runv start container 流程分析

1、runv/start.go

func startContainer(context *cli.Context, container, address string, config *spec.Spec)

该函数所做的工作很简单,首先构建CreateContainerRequest的grpc请求,如下所示:

r  := &types.CreateContainerRequest {

  Id:       container,
  BundlePath:     context.String("bundle"),
  Stdin:     fmt.Sprintf("/proc/%d/fd/0", pid),
  Stdout:      fmt.Sprintf("/proc/%d/fd/1", pid),
  Stderr:    fmt.Sprintf("/proc/%d/fd/2", pid),
}

  

再通过调用c := getClient(address)获取grpc client,再通过调用c.CreateContainer(netcontext.Background(), r)将创建容器的请求发送给grpc server。

 

2、runv/containerd/api/grpc/server/server.go

func (s *apiServer) CreateContainer(ctx context.Context, r *types.CreateContainerRequest) (*types.CreateContainerResponse, error)

(1)、调用ocfData, err := ioutil.ReadFile(filepath.Join(r.BundlePath, "config.json"))和json.Unmarshal(ocfData, &spec)加载spec

(2)、调用c, p, err := s.sv.CreateContainer(r.Id, r.BundlePath, r.Stdin, r.Stdout, r.Stderr, &spec)

(3)、最后调用apiP := supervisorProcess2ApiProcess(p), apiC := supervisorContainer2ApiContainer(c), addApiProcess2ApiContainer(apiC, apiP)完成api的转换,并且将进程apiP添加到apiC中

 

3、runv/supervisor/supervisor.go

func (sv *Supervisor) CreateContainer(container, bundlePath, stdin, stdout, stderr string, spec *specs.Spec) (*Container, *Process, error)

(1)、调用hp, err := sv.getHyperPod(container, spec),获取一个hyper pod 实例

(2)、c, err := hp.createContainer(container, bundlePath, stdin, stdout, stderr, spec) 创建一个hyper container实例

(3)、sv.Containers[container] = c, 再 return c, c.Processes["init"], nil

 

HyperPod的结构如下所示:

type HyperPod struct {

  Containers  map[string]*Container
  Processes  map[string]*Process
  userPod    *pod.UserPod
  podStatus  *hypervisor.PodStatus
  vm     *hypervisor.Vm
  sv      *supervisor
  nslistener  *nsListener
}

  

4、runv/supervisor/supervisor.go

// find shared pod or create a new one

func (sv *Supervisor) getHyperPod(container string, spec *specs.Spec) (hp *HyperPod, err error)

(1)、当spec.Linux.Namespaces不为空时,先进行相关的处理,这里暂时不讨论。

(2)、当hp为nil时,先调用hp, err = createHyperPod(sv.Factory, spec, sv.defaultCpus, sv.defaultMemory),再调用hp.sv = sv将hypervisor和supervisor串联起来

(3)、return hp, nil

 

5、runv/supervisor/hyperpod.go

func createHyperPod(f factory, spec *specs.Spec, defaultCpus int, defaultMemory int) (*HyperPod, error)

(1)、先获取一个podId,再调用userPod := pod.ConvertOCF2PureUserPod(spec)(并没有container的信息)和podStatus := hypervisor.NewPod(podId, userPod, nil)

(2)、获取cpu, mem, kernel, initrd的信息,再启动虚拟机,虽然根据获得的kernel和initrd情况略有不同,最终调用都是调用hypervisor.GetVm("", boot, bool, bool),启动虚拟机

(3)、调用Response := vm.StartPod(podStatus, userPod, nil, nil)

(4)、填充HyperPod, hp := &HyperPod {

              userPod:  userPod,

              podStatus: podStatus,

              vm:    vm,

              Containers: make(map[string]*Container),

              Processes: make(map[string]*Process)

            }

(5)、调用hp.startNsListener(), 返回 return hp, nil


6、runv/supervisor/hyperpod.go

func (hp *HyperPod) createContainer(container, bundlePath, stdin, stdout, stderr string, spec *specs.Spec) (*Container, error)

(1)、inerProcessId := container + "-init",并且填充数据结构获得 c := &Container{...},p := &Process{...}

(2)、调用p.setupIO()

(3)、对各个数据结构进行关联,c.Processes["init"] = p, c.ownerPod.Processes[inerProcessId] = p, c.ownerPod.Containers[container] = c

(4)、c.run(p),return c, nil

 

7、runv/supervisor/container.go

func (c *Container) run(p *Process)

(1)、该函数仅仅启动一个goroutine,首先调用c.start(p)启动容器

(2)、创建e := Event{ID: c.Id, Type: EventContainerStart, Timestamp: time.Now(),},调用c.ownerPod.sv.Events.notifySubscribers(e)将event发出

(3)、调用c.wait(p)等待容器退出

(4)、创建e := Event{ID: c.Id, EventExit, Timestamp: time.Now(), PID: p.Id, Status: -1},若正常退出,则将Status设置为容器的退出码

 

8、runv/supervisor/container.go

func (c *Container) start(p *Process) error

(1)、创建目录/run/runv/container-id,再在其中生成文件state.go,将state := &specs.State{Version: c.Spec.Version, ID: c.Id, Pid: c.ownerPod.getNsPid(), BundlePath: c.BundlePath}写入

(2)、先调用u := pod.ConvertOCF2UserContainer(c.Spec)将spec转换为UserPod,镜像目录为u.Image,镜像的挂载目录为vmRootfs := filepath.Join(hypervisor.BaseDir, c.ownerPod.vm.Id, hypervisor.ShareDirTag, c.Id, "rootfs")(和hyperd兼容),再调用utils.Mount(u.Image, vmRootfs, c.Spec.Root.Readonly)将镜像挂载到vmRootfs。如果c.Spec.Mounts不为空的话,调用mountToRootfs(&m, vmRootfs, ""),挂载相应的目录或文件

(3)、调用info := &hypervisor.ContainerInfo{Id: c.Id, Rootfs: "rootfs", Image: pod.userVolume{Source: c.Id}, Fstype: "dir", Cmd: u.Command, Envs: envs,}

(4)、调用c.ownerPod.vm.Attach(p.stdio, c.Id, nil)连接vm, 再调用execPrestartHooks(c.Spec, state),再调用c.ownerPod.initPodNetwork(c)初始化网络

(5)、调用c.ownerPod.podStatus.AddContainer(c.Id, c.ownerPod.podStatus.Id, "", []string{}, types.S_POD_CREATED) ---> 仅仅只是将ContainerStatus添加到podStatus.Containers中

(6)、return c.ownerPod.vm.NewContainer(u, info) ---> 真正在虚拟机中完成container的创建

 

posted on 2016-10-21 17:08  姚灯灯!  阅读(922)  评论(0编辑  收藏  举报

导航