support multi-interest for UserProcessor

BOLT use -Dcom.alipay.remoting.client.log.encode to control bolt framework log encoding, why not use file.encoding directly? Just like sofa-rpc log-conf.xml, so developers can use one configuration to control all logs encoding.

Your question

Processor默认线程池和RemotingProcessor线程池的设定需要在RpcServer.start()之后，代码如下：

RpcServer server = new RpcServer(8888);
server.start(); // 启动服务
server.registerDefaultExecutor(RpcProtocol.PROTOCOL_CODE, Executors.newCachedThreadPool());
server.registerDefaultExecutor(RpcProtocolV2.PROTOCOL_CODE, Executors.newCachedThreadPool());
server.registerProcessor(RpcProtocol.PROTOCOL_CODE, RpcCommandCode.RPC_REQUEST, new RpcRequestProcessor(Executors.newCachedThreadPool()));

如上这样使用，注册是需要发生在start()之后的，因为注册的代码需要用到ProccessorManager，在start()中的doInit()的时候，才会创建ProccessorManager。
如果注册必须发生在start()之后，那么如果在 RpcServer 启动之后并且在设置 registerDefaultExecutor 之前来了一个请求，请求的处理就放在原始的默认线程池去做了，与用户的初衷不同。

Your advice

是不是我的使用姿势不对？如果确实是这样使用的，那么建议 RpcServer 的 init() 和 start() 应该分开，提供类似于如下的代码，这样我们就可以在 init() 之后进行线程池的设置了，设置完成之后再 start()。

init() {
  doInit();
}
startOnly() {
  doStart();
}
start() {
  init();
  startOnly();
}

Environment

• SOFABolt version: 1.5.1
• JVM version (e.g. java -version):
• OS version (e.g. uname -a):
• Maven version:
• IDE version:

bug description

thx @YangLi for pointing out these problems:

this bug is a mis-use of api, we should check key not value existed and then print warn log to notify user.

impact analysis

actually this bug have no bad impact except some warning log missing

ProtocolCodeBasedDecoder
代码中没有校验报文的开始或结束符，万一报文有问题（例如报文错乱），那会不会一直导致误以为这连接对应的流还是可信的，一直在误读？

Is your feature request related to a problem? Please describe.

I want to implement the protocol in HTTP pipeline semantic, just like

https://github.com/spinscale/netty4-http-pipelining/blob/master/src/main/java/de/spinscale/netty/http/pipelining/HttpPipeliningHandler.java

But there is no way to add a such custom ChannelDuplexHandler to netty under bolt framework.

Describe the solution you'd like

1. Supports user custom netty ChannelDuplexHandler to be added.
2. Or a bolt pipeline handler to be extendted that wraps netty handler, do not export netty directly.

Describe alternatives you've considered
A clear and concise description of any alternative solutions or features you've considered.

Additional context
Add any other context or screenshots about the feature request here.

advice

As sofa-bolt build on top of netty-4.1.25.Final, and netty 4 support epoll nativelly.
For performance, bolt should support epoll per user-like.
So here I add an option transport.use.epoll so that user can determine whether enabled it or not under Linux system.

Environment

• SOFABOLT version: v1.4.2
• OS version: Linux Family

问题

Andorid端能否使用Bolt进行协议通讯

使用场景

长连接推送服务
实现细节:服务端使用Bolt作为远程通讯框架,Android通过TCP协议连接服务端,那么Android支持解析Bolt协议吗?

Environment

• SOFABolt version: 1.4.2
• JVM version (JDK 8):
• OS version (CentOS7):
• Maven version: Maven 3
• IDE version: IDEA

Describe the bug
Currently system settings for global switch take effect before user settings.
So there may be two instance of RpcClient in one java process, and the configs of the first instance setting by system property will change the default behavior of the second one.

 @Override
    public boolean isOn(int switchIndex) {
        return this.userSettings.get(switchIndex) || systemSettings.get(switchIndex);
    }

Expected behavior

User settings takes effect before system settings

Actual behavior

System settings takes effect before user settings

InvokeFuture应该支持泛型，以实现用户将Bolt作为一个基础通信框架使用，但是不使用Bolt协议的场景。
（不限制用户一定要使用ProtocolCode和RemotingCommand）

1. 用户无法指定自己的协议
   • ConnectionFactory无法使用用户的Encoder和Decoder，需要继承ConnectionFactory实现自己的ConnectionFactory
2. ConnectionFactory接口不应当包含registerUserProcessor的方法
   • ConnectionFactory应当只负责Connection相关的逻辑，比如Connection的创建
   • Processor是用户的处理逻辑，应该注册到Netty的handler中

• detail see: sofastack/sofa-rpc#190

Your question

1. 假设现在想增加一个自定义配置项 NETTY_TCP_NODELAY，由于 ConfigItem 是一个 enum 类，无法继承，这时候我们应该怎么添加一个自定义的的扩展项呢？

2. ConfigItem 和 Configs 类冗余，建议去掉 ConfigItem 类，
   

ConfigContainer中的第二个参数用 Configs 中的String 来填充。

• SOFABolt version: 1.5.1
• JVM version (e.g. java -version):
• OS version (e.g. uname -a):
• Maven version:
• IDE version:

Considering the compatibility with IDE, @yangl suggest to upgrade the format plugin

<groupId>net.revelc.code.formatter</groupId>
<artifactId>formatter-maven-plugin</artifactId>
<version>2.7.3</version>

@xmtsui ，能否在开源版本上支持协程？据了解阿里AliJDK协程是按照JKU即http://ssw.jku.at/General/Staff/LS/coro/做的修改，所以能否把这块在sofa-bolt开源版本按照这个做个修改支持协程呢？

@MarkLinHz 反馈license header的注释问题
mathieucarbou/license-maven-plugin#118

The method initRpcRemoting() of RpcServer previously need an arg RpcRemoting, actually it do not need this. We can just provide a default implementation, and change the visibility to protected, so the sub class and override this to create a specific RpcRemoting instance.

Upgrade to netty-4.1.25.Final, no new feature uesed, just be in keeping with SOFARPC

我使用dubbo做了一个简易的rpc测试，使用AbstractBatchDecoder代替ByteToMessageDecoder之后，在handler中判断List还是单个response。经过高并发的压测，发现始终有个别请求会响应失败，异常未知；地并发下很稳定。官方是否有提供过压测脚本，AbstractBatchDecoder到底能提升多少qps？以及是否存在响应失败的问题

Is your feature request related to a problem? Please describe.
Netty has used ByteBufAllocator as default

Describe the solution you'd like
change the default config in bolt to true

    public static final String NETTY_BUFFER_POOLED                   = "bolt.netty.buffer.pooled";
    public static final String NETTY_BUFFER_POOLED_DEFAULT           = "false";

: is an invalid character in resource name 'com.alipay.sofa:bolt'.
导入eclipse 报错

我看了 com.alipay.remoting.rpc.DefaultInvokeFuture 的相关代码，使用 CountDown 的代价是否有点大呢？如果使用 volatile 变量加上 LockSupport 是不是更轻量呢（貌似能够代替目前的功能？）？毕竟 AQS 还是挺重的。而且还需要实例化一个 CountDownLatch 对象。

1. RemotingServer抽象成接口比当前的抽象类要好（逻辑更加清晰）
2. 代码中有很多警告
3. 一些拼写错误

bug description

RemotingContext

this bug lead to an incorrect set of member variable

• this issue should be scanned out by pmd, but rule UnusedFormalParameter failed to discover this. And I have issued in pmd repo
• test case of this feature also have problems, need to be fixed also

impact analysis

this bug will cause timeout fail-fast switch do not work. If you have used this feature, should pay attention to our release note and do an upgrade.

我在win7下使用IntelliJ IDEA运行程序，发现只能打印部分log。
不能打印log的java文件中Logger定义如下：
private static final Logger logger = LoggerFactory.getLogger(XXX.class); //其中XXX为java类名。
能打印log的java文件中Logger定义如下：
private static final Logger logger = BoltLoggerFactory.getLogger("XXXXX");//其中XXX为log-conf.xml中定义的logger name。

请问：如果我需要看到所有的日志，该怎么做?

https://github.com/alipay/sofa-bolt/wiki/SOFA-Bolt-Handbook

2.3 建立多连接与连接预热
通常来说，点对点的直连通信，客户端和服务端，一个 IP 一个连接对象就够用了。不管是吞吐能力还是并发度，都能满足一般业务的通信需求。而有一些场景，比如不是点对点直连通信，而是经过了 LVS VIP，或者 F5 设备的连接，此时，为了负载均衡和容错，会针对一个 URL 地址建立多个连接。我们提供如下方式来建立多连接，即发起调用时传入的 URL 增加如下参数 127.0.0.1:12200?_CONNECTIONNUM=30&_CONNECTIONWARMUP=true，表示针对这个 IP 地址，需要建立30个连接，同时需要预热连接。其中预热与不预热的区别是：

预热：即第一次调用（比如 Sync 同步调用），就建立30个连接
不预热：每一次调用，创建一个连接，直到创建满30个连接

不预热应该是：第一次调用，先同步创建一个链接，然后开启异步线程池去做剩余的连接操作才对

callback异步调用方式，必须和 final CountDownLatch latch结合使用，否则就会出现
com.alipay.remoting.exception.ConnectionClosedException: Connection closed when invoke with callback.The address is 127.0.0.1:8999
感觉不是很方便，这有什么改进方法么？

@killme2008 point out some threadpools ( netty eventloop etc.) do not shutdown after client or server shutdown

In

com.alipay.remoting.rpc.RpcResponseResolver#preProcess

bolt only return custom errMsg, without real exception

 case SERVER_SERIAL_EXCEPTION:
                    msg = "Server serialize response exception! the address is " + addr + ", id="
                          + responseCommand.getId() + ", serverSide=true";
                    e = new SerializationException(msg, true);
                    break;
                case SERVER_DESERIAL_EXCEPTION:
                    msg = "Server deserialize request exception! the address is " + addr + ", id="
                          + responseCommand.getId() + ", serverSide=true";
                    e = new DeserializationException(msg, true);
                    break;

when client receive exception info ,they need to login to server to determine the real exception

com.alipay.remoting.RemotingServer

now RemotingServer won't throw exception to outside.

 /**
     * Start the server with ip and port.
     */
    public boolean start(String ip) {
        this.init();
        if (started.compareAndSet(false, true)) {
            try {
                logger.warn("Server started on " + ip + ":" + port);
                return this.doStart(ip);
            } catch (Throwable t) {
                started.set(false);
                logger.error("ERROR: Failed to start the Server!", t);
                return false;
            }
        } else {
            logger.error("ERROR: The server has already started!");
            return false;
        }
    }

so rpc framework can not get the accurate info about the error. we only can inform users to see bolt log like this

com.alipay.sofa.rpc.core.exception.SofaRpcRuntimeException: Failed to start bolt server, see more detail from bolt log.
	at com.alipay.sofa.rpc.server.bolt.BoltServer.start(BoltServer.java:112) ~[sofa-rpc-all-5.4.0.jar:5.4.0]
	at com.alipay.sofa.rpc.boot.container.ServerConfigContainer.startServers(ServerConfigContainer.java:78) ~[rpc-sofa-boot-starter-5.4.0.jar:5.4.0]
	at com.alipay.sofa.rpc.boot.context.SofaBootRpcStartListener.onApplicationEvent(SofaBootRpcStartListener.java:68) ~[rpc-sofa-boot-starter-5.4.0.jar:5.4.0]
	at com.alipay.sofa.rpc.boot.context.SofaBootRpcStartListener.onApplicationEvent(SofaBootRpcStartListener.java:38) ~[rpc-sofa-boot-starter-5.4.0.jar:5.4.0]
	at org.springframework.context.event.SimpleApplicationEventMulticaster.invokeListener(SimpleApplicationEventMulticaster.java:166) ~[spring-context-4.3.4.RELEASE.jar:4.3.4.RELEASE]
	at org.springframework.context.event.SimpleApplicationEventMulticaster.multicastEvent(SimpleApplicationEventMulticaster.java:138) ~[spring-context-4.3.4.RELEASE.jar:4.3.4.RELEASE]

For security reasons, please upgrade Hessian to 3.3.2.

https://github.com/alipay/sofa-bolt/blob/release/1.5.0/pom.xml#L69

Describe the bug
可以多次调用AbstractRemotingServer#stop，并没有抛出预期的IllegalStateException

Expected behavior

第一次调用AbstractRemotingServer#stop，执行stop逻辑
第二次调用AbstractRemotingServer#stop，抛出IllegalStateException

Actual behavior

第一次调用AbstractRemotingServer#stop，执行stop逻辑
第二次调用AbstractRemotingServer#stop，再次执行stop逻辑
第三次及之后调用AbstractRemotingServer#stop，抛出IllegalStateException

Steps to reproduce

执行RemotingServerTest#testStopRepeatedly的单元测试

Minimal yet complete reproducer code (or GitHub URL to code)

class AbstractRemotingServer
@Override
    public boolean stop() {
        // 逻辑或在前一个条件为true的情况下，不会之后后面的操作，所以第二次执行stop时started.compareAndSet(true, false)会返回true导致可以再次执行doStop
        if (inited.compareAndSet(true, false) || started.compareAndSet(true, false)) {
            return this.doStop();
        } else {
            throw new IllegalStateException("ERROR: The server has already stopped!");
        }
    }

Environment

• SOFABolt version: 1.5.0
• JVM version (e.g. java -version):
• OS version (e.g. uname -a):
• Maven version:
• IDE version:

Describe the bug
代码风格不符合规范，比如Scannable接口的方法被定义为public；RemotingCommand的方法缺少必要的注释等

Expected behavior

无

Actual behavior

无

Steps to reproduce

无

Minimal yet complete reproducer code (or GitHub URL to code)

Environment

• SOFABolt version:
• JVM version (e.g. java -version):
• OS version (e.g. uname -a):
• Maven version:
• IDE version:

#96

integrated plugins

• PMD
• formatter

refer

1. PMD usage
   • alibaba p3c rule set
   • java rule set
   • how to customized rule set priority
2. introduce common used plugins
3. formatter plugin upgrade: #32

RpcServerDemoByMain 程序，启动报
Sofa-Middleware-Log SLF4J : Actual binding is of type [ com.alipay.remoting Log4j ]
INFO - Sofa-Middleware-Log SLF4J : Actual binding is of type [ com.alipay.remoting Log4j ]
server start failed!

This Demo import BoltServer which not belong to sofa-bolt, it's a class in sofa-rpc.

@ujjboy 提出版本协商的需求，同时考虑运行时，客户端与服务端需要有控制类信息的交互。
可以统一定义一套控制命令的通信方式来支持此类需求。

无钩提出，需要类似http2的go away帧的指令需求

Describe the bug

1. RunStateRecordedFutureTask#getAfterRun方法可能会阻塞在super.get()上，如果RunStateRecordedFutureTask执行的Callable被阻塞
2. Callable实际执行的是ConnectionPoolCall，执行建连操作，可能会阻塞在获取netty的建连结果上
3. 那么可能出现ConnectionPoolCall再等待IO线程建连，而IO线程在等待ConnectionPoolCall的执行结果，出现死锁

Expected behavior

Actual behavior

• IO线程被阻塞在获取callable的结果上，建连的线程被阻塞在获取建连的结果上

Steps to reproduce

• 并发执行主动的建连和close操作，使channelInactive时拿到的task对象为后续建连产生的新的task对象，且这个对象的hasRun状态已经为true但是callable未执行

Minimal yet complete reproducer code (or GitHub URL to code)

Environment

• SOFABolt version: 1.4.x 1.5.x
• JVM version (e.g. java -version):
• OS version (e.g. uname -a):
• Maven version:
• IDE version:

现在线上的项目有使用socket协议, 我方位服务端或客户端的其中一端, 协议由双方共同协定, 在此框架上自定义私有协议, 包括 Command 类型、Command 处理器、编解码处理器 还是挺有难度的, 能否提供自定义协议的单元测试模块, 供参考, 谢谢!

problem

For now, the stop method of RemotingServer will throw IllegalStateException when do repeat call.
But the start method of RemotingServer do not throw IllegalStateException when do repeat call.

how to fix

So we change the logic for starting RemotingServer, if start failed, we will return false. But if start repeatedly, we will throw IllegalStateException to warn the server have been started already.

当前bolt打印日志都指定了log名称，因此没有加root logger。但 @ujjboy 反馈，编译过程有warning报出。因此做一下优化，把root logger统一加上。

Describe the bug

• RpcClient通过init方法启动，shutdown方法关闭
• RemotingServer通过start方法启动，stop方法关闭

为什么一个项目中生命周期相关的抽象不统一？

Expected behavior

提供统一的生命周期相关的接口抽象

Actual behavior

Steps to reproduce

Minimal yet complete reproducer code (or GitHub URL to code)

Environment

• SOFABolt version:
• JVM version (e.g. java -version):
• OS version (e.g. uname -a):
• Maven version:
• IDE version:

now we only support set netty buffer watermark using system property, some scenarios need to set by user class api.

@hongweiyi 反馈

4.1.x之后，netty默认开启了PooledByteBufAllocator，详见release note #PooledByteBufAllocator as the default allocator
bolt之前的系统开关已经无用。需要修改下，改成支持关闭的形式，客户端与server端都需要修改：
由：

     boolean pooledBuffer = SystemProperties.netty_buffer_pooled();
        if (pooledBuffer) {
            this.bootstrap.option(ChannelOption.ALLOCATOR, PooledByteBufAllocator.DEFAULT)
                .childOption(ChannelOption.ALLOCATOR, PooledByteBufAllocator.DEFAULT);
        }

改为：

     boolean pooledBuffer = SystemProperties.netty_buffer_pooled();
        if (!pooledBuffer) {
            this.bootstrap.option(ChannelOption.ALLOCATOR, UnpooledByteBufAllocator.DEFAULT)
                .childOption(ChannelOption.ALLOCATOR, UnpooledByteBufAllocator.DEFAULT);
        }

在release/1.5.0分支中定义了Codec接口并且就这个接口的增加修改了ConnectionFactory的部分代码，使得代码整体的解耦性提高了，但是在RpcServer中似乎并没有适用起来，详情见下图：
使用了Codec

未使用Codec

see RandomSelectStrategy#select

https://github.com/alipay/sofa-bolt/blob/2aabea02cacb7e134ac50cc1c965053fdb278bc7/src/main/java/com/alipay/remoting/RandomSelectStrategy.java#L78

Describe the bug
ConnectionFactory中workerGroup非全局静态，这样在创建多个Bolt客户单的场景中，会创建过多的线程。

Expected behavior

Actual behavior

Steps to reproduce

Minimal yet complete reproducer code (or GitHub URL to code)

Environment

• SOFABolt version:
• JVM version (e.g. java -version):
• OS version (e.g. uname -a):
• Maven version:
• IDE version:

http://netty.io/wiki/native-transports.html