RocketMQ中broker消息存儲之如何實(shí)現(xiàn)消息轉(zhuǎn)儲

小編給大家分享一下RocketMQ中broker消息存儲之如何實(shí)現(xiàn)消息轉(zhuǎn)儲，希望大家閱讀完這篇文章之后都有所收獲，下面讓我們一起去探討吧！

broker在接收到producer發(fā)送的消息之后，首先會將消息存儲到CommitLog的末尾，然后通過一個(gè)異步的分發(fā)線程ReputMessageService將消息轉(zhuǎn)儲到ConsumeQueue以及IndexFile中。

轉(zhuǎn)儲的核心邏輯在ReputMessageService.doReput中：

    // DefaultMessageStore.ReputMessageService    
    private void doReput() {
            if (this.reputFromOffset < DefaultMessageStore.this.commitLog.getMinOffset()) {
                log.warn("The reputFromOffset={} is smaller than minPyOffset={}, this usually indicate that the dispatch behind too much and the commitlog has expired.",
                    this.reputFromOffset, DefaultMessageStore.this.commitLog.getMinOffset());
                this.reputFromOffset = DefaultMessageStore.this.commitLog.getMinOffset();
            }
            for (boolean doNext = true; this.isCommitLogAvailable() && doNext; ) {

                if (DefaultMessageStore.this.getMessageStoreConfig().isDuplicationEnable()
                    && this.reputFromOffset >= DefaultMessageStore.this.getConfirmOffset()) {
                    break;
                }

                // 1. 獲取reputFromOffset偏移所指向的數(shù)據(jù)
                SelectMappedBufferResult result = DefaultMessageStore.this.commitLog.getData(reputFromOffset);
                if (result != null) {
                    try {
                        this.reputFromOffset = result.getStartOffset();

                        for (int readSize = 0; readSize < result.getSize() && doNext; ) {
                            // 2. 解析消息體
                            DispatchRequest dispatchRequest =
                                DefaultMessageStore.this.commitLog.checkMessageAndReturnSize(result.getByteBuffer(), false, false);
                            int size = dispatchRequest.getBufferSize() == -1 ? dispatchRequest.getMsgSize() : dispatchRequest.getBufferSize();

                            if (dispatchRequest.isSuccess()) {
                                if (size > 0) {
                                    // 3. 執(zhí)行分發(fā)
                                    DefaultMessageStore.this.doDispatch(dispatchRequest);

                                    // ...

                                    this.reputFromOffset += size;
                                    readSize += size;
                                    // ...
                                } else if (size == 0) {
                                    this.reputFromOffset = DefaultMessageStore.this.commitLog.rollNextFile(this.reputFromOffset);
                                    readSize = result.getSize();
                                }
                            } else if (!dispatchRequest.isSuccess()) {

                               // ...
                            }
                        }
                    } finally {
                        result.release();
                    }
                } else {
                    doNext = false;
                }
            }
        }

ConsumeQueue的插入操作如下：

    // ConsumeQueue.java
    private boolean putMessagePositionInfo(final long offset, final int size, final long tagsCode,
        final long cqOffset) {

        if (offset + size <= this.maxPhysicOffset) {
            log.warn("Maybe try to build consume queue repeatedly maxPhysicOffset={} phyOffset={}", maxPhysicOffset, offset);
            return true;
        }

        // 1. 將commitlog offset/msg size/tags code寫到內(nèi)存緩存
        this.byteBufferIndex.flip();
        this.byteBufferIndex.limit(CQ_STORE_UNIT_SIZE);
        this.byteBufferIndex.putLong(offset);
        this.byteBufferIndex.putInt(size);
        this.byteBufferIndex.putLong(tagsCode);

        final long expectLogicOffset = cqOffset * CQ_STORE_UNIT_SIZE; // ConsumeQueue中偏移

        // 2. 獲取最后一個(gè)MappedFile
        MappedFile mappedFile = this.mappedFileQueue.getLastMappedFile(expectLogicOffset);
        if (mappedFile != null) {

            // ...
            this.maxPhysicOffset = offset + size;
            // 3. 寫入索引數(shù)據(jù)
            return mappedFile.appendMessage(this.byteBufferIndex.array());
        }
        return false;
    }

IndexFile的寫入邏輯如下：

    // IndexService.java
    public void buildIndex(DispatchRequest req) {
        IndexFile indexFile = retryGetAndCreateIndexFile();
        if (indexFile != null) {
            long endPhyOffset = indexFile.getEndPhyOffset();
            DispatchRequest msg = req;
            String topic = msg.getTopic();
            String keys = msg.getKeys();
            if (msg.getCommitLogOffset() < endPhyOffset) {
                return;
            }

            // ...
         
            if (keys != null && keys.length() > 0) {
                String[] keyset = keys.split(MessageConst.KEY_SEPARATOR);
                for (int i = 0; i < keyset.length; i++) { // 為每個(gè)key執(zhí)行寫入
                    String key = keyset[i];
                    if (key.length() > 0) {
                        indexFile = putKey(indexFile, msg, buildKey(topic, key));
                        if (indexFile == null) {
                            log.error("putKey error commitlog {} uniqkey {}", req.getCommitLogOffset(), req.getUniqKey());
                            return;
                        }
                    }
                }
            }
        } else {
            log.error("build index error, stop building index");
        }
    }

    private IndexFile putKey(IndexFile indexFile, DispatchRequest msg, String idxKey) {
        for (boolean ok = indexFile.putKey(idxKey, msg.getCommitLogOffset(), msg.getStoreTimestamp()); !ok; ) {
            log.warn("Index file [" + indexFile.getFileName() + "] is full, trying to create another one");

            indexFile = retryGetAndCreateIndexFile(); // 文件已滿，重試
            if (null == indexFile) {
                return null;
            }

            ok = indexFile.putKey(idxKey, msg.getCommitLogOffset(), msg.getStoreTimestamp());
        }

        return indexFile;
    }

消息轉(zhuǎn)儲的整體流程如下圖：

看完了這篇文章，相信你對“RocketMQ中broker消息存儲之如何實(shí)現(xiàn)消息轉(zhuǎn)儲”有了一定的了解，如果想了解更多相關(guān)知識，歡迎關(guān)注億速云行業(yè)資訊頻道，感謝各位的閱讀！