nsq源码分析（3）：nsqd之数据持久化2

nsq源码分析（3）：nsqd之数据持久化2

当nsqd进程退出时，将内存中的数据写入到磁盘
当nsqd进程启动时，将磁盘中的数据读入到内存

nsqd数据信息结构体

nsqd/nsqd.go

type meta struct {
    Topics []struct {
        Name     string `json:"name"`   // topic名称
        Paused   bool   `json:"paused"` // topic状态
        Channels []struct {
            Name   string `json:"name"`   // channel名称
            Paused bool   `json:"paused"` // channel状态
        } `json:"channels"`
    } `json:"topics"`
}

nsqd退出时写入本地元数据

nsqd/nsqd.go

func (n *NSQD) Exit() {
    ...
    n.Lock()
    err := n.PersistMetadata()
    if err != nil {
        n.logf("ERROR: failed to persist metadata - %s", err)
    }
    n.logf("NSQ: closing topics")
    for _, topic := range n.topicMap {
        topic.Close()
    }
    n.Unlock()
    ...
}

nsqd/nsqd.go

// 将topic和channel的信息写入磁盘
func (n *NSQD) PersistMetadata() error {
    // persist metadata about what topics/channels we have, across restarts
    // fileName：nsqd数据文件名。保存topic和channels的相关信息
    // 根据用户传参data-path指定数据目录，文件命名为nsqd.dat
    fileName := newMetadataFile(n.getOpts())
    // old metadata filename with ID, maintained in parallel to enable roll-back
    // fileNameID：id json数据文件名称。
    // fileNameID文件是fileName文件的软连接，内容和fileName一样，用于回滚操作。文件命名为nsqd.867.dat
    // nsqd.867.dat这里的867是opts.ID，根据主机名的哈希值算出的0-1024之间的值
    fileNameID := oldMetadataFile(n.getOpts())

    n.logf("NSQ: persisting topic/channel metadata to %s", fileName)

    // 遍历所有的topic和channel。注：在调用PersistMetadata之前就加上了lock锁
    // 拿到所有topic和channel的name和paused字段，paused描述该topic或channel的状态
    js := make(map[string]interface{})
    topics := []interface{}{}
    for _, topic := range n.topicMap {
        if topic.ephemeral {
            continue
        }
        topicData := make(map[string]interface{})
        topicData["name"] = topic.name
        topicData["paused"] = topic.IsPaused()
        channels := []interface{}{}
        topic.Lock()
        for _, channel := range topic.channelMap {
            channel.Lock()
            if channel.ephemeral {
                channel.Unlock()
                continue
            }
            channelData := make(map[string]interface{})
            channelData["name"] = channel.name
            channelData["paused"] = channel.IsPaused()
            channels = append(channels, channelData)
            channel.Unlock()
        }
        topic.Unlock()
        topicData["channels"] = channels
        topics = append(topics, topicData)
    }
    js["version"] = version.Binary
    js["topics"] = topics

    // 序列化成json字符串
    data, err := json.Marshal(&js)
    if err != nil {
        return err
    }

    // 写入nsqd数据文件（nsqd.dat）
    tmpFileName := fmt.Sprintf("%s.%d.tmp", fileName, rand.Int())

    err = writeSyncFile(tmpFileName, data)
    if err != nil {
        return err
    }
    err = os.Rename(tmpFileName, fileName)
    if err != nil {
        return err
    }
    // technically should fsync DataPath here

    stat, err := os.Lstat(fileNameID)
    if err == nil && stat.Mode()&os.ModeSymlink != 0 {
        return nil
    }

    // if no symlink (yet), race condition:
    // crash right here may cause next startup to see metadata conflict and abort

    // 创建nsqd数据文件的软连接（nsqd.867.dat）
    tmpFileNameID := fmt.Sprintf("%s.%d.tmp", fileNameID, rand.Int())

    if runtime.GOOS != "windows" {
        err = os.Symlink(fileName, tmpFileNameID)
    } else {
        // on Windows need Administrator privs to Symlink
        // instead write copy every time
        err = writeSyncFile(tmpFileNameID, data)
    }
    if err != nil {
        return err
    }

    err = os.Rename(tmpFileNameID, fileNameID)
    if err != nil {
        return err
    }
    // technically should fsync DataPath here

    return nil
}

nsqd启动时加载本地元数据

apps/nsqd/nsqd.go

func (p *program) Start() error {
    ...
    // 加载元数据
    err := nsqd.LoadMetadata()
    if err != nil {
        log.Fatalf("ERROR: %s", err.Error())
    }
    ...
}

nsqd/nsqd.go

// 加载磁盘中的topic和channel信息
func (n *NSQD) LoadMetadata() error {
    atomic.StoreInt32(&n.isLoading, 1)
    defer atomic.StoreInt32(&n.isLoading, 0)

    fn := newMetadataFile(n.getOpts())
    // old metadata filename with ID, maintained in parallel to enable roll-back
    fnID := oldMetadataFile(n.getOpts())

    // 读取nsqd数据文件和nsqd数据文件的软连接文件（用于回滚操作）
    data, err := readOrEmpty(fn)
    if err != nil {
        return err
    }
    dataID, errID := readOrEmpty(fnID)
    if errID != nil {
        return errID
    }

    if data == nil && dataID == nil {
        return nil // fresh start
    }
    if data != nil && dataID != nil {
        if bytes.Compare(data, dataID) != 0 {
            return fmt.Errorf("metadata in %s and %s do not match (delete one)", fn, fnID)
        }
    }
    if data == nil {
        // only old metadata file exists, use it
        fn = fnID
        data = dataID
    }

    var m meta
    err = json.Unmarshal(data, &m)
    if err != nil {
        return fmt.Errorf("failed to parse metadata in %s - %s", fn, err)
    }

    // 将nsqd数据文件的json反序列化到Topic和Channels
    for _, t := range m.Topics {
        if !protocol.IsValidTopicName(t.Name) {
            n.logf("WARNING: skipping creation of invalid topic %s", t.Name)
            continue
        }
        topic := n.GetTopic(t.Name)
        if t.Paused {
            topic.Pause()
        }

        for _, c := range t.Channels {
            if !protocol.IsValidChannelName(c.Name) {
                n.logf("WARNING: skipping creation of invalid channel %s", c.Name)
                continue
            }
            channel := topic.GetChannel(c.Name)
            if c.Paused {
                channel.Pause()
            }
        }
    }
    return nil
}