Flume 实战(2)--Flume-ng-sdk源码分析

提示：扫一扫查出行
【扫一扫了解最新限行尾号】
复制提示

具体参考: 官方用户手册和开发指南

http://flume.apache.org/FlumeDeveloperGuide.html

*) 定位和简单例子

1). Flume-ng-sdk是用于编写往flume agent发送数据的client sdk
2). 简单示例

RpcClient client = null;
try {
client = RpcClientFactory.getDefaultInstance("127.0.0.1", 41414);
Event event = EventBuilder.withBody("hello flume", Charset.forName("UTF-8"));
client.append(event);
} catch (EventDeliveryException e) {
e.printStackTrace();
} finally {
if ( client != null ) {
client.close();
}
}

*) Event设计和类层次结构

1. Event类设计
在Flume中Event是个接口类

public interface Event {
public Map<String, String> getHeaders();
public void setHeaders(Map<String, String> headers);
public byte[] getBody();
public void setBody(byte[] body);
}

由代码可得, Event由Header集合和消息负载两部分构成.

2. Builder设计模式
在org.apache.flume.event下, 有两个Event的具体实现类: SimpleEvent, JSonEvent.
EventBuilder类顾名思义, 采用Builder的方式来组装对象的成员, 并产生最终的对象.

public class EventBuilder {public static Event withBody(byte[] body, Map<String, String> headers) {
Event event = new SimpleEvent();
if(body == null) {
body = new byte[0];
}
event.setBody(body);
if (headers != null) {
event.setHeaders(new HashMap<String, String>(headers));
}
return event;
}public static Event withBody(byte[] body) {
return withBody(body, null);
}public static Event withBody(String body, Charset charset,
Map<String, String> headers) {
return withBody(body.getBytes(charset), headers);
}public static Event withBody(String body, Charset charset) {
return withBody(body, charset, null);
}}

java的访问控制符: public/default/protected/private, default表示同package可见
不过另人意外的是, 其对应的SimpleEvent的构造函数的修饰符是public, 即不是default, 也不是protected, 这点让EventBuilder的引入有些失败.把Builder模式, 用到极致的是Google Protocol Buffer(java), 其每个PB对象, 都是用相应的Builder类来组装和生成. 采用这种Builder模式的用途是, 把一个对象元素的修改和读取彻底分离, 使得一个PB对象,从诞生后就是一个immutable对象, 只能读取其属性信息, 而不能修改其属性.

*) RpcClient设计和类层次结构

1. RpcClient的接口定义:

public interface RpcClient {
public int getBatchSize();
public void append(Event event) throws EventDeliveryException;
public void appendBatch(List<Event> events) throws EventDeliveryException;
public boolean isActive();
public void close() throws FlumeException;
}

2. AbstractRpcClient的抽象类定义:

public abstract class AbstractRpcClient implements RpcClient { protected int batchSize = RpcClientConfigurationConstants.DEFAULT_BATCH_SIZE;
protected long connectTimeout = RpcClientConfigurationConstants.DEFAULT_CONNECT_TIMEOUT_MILLIS;
protected long requestTimeout = RpcClientConfigurationConstants.DEFAULT_REQUEST_TIMEOUT_MILLIS; @Override
public int getBatchSize(){
return batchSize;
} protected abstract void configure(Properties properties) throws FlumeException;}

新增了一些常量定义, 和新的抽象函数configure(Properties prop);

3. RpcClient工厂类的使用
RpcClientFactory的定义

public class RpcClientFactory { public static RpcClient getInstance(Properties properties) throws FlumeException {
// 1). 获取具体rpcclient的类型信息
properties.getProperty(RpcClientConfigurationConstants.CONFIG_CLIENT_TYPE);
// 2). 利用反射,获取类的class
Class<? extends AbstractRpcClient> clazz = (Class<? extends AbstractRpcClient>) Class.forName(...);
// 3). 产生类对象
RpcClient client = clazz.newInstance();
// 4). 进行具体rpcclient实例的配置初始化
client.configure(properties);
// 5). 返回对象
return client;
}}

RpcClientFactory借助静态方法getInstance, 其依据Properties里的相应key/value来, 来产生不同的对象实例, 配置不同的属性. 同时RpcClient的具体实例, 其构造方法的访问限定符都是protected, 这一点做的, 比之前EventBuilder设计和实现要规范和清晰.

clazz = Class.forName(...);
client = class.newInstance();
client.configure(...);

是种非常好的实践代码, 把面向对象的多态性用到极致

4. 具体的RpcClient类的实现
其SDK提供了两大类, 具体的实现类ThriftRpcClient和AvroRpcClient
4.1. 对ThriftRpcClient的解读
4.1.1 thrift idl的定义
idl文件(src/main/thrift/flume.thrift)的定义

namespace java org.apache.flume.thriftstruct ThriftFlumeEvent {
1: required map <string, string> headers,
2: required binary body,
}enum Status {
OK,
FAILED,
ERROR,
UNKNOWN
}service ThriftSourceProtocol {
Status append(1: ThriftFlumeEvent event),
Status appendBatch(1: list<ThriftFlumeEvent> events),
}

分别对应源码包org.apache.flume.thrift下

Status, ThriftFlumeEvent, ThriftSourceProtocol类
4.1.2 ThriftRpcClient的实现
ThriftRpcClient并不是简单对ThriftSourceProtocol的客户端的简单封装

public class ThriftRpcClient extends AbstractRpcClient {
private ConnectionPoolManager connectionManager;
private final ExecutorService callTimeoutPool;
private final AtomicLong threadCounter;
// ......
}

评注: 粗略观察其类成员, 其借助线程池(ExecutorService)和连接池(ConnectionManager)管理, 来实现RpcClient的发送接口, 这样append(), appendBatch()的接口都是线程安全的, 该客户端的实例能用于多线程并发使用.

AvroRpcClient代码结构差不多, 先一笔带过.

5. 两个重要的实现类
FailOverRpcClient的源码解析:
这边采用装饰模式(Decorator Pattern), FailOverRpcClient继承自RpcClient, 同时又拥有实际的RpcClient实例, 只是在实际RpcClient基础上, 添加了失败后重试的能力.

FailOver是失败后重试的机制, 通常借助带尝试次数的重试来实现
其append(Event e)方法中:

int tries = 0;
while (tries < maxTries) {
try {
tries++;
localClient = getClient();
localClient.append(event);
return;
} catch (EventDeliveryException e) {
localClient.close();
localClient = null;
} catch (Exception e2) {
throw new EventDeliveryException(
"Failed to send event. Exception follows: ", e2);
}
}

这段代码采用相对简单的.

getNextClient()的实现如下:

for (int count = lastCheckedhost + 1; count < limit; count++) {
HostInfo hostInfo = hosts.get(count);
try {
setDefaultProperties(hostInfo, props);
localClient = RpcClientFactory.getInstance(props);
lastCheckedhost = count;
return localClient;
} catch (FlumeException e) {
logger.info("Could not connect to " + hostInfo, e);
continue;
}
}
for(int count = 0; count <= lastCheckedhost; count++) {
HostInfo hostInfo = hosts.get(count);
try {
setDefaultProperties(hostInfo, props);
localClient = RpcClientFactory.getInstance(props);
lastCheckedhost = count;
return localClient;
} catch (FlumeException e) {
logger.info("Could not connect to " + hostInfo, e);
continue;
}
}

HostInfo封装了一个远端服务的ip地址
FailOver简单的轮询了各个服务地址.

LoadBalancingRpcClient的源码解析:
LoadBalancingRpcClient顾名思义, 采用负载均衡的策略来实现, 其还是采用遍历(轮询/随机)+反馈的机制, 来动态的调整服务列表的候选顺序.
在append(Event)方法中:

Iterator<HostInfo> it = selector.createHostIterator();
while (it.hasNext()) {
HostInfo host = it.next();
try {
RpcClient client = getClient(host);
client.append(event);
eventSent = true;
break;
} catch (Exception ex) {
selector.informFailure(host);
LOGGER.warn("Failed to send event to host " + host, ex);
}
}
if (!eventSent) {
throw new EventDeliveryException("Unable to send event to any host");
}

selector.createHostIterator() 创建当前服务候选列表的一个快照, 同时递进一个轮询单元.
selector.informFailure(host) 是对失败的服务进行降级处理

而HostSelector接口定义如下:

public interface HostSelector {
void setHosts(List<HostInfo> hosts);
Iterator<HostInfo> createHostIterator();
void informFailure(HostInfo failedHost);
}

其具体实现类

#). RoundRobinHostSelector, 借助轮询的方式来实现
#). RandomOrderHostSelector, 借助随机的方式来实现
这两个类, 都是借助OrderSelector<T>的实现类来实现, OrderSelector封装了对错误服务机器列表的屏蔽策略
该屏蔽策略如下所示:
失败一次, 设置一个恢复时间点, 未到该恢复时间点, 则不允许获取该机器ip/port
同时为了惩罚多次失败, 减少获取该服务机器的ip/port, 采用1000 * (1 << sequentialFails), 连续失败次数, 其恢复时间的间隔要加大.

*) Properties的属性配置
基本的属性配置

client.type = default (for avro) or thrift (for thrift)
hosts = h1 # default client accepts only 1 host
hosts.h1 = host1.example.org:41414 # host and port must both be specified
batch-size = 100 # Must be >=1 (default: 100)
connect-timeout = 20000 # Must be >=1000 (default: 20000)
request-timeout = 20000 # Must be >=1000 (default: 20000)

FailOver支持的配置

client.type = default_failover
hosts = h1 h2 h3 # at least one is required, but 2 or more makes better sense
max-attempts = 3 # Must be >=0 (default: number of hosts

Balancing支持的配置

client.type = default_loadbalance
hosts = h1 h2 h3 # At least 2 hosts are required
backoff = false # Specifies whether the client should back-off from a failed host
maxBackoff = 0 # Max timeout in millis
host-selector = round_robin # The host selection strategy used

*) 异常类定义

EventDeliveryException和FlumeException