日期：2016年6月20日

Engine.Connect()方法用来完成对Docker engine的连接工作：

// Connect will initialize a connection to the Docker daemon running on the
// host, gather machine specs (memory, cpu, ...) and monitor state changes.
func (e *Engine) Connect(config *tls.Config) error {
    host, _, err := net.SplitHostPort(e.Addr)
    if err != nil {
        return err
    }

    addr, err := net.ResolveIPAddr("ip4", host)
    if err != nil {
        return err
    }
    e.IP = addr.IP.String()

    c, err := dockerclient.NewDockerClientTimeout("tcp://"+e.Addr, config, time.Duration(requestTimeout), setTCPUserTimeout)
    if err != nil {
        return err
    }
    // Use HTTP Client used by dockerclient to create engine-api client
    apiClient, err := engineapi.NewClient("tcp://"+e.Addr, "", c.HTTPClient, nil)
    if err != nil {
        return err
    }

    return e.ConnectWithClient(c, apiClient)
}

Engine.Connect()利用了另外两个Docker项目：dockerclient和engine-api。首先调用dockerclient.NewDockerClientTimeout("tcp://"+e.Addr, config, time.Duration(requestTimeout), setTCPUserTimeout)生成一个DockerClient结构体：

type DockerClient struct {
    URL           *url.URL
    HTTPClient    *http.Client
    TLSConfig     *tls.Config
    monitorStats  int32
    eventStopChan chan (struct{})
}

再把DockerClient.HTTPClient传入engineapi.NewClient()函数，作为engine-api的http.Client，作为HTTP请求的传输载体。最后调用Engine.ConnectWithClient()连接Docker engine，获得Docker信息。

Cluster.validatePendingEngine是实际上用来连接Docker engine的代码：

// validatePendingEngine connects to the engine,
func (c *Cluster) validatePendingEngine(engine *cluster.Engine) bool {
    // Attempt a connection to the engine. Since this is slow, don't get a hold
    // of the lock yet.
    if err := engine.Connect(c.TLSConfig); err != nil {
        log.WithFields(log.Fields{"Addr": engine.Addr}).Debugf("Failed to validate pending node: %s", err)
        return false
    }

    // The following is critical and fast. Grab a lock.
    c.Lock()
    defer c.Unlock()

    // Only validate engines from pendingEngines list
    if _, exists := c.pendingEngines[engine.Addr]; !exists {
        return false
    }

    // Make sure the engine ID is unique.
    if old, exists := c.engines[engine.ID]; exists {
        if old.Addr != engine.Addr {
            log.Errorf("ID duplicated. %s shared by %s and %s", engine.ID, old.Addr, engine.Addr)
            // Keep this engine in pendingEngines table and show its error.
            // If it's ID duplication from VM clone, user see this message and can fix it.
            // If the engine is rebooted and get new IP from DHCP, previous address will be removed
            // from discovery after a while.
            // In both cases, retry may fix the problem.
            engine.HandleIDConflict(old.Addr)
        } else {
            log.Debugf("node %q (name: %q) with address %q is already registered", engine.ID, engine.Name, engine.Addr)
            engine.Disconnect()
            // Remove it from pendingEngines table
            delete(c.pendingEngines, engine.Addr)
        }
        return false
    }

    // Engine validated, move from pendingEngines table to engines table
    delete(c.pendingEngines, engine.Addr)
    // set engine state to healthy, and start refresh loop
    engine.ValidationComplete()
    c.engines[engine.ID] = engine

    log.Infof("Registered Engine %s at %s", engine.Name, engine.Addr)
    return true
}

Cluster.validatePendingEngine永远操作Cluster.pendingEngines里的Engine，如果连接Docker engine成功，就会把这个Engine移到Cluster.engines这个map里。