectDatumWriter<Object>(schema);  
            encoder = EncoderFactory.get().binaryEncoder(out, null);  
        }  
        out.reset();  
        try {  
            writer.write(datum, encoder);  
            encoder.flush();  
            return out.toByteArray();  
        } catch (IOException e) {  
            throw new FlumeException(e);  
        }  
    }  
  
    // This function should be synchronized to make sure one thread  
    // does not close an appender another thread is using, and hence risking  
    // a null pointer exception.  
    /** 
     * Closes underlying client. If <tt>append()</tt> is called after this 
     * function is called, it will throw an exception. 
     *  
     * @throws FlumeException 
     *             if errors occur during close 
     */  
    @Override  
    public synchronized void close() throws FlumeException {  
        // Any append calls after this will result in an Exception.  
        if (rpcClient != null) {  
            try {  
                rpcClient.close();  
            } catch (FlumeException ex) {  
                LogLog.error("Error while trying to close RpcClient.", ex);  
                if (unsafeMode) {  
                    return;  
                }  
                throw ex;  
            } finally {  
                rpcClient = null;  
            }  
        } else {  
            String errorMsg = "Flume log4jappender already closed!";  
            LogLog.error(errorMsg);  
            if (unsafeMode) {  
                return;  
            }  
            throw new FlumeException(errorMsg);  
        }  
    }  
  
    @Override  
    public boolean requiresLayout() {  
        // This method is named quite incorrectly in the interface. It should  
        // probably be called canUseLayout or something. According to the docs,  
        // even if the appender can work without a layout, if it can work with  
        // one,  
        // this method must return true.  
        return true;  
    }  
  
    /** 
     * Set the first flume hop hostname. 
     *  
     * @param hostname 
     *            The first hop where the client should connect to. 
     */  
    public void setHostname(String hostname) {  
        this.hostname = hostname;  
    }  
  
    /** 
     * Set the port on the hostname to connect to. 
     *  
     * @param port 
     *            The port to connect on the host. 
     */  
    public void setPort(int port) {  
        this.port = port;  
    }  
  
    public void setUnsafeMode(boolean unsafeMode) {  
        this.unsafeMode = unsafeMode;  
    }  
  
    public boolean getUnsafeMode() {  
        return unsafeMode;  
    }  
  
    public void setTimeout(long timeout) {  
        this.timeout = timeout;  
    }  
  
    public long getTimeout() {  
        return this.timeout;  
    }  
  
    public void setAvroReflectionEnabled(boolean avroReflectionEnabled) {  
        this.avroReflectionEnabled = avroReflectionEnabled;  
    }  
  
    public void setAvroSchemaUrl(String avroSchemaUrl) {  
        this.avroSchemaUrl = avroSchemaUrl;  
    }  
  
    /** 
     * Activate the options set using <tt>setPort()</tt> and 
     * <tt>setHostname()</tt> 
     *  
     * @throws FlumeException 
     *             if the <tt>hostname</tt> and <tt>port</tt> combination is 
     *             invalid. 
     */  
    @Override  
    public void activateOptions() throws FlumeException {  
        Properties props = new Properties();  
        props.setProperty(RpcClientConfigurationConstants.CONFIG_HOSTS, "h1");  
        props.setProperty(RpcClientConfigurationConstants.CONFIG_HOSTS_PREFIX  
                + "h1", hostname + ":" + port);  
        props.setProperty(  
                RpcClientConfigurationConstants.CONFIG_CONNECT_TIMEOUT,  
                String.valueOf(timeout));  
        props.setProperty(  
                RpcClientConfigurationConstants.CONFIG_REQUEST_TIMEOUT,  
                String.valueOf(timeout));  
        try {  
            rpcClient = RpcClientFactory.getInstance(props);  
            if (layout != null) {  
                layout.activateOptions();  
            }  
        } catch (FlumeException e) {  
            String errormsg = "RPC client creation failed! " + e.getMessage();  
            LogLog.error(errormsg);  
            if (unsafeMode) {  
                return;  
            }  
            throw e;  
        }  
    }  
  
    /** 
     * Make it easy to reconnect on failure 
     *  
     * @throws FlumeException 
     */  
    private void reconnect() throws FlumeException {  
        close();  
        activateOptions();  
    }  
}  
复制代码

然后然后将这个类打了一个jar包：Log4jExtAppender.jar，扔在了flumedemo和flumedemo2的lib目录下。
这时候flumedemo的log4j.properties如下：

log4j.rootLogger=INFO  
  
  
log4j.category.com.besttone=INFO,flume,console,LogFile  
  
#log4j.appender.flume = org.apache.flume.clients.log4jappender.Log4jExtAppender  
log4j.appender.flume = com.besttone.flume.Log4jExtAppender  
log4j.appender.flume.Hostname = localhost  
log4j.appender.flume.Port = 44444  
log4j.appender.flume.UnsafeMode = false  
log4j.appender.flume.Source = app1  
  
  
log4j.appender.console= org.apache.log4j.ConsoleAppender  
log4j.appender.console.Target= System.out  
log4j.appender.console.layout= org.apache.log4j.PatternLayout  
log4j.appender.console.layout.ConversionPattern= %d{yyyy-MM-dd HH:mm:ss} %5p %c{1}: %L - %m%n  
  
log4j.appender.LogFile= org.apache.log4j.DailyRollingFileAppender  
log4j.appender.LogFile.File= logs/app.log  
log4j.appender.LogFile.MaxFileSize=10KB  
log4j.appender.LogFile.Append= true  
log4j.appender.LogFile.Threshold= DEBUG  
log4j.appender.LogFile.layout= org.apache.log4j.PatternLayout  
log4j.appender.LogFile.layout.ConversionPattern= %-d{yyyy-MM-dd HH:mm:ss} [%t:%r] - [%5p] %m%n
复制代码

flumedemo2的如下：

log4j.rootLogger=INFO  
  
  
log4j.category.com.besttone=INFO,flume,console,LogFile  
  
#log4j.appender.flume = org.apache.flume.clients.log4jappender.Log4jExtAppender  
log4j.appender.flume = com.besttone.flume.Log4jExtAppender  
log4j.appender.flume.Hostname = localhost  
log4j.appender.flume.Port = 44444  
log4j.appender.flume.UnsafeMode = false  
log4j.appender.flume.Source = app2  
  
  
log4j.appender.console= org.apache.log4j.ConsoleAppender  
log4j.appender.console.Target= System.out  
log4j.appender.console.layout= org.apache.log4j.PatternLayout  
log4j.appender.console.layout.ConversionPattern= %d{yyyy-MM-dd HH:mm:ss} %5p %c{1}: %L - %m%n  
  
log4j.appender.LogFile= org.apache.log4j.DailyRollingFileAppender  
log4j.appender.LogFile.File= logs/app.log  
log4j.appender.LogFile.MaxFileSize=10KB  
log4j.appender.LogFile.Append= true  
log4j.appender.LogFile.Threshold= DEBUG  
log4j.appender.LogFile.layout= org.apache.log4j.PatternLayout  
log4j.appender.LogFile.layout.ConversionPattern= %-d{yyyy-MM-dd HH:mm:ss} [%t:%r] - [%5p] %m%n  
复制代码

将原来的log4j.appender.flume 由org.apache.flume.clients.log4jappender.Log4jExtAppender改为了我重新实现添加了source参数的com.besttone.flume.Log4jExtAppender

然后flumedemo的log4j.appender.flume.Source = app1，flumedemo2的log4j.appender.flume.Source = app2。
运行flumedemo的WriteLog类，和flumedemo2的WriteLog2类，分别去hdfs上和agent的log文件中看看内容，发现hdfs上都是app1的日志，log文件中都是app2的日志，功能实现。

完整的flume.conf如下：

tier1.sources=source1  
tier1.channels=channel1 channel2  
tier1.sinks=sink1 sink2  
tier1.sources.source1.type=avro  
tier1.sources.source1.bind=0.0.0.0  
tier1.sources.source1.port=44444  
tier1.sources.source1.channels=channel1 channel2  
tier1.sources.source1.selector.type=multiplexing  
tier1.sources.source1.selector.header=flume.client.log4j.logger.source  
tier1.sources.source1.selector.mapping.app1=channel1  
tier1.sources.source1.selector.mapping.app2=channel2  
tier1.sources.source1.interceptors=i1 i2  
tier1.sources.source1.interceptors.i1.type=regex_filter  
tier1.sources.source1.interceptors.i1.regex=\\{.*\\}  
tier1.sources.source1.interceptors.i2.type=timestamp  
tier1.channels.channel1.type=memory  
tier1.channels.channel1.capacity=10000  
tier1.channels.channel1.transactionCapacity=1000  
tier1.channels.channel1.keep-alive=30  
tier1.channels.channel2.type=memory  
tier1.channels.channel2.capacity=10000  
tier1.channels.channel2.transactionCapacity=1000  
tier1.channels.channel2.keep-alive=30  
tier1.sinks.sink1.type=hdfs  
tier1.sinks.sink1.channel=channel1  
tier1.sinks.sink1.hdfs.path=hdfs://master68:8020/flume/events/%y-%m-%d  
tier1.sinks.sink1.hdfs.round=true  
tier1.sinks.sink1.hdfs.roundValue=10  
tier1.sinks.sink1.hdfs.roundUnit=minute  
tier1.sinks.sink1.hdfs.fileType=DataStream  
tier1.sinks.sink1.hdfs.writeFormat=Text  
tier1.sinks.sink1.hdfs.rollInterval=0  
tier1.sinks.sink1.hdfs.rollSize=10240  
tier1.sinks.sink1.hdfs.rollCount=0  
tier1.sinks.sink1.hdfs.idleTimeout=60  
tier1.sinks.sink2.type=logger  
tier1.sinks.sink2.channel=channel2  
复制代码