一般在docker启动时,containerd的启动命令如下所示:
root 2090 0.0 0.1 292780 11008 ? Ssl 10月22 0:12 docker-containerd -l unix:///var/run/docker/libcontainerd/docker-containerd.sock --shim docker-containerd-shim
--metrics-interval=0 --start-timeout 2m --state-dir /var/run/docker/libcontainerd/containerd --runtime docker-runc
1、containerd/containerd/main.go
func daemon(context *cli.Context) error
(1)、首先调用:
sv, err := supervisor.New( context.String("state-dir"), context.String("runtime"), context.String("shim"), context.String("runtime-args"), context.String("start-timeout"), context.Int("retain-count"), )
(2)、for循环10次,调用w := supervisor.NewWorker(sv, wg),再go w.Start()
(3)、调用sv.Start(),启动supervisor
(4)、调用server, err := startServer(listenParts[0], listenParts[1], sv),启动grpc server
supervisor的数据结构定义如下所示:
// Supervisor represents a container supervisor type Supervisor struct { // stateDir is the directory on the system to store container runtime state information stateDir string // name of the OCI compatible runtime used to execute containers runtime string runtimeArgs []string shim string containers map[string]*containerInfo startTasks chan *startTask // we need a lock around the subscribers map only because addtions and deletions from // the map via the API so we cannot really control the currency subscriberLock sync.RWMutex subscribers map[chan Event]struct{} machine Machine tasks chan Task monitor *Monitor eventLog []Event eventLock sync.Mutex timeout time.Duration }
2、containerd/supervisor/supervisor.go
// New returns an initialized Process supervisor
func New(stateDir string, runtimeName, shimName string, runtimeArgs []string, timeout time.Duration, retainCount int) (*Supervisor, error)
(1)、调用machine, err := CollectionMachineInformation(),获取当前宿主机的CPU数和RAM总量
(2)、调用monitor, err := NewMonitor(),启动并返回一个监视器
(3)、填充数据结构Supervisor:
s := &Supervisor{ stateDir: stateDir, containers: make(map[string]*ContainerInfo), startTasks: startTasks, machine: machine, subscriber: make(map[chan Event]struct{}), tasks: make(chan Task, defaultBufferSize), monitor: monitor, runtime: runtimeName, runtimeArgs: runtimeArgs, shim: shimName, timeout: timeout, }
(4)、调用setupEventLog(s, retainCount)设置event log
(5)、生成两个goroutine,s.exitHandler()和s.oomHandler()
(6)、最后,调用s.restore(),加载之前已经存在的容器
3、containerd/supervisor/supervisor.go
func (s *Supvervisor) restore() error
(1)、遍历目录s.stateDir(其实就是/var/run/docker/libcontainerd/containerd)
(2)、调用id := d.Name()获取容器id,再调用container, err := runtime.Load(s.stateDir, id, s.shim, s.timeout),load的作用就是加载s.stateDir/id/state.json获取容器实例。之后,再遍历s.stateDir/id/下的pid 文件,加载容器中的process。
(3)、调用processes, err := container.Processes(),加载容器中的process,如果process的状态为running,则调用s.monitorProcess(p)对其进行监控,并对其中不在运行的process进行处理。
4、containerd/supervisor/supervisor.go
// Start is a non-blocking call that runs the supervisor for monitoring container processes and executing new containers
// This event loop is the only thing that is allowed to modify state of containers and processes, therefore it is save to do operations
// in the handlers that modify state of the system or state of the Supervisor
func (s *Supervisor) Start() error
该函数所做的工作很简单,就是启动一个goroutine,再for i := range s.tasks,调用s.handlerTask(i)
Task的数据结构如下所示:
// Task executes an action returning an error chan with either nil or the error from excuting the task type Task interface { // ErrorCh returns a channel used to report and error from an async task ErrorCh() chan error }
5、containerd/supervisor/supervisor.go
func (s *Supervisor) handleTask(i Task)
该函数根据i的类型,调用相应的处理函数进行处理。例如,i.(type)为*StartTask时,则调用s.start(t),若i.(type)为*DeleteTask时,则调用s.delete(t)。
----------------------------------------------------------------------- worker的工作 -------------------------------------------------------------------------
worker的数据结构如下所示:
type Work interface { Start() } type worker struct { wg *sync.WaitGroup s *Supervisor }
4、containerd/supervisor/worker.go
func NewWorker(s *Supervisor, wg *sync.WaitGroup) Worker
这个函数只是简单地填充数据结构,return &worker{s: s, wg: wg}
5、containerd/supervisor/worker.go
// Start runs a loop in charge of starting new containers
func (w *worker) Start()
(1)、遍历w.s.startTasks,调用process, err := t.container.Start(t.checkPointPath, runtime.NewStdio(t.Stdin, t.Stdout, t.Stderr))
(2)、调用w.s.monitor.MonitorOOM(t.Container)和w.s.monitorProcess(process)对container和process进行监控
(3)、当我们从checkpoint restore一个容器的时候,不需要start process。因此,在t.CheckpointPath == ""的时候,调用process.Start()
(4)、调用ContainerStartTimer.UpdateSince(started),started是当前的时间
(5)、最后,调用t.Err <- nil, t.StartResponse <- StartResponse{Container: t.Container},和w.s.notifySubscribers(Event{Timestamp: time.Now, ID: t.container.ID(), Type: StateStart}),进行消息通知
---------------------------------------------------------------------------- monitor 分析 -----------------------------------------------------------------------
Monitor的数据结构如下所示:
// Monitor represents a runtime.Process monitor type Monitor struct { m sync.Mutext receivers map[int]interface{} exits chan runtime.Process ooms chan string epollFd int }
1、containerd/supervisor/monitor_linux.go
// NewMonitor starts a new process monitor and returns it
(1)、首先获取一个monitor实例,m := &Monitor{receivers: make(map[int]interface{}), exits: make(chan runtime.Process, 1024), oom: make(chan string, 1024)}
(2)、调用fd, err := archutils.EpollCreate1(0),创建一个epoll fd,接着将fd赋值给m.epollFd
(3)、生成一个goroutine,go m.start()
2、containerd/supervisor/monitor_linux.go
func (m *Monitor) start()
(1)、该函数就是对各种syscall.EpollEvent进行处理,每次通过调用n, err := archutils.EpollWait(m.epollFd, events[:], -1),获取n个EpollEvent。
(2)、再通过fd := int(events[i].Fd),r := m.receivers[fd]找到对应的runtimeProcess或者runtime.OOM。
(3)、最后,t := r.(type),再分别对runtime.Process和runtime.OOM进行处理
3、containerd/supervisor/monitor_linux.go
// Monitor adds a process to the list of the one being monitored
func (m *Monitor) Monitor(p runtime.Process) error
(1)、调用fd := p.ExitFD() ---> ExitFD returns the fd of the exit pipe,再根据fd新建一个event := syscall.EpollEvent{Fd: int32(fd), Events: syscall.EPOLLHUP,}
(2)、调用archutils.EpollCtl(m.epollFd, syscall.EPOLL_CTL_ADD, fd, &event)
(3)、最后,调用EpollFdCounter.Inc(1),m.receivers[fd] = p