docker health check low performance

[supermeng]

in containerd 0.2.x,
docker exec new process need 70ms as issue moby/moby#33933 said,
and containerd supervisord exec task with single goroutine, so many context cancelled when too many containers in a node need healthcheck.
cost 70*n ms.

[supermeng]

func (s *Supervisor) Start() error {
logrus.WithFields(logrus.Fields{
"stateDir": s.stateDir,
"runtime": s.runtime,
"runtimeArgs": s.runtimeArgs,
"memory": s.machine.Memory,
"cpus": s.machine.Cpus,
}).Debug("containerd: supervisor running")
go func() {
for i := range s.tasks {
s.handleTask(i)
}
}()
return nil
}

func (s *Supervisor) handleTask(i Task) {
var err error
switch t := i.(type) {
case *AddProcessTask:
err = s.addProcess(t)
case *CreateCheckpointTask:
err = s.createCheckpoint(t)
case *DeleteCheckpointTask:
err = s.deleteCheckpoint(t)
case *StartTask:
err = s.start(t)
case *DeleteTask:
err = s.delete(t)
case *ExitTask:
err = s.exit(t)
case *GetContainersTask:
err = s.getContainers(t)
case *SignalTask:
err = s.signal(t)
case *StatsTask:
err = s.stats(t)
case *UpdateTask:
err = s.updateContainer(t)
case *UpdateProcessTask:
err = s.updateProcess(t)
case *OOMTask:
err = s.oom(t)
default:
err = ErrUnknownTask
}
if err != errDeferredResponse {
i.ErrorCh() <- err
close(i.ErrorCh())
}
}