1458 lines
49 KiB
Markdown
1458 lines
49 KiB
Markdown
# 精尽 Netty 源码解析 —— ChannelHandler(五)之 IdleStateHandler
|
||
|
||
# 1. 概述
|
||
|
||
在 `netty-handler` 模块的 `timeout` 包,实现 Channel 的读写操作的**空闲**检测。可能有胖友不太了解空闲检测的具体用途。请先研读理解下 [《简易RPC框架-心跳与重连机制》](https://www.cnblogs.com/ASPNET2008/p/7615973.html) 。
|
||
|
||
# 2. 类
|
||
|
||
`timeout` 包,包含的类,如下图所示:[之 IdleStateHandler.assets/01.png)](http://static.iocoder.cn/images/Netty/2018_10_13/01.png)`timeout` 包
|
||
|
||
一共有 3 个 ChannelHandler 实现类:
|
||
|
||
- IdleStateHandler ,当 Channel 的
|
||
|
||
读或者写
|
||
|
||
空闲时间太长时,将会触发一个 IdleStateEvent 事件。然后,你可以自定义一个 ChannelInboundHandler ,重写
|
||
|
||
|
||
|
||
```
|
||
#userEventTriggered(ChannelHandlerContext ctx, Object evt)
|
||
```
|
||
|
||
|
||
|
||
方法,处理该事件。
|
||
|
||
- ReadTimeoutHandler ,继承 IdleStateHandler 类,当 Channel 的**读**空闲时间( 读或者写 )太长时,抛出 ReadTimeoutException 异常,并自动关闭该 Channel 。然后,你可以自定一个 ChannelInboundHandler ,重写 `#exceptionCaught(ChannelHandlerContext ctx, Throwable cause)` 方法,处理该异常。
|
||
|
||
- WriteTimeoutHandler ,当一个**写**操作不能在指定时间内完成时,抛出 WriteTimeoutException 异常,并自动关闭对应 Channel 。然后,你可以自定一个 ChannelInboundHandler ,重写 `#exceptionCaught(ChannelHandlerContext ctx, Throwable cause)` 方法,处理该异常。
|
||
|
||
😈 从 WriteTimeoutHandler 可以看出,本文实际不仅仅分享 IdleStateHandler ,更准确的是分享 Timeout 相关的 ChannelHandler 。考虑到大多数胖友对 IdleStateHandler 比较熟悉,也相对常用,所以标题才取了 [《精尽 Netty 源码解析 —— ChannelHandler(五)之 IdleStateHandler》](https://svip.iocoder.cn/Netty/ChannelHandler-5-idle/#) 。
|
||
|
||
# 3. IdleState
|
||
|
||
`io.netty.handler.timeout.IdleState` ,空闲状态**枚举**。代码如下:
|
||
|
||
```
|
||
/**
|
||
* 空闲状态枚举
|
||
*
|
||
* An {@link Enum} that represents the idle state of a {@link Channel}.
|
||
*/
|
||
public enum IdleState {
|
||
|
||
/**
|
||
* No data was received for a while.
|
||
*
|
||
* 读空闲
|
||
*/
|
||
READER_IDLE,
|
||
/**
|
||
* No data was sent for a while.
|
||
*
|
||
* 写空闲
|
||
*/
|
||
WRITER_IDLE,
|
||
/**
|
||
* No data was either received or sent for a while.
|
||
*
|
||
* 读或写任一空闲
|
||
*/
|
||
ALL_IDLE
|
||
|
||
}
|
||
```
|
||
|
||
- 一共有 3 种状态。其中,`ALL_IDLE` 表示的是,读**或**写任一空闲,注意是“或”。
|
||
|
||
## 3.1 IdleStateEvent
|
||
|
||
`io.netty.handler.timeout.IdleStateEvent` ,空闲事件。代码如下:
|
||
|
||
```
|
||
public class IdleStateEvent {
|
||
|
||
// READ
|
||
public static final IdleStateEvent FIRST_READER_IDLE_STATE_EVENT = new IdleStateEvent(IdleState.READER_IDLE, true); // 首次
|
||
public static final IdleStateEvent READER_IDLE_STATE_EVENT = new IdleStateEvent(IdleState.READER_IDLE, false);
|
||
// WRITE
|
||
public static final IdleStateEvent FIRST_WRITER_IDLE_STATE_EVENT = new IdleStateEvent(IdleState.WRITER_IDLE, true); // 首次
|
||
public static final IdleStateEvent WRITER_IDLE_STATE_EVENT = new IdleStateEvent(IdleState.WRITER_IDLE, false);
|
||
// ALL
|
||
public static final IdleStateEvent FIRST_ALL_IDLE_STATE_EVENT = new IdleStateEvent(IdleState.ALL_IDLE, true); // 首次
|
||
public static final IdleStateEvent ALL_IDLE_STATE_EVENT = new IdleStateEvent(IdleState.ALL_IDLE, false);
|
||
|
||
/**
|
||
* 空闲状态类型
|
||
*/
|
||
private final IdleState state;
|
||
/**
|
||
* 是否首次
|
||
*/
|
||
private final boolean first;
|
||
|
||
/**
|
||
* Constructor for sub-classes.
|
||
*
|
||
* @param state the {@link IdleStateEvent} which triggered the event.
|
||
* @param first {@code true} if its the first idle event for the {@link IdleStateEvent}.
|
||
*/
|
||
protected IdleStateEvent(IdleState state, boolean first) {
|
||
this.state = ObjectUtil.checkNotNull(state, "state");
|
||
this.first = first;
|
||
}
|
||
|
||
/**
|
||
* Returns the idle state.
|
||
*/
|
||
public IdleState state() {
|
||
return state;
|
||
}
|
||
|
||
/**
|
||
* Returns {@code true} if this was the first event for the {@link IdleState}
|
||
*/
|
||
public boolean isFirst() {
|
||
return first;
|
||
}
|
||
|
||
}
|
||
```
|
||
|
||
- 3 **类**( `state` )空闲事件,再组合上是否首次( `first` ),一共有 6 种空闲事件。
|
||
|
||
# 4. TimeoutException
|
||
|
||
`io.netty.handler.timeout.TimeoutException` ,继承 ChannelException 类,超时异常。代码如下:
|
||
|
||
```
|
||
public class TimeoutException extends ChannelException {
|
||
|
||
TimeoutException() { }
|
||
|
||
@Override
|
||
public Throwable fillInStackTrace() {
|
||
return this;
|
||
}
|
||
|
||
}
|
||
```
|
||
|
||
## 4.1 ReadTimeoutException
|
||
|
||
`io.netty.handler.timeout.ReadTimeoutException` ,继承 TimeoutException 类,读超时( 空闲 )异常。代码如下:
|
||
|
||
```
|
||
public final class ReadTimeoutException extends TimeoutException {
|
||
|
||
/**
|
||
* 单例
|
||
*/
|
||
public static final ReadTimeoutException INSTANCE = new ReadTimeoutException();
|
||
|
||
private ReadTimeoutException() { }
|
||
|
||
}
|
||
```
|
||
|
||
## 4.2 WriteTimeoutException
|
||
|
||
`io.netty.handler.timeout.WriteTimeoutException` ,继承 TimeoutException 类,写超时( 空闲 )异常。代码如下:
|
||
|
||
```
|
||
public final class WriteTimeoutException extends TimeoutException {
|
||
|
||
/**
|
||
* 单例
|
||
*/
|
||
public static final WriteTimeoutException INSTANCE = new WriteTimeoutException();
|
||
|
||
private WriteTimeoutException() { }
|
||
|
||
}
|
||
```
|
||
|
||
# 5. IdleStateHandler
|
||
|
||
`io.netty.handler.timeout.IdleStateHandler` ,继承 ChannelDuplexHandler 类,当 Channel 的**读或者写**空闲时间太长时,将会触发一个 IdleStateEvent 事件。
|
||
|
||
## 5.1 构造方法
|
||
|
||
> 老艿艿:高能预警,IdleStateHandler 的属性有点点多。
|
||
|
||
```
|
||
/**
|
||
* 最小的超时时间,单位:纳秒
|
||
*/
|
||
private static final long MIN_TIMEOUT_NANOS = TimeUnit.MILLISECONDS.toNanos(1);
|
||
|
||
/**
|
||
* 写入任务监听器
|
||
*/
|
||
// Not create a new ChannelFutureListener per write operation to reduce GC pressure.
|
||
private final ChannelFutureListener writeListener = new ChannelFutureListener() {
|
||
|
||
@Override
|
||
public void operationComplete(ChannelFuture future) throws Exception {
|
||
// 记录最后写时间
|
||
lastWriteTime = ticksInNanos();
|
||
// 重置 firstWriterIdleEvent 和 firstAllIdleEvent 为 true
|
||
firstWriterIdleEvent = firstAllIdleEvent = true;
|
||
}
|
||
|
||
};
|
||
|
||
/**
|
||
* 是否观察 {@link ChannelOutboundBuffer} 写入队列
|
||
*/
|
||
private final boolean observeOutput;
|
||
/**
|
||
* 配置的读空闲时间,单位:纳秒
|
||
*/
|
||
private final long readerIdleTimeNanos;
|
||
/**
|
||
* 配置的写空闲时间,单位:纳秒
|
||
*/
|
||
private final long writerIdleTimeNanos;
|
||
/**
|
||
* 配置的All( 读或写任一 ),单位:纳秒
|
||
*/
|
||
private final long allIdleTimeNanos;
|
||
|
||
/**
|
||
* 读空闲的定时检测任务
|
||
*/
|
||
private ScheduledFuture<?> readerIdleTimeout;
|
||
/**
|
||
* 最后读时间
|
||
*/
|
||
private long lastReadTime;
|
||
/**
|
||
* 是否首次读空闲
|
||
*/
|
||
private boolean firstReaderIdleEvent = true;
|
||
|
||
/**
|
||
* 写空闲的定时检测任务
|
||
*/
|
||
private ScheduledFuture<?> writerIdleTimeout;
|
||
/**
|
||
* 最后写时间
|
||
*/
|
||
private long lastWriteTime;
|
||
/**
|
||
* 是否首次写空闲
|
||
*/
|
||
private boolean firstWriterIdleEvent = true;
|
||
|
||
/**
|
||
* All 空闲时间,单位:纳秒
|
||
*/
|
||
private ScheduledFuture<?> allIdleTimeout;
|
||
/**
|
||
* 是否首次 All 空闲
|
||
*/
|
||
private boolean firstAllIdleEvent = true;
|
||
|
||
/**
|
||
* 状态
|
||
*
|
||
* 0 - none ,未初始化
|
||
* 1 - initialized ,已经初始化
|
||
* 2 - destroyed ,已经销毁
|
||
*/
|
||
private byte state; // 0 - none, 1 - initialized, 2 - destroyed
|
||
/**
|
||
* 是否正在读取
|
||
*/
|
||
private boolean reading;
|
||
|
||
/**
|
||
* 最后检测到 {@link ChannelOutboundBuffer} 发生变化的时间
|
||
*/
|
||
private long lastChangeCheckTimeStamp;
|
||
/**
|
||
* 第一条准备 flash 到对端的消息( {@link ChannelOutboundBuffer#current()} )的 HashCode
|
||
*/
|
||
private int lastMessageHashCode;
|
||
/**
|
||
* 总共等待 flush 到对端的内存大小( {@link ChannelOutboundBuffer#totalPendingWriteBytes()} )
|
||
*/
|
||
private long lastPendingWriteBytes;
|
||
|
||
public IdleStateHandler(int readerIdleTimeSeconds, int writerIdleTimeSeconds, int allIdleTimeSeconds) {
|
||
this(readerIdleTimeSeconds, writerIdleTimeSeconds, allIdleTimeSeconds, TimeUnit.SECONDS);
|
||
}
|
||
|
||
public IdleStateHandler(long readerIdleTime, long writerIdleTime, long allIdleTime, TimeUnit unit) {
|
||
this(false, readerIdleTime, writerIdleTime, allIdleTime, unit);
|
||
}
|
||
|
||
/**
|
||
* Creates a new instance firing {@link IdleStateEvent}s.
|
||
*
|
||
* @param observeOutput
|
||
* whether or not the consumption of {@code bytes} should be taken into
|
||
* consideration when assessing write idleness. The default is {@code false}.
|
||
* @param readerIdleTime
|
||
* an {@link IdleStateEvent} whose state is {@link IdleState#READER_IDLE}
|
||
* will be triggered when no read was performed for the specified
|
||
* period of time. Specify {@code 0} to disable.
|
||
* @param writerIdleTime
|
||
* an {@link IdleStateEvent} whose state is {@link IdleState#WRITER_IDLE}
|
||
* will be triggered when no write was performed for the specified
|
||
* period of time. Specify {@code 0} to disable.
|
||
* @param allIdleTime
|
||
* an {@link IdleStateEvent} whose state is {@link IdleState#ALL_IDLE}
|
||
* will be triggered when neither read nor write was performed for
|
||
* the specified period of time. Specify {@code 0} to disable.
|
||
* @param unit
|
||
* the {@link TimeUnit} of {@code readerIdleTime},
|
||
* {@code writeIdleTime}, and {@code allIdleTime}
|
||
*/
|
||
public IdleStateHandler(boolean observeOutput, long readerIdleTime, long writerIdleTime, long allIdleTime, TimeUnit unit) {
|
||
if (unit == null) {
|
||
throw new NullPointerException("unit");
|
||
}
|
||
|
||
this.observeOutput = observeOutput;
|
||
|
||
if (readerIdleTime <= 0) {
|
||
readerIdleTimeNanos = 0;
|
||
} else {
|
||
readerIdleTimeNanos = Math.max(unit.toNanos(readerIdleTime), MIN_TIMEOUT_NANOS); // 保证大于等于 MIN_TIMEOUT_NANOS
|
||
}
|
||
if (writerIdleTime <= 0) {
|
||
writerIdleTimeNanos = 0;
|
||
} else {
|
||
writerIdleTimeNanos = Math.max(unit.toNanos(writerIdleTime), MIN_TIMEOUT_NANOS); // 保证大于等于 MIN_TIMEOUT_NANOS
|
||
}
|
||
if (allIdleTime <= 0) {
|
||
allIdleTimeNanos = 0;
|
||
} else {
|
||
allIdleTimeNanos = Math.max(unit.toNanos(allIdleTime), MIN_TIMEOUT_NANOS); // 保证大于等于 MIN_TIMEOUT_NANOS
|
||
}
|
||
}
|
||
```
|
||
|
||
- 属性比较多,保持耐心和淡定,我们继续来整理一波。
|
||
|
||
- `MIN_TIMEOUT_NANOS` 静态属性,最小的超时时间为 **1** ,单位:纳秒。因为 IdleStateHandler 创建的,检测定时任务的时间,以纳秒为单位。
|
||
|
||
- `state` 属性,IdleStateHandler 的状态。一共有三种,见注释。
|
||
|
||
- Read 空闲相关属性
|
||
|
||
- `readerIdleTimeNanos` 属性,配置的读空闲时间,单位:纳秒。
|
||
- `readerIdleTimeout` 属性,读空闲的定时检测任务。
|
||
- `lastReadTime` 属性,读空闲的定时检测任务。
|
||
- `firstReaderIdleEvent` 属性,是否首次读空闲。
|
||
- 【**独有**】 `reading` 属性,是否正在读取。
|
||
|
||
- Write 空闲相关属性
|
||
|
||
- `writerIdleTimeNanos` 属性,配置的写空闲时间,单位:纳秒。
|
||
|
||
- `writerIdleTimeout` 属性,写空闲的定时检测任务。
|
||
|
||
- `lastWriteTime` 属性,最后写时间。
|
||
|
||
- `writeListener` 属性,写入操作,完成 flush 到对端的回调监听器。初始时,创建好,避免重复创建,从而减轻 GC 压力。
|
||
|
||
- 【
|
||
|
||
独有
|
||
|
||
】ChannelOutboundBuffer 相关属性
|
||
|
||
- `observeOutput` 属性, 是否观察 ChannelOutboundBuffer 写入队列。
|
||
- `lastChangeCheckTimeStamp` 属性,最后检测到 ChannelOutboundBuffer 发生变化的时间。
|
||
- `lastMessageHashCode` 属性,第一条准备 flash 到对端的消息的 HashCode 。
|
||
- `lastPendingWriteBytes` 属性,总共等待 flush 到对端的内存大小。
|
||
- 关于这几个属性,跟着 [「5.7 hasOutputChanged」](https://svip.iocoder.cn/Netty/ChannelHandler-5-idle/#) 一起理解。
|
||
|
||
- ALL 空闲相关属性
|
||
|
||
- 因为 ALL 是 Write 和 Read 任一,所以共用它们的一些属性。
|
||
- `allIdleTimeNanos` 属性,配置的All( 读或写任一 ),单位:纳秒。
|
||
|
||
## 5.2 initialize
|
||
|
||
`#initialize(ChannelHandlerContext ctx)` 方法,初始化 IdleStateHandler 。代码如下:
|
||
|
||
```
|
||
1: private void initialize(ChannelHandlerContext ctx) {
|
||
2: // 校验状态,避免因为 `#destroy()` 方法在 `#initialize(ChannelHandlerContext ctx)` 方法,执行之前
|
||
3: // Avoid the case where destroy() is called before scheduling timeouts.
|
||
4: // See: https://github.com/netty/netty/issues/143
|
||
5: switch (state) {
|
||
6: case 1:
|
||
7: case 2:
|
||
8: return;
|
||
9: }
|
||
10:
|
||
11: // 标记为已初始化
|
||
12: state = 1;
|
||
13: // 初始化 ChannelOutboundBuffer 相关属性
|
||
14: initOutputChanged(ctx);
|
||
15:
|
||
16: // 初始相应的定时任务
|
||
17: lastReadTime = lastWriteTime = ticksInNanos();
|
||
18: if (readerIdleTimeNanos > 0) {
|
||
19: readerIdleTimeout = schedule(ctx, new ReaderIdleTimeoutTask(ctx), readerIdleTimeNanos, TimeUnit.NANOSECONDS);
|
||
20: }
|
||
21: if (writerIdleTimeNanos > 0) {
|
||
22: writerIdleTimeout = schedule(ctx, new WriterIdleTimeoutTask(ctx), writerIdleTimeNanos, TimeUnit.NANOSECONDS);
|
||
23: }
|
||
24: if (allIdleTimeNanos > 0) {
|
||
25: allIdleTimeout = schedule(ctx, new AllIdleTimeoutTask(ctx), allIdleTimeNanos, TimeUnit.NANOSECONDS);
|
||
26: }
|
||
27: }
|
||
```
|
||
|
||
- 第 2 至 9 行:校验状态,避免因为 `#destroy()` 方法在 `#initialize(ChannelHandlerContext ctx)` 方法,执行之前。
|
||
|
||
- 第 12 行:标记 `state` 为已初始化。
|
||
|
||
- 第 14 行:调用 `#initOutputChanged(ChannelHandlerContext ctx)` 方法,初始化 ChannelOutboundBuffer 相关属性。代码如下:
|
||
|
||
```
|
||
private void initOutputChanged(ChannelHandlerContext ctx) {
|
||
if (observeOutput) {
|
||
Channel channel = ctx.channel();
|
||
Unsafe unsafe = channel.unsafe();
|
||
ChannelOutboundBuffer buf = unsafe.outboundBuffer();
|
||
|
||
if (buf != null) {
|
||
// 记录第一条准备 flash 到对端的消息的 HashCode
|
||
lastMessageHashCode = System.identityHashCode(buf.current());
|
||
// 记录总共等待 flush 到对端的内存大小
|
||
lastPendingWriteBytes = buf.totalPendingWriteBytes();
|
||
}
|
||
}
|
||
}
|
||
```
|
||
|
||
- 初始化 `lastMessageHashCode` 和 `lastPendingWriteBytes` 属性。
|
||
|
||
- 第 17 至 26 行:根据配置,分别调用 `#schedule(hannelHandlerContext ctx, Runnable task, long delay, TimeUnit unit)` 方法,初始相应的定时任务。代码如下:
|
||
|
||
```
|
||
ScheduledFuture<?> schedule(ChannelHandlerContext ctx, Runnable task, long delay, TimeUnit unit) {
|
||
return ctx.executor().schedule(task, delay, unit);
|
||
}
|
||
```
|
||
|
||
- 一共有 ReaderIdleTimeoutTask、WriterIdleTimeoutTask、AllIdleTimeoutTask 三种任务,下文我们详细解析。
|
||
|
||
------
|
||
|
||
该方法,会在多个 Channel **事件**中被调用。代码如下:
|
||
|
||
```
|
||
// <2>
|
||
@Override
|
||
public void handlerAdded(ChannelHandlerContext ctx) throws Exception {
|
||
if (ctx.channel().isActive() && ctx.channel().isRegistered()) {
|
||
// 初始化
|
||
// channelActive() event has been fired already, which means this.channelActive() will
|
||
// not be invoked. We have to initialize here instead.
|
||
initialize(ctx);
|
||
} else {
|
||
// channelActive() event has not been fired yet. this.channelActive() will be invoked
|
||
// and initialization will occur there.
|
||
}
|
||
}
|
||
|
||
// <3>
|
||
@Override
|
||
public void channelRegistered(ChannelHandlerContext ctx) throws Exception {
|
||
// 初始化
|
||
// Initialize early if channel is active already.
|
||
if (ctx.channel().isActive()) {
|
||
initialize(ctx);
|
||
}
|
||
// 继续传播 Channel Registered 事件到下一个节点
|
||
super.channelRegistered(ctx);
|
||
}
|
||
|
||
// <1>
|
||
@Override
|
||
public void channelActive(ChannelHandlerContext ctx) throws Exception {
|
||
// 初始化
|
||
// This method will be invoked only if this handler was added
|
||
// before channelActive() event is fired. If a user adds this handler
|
||
// after the channelActive() event, initialize() will be called by beforeAdd().
|
||
initialize(ctx);
|
||
// 继续传播 Channel Registered 事件到下一个节点
|
||
super.channelActive(ctx);
|
||
}
|
||
```
|
||
|
||
- `<1>` :当客户端与服务端成功建立连接后,Channel 被激活,此时 channelActive 方法,的初始化被调用。
|
||
- `<2>` :当 Channel 被激活后,动态添加此 Handler ,则 handlerAdded 方法的初始化被调用。
|
||
- `<3>` :当 Channel 被激活后,用户主动切换 Channel 的所在的 EventLoop ,则 channelRegistered 方法的初始化被调用。
|
||
|
||
## 5.3 destroy
|
||
|
||
`#destroy()` 方法,销毁 IdleStateHandler 。代码如下:
|
||
|
||
```
|
||
private void destroy() {
|
||
// 标记为销毁
|
||
state = 2;
|
||
|
||
// 销毁相应的定时任务
|
||
if (readerIdleTimeout != null) {
|
||
readerIdleTimeout.cancel(false);
|
||
readerIdleTimeout = null;
|
||
}
|
||
if (writerIdleTimeout != null) {
|
||
writerIdleTimeout.cancel(false);
|
||
writerIdleTimeout = null;
|
||
}
|
||
if (allIdleTimeout != null) {
|
||
allIdleTimeout.cancel(false);
|
||
allIdleTimeout = null;
|
||
}
|
||
}
|
||
```
|
||
|
||
- 标记 `state` 为已销毁。
|
||
- 销毁响应的定时任务。
|
||
|
||
------
|
||
|
||
该方法,会在多个 Channel **事件**中被调用。代码如下:
|
||
|
||
```
|
||
@Override
|
||
public void handlerRemoved(ChannelHandlerContext ctx) throws Exception {
|
||
// 销毁
|
||
destroy();
|
||
}
|
||
|
||
@Override
|
||
public void channelInactive(ChannelHandlerContext ctx) throws Exception {
|
||
// 销毁
|
||
destroy();
|
||
// 继续传播 Channel Incative 事件到下一个节点
|
||
super.channelInactive(ctx);
|
||
}
|
||
```
|
||
|
||
## 5.4 channelIdle
|
||
|
||
在定时任务中,如果检测到**空闲**:
|
||
|
||
① 首先,调用 `#newIdleStateEvent(IdleState state, boolean first)` 方法,创建对应的空闲事件。代码如下:
|
||
|
||
```
|
||
protected IdleStateEvent newIdleStateEvent(IdleState state, boolean first) {
|
||
switch (state) {
|
||
case ALL_IDLE:
|
||
return first ? IdleStateEvent.FIRST_ALL_IDLE_STATE_EVENT : IdleStateEvent.ALL_IDLE_STATE_EVENT;
|
||
case READER_IDLE:
|
||
return first ? IdleStateEvent.FIRST_READER_IDLE_STATE_EVENT : IdleStateEvent.READER_IDLE_STATE_EVENT;
|
||
case WRITER_IDLE:
|
||
return first ? IdleStateEvent.FIRST_WRITER_IDLE_STATE_EVENT : IdleStateEvent.WRITER_IDLE_STATE_EVENT;
|
||
default:
|
||
throw new IllegalArgumentException("Unhandled: state=" + state + ", first=" + first);
|
||
}
|
||
}
|
||
```
|
||
|
||
② 然后,调用 `#channelIdle(ChannelHandlerContext ctx, IdleStateEvent evt)` 方法,在 pipeline 中,触发 UserEvent 事件。代码如下:
|
||
|
||
```
|
||
/**
|
||
* Is called when an {@link IdleStateEvent} should be fired. This implementation calls
|
||
* {@link ChannelHandlerContext#fireUserEventTriggered(Object)}.
|
||
*/
|
||
protected void channelIdle(ChannelHandlerContext ctx, IdleStateEvent evt) throws Exception {
|
||
ctx.fireUserEventTriggered(evt);
|
||
}
|
||
```
|
||
|
||
## 5.5 channelRead
|
||
|
||
`#channelRead(ChannelHandlerContext ctx, Object msg)` 方法,代码如下:
|
||
|
||
```
|
||
@Override
|
||
public void channelRead(ChannelHandlerContext ctx, Object msg) throws Exception {
|
||
// 开启了 read 或 all 的空闲检测
|
||
if (readerIdleTimeNanos > 0 || allIdleTimeNanos > 0) {
|
||
// 标记正在读取
|
||
reading = true;
|
||
// 重置 firstWriterIdleEvent 和 firstAllIdleEvent 为 true
|
||
firstReaderIdleEvent = firstAllIdleEvent = true;
|
||
}
|
||
// 继续传播 Channel Read 事件到下一个节点
|
||
ctx.fireChannelRead(msg);
|
||
}
|
||
```
|
||
|
||
在开启 read 或 all 的空闲检测的情况下,在【继续传播 Channel Read 事件到下一个节点】**之前**,会:
|
||
|
||
- 标记 `reading` 为正在读取。
|
||
- 重置 `firstWriterIdleEvent` 和 `firstAllIdleEvent` 为 `true` ,即又变成**首次**。
|
||
|
||
------
|
||
|
||
那么什么时候记录 `lastReadTime` 最后读取时间呢?答案在 `#channelReadComplete(ChannelHandlerContext ctx)` 方法中。代码如下:
|
||
|
||
```
|
||
@Override
|
||
public void channelReadComplete(ChannelHandlerContext ctx) throws Exception {
|
||
// 开启了 read 或 all 的空闲检测
|
||
if ((readerIdleTimeNanos > 0 || allIdleTimeNanos > 0) && reading) {
|
||
// 记录最后读时间
|
||
lastReadTime = ticksInNanos();
|
||
// 标记不在读取
|
||
reading = false;
|
||
}
|
||
// 继续传播 Channel ReadComplete 事件到下一个节点
|
||
ctx.fireChannelReadComplete();
|
||
}
|
||
```
|
||
|
||
在开启 read 或 all 的空闲检测的情况下,在【继续传播 Channel ReadComplete 事件到下一个节点】**之前**,会:
|
||
|
||
- 记录 `lastReadTime` 最后读取时间为 `#ticksInNanos()` 方法,代码如下:
|
||
|
||
```
|
||
long ticksInNanos() {
|
||
return System.nanoTime();
|
||
}
|
||
```
|
||
|
||
- 当前时间,单位:纳秒。
|
||
|
||
- 标记 `reading` 为不在读取。
|
||
|
||
## 5.6 write
|
||
|
||
`#write(ChannelHandlerContext ctx, Object msg, ChannelPromise promise)` 方法,代码如下:
|
||
|
||
```
|
||
@Override
|
||
public void write(ChannelHandlerContext ctx, Object msg, ChannelPromise promise) throws Exception {
|
||
// 开启了 write 或 all 的空闲检测
|
||
// Allow writing with void promise if handler is only configured for read timeout events.
|
||
if (writerIdleTimeNanos > 0 || allIdleTimeNanos > 0) {
|
||
// 写入,并添加写入监听器
|
||
ctx.write(msg, promise.unvoid()).addListener(writeListener);
|
||
} else {
|
||
// 写入,不添加监听器
|
||
ctx.write(msg, promise);
|
||
}
|
||
}
|
||
```
|
||
|
||
在开启 write 或 all 的空闲检测的情况下,写入的时候,会添加写入监听器 `writeListener` 。该监听器会在消息( 数据 ) flush 到对端后,**回调**,修改最后写入时间 `lastWriteTime` 为 `#ticksInNanos()` 。代码如下:
|
||
|
||
```
|
||
// Not create a new ChannelFutureListener per write operation to reduce GC pressure.
|
||
private final ChannelFutureListener writeListener = new ChannelFutureListener() {
|
||
|
||
@Override
|
||
public void operationComplete(ChannelFuture future) throws Exception {
|
||
// 记录最后写时间
|
||
lastWriteTime = ticksInNanos();
|
||
// 重置 firstWriterIdleEvent 和 firstAllIdleEvent 为 true
|
||
firstWriterIdleEvent = firstAllIdleEvent = true;
|
||
}
|
||
|
||
}
|
||
```
|
||
|
||
## 5.7 hasOutputChanged
|
||
|
||
> 老艿艿:关于这个方法,看完 [「5.8.2 WriterIdelTimeoutTask」](https://svip.iocoder.cn/Netty/ChannelHandler-5-idle/#) 后,再回过头理解。
|
||
|
||
`#hasOutputChanged(ChannelHandlerContext ctx, boolean first)` 方法,判断 ChannelOutboundBuffer 是否发生变化。代码如下:
|
||
|
||
```
|
||
/**
|
||
* Returns {@code true} if and only if the {@link IdleStateHandler} was constructed
|
||
* with {@link #observeOutput} enabled and there has been an observed change in the
|
||
* {@link ChannelOutboundBuffer} between two consecutive calls of this method.
|
||
*
|
||
* https://github.com/netty/netty/issues/6150
|
||
*/
|
||
1: private boolean hasOutputChanged(ChannelHandlerContext ctx, boolean first) {
|
||
2: // 开启观察 ChannelOutboundBuffer 队列
|
||
3: if (observeOutput) {
|
||
4:
|
||
5: // We can take this shortcut if the ChannelPromises that got passed into write()
|
||
6: // appear to complete. It indicates "change" on message level and we simply assume
|
||
7: // that there's change happening on byte level. If the user doesn't observe channel
|
||
8: // writability events then they'll eventually OOME and there's clearly a different
|
||
9: // problem and idleness is least of their concerns.
|
||
10: // 如果 lastChangeCheckTimeStamp 和 lastWriteTime 不一样,说明写操作进行过了,则更新此值
|
||
11: if (lastChangeCheckTimeStamp != lastWriteTime) {
|
||
12: lastChangeCheckTimeStamp = lastWriteTime;
|
||
13:
|
||
14: // But this applies only if it's the non-first call.
|
||
15: if (!first) { // 非首次
|
||
16: return true;
|
||
17: }
|
||
18: }
|
||
19:
|
||
20: Channel channel = ctx.channel();
|
||
21: Unsafe unsafe = channel.unsafe();
|
||
22: ChannelOutboundBuffer buf = unsafe.outboundBuffer();
|
||
23:
|
||
24: if (buf != null) {
|
||
25: // 获得新的 messageHashCode 和 pendingWriteBytes
|
||
26: int messageHashCode = System.identityHashCode(buf.current());
|
||
27: long pendingWriteBytes = buf.totalPendingWriteBytes();
|
||
28:
|
||
29: // 发生了变化
|
||
30: if (messageHashCode != lastMessageHashCode || pendingWriteBytes != lastPendingWriteBytes) {
|
||
31: // 修改最后一次的 lastMessageHashCode 和 lastPendingWriteBytes
|
||
32: lastMessageHashCode = messageHashCode;
|
||
33: lastPendingWriteBytes = pendingWriteBytes;
|
||
34:
|
||
35: if (!first) { // 非首次
|
||
36: return true;
|
||
37: }
|
||
38: }
|
||
39: }
|
||
40: }
|
||
41:
|
||
42: return false;
|
||
43: }
|
||
```
|
||
|
||
- 第 3 行:判断开启观察 ChannelOutboundBuffer 队列。
|
||
|
||
- 如果 `lastChangeCheckTimeStamp` 和 `lastWriteTime` 不一样,说明写操作进行过了,则更新此值。
|
||
- 第 14 至 17 行:这段逻辑,理论来说不会发生。因为 `lastWriteTime` 属性,只会在 `writeListener` 回调中修改,那么如果发生 `lastChangeCheckTimeStamp` 和 `lastWriteTime` 不相等,`first` 必然为 `true` 。因为,Channel 相关的事件逻辑,都在它所在的 EventLoop 中,不会出现并发的情况。关于这一块,基友【莫那一鲁道】在 https://github.com/netty/netty/issues/8251 已经进行提问,坐等结果。
|
||
|
||
- 第 25 至 27 行:获得新的 `messageHashCode` 和 `pendingWriteBytes` 的。
|
||
|
||
- 第 29 至 33 行:若发生了变化,则修改最后一次的
|
||
|
||
|
||
|
||
```
|
||
lastMessageHashCode
|
||
```
|
||
|
||
|
||
|
||
和
|
||
|
||
|
||
|
||
```
|
||
lastPendingWriteBytes
|
||
```
|
||
|
||
|
||
|
||
。
|
||
|
||
- `messageHashCode != lastMessageHashCode` 成立,① 有可能对端接收数据比较慢,导致一个消息发送了一部分;② 又或者,发送的消息**非常非常非常大**,导致一个消息发送了一部分,就将发送缓存区写满。如果是这种情况下,可以使用 ChunkedWriteHandler ,一条大消息,拆成多条小消息。
|
||
- `pendingWriteBytes != lastPendingWriteBytes` 成立,① 有新的消息,写到 ChannelOutboundBuffer 内存队列中;② 有几条消息成功写到对端。这种情况,此处不会发生。
|
||
|
||
- 第 35 至 37 行:当且仅当
|
||
|
||
|
||
|
||
```
|
||
first
|
||
```
|
||
|
||
|
||
|
||
为
|
||
|
||
|
||
|
||
```
|
||
true
|
||
```
|
||
|
||
|
||
|
||
时,即非首次,才返回
|
||
|
||
|
||
|
||
```
|
||
true
|
||
```
|
||
|
||
|
||
|
||
,表示 ChannelOutboundBuffer 发生变化。
|
||
|
||
- 这是一个有点“神奇”的设定,笔者表示不太理解。理论来说,ChannelOutboundBuffer 是否发生变化,只需要考虑【第 30 行】代码的判断。如果加了 `!first` 的判断,导致的结果是在 WriterIdleTimeoutTask 和 AllIdleTimeoutTask 任务中,ChannelOutboundBuffer 即使发生了变化,在**首次**还是会触发 write 和 all 空闲事件,在**非首次**不会触发 write 和 all 空闲事件。
|
||
- 关于上述的困惑,[《Netty 那些事儿 ——— 关于 “Netty 发送大数据包时 触发写空闲超时” 的一些思考》](https://www.jianshu.com/p/8fe70d313d78) 一文的作者,也表达了相同的困惑。后续,找闪电侠面基沟通下。
|
||
- 关于上述的困惑,[《Netty 心跳服务之 IdleStateHandler 源码分析》](https://www.jianshu.com/p/f2ed73cf4df8) 一文的作者,表达了自己的理解。感兴趣的胖友,可以看看。
|
||
- 当然,这块如果不理解的胖友,也不要方。从笔者目前了解下来,`observeOutput` 都是设置为 `false` 。也就说,不会触发这个方法的执行。
|
||
|
||
- 第 42 行:返回 `false` ,表示 ChannelOutboundBuffer 未发生变化。
|
||
|
||
## 5.8 AbstractIdleTask
|
||
|
||
AbstractIdleTask ,实现 Runnable 接口,空闲任务抽象类。代码如下:
|
||
|
||
> AbstractIdleTask 是 IdleStateHandler 的内部静态类。
|
||
|
||
```
|
||
private abstract static class AbstractIdleTask implements Runnable {
|
||
|
||
private final ChannelHandlerContext ctx;
|
||
|
||
AbstractIdleTask(ChannelHandlerContext ctx) {
|
||
this.ctx = ctx;
|
||
}
|
||
|
||
@Override
|
||
public void run() {
|
||
// <1> 忽略未打开的 Channel
|
||
if (!ctx.channel().isOpen()) {
|
||
return;
|
||
}
|
||
|
||
// <2> 执行任务
|
||
run(ctx);
|
||
}
|
||
|
||
protected abstract void run(ChannelHandlerContext ctx);
|
||
|
||
}
|
||
```
|
||
|
||
- `<1>` 处,忽略未打开的 Channel 。
|
||
- `<2>` 处,子类实现 `#run()` **抽象**方法,实现自定义的空闲检测逻辑。
|
||
|
||
### 5.8.1 ReaderIdleTimeoutTask
|
||
|
||
ReaderIdleTimeoutTask ,继承 AbstractIdleTask 抽象类,检测 Read 空闲超时**定时**任务。代码如下:
|
||
|
||
> ReaderIdleTimeoutTask 是 IdleStateHandler 的内部静态类。
|
||
|
||
```
|
||
1: private final class ReaderIdleTimeoutTask extends AbstractIdleTask {
|
||
2:
|
||
3: ReaderIdleTimeoutTask(ChannelHandlerContext ctx) {
|
||
4: super(ctx);
|
||
5: }
|
||
6:
|
||
7: @Override
|
||
8: protected void run(ChannelHandlerContext ctx) {
|
||
9: // 计算下一次检测的定时任务的延迟
|
||
10: long nextDelay = readerIdleTimeNanos;
|
||
11: if (!reading) {
|
||
12: nextDelay -= ticksInNanos() - lastReadTime;
|
||
13: }
|
||
14:
|
||
15: // 如果小于等于 0 ,说明检测到读空闲
|
||
16: if (nextDelay <= 0) {
|
||
17: // 延迟时间为 readerIdleTimeNanos ,即再次检测
|
||
18: // Reader is idle - set a new timeout and notify the callback.
|
||
19: readerIdleTimeout = schedule(ctx, this, readerIdleTimeNanos, TimeUnit.NANOSECONDS);
|
||
20:
|
||
21: // 获得当前是否首次检测到读空闲
|
||
22: boolean first = firstReaderIdleEvent;
|
||
23: // 标记 firstReaderIdleEvent 为 false 。也就说,下次检测到空闲,就非首次了。
|
||
24: firstReaderIdleEvent = false;
|
||
25:
|
||
26: try {
|
||
27: // 创建读空闲事件
|
||
28: IdleStateEvent event = newIdleStateEvent(IdleState.READER_IDLE, first);
|
||
29: // 通知通道空闲事件
|
||
30: channelIdle(ctx, event);
|
||
31: } catch (Throwable t) {
|
||
32: // 触发 Exception Caught 到下一个节点
|
||
33: ctx.fireExceptionCaught(t);
|
||
34: }
|
||
35: // 如果大于 0 ,说明未检测到读空闲
|
||
36: } else {
|
||
37: // 延迟时间为 nextDelay ,即按照最后一次读的时间作为开始计数
|
||
38: // Read occurred before the timeout - set a new timeout with shorter delay.
|
||
39: readerIdleTimeout = schedule(ctx, this, nextDelay, TimeUnit.NANOSECONDS);
|
||
40: }
|
||
41: }
|
||
42: }
|
||
```
|
||
|
||
- 第 9 至 13 行:计算下一次检测的定时任务的
|
||
|
||
延迟
|
||
|
||
。
|
||
|
||
- `reading` 为 `true` 时,意味着正在读取,**不会**被检测为读空闲。
|
||
- `reading` 为 `false` 时,实际 `nextDelay` 的计算为 `readerIdleTimeNanos - (ticksInNanos() - lastReadTime)` 。如果小于等于 0 ,意味着 `ticksInNanos() - lastReadTime >= readerIdleTimeNanos` ,超时。
|
||
|
||
- ① 第 35 至 40 行:如果
|
||
|
||
大于
|
||
|
||
|
||
|
||
0 ,说明未检测到读空闲。
|
||
|
||
- 第 39 行:调用 `#schedule(ChannelHandlerContext ctx, Runnable task, long delay, TimeUnit unit)` 方法,初始**下一次**的 ReaderIdleTimeoutTask 定时任务。其中,延迟时间为 `nextDelay` ,即按照最后一次读的时间作为开始计数。
|
||
|
||
- ② 第 15 至 34 行:如果
|
||
|
||
小于等于
|
||
|
||
|
||
|
||
0 ,说明检测到读空闲。
|
||
|
||
- 第 19 行:调用 `#schedule(ChannelHandlerContext ctx, Runnable task, long delay, TimeUnit unit)` 方法,初始**下一次**的 ReaderIdleTimeoutTask 定时任务。其中,延迟时间为 `readerIdleTimeNanos` ,即重新计数。
|
||
|
||
- 第 21 行:获得当前是否首次检测到读空闲。
|
||
|
||
- 第 24 行:标记 `firstReaderIdleEvent` 为 `false` 。也就说,下次检测到空闲,就**非首次**了。
|
||
|
||
- 第 28 行:调用
|
||
|
||
|
||
|
||
```
|
||
#newIdleStateEvent(IdleState state, boolean first)
|
||
```
|
||
|
||
|
||
|
||
方法,创建创建
|
||
|
||
读
|
||
|
||
空闲事件。
|
||
|
||
- 第 30 行: 调用 `#channelIdle(ChannelHandlerContext ctx, IdleStateEvent evt)` 方法,在 pipeline 中,触发 UserEvent 事件。
|
||
|
||
- 第 31 至 34 行:如果**发生异常**,触发 Exception Caught 事件到下一个节点,处理异常。
|
||
|
||
### 5.8.2 WriterIdleTimeoutTask
|
||
|
||
WriterIdleTimeoutTask ,继承 AbstractIdleTask 抽象类,检测 Write 空闲超时**定时**任务。代码如下:
|
||
|
||
> WriterIdleTimeoutTask 是 IdleStateHandler 的内部静态类。
|
||
|
||
```
|
||
1: private final class WriterIdleTimeoutTask extends AbstractIdleTask {
|
||
2:
|
||
3: WriterIdleTimeoutTask(ChannelHandlerContext ctx) {
|
||
4: super(ctx);
|
||
5: }
|
||
6:
|
||
7: @Override
|
||
8: protected void run(ChannelHandlerContext ctx) {
|
||
9: // 计算下一次检测的定时任务的延迟
|
||
10: long lastWriteTime = IdleStateHandler.this.lastWriteTime;
|
||
11: long nextDelay = writerIdleTimeNanos - (ticksInNanos() - lastWriteTime);
|
||
12:
|
||
13: // 如果小于等于 0 ,说明检测到写空闲
|
||
14: if (nextDelay <= 0) {
|
||
15: // 延迟时间为 writerIdleTimeout ,即再次检测
|
||
16: // Writer is idle - set a new timeout and notify the callback.
|
||
17: writerIdleTimeout = schedule(ctx, this, writerIdleTimeNanos, TimeUnit.NANOSECONDS);
|
||
18:
|
||
19: // 获得当前是否首次检测到写空闲
|
||
20: boolean first = firstWriterIdleEvent;
|
||
21: // 标记 firstWriterIdleEvent 为 false 。也就说,下次检测到空闲,就非首次了。
|
||
22: firstWriterIdleEvent = false;
|
||
23:
|
||
24: try {
|
||
25: // 判断 ChannelOutboundBuffer 是否发生变化
|
||
26: if (hasOutputChanged(ctx, first)) {
|
||
27: return;
|
||
28: }
|
||
29:
|
||
30: // 创建写空闲事件
|
||
31: IdleStateEvent event = newIdleStateEvent(IdleState.WRITER_IDLE, first);
|
||
32: // 通知通道空闲事件
|
||
33: channelIdle(ctx, event);
|
||
34: } catch (Throwable t) {
|
||
35: // 触发 Exception Caught 到下一个节点
|
||
36: ctx.fireExceptionCaught(t);
|
||
37: }
|
||
38: // 如果大于 0 ,说明未检测到读空闲
|
||
39: } else {
|
||
40: // Write occurred before the timeout - set a new timeout with shorter delay.
|
||
41: writerIdleTimeout = schedule(ctx, this, nextDelay, TimeUnit.NANOSECONDS);
|
||
42: }
|
||
43: }
|
||
44: }
|
||
```
|
||
|
||
- 第 9 至 11 行:计算下一次检测的定时任务的**延迟**。
|
||
|
||
- ① 第 38 至 42 行:如果
|
||
|
||
大于
|
||
|
||
|
||
|
||
0 ,说明未检测到写空闲。
|
||
|
||
- 第 39 行:调用 `#schedule(ChannelHandlerContext ctx, Runnable task, long delay, TimeUnit unit)` 方法,初始**下一次**的 WriterIdleTimeoutTask 定时任务。其中,延迟时间为 `nextDelay` ,即按照最后一次写的时间作为开始计数。
|
||
|
||
- ② 第 13 至 37 行:如果
|
||
|
||
小于等于
|
||
|
||
|
||
|
||
0 ,说明检测到写空闲。
|
||
|
||
- 第 17 行:调用 `#schedule(ChannelHandlerContext ctx, Runnable task, long delay, TimeUnit unit)` 方法,初始**下一次**的 WriterIdleTimeoutTask 定时任务。其中,延迟时间为 `readerIdleTimeNanos` ,即重新计数。
|
||
|
||
- 第 20 行:获得当前是否首次检测到写空闲。
|
||
|
||
- 第 22 行:标记 `firstWriterIdleEvent` 为 `false` 。也就说,下次检测到空闲,就**非首次**了。
|
||
|
||
- 第 25 至 28 行:判断 ChannelOutboundBuffer 是否发生变化。如果有变化,不触发写空闲时间。
|
||
|
||
- 第 31 行:调用
|
||
|
||
|
||
|
||
```
|
||
#newIdleStateEvent(IdleState state, boolean first)
|
||
```
|
||
|
||
|
||
|
||
方法,创建创建
|
||
|
||
写
|
||
|
||
空闲事件。
|
||
|
||
- 第 33 行: 调用 `#channelIdle(ChannelHandlerContext ctx, IdleStateEvent evt)` 方法,在 pipeline 中,触发 UserEvent 事件。
|
||
|
||
- 第 34 至 37 行:如果**发生异常**,触发 Exception Caught 事件到下一个节点,处理异常。
|
||
|
||
### 5.8.3 AllIdleTimeoutTask
|
||
|
||
AllIdleTimeoutTask ,继承 AbstractIdleTask 抽象类,检测 All 空闲超时**定时**任务。代码如下:
|
||
|
||
> AllIdleTimeoutTask 是 IdleStateHandler 的内部静态类。
|
||
|
||
```
|
||
private final class AllIdleTimeoutTask extends AbstractIdleTask {
|
||
|
||
AllIdleTimeoutTask(ChannelHandlerContext ctx) {
|
||
super(ctx);
|
||
}
|
||
|
||
@Override
|
||
protected void run(ChannelHandlerContext ctx) {
|
||
// 计算下一次检测的定时任务的延迟
|
||
long nextDelay = allIdleTimeNanos;
|
||
if (!reading) {
|
||
nextDelay -= ticksInNanos() - Math.max(lastReadTime, lastWriteTime); // <1> 取大值
|
||
}
|
||
|
||
// 如果小于等于 0 ,说明检测到 all 空闲
|
||
if (nextDelay <= 0) {
|
||
// 延迟时间为 allIdleTimeNanos ,即再次检测
|
||
// Both reader and writer are idle - set a new timeout and
|
||
// notify the callback.
|
||
allIdleTimeout = schedule(ctx, this, allIdleTimeNanos, TimeUnit.NANOSECONDS);
|
||
|
||
// 获得当前是否首次检测到 all 空闲
|
||
boolean first = firstAllIdleEvent;
|
||
// 标记 firstAllIdleEvent 为 false 。也就说,下次检测到空闲,就非首次了。
|
||
firstAllIdleEvent = false;
|
||
|
||
try {
|
||
// 判断 ChannelOutboundBuffer 是否发生变化
|
||
if (hasOutputChanged(ctx, first)) {
|
||
return;
|
||
}
|
||
|
||
// 创建 all 空闲事件
|
||
IdleStateEvent event = newIdleStateEvent(IdleState.ALL_IDLE, first);
|
||
// 通知通道空闲事件
|
||
channelIdle(ctx, event);
|
||
} catch (Throwable t) {
|
||
ctx.fireExceptionCaught(t);
|
||
}
|
||
// 如果大于 0 ,说明未检测到 all 空闲
|
||
} else {
|
||
// Either read or write occurred before the timeout - set a new
|
||
// timeout with shorter delay.
|
||
allIdleTimeout = schedule(ctx, this, nextDelay, TimeUnit.NANOSECONDS);
|
||
}
|
||
}
|
||
}
|
||
```
|
||
|
||
- 因为 All 是 Write 和 Read **任一**一种空闲即可,所以 AllIdleTimeoutTask 是 ReaderIdleTimeoutTask 和 WriterIdleTimeoutTask 的**综合**。
|
||
- `<1>` 处,取 `lastReadTime` 和 `lastWriteTime` 中的**大**值,从而来判断,是否有 Write 和 Read **任一**一种空闲。
|
||
- WriterIdleTimeoutTask 就不详细解析,胖友自己读读代码即可。
|
||
|
||
# 6. ReadTimeoutHandler
|
||
|
||
`io.netty.handler.timeout.ReadTimeoutHandler` ,继承 IdleStateHandler 类,当 Channel 的**读**空闲时间( 读或者写 )太长时,抛出 ReadTimeoutException 异常,并自动关闭该 Channel 。
|
||
|
||
## 6.1 构造方法
|
||
|
||
```
|
||
/**
|
||
* Channel 是否关闭
|
||
*/
|
||
private boolean closed;
|
||
|
||
public ReadTimeoutHandler(int timeoutSeconds) {
|
||
this(timeoutSeconds, TimeUnit.SECONDS);
|
||
}
|
||
|
||
public ReadTimeoutHandler(long timeout, TimeUnit unit) {
|
||
// 禁用 Write / All 的空闲检测
|
||
super(timeout, 0, 0, unit); // <1>
|
||
}
|
||
```
|
||
|
||
- `closed` 属性,Channel 是否关闭。
|
||
- `<1>` 处,禁用 Write / All 的空闲检测,只根据 `timeout` 方法参数,开启 Read 的空闲检测。
|
||
|
||
## 6.2 channelIdle
|
||
|
||
`#channelIdle(ChannelHandlerContext ctx, IdleStateEvent evt)` 方法,覆写父类方法,代码如下:
|
||
|
||
```
|
||
@Override
|
||
protected final void channelIdle(ChannelHandlerContext ctx, IdleStateEvent evt) throws Exception {
|
||
assert evt.state() == IdleState.READER_IDLE;
|
||
readTimedOut(ctx);
|
||
}
|
||
|
||
/**
|
||
* Is called when a read timeout was detected.
|
||
*/
|
||
protected void readTimedOut(ChannelHandlerContext ctx) throws Exception {
|
||
if (!closed) {
|
||
// <1> 触发 Exception Caught 事件到 pipeline 中,异常为 ReadTimeoutException
|
||
ctx.fireExceptionCaught(ReadTimeoutException.INSTANCE);
|
||
// <2> 关闭 Channel 通道
|
||
ctx.close();
|
||
// <3> 标记 Channel 为已关闭
|
||
closed = true;
|
||
}
|
||
}
|
||
```
|
||
|
||
- `<1>` 处,触发 Exception Caught 事件到 pipeline 中,异常为 ReadTimeoutException 。
|
||
- `<2>` 处,关闭 Channel 通道。
|
||
- `<3>` 处,标记 Channel 为已关闭。
|
||
|
||
# 7. WriteTimeoutHandler
|
||
|
||
`io.netty.handler.timeout.WriteTimeoutHandler` ,继承 ChannelOutboundHandlerAdapter 类,当一个**写**操作不能在指定时间内完成时,抛出 WriteTimeoutException 异常,并自动关闭对应 Channel 。
|
||
|
||
😈 **注意,这里写入,指的是 flush 到对端 Channel ,而不仅仅是写到 ChannelOutboundBuffer 队列**。
|
||
|
||
## 7.1 构造方法
|
||
|
||
```
|
||
/**
|
||
* 最小的超时时间,单位:纳秒
|
||
*/
|
||
private static final long MIN_TIMEOUT_NANOS = TimeUnit.MILLISECONDS.toNanos(1);
|
||
|
||
/**
|
||
* 超时时间,单位:纳秒
|
||
*/
|
||
private final long timeoutNanos;
|
||
|
||
/**
|
||
* WriteTimeoutTask 双向链表。
|
||
*
|
||
* lastTask 为链表的尾节点
|
||
*
|
||
* A doubly-linked list to track all WriteTimeoutTasks
|
||
*/
|
||
private WriteTimeoutTask lastTask;
|
||
|
||
/**
|
||
* Channel 是否关闭
|
||
*/
|
||
private boolean closed;
|
||
|
||
public WriteTimeoutHandler(int timeoutSeconds) {
|
||
this(timeoutSeconds, TimeUnit.SECONDS);
|
||
}
|
||
|
||
public WriteTimeoutHandler(long timeout, TimeUnit unit) {
|
||
if (unit == null) {
|
||
throw new NullPointerException("unit");
|
||
}
|
||
|
||
if (timeout <= 0) {
|
||
timeoutNanos = 0;
|
||
} else {
|
||
timeoutNanos = Math.max(unit.toNanos(timeout), MIN_TIMEOUT_NANOS); // 保证大于等于 MIN_TIMEOUT_NANOS
|
||
}
|
||
}
|
||
```
|
||
|
||
- ```
|
||
timeoutNanos
|
||
```
|
||
|
||
|
||
|
||
属性,写入超时时间,单位:纳秒。
|
||
|
||
- `MIN_TIMEOUT_NANOS` 属性,最小的超时时间,单位:纳秒。
|
||
|
||
- `lastTask` 属性,WriteTimeoutTask 双向链表。其中,`lastTask` 为链表的**尾节点**。
|
||
|
||
- `closed` 属性,Channel 是否关闭。
|
||
|
||
## 7.2 handlerRemoved
|
||
|
||
`#handlerRemoved(ChannelHandlerContext ctx)` 方法,移除所有 WriteTimeoutTask 任务,并取消。代码如下:
|
||
|
||
```
|
||
@Override
|
||
public void handlerRemoved(ChannelHandlerContext ctx) throws Exception {
|
||
WriteTimeoutTask task = lastTask;
|
||
// 置空 lastTask
|
||
lastTask = null;
|
||
// 循环移除,知道为空
|
||
while (task != null) {
|
||
// 取消当前任务的定时任务
|
||
task.scheduledFuture.cancel(false);
|
||
// 记录前一个任务
|
||
WriteTimeoutTask prev = task.prev;
|
||
// 置空当前任务的前后节点
|
||
task.prev = null;
|
||
task.next = null;
|
||
// 跳到前一个任务
|
||
task = prev;
|
||
}
|
||
}
|
||
```
|
||
|
||
- 代码比较简单,胖友自己看注释。
|
||
|
||
## 7.3 write
|
||
|
||
`#write(ChannelHandlerContext ctx, Object msg, ChannelPromise promise)` 方法,代码如下:
|
||
|
||
```
|
||
@Override
|
||
public void write(ChannelHandlerContext ctx, Object msg, ChannelPromise promise) throws Exception {
|
||
if (timeoutNanos > 0) {
|
||
// 如果 promise 是 VoidPromise ,则包装成非 VoidPromise ,为了后续的回调。
|
||
promise = promise.unvoid(); <1》
|
||
// 创建定时任务
|
||
scheduleTimeout(ctx, promise);
|
||
}
|
||
// 写入
|
||
ctx.write(msg, promise);
|
||
}
|
||
```
|
||
|
||
- `<1>` 处,如果 `promise` 类型是 VoidPromise ,则包装成非 VoidPromise ,为了后续的回调。因为 VoidPromise 无法接收到回调。
|
||
- `<2>` 处,调用 `#scheduleTimeout(final ChannelHandlerContext ctx, final ChannelPromise promise)` 方法,创建定时任务。详细解析,见 [「7.4 scheduleTimeout」](https://svip.iocoder.cn/Netty/ChannelHandler-5-idle/#) 。
|
||
|
||
## 7.4 scheduleTimeout
|
||
|
||
`#scheduleTimeout(final ChannelHandlerContext ctx, final ChannelPromise promise)` 方法,创建定时任务。代码如下:
|
||
|
||
```
|
||
1: private void scheduleTimeout(final ChannelHandlerContext ctx, final ChannelPromise promise) {
|
||
2: // Schedule a timeout.
|
||
3: // 创建 WriteTimeoutTask 任务
|
||
4: final WriteTimeoutTask task = new WriteTimeoutTask(ctx, promise);
|
||
5: // 定时任务
|
||
6: task.scheduledFuture = ctx.executor().schedule(task, timeoutNanos, TimeUnit.NANOSECONDS);
|
||
7:
|
||
8: if (!task.scheduledFuture.isDone()) {
|
||
9: // 添加到链表
|
||
10: addWriteTimeoutTask(task);
|
||
11:
|
||
12: // Cancel the scheduled timeout if the flush promise is complete.
|
||
13: // 将 task 作为监听器,添加到 promise 中。在写入完成后,可以移除该定时任务
|
||
14: promise.addListener(task);
|
||
15: }
|
||
16: }
|
||
```
|
||
|
||
- 第 2 至 6 行:创建 WriteTimeoutTask 任务,并发起**定时任务**。
|
||
- 第 8 行:如果定时任务**已经执行完成**,则不需要进行监听。否则,需要执行【第 10 至 14 行】的代码逻辑。
|
||
- 第 10 行:调用 `#addWriteTimeoutTask(WriteTimeoutTask task)` 方法,添加到链表。详细解析,见 [「7.5 addWriteTimeoutTask」](https://svip.iocoder.cn/Netty/ChannelHandler-5-idle/#) 。
|
||
- 第 14 行:将 `task` 作为监听器,添加到 `promise` 中。在写入完成后,可以移除该定时任务。也就说,调用链是 `flush => 回调 => promise => 回调 => task` 。
|
||
|
||
## 7.5 addWriteTimeoutTask
|
||
|
||
`#addWriteTimeoutTask(WriteTimeoutTask task)` 方法,添加到链表。代码如下:
|
||
|
||
```
|
||
private void addWriteTimeoutTask(WriteTimeoutTask task) {
|
||
// 添加到链表的尾节点
|
||
if (lastTask != null) {
|
||
lastTask.next = task;
|
||
task.prev = lastTask;
|
||
}
|
||
// 修改 lastTask 为当前任务
|
||
lastTask = task;
|
||
}
|
||
```
|
||
|
||
添加到链表的尾节点,并修改 `lastTask` 为**当前**任务。
|
||
|
||
## 7.6 removeWriteTimeoutTask
|
||
|
||
`#removeWriteTimeoutTask(WriteTimeoutTask task)` 方法,移除出链表。代码如下:
|
||
|
||
```
|
||
private void removeWriteTimeoutTask(WriteTimeoutTask task) {
|
||
// 从双向链表中,移除自己
|
||
if (task == lastTask) { // 尾节点
|
||
// task is the tail of list
|
||
assert task.next == null;
|
||
lastTask = lastTask.prev;
|
||
if (lastTask != null) {
|
||
lastTask.next = null;
|
||
}
|
||
} else if (task.prev == null && task.next == null) { // 已经被移除
|
||
// Since task is not lastTask, then it has been removed or not been added.
|
||
return;
|
||
} else if (task.prev == null) { // 头节点
|
||
// task is the head of list and the list has at least 2 nodes
|
||
task.next.prev = null;
|
||
} else { // 中间的节点
|
||
task.prev.next = task.next;
|
||
task.next.prev = task.prev;
|
||
}
|
||
// 重置 task 前后节点为空
|
||
task.prev = null;
|
||
task.next = null;
|
||
}
|
||
```
|
||
|
||
该方法的调用,在 [「7.8 WriteTimeoutTask」](https://svip.iocoder.cn/Netty/ChannelHandler-5-idle/#) 会看到。
|
||
|
||
## 7.7 writeTimedOut
|
||
|
||
`#writeTimedOut(ChannelHandlerContext ctx)` 方法,写入超时,关闭 Channel 通道。代码如下:
|
||
|
||
```
|
||
/**
|
||
* Is called when a write timeout was detected
|
||
*/
|
||
protected void writeTimedOut(ChannelHandlerContext ctx) throws Exception {
|
||
if (!closed) {
|
||
// 触发 Exception Caught 事件到 pipeline 中,异常为 WriteTimeoutException
|
||
ctx.fireExceptionCaught(WriteTimeoutException.INSTANCE);
|
||
// 关闭 Channel 通道
|
||
ctx.close();
|
||
// 标记 Channel 为已关闭
|
||
closed = true;
|
||
}
|
||
}
|
||
```
|
||
|
||
- 和 `ReadTimeoutHandler#readTimeout(ChannelHandlerContext ctx)` 方法,基本类似。
|
||
|
||
该方法的调用,在 [「7.8 WriteTimeoutTask」](https://svip.iocoder.cn/Netty/ChannelHandler-5-idle/#) 会看到。
|
||
|
||
## 7.8 WriteTimeoutTask
|
||
|
||
WriteTimeoutTask ,实现 Runnable 和 ChannelFutureListener 接口,写入超时任务。
|
||
|
||
> WriteTimeoutTask 是 WriteTimeoutHandler 的内部类。
|
||
|
||
### 7.8.1 构造方法
|
||
|
||
```
|
||
private final ChannelHandlerContext ctx;
|
||
/**
|
||
* 写入任务的 Promise 对象
|
||
*/
|
||
private final ChannelPromise promise;
|
||
|
||
// WriteTimeoutTask is also a node of a doubly-linked list
|
||
/**
|
||
* 前一个 task
|
||
*/
|
||
WriteTimeoutTask prev;
|
||
/**
|
||
* 后一个 task
|
||
*/
|
||
WriteTimeoutTask next;
|
||
/**
|
||
* 定时任务
|
||
*/
|
||
ScheduledFuture<?> scheduledFuture;
|
||
|
||
WriteTimeoutTask(ChannelHandlerContext ctx, ChannelPromise promise) {
|
||
this.ctx = ctx;
|
||
this.promise = promise;
|
||
}
|
||
```
|
||
|
||
### 7.8.2 run
|
||
|
||
当定时任务执行,说明写入任务执行超时。代码如下:
|
||
|
||
```
|
||
@Override
|
||
public void run() {
|
||
// Was not written yet so issue a write timeout
|
||
// The promise itself will be failed with a ClosedChannelException once the close() was issued
|
||
// See https://github.com/netty/netty/issues/2159
|
||
if (!promise.isDone()) { // 未完成,说明写入超时
|
||
try {
|
||
// <1> 写入超时,关闭 Channel 通道
|
||
writeTimedOut(ctx);
|
||
} catch (Throwable t) {
|
||
// 触发 Exception Caught 事件到 pipeline 中
|
||
ctx.fireExceptionCaught(t);
|
||
}
|
||
}
|
||
// <2> 移除出链表
|
||
removeWriteTimeoutTask(this);
|
||
}
|
||
```
|
||
|
||
- `<1>` 处,调用 `#writeTimedOut(ChannelHandlerContext ctx)` 方法,写入超时,关闭 Channel 通道。
|
||
- `<2>` 处,调用 `#removeWriteTimeoutTask(WriteTimeoutTask task)` 方法,移除出链表。
|
||
|
||
### 7.8.3 operationComplete
|
||
|
||
当回调方法执行,说明写入任务执行完成。代码如下:
|
||
|
||
```
|
||
@Override
|
||
public void operationComplete(ChannelFuture future) throws Exception {
|
||
// scheduledFuture has already be set when reaching here
|
||
// <1> 取消定时任务
|
||
scheduledFuture.cancel(false);
|
||
// <2> 移除出链表
|
||
removeWriteTimeoutTask(this);
|
||
}
|
||
```
|
||
|
||
- `<1>` 处,取消定时任务。
|
||
- `<2>` 处,调用 `#removeWriteTimeoutTask(WriteTimeoutTask task)` 方法,移除出链表。
|
||
|
||
# 666. 彩蛋
|
||
|
||
和 「5.7 hasOutputChanged」(#) 小节,这个方法较真了好久。感谢中间,基友【莫那一鲁道】的沟通。
|
||
|
||
推荐阅读文章:
|
||
|
||
- 莫那一鲁道 [《Netty 心跳服务之 IdleStateHandler 源码分析》](https://www.jianshu.com/p/f2ed73cf4df8)
|
||
- Hypercube [自顶向下深入分析Netty(八)–ChannelHandler](https://www.jianshu.com/p/a9bcd89553f5) |