问题列表
Broker 怎么响应Consumer请求?
Broker 怎么维护ConsumeQueue?
Broker 怎么处理事务消息的 ConsumeQueue ?
Broker 怎么处理定时消息的 ConsumeQueue?
Broker 怎么处理回溯消费请求?
Broker 的消息是 at least once还是exactly only once?
怎么响应Consumer请求?
原理:
如上图所示,RocketMQ将所有消息都放在CommitLog里面,消费是维护一个ConsumeQueue帮助Consumer消费.pull操作要读两次,先读ConsumeQueue得到offset，再读CommitLog得到消息内容.
ConsumeQueue有一个长度20的ByteBufferb变量byteBufferIndex,里面维护者消息偏移量,消息发现,tags的hashcode
//消息偏移量
this.byteBufferIndex.putLong(offset);
//消息大小
this.byteBufferIndex.putInt(size);
//tags的HashCode
this.byteBufferIndex.putLong(tagsCode);BrokerController.initialize方法中会注册PullMessageProcessor来处理pull message 请求
this.remotingServer.registerProcessor(RequestCode.PULL_MESSAGE, this.pullMessageProcessor, this.pullMessageExecutor);
this.pullMessageProcessor.registerConsumeMessageHook(consumeMessageHookList);在PullMessageProcessor.processRequest中又委托给DefaultMessageStore获取
final GetMessageResult getMessageResult =
this.brokerController.getMessageStore().getMessage(requestHeader.getConsumerGroup(), requestHeader.getTopic(),
requestHeader.getQueueId(), requestHeader.getQueueOffset(), requestHeader.getMaxMsgNums(), messageFilter);
DefaultMessageStore.getMessage
public GetMessageResult getMessage(final String group, final String topic, final int queueId, final long offset, final int maxMsgNums,
final MessageFilter messageFilter) {
if (this.shutdown) {
log.warn("message store has shutdown, so getMessage is forbidden");
return null;
}
if (!this.runningFlags.isReadable()) {
log.warn("message store is not readable, so getMessage is forbidden " + this.runningFlags.getFlagBits());
return null;
}
long beginTime = this.getSystemClock().now();
GetMessageStatus status = GetMessageStatus.NO_MESSAGE_IN_QUEUE;
long nextBeginOffset = offset;
long minOffset = 0;
long maxOffset = 0;
GetMessageResult getResult = new GetMessageResult();
final long maxOffsetPy = this.commitLog.getMaxOffset();
//根据topic和queueId找到ConsumeQueue
ConsumeQueue consumeQueue = findConsumeQueue(topic, queueId);
if (consumeQueue != null) {
minOffset = consumeQueue.getMinOffsetInQueue();
maxOffset = consumeQueue.getMaxOffsetInQueue();
if (maxOffset == 0) {
status = GetMessageStatus.NO_MESSAGE_IN_QUEUE;
nextBeginOffset = nextOffsetCorrection(offset, 0);
} else if (offset < minOffset) {
status = GetMessageStatus.OFFSET_TOO_SMALL;
nextBeginOffset = nextOffsetCorrection(offset, minOffset);
} else if (offset == maxOffset) {
status = GetMessageStatus.OFFSET_OVERFLOW_ONE;
nextBeginOffset = nextOffsetCorrection(offset, offset);
} else if (offset > maxOffset) {
status = GetMessageStatus.OFFSET_OVERFLOW_BADLY;
if (0 == minOffset) {
nextBeginOffset = nextOffsetCorrection(offset, minOffset);
} else {
nextBeginOffset = nextOffsetCorrection(offset, maxOffset);
}
} else {
//前面都在处理异常,这里开始真正获取
//获取consumeQueue offset之后所有可读的offset,consumeQueue也是文件存储
SelectMappedBufferResult bufferConsumeQueue = consumeQueue.getIndexBuffer(offset);
if (bufferConsumeQueue != null) {
try {
status = GetMessageStatus.NO_MATCHED_MESSAGE;
long nextPhyFileStartOffset = Long.MIN_VALUE;
long maxPhyOffsetPulling = 0;
int i = 0;
final int maxFilterMessageCount = Math.max(16000, maxMsgNums * ConsumeQueue.CQ_STORE_UNIT_SIZE);
final boolean diskFallRecorded = this.messageStoreConfig.isDiskFallRecorded();
ConsumeQueueExt.CqExtUnit cqExtUnit = new ConsumeQueueExt.CqExtUnit();
//循环单个获取
for (; i < bufferConsumeQueue.getSize() && i < maxFilterMessageCount; i += ConsumeQueue.CQ_STORE_UNIT_SIZE) {
long offsetPy = bufferConsumeQueue.getByteBuffer().getLong();
int sizePy = bufferConsumeQueue.getByteBuffer().getInt();
long tagsCode = bufferConsumeQueue.getByteBuffer().getLong();
maxPhyOffsetPulling = offsetPy;
if (nextPhyFileStartOffset != Long.MIN_VALUE) {
if (offsetPy < nextPhyFileStartOffset)
continue;
}
boolean isInDisk = checkInDiskByCommitOffset(offsetPy, maxOffsetPy);
if (this.isTheBatchFull(sizePy, maxMsgNums, getResult.getBufferTotalSize(), getResult.getMessageCount(),
isInDisk)) {
break;
}
boolean extRet = false;
if (consumeQueue.isExtAddr(tagsCode)) {
extRet = consumeQueue.getExt(tagsCode, cqExtUnit);
if (extRet) {
tagsCode = cqExtUnit.getTagsCode();
} else {
// can't find ext content.Client will filter messages by tag also.
log.error("[BUG] can't find consume queue extend file content!addr={}, offsetPy={}, sizePy={}, topic={}, group={}",
tagsCode, offsetPy, sizePy, topic, group);
}
}
if (messageFilter != null
&& !messageFilter.isMatchedByConsumeQueue(tagsCode, extRet ? cqExtUnit : null)) {
if (getResult.getBufferTotalSize() == 0) {
status = GetMessageStatus.NO_MATCHED_MESSAGE;
}
continue;
}
//根据偏移量和大小从CommitLog获取消息
SelectMappedBufferResult selectResult = this.commitLog.getMessage(offsetPy, sizePy);
if (null == selectResult) {
if (getResult.getBufferTotalSize() == 0) {
status = GetMessageStatus.MESSAGE_WAS_REMOVING;
}
nextPhyFileStartOffset = this.commitLog.rollNextFile(offsetPy);
continue;
}
if (messageFilter != null
&& !messageFilter.isMatchedByCommitLog(selectResult.getByteBuffer().slice(), null)) {
if (getResult.getBufferTotalSize() == 0) {
status = GetMessageStatus.NO_MATCHED_MESSAGE;
}
// release...
selectResult.release();
continue;
}
this.storeStatsService.getGetMessageTransferedMsgCount().incrementAndGet();
//将查到的消息放入返回值
getResult.addMessage(selectResult);
status = GetMessageStatus.FOUND;
nextPhyFileStartOffset = Long.MIN_VALUE;
}
if (diskFallRecorded) {
long fallBehind = maxOffsetPy - maxPhyOffsetPulling;
brokerStatsManager.recordDiskFallBehindSize(group, topic, queueId, fallBehind);
}
nextBeginOffset = offset + (i / ConsumeQueue.CQ_STORE_UNIT_SIZE);
long diff = maxOffsetPy - maxPhyOffsetPulling;
long memory = (long) (StoreUtil.TOTAL_PHYSICAL_MEMORY_SIZE
* (this.messageStoreConfig.getAccessMessageInMemoryMaxRatio() / 100.0));
getResult.setSuggestPullingFromSlave(diff > memory);
} finally {
bufferConsumeQueue.release();
}
} else {
status = GetMessageStatus.OFFSET_FOUND_NULL;
nextBeginOffset = nextOffsetCorrection(offset, consumeQueue.rollNextFile(offset));
log.warn("consumer request topic: " + topic + "offset: " + offset + " minOffset: " + minOffset + " maxOffset: "
+ maxOffset + ", but access logic queue failed.");
}
}
} else {
status = GetMessageStatus.NO_MATCHED_LOGIC_QUEUE;
nextBeginOffset = nextOffsetCorrection(offset, 0);
}
if (GetMessageStatus.FOUND == status) {
this.storeStatsService.getGetMessageTimesTotalFound().incrementAndGet();
} else {
this.storeStatsService.getGetMessageTimesTotalMiss().incrementAndGet();
}
long eclipseTime = this.getSystemClock().now() - beginTime;
this.storeStatsService.setGetMessageEntireTimeMax(eclipseTime);
getResult.setStatus(status);
getResult.setNextBeginOffset(nextBeginOffset);
getResult.setMaxOffset(maxOffset);
getResult.setMinOffset(minOffset);
return getResult;
}CommitLog.getMessage
public SelectMappedBufferResult getMessage(final long offset, final int size) {
int mappedFileSize = this.defaultMessageStore.getMessageStoreConfig().getMapedFileSizeCommitLog();
//根据offset获取到映射文件
MappedFile mappedFile = this.mappedFileQueue.findMappedFileByOffset(offset, offset == 0);
if (mappedFile != null) {
int pos = (int) (offset % mappedFileSize);
//从映射文件中获取指定位置的数据
return mappedFile.selectMappedBuffer(pos, size);
}
return null;
}MappedFile.selectMappedBuffer
public SelectMappedBufferResult selectMappedBuffer(int pos, int size) {
int readPosition = getReadPosition();
if ((pos + size) <= readPosition) {
if (this.hold()) {
//Java NIO 获取指定pos数据
ByteBuffer byteBuffer = this.mappedByteBuffer.slice();
byteBuffer.position(pos);
ByteBuffer byteBufferNew = byteBuffer.slice();
byteBufferNew.limit(size);
return new SelectMappedBufferResult(this.fileFromOffset + pos, byteBufferNew, size, this);
} else {
log.warn("matched, but hold failed, request pos: " + pos + ", fileFromOffset: "
+ this.fileFromOffset);
}
} else {
log.warn("selectMappedBuffer request pos invalid, request pos: " + pos + ", size: " + size
+ ", fileFromOffset: " + this.fileFromOffset);
}
return null;
}消息获取流程明白了,看下怎么获取ConsumeQueue,它是维护在一个以queueId为key的ConcurrentMap中,没有就新建一个.
public ConsumeQueue findConsumeQueue(String topic, int queueId) {
ConcurrentMap<Integer, ConsumeQueue> map = consumeQueueTable.get(topic);
if (null == map) {
ConcurrentMap<Integer, ConsumeQueue> newMap = new ConcurrentHashMap<Integer, ConsumeQueue>(128);
ConcurrentMap<Integer, ConsumeQueue> oldMap = consumeQueueTable.putIfAbsent(topic, newMap);
if (oldMap != null) {
map = oldMap;
} else {
map = newMap;
}
}
ConsumeQueue logic = map.get(queueId);
if (null == logic) {
//m
ConsumeQueue newLogic = new ConsumeQueue(//
topic, //
queueId, //
StorePathConfigHelper.getStorePathConsumeQueue(this.messageStoreConfig.getStorePathRootDir()), //
this.getMessageStoreConfig().getMapedFileSizeConsumeQueue(), //
this);
ConsumeQueue oldLogic = map.putIfAbsent(queueId, newLogic);
if (oldLogic != null) {
logic = oldLogic;
} else {
logic = newLogic;
}
}
return logic;
}Broker 怎么维护ConsumeQueue?
很容易的想到,是在存放Message的时候维护了ConsumeQueue.但是相关代码没找到,我们从顶层往上层找
CosumeQueue.putMessagePositionInfo
-> CosumeQueue.putMessagePositionInfoWrapper
-> DefaultMessageStore.putMessagePositionInfo
-> CommitLogDispatcherBuildConsumeQueue.dispatch
-> DefaultMessageStore.doDispatch
-> ReputMessageService.doReput
private void doReput() {
for (boolean doNext = true; this.isCommitLogAvailable() && doNext; ) {
if (DefaultMessageStore.this.getMessageStoreConfig().isDuplicationEnable() //
&& this.reputFromOffset >= DefaultMessageStore.this.getConfirmOffset()) {
break;
}
//从CommitLog中读取上次偏移量之后的新消息
SelectMappedBufferResult result = DefaultMessageStore.this.commitLog.getData(reputFromOffset);
if (result != null) {
try {
this.reputFromOffset = result.getStartOffset();
//顺序读
for (int readSize = 0; readSize < result.getSize() && doNext; ) {
//构建请求去维护consumeQueue的请求
DispatchRequest dispatchRequest =
DefaultMessageStore.this.commitLog.checkMessageAndReturnSize(result.getByteBuffer(), false, false);
int size = dispatchRequest.getMsgSize();
if (dispatchRequest.isSuccess()) {
if (size > 0) {
DefaultMessageStore.this.doDispatch(dispatchRequest);
if (BrokerRole.SLAVE != DefaultMessageStore.this.getMessageStoreConfig().getBrokerRole()
&& DefaultMessageStore.this.brokerConfig.isLongPollingEnable()) {
DefaultMessageStore.this.messageArrivingListener.arriving(dispatchRequest.getTopic(),
dispatchRequest.getQueueId(), dispatchRequest.getConsumeQueueOffset() + 1,
dispatchRequest.getTagsCode(), dispatchRequest.getStoreTimestamp(),
dispatchRequest.getBitMap(), dispatchRequest.getPropertiesMap());
}
// FIXED BUG By shijia
this.reputFromOffset += size;
readSize += size;
if (DefaultMessageStore.this.getMessageStoreConfig().getBrokerRole() == BrokerRole.SLAVE) {
DefaultMessageStore.this.storeStatsService
.getSinglePutMessageTopicTimesTotal(dispatchRequest.getTopic()).incrementAndGet();
DefaultMessageStore.this.storeStatsService
.getSinglePutMessageTopicSizeTotal(dispatchRequest.getTopic())
.addAndGet(dispatchRequest.getMsgSize());
}
} else if (size == 0) {
this.reputFromOffset = DefaultMessageStore.this.commitLog.rollNextFile(this.reputFromOffset);
readSize = result.getSize();
}
} else if (!dispatchRequest.isSuccess()) {
if (size > 0) {
log.error("[BUG]read total count not equals msg total size. reputFromOffset={}", reputFromOffset);
this.reputFromOffset += size;
} else {
doNext = false;
if (DefaultMessageStore.this.brokerConfig.getBrokerId() == MixAll.MASTER_ID) {
log.error("[BUG]the master dispatch message to consume queue error, COMMITLOG OFFSET: {}",
this.reputFromOffset);
this.reputFromOffset += result.getSize() - readSize;
}
}
}
}
} finally {
result.release();
}
} else {
doNext = false;
}
}
}逻辑:启个线程,不断地查询有没有新的提交,如果有就请求维护consumeQueue.
Broker 怎么处理事务消息的ConsumeQueue?
逻辑:
1.prepare消息不会放入ConsumeQueue
2.commit消息才能被ConsumeQueue
CommitLogDispatcherBuildConsumeQueue.dispatch
class CommitLogDispatcherBuildConsumeQueue implements CommitLogDispatcher {
@Override
public void dispatch(DispatchRequest request) {
final int tranType = MessageSysFlag.getTransactionValue(request.getSysFlag());
switch (tranType) {
case MessageSysFlag.TRANSACTION_NOT_TYPE:
// commit操作会放入consumeQueue
case MessageSysFlag.TRANSACTION_COMMIT_TYPE:
DefaultMessageStore.this.putMessagePositionInfo(request);
break;
// prepare消息不会放入ConsumeQueue
case MessageSysFlag.TRANSACTION_PREPARED_TYPE:
case MessageSysFlag.TRANSACTION_ROLLBACK_TYPE:
break;
}
}
}Broker 怎么处理定时消息的 ConsumeQueue?
Producer那篇有讲,定时消息是吧topic和queueId存储到属性中,真正存储与ScheduleTopic上,时间到了才放入原有topic,故ConsumeQueue无需特殊维护.
Broker 怎么处理回溯消费请求?
逻辑:
1.在ConsumeQueue根据时间查询offset
2.请求Consumer Group中所有Consumer重置offset
代码:
AdminBrokerProcessor.processRequest
case RequestCode.INVOKE_BROKER_TO_RESET_OFFSET:
return this.resetOffset(ctx, request);Broker2Client.resetOffset
public RemotingCommand resetOffset(ChannelHandlerContext ctx, RemotingCommand request) throws RemotingCommandException {
final ResetOffsetRequestHeader requestHeader =
(ResetOffsetRequestHeader) request.decodeCommandCustomHeader(ResetOffsetRequestHeader.class);
log.info("[reset-offset] reset offset started by {}. topic={}, group={}, timestamp={}, isForce={}",
RemotingHelper.parseChannelRemoteAddr(ctx.channel()), requestHeader.getTopic(), requestHeader.getGroup(),
requestHeader.getTimestamp(), requestHeader.isForce());
boolean isC = false;
LanguageCode language = request.getLanguage();
switch (language) {
case CPP:
isC = true;
break;
}
return this.brokerController.getBroker2Client().resetOffset(requestHeader.getTopic(), requestHeader.getGroup(),
requestHeader.getTimestamp(), requestHeader.isForce(), isC);
}主要逻辑在此方法
Broker2Client.resetOffset
public RemotingCommand resetOffset(String topic, String group, long timeStamp, boolean isForce,
boolean isC) {
final RemotingCommand response = RemotingCommand.createResponseCommand(null);
TopicConfig topicConfig = this.brokerController.getTopicConfigManager().selectTopicConfig(topic);
if (null == topicConfig) {
log.error("[reset-offset] reset offset failed, no topic in this broker. topic={}", topic);
response.setCode(ResponseCode.SYSTEM_ERROR);
response.setRemark("[reset-offset] reset offset failed, no topic in this broker. topic=" + topic);
return response;
}
Map<MessageQueue, Long> offsetTable = new HashMap<MessageQueue, Long>();
for (int i = 0; i < topicConfig.getWriteQueueNums(); i++) {
MessageQueue mq = new MessageQueue();
mq.setBrokerName(this.brokerController.getBrokerConfig().getBrokerName());
mq.setTopic(topic);
mq.setQueueId(i);
long consumerOffset =
this.brokerController.getConsumerOffsetManager().queryOffset(group, topic, i);
if (-1 == consumerOffset) {
response.setCode(ResponseCode.SYSTEM_ERROR);
response.setRemark(String.format("THe consumer group <%s> not exist", group));
return response;
}
long timeStampOffset;
if (timeStamp == -1) {
timeStampOffset = this.brokerController.getMessageStore().getMaxOffsetInQueue(topic, i);
} else {
//根据时间找offset
timeStampOffset = this.brokerController.getMessageStore().getOffsetInQueueByTime(topic, i, timeStamp);
}
if (timeStampOffset < 0) {
log.warn("reset offset is invalid. topic={}, queueId={}, timeStampOffset={}", topic, i, timeStampOffset);
timeStampOffset = 0;
}
if (isForce || timeStampOffset < consumerOffset) {
offsetTable.put(mq, timeStampOffset);
} else {
offsetTable.put(mq, consumerOffset);
}
}
ResetOffsetRequestHeader requestHeader = new ResetOffsetRequestHeader();
requestHeader.setTopic(topic);
requestHeader.setGroup(group);
requestHeader.setTimestamp(timeStamp);
//给Consumer发送请求重置offset
RemotingCommand request =
RemotingCommand.createRequestCommand(RequestCode.RESET_CONSUMER_CLIENT_OFFSET, requestHeader);
if (isC) {
// c++ language
ResetOffsetBodyForC body = new ResetOffsetBodyForC();
List<MessageQueueForC> offsetList = convertOffsetTable2OffsetList(offsetTable);
body.setOffsetTable(offsetList);
request.setBody(body.encode());
} else {
// other language
ResetOffsetBody body = new ResetOffsetBody();
body.setOffsetTable(offsetTable);
request.setBody(body.encode());
}
ConsumerGroupInfo consumerGroupInfo =
this.brokerController.getConsumerManager().getConsumerGroupInfo(group);
if (consumerGroupInfo != null && !consumerGroupInfo.getAllChannel().isEmpty()) {
ConcurrentMap<Channel, ClientChannelInfo> channelInfoTable =
consumerGroupInfo.getChannelInfoTable();
for (Map.Entry<Channel, ClientChannelInfo> entry : channelInfoTable.entrySet()) {
int version = entry.getValue().getVersion();
if (version >= MQVersion.Version.V3_0_7_SNAPSHOT.ordinal()) {
try {
this.brokerController.getRemotingServer().invokeOneway(entry.getKey(), request, 5000);
log.info("[reset-offset] reset offset success. topic={}, group={}, clientId={}",
topic, group, entry.getValue().getClientId());
} catch (Exception e) {
log.error("[reset-offset] reset offset exception. topic={}, group={}",
new Object[] {topic, group}, e);
}
} else {
response.setCode(ResponseCode.SYSTEM_ERROR);
response.setRemark("the client does not support this feature. version="
+ MQVersion.getVersionDesc(version));
log.warn("[reset-offset] the client does not support this feature. version={}",
RemotingHelper.parseChannelRemoteAddr(entry.getKey()), MQVersion.getVersionDesc(version));
return response;
}
}
} else {
String errorInfo =
String.format("Consumer not online, so can not reset offset, Group: %s Topic: %s Timestamp: %d",
requestHeader.getGroup(),
requestHeader.getTopic(),
requestHeader.getTimestamp());
log.error(errorInfo);
response.setCode(ResponseCode.CONSUMER_NOT_ONLINE);
response.setRemark(errorInfo);
return response;
}
response.setCode(ResponseCode.SUCCESS);
ResetOffsetBody resBody = new ResetOffsetBody();
resBody.setOffsetTable(offsetTable);
response.setBody(resBody.encode());
return response;
}ConsumeQueue中没有时间戳啊,还是得去CommitLog中找
public long getOffsetInQueueByTime(final long timestamp) {
MappedFile mappedFile = this.mappedFileQueue.getMappedFileByTime(timestamp);
if (mappedFile != null) {
long offset = 0;
int low = minLogicOffset > mappedFile.getFileFromOffset() ? (int) (minLogicOffset - mappedFile.getFileFromOffset()) : 0;
int high = 0;
int midOffset = -1, targetOffset = -1, leftOffset = -1, rightOffset = -1;
long leftIndexValue = -1L, rightIndexValue = -1L;
long minPhysicOffset = this.defaultMessageStore.getMinPhyOffset();
SelectMappedBufferResult sbr = mappedFile.selectMappedBuffer(0);
if (null != sbr) {
ByteBuffer byteBuffer = sbr.getByteBuffer();
high = byteBuffer.limit() - CQ_STORE_UNIT_SIZE;
try {
//二分查找?
while (high >= low) {
midOffset = (low + high) / (2 * CQ_STORE_UNIT_SIZE) * CQ_STORE_UNIT_SIZE;
byteBuffer.position(midOffset);
long phyOffset = byteBuffer.getLong();
int size = byteBuffer.getInt();
if (phyOffset < minPhysicOffset) {
low = midOffset + CQ_STORE_UNIT_SIZE;
leftOffset = midOffset;
continue;
}
//获取指定偏移量的时间戳
long storeTime =
this.defaultMessageStore.getCommitLog().pickupStoreTimestamp(phyOffset, size);
if (storeTime < 0) {
return 0;
} else if (storeTime == timestamp) {
targetOffset = midOffset;
break;
} else if (storeTime > timestamp) {
high = midOffset - CQ_STORE_UNIT_SIZE;
rightOffset = midOffset;
rightIndexValue = storeTime;
} else {
low = midOffset + CQ_STORE_UNIT_SIZE;
leftOffset = midOffset;
leftIndexValue = storeTime;
}
}
if (targetOffset != -1) {
offset = targetOffset;
} else {
if (leftIndexValue == -1) {
offset = rightOffset;
} else if (rightIndexValue == -1) {
offset = leftOffset;
} else {
offset =
Math.abs(timestamp - leftIndexValue) > Math.abs(timestamp
- rightIndexValue) ? rightOffset : leftOffset;
}
}
return (mappedFile.getFileFromOffset() + offset) / CQ_STORE_UNIT_SIZE;
} finally {
sbr.release();
}
}
}
return 0;
}
再来关注下消息存储时存储的额外信息 MessageExt
// 队列ID <PUT>
private int queueId;
//存储消息大小
private int storeSize;
//队列的offset
private long queueOffset;
//消息标志位
private int sysFlag;
//消息在客户端创建时间戳
private long bornTimestamp;
//producer 地址
private SocketAddress bornHost;
//消息在服务器存储时间戳
private long storeTimestamp;
//存储的broker地址
private SocketAddress storeHost;
//消息id
private String msgId;
//消息对应的Commit Log Offset
private long commitLogOffset;
//消息体CRC
private int bodyCRC;
// 当前消息被某个订阅组重新消费了几次（订阅组之间独立计数）
private int reconsumeTimes;Broker 的消息是 at least once还是exactly only once?
at least once:
是指每个消息必须投递一次,RocketMQ Consumer 先 pull 消息到本地，消费完成后，才向服务器返回 ack，如果没有消费一定不会ack消息，所以RocketMQ可以很好的支持此特性。
exactly only once:
(1). 发送消息阶段，不允许发送重复的消息。
(2). 消费消息阶段，不允许消费重复的消息。只有以上两个条件都满足情况下，才能认为消息是“Exactly Only Once”，而要实现以上两点，在分布式系统环境下，不可避免要产生巨大的开销。所以 RocketMQ 为了追求高性能，并不保证此特性，要求在业务上进行去重， 也就是说消费消息要做到幂等性。RocketMQ 虽然不能严格保证不重复，但是正常情况下很少会出现重复发送、消 费情况，只有网络异常，Consumer 启停等异常情况下会出现消息重复。此问题的本质原因是网络调用存在不确定性，即既不成功也不失败的第三种状态，所以才产生了消息重复性问 题。
java
**粗体** _斜体_ [链接](http://example.com) `代码` - 列表 > 引用。你还可以使用@来通知其他用户。