containerd如何监控容器oom,重启次数

demo

func handle(e interface{}) error {
    switch ev := e.(type) {
    case *events.TaskOOM:
        ...
        }
}

func main(){
    client, err := containerd.New(defaults.DefaultAddress)
    if err != nil {
        panic(err)
    }
    ctx := namespaces.WithNamespace(context.TODO(), "k8s.io")

    evts, errs := client.EventService().Subscribe(ctx)
    for {
        select {
        case err = <-errs:
            panic(err)
        case evt := <-evts:
            if evt.Event != nil {
                v, err := typeurl.UnmarshalAny(evt.Event)
                if err != nil {
                    panic(err)
                }

                handle(v)
            }
        }
    }
}

相关源码

依赖于containerd的events service

services/events/service.go中

func (s *service) Subscribe(req *api.SubscribeRequest, srv api.Events_SubscribeServer) error {
...
    eventq, errq := s.events.Subscribe(ctx, req.Filters...)
...
}

events/exchange/exchange.go中

func (e *Exchange) Subscribe(ctx context.Context, fs ...string) (ch <-chan *events.Envelope, errs <-chan error) {
...
    go func() {
        defer closeAll()

        var err error
    loop:
        for {
            select {
            case ev := <-channel.C:
                env, ok := ev.(*events.Envelope)
                if !ok {
                    // TODO(stevvooe): For the most part, we are well protected
                    // from this condition. Both Forward and Publish protect
                    // from this.
                    err = fmt.Errorf("invalid envelope encountered %#v; please file a bug", ev)
                    break
                }

                select {
                case evch <- env:
                case <-ctx.Done():
                    break loop
                }
            case <-ctx.Done():
                break loop
            }
        }

        if err == nil {
            if cerr := ctx.Err(); cerr != context.Canceled {
                err = cerr
            }
        }

        errq <- err
    }()
...
}

runtime/events.go中

topic有如下
const (
    // TaskCreateEventTopic for task create
    TaskCreateEventTopic = "/tasks/create"
    // TaskStartEventTopic for task start
    TaskStartEventTopic = "/tasks/start"
    // TaskOOMEventTopic for task oom
    TaskOOMEventTopic = "/tasks/oom"
    // TaskExitEventTopic for task exit
    TaskExitEventTopic = "/tasks/exit"
    // TaskDeleteEventTopic for task delete
    TaskDeleteEventTopic = "/tasks/delete"
    // TaskExecAddedEventTopic for task exec create
    TaskExecAddedEventTopic = "/tasks/exec-added"
    // TaskExecStartedEventTopic for task exec start
    TaskExecStartedEventTopic = "/tasks/exec-started"
    // TaskPausedEventTopic for task pause
    TaskPausedEventTopic = "/tasks/paused"
    // TaskResumedEventTopic for task resume
    TaskResumedEventTopic = "/tasks/resumed"
    // TaskCheckpointedEventTopic for task checkpoint
    TaskCheckpointedEventTopic = "/tasks/checkpointed"
    // TaskUnknownTopic for unknown task events
    TaskUnknownTopic = "/tasks/?"
)

topic对应结构
func GetTopic(e interface{}) string {
    switch e.(type) {
    case *events.TaskCreate:
        return TaskCreateEventTopic
    case *events.TaskStart:
        return TaskStartEventTopic
    case *events.TaskOOM:
        return TaskOOMEventTopic
    case *events.TaskExit:
        return TaskExitEventTopic
    case *events.TaskDelete:
        return TaskDeleteEventTopic
    case *events.TaskExecAdded:
        return TaskExecAddedEventTopic
    case *events.TaskExecStarted:
        return TaskExecStartedEventTopic
    case *events.TaskPaused:
        return TaskPausedEventTopic
    case *events.TaskResumed:
        return TaskResumedEventTopic
    case *events.TaskCheckpointed:
        return TaskCheckpointedEventTopic
    default:
        log.L.Warnf("no topic for type %#v", e)
    }
    return TaskUnknownTopic
}

开源实现

https://github.com/draganm/missing-container-metrics