Enable force kill container with SIGKILL if timeout

[Crazykev]

ref: #549

• I use FutureSet workflow to re-implement stopContainers() to enable force kill container after timeout, change the way it works, original graceful guarantee stopContianers() will return in graceful time, now ensure container have graceful time to stop with default signal before forcibly kill it.
• remove cMap and eMap, since we don't need it in current approach(Am I correct?)
• Also add timeout in StopContainer grpc api since CRI need it(/cc @feiskyer )

[gnawux]

IMHO, if graceful == 0, we should not wait, and fire force kill directly.

[Crazykev]

yes, if graceful == 0, it is force kill directly:

...
forceKill := graceful == 0
resChan := p.sandbox.WaitProcess(true, []string{c.Id()}, -1)
err := c.terminate(forceKill)
...

This waitTime is the time we wait for container finish event before abort this mission(or try it with SIGKILL in non-zero graceful)

[gnawux]

Then, do we need wait 5 seconds and kill again?

[Crazykev]

If graceful == 0, it just wait 5s and won't kill again(return timeout error); if kill with image specified signal(graceful != 0), it will wait graceful time to kill with SIGKILL again.
Is this 5 seconds too long?

[gnawux]

If graceful == 0, does this mean the user don't want to wait anything, just return at once?

[Crazykev]

Nop if we use docker:

$ docker stop --help

Usage:	docker stop [OPTIONS] CONTAINER [CONTAINER...]

Stop a running container.
Sending SIGTERM and then SIGKILL after a grace period

  --help             Print usage
  -t, --time=10      Seconds to wait for stop before killing it

or k8s way:
https://github.com/kubernetes/kubernetes/blob/master/pkg/kubelet/api/v1alpha1/runtime/api.proto#L653-L658

[gnawux]

@laijs could you review this patch? Its size is far beyond what I imagined before, but looks good except for the grace==0 case.

[laijs]

LGTM