2343 lines
86 KiB
Markdown
2343 lines
86 KiB
Markdown
# 精尽 Netty 源码解析 —— Channel(五)之 flush 操作
|
||
|
||
# 1. 概述
|
||
|
||
本文接 [《精尽 Netty 源码解析 —— Channel(四)之 write 操作》](http://svip.iocoder.cn/Netty/Channel-4-write/) ,分享 Netty Channel 的 `#flush()` 方法,刷新**内存队列**,将其中的数据写入到对端。
|
||
|
||
在本文中,我们会发现,`#flush()` 方法和 `#write(Object msg, ...)` **正常**情况下,经历的流程是**差不多**的,例如在 pipeline 中对事件的传播,从 `tail` 节点传播到 `head` 节点,最终交由 Unsafe 处理,而差异点就是 Unsafe 的处理方式**不同**:
|
||
|
||
- write 方法:将数据写到**内存队列**中。
|
||
- flush 方法:刷新**内存队列**,将其中的数据写入到对端。
|
||
|
||
当然,上述描述仅仅指的是**正常**情况下,在**异常**情况下会有所不同。我们知道,Channel 大多数情况下是**可写**的,所以不需要专门去注册 `SelectionKey.OP_WRITE` 事件。所以在 Netty 的实现中,默认 Channel 是**可写**的,当写入失败的时候,再去注册 `SelectionKey.OP_WRITE` 事件。这意味着什么呢?在 `#flush()` 方法中,如果写入数据到 Channel 失败,会通过注册 `SelectionKey.OP_WRITE` 事件,然后在轮询到 Channel **可写** 时,再“回调” `#forceFlush()` 方法。
|
||
|
||
是不是非常巧妙?!让我直奔代码,大口吃肉,潇洒撸码。
|
||
|
||
> 下文的 [「2.」](https://svip.iocoder.cn/Netty/Channel-5-flush/#)、[「3.」](https://svip.iocoder.cn/Netty/Channel-5-flush/#)、[「4.」](https://svip.iocoder.cn/Netty/Channel-5-flush/#)、[「5.」](https://svip.iocoder.cn/Netty/Channel-5-flush/#) 和 [《精尽 Netty 源码解析 —— Channel(四)之 write 操作》](http://svip.iocoder.cn/Netty/Channel-4-write) 非常**类似**,所以胖友可以快速浏览。真正的**差异**,从 [「6.」](https://svip.iocoder.cn/Netty/Channel-5-flush/#) 开始。
|
||
|
||
# 2. AbstractChannel
|
||
|
||
AbstractChannel 对 `#flush()` 方法的实现,代码如下:
|
||
|
||
```
|
||
@Override
|
||
public Channel flush() {
|
||
pipeline.flush();
|
||
return this;
|
||
}
|
||
```
|
||
|
||
- 在方法内部,会调用对应的
|
||
|
||
|
||
|
||
```
|
||
ChannelPipeline#flush()
|
||
```
|
||
|
||
|
||
|
||
方法,将 flush 事件在 pipeline 上传播。详细解析,见
|
||
|
||
|
||
|
||
「3. DefaultChannelPipeline」
|
||
|
||
|
||
|
||
。
|
||
|
||
- 最终会传播 flush 事件到 `head` 节点,刷新**内存队列**,将其中的数据写入到对端。详细解析,见 [「5. HeadContext」](https://svip.iocoder.cn/Netty/Channel-5-flush/#) 。
|
||
|
||
# 3. DefaultChannelPipeline
|
||
|
||
`DefaultChannelPipeline#flush()` 方法,代码如下:
|
||
|
||
```
|
||
@Override
|
||
public final ChannelPipeline flush() {
|
||
tail.flush();
|
||
return this;
|
||
}
|
||
```
|
||
|
||
- 在方法内部,会调用 `TailContext#flush()` 方法,将 flush 事件在 pipeline 中,从尾节点向头节点传播。详细解析,见 [「4. TailContext」](https://svip.iocoder.cn/Netty/Channel-5-flush/#) 。
|
||
|
||
# 4. TailContext
|
||
|
||
TailContext 对 `TailContext#flush()` 方法的实现,是从 AbstractChannelHandlerContext 抽象类继承,代码如下:
|
||
|
||
```
|
||
1: @Override
|
||
2: public ChannelHandlerContext flush() {
|
||
3: // 获得下一个 Outbound 节点
|
||
4: final AbstractChannelHandlerContext next = findContextOutbound();
|
||
5: EventExecutor executor = next.executor();
|
||
6: // 在 EventLoop 的线程中
|
||
7: if (executor.inEventLoop()) {
|
||
8: // 执行 flush 事件到下一个节点
|
||
9: next.invokeFlush();
|
||
10: // 不在 EventLoop 的线程中
|
||
11: } else {
|
||
12: // 创建 flush 任务
|
||
13: Runnable task = next.invokeFlushTask;
|
||
14: if (task == null) {
|
||
15: next.invokeFlushTask = task = new Runnable() {
|
||
16: @Override
|
||
17: public void run() {
|
||
18: next.invokeFlush();
|
||
19: }
|
||
20: };
|
||
21: }
|
||
22: // 提交到 EventLoop 的线程中,执行该任务
|
||
23: safeExecute(executor, task, channel().voidPromise(), null);
|
||
24: }
|
||
25:
|
||
26: return this;
|
||
27: }
|
||
```
|
||
|
||
- 第 4 行:调用 `#findContextOutbound()` 方法,获得**下一个** Outbound 节点。
|
||
|
||
- 第 7 行:
|
||
|
||
在
|
||
|
||
|
||
|
||
EventLoop 的线程中。
|
||
|
||
- 第 12 至 15 行:调用 `AbstractChannelHandlerContext#invokeFlush()()` 方法,执行 flush 事件到下一个节点。
|
||
- 后续的逻辑,和 [《精尽 Netty 源码解析 —— ChannelPipeline(四)之 Outbound 事件的传播》](http://svip.iocoder.cn/Netty/Pipeline-4-outbound/) 分享的 **bind** 事件在 pipeline 中的传播是**基本一致**的。
|
||
- 随着 flush **事件**不断的向下一个节点传播,最终会到达 HeadContext 节点。详细解析,见 [「5. HeadContext」](https://svip.iocoder.cn/Netty/Channel-5-flush/#) 。
|
||
|
||
- 第 16 行:
|
||
|
||
不在
|
||
|
||
|
||
|
||
EventLoop 的线程中。
|
||
|
||
- 第 12 至 21 行:创建 flush 任务。该任务的内部的调用【第 18 行】的代码,和【第 9 行】的代码是**一致**的。
|
||
- 第 23 行:调用 `#safeExecute(executor, task, promise, m)` 方法,提交到 EventLoop 的线程中,执行该任务。从而实现,**在** EventLoop 的线程中,执行 flush 事件到下一个节点。
|
||
|
||
# 5. HeadContext
|
||
|
||
在 pipeline 中,flush 事件最终会到达 HeadContext 节点。而 HeadContext 的 `#flush()` 方法,会处理该事件,代码如下:
|
||
|
||
```
|
||
@Override
|
||
public void flush(ChannelHandlerContext ctx) throws Exception {
|
||
unsafe.flush();
|
||
}
|
||
```
|
||
|
||
- 在方法内部,会调用 `AbstractUnsafe#flush()` 方法,刷新**内存队列**,将其中的数据写入到对端。详细解析,见 [「6. AbstractUnsafe」](https://svip.iocoder.cn/Netty/Channel-5-flush/#) 。
|
||
|
||
# 6. AbstractUnsafe
|
||
|
||
`AbstractUnsafe#flush()` 方法,刷新**内存队列**,将其中的数据写入到对端。代码如下:
|
||
|
||
```
|
||
1: @Override
|
||
2: public final void flush() {
|
||
3: assertEventLoop();
|
||
4:
|
||
5: // 内存队列为 null ,一般是 Channel 已经关闭,所以直接返回。
|
||
6: ChannelOutboundBuffer outboundBuffer = this.outboundBuffer;
|
||
7: if (outboundBuffer == null) {
|
||
8: return;
|
||
9: }
|
||
10:
|
||
11: // 标记内存队列开始 flush
|
||
12: outboundBuffer.addFlush();
|
||
13: // 执行 flush
|
||
14: flush0();
|
||
15: }
|
||
```
|
||
|
||
- 第 5 至 9 行:内存队列为 `null` ,一般是 Channel **已经关闭**,所以直接返回。
|
||
|
||
- 第 12 行:调用 `ChannelOutboundBuffer#addFlush()` 方法,标记内存队列开始 **flush** 。详细解析,见 [「8.4 addFlush」](https://svip.iocoder.cn/Netty/Channel-5-flush/#) 。
|
||
|
||
- 第 14 行:调用 `#flush0()` 方法,执行 flush 操作。代码如下:
|
||
|
||
```
|
||
/**
|
||
* 是否正在 flush 中,即正在调用 {@link #flush0()} 中
|
||
*/
|
||
private boolean inFlush0;
|
||
|
||
1: @SuppressWarnings("deprecation")
|
||
2: protected void flush0() {
|
||
3: // 正在 flush 中,所以直接返回。
|
||
4: if (inFlush0) {
|
||
5: // Avoid re-entrance
|
||
6: return;
|
||
7: }
|
||
8:
|
||
9: // 内存队列为 null ,一般是 Channel 已经关闭,所以直接返回。
|
||
10: // 内存队列为空,无需 flush ,所以直接返回
|
||
11: final ChannelOutboundBuffer outboundBuffer = this.outboundBuffer;
|
||
12: if (outboundBuffer == null || outboundBuffer.isEmpty()) {
|
||
13: return;
|
||
14: }
|
||
15:
|
||
16: // 标记正在 flush 中。
|
||
17: inFlush0 = true;
|
||
18:
|
||
19: // 若未激活,通知 flush 失败
|
||
20: // Mark all pending write requests as failure if the channel is inactive.
|
||
21: if (!isActive()) {
|
||
22: try {
|
||
23: if (isOpen()) {
|
||
24: outboundBuffer.failFlushed(FLUSH0_NOT_YET_CONNECTED_EXCEPTION, true);
|
||
25: } else {
|
||
26: // Do not trigger channelWritabilityChanged because the channel is closed already.
|
||
27: outboundBuffer.failFlushed(FLUSH0_CLOSED_CHANNEL_EXCEPTION, false);
|
||
28: }
|
||
29: } finally {
|
||
30: // 标记不在 flush 中。
|
||
31: inFlush0 = false;
|
||
32: }
|
||
33: return;
|
||
34: }
|
||
35:
|
||
36: // 执行真正的写入到对端
|
||
37: try {
|
||
38: doWrite(outboundBuffer);
|
||
39: } catch (Throwable t) {
|
||
40: // TODO 芋艿 细节
|
||
41: if (t instanceof IOException && config().isAutoClose()) {
|
||
42: /**
|
||
43: * Just call {@link #close(ChannelPromise, Throwable, boolean)} here which will take care of
|
||
44: * failing all flushed messages and also ensure the actual close of the underlying transport
|
||
45: * will happen before the promises are notified.
|
||
46: *
|
||
47: * This is needed as otherwise {@link #isActive()} , {@link #isOpen()} and {@link #isWritable()}
|
||
48: * may still return {@code true} even if the channel should be closed as result of the exception.
|
||
49: */
|
||
50: close(voidPromise(), t, FLUSH0_CLOSED_CHANNEL_EXCEPTION, false);
|
||
51: } else {
|
||
52: try {
|
||
53: shutdownOutput(voidPromise(), t);
|
||
54: } catch (Throwable t2) {
|
||
55: close(voidPromise(), t2, FLUSH0_CLOSED_CHANNEL_EXCEPTION, false);
|
||
56: }
|
||
57: }
|
||
58: } finally {
|
||
59: // 标记不在 flush 中。
|
||
60: inFlush0 = false;
|
||
61: }
|
||
62: }
|
||
```
|
||
|
||
- `inFlush0` 字段,是否正在 flush 中,即正在调用 `#flush0()` 中。
|
||
|
||
- 第 3 至 7 行:正在 flush 中,所以直接返回。
|
||
|
||
- 第 9 至 14 行:
|
||
|
||
- `outboundBuffer == null` ,内存队列为 `null` ,一般是 Channel 已经**关闭**,所以直接返回。
|
||
- `outboundBuffer.isEmpty()` ,内存队列为空,无需 flush ,所以直接返回。
|
||
|
||
- 第 17 行:设置 `inFlush0` 为 `true` ,表示正在 flush 中。
|
||
|
||
- 第 19 至 34 行:调用
|
||
|
||
|
||
|
||
```
|
||
#isActive()
|
||
```
|
||
|
||
|
||
|
||
方法,发现 Channel
|
||
|
||
|
||
|
||
未激活
|
||
|
||
,在根据 Channel
|
||
|
||
|
||
|
||
是否打开
|
||
|
||
,调用
|
||
|
||
|
||
|
||
```
|
||
ChannelOutboundBuffer#failFlushed(Throwable cause, boolean notify)
|
||
```
|
||
|
||
|
||
|
||
方法,通知 flush 失败
|
||
|
||
异常
|
||
|
||
。详细解析,见
|
||
|
||
|
||
|
||
「8.6 failFlushed」
|
||
|
||
|
||
|
||
。
|
||
|
||
- 第 29 至 33 行:最终,设置 `inFlush0` 为 `false` ,表示结束 flush 操作,最后 `return` 返回。
|
||
|
||
- 第 38 行:调用
|
||
|
||
|
||
|
||
```
|
||
AbstractChannel#doWrite(outboundBuffer)
|
||
```
|
||
|
||
|
||
|
||
方法,
|
||
|
||
执行真正的写入到对端
|
||
|
||
。详细解析,见
|
||
|
||
|
||
|
||
「7. NioSocketChannel」
|
||
|
||
|
||
|
||
。
|
||
|
||
- 第 39 至 57 行:TODO 芋艿 细节
|
||
- 第 58 至 61 行:同【第 29 至 33】的代码和目的。
|
||
|
||
- 实际上,AbstractNioUnsafe **重写**了 `#flush0()` 方法,代码如下:
|
||
|
||
```
|
||
@Override
|
||
protected final void flush0() {
|
||
// Flush immediately only when there's no pending flush.
|
||
// If there's a pending flush operation, event loop will call forceFlush() later,
|
||
// and thus there's no need to call it now.
|
||
if (!isFlushPending()) {
|
||
super.flush0();
|
||
}
|
||
}
|
||
```
|
||
|
||
- 在执行父类 AbstractUnsafe 的 `#flush0()` 方法时,先调用 `AbstractNioUnsafe#isFlushPending()` 判断,是否已经处于 flush **准备**中。代码如下:
|
||
|
||
```
|
||
private boolean isFlushPending() {
|
||
SelectionKey selectionKey = selectionKey();
|
||
return selectionKey.isValid() // 合法
|
||
&& (selectionKey.interestOps() & SelectionKey.OP_WRITE) != 0; // 对 SelectionKey.OP_WRITE 事件不感兴趣。
|
||
}
|
||
```
|
||
|
||
- 是不是有点懵 x ?在文初,我们提到:“所以在 Netty 的实现中,默认 Channel 是**可写**的,当写入失败的时候,再去注册 `SelectionKey.OP_WRITE` 事件。这意味着什么呢?在 `#flush()` 方法中,如果写入数据到 Channel 失败,会通过注册 `SelectionKey.OP_WRITE` 事件,然后在轮询到 Channel **可写** 时,再“回调” `#forceFlush()` 方法”。
|
||
- 这就是这段代码的目的,如果处于对 `SelectionKey.OP_WRITE` 事件感兴趣,说明 Channel 此时是**不可写**的,那么调用父类 AbstractUnsafe 的 `#flush0()` 方法,**也没有意义**,所以就不调用。
|
||
- 😈 逻辑上,略微有点复杂,胖友好好理解下。
|
||
|
||
# 7. NioSocketChannel
|
||
|
||
`AbstractChannel#doWrite(ChannelOutboundBuffer in)` **抽象**方法,**执行真正的写入到对端**。定义在 AbstractChannel **抽象**类中,代码如下:
|
||
|
||
```
|
||
/**
|
||
* Flush the content of the given buffer to the remote peer.
|
||
*/
|
||
protected abstract void doWrite(ChannelOutboundBuffer in) throws Exception;
|
||
```
|
||
|
||
------
|
||
|
||
NioSocketChannel 对该**抽象**方法,实现代码如下:
|
||
|
||
```
|
||
1: @Override
|
||
2: protected void doWrite(ChannelOutboundBuffer in) throws Exception {
|
||
3: SocketChannel ch = javaChannel();
|
||
4: // 获得自旋写入次数
|
||
5: int writeSpinCount = config().getWriteSpinCount();
|
||
6: do {
|
||
7: // 内存队列为空,结束循环,直接返回
|
||
8: if (in.isEmpty()) {
|
||
9: // 取消对 SelectionKey.OP_WRITE 的感兴趣
|
||
10: // All written so clear OP_WRITE
|
||
11: clearOpWrite();
|
||
12: // Directly return here so incompleteWrite(...) is not called.
|
||
13: return;
|
||
14: }
|
||
15:
|
||
16: // 获得每次写入的最大字节数
|
||
17: // Ensure the pending writes are made of ByteBufs only.
|
||
18: int maxBytesPerGatheringWrite = ((NioSocketChannelConfig) config).getMaxBytesPerGatheringWrite();
|
||
19: // 从内存队列中,获得要写入的 ByteBuffer 数组
|
||
20: ByteBuffer[] nioBuffers = in.nioBuffers(1024, maxBytesPerGatheringWrite);
|
||
21: // 写入的 ByteBuffer 数组的个数
|
||
22: int nioBufferCnt = in.nioBufferCount();
|
||
23:
|
||
24: // 写入 ByteBuffer 数组,到对端
|
||
25: // Always us nioBuffers() to workaround data-corruption.
|
||
26: // See https://github.com/netty/netty/issues/2761
|
||
27: switch (nioBufferCnt) {
|
||
28: case 0:
|
||
29: // 芋艿 TODO 1014 扣 doWrite0 的细节
|
||
30: // We have something else beside ByteBuffers to write so fallback to normal writes.
|
||
31: writeSpinCount -= doWrite0(in);
|
||
32: break;
|
||
33: case 1: {
|
||
34: // Only one ByteBuf so use non-gathering write
|
||
35: // Zero length buffers are not added to nioBuffers by ChannelOutboundBuffer, so there is no need
|
||
36: // to check if the total size of all the buffers is non-zero.
|
||
37: ByteBuffer buffer = nioBuffers[0];
|
||
38: int attemptedBytes = buffer.remaining();
|
||
39: // 执行 NIO write 调用,写入单个 ByteBuffer 对象到对端
|
||
40: final int localWrittenBytes = ch.write(buffer);
|
||
41: // 写入字节小于等于 0 ,说明 NIO Channel 不可写,所以注册 SelectionKey.OP_WRITE ,等待 NIO Channel 可写,并返回以结束循环
|
||
42: if (localWrittenBytes <= 0) {
|
||
43: incompleteWrite(true);
|
||
44: return;
|
||
45: }
|
||
46: // TODO 芋艿 调整每次写入的最大字节数
|
||
47: adjustMaxBytesPerGatheringWrite(attemptedBytes, localWrittenBytes, maxBytesPerGatheringWrite);
|
||
48: // 从内存队列中,移除已经写入的数据( 消息 )
|
||
49: in.removeBytes(localWrittenBytes);
|
||
50: // 写入次数减一
|
||
51: --writeSpinCount;
|
||
52: break;
|
||
53: }
|
||
54: default: {
|
||
55: // Zero length buffers are not added to nioBuffers by ChannelOutboundBuffer, so there is no need
|
||
56: // to check if the total size of all the buffers is non-zero.
|
||
57: // We limit the max amount to int above so cast is safe
|
||
58: long attemptedBytes = in.nioBufferSize();
|
||
59: // 执行 NIO write 调用,写入多个 ByteBuffer 到对端
|
||
60: final long localWrittenBytes = ch.write(nioBuffers, 0, nioBufferCnt);
|
||
61: // 写入字节小于等于 0 ,说明 NIO Channel 不可写,所以注册 SelectionKey.OP_WRITE ,等待 NIO Channel 可写,并返回以结束循环
|
||
62: if (localWrittenBytes <= 0) {
|
||
63: incompleteWrite(true);
|
||
64: return;
|
||
65: }
|
||
66: // TODO 芋艿 调整每次写入的最大字节数
|
||
67: // Casting to int is safe because we limit the total amount of data in the nioBuffers to int above.
|
||
68: adjustMaxBytesPerGatheringWrite((int) attemptedBytes, (int) localWrittenBytes, maxBytesPerGatheringWrite);
|
||
69: // 从内存队列中,移除已经写入的数据( 消息 )
|
||
70: in.removeBytes(localWrittenBytes);
|
||
71: // 写入次数减一
|
||
72: --writeSpinCount;
|
||
73: break;
|
||
74: }
|
||
75: }
|
||
76: } while (writeSpinCount > 0); // 循环自旋写入
|
||
77:
|
||
78: // 内存队列中的数据未完全写入,说明 NIO Channel 不可写,所以注册 SelectionKey.OP_WRITE ,等待 NIO Channel 可写
|
||
79: incompleteWrite(writeSpinCount < 0);
|
||
80: }
|
||
```
|
||
|
||
- 第 3 行:调用 `#javaChannel()` 方法,获得 Java NIO **原生** SocketChannel 。
|
||
|
||
- 第 5 行:调用 `ChannelConfig#getWriteSpinCount()` 方法,获得**自旋**写入次数 N 。在【第 6 至 76 行】的代码,我们可以看到,不断**自旋**写入 N 次,直到完成写入结束。关于该配置项,官方注释如下:
|
||
|
||
```
|
||
/**
|
||
* Returns the maximum loop count for a write operation until {@link WritableByteChannel#write(ByteBuffer)} returns a non-zero value.
|
||
* It is similar to what a spin lock is used for in concurrency programming.
|
||
* It improves memory utilization and write throughput depending on the platform that JVM runs on. The default value is {@code 16}.
|
||
*/
|
||
int getWriteSpinCount();
|
||
```
|
||
|
||
- 默认值为 `DefaultChannelConfig.writeSpinCount = 16` ,可配置修改,一般不需要。
|
||
|
||
- 第 6 至 76 行:不断**自旋**写入 N 次,直到完成写入结束。
|
||
|
||
- 第 8 行:调用 `ChannelOutboundBuffer#isEmpty()` 方法,内存队列为空,结束循环,直接返回。
|
||
|
||
- 第 10 行:因为在 Channel **不可写**的时候,会注册 `SelectionKey.OP_WRITE` ,等待 NIO Channel 可写。而后会”回调” `#forceFlush()` 方法,该方法内部也会调用 `#doWrite(ChannelOutboundBuffer in)` 方法。所以在完成内部队列的数据向对端写入时候,需要调用 `#clearOpWrite()` 方法,代码如下:
|
||
|
||
```
|
||
protected final void clearOpWrite() {
|
||
final SelectionKey key = selectionKey();
|
||
// Check first if the key is still valid as it may be canceled as part of the deregistration
|
||
// from the EventLoop
|
||
// See https://github.com/netty/netty/issues/2104
|
||
if (!key.isValid()) { // 合法
|
||
return;
|
||
}
|
||
final int interestOps = key.interestOps();
|
||
// 若注册了 SelectionKey.OP_WRITE ,则进行取消
|
||
if ((interestOps & SelectionKey.OP_WRITE) != 0) {
|
||
key.interestOps(interestOps & ~SelectionKey.OP_WRITE);
|
||
}
|
||
}
|
||
```
|
||
|
||
- 😈 胖友看下代码注释。
|
||
|
||
- 第 18 行:调用 `NioSocketChannelConfig#getMaxBytesPerGatheringWrite()` 方法,获得每次写入的最大字节数。// TODO 芋艿 调整每次写入的最大字节数
|
||
|
||
- 第 20 行:调用
|
||
|
||
|
||
|
||
```
|
||
ChannelOutboundBuffer#nioBuffers(int maxCount, long maxBytes)
|
||
```
|
||
|
||
|
||
|
||
方法,从内存队列中,获得要写入的 ByteBuffer 数组。
|
||
|
||
注意
|
||
|
||
,如果内存队列中数据量很大,可能获得的仅仅是一部分数据。详细解析,见
|
||
|
||
|
||
|
||
「8.5 nioBuffers」
|
||
|
||
|
||
|
||
。
|
||
|
||
- 第 22 行:获得写入的 ByteBuffer 数组的个数。为什么不直接调用数组的 `#length()` 方法呢?因为返回的 ByteBuffer 数组是**预先生成的数组缓存**,存在不断重用的情况,所以不能直接使用 `#length()` 方法,而是要调用 `ChannelOutboundBuffer#nioBufferCount()` 方法,获得写入的 ByteBuffer 数组的个数。详细解析,见 [「8.5 nioBuffers」](https://svip.iocoder.cn/Netty/Channel-5-flush/#) 。
|
||
- 后续根据 `nioBufferCnt` 的数值,分成**三种**情况。
|
||
|
||
- **(づ ̄3 ̄)づ╭❤~ 第一种**,`nioBufferCnt = 0` 。
|
||
|
||
- 芋艿 TODO 1014 扣 doWrite0 的细节,应该是内部的数据为 FileRegion ,可以暂时无视,不影响本文理解。
|
||
|
||
- **(づ ̄3 ̄)づ╭❤~ 第二种**,`nioBufferCnt = 1` 。
|
||
|
||
- 第 40 行:调用 Java **原生** `SocketChannel#write(ByteBuffer buffer)` 方法,执行 NIO write 调用,写入**单个** ByteBuffer 对象到对端。
|
||
|
||
- 第 42 行:写入字节小于等于 0 ,说明 NIO Channel **不可写**,所以注册 `SelectionKey.OP_WRITE` ,等待 NIO Channel **可写**,并返回以结束循环。
|
||
|
||
- 第 43 行:调用 `AbstractNioByteChannel#incompleteWrite(true)` 方法,代码如下:
|
||
|
||
```
|
||
protected final void incompleteWrite(boolean setOpWrite) {
|
||
// Did not write completely.
|
||
// true ,注册对 SelectionKey.OP_WRITE 事件感兴趣
|
||
if (setOpWrite) {
|
||
setOpWrite();
|
||
// false ,取消对 SelectionKey.OP_WRITE 事件感兴趣
|
||
} else {
|
||
// It is possible that we have set the write OP, woken up by NIO because the socket is writable, and then
|
||
// use our write quantum. In this case we no longer want to set the write OP because the socket is still
|
||
// writable (as far as we know). We will find out next time we attempt to write if the socket is writable
|
||
// and set the write OP if necessary.
|
||
clearOpWrite();
|
||
|
||
// Schedule flush again later so other tasks can be picked up in the meantime
|
||
// 立即发起下一次 flush 任务
|
||
eventLoop().execute(flushTask); // <1>
|
||
}
|
||
}
|
||
```
|
||
|
||
- `setOpWrite` 为 `true` ,调用 `#setOpWrite()` 方法,注册对 `SelectionKey.OP_WRITE` 事件感兴趣。代码如下:
|
||
|
||
```
|
||
protected final void setOpWrite() {
|
||
final SelectionKey key = selectionKey();
|
||
// Check first if the key is still valid as it may be canceled as part of the deregistration
|
||
// from the EventLoop
|
||
// See https://github.com/netty/netty/issues/2104
|
||
if (!key.isValid()) { // 合法
|
||
return;
|
||
}
|
||
final int interestOps = key.interestOps();
|
||
// 注册 SelectionKey.OP_WRITE 事件的感兴趣
|
||
if ((interestOps & SelectionKey.OP_WRITE) == 0) {
|
||
key.interestOps(interestOps | SelectionKey.OP_WRITE);
|
||
}
|
||
}
|
||
```
|
||
|
||
- 【第 43 行】的代码,就是这种情况。
|
||
|
||
- `setOpWrite` 为 `false` ,调用 `#clearOpWrite()` 方法,取消对 SelectionKey.OP_WRITE 事件感兴趣。而后,在 `<1>` 处,立即发起下一次 flush 任务。
|
||
|
||
- 第 47 行:TODO 芋艿 调整每次写入的最大字节数
|
||
|
||
- 第 49 行:调用 `ChannelOutboundBuffer#removeBytes(long writtenBytes)` 方法啊,从内存队列中,移除已经写入的数据( 消息 )。详细解析,见 [「8.7 removeBytes」](https://svip.iocoder.cn/Netty/Channel-5-flush/#) 。
|
||
|
||
- 第 51 行:写入次数减一。
|
||
|
||
- **(づ ̄3 ̄)づ╭❤~ 第三种**,`nioBufferCnt > 1` 。和【第二种】基本相同,差别是在于【第 60 行】的代码,调用 Java **原生** `SocketChannel#write(ByteBuffer[] srcs, int offset, int length)` 方法,执行 NIO write 调用,写入**多个** ByteBuffer 对象到对端。😈 批量一次性写入,提升性能。
|
||
|
||
- =========== 结束 ===========
|
||
|
||
- 第 79 行:通过
|
||
|
||
|
||
|
||
```
|
||
writeSpinCount < 0
|
||
```
|
||
|
||
|
||
|
||
来判断,内存队列中的数据
|
||
|
||
是否
|
||
|
||
未完全写入。从目前逻辑看下来,笔者认为只会返回
|
||
|
||
|
||
|
||
```
|
||
true
|
||
```
|
||
|
||
|
||
|
||
,即内存队列中的数据未完全写入,说明 NIO Channel 不可写,所以注册
|
||
|
||
|
||
|
||
```
|
||
SelectionKey.OP_WRITE
|
||
```
|
||
|
||
|
||
|
||
,等待 NIO Channel 可写。因此,调用
|
||
|
||
|
||
|
||
```
|
||
#incompleteWrite(true)
|
||
```
|
||
|
||
|
||
|
||
方法。
|
||
|
||
- 举个例子,最后一次写入,Channel 的缓冲区还剩下 10 字节可写,内存队列中剩余 90 字节,那么可以成功写入 10 字节,剩余 80 字节。😈 也就说,此时 Channel 不可写落。
|
||
|
||
## 7.1 乱入
|
||
|
||
> 老艿艿:临时插入下 AbstractNioByteChannel 和 AbstractNioMessageChannel 实现类对 `#doWrite(ChannelOutboundBuffer in)` 方法的实现。不感兴趣的胖友,可以直接跳过。
|
||
|
||
**AbstractNioByteChannel**
|
||
|
||
虽然,AbstractNioByteChannel 实现了 `#doWrite(ChannelOutboundBuffer in)` 方法,但是子类 NioSocketChannel 又覆盖实现了该方法,所以可以忽略 AbstractNioByteChannel 的实现方法了。
|
||
|
||
那么为什么 AbstractNioByteChannel 会实现了 `#doWrite(ChannelOutboundBuffer in)` 方法呢?因为 NioUdtByteConnectorChannel 和 NioUdtByteRendezvousChannel 会使用到该方法。但是,这两个类已经被**标记废弃**,因为:
|
||
|
||
```
|
||
transport udt is deprecated and so the user knows it will be removed in the future.
|
||
```
|
||
|
||
- 来自 Netty 官方提交的注释说明。
|
||
|
||
**AbstractNioMessageChannel**
|
||
|
||
虽然,AbstractNioMessageChannel 实现了 `#doWrite(ChannelOutboundBuffer in)` 方法,但是对于 NioServerSocketChannel 来说,暂时没有意义,因为:
|
||
|
||
```
|
||
@Override
|
||
protected boolean doWriteMessage(Object msg, ChannelOutboundBuffer in) throws Exception {
|
||
throw new UnsupportedOperationException();
|
||
}
|
||
|
||
@Override
|
||
protected final Object filterOutboundMessage(Object msg) throws Exception {
|
||
throw new UnsupportedOperationException();
|
||
}
|
||
```
|
||
|
||
- 两个方法,都是直接抛出 UnsupportedOperationException 异常。
|
||
|
||
那么为什么 AbstractNioMessageChannel 会实现了 `#doWrite(ChannelOutboundBuffer in)` 方法呢?因为 NioDatagramChannel 和 NioSctpChannel **等等**会使用到该方法。感兴趣的胖友,可以自己研究下。
|
||
|
||
# 8. ChannelOutboundBuffer
|
||
|
||
`io.netty.channel.ChannelOutboundBuffer` ,**内存队列**。
|
||
|
||
- 在 write 操作时,将数据写到 ChannelOutboundBuffer 中。
|
||
- 在 flush 操作时,将 ChannelOutboundBuffer 的数据写入到对端。
|
||
|
||
## 8.1 Entry
|
||
|
||
在 write 操作时,将数据写到 ChannelOutboundBuffer 中,都会产生一个 Entry 对象。代码如下:
|
||
|
||
```
|
||
/**
|
||
* Recycler 对象,用于重用 Entry 对象
|
||
*/
|
||
private static final Recycler<Entry> RECYCLER = new Recycler<Entry>() {
|
||
@Override
|
||
protected Entry newObject(Handle<Entry> handle) {
|
||
return new Entry(handle);
|
||
}
|
||
};
|
||
|
||
/**
|
||
* Recycler 处理器
|
||
*/
|
||
private final Handle<Entry> handle;
|
||
/**
|
||
* 下一条 Entry
|
||
*/
|
||
Entry next;
|
||
/**
|
||
* 消息(数据)
|
||
*/
|
||
Object msg;
|
||
/**
|
||
* {@link #msg} 转化的 NIO ByteBuffer 数组
|
||
*/
|
||
ByteBuffer[] bufs;
|
||
/**
|
||
* {@link #msg} 转化的 NIO ByteBuffer 对象
|
||
*/
|
||
ByteBuffer buf;
|
||
/**
|
||
* Promise 对象
|
||
*/
|
||
ChannelPromise promise;
|
||
/**
|
||
* 已写入的字节数
|
||
*/
|
||
long progress;
|
||
/**
|
||
* 长度,可读字节数数。
|
||
*/
|
||
long total;
|
||
/**
|
||
* 每个 Entry 预计占用的内存大小,计算方式为消息( {@link #msg} )的字节数 + Entry 对象自身占用内存的大小。
|
||
*/
|
||
int pendingSize;
|
||
/**
|
||
* {@link #msg} 转化的 NIO ByteBuffer 的数量。
|
||
*
|
||
* 当 = 1 时,使用 {@link #buf}
|
||
* 当 > 1 时,使用 {@link #bufs}
|
||
*/
|
||
int count = -1;
|
||
/**
|
||
* 是否取消写入对端
|
||
*/
|
||
boolean cancelled;
|
||
|
||
private Entry(Handle<Entry> handle) {
|
||
this.handle = handle;
|
||
}
|
||
```
|
||
|
||
- ```
|
||
RECYCLER
|
||
```
|
||
|
||
|
||
|
||
静态
|
||
|
||
属性,用于
|
||
|
||
重用
|
||
|
||
|
||
|
||
Entry 对象。
|
||
|
||
- `handle` 属性,Recycler 处理器,用于**回收** Entry 对象。
|
||
|
||
- `next` 属性,指向**下一条** Entry 。通过它,形成 ChannelOutboundBuffer 内部的链式存储**每条写入数据**的数据结构。
|
||
|
||
- `msg` 属性,写入的消息( 数据 )。
|
||
|
||
- `promise` 属性,Promise 对象。当数据写入成功后,可以通过它回调通知结果。
|
||
|
||
- `total` 属性,长度,可读字节数。通过 `#total(Object msg)` 方法来计算。代码如下:
|
||
|
||
```
|
||
private static long total(Object msg) {
|
||
if (msg instanceof ByteBuf) {
|
||
return ((ByteBuf) msg).readableBytes();
|
||
}
|
||
if (msg instanceof FileRegion) {
|
||
return ((FileRegion) msg).count();
|
||
}
|
||
if (msg instanceof ByteBufHolder) {
|
||
return ((ByteBufHolder) msg).content().readableBytes();
|
||
}
|
||
return -1;
|
||
}
|
||
```
|
||
|
||
- 从这个方法,我们看到,`msg` 的类型,有 ByteBuf、FileRegion、ByteBufHolder 。
|
||
|
||
- `process` 属性,已写入的字节数。详细解析,见 [「8.7.1 process」](https://svip.iocoder.cn/Netty/Channel-5-flush/#) 。
|
||
|
||
- ```
|
||
count
|
||
```
|
||
|
||
|
||
|
||
属性,
|
||
|
||
```
|
||
msg
|
||
```
|
||
|
||
|
||
|
||
属性转化的 NIO ByteBuffer 的数量。
|
||
|
||
- `bufs` 属性,当 `count > 0` 时使用,表示 `msg` 属性转化的 NIO ByteBuffer 数组。
|
||
- `buf` 属性,当 `count = 0` 时使用,表示 `msg` 属性转化的 NIO ByteBuffer 对象。
|
||
|
||
- `cancelled` 属性,是否取消写入对端。
|
||
|
||
- `pendingSize` 属性,每个 Entry 预计占用的内存大小,计算方式为消息( `msg` )的字节数 + Entry 对象自身占用内存的大小。
|
||
|
||
### 8.1.1 newInstance
|
||
|
||
`#newInstance(Object msg, int size, long total, ChannelPromise promise)` **静态**方法,创建 Entry 对象。代码如下:
|
||
|
||
```
|
||
static Entry newInstance(Object msg, int size, long total, ChannelPromise promise) {
|
||
// 通过 Recycler 重用对象
|
||
Entry entry = RECYCLER.get();
|
||
// 初始化属性
|
||
entry.msg = msg;
|
||
entry.pendingSize = size + CHANNEL_OUTBOUND_BUFFER_ENTRY_OVERHEAD;
|
||
entry.total = total;
|
||
entry.promise = promise;
|
||
return entry;
|
||
}
|
||
```
|
||
|
||
- 通过 Recycler 来**重用** Entry 对象。
|
||
|
||
### 8.1.2 recycle
|
||
|
||
`#recycle()` 方法,**回收** Entry 对象,以为下次**重用**该对象。代码如下:
|
||
|
||
```
|
||
void recycle() {
|
||
// 重置属性
|
||
next = null;
|
||
bufs = null;
|
||
buf = null;
|
||
msg = null;
|
||
promise = null;
|
||
progress = 0;
|
||
total = 0;
|
||
pendingSize = 0;
|
||
count = -1;
|
||
cancelled = false;
|
||
// 回收 Entry 对象
|
||
handle.recycle(this);
|
||
}
|
||
```
|
||
|
||
### 8.1.3 recycleAndGetNext
|
||
|
||
`#recycleAndGetNext()` 方法,获得下一个 Entry 对象,并**回收**当前 Entry 对象。代码如下:
|
||
|
||
```
|
||
Entry recycleAndGetNext() {
|
||
// 获得下一个 Entry 对象
|
||
Entry next = this.next;
|
||
// 回收当前 Entry 对象
|
||
recycle();
|
||
// 返回下一个 Entry 对象
|
||
return next;
|
||
}
|
||
```
|
||
|
||
### 8.1.4 cancel
|
||
|
||
`#cancel()` 方法,标记 Entry 对象,取消写入到对端。在 ChannelOutboundBuffer 里,Entry 数组是通过**链式**的方式进行组织,而当某个 Entry 对象( **节点** )如果需要取消写入到对端,是通过设置 `canceled = true` 来**标记删除**。代码如下:
|
||
|
||
```
|
||
int cancel() {
|
||
if (!cancelled) {
|
||
// 标记取消
|
||
cancelled = true;
|
||
int pSize = pendingSize;
|
||
|
||
// 释放消息( 数据 )相关的资源
|
||
// release message and replace with an empty buffer
|
||
ReferenceCountUtil.safeRelease(msg);
|
||
// 设置为空 ByteBuf
|
||
msg = Unpooled.EMPTY_BUFFER;
|
||
|
||
// 置空属性
|
||
pendingSize = 0;
|
||
total = 0;
|
||
progress = 0;
|
||
bufs = null;
|
||
buf = null;
|
||
|
||
// 返回 pSize
|
||
return pSize;
|
||
}
|
||
return 0;
|
||
}
|
||
```
|
||
|
||
## 8.2 构造方法
|
||
|
||
```
|
||
/**
|
||
* Entry 对象自身占用内存的大小
|
||
*/
|
||
// Assuming a 64-bit JVM:
|
||
// - 16 bytes object header
|
||
// - 8 reference fields
|
||
// - 2 long fields
|
||
// - 2 int fields
|
||
// - 1 boolean field
|
||
// - padding
|
||
static final int CHANNEL_OUTBOUND_BUFFER_ENTRY_OVERHEAD = SystemPropertyUtil.getInt("io.netty.transport.outboundBufferEntrySizeOverhead", 96);
|
||
|
||
private static final InternalLogger logger = InternalLoggerFactory.getInstance(ChannelOutboundBuffer.class);
|
||
|
||
/**
|
||
* 线程对应的 ByteBuffer 数组缓存
|
||
*
|
||
* 每次调用 {@link #nioBuffers(int, long)} 会重新生成
|
||
*/
|
||
private static final FastThreadLocal<ByteBuffer[]> NIO_BUFFERS = new FastThreadLocal<ByteBuffer[]>() {
|
||
|
||
@Override
|
||
protected ByteBuffer[] initialValue() throws Exception {
|
||
return new ByteBuffer[1024];
|
||
}
|
||
|
||
};
|
||
|
||
/**
|
||
* Channel 对象
|
||
*/
|
||
private final Channel channel;
|
||
|
||
// Entry(flushedEntry) --> ... Entry(unflushedEntry) --> ... Entry(tailEntry)
|
||
//
|
||
/**
|
||
* 第一个( 开始 ) flush Entry
|
||
*/
|
||
// The Entry that is the first in the linked-list structure that was flushed
|
||
private Entry flushedEntry;
|
||
/**
|
||
* 第一个未 flush Entry
|
||
*/
|
||
// The Entry which is the first unflushed in the linked-list structure
|
||
private Entry unflushedEntry;
|
||
/**
|
||
* 尾 Entry
|
||
*/
|
||
// The Entry which represents the tail of the buffer
|
||
private Entry tailEntry;
|
||
/**
|
||
* 已 flush 但未写入对端的 Entry 数量
|
||
*
|
||
* {@link #addFlush()}
|
||
*
|
||
* The number of flushed entries that are not written yet
|
||
*/
|
||
private int flushed;
|
||
|
||
/**
|
||
* {@link #NIO_BUFFERS} 数组大小
|
||
*/
|
||
private int nioBufferCount;
|
||
/**
|
||
* {@link #NIO_BUFFERS} 字节数
|
||
*/
|
||
private long nioBufferSize;
|
||
/**
|
||
* 正在通知 flush 失败中
|
||
*/
|
||
private boolean inFail;
|
||
|
||
/**
|
||
* {@link #totalPendingSize} 的原子更新器
|
||
*/
|
||
private static final AtomicLongFieldUpdater<ChannelOutboundBuffer> TOTAL_PENDING_SIZE_UPDATER = AtomicLongFieldUpdater.newUpdater(ChannelOutboundBuffer.class, "totalPendingSize");
|
||
/**
|
||
* 总共等待 flush 到对端的内存大小,通过 {@link Entry#pendingSize} 来合计。
|
||
*/
|
||
@SuppressWarnings("UnusedDeclaration")
|
||
private volatile long totalPendingSize;
|
||
|
||
/**
|
||
* {@link #unwritable} 的原子更新器
|
||
*/
|
||
private static final AtomicIntegerFieldUpdater<ChannelOutboundBuffer> UNWRITABLE_UPDATER = AtomicIntegerFieldUpdater.newUpdater(ChannelOutboundBuffer.class, "unwritable");
|
||
/**
|
||
* 是否不可写
|
||
*/
|
||
@SuppressWarnings("UnusedDeclaration")
|
||
private volatile int unwritable;
|
||
|
||
/**
|
||
* 触发 Channel 可写的改变的任务
|
||
*/
|
||
private volatile Runnable fireChannelWritabilityChangedTask;
|
||
|
||
ChannelOutboundBuffer(AbstractChannel channel) {
|
||
this.channel = channel;
|
||
}
|
||
```
|
||
|
||
- `channel` 属性,所属的 Channel 对象。
|
||
|
||
- 链式结构
|
||
|
||
- `flushedEntry` 属性,第一个( 开始 ) flush Entry 。
|
||
- `unflushedEntry` 属性,第一个**未** flush Entry 。
|
||
- `tailEntry` 属性,尾 Entry 。
|
||
- `flushed` 属性, 已 flush 但未写入对端的 Entry 数量。
|
||
- 指向关系是 `Entry(flushedEntry) --> ... Entry(unflushedEntry) --> ... Entry(tailEntry)` 。这样看,可能有点抽象,下文源码解析详细理解。
|
||
|
||
- ```
|
||
NIO_BUFFERS
|
||
```
|
||
|
||
|
||
|
||
静态
|
||
|
||
属性,线程对应的 NIO ByteBuffer 数组缓存。在
|
||
|
||
|
||
|
||
```
|
||
AbstractChannel#doWrite(ChannelOutboundBuffer)
|
||
```
|
||
|
||
|
||
|
||
方法中,会调用
|
||
|
||
|
||
|
||
```
|
||
ChannelOutbound#nioBuffers(int maxCount, long maxBytes)
|
||
```
|
||
|
||
|
||
|
||
方法,初始化数组缓存。 详细解析,见
|
||
|
||
|
||
|
||
「8.6 nioBuffers」
|
||
|
||
|
||
|
||
中。
|
||
|
||
- `nioBufferCount` 属性:NIO ByteBuffer 数组的**数组**大小。
|
||
- `nioBufferSize` 属性:NIO ByteBuffer 数组的字**节**大小。
|
||
|
||
- `inFail` 属性,正在通知 flush 失败中。详细解析,见 [「8.8 failFlushed」](https://svip.iocoder.cn/Netty/Channel-5-flush/#) 中。
|
||
|
||
- ChannelOutboundBuffer 写入控制相关。😈 详细解析,见
|
||
|
||
|
||
|
||
「10. ChannelOutboundBuffer」
|
||
|
||
|
||
|
||
。
|
||
|
||
- ```
|
||
unwritable
|
||
```
|
||
|
||
|
||
|
||
属性,是否不可写。
|
||
|
||
- `UNWRITABLE_UPDATER` 静态属性,`unwritable` 属性的原子更新器。
|
||
|
||
- ```
|
||
totalPendingSize
|
||
```
|
||
|
||
|
||
|
||
属性,所有 Entry 预计占用的内存大小,通过
|
||
|
||
|
||
|
||
```
|
||
Entry.pendingSize
|
||
```
|
||
|
||
|
||
|
||
来合计。
|
||
|
||
- `TOTAL_PENDING_SIZE_UPDATER` 静态属性,`totalPendingSize` 属性的原子更新器。
|
||
|
||
- `fireChannelWritabilityChangedTask` 属性,触发 Channel 可写的改变的**任务**。
|
||
|
||
- ```
|
||
CHANNEL_OUTBOUND_BUFFER_ENTRY_OVERHEAD
|
||
```
|
||
|
||
|
||
|
||
静态
|
||
|
||
属性,每个 Entry 对象自身占用内存的大小。为什么占用的 96 字节呢?
|
||
|
||
- `- 16 bytes object header` ,对象头,16 字节。
|
||
- `- 8 reference fields` ,实际是 6 个**对象引用**字段,6 * 8 = 48 字节。
|
||
- `- 2 long fields` ,2 个 `long` 字段,2 * 8 = 16 字节。
|
||
- `- 2 int fields` ,1 个 `int` 字段,2 * 4 = 8 字节。
|
||
- `- 1 boolean field` ,1 个 `boolean` 字段,1 字节。
|
||
- `padding` ,补齐 8 字节的整数倍,因此 7 字节。
|
||
- 因此,合计 96 字节( 64 位的 JVM 虚拟机,并且不考虑压缩 )。
|
||
- 如果不理解的胖友,可以看看 [《JVM中 对象的内存布局 以及 实例分析》](https://www.jianshu.com/p/12a3c97dc2b7) 。
|
||
|
||
## 8.3 addMessage
|
||
|
||
`#addMessage(Object msg, int size, ChannelPromise promise)` 方法,写入消息( 数据 )到内存队列。**注意**,`promise` 只有在真正完成写入到对端操作,才会进行通知。代码如下:
|
||
|
||
```
|
||
1: /**
|
||
2: * Add given message to this {@link ChannelOutboundBuffer}. The given {@link ChannelPromise} will be notified once
|
||
3: * the message was written.
|
||
4: */
|
||
5: public void addMessage(Object msg, int size, ChannelPromise promise) {
|
||
6: // 创建新 Entry 对象
|
||
7: Entry entry = Entry.newInstance(msg, size, total(msg), promise);
|
||
8: // 若 tailEntry 为空,将 flushedEntry 也设置为空。防御型编程,实际不会出现
|
||
9: if (tailEntry == null) {
|
||
10: flushedEntry = null;
|
||
11: // 若 tailEntry 非空,将原 tailEntry 指向新 Entry
|
||
12: } else {
|
||
13: Entry tail = tailEntry;
|
||
14: tail.next = entry;
|
||
15: }
|
||
16: // 更新 tailEntry 为新 Entry
|
||
17: tailEntry = entry;
|
||
18: // 若 unflushedEntry 为空,更新为新 Entry
|
||
19: if (unflushedEntry == null) {
|
||
20: unflushedEntry = entry;
|
||
21: }
|
||
22:
|
||
23: // 增加 totalPendingSize 计数
|
||
24: // increment pending bytes after adding message to the unflushed arrays.
|
||
25: // See https://github.com/netty/netty/issues/1619
|
||
26: incrementPendingOutboundBytes(entry.pendingSize, false);
|
||
27: }
|
||
```
|
||
|
||
- 第 7 行:调用 `#newInstance(Object msg, int size, long total, ChannelPromise promise)` **静态**方法,创建 Entry 对象。
|
||
|
||
- 第 11 至 17 行:修改
|
||
|
||
尾
|
||
|
||
节点
|
||
|
||
|
||
|
||
```
|
||
tailEntry
|
||
```
|
||
|
||
|
||
|
||
为新的 Entry 节点。
|
||
|
||
- 第 8 至 10 行:若 `tailEntry` 为空,将 `flushedEntry` 也设置为空。防御型编程,实际不会出现,胖友可以忽略。😈 当然,原因在 `#removeEntry(Entry e)` 方法。
|
||
- 第 11 至 15 行:若 `tailEntry` 非空,将原 `tailEntry.next` 指向**新** Entry 。
|
||
- 第 17 行:更新原 `tailEntry` 为新 Entry 。
|
||
|
||
- 第 18 至 21 行:若 `unflushedEntry` 为空,则更新为新 Entry ,此时相当于**首**节点。
|
||
|
||
- 第 23 至 26 行:`#incrementPendingOutboundBytes(long size, ...)` 方法,增加 `totalPendingSize` 计数。详细解析,见 [「10.1 incrementPendingOutboundBytes」](https://svip.iocoder.cn/Netty/Channel-5-flush/#) 。
|
||
|
||
------
|
||
|
||
可能有点抽象,我们来看看基友【闪电侠】对这块的解析:
|
||
|
||
> FROM 闪电侠 [《netty 源码分析之 writeAndFlush 全解析》](https://www.jianshu.com/p/feaeaab2ce56)
|
||
>
|
||
> 初次调用 `addMessage` 之后,各个指针的情况为
|
||
>
|
||
> [之 flush 操作.assets/1ff7a5d2b08b9e6160dd92e74e68145f.png)](http://static.iocoder.cn/1ff7a5d2b08b9e6160dd92e74e68145f)
|
||
>
|
||
> `fushedEntry`指向空,`unFushedEntry`和 `tailEntry` 都指向新加入的节点
|
||
>
|
||
> 第二次调用 `addMessage` 之后,各个指针的情况为
|
||
>
|
||
> [之 flush 操作.assets/1f939423f079ff491b90c8300e7ef3ea.png)](http://static.iocoder.cn/1f939423f079ff491b90c8300e7ef3ea)
|
||
>
|
||
> 第 n 次调用 `addMessage`之后,各个指针的情况为
|
||
>
|
||
> [之 flush 操作.assets/c0077b0dc86ecf1b791a99eeb9664fc3.png)](http://static.iocoder.cn/c0077b0dc86ecf1b791a99eeb9664fc3)
|
||
>
|
||
> 可以看到,调用 n 次 `addMessage` ,`flushedEntry` 指针一直指向 NULL ,表示现在还未有节点需要写出到 Socket 缓冲区,而`unFushedEntry` 之后有 n 个节点,表示当前还有n个节点尚未写出到 Socket 缓冲区中去
|
||
|
||
## 8.4 addFlush
|
||
|
||
`#addFlush()` 方法,标记内存队列每个 Entry 对象,开始 **flush** 。代码如下:
|
||
|
||
> 老艿艿:总觉得这个方法名取的有点奇怪,胖友可以直接看英文注释。😈 我“翻译”不好,哈哈哈。
|
||
|
||
```
|
||
1: public void addFlush() {
|
||
2: // There is no need to process all entries if there was already a flush before and no new messages
|
||
3: // where added in the meantime.
|
||
4: //
|
||
5: // See https://github.com/netty/netty/issues/2577
|
||
6: Entry entry = unflushedEntry;
|
||
7: if (entry != null) {
|
||
8: // 若 flushedEntry 为空,赋值为 unflushedEntry ,用于记录第一个( 开始 ) flush 的 Entry 。
|
||
9: if (flushedEntry == null) {
|
||
10: // there is no flushedEntry yet, so start with the entry
|
||
11: flushedEntry = entry;
|
||
12: }
|
||
13: // 计算 flush 的数量,并设置每个 Entry 对应的 Promise 不可取消
|
||
14: do {
|
||
15: // 增加 flushed
|
||
16: flushed ++;
|
||
17: // 设置 Promise 不可取消
|
||
18: if (!entry.promise.setUncancellable()) { // 设置失败
|
||
19: // 减少 totalPending 计数
|
||
20: // Was cancelled so make sure we free up memory and notify about the freed bytes
|
||
21: int pending = entry.cancel();
|
||
22: decrementPendingOutboundBytes(pending, false, true);
|
||
23: }
|
||
24: // 获得下一个 Entry
|
||
25: entry = entry.next;
|
||
26: } while (entry != null);
|
||
27:
|
||
28: // 设置 unflushedEntry 为空,表示所有都 flush
|
||
29: // All flushed so reset unflushedEntry
|
||
30: unflushedEntry = null;
|
||
31: }
|
||
32: }
|
||
```
|
||
|
||
- 第 6 至 7 行:若 `unflushedEntry` 为空,说明每个 Entry 对象已经“标记” flush 。**注意**,“标记”的方式,不是通过 Entry 对象有一个 `flushed` 字段,而是 `flushedEntry` 属性,指向第一个( 开始 ) flush 的 Entry ,而 `unflushedEntry` 置空。
|
||
|
||
- 第 8 至 12 行:若 `flushedEntry` 为空,赋值为 `unflushedEntry` ,用于记录第一个( 开始 ) flush 的 Entry 。
|
||
|
||
- 第 13 至 26 行:计算需要 flush 的 Entry 数量,并设置每个 Entry 对应的 Promise
|
||
|
||
|
||
|
||
不可取消
|
||
|
||
。
|
||
|
||
- 第 18 至 23 行:`#decrementPendingOutboundBytes(long size, ...)` 方法,减少 `totalPendingSize` 计数。
|
||
|
||
- 第 30 行:设置 `unflushedEntry` 为空。
|
||
|
||
------
|
||
|
||
可能有点抽象,我们来看看基友【闪电侠】对这块的解析:
|
||
|
||
> FROM 闪电侠 [《netty 源码分析之 writeAndFlush 全解析》](https://www.jianshu.com/p/feaeaab2ce56)
|
||
>
|
||
> 可以结合前面的图来看,首先拿到 `unflushedEntry` 指针,然后将 `flushedEntry` 指向 `unflushedEntry` 所指向的节点,调用完毕之后,三个指针的情况如下所示
|
||
>
|
||
> [之 flush 操作.assets/ecb3df153a3df70464b524838b559232.png)](http://static.iocoder.cn/ecb3df153a3df70464b524838b559232)
|
||
|
||
------
|
||
|
||
> 老艿艿:再次切回到老艿艿的频道,呼呼。
|
||
|
||
当一次需要从内存队列写到对端的数据量非常大,那么可能写着写着 Channel 的缓存区不够,导致 Channel 此时不可写。但是,这一轮 `#addFlush(...)` 标记的 Entry 对象并没有都写到对端。例如,准备写到对端的 Entry 的数量是 `flush = 10` 个,结果只写了 6 个,那么就剩下 `flush = 4` 。
|
||
|
||
但是的但是,`#addMessage(...)` 可能又不断写入新的消息( 数据 )到 ChannelOutboundBuffer 中。那会出现什么情况呢?会“分”成两段:
|
||
|
||
- `<1>` 段:自节点 `flushedEntry` 开始的 `flush` 个 Entry 节点,需要写入到对端。
|
||
- `<2>` 段:自节点 `unFlushedEntry` 开始的 Entry 节点,需要调用 `#addFlush()` 方法,添加到 `<1>` 段中。
|
||
|
||
这就很好的解释两个事情:
|
||
|
||
1. 为什么 `#addFlush()` 方法,命名是以 `"add"` 开头。
|
||
2. ChannelOutboundBuffer 的链式结构,为什么不是 `head` 和 `tail` **两个**节点,而是 `flushedEntry`、`unFlushedEntry`、`flushedEntry` **三个**节点。在此处,请允许老艿艿爆个粗口:真他 x 的巧妙啊。
|
||
|
||
### 8.4.1 size
|
||
|
||
`#size()` 方法,获得 `flushed` 属性。代码如下:
|
||
|
||
```
|
||
/**
|
||
* Returns the number of flushed messages in this {@link ChannelOutboundBuffer}.
|
||
*/
|
||
public int size() {
|
||
return flushed;
|
||
}
|
||
```
|
||
|
||
### 8.4.2 isEmpty
|
||
|
||
`#isEmpty()` 方法,是否为空。代码如下:
|
||
|
||
```
|
||
/**
|
||
* Returns {@code true} if there are flushed messages in this {@link ChannelOutboundBuffer} or {@code false}
|
||
* otherwise.
|
||
*/
|
||
public boolean isEmpty() {
|
||
return flushed == 0;
|
||
}
|
||
```
|
||
|
||
## 8.5 current
|
||
|
||
`#current()` 方法,获得**当前**要写入对端的消息( 数据 )。代码如下:
|
||
|
||
```
|
||
/**
|
||
* Return the current message to write or {@code null} if nothing was flushed before and so is ready to be written.
|
||
*/
|
||
public Object current() {
|
||
Entry entry = flushedEntry;
|
||
if (entry == null) {
|
||
return null;
|
||
}
|
||
|
||
return entry.msg;
|
||
}
|
||
```
|
||
|
||
- 即,返回的是 `flushedEntry` 的消息( 数据 )。
|
||
|
||
## 8.6 nioBuffers
|
||
|
||
`#nioBuffers(int maxCount, long maxBytes)` 方法,获得当前要写入到对端的 NIO ByteBuffer 数组,并且获得的数组大小不得超过 `maxCount` ,字节数不得超过 `maxBytes` 。我们知道,在写入数据到 ChannelOutboundBuffer 时,一般使用的是 Netty ByteBuf 对象,但是写到 NIO SocketChannel 时,则必须使用 NIO ByteBuffer 对象,因此才有了这个方法。考虑到性能,这个方法里会使用到“**缓存**”,所以看起来会比较绕一丢丢。OK,开始看代码落:
|
||
|
||
```
|
||
/**
|
||
* Returns an array of direct NIO buffers if the currently pending messages are made of {@link ByteBuf} only.
|
||
* {@link #nioBufferCount()} and {@link #nioBufferSize()} will return the number of NIO buffers in the returned
|
||
* array and the total number of readable bytes of the NIO buffers respectively.
|
||
* <p>
|
||
* Note that the returned array is reused and thus should not escape
|
||
* {@link AbstractChannel#doWrite(ChannelOutboundBuffer)}.
|
||
* Refer to {@link NioSocketChannel#doWrite(ChannelOutboundBuffer)} for an example.
|
||
* </p>
|
||
* @param maxCount The maximum amount of buffers that will be added to the return value.
|
||
* @param maxBytes A hint toward the maximum number of bytes to include as part of the return value. Note that this
|
||
* value maybe exceeded because we make a best effort to include at least 1 {@link ByteBuffer}
|
||
* in the return value to ensure write progress is made.
|
||
*/
|
||
|
||
1: public ByteBuffer[] nioBuffers(int maxCount, long maxBytes) {
|
||
2: assert maxCount > 0;
|
||
3: assert maxBytes > 0;
|
||
4: long nioBufferSize = 0;
|
||
5: int nioBufferCount = 0;
|
||
6: // 获得当前线程的 NIO ByteBuffer 数组缓存。
|
||
7: final InternalThreadLocalMap threadLocalMap = InternalThreadLocalMap.get();
|
||
8: ByteBuffer[] nioBuffers = NIO_BUFFERS.get(threadLocalMap);
|
||
9: // 从 flushedEntry 节点,开始向下遍历
|
||
10: Entry entry = flushedEntry;
|
||
11: while (isFlushedEntry(entry) && entry.msg instanceof ByteBuf) {
|
||
12: // 若 Entry 节点已经取消,忽略。
|
||
13: if (!entry.cancelled) {
|
||
14: ByteBuf buf = (ByteBuf) entry.msg;
|
||
15: // 获得消息( 数据 )开始读取位置
|
||
16: final int readerIndex = buf.readerIndex();
|
||
17: // 获得消息( 数据 )可读取的字节数
|
||
18: final int readableBytes = buf.writerIndex() - readerIndex;
|
||
19:
|
||
20: // 若无可读取的数据,忽略。
|
||
21: if (readableBytes > 0) {
|
||
22: // 前半段,可读取的字节数,不能超过 maxBytes
|
||
23: // 后半段,如果第一条数据,就已经超过 maxBytes ,那么只能“强行”读取,否则会出现一直无法读取的情况。
|
||
24: if (maxBytes - readableBytes < nioBufferSize && nioBufferCount != 0) {
|
||
25: // If the nioBufferSize + readableBytes will overflow maxBytes, and there is at least one entry
|
||
26: // we stop populate the ByteBuffer array. This is done for 2 reasons:
|
||
27: // 1. bsd/osx don't allow to write more bytes then Integer.MAX_VALUE with one writev(...) call
|
||
28: // and so will return 'EINVAL', which will raise an IOException. On Linux it may work depending
|
||
29: // on the architecture and kernel but to be safe we also enforce the limit here.
|
||
30: // 2. There is no sense in putting more data in the array than is likely to be accepted by the
|
||
31: // OS.
|
||
32: //
|
||
33: // See also:
|
||
34: // - https://www.freebsd.org/cgi/man.cgi?query=write&sektion=2
|
||
35: // - http://linux.die.net/man/2/writev
|
||
36: break;
|
||
37: }
|
||
38: // 增加 nioBufferSize
|
||
39: nioBufferSize += readableBytes;
|
||
40: // 初始 Entry 节点的 NIO ByteBuffer 数量
|
||
41: int count = entry.count;
|
||
42: if (count == -1) {
|
||
43: //noinspection ConstantValueVariableUse
|
||
44: entry.count = count = buf.nioBufferCount();
|
||
45: }
|
||
46: // 如果超过 NIO ByteBuffer 数组的大小,进行扩容。
|
||
47: int neededSpace = min(maxCount, nioBufferCount + count);
|
||
48: if (neededSpace > nioBuffers.length) {
|
||
49: nioBuffers = expandNioBufferArray(nioBuffers, neededSpace, nioBufferCount);
|
||
50: NIO_BUFFERS.set(threadLocalMap, nioBuffers);
|
||
51: }
|
||
52: // 初始化 Entry 节点的 buf / bufs 属性
|
||
53: if (count == 1) {
|
||
54: ByteBuffer nioBuf = entry.buf;
|
||
55: if (nioBuf == null) {
|
||
56: // cache ByteBuffer as it may need to create a new ByteBuffer instance if its a
|
||
57: // derived buffer
|
||
58: entry.buf = nioBuf = buf.internalNioBuffer(readerIndex, readableBytes);
|
||
59: }
|
||
60: nioBuffers[nioBufferCount++] = nioBuf;
|
||
61: } else {
|
||
62: ByteBuffer[] nioBufs = entry.bufs;
|
||
63: if (nioBufs == null) {
|
||
64: // cached ByteBuffers as they may be expensive to create in terms
|
||
65: // of Object allocation
|
||
66: entry.bufs = nioBufs = buf.nioBuffers();
|
||
67: }
|
||
68: for (int i = 0; i < nioBufs.length && nioBufferCount < maxCount; ++i) {
|
||
69: ByteBuffer nioBuf = nioBufs[i];
|
||
70: if (nioBuf == null) {
|
||
71: break;
|
||
72: } else if (!nioBuf.hasRemaining()) {
|
||
73: continue;
|
||
74: }
|
||
75: nioBuffers[nioBufferCount++] = nioBuf;
|
||
76: }
|
||
77: }
|
||
78:
|
||
79: // 到达 maxCount 上限,结束循环。老艿艿的想法,这里最好改成 nioBufferCount >= maxCount ,是有可能会超过的
|
||
80: if (nioBufferCount == maxCount) {
|
||
81: break;
|
||
82: }
|
||
83: }
|
||
84: }
|
||
85:
|
||
86: // 下一个 Entry节点
|
||
87: entry = entry.next;
|
||
88: }
|
||
89:
|
||
90: // 设置 nioBufferCount 和 nioBufferSize 属性
|
||
91: this.nioBufferCount = nioBufferCount;
|
||
92: this.nioBufferSize = nioBufferSize;
|
||
93:
|
||
94: return nioBuffers;
|
||
95: }
|
||
```
|
||
|
||
- 第 4 至 5 行:初始 `nioBufferSize`、`nioBufferCount` 计数。
|
||
|
||
- 第 6 至 8 行:获得当前线程的 NIO ByteBuffer 数组缓存。
|
||
|
||
- 关于 InternalThreadLocalMap 和 FastThreadLocal ,胖友可以暂时忽略,后续的文章,详细解析。
|
||
|
||
- 第 10 至 11 行:从 `flushedEntry` 节点,开始向下遍历。
|
||
|
||
- 调用 `#isFlushedEntry(Entry entry)` 方法,判断是否为已经“标记”为 flush 的 Entry 节点。代码如下:
|
||
|
||
```
|
||
private boolean isFlushedEntry(Entry e) {
|
||
return e != null && e != unflushedEntry;
|
||
}
|
||
```
|
||
|
||
- `e != unflushedEntry` ,就是我们在 [「8.4 addFlush」](https://svip.iocoder.cn/Netty/Channel-5-flush/#) 最后部分讲的,思考下。
|
||
|
||
- `entry.msg instanceof ByteBuf` ,消息( 数据 )类型为 ByteBuf 。实际上,`msg` 的类型也可能是 FileRegion 。如果 ChannelOutboundBuffer 里的消息都是 FileRegion 类型,那就会导致这个方法返回为**空** NIO ByteBuffer 数组。
|
||
|
||
- 第 13 行:若 Entry 节点已经取消,忽略。
|
||
|
||
- 第 14 至 18 行:获得消息( 数据 )开始读取位置和可读取的字节数。
|
||
|
||
- 第 21 行:若无可读取的数据,忽略。
|
||
|
||
- 第 22 至 37 行:
|
||
|
||
- 前半段 `maxBytes - readableBytes < nioBufferSize` ,当前 ByteBuf 可读取的字节数,不能超过 `maxBytes` 。这个比较好理解。
|
||
- 后半段 `nioBufferCount != 0` ,如果**第一条**数据,就已经超过 `maxBytes` ,那么只能“强行”读取,否则会出现一直无法读取的情况( 因为不能跳过这条 😈 )。
|
||
|
||
- 第 39 行:增加 `nioBufferSize` 。
|
||
|
||
- 第 40 至 45 行:调用
|
||
|
||
|
||
|
||
```
|
||
ByteBuf#nioBufferCount()
|
||
```
|
||
|
||
|
||
|
||
方法,初始 Entry 节点的
|
||
|
||
|
||
|
||
```
|
||
count
|
||
```
|
||
|
||
|
||
|
||
属性( NIO ByteBuffer 数量)。
|
||
|
||
- 使用 `count == -1` 的原因是,`Entry.count` 未初始化时,为 `-1` 。
|
||
|
||
- 第 47 至 51 行:如果超过 NIO ByteBuffer 数组的大小,调用 `#expandNioBufferArray(ByteBuffer[] array, int neededSpace, int size)` 方法,进行扩容。详细解析,见 [「8.6.1 expandNioBufferArray」](https://svip.iocoder.cn/Netty/Channel-5-flush/#) 。
|
||
|
||
- 第 52 至 77 行:初始 Entry 节点的
|
||
|
||
|
||
|
||
```
|
||
buf
|
||
```
|
||
|
||
|
||
|
||
或
|
||
|
||
|
||
|
||
```
|
||
bufs
|
||
```
|
||
|
||
|
||
|
||
属性。
|
||
|
||
- 当 `count = 1` 时,调用 `ByteBuf#internalNioBuffer(readerIndex, readableBytes)` 方法,获得 NIO ByteBuffer 对象。
|
||
- 当 `count > 1` 时,调用 `ByteBuf#nioBuffers()` 方法,获得 NIO ByteBuffer 数组。
|
||
- 通过 `nioBuffers[nioBufferCount++] = nioBuf` ,将 NIO ByteBuffer 赋值到结果数组 `nioBuffers` 中,并增加 `nioBufferCount` 。
|
||
|
||
- 第 79 至 82 行:到达 `maxCount` 上限,结束循环。老艿艿的想法,这里最好改成 `nioBufferCount >= maxCount` ,是有可能会超过的。
|
||
|
||
- 第 87 行:**下一个 Entry 节点**。
|
||
|
||
- 第 90 至 92 行:设置 ChannelOutboundBuffer 的 `nioBufferCount` 和 `nioBufferSize` 属性。
|
||
|
||
### 8.6.1 expandNioBufferArray
|
||
|
||
`#expandNioBufferArray(ByteBuffer[] array, int neededSpace, int size)` 方法,进行 NIO ByteBuff 数组的**扩容**。代码如下:
|
||
|
||
```
|
||
private static ByteBuffer[] expandNioBufferArray(ByteBuffer[] array, int neededSpace, int size) {
|
||
// 计算扩容后的数组的大小,按照 2 倍计算
|
||
int newCapacity = array.length;
|
||
do {
|
||
// double capacity until it is big enough
|
||
// See https://github.com/netty/netty/issues/1890
|
||
newCapacity <<= 1;
|
||
|
||
if (newCapacity < 0) {
|
||
throw new IllegalStateException();
|
||
}
|
||
|
||
} while (neededSpace > newCapacity);
|
||
|
||
// 创建新的 ByteBuffer 数组
|
||
ByteBuffer[] newArray = new ByteBuffer[newCapacity];
|
||
|
||
// 复制老的 ByteBuffer 数组到新的 ByteBuffer 数组中
|
||
System.arraycopy(array, 0, newArray, 0, size);
|
||
|
||
return newArray;
|
||
}
|
||
```
|
||
|
||
- 代码比较简单,胖友自己看下注释。
|
||
|
||
### 8.6.2 nioBufferCount
|
||
|
||
`#nioBufferCount()` 方法,返回 `nioBufferCount` 属性。代码如下:
|
||
|
||
```
|
||
/**
|
||
* Returns the number of {@link ByteBuffer} that can be written out of the {@link ByteBuffer} array that was
|
||
* obtained via {@link #nioBuffers()}. This method <strong>MUST</strong> be called after {@link #nioBuffers()}
|
||
* was called.
|
||
*/
|
||
public int nioBufferCount() {
|
||
return nioBufferCount;
|
||
}
|
||
```
|
||
|
||
### 8.6.3 nioBufferSize
|
||
|
||
`#nioBufferSize()` 方法,返回 `nioBufferSize` 属性。代码如下:
|
||
|
||
```
|
||
/**
|
||
* Returns the number of bytes that can be written out of the {@link ByteBuffer} array that was
|
||
* obtained via {@link #nioBuffers()}. This method <strong>MUST</strong> be called after {@link #nioBuffers()}
|
||
* was called.
|
||
*/
|
||
public long nioBufferSize() {
|
||
return nioBufferSize;
|
||
}
|
||
```
|
||
|
||
## 8.7 removeBytes
|
||
|
||
`#removeBytes(long writtenBytes)` 方法,移除已经写入 `writtenBytes` 字节对应的 Entry 对象 / 对象们。代码如下:
|
||
|
||
```
|
||
1: public void removeBytes(long writtenBytes) {
|
||
2: // 循环移除
|
||
3: for (;;) {
|
||
4: // 获得当前消息( 数据 )
|
||
5: Object msg = current();
|
||
6: if (!(msg instanceof ByteBuf)) {
|
||
7: assert writtenBytes == 0;
|
||
8: break;
|
||
9: }
|
||
10:
|
||
11: final ByteBuf buf = (ByteBuf) msg;
|
||
12: // 获得消息( 数据 )开始读取位置
|
||
13: final int readerIndex = buf.readerIndex();
|
||
14: // 获得消息( 数据 )可读取的字节数
|
||
15: final int readableBytes = buf.writerIndex() - readerIndex;
|
||
16:
|
||
17: // 当前消息( 数据 )已被写完到对端
|
||
18: if (readableBytes <= writtenBytes) {
|
||
19: if (writtenBytes != 0) {
|
||
20: // 处理当前消息的 Entry 的写入进度
|
||
21: progress(readableBytes);
|
||
22: // 减小 writtenBytes
|
||
23: writtenBytes -= readableBytes;
|
||
24: }
|
||
25: // 移除当前消息对应的 Entry
|
||
26: remove();
|
||
27: // 当前消息( 数据 )未被写完到对端
|
||
28: } else { // readableBytes > writtenBytes
|
||
29: if (writtenBytes != 0) {
|
||
30: // 标记当前消息的 ByteBuf 的读取位置
|
||
31: buf.readerIndex(readerIndex + (int) writtenBytes);
|
||
32: // 处理当前消息的 Entry 的写入进度
|
||
33: progress(writtenBytes);
|
||
34: }
|
||
35: break;
|
||
36: }
|
||
37: }
|
||
38:
|
||
39: // 清除 NIO ByteBuff 数组的缓存
|
||
40: clearNioBuffers();
|
||
41: }
|
||
```
|
||
|
||
- 第 3 行:
|
||
|
||
循环
|
||
|
||
,移除已经写入
|
||
|
||
|
||
|
||
```
|
||
writtenBytes
|
||
```
|
||
|
||
|
||
|
||
字节对应的 Entry 对象。
|
||
|
||
- 第 5 行:调用 `#current()` 方法,获得当前消息( 数据 )。
|
||
- 第 12 至 15 行:获得消息( 数据 )开始读取位置和可读取的字节数。
|
||
- `<1>` 当前消息( 数据 )**已**被写完到对端。
|
||
- 第 21 行:调用 `#progress(long amount)` 方法,处理当前消息的 Entry 的写入进度。详细解析,见 [「8.7.1 progress」](https://svip.iocoder.cn/Netty/Channel-5-flush/#) 。
|
||
- 第 23 行:减小 `writtenBytes` 。
|
||
- 第 26 行:调用 `#remove()` 方法,移除当前消息对应的 Entry 对象。详细解析,见 [「8.7.2 remove」](https://svip.iocoder.cn/Netty/Channel-5-flush/#) 。
|
||
- `<2》` 当前消息( 数据 )**未**被写完到对端。
|
||
- 第 31 行:调用 `ByteBuf#readerIndex(readerIndex)` 方法,标记当前消息的 ByteBuf 的**读取位置**。
|
||
- 第 33 行:调用 `#progress(long amount)` 方法,处理当前消息的 Entry 的写入进度。
|
||
- 第 35 行:`break` ,结束循环。
|
||
|
||
- 第 40 行:调用 `#clearNioBuffers()` 方法,**清除** NIO ByteBuff 数组的缓存。详细解析,见 [「8.7.4 clearNioBuffers」](https://svip.iocoder.cn/Netty/Channel-5-flush/#) 。
|
||
|
||
### 8.7.1 progress
|
||
|
||
`#progress(long amount)` 方法,处理当前消息的 Entry 的写入进度,主要是**通知** Promise 消息写入的进度。代码如下:
|
||
|
||
```
|
||
/**
|
||
* Notify the {@link ChannelPromise} of the current message about writing progress.
|
||
*/
|
||
1: public void progress(long amount) {
|
||
2: Entry e = flushedEntry;
|
||
3: assert e != null;
|
||
4: ChannelPromise p = e.promise;
|
||
5: if (p instanceof ChannelProgressivePromise) {
|
||
6: // 设置 Entry 对象的 progress 属性
|
||
7: long progress = e.progress + amount;
|
||
8: e.progress = progress;
|
||
9: // 通知 ChannelProgressivePromise 进度
|
||
10: ((ChannelProgressivePromise) p).tryProgress(progress, e.total);
|
||
11: }
|
||
12: }
|
||
```
|
||
|
||
- 第 5 行:若 `promise` 的类型是 ChannelProgressivePromise 类型。
|
||
- 第 6 至 8 行:设置 Entry 对象的 `progress` 属性。
|
||
- 第 10 行:调用 `ChannelProgressivePromise#tryProgress(progress, total)` 方法,通知 ChannelProgressivePromise 进度。
|
||
|
||
### 8.7.2 remove
|
||
|
||
`#remove()` 方法,移除当前消息对应的 Entry 对象,并 Promise 通知成功。代码如下:
|
||
|
||
```
|
||
1: public boolean remove() {
|
||
2: Entry e = flushedEntry;
|
||
3: if (e == null) {
|
||
4: // 清除 NIO ByteBuff 数组的缓存
|
||
5: clearNioBuffers();
|
||
6: return false;
|
||
7: }
|
||
8: Object msg = e.msg;
|
||
9:
|
||
10: ChannelPromise promise = e.promise;
|
||
11: int size = e.pendingSize;
|
||
12:
|
||
13: // 移除指定 Entry 对象
|
||
14: removeEntry(e);
|
||
15:
|
||
16: if (!e.cancelled) {
|
||
17: // 释放消息( 数据 )相关的资源
|
||
18: // only release message, notify and decrement if it was not canceled before.
|
||
19: ReferenceCountUtil.safeRelease(msg);
|
||
20: // 通知 Promise 执行成功
|
||
21: safeSuccess(promise);
|
||
22: // 减少 totalPending 计数
|
||
23: decrementPendingOutboundBytes(size, false, true);
|
||
24: }
|
||
25:
|
||
26: // 回收 Entry 对象
|
||
27: // recycle the entry
|
||
28: e.recycle();
|
||
29:
|
||
30: return true;
|
||
31: }
|
||
```
|
||
|
||
- 第 14 行:调用 `#removeEntry(Entry e)` 方法,移除**指定** Entry 对象。详细解析,见 [「8.7.3 removeEntry」](https://svip.iocoder.cn/Netty/Channel-5-flush/#) 。
|
||
|
||
- 第 16 行:若 Entry 已取消,则忽略。
|
||
|
||
- 第 19 行:`ReferenceCountUtil#safeRelease(msg)` 方法,释放消息( 数据 )相关的资源。
|
||
|
||
- 第 21 行:【**重要**】调用 `#safeSuccess(promise)` 方法,通知 Promise 执行成功。此处才是,真正触发 `Channel#write(...)` 或 `Channel#writeAndFlush(...)` 方法,返回的 Promise 的通知。`#safeSuccess(promise)` 方法的代码如下:
|
||
|
||
```
|
||
private static void safeSuccess(ChannelPromise promise) {
|
||
// Only log if the given promise is not of type VoidChannelPromise as trySuccess(...) is expected to return
|
||
// false.
|
||
PromiseNotificationUtil.trySuccess(promise, null, promise instanceof VoidChannelPromise ? null : logger);
|
||
}
|
||
```
|
||
|
||
- 第 23 行:`#decrementPendingOutboundBytes(long size, ...)` 方法,减少 `totalPendingSize` 计数。
|
||
|
||
- 第 28 行:调用 `Entry#recycle()` 方法,**回收** Entry 对象。
|
||
|
||
### 8.7.3 removeEntry
|
||
|
||
`#removeEntry(Entry e)` 方法,移除**指定** Entry 对象。代码如下:
|
||
|
||
```
|
||
1: private void removeEntry(Entry e) {
|
||
2: // 已移除完已 flush 的 Entry 节点,置空 flushedEntry、tailEntry、unflushedEntry 。
|
||
3: if (-- flushed == 0) {
|
||
4: // processed everything
|
||
5: flushedEntry = null;
|
||
6: if (e == tailEntry) {
|
||
7: tailEntry = null;
|
||
8: unflushedEntry = null;
|
||
9: }
|
||
10: // 未移除完已 flush 的 Entry 节点,flushedEntry 指向下一个 Entry 对象
|
||
11: } else {
|
||
12: flushedEntry = e.next;
|
||
13: }
|
||
14: }
|
||
```
|
||
|
||
- 第 3 至 9 行:**已**移除完已 flush 的**所有** Entry 节点,置空 `flushedEntry`、`tailEntry`、`unflushedEntry` 。
|
||
- 第 10 至 13 行:**未**移除完已 flush 的**所有** Entry 节点,`flushedEntry` 指向**下一个** Entry 对象。
|
||
|
||
### 8.7.4 clearNioBuffers
|
||
|
||
`#clearNioBuffers()` 方法,**清除** NIO ByteBuff 数组的缓存。代码如下:
|
||
|
||
```
|
||
// Clear all ByteBuffer from the array so these can be GC'ed.
|
||
// See https://github.com/netty/netty/issues/3837
|
||
private void clearNioBuffers() {
|
||
int count = nioBufferCount;
|
||
if (count > 0) {
|
||
// 归零 nioBufferCount 。老艿艿觉得,应该把 nioBufferSize 也归零
|
||
nioBufferCount = 0;
|
||
// 置空 NIO ByteBuf 数组
|
||
Arrays.fill(NIO_BUFFERS.get(), 0, count, null);
|
||
}
|
||
}
|
||
```
|
||
|
||
- 代码比较简单,胖友自己看注释。主要目的是 help gc 。
|
||
|
||
## 8.8 failFlushed
|
||
|
||
`#failFlushed(Throwable cause, boolean notify)` 方法,写入数据到对端**失败**,进行后续的处理,详细看代码。代码如下:
|
||
|
||
```
|
||
1: void failFlushed(Throwable cause, boolean notify) {
|
||
2: // 正在通知 flush 失败中,直接返回
|
||
3: // Make sure that this method does not reenter. A listener added to the current promise can be notified by the
|
||
4: // current thread in the tryFailure() call of the loop below, and the listener can trigger another fail() call
|
||
5: // indirectly (usually by closing the channel.)
|
||
6: //
|
||
7: // See https://github.com/netty/netty/issues/1501
|
||
8: if (inFail) {
|
||
9: return;
|
||
10: }
|
||
11:
|
||
12: try {
|
||
13: // 标记正在通知 flush 失败中
|
||
14: inFail = true;
|
||
15: // 循环,移除所有已 flush 的 Entry 节点们
|
||
16: for (;;) {
|
||
17: if (!remove0(cause, notify)) {
|
||
18: break;
|
||
19: }
|
||
20: }
|
||
21: } finally {
|
||
22: // 标记不在通知 flush 失败中
|
||
23: inFail = false;
|
||
24: }
|
||
25: }
|
||
```
|
||
|
||
- 第 2 至 10 行:正在通知 flush 失败中,直接返回。
|
||
- 第 14 行:标记正在通知 flush 失败中,即 `inFail = true` 。
|
||
- 第 15 至 20 行:循环,调用 `#remove0(Throwable cause, boolean notifyWritability)` 方法,移除**所有**已 flush 的 Entry 节点们。详细解析,见 [「8. remove0」](https://svip.iocoder.cn/Netty/Channel-5-flush/#) 中。
|
||
- 第 21 至 24 行:标记不在通知 flush 失败中,即 `inFail = false` 。
|
||
|
||
### 8.8.1 remove0
|
||
|
||
`#remove0(Throwable cause, boolean notifyWritability)` 方法,移除当前消息对应的 Entry 对象,并 Promise 通知异常。代码如下:
|
||
|
||
```
|
||
1: private boolean remove0(Throwable cause, boolean notifyWritability) {
|
||
2: Entry e = flushedEntry;
|
||
3: // 所有 flush 的 Entry 节点,都已经写到对端
|
||
4: if (e == null) {
|
||
5: // // 清除 NIO ByteBuff 数组的缓存
|
||
6: clearNioBuffers();
|
||
7: return false; // 没有后续的 flush 的 Entry 节点
|
||
8: }
|
||
9: Object msg = e.msg;
|
||
10:
|
||
11: ChannelPromise promise = e.promise;
|
||
12: int size = e.pendingSize;
|
||
13:
|
||
14: removeEntry(e);
|
||
15:
|
||
16: if (!e.cancelled) {
|
||
17: // 释放消息( 数据 )相关的资源
|
||
18: // only release message, fail and decrement if it was not canceled before.
|
||
19: ReferenceCountUtil.safeRelease(msg);
|
||
20: // 通知 Promise 执行失败
|
||
21: safeFail(promise, cause);
|
||
22: // 减少 totalPendingSize 计数
|
||
23: decrementPendingOutboundBytes(size, false, notifyWritability);
|
||
24: }
|
||
25:
|
||
26: // 回收 Entry 对象
|
||
27: // recycle the entry
|
||
28: e.recycle();
|
||
29:
|
||
30: return true; // 还有后续的 flush 的 Entry 节点
|
||
31: }
|
||
```
|
||
|
||
- 第 3 至 8 行:若**所有** flush 的 Entry 节点,都已经写到对端,则调用 `#clearNioBuffers()` 方法,清除 NIO ByteBuff 数组的缓存。
|
||
|
||
- 第 14 行:调用 `#removeEntry(Entry e)` 方法,移除**指定** Entry 对象。详细解析,见 [「8.7.3 removeEntry」](https://svip.iocoder.cn/Netty/Channel-5-flush/#) 。
|
||
|
||
- 第 16 行:若 Entry 已取消,则忽略。
|
||
|
||
- 第 19 行:`ReferenceCountUtil#safeRelease(msg)` 方法,释放消息( 数据 )相关的资源。
|
||
|
||
- 第 21 行:【**重要**】调用 `#safeFail(promise)` 方法,通知 Promise 执行失败。此处才是,真正触发 `Channel#write(...)` 或 `Channel#writeAndFlush(...)` 方法,返回的 Promise 的通知。`#safeFail(promise)` 方法的代码如下:
|
||
|
||
```
|
||
private static void safeFail(ChannelPromise promise, Throwable cause) {
|
||
// Only log if the given promise is not of type VoidChannelPromise as tryFailure(...) is expected to return
|
||
// false.
|
||
PromiseNotificationUtil.tryFailure(promise, cause, promise instanceof VoidChannelPromise ? null : logger);
|
||
}
|
||
```
|
||
|
||
- 第 23 行:调用 `#decrementPendingOutboundBytes(long size, ...)` 方法,减少 `totalPendingSize` 计数。
|
||
|
||
- 第 28 行:调用 `Entry#recycle()` 方法,**回收** Entry 对象。
|
||
|
||
## 8.9 forEachFlushedMessage
|
||
|
||
TODO 1015 forEachFlushedMessage 在 `netty-transport-native-poll` 和 `netty-transport-native-kqueue` 中使用,在后续的文章解析。
|
||
|
||
## 8.10 close
|
||
|
||
`#close(...)` 方法,关闭 ChannelOutboundBuffer ,进行后续的处理,详细看代码。代码如下:
|
||
|
||
```
|
||
void close(ClosedChannelException cause) {
|
||
close(cause, false);
|
||
}
|
||
|
||
1: void close(final Throwable cause, final boolean allowChannelOpen) {
|
||
2: // 正在通知 flush 失败中
|
||
3: if (inFail) {
|
||
4: // 提交 EventLoop 的线程中,执行关闭
|
||
5: channel.eventLoop().execute(new Runnable() {
|
||
6: @Override
|
||
7: public void run() {
|
||
8: close(cause, allowChannelOpen);
|
||
9: }
|
||
10: });
|
||
11: // 返回
|
||
12: return;
|
||
13: }
|
||
14:
|
||
15: // 标记正在通知 flush 失败中
|
||
16: inFail = true;
|
||
17:
|
||
18: if (!allowChannelOpen && channel.isOpen()) {
|
||
19: throw new IllegalStateException("close() must be invoked after the channel is closed.");
|
||
20: }
|
||
21:
|
||
22: if (!isEmpty()) {
|
||
23: throw new IllegalStateException("close() must be invoked after all flushed writes are handled.");
|
||
24: }
|
||
25:
|
||
26: // Release all unflushed messages.
|
||
27: try {
|
||
28: // 从 unflushedEntry 节点,开始向下遍历
|
||
29: Entry e = unflushedEntry;
|
||
30: while (e != null) {
|
||
31: // 减少 totalPendingSize
|
||
32: // Just decrease; do not trigger any events via decrementPendingOutboundBytes()
|
||
33: int size = e.pendingSize;
|
||
34: TOTAL_PENDING_SIZE_UPDATER.addAndGet(this, -size);
|
||
35:
|
||
36: if (!e.cancelled) {
|
||
37: // 释放消息( 数据 )相关的资源
|
||
38: ReferenceCountUtil.safeRelease(e.msg);
|
||
39: // 通知 Promise 执行失败
|
||
40: safeFail(e.promise, cause);
|
||
41: }
|
||
42: // 回收当前节点,并获得下一个 Entry 节点
|
||
43: e = e.recycleAndGetNext();
|
||
44: }
|
||
45: } finally {
|
||
46: // 标记在在通知 flush 失败中
|
||
47: inFail = false;
|
||
48: }
|
||
49:
|
||
50: // 清除 NIO ByteBuff 数组的缓存。
|
||
51: clearNioBuffers();
|
||
52: }
|
||
```
|
||
|
||
- 第 3 行:正在通知 flush 失败中:
|
||
|
||
- 第 5 至 10 行: 提交 EventLoop 的线程中,执行关闭。
|
||
- 第 12 行:`return` 返回。
|
||
|
||
- 第 16 行:标记正在通知 flush 失败中,即 `inFail = true` 。
|
||
|
||
- 第 28 至 30 行:从
|
||
|
||
|
||
|
||
```
|
||
unflushedEntry
|
||
```
|
||
|
||
|
||
|
||
节点,开始向下遍历。
|
||
|
||
- 第 31 至 34 行:减少 `totalPendingSize` 计数。
|
||
- 第 36 行:若 Entry 已取消,则忽略。
|
||
- 第 38 行:调用 `ReferenceCountUtil#safeRelease(msg)` 方法,释放消息( 数据 )相关的资源。
|
||
- 第 40 行:【**重要**】调用 `#safeFail(promise)` 方法,通知 Promise 执行失败。此处才是,真正触发 `Channel#write(...)` 或 `Channel#writeAndFlush(...)` 方法,返回的 Promise 的通知。
|
||
- 第 43 行:调用 `Entry#recycleAndGetNext()` 方法,回收当前节点,并获得下一个 Entry 节点。
|
||
|
||
- 第 45 至 48 行:标记不在通知 flush 失败中,即 `inFail = false` 。
|
||
|
||
- 第 51 行:调用 `#clearNioBuffers()` 方法,**清除** NIO ByteBuff 数组的缓存。
|
||
|
||
# 9. NioEventLoop
|
||
|
||
在上文 [「7. NioSocketChannel」](https://svip.iocoder.cn/Netty/Channel-5-flush/#) 中,在写入到 Channel 到对端,若 TCP 数据发送缓冲区**已满**,这将导致 Channel **不写可**,此时会注册对该 Channel 的 `SelectionKey.OP_WRITE` 事件感兴趣。从而实现,再在 Channel 可写后,进行**强制** flush 。这块的逻辑,在 `NioEventLoop#processSelectedKey(SelectionKey k, AbstractNioChannel ch)` 中实现,代码如下:
|
||
|
||
```
|
||
// OP_WRITE 事件就绪
|
||
// Process OP_WRITE first as we may be able to write some queued buffers and so free memory.
|
||
if ((readyOps & SelectionKey.OP_WRITE) != 0) {
|
||
// Call forceFlush which will also take care of clear the OP_WRITE once there is nothing left to write
|
||
// 向 Channel 写入数据
|
||
ch.unsafe().forceFlush();
|
||
}
|
||
```
|
||
|
||
- 通过 Selector 轮询到 Channel 的 `OP_WRITE` 就绪时,调用 `AbstractNioUnsafe#forceFlush()` 方法,强制 flush 。代码如下:
|
||
|
||
```
|
||
// AbstractNioUnsafe.java
|
||
@Override
|
||
public final void forceFlush() {
|
||
// directly call super.flush0() to force a flush now
|
||
super.flush0();
|
||
}
|
||
```
|
||
|
||
- 后续的逻辑,又回到 [「6. AbstractUnsafe」](https://svip.iocoder.cn/Netty/Channel-5-flush/#) 小节的 `#flush0()` 流程。
|
||
- 在完成强制 flush 之后,会取消对 `SelectionKey.OP_WRITE` 事件的感兴趣。
|
||
|
||
## 9.1 如何模拟
|
||
|
||
1. 配置服务端 ServerBootstrap 的启动参数如下:
|
||
|
||
```
|
||
.childOption(ChannelOption.SO_SNDBUF, 5) // Socket 参数,TCP 数据发送缓冲区大小。
|
||
```
|
||
|
||
2. `telnet` 到启动的服务端,发送相对长的命令,例如 `"abcdefghijklmnopqrstuvw11321321321nhdkslk"` 。
|
||
|
||
# 10. ChannelOutboundBuffer 写入控制
|
||
|
||
当我们不断调用 `#addMessage(Object msg, int size, ChannelPromise promise)` 方法,添加消息到 ChannelOutboundBuffer 内存队列中,如果**不及时** flush 写到对端( 例如程序一直未调用 `Channel#flush()` 方法,或者对端接收数据比较慢导致 Channel 不可写 ),可能会导致 **OOM 内存溢出**。所以,在 ChannelOutboundBuffer 使用 `totalPendingSize` 属性,存储所有 Entry 预计占用的内存大小( `pendingSize` )。
|
||
|
||
- 在 `totalPendingSize` 大于高水位阀值时( `ChannelConfig.writeBufferHighWaterMark` ,默认值为 64 KB ),**关闭**写开关( `unwritable` )。详细解析,见 [「10.1 incrementPendingOutboundBytes」](https://svip.iocoder.cn/Netty/Channel-5-flush/#) 。
|
||
- 在 `totalPendingSize` 小于低水位阀值时( `ChannelConfig.writeBufferLowWaterMark` ,默认值为 32 KB ),**打开**写开关( `unwritable` )。详细解析,见 [「10.2 decrementPendingOutboundBytes」](https://svip.iocoder.cn/Netty/Channel-5-flush/#) 。
|
||
|
||
该功能,对应 Github 提交为 [《Take memory overhead of ChannelOutboundBuffer / PendingWriteQueue into account》](https://github.com/netty/netty/commit/e3cb9935c0b63357e3d51867cffe624129e7e1dd) 。
|
||
|
||
## 10.1 incrementPendingOutboundBytes
|
||
|
||
`#incrementPendingOutboundBytes(long size, ...)` 方法,增加 `totalPendingSize` 计数。代码如下:
|
||
|
||
```
|
||
1: /**
|
||
2: * Increment the pending bytes which will be written at some point.
|
||
3: * This method is thread-safe!
|
||
4: */
|
||
5: void incrementPendingOutboundBytes(long size) {
|
||
6: incrementPendingOutboundBytes(size, true);
|
||
7: }
|
||
8:
|
||
9: private void incrementPendingOutboundBytes(long size, boolean invokeLater) {
|
||
10: if (size == 0) {
|
||
11: return;
|
||
12: }
|
||
13:
|
||
14: // 增加 totalPendingSize 计数
|
||
15: long newWriteBufferSize = TOTAL_PENDING_SIZE_UPDATER.addAndGet(this, size);
|
||
16: // totalPendingSize 大于高水位阀值时,设置为不可写
|
||
17: if (newWriteBufferSize > channel.config().getWriteBufferHighWaterMark()) {
|
||
18: setUnwritable(invokeLater);
|
||
19: }
|
||
20: }
|
||
```
|
||
|
||
- 第 15 行:增加 `totalPendingSize` 计数。
|
||
|
||
- 第 16 至 19 行:`totalPendingSize` 大于高水位阀值时,调用 `#setUnwritable(boolean invokeLater)` 方法,设置为不可写。代码如下:
|
||
|
||
```
|
||
1: private void setUnwritable(boolean invokeLater) {
|
||
2: for (;;) {
|
||
3: final int oldValue = unwritable;
|
||
4: // 或位操作,修改第 0 位 bits 为 1
|
||
5: final int newValue = oldValue | 1;
|
||
6: // CAS 设置 unwritable 为新值
|
||
7: if (UNWRITABLE_UPDATER.compareAndSet(this, oldValue, newValue)) {
|
||
8: // 若之前可写,现在不可写,触发 Channel WritabilityChanged 事件到 pipeline 中。
|
||
9: if (oldValue == 0 && newValue != 0) {
|
||
10: fireChannelWritabilityChanged(invokeLater);
|
||
11: }
|
||
12: break;
|
||
13: }
|
||
14: }
|
||
15: }
|
||
```
|
||
|
||
- 第 2 行:`for` 循环,直到 CAS 修改成功
|
||
- 第 5 行:或位操作,修改第 0 位 bits 为 1 。😈 比较神奇的是,`unwritable` 的类型不是 `boolean` ,而是 `int` 类型。通过每个 bits ,来表示**哪种**类型不可写。感兴趣的胖友,可以看看 `io.netty.handler.traffic.AbstractTrafficShapingHandler` ,使用了第 1、2、3 bits 。
|
||
- 第 7 行:CAS 设置 `unwritable` 为新值。
|
||
- 第 8 至 11 行:若之前可写,现在不可写,调用 `#fireChannelWritabilityChanged(boolean invokeLater)` 方法,触发 Channel WritabilityChanged 事件到 pipeline 中。详细解析,见 [「10.3 fireChannelWritabilityChanged」](https://svip.iocoder.cn/Netty/Channel-5-flush/#) 。
|
||
|
||
### 10.1.1 bytesBeforeUnwritable
|
||
|
||
`#bytesBeforeUnwritable()` 方法,获得距离**不可写**还有多少字节数。代码如下:
|
||
|
||
```
|
||
public long bytesBeforeUnwritable() {
|
||
long bytes = channel.config().getWriteBufferHighWaterMark() - totalPendingSize;
|
||
// If bytes is negative we know we are not writable, but if bytes is non-negative we have to check writability.
|
||
// Note that totalPendingSize and isWritable() use different volatile variables that are not synchronized
|
||
// together. totalPendingSize will be updated before isWritable().
|
||
if (bytes > 0) {
|
||
return isWritable() ? bytes : 0; // 判断 #isWritable() 的原因是,可能已经被设置不可写
|
||
}
|
||
return 0;
|
||
}
|
||
```
|
||
|
||
- 基于**高水位**阀值来判断。
|
||
|
||
## 10.2 decrementPendingOutboundBytes
|
||
|
||
`#decrementPendingOutboundBytes(long size, ...)` 方法,减少 `totalPendingSize` 计数。代码如下:
|
||
|
||
```
|
||
1: /**
|
||
2: * Decrement the pending bytes which will be written at some point.
|
||
3: * This method is thread-safe!
|
||
4: */
|
||
5: void decrementPendingOutboundBytes(long size) {
|
||
6: decrementPendingOutboundBytes(size, true, true);
|
||
7: }
|
||
8:
|
||
9: private void decrementPendingOutboundBytes(long size, boolean invokeLater, boolean notifyWritability) {
|
||
10: if (size == 0) {
|
||
11: return;
|
||
12: }
|
||
13:
|
||
14: // 减少 totalPendingSize 计数
|
||
15: long newWriteBufferSize = TOTAL_PENDING_SIZE_UPDATER.addAndGet(this, -size);
|
||
16: // totalPendingSize 小于低水位阀值时,设置为可写
|
||
17: if (notifyWritability && newWriteBufferSize < channel.config().getWriteBufferLowWaterMark()) {
|
||
18: setWritable(invokeLater);
|
||
19: }
|
||
20: }
|
||
```
|
||
|
||
- 第 15 行:减少 `totalPendingSize` 计数。
|
||
|
||
- 第 16 至 19 行:`totalPendingSize` 小于低水位阀值时,调用 `#setWritable(boolean invokeLater)` 方法,设置为可写。代码如下:
|
||
|
||
```
|
||
1: private void setWritable(boolean invokeLater) {
|
||
2: for (;;) {
|
||
3: final int oldValue = unwritable;
|
||
4: // 并位操作,修改第 0 位 bits 为 0
|
||
5: final int newValue = oldValue & ~1;
|
||
6: // CAS 设置 unwritable 为新值
|
||
7: if (UNWRITABLE_UPDATER.compareAndSet(this, oldValue, newValue)) {
|
||
8: // 若之前不可写,现在可写,触发 Channel WritabilityChanged 事件到 pipeline 中。
|
||
9: if (oldValue != 0 && newValue == 0) {
|
||
10: fireChannelWritabilityChanged(invokeLater);
|
||
11: }
|
||
12: break;
|
||
13: }
|
||
14: }
|
||
15: }
|
||
```
|
||
|
||
- 第 2 行:`for` 循环,直到 CAS 修改成功
|
||
- 第 5 行:并位操作,修改第 0 位 bits 为 0 。
|
||
- 第 7 行:CAS 设置 `unwritable` 为新值。
|
||
- 第 8 至 11 行:若之前可写,现在不可写,调用 `#fireChannelWritabilityChanged(boolean invokeLater)` 方法,触发 Channel WritabilityChanged 事件到 pipeline 中。详细解析,见 [「10.3 fireChannelWritabilityChanged」](https://svip.iocoder.cn/Netty/Channel-5-flush/#) 。
|
||
|
||
### 10.2.1 bytesBeforeWritable
|
||
|
||
`#bytesBeforeWritable()` 方法,获得距离**可写**还要多少字节数。代码如下:
|
||
|
||
```
|
||
/**
|
||
* Get how many bytes must be drained from the underlying buffer until {@link #isWritable()} returns {@code true}.
|
||
* This quantity will always be non-negative. If {@link #isWritable()} is {@code true} then 0.
|
||
*/
|
||
public long bytesBeforeWritable() {
|
||
long bytes = totalPendingSize - channel.config().getWriteBufferLowWaterMark();
|
||
// If bytes is negative we know we are writable, but if bytes is non-negative we have to check writability.
|
||
// Note that totalPendingSize and isWritable() use different volatile variables that are not synchronized
|
||
// together. totalPendingSize will be updated before isWritable().
|
||
if (bytes > 0) {
|
||
return isWritable() ? 0 : bytes; // 判断 #isWritable() 的原因是,可能已经被设置不可写
|
||
}
|
||
return 0;
|
||
}
|
||
```
|
||
|
||
- 基于**低水位**阀值来判断。
|
||
|
||
## 10.3 fireChannelWritabilityChanged
|
||
|
||
`#fireChannelWritabilityChanged(boolean invokeLater)` 方法,触发 Channel WritabilityChanged 事件到 pipeline 中。代码如下:
|
||
|
||
```
|
||
private void fireChannelWritabilityChanged(boolean invokeLater) {
|
||
final ChannelPipeline pipeline = channel.pipeline();
|
||
// 延迟执行,即提交 EventLoop 中触发 Channel WritabilityChanged 事件到 pipeline 中
|
||
if (invokeLater) {
|
||
Runnable task = fireChannelWritabilityChangedTask;
|
||
if (task == null) {
|
||
fireChannelWritabilityChangedTask = task = new Runnable() {
|
||
@Override
|
||
public void run() {
|
||
pipeline.fireChannelWritabilityChanged();
|
||
}
|
||
};
|
||
}
|
||
channel.eventLoop().execute(task);
|
||
// 直接触发 Channel WritabilityChanged 事件到 pipeline 中
|
||
} else {
|
||
pipeline.fireChannelWritabilityChanged();
|
||
}
|
||
}
|
||
```
|
||
|
||
- 根据 `invokeLater` 的值,分成两种方式,调用 `ChannelPipeline#fireChannelWritabilityChanged()` 方法,触发 Channel WritabilityChanged 事件到 pipeline 中。具体,胖友看下代码注释。
|
||
|
||
- 后续的流程,就是 [《精尽 Netty 源码解析 —— ChannelPipeline(五)之 Inbound 事件的传播》](http://svip.iocoder.cn/Netty/Pipeline-5-inbound/) 。
|
||
|
||
- 通过 Channel WritabilityChanged 事件,配合
|
||
|
||
|
||
|
||
```
|
||
io.netty.handler.stream.ChunkedWriteHandler
|
||
```
|
||
|
||
|
||
|
||
处理器,实现 ChannelOutboundBuffer 写入的控制,避免 OOM 。ChunkedWriteHandler 的具体代码实现,我们在后续的文章,详细解析。
|
||
|
||
- 所以,有一点要注意,ChannelOutboundBuffer 的 `unwritable` 属性,仅仅作为一个是否不可写的**开关**,具体需要配合响应的 ChannelHandler 处理器,才能实现“不可写”的功能。
|
||
|
||
## 10.4 isWritable
|
||
|
||
`#isWritable()` 方法,是否可写。代码如下:
|
||
|
||
```
|
||
/**
|
||
* Returns {@code true} if and only if {@linkplain #totalPendingWriteBytes() the total number of pending bytes} did
|
||
* not exceed the write watermark of the {@link Channel} and
|
||
* no {@linkplain #setUserDefinedWritability(int, boolean) user-defined writability flag} has been set to
|
||
* {@code false}.
|
||
*/
|
||
public boolean isWritable() {
|
||
return unwritable == 0;
|
||
}
|
||
```
|
||
|
||
- 如果 `unwritable` 大于 0 ,则表示不可写。😈 一定要注意!!!
|
||
|
||
### 10.4.1 getUserDefinedWritability
|
||
|
||
`#getUserDefinedWritability(int index)` 方法,获得指定 bits 是否可写。代码如下:
|
||
|
||
```
|
||
/**
|
||
* Returns {@code true} if and only if the user-defined writability flag at the specified index is set to
|
||
* {@code true}.
|
||
*/
|
||
public boolean getUserDefinedWritability(int index) {
|
||
return (unwritable & writabilityMask(index)) == 0;
|
||
}
|
||
|
||
private static int writabilityMask(int index) {
|
||
// 不能 < 1 ,因为第 0 bits 为 ChannelOutboundBuffer 自己使用
|
||
// 不能 > 31 ,因为超过 int 的 bits 范围
|
||
if (index < 1 || index > 31) {
|
||
throw new IllegalArgumentException("index: " + index + " (expected: 1~31)");
|
||
}
|
||
return 1 << index;
|
||
}
|
||
```
|
||
|
||
- 为什么方法名字上会带有 `"UserDefined"` 呢?因为 `index` 不能使用 0 ,表示只允许使用用户定义( `"UserDefined"` ) bits 位,即 `[1, 31]` 。
|
||
|
||
### 10.4.2 setUserDefinedWritability
|
||
|
||
`#setUserDefinedWritability(int index, boolean writable)` 方法,设置指定 bits 是否可写。代码如下:
|
||
|
||
```
|
||
/**
|
||
* Sets a user-defined writability flag at the specified index.
|
||
*/
|
||
public void setUserDefinedWritability(int index, boolean writable) {
|
||
// 设置可写
|
||
if (writable) {
|
||
setUserDefinedWritability(index);
|
||
// 设置不可写
|
||
} else {
|
||
clearUserDefinedWritability(index);
|
||
}
|
||
}
|
||
|
||
private void setUserDefinedWritability(int index) {
|
||
final int mask = ~writabilityMask(index);
|
||
for (;;) {
|
||
final int oldValue = unwritable;
|
||
final int newValue = oldValue & mask;
|
||
// CAS 设置 unwritable 为新值
|
||
if (UNWRITABLE_UPDATER.compareAndSet(this, oldValue, newValue)) {
|
||
// 若之前不可写,现在可写,触发 Channel WritabilityChanged 事件到 pipeline 中。
|
||
if (oldValue != 0 && newValue == 0) {
|
||
fireChannelWritabilityChanged(true);
|
||
}
|
||
break;
|
||
}
|
||
}
|
||
}
|
||
|
||
private void clearUserDefinedWritability(int index) {
|
||
final int mask = writabilityMask(index);
|
||
for (;;) {
|
||
final int oldValue = unwritable;
|
||
final int newValue = oldValue | mask;
|
||
if (UNWRITABLE_UPDATER.compareAndSet(this, oldValue, newValue)) {
|
||
// 若之前可写,现在不可写,触发 Channel WritabilityChanged 事件到 pipeline 中。
|
||
if (oldValue == 0 && newValue != 0) {
|
||
fireChannelWritabilityChanged(true);
|
||
}
|
||
break;
|
||
}
|
||
}
|
||
}
|
||
```
|
||
|
||
- 代码比较简单,胖友自己看噢。
|
||
|
||
# 666. 彩蛋
|
||
|
||
比想象中,长的多的多的一篇文章。总的来说,绝大部分细节,都已经扣到,美滋滋。如果有解释不够清晰或错误的细节,一起多多沟通呀。
|
||
|
||
写完这篇,我简直疯了。。。。
|
||
|
||
推荐阅读文章:
|
||
|
||
- 莫那一鲁道 [《Netty 出站缓冲区 ChannelOutboundBuffer 源码解析(isWritable 属性的重要性)》](https://www.jianshu.com/p/311425d1c72f)
|
||
- tomas家的小拨浪鼓 [《Netty 源码解析 ——— writeAndFlush流程分析》](https://www.jianshu.com/p/a3443cacd081)
|
||
- 闪电侠 [《netty 源码分析之 writeAndFlush 全解析》](https://www.jianshu.com/p/feaeaab2ce56)
|
||
- 占小狼 [《深入浅出 Netty write》](https://www.jianshu.com/p/1ad424c53e80)
|
||
- Hypercube [《自顶向下深入分析Netty(六)–Channel源码实现》](https://www.jianshu.com/p/9258af254e1d) |