从本文开始,咱们来剖析rocketMq音讯接管、散发以及投递流程。
RocketMq音讯解决整个流程如下:
- 音讯接管:音讯接管是指接管producer的音讯,解决类是SendMessageProcessor,将音讯写入到commigLog文件后,接管流程处理完毕;
- 音讯散发:broker解决音讯散发的类是ReputMessageService,它会启动一个线程,一直地将commitLong分到到对应的consumerQueue,这一步操作会写两个文件:consumerQueue与indexFile,写入后,音讯散发流程解决 结束;
- 音讯投递:音讯投递是指将音讯发往consumer的流程,consumer会发动获取音讯的申请,broker收到申请后,调用PullMessageProcessor类解决,从consumerQueue文件获取音讯,返回给consumer后,投递流程处理完毕。
以上就是rocketMq解决音讯的流程了,接下来咱们就从源码来看相干流程的实现。
1. remotingServer的启动流程
在正式剖析接管与投递流程前,咱们来理解下remotingServer的启动。
remotingServer是一个netty服务,他开启了一个端口用来解决producer与consumer的网络申请。
remotingServer是在BrokerController#start中启动的,代码如下:
public void start() throws Exception {
// 启动各组件
...
if (this.remotingServer != null) {
this.remotingServer.start();
}
...
}
持续查看remotingServer的启动流程,进入NettyRemotingServer#start办法:
public void start() {
...
ServerBootstrap childHandler =
this.serverBootstrap.group(this.eventLoopGroupBoss, this.eventLoopGroupSelector)
...
.childHandler(new ChannelInitializer<SocketChannel>() {
@Override
public void initChannel(SocketChannel ch) throws Exception {
ch.pipeline()
.addLast(defaultEventExecutorGroup,
HANDSHAKE_HANDLER_NAME, handshakeHandler)
.addLast(defaultEventExecutorGroup,
encoder,
new NettyDecoder(),
new IdleStateHandler(0, 0,
nettyServerConfig.getServerChannelMaxIdleTimeSeconds()),
connectionManageHandler,
// 解决业务申请的handler
serverHandler
);
}
});
...
}
这就是一个规范的netty服务启动流程了,套路与nameServer的启动是一样的。对于netty的相干内容,这里咱们仅关注pipeline上的channelHandler,在netty中,解决读写申请的操作为一个个ChannelHandler,remotingServer中解决读写申请的ChanelHandler为NettyServerHandler,代码如下:
@ChannelHandler.Sharable
class NettyServerHandler extends SimpleChannelInboundHandler<RemotingCommand> {
@Override
protected void channelRead0(ChannelHandlerContext ctx, RemotingCommand msg) throws Exception {
processMessageReceived(ctx, msg);
}
}
这块的操作与nameServer对外提供的服务极类似(就是同一个类),最终调用的是NettyRemotingAbstract#processRequestCommand办法:
public void processRequestCommand(final ChannelHandlerContext ctx, final RemotingCommand cmd) {
// 依据 code 从 processorTable 获取 Pair
final Pair<NettyRequestProcessor, ExecutorService> matched
= this.processorTable.get(cmd.getCode());
// 找不到默认值
final Pair<NettyRequestProcessor, ExecutorService> pair =
null == matched ? this.defaultRequestProcessor : matched;
...
// 从 pair 中拿到 Processor 进行解决
NettyRequestProcessor processor = pair.getObject1();
// 解决申请
RemotingCommand response = processor.processRequest(ctx, cmd);
....
}
如果进入源码去看,会发现这个办法十分长,这里省略了异步解决、异样解决及返回值结构等,仅列出了关键步骤:
- 依据code从processorTable拿到对应的Pair
- 从Pair里获取Processor
最终解决申请的就是Processor了。
2. Processor的注册
从下面的剖析中可知, Processor是解决音讯的要害,它是从processorTable中获取的,这个processorTable是啥呢?
processorTable是NettyRemotingAbstract成员变量,外面的内容是BrokerController在初始化时(执行BrokerController#initialize办法)注册的。之前在剖析BrokerController的初始化流程时,就提到过Processor的提供操作,这里再回顾下:
BrokerController的初始化办法initialize会调用 BrokerController#registerProcessor,Processor的注册操作就在这个办法里:
public class BrokerController {
private final PullMessageProcessor pullMessageProcessor;
/**
* 构造方法
*/
public BrokerController(...) {
// 解决 consumer 拉音讯申请的
this.pullMessageProcessor = new PullMessageProcessor(this);
}
/**
* 注册操作
*/
public void registerProcessor() {
// SendMessageProcessor
SendMessageProcessor sendProcessor = new SendMessageProcessor(this);
sendProcessor.registerSendMessageHook(sendMessageHookList);
sendProcessor.registerConsumeMessageHook(consumeMessageHookList);
// 解决 Processor
this.remotingServer.registerProcessor(RequestCode.SEND_MESSAGE,
sendProcessor, this.sendMessageExecutor);
this.remotingServer.registerProcessor(RequestCode.SEND_MESSAGE_V2,
sendProcessor, this.sendMessageExecutor);
this.remotingServer.registerProcessor(RequestCode.SEND_BATCH_MESSAGE,
sendProcessor, this.sendMessageExecutor);
// PullMessageProcessor
this.remotingServer.registerProcessor(RequestCode.PULL_MESSAGE,
this.pullMessageProcessor, this.pullMessageExecutor);
// 省略其余许许多多的Processor注册
...
}
...
须要指明的是,sendProcessor用来解决producer申请过去的音讯,pullMessageProcessor用来解决consumer拉取音讯的申请。
3. 接管producer音讯
理解完remotingServer的启动与Processor的注册内容后,接下来咱们就能够剖析接管producer音讯的流程了。
producer发送音讯到broker时,发送的申请code为SEND_MESSAGE(RocketMQ源码5-producer 同步发送和单向发送 第1.4大节),依据下面的剖析,当音讯过去时,会应用NettyServerHandler这个ChannelHandler来解决,之后会调用到NettyRemotingAbstract#processRequestCommand办法。
在NettyRemotingAbstract#processRequestCommand办法中,会依据音讯的code获取对应的Processor来解决,从Processor的注册流程来看,解决该SEND_MESSAGE的Processor为SendMessageProcessor,咱们进入SendMessageProcessor#processRequest看看它的流程:
public RemotingCommand processRequest(ChannelHandlerContext ctx,
RemotingCommand request) throws RemotingCommandException {
RemotingCommand response = null;
try {
// broker解决接管音讯
response = asyncProcessRequest(ctx, request).get();
} catch (InterruptedException | ExecutionException e) {
log.error("process SendMessage error, request : " + request.toString(), e);
}
return response;
}
没干啥事,一路跟上来,间接看一般音讯的流程,进入SendMessageProcessor#asyncSendMessage办法:
private CompletableFuture<RemotingCommand> asyncSendMessage(ChannelHandlerContext ctx,
RemotingCommand request, SendMessageContext mqtraceContext,
SendMessageRequestHeader requestHeader) {
final RemotingCommand response = preSend(ctx, request, requestHeader);
final SendMessageResponseHeader responseHeader
= (SendMessageResponseHeader)response.readCustomHeader();
if (response.getCode() != -1) {
return CompletableFuture.completedFuture(response);
}
final byte[] body = request.getBody();
int queueIdInt = requestHeader.getQueueId();
TopicConfig topicConfig = this.brokerController.getTopicConfigManager()
.selectTopicConfig(requestHeader.getTopic());
// 如果没指定队列,就随机指定一个队列
if (queueIdInt < 0) {
queueIdInt = randomQueueId(topicConfig.getWriteQueueNums());
}
// 将音讯包装为 MessageExtBrokerInner
MessageExtBrokerInner msgInner = new MessageExtBrokerInner();
msgInner.setTopic(requestHeader.getTopic());
msgInner.setQueueId(queueIdInt);
// 省略解决 msgInner 的流程
...
CompletableFuture<PutMessageResult> putMessageResult = null;
Map<String, String> origProps = MessageDecoder
.string2messageProperties(requestHeader.getProperties());
String transFlag = origProps.get(MessageConst.PROPERTY_TRANSACTION_PREPARED);
// 发送事务音讯
if (transFlag != null && Boolean.parseBoolean(transFlag)) {
...
// 发送事务音讯
putMessageResult = this.brokerController.getTransactionalMessageService()
.asyncPrepareMessage(msgInner);
} else {
// 发送一般音讯
putMessageResult = this.brokerController.getMessageStore().asyncPutMessage(msgInner);
}
return handlePutMessageResultFuture(putMessageResult, response, request, msgInner,
responseHeader, mqtraceContext, ctx, queueIdInt);
}
这个办法是在筹备音讯的发送数据,所做的工作如下:
- 如果没指定队列,就随机指定一个队列,个别状况下不会给音讯指定队列的,但如果要发送程序音讯,就须要指定队列了,这点前面再剖析。
- 结构MessageExtBrokerInner对象,就是将producer上送的音讯包装下,加上一些额定的信息,如音讯标识msgId、发送工夫、topic、queue等。
- 发送音讯,这里只是分为两类:事务音讯与一般音讯,这里咱们次要关注一般音讯,事务音讯前面再剖析。
进入一般音讯的发送办法DefaultMessageStore#asyncPutMessage:
public CompletableFuture<PutMessageResult> asyncPutMessage(MessageExtBrokerInner msg) {
...
// 保留到 commitLog
CompletableFuture<PutMessageResult> putResultFuture = this.commitLog.asyncPutMessage(msg);
...
}
3.1 commitLog写入原理
一个broker逻辑上对应着一个commitLog,你能够把它看作一个大文件,而后这个broker收到的所有音讯都写到这个外面,然而物理上ROCKET_HOME/commitlog/00000000000000000000这个门路存储的,它是由若干个文件组成的,每个文件默认大小是1G,而后每个文件都对应这个一个MappedFile,00000000000000000000 就是第一个MappedFile对应的物理文件,每个文件的文件名就是在commitLog外面的一个其实offset,第二个文件名就是00000000001073741824,也就是上一个MappedFile文件起始offset加上每个文件的大小,这个MappedFile就是RocketMQ的黑科技,应用了内存映射技术来进步文件的访问速度与写入速度,而后都是采纳追加写的形式进步写入速度。
咱们间接看官网的形容(链接:github.com/apache/rock…):
rocketMq 音讯存储架构图
音讯存储架构图中次要有上面三个跟音讯存储相干的文件形成。
(1) CommitLog:音讯主体以及元数据的存储主体,存储Producer端写入的音讯主体内容,音讯内容不是定长的。单个文件大小默认1G ,文件名长度为20位,右边补零,残余为起始偏移量,比方00000000000000000000代表了第一个文件,起始偏移量为0,文件大小为1G=1073741824;当第一个文件写满了,第二个文件为00000000001073741824,起始偏移量为1073741824,以此类推。音讯次要是程序写入日志文件,当文件满了,写入下一个文件;
(2) ConsumeQueue:音讯生产队列,引入的目标次要是进步音讯生产的性能,因为RocketMQ是基于主题topic的订阅模式,音讯生产是针对主题进行的,如果要遍历commitlog文件中依据topic检索音讯是十分低效的。Consumer即可依据ConsumeQueue来查找待生产的音讯。其中,ConsumeQueue(逻辑生产队列)作为生产音讯的索引,保留了指定Topic下的队列音讯在CommitLog中的起始物理偏移量offset,音讯大小size和音讯Tag的HashCode值。consumequeue文件能够看成是基于topic的commitlog索引文件,故consumequeue文件夹的组织形式如下:topic/queue/file三层组织构造,具体存储门路为:$HOME/store/consumequeue/{topic}/{queueId}/{fileName}。同样consumequeue文件采取定长设计,每一个条目共20个字节,别离为8字节的commitlog物理偏移量、4字节的音讯长度、8字节tag hashcode,单个文件由30W个条目组成,能够像数组一样随机拜访每一个条目,每个ConsumeQueue文件大小约5.72M;
(3) IndexFile:IndexFile(索引文件)提供了一种能够通过key或工夫区间来查问音讯的办法。Index文件的存储地位是:HOME\store\index{fileName},文件名fileName是以创立时的工夫戳命名的,固定的单个IndexFile文件大小约为400M,一个IndexFile能够保留 2000W个索引,IndexFile的底层存储设计为在文件系统中实现HashMap构造,故rocketmq的索引文件其底层实现为hash索引。
在下面的RocketMQ的音讯存储整体架构图中能够看出,RocketMQ采纳的是混合型的存储构造,即为Broker单个实例下所有的队列共用一个日志数据文件(即为CommitLog)来存储。RocketMQ的混合型存储构造(多个Topic的音讯实体内容都存储于一个CommitLog中)针对Producer和Consumer别离采纳了数据和索引局部相拆散的存储构造,Producer发送音讯至Broker端,而后Broker端应用同步或者异步的形式对音讯刷盘长久化,保留至CommitLog中。只有音讯被刷盘长久化至磁盘文件CommitLog中,那么Producer发送的音讯就不会失落。
正因为如此,Consumer也就必定有机会去生产这条音讯。当无奈拉取到音讯后,能够等下一次音讯拉取,同时服务端也反对长轮询模式,如果一个音讯拉取申请未拉取到音讯,Broker容许期待30s的工夫,只有这段时间内有新音讯达到,将间接返回给生产端。这里,RocketMQ的具体做法是,应用Broker端的后盾服务线程—ReputMessageService不停地散发申请并异步构建ConsumeQueue(逻辑生产队列)和IndexFile(索引文件)数据。
3.2 CommitLog#asyncPutMessage
持续进入CommitLog#asyncPutMessage办法,这个办法有点长,咱们分局部:
public CompletableFuture<PutMessageResult> asyncPutMessage(final MessageExtBrokerInner msg) {
// Set the storage time 设置存储工夫
msg.setStoreTimestamp(System.currentTimeMillis());
// Set the message body BODY CRC (consider the most appropriate setting
// on the client)
// 设置crc
msg.setBodyCRC(UtilAll.crc32(msg.getBody()));
// Back to Results
AppendMessageResult result = null;
StoreStatsService storeStatsService = this.defaultMessageStore.getStoreStatsService();
String topic = msg.getTopic();
int queueId = msg.getQueueId();
// 获取事务状态
final int tranType = MessageSysFlag.getTransactionValue(msg.getSysFlag());
// 事务
if (tranType == MessageSysFlag.TRANSACTION_NOT_TYPE
|| tranType == MessageSysFlag.TRANSACTION_COMMIT_TYPE) {
// Delay Delivery 延时音讯的解决
if (msg.getDelayTimeLevel() > 0) {
if (msg.getDelayTimeLevel() > this.defaultMessageStore.getScheduleMessageService().getMaxDelayLevel()) {
msg.setDelayTimeLevel(this.defaultMessageStore.getScheduleMessageService().getMaxDelayLevel());
}
// 设置提早队列
topic = TopicValidator.RMQ_SYS_SCHEDULE_TOPIC;
queueId = ScheduleMessageService.delayLevel2QueueId(msg.getDelayTimeLevel());
// Backup real topic, queueId
MessageAccessor.putProperty(msg, MessageConst.PROPERTY_REAL_TOPIC, msg.getTopic());
MessageAccessor.putProperty(msg, MessageConst.PROPERTY_REAL_QUEUE_ID, String.valueOf(msg.getQueueId()));
msg.setPropertiesString(MessageDecoder.messageProperties2String(msg.getProperties()));
msg.setTopic(topic);
msg.setQueueId(queueId);
}
}
...
这一部分其实就是从msg中获取一些信息,判断解决一下这个延时音讯。
long elapsedTimeInLock = 0;
MappedFile unlockMappedFile = null;
// 获取最初一个 MappedFile
MappedFile mappedFile = this.mappedFileQueue.getLastMappedFile();
// 获取写入锁
putMessageLock.lock(); //spin or ReentrantLock ,depending on store config
try {
long beginLockTimestamp = this.defaultMessageStore.getSystemClock().now();
// 开始在锁里的工夫
this.beginTimeInLock = beginLockTimestamp;
// Here settings are stored timestamp, in order to ensure an orderly
// global
// 设置写入的工夫戳,确保它是有序的,
msg.setStoreTimestamp(beginLockTimestamp);
// 判断MappedFile 是否是 null 或者是否是满了
if (null == mappedFile || mappedFile.isFull()) {
mappedFile = this.mappedFileQueue.getLastMappedFile(0); // Mark: NewFile may be cause noise
}
if (null == mappedFile) {
log.error("create mapped file1 error, topic: " + msg.getTopic() + " clientAddr: " + msg.getBornHostString());
beginTimeInLock = 0;
return CompletableFuture.completedFuture(new PutMessageResult(PutMessageStatus.CREATE_MAPEDFILE_FAILED, null));
}
// todo 往mappedFile追加音讯
result = mappedFile.appendMessage(msg, this.appendMessageCallback);
...
这一部分就比拟重要了,首先是从mappedFileQueue中获取最初一个MappedFile,这个就是拿汇合最初一个元素,因为都是有序的,最初一个元素就是最初一个MappedFile,接着就是获取锁了,这个锁也是比拟有考究的,能够设置应用ReentrantLock 也能够设置应用cas,默认是应用cas,接着就是设置beginTimeInLock这个变量了,这个变量咱们在判断os page cache忙碌的时候说过,就是获取到锁的一个工夫戳,在开释锁之前会重置成0,接着就是判断mappedFile是不是null或者是不是满了,如果是的话就要新建一个了。
接着就是最最最重要的了 往mappedFile中追加音讯,
mappedFile.appendMessage
/**
* 将音讯追加到MappedFile文件中
*/
public AppendMessageResult appendMessagesInner(final MessageExt messageExt, final AppendMessageCallback cb) {
assert messageExt != null;
assert cb != null;
// 获取MappedFile以后文件写指针
int currentPos = this.wrotePosition.get();
// 如果currentPos小于文件大小
if (currentPos < this.fileSize) {
ByteBuffer byteBuffer = writeBuffer != null ? writeBuffer.slice() : this.mappedByteBuffer.slice();
byteBuffer.position(currentPos);
AppendMessageResult result;
// 单个音讯
if (messageExt instanceof MessageExtBrokerInner) {
// todo
result = cb.doAppend(this.getFileFromOffset(), byteBuffer, this.fileSize - currentPos, (MessageExtBrokerInner) messageExt);
// 批量音讯
} else if (messageExt instanceof MessageExtBatch) {
result = cb.doAppend(this.getFileFromOffset(), byteBuffer, this.fileSize - currentPos, (MessageExtBatch) messageExt);
} else {
return new AppendMessageResult(AppendMessageStatus.UNKNOWN_ERROR);
}
this.wrotePosition.addAndGet(result.getWroteBytes());
this.storeTimestamp = result.getStoreTimestamp();
return result;
}
// 如果currentPos大于或等于文件大小,表明文件已写满,抛出异样
log.error("MappedFile.appendMessage return null, wrotePosition: {} fileSize: {}", currentPos, this.fileSize);
return new AppendMessageResult(AppendMessageStatus.UNKNOWN_ERROR);
}
这里首先获取了一下这个mappedFile写到哪个地位了,它这个地位是从0开始的,而后判断一下以后地位与文件大小做比照,要是大于的就超了文件大小了,接着是获取writerbuffer因为这里是没有开启transientStorePool的,所以它是个空的,就会应用mmapedByteBuffer,接着就是调用回调的doAppend追加音讯了,咱们看下它的参数, 第一个是开始offset,这个offset是commitlog的一个offset,举个例子,第一个MappedFile的开始offset是0,而后一个MappedFile 的大小是1g,而后第二个MappedFile就得从1073741824(1g)开始了,第二个参数是bytebuffer,这个不必多说,第三个是这个MappedFile还空多少字节没用,第四个就是音讯了。
咱们来看下这个doAppend办法,这个也有点长,咱们须要离开看下:
/**
* // 只是将音讯追加到内存中
* @param fileFromOffset 文件的第一个偏移量(就是MappedFile是从哪个中央开始的)
*/
public AppendMessageResult doAppend(final long fileFromOffset, final ByteBuffer byteBuffer, final int maxBlank,
final MessageExtBrokerInner msgInner) {
// STORETIMESTAMP + STOREHOSTADDRESS + OFFSET <br>
// PHY OFFSET
long wroteOffset = fileFromOffset + byteBuffer.position();
int sysflag = msgInner.getSysFlag();
int bornHostLength = (sysflag & MessageSysFlag.BORNHOST_V6_FLAG) == 0 ? 4 + 4 : 16 + 4;
int storeHostLength = (sysflag & MessageSysFlag.STOREHOSTADDRESS_V6_FLAG) == 0 ? 4 + 4 : 16 + 4;
ByteBuffer bornHostHolder = ByteBuffer.allocate(bornHostLength);
ByteBuffer storeHostHolder = ByteBuffer.allocate(storeHostLength);
this.resetByteBuffer(storeHostHolder, storeHostLength);
// 创立全局惟一音讯id
String msgId;
if ((sysflag & MessageSysFlag.STOREHOSTADDRESS_V6_FLAG) == 0) {
msgId = MessageDecoder.createMessageId(this.msgIdMemory, msgInner.getStoreHostBytes(storeHostHolder), wroteOffset);
} else {
msgId = MessageDecoder.createMessageId(this.msgIdV6Memory, msgInner.getStoreHostBytes(storeHostHolder), wroteOffset);
}
// Record ConsumeQueue information
keyBuilder.setLength(0);
keyBuilder.append(msgInner.getTopic());
keyBuilder.append('-');
keyBuilder.append(msgInner.getQueueId());
String key = keyBuilder.toString();
// 获取该音讯在音讯队列的物理偏移量
Long queueOffset = CommitLog.this.topicQueueTable.get(key);
if (null == queueOffset) {
queueOffset = 0L;
CommitLog.this.topicQueueTable.put(key, queueOffset);
}
// Transaction messages that require special handling
final int tranType = MessageSysFlag.getTransactionValue(msgInner.getSysFlag());
switch (tranType) {
// Prepared and Rollback message is not consumed, will not enter the
// consumer queue
case MessageSysFlag.TRANSACTION_PREPARED_TYPE:
case MessageSysFlag.TRANSACTION_ROLLBACK_TYPE:
queueOffset = 0L;
break;
case MessageSysFlag.TRANSACTION_NOT_TYPE:
case MessageSysFlag.TRANSACTION_COMMIT_TYPE:
default:
break;
}
这一部分次要就是 计算了一下这个音讯写在commitlog中的一个offset,接着就是生成一个msgId,而后依据topic 与queueId从缓存中获取了一下这个queueId对应的一个queue的offset,这个其实就是增加一个音讯加1,而后就是事务的货色了,如果有事务,而后还在筹备阶段或者回滚阶段,就将queue offset 设置成0,再往下其实就是解决音讯,而后写到buffer中了。
/**
* Serialize message
*/
final byte[] propertiesData =
msgInner.getPropertiesString() == null ? null : msgInner.getPropertiesString().getBytes(MessageDecoder.CHARSET_UTF8);
final int propertiesLength = propertiesData == null ? 0 : propertiesData.length;
if (propertiesLength > Short.MAX_VALUE) {
log.warn("putMessage message properties length too long. length={}", propertiesData.length);
return new AppendMessageResult(AppendMessageStatus.PROPERTIES_SIZE_EXCEEDED);
}
final byte[] topicData = msgInner.getTopic().getBytes(MessageDecoder.CHARSET_UTF8);
final int topicLength = topicData.length;
final int bodyLength = msgInner.getBody() == null ? 0 : msgInner.getBody().length;
// todo 计算音讯总长度
final int msgLen = calMsgLength(msgInner.getSysFlag(), bodyLength, topicLength, propertiesLength);
// Exceeds the maximum message
if (msgLen > this.maxMessageSize) { // 最大音讯长度 65536
CommitLog.log.warn("message size exceeded, msg total size: " + msgLen + ", msg body size: " + bodyLength
+ ", maxMessageSize: " + this.maxMessageSize);
return new AppendMessageResult(AppendMessageStatus.MESSAGE_SIZE_EXCEEDED);
}
...
这里首先是获取了一下音讯外面的properties,将它转成字节数组,计算了一下长度,接着就是将topic转成字节数据,计算了一下长度,获取了一下body的长度,就是你往Message塞得内容长度,重点来,计算 音讯的总长度,而后判断一下长度是否超长。其中calMsgLength如下:
// 计算音讯长度
protected static int calMsgLength(int sysFlag, int bodyLength, int topicLength, int propertiesLength) {
int bornhostLength = (sysFlag & MessageSysFlag.BORNHOST_V6_FLAG) == 0 ? 8 : 20;
int storehostAddressLength = (sysFlag & MessageSysFlag.STOREHOSTADDRESS_V6_FLAG) == 0 ? 8 : 20;
final int msgLen = 4 //TOTALSIZE 音讯条目总长度,4字节
+ 4 //MAGICCODE 魔数,4字节。固定值0xdaa320a7
+ 4 //BODYCRC 音讯体的crc校验码,4字节
+ 4 //QUEUEID 音讯生产队列ID,4字节
+ 4 //FLAG 音讯标记,RocketMQ对其不做解决,供应用程序应用, 默认4字节
+ 8 //QUEUEOFFSET 音讯在ConsumeQuene文件中的物理偏移量,8字节。
+ 8 //PHYSICALOFFSET 音讯在CommitLog文件中的物理偏移量,8字节
+ 4 //SYSFLAG 音讯零碎标记,例如是否压缩、是否是事务音讯 等,4字节
+ 8 //BORNTIMESTAMP 音讯生产者调用音讯发送API的工夫戳,8字 节
+ bornhostLength //BORNHOST 音讯发送者IP、端口号,8字节
+ 8 //STORETIMESTAMP 音讯存储工夫戳,8字节
+ storehostAddressLength //STOREHOSTADDRESS Broker服务器IP+端口号,8字节
+ 4 //RECONSUMETIMES 音讯重试次数,4字节
+ 8 //Prepared Transaction Offset 事务音讯的物理偏移量,8 字节。
+ 4 // 音讯体长度,4字节
+ (bodyLength > 0 ? bodyLength : 0) // BODY 音讯体内容,长度为bodyLenth中存储的值
+ 1 // 主题存储长度,1字节,示意主题名称不能超过255个字符。
+ topicLength //TOPIC 主题,长度为TopicLength中存储的值
+ 2 // 音讯属性长度,2字节,示意音讯属性长 度不能超过65536个字符。
+ (propertiesLength > 0 ? propertiesLength : 0) //propertiesLength 音讯属性,长度为PropertiesLength中存储的 值
+ 0;
return msgLen;
}
持续:
...
// todo 音讯长度+END_FILE_MIN_BLANK_LENGTH 大于commitLog的闲暇空间,则返回END_OF_FILE
if ((msgLen + END_FILE_MIN_BLANK_LENGTH) > maxBlank) {
this.resetByteBuffer(this.msgStoreItemMemory, maxBlank);
// 1 TOTALSIZE 4字节存储以后文件的残余空间
this.msgStoreItemMemory.putInt(maxBlank);
// 2 MAGICCODE 4字节存储魔数
this.msgStoreItemMemory.putInt(CommitLog.BLANK_MAGIC_CODE);
// 3 The remaining space may be any value
// Here the length of the specially set maxBlank
final long beginTimeMills = CommitLog.this.defaultMessageStore.now();
byteBuffer.put(this.msgStoreItemMemory.array(), 0, maxBlank);
return new AppendMessageResult(AppendMessageStatus.END_OF_FILE, wroteOffset, maxBlank, msgId, msgInner.getStoreTimestamp(),
queueOffset, CommitLog.this.defaultMessageStore.now() - beginTimeMills);
}
...
判断剩下的空间能不能放开,如果放不开的话,就塞上一个完结的货色,8个字节是正经的,剩下的随便,而后返回文件满了的状态。
...
// Initialization of storage space
// 初始化存储空间
this.resetByteBuffer(msgStoreItemMemory, msgLen);
// 1 TOTALSIZE
this.msgStoreItemMemory.putInt(msgLen);
// 2 MAGICCODE
this.msgStoreItemMemory.putInt(CommitLog.MESSAGE_MAGIC_CODE);
// 3 BODYCRC
this.msgStoreItemMemory.putInt(msgInner.getBodyCRC());
// 4 QUEUEID
this.msgStoreItemMemory.putInt(msgInner.getQueueId());
// 5 FLAG
this.msgStoreItemMemory.putInt(msgInner.getFlag());
// 6 QUEUEOFFSET
this.msgStoreItemMemory.putLong(queueOffset);
// 7 PHYSICALOFFSET
this.msgStoreItemMemory.putLong(fileFromOffset + byteBuffer.position());
// 8 SYSFLAG
this.msgStoreItemMemory.putInt(msgInner.getSysFlag());
// 9 BORNTIMESTAMP
this.msgStoreItemMemory.putLong(msgInner.getBornTimestamp());
// 10 BORNHOST
this.resetByteBuffer(bornHostHolder, bornHostLength);
this.msgStoreItemMemory.put(msgInner.getBornHostBytes(bornHostHolder));
// 11 STORETIMESTAMP
this.msgStoreItemMemory.putLong(msgInner.getStoreTimestamp());
// 12 STOREHOSTADDRESS
this.resetByteBuffer(storeHostHolder, storeHostLength);
this.msgStoreItemMemory.put(msgInner.getStoreHostBytes(storeHostHolder));
// 13 RECONSUMETIMES
this.msgStoreItemMemory.putInt(msgInner.getReconsumeTimes());
// 14 Prepared Transaction Offset
this.msgStoreItemMemory.putLong(msgInner.getPreparedTransactionOffset());
// 15 BODY
this.msgStoreItemMemory.putInt(bodyLength);
if (bodyLength > 0)
this.msgStoreItemMemory.put(msgInner.getBody());
// 16 TOPIC
this.msgStoreItemMemory.put((byte) topicLength);
this.msgStoreItemMemory.put(topicData);
// 17 PROPERTIES
this.msgStoreItemMemory.putShort((short) propertiesLength);
if (propertiesLength > 0)
this.msgStoreItemMemory.put(propertiesData);
final long beginTimeMills = CommitLog.this.defaultMessageStore.now();
// Write messages to the queue buffer
byteBuffer.put(this.msgStoreItemMemory.array(), 0, msgLen);
...
这个就是封装音讯了,最初将音讯放到byteBuffer中。
...
// 创立AppendMessageResult
AppendMessageResult result = new AppendMessageResult(AppendMessageStatus.PUT_OK, wroteOffset, msgLen, msgId,
msgInner.getStoreTimestamp(), queueOffset, CommitLog.this.defaultMessageStore.now() - beginTimeMills);
switch (tranType) {
case MessageSysFlag.TRANSACTION_PREPARED_TYPE:
case MessageSysFlag.TRANSACTION_ROLLBACK_TYPE:
break;
case MessageSysFlag.TRANSACTION_NOT_TYPE:
case MessageSysFlag.TRANSACTION_COMMIT_TYPE:
// The next update ConsumeQueue information
// 更新音讯队列的逻辑偏移量
CommitLog.this.topicQueueTable.put(key, ++queueOffset);
break;
default:
break;
}
return result;
}
最初就是封装追加音讯的后果是put_ok,而后更新queue offset ,其实就是+1。
接下来咱们回过头来看下appendMessagesInner的后半局部,
...
this.wrotePosition.addAndGet(result.getWroteBytes());
this.storeTimestamp = result.getStoreTimestamp();
return result;
这里其实就是更新了一下 这个MappedFile 写到哪个中央了,更新了下写入工夫。
回到commitLog的putMessage办法:
...
// todo 往mappedFile追加音讯
result = mappedFile.appendMessage(msg, this.appendMessageCallback);
switch (result.getStatus()) {
case PUT_OK:
break;
case END_OF_FILE:
unlockMappedFile = mappedFile;
// Create a new file, re-write the message
mappedFile = this.mappedFileQueue.getLastMappedFile(0);
if (null == mappedFile) {
// XXX: warn and notify me
log.error("create mapped file2 error, topic: " + msg.getTopic() + " clientAddr: " + msg.getBornHostString());
beginTimeInLock = 0;
return CompletableFuture.completedFuture(new PutMessageResult(PutMessageStatus.CREATE_MAPEDFILE_FAILED, result));
}
result = mappedFile.appendMessage(msg, this.appendMessageCallback);
break;
case MESSAGE_SIZE_EXCEEDED:
case PROPERTIES_SIZE_EXCEEDED:
beginTimeInLock = 0;
return CompletableFuture.completedFuture(new PutMessageResult(PutMessageStatus.MESSAGE_ILLEGAL, result));
case UNKNOWN_ERROR:
beginTimeInLock = 0;
return CompletableFuture.completedFuture(new PutMessageResult(PutMessageStatus.UNKNOWN_ERROR, result));
default:
beginTimeInLock = 0;
return CompletableFuture.completedFuture(new PutMessageResult(PutMessageStatus.UNKNOWN_ERROR, result));
}
elapsedTimeInLock = this.defaultMessageStore.getSystemClock().now() - beginLockTimestamp;
beginTimeInLock = 0;
} finally {
putMessageLock.unlock();
}
...
这里追加实现了,就须要判断追加状态了,如果是那种MappedFile放不开音讯的状况,它会从新获取一个MappedFile,而后从新追加,在开释锁之前,它还会将beginTimeInLock这个字段重置为0;
...
if (elapsedTimeInLock > 500) {
log.warn("[NOTIFYME]putMessage in lock cost time(ms)={}, bodyLength={} AppendMessageResult={}", elapsedTimeInLock, msg.getBody().length, result);
}
if (null != unlockMappedFile && this.defaultMessageStore.getMessageStoreConfig().isWarmMapedFileEnable()) {
this.defaultMessageStore.unlockMappedFile(unlockMappedFile);
}
PutMessageResult putMessageResult = new PutMessageResult(PutMessageStatus.PUT_OK, result);
// Statistics
storeStatsService.getSinglePutMessageTopicTimesTotal(msg.getTopic()).incrementAndGet();
storeStatsService.getSinglePutMessageTopicSizeTotal(topic).addAndGet(result.getWroteBytes());
// todo 音讯首先进入pagecache,而后执行刷盘操作,
CompletableFuture<PutMessageStatus> flushResultFuture = submitFlushRequest(result, msg);
// todo 接着调用submitReplicaRequest办法将音讯提交到HaService,进行数据复制
CompletableFuture<PutMessageStatus> replicaResultFuture = submitReplicaRequest(result, msg);
// todo 这里应用了ComplateFuture的thenCombine办法,将刷盘、复制当成一
// todo 个联结工作执行,这里设置音讯追加的最终状态
return flushResultFuture.thenCombine(replicaResultFuture, (flushStatus, replicaStatus) -> {
if (flushStatus != PutMessageStatus.PUT_OK) {
putMessageResult.setPutMessageStatus(flushStatus);
}
if (replicaStatus != PutMessageStatus.PUT_OK) {
putMessageResult.setPutMessageStatus(replicaStatus);
if (replicaStatus == PutMessageStatus.FLUSH_SLAVE_TIMEOUT) {
log.error("do sync transfer other node, wait return, but failed, topic: {} tags: {} client address: {}",
msg.getTopic(), msg.getTags(), msg.getBornHostNameString());
}
}
return putMessageResult;
});
}
判断了一下耗时,如果是大于500ms的话,打印正告,封装put音讯的后果,统计store,能够看到前面调用了2个办法,一个是刷盘的,一个是同步音讯的,咱们这里要看下这个刷盘动作:
public CompletableFuture<PutMessageStatus> submitFlushRequest(AppendMessageResult result, MessageExt messageExt) {
// Synchronization flush 同步刷盘
if (FlushDiskType.SYNC_FLUSH == this.defaultMessageStore.getMessageStoreConfig().getFlushDiskType()) {
final GroupCommitService service = (GroupCommitService) this.flushCommitLogService;
if (messageExt.isWaitStoreMsgOK()) {
// 构建GroupCommitRequest同步工作并提交到GroupCommitRequest
GroupCommitRequest request = new GroupCommitRequest(result.getWroteOffset() + result.getWroteBytes(),
this.defaultMessageStore.getMessageStoreConfig().getSyncFlushTimeout());
// 刷盘申请
service.putRequest(request);
return request.future();
} else {
service.wakeup();
return CompletableFuture.completedFuture(PutMessageStatus.PUT_OK);
}
}
// Asynchronous flush 异步刷盘 这个就是靠os
else {
if (!this.defaultMessageStore.getMessageStoreConfig().isTransientStorePoolEnable()) {
flushCommitLogService.wakeup();
} else {
commitLogService.wakeup();
}
return CompletableFuture.completedFuture(PutMessageStatus.PUT_OK);
}
}
如果broker配置的SYNC_FLUSH 并且是个同步音讯,这个时候就会创立一个刷盘申请,而后提交刷盘申请,这个时候会等着刷盘实现,默认就是5s。
接着就是到存储器的putMessage办法的后半局部了:
...
// todo 存储音讯
CompletableFuture<PutMessageResult> putResultFuture = this.commitLog.asyncPutMessage(msg);
putResultFuture.thenAccept((result) -> {
long elapsedTime = this.getSystemClock().now() - beginTime;
if (elapsedTime > 500) {
log.warn("putMessage not in lock elapsed time(ms)={}, bodyLength={}", elapsedTime, msg.getBody().length);
}
// 记录状态
this.storeStatsService.setPutMessageEntireTimeMax(elapsedTime);
if (null == result || !result.isOk()) {
// 记录状态
this.storeStatsService.getPutMessageFailedTimes().incrementAndGet();
}
});
return putResultFuture;
commitlog存入音讯之后,咱们这块也就算是实现了,最初就是回到那个processor,而后将put后果写入对应的channel给返回去,通知音讯生产者音讯写入后果 。音讯存储其实就是找对应的MappedFile,依照肯定的格局往文件外面写入,须要留神的是内存映射文件。
这里附一张音讯存储字段存储程序与字段长度的图:
4. 总结
本文次要剖析了 broker 接管producer音讯的流程,流程如下:
- 解决音讯接管的底层服务为 netty,在BrokerController#start办法中启动
- netty服务中,解决音讯接管的channelHandler为NettyServerHandler,最终会调用SendMessageProcessor#processRequest来解决音讯接管
- 音讯接管流程的最初,MappedFile#appendMessage(…)办法会将音讯内容写入到commitLog文件中。
发表回复