Docker Swarm代码分析笔记（16）——Node结构体

Docker Swarm的scheduler会选择符合要求的node来创建container：

candidates, err := s.selectNodesForContainer(nodes, config, true)

node定义在scheduler/node/node.go：

// Node is an abstract type used by the scheduler.
type Node struct {
    ID         string
    IP         string
    Addr       string
    Name       string
    Labels     map[string]string
    Containers cluster.Containers
    Images     []*cluster.Image

    UsedMemory  int64
    UsedCpus    int64
    TotalMemory int64
    TotalCpus   int64

    HealthIndicator int64
}

Cluster.listNodes方法实现如下：

// listNodes returns all validated engines in the cluster, excluding pendingEngines.
func (c *Cluster) listNodes() []*node.Node {
    c.RLock()
    defer c.RUnlock()

    out := make([]*node.Node, 0, len(c.engines))
    for _, e := range c.engines {
        node := node.NewNode(e)
        for _, pc := range c.pendingContainers {
            if pc.Engine.ID == e.ID && node.Container(pc.Config.SwarmID()) == nil {
                node.AddContainer(pc.ToContainer())
            }
        }
        out = append(out, node)
    }

    return out
}

其实就是从Cluster.engines构建node列表（因为Cluster.pendingEngines还处在待定状态）。后续scheduler就会从这个node列表中选择合适的node。