Add support for adjusting OOM score adjustment.

[zxj874478410]

The OomScoreAdj field is added to the LinuxContainerAdjustment structure of the NRI plugin. This field records the value of oom_score_adj that needs to be adjusted during CreateContainer. Added the appropriate set method and the method to write back to the NRI response.
Fix #92

TEST METHOD：
We build a NRI plugin with CreateContainer function like this:

func (p *plugin) CreateContainer(_ context.Context, pod *api.PodSandbox, ctr *api.Container) (*api.ContainerAdjustment, []*api.ContainerUpdate, error) {
	log.Infof("Creating container %s/%s/%s...", pod.GetNamespace(), pod.GetName(), ctr.GetName())

	//
	// This is the container creation request handler. Because the container
	// has not been created yet, this is the lifecycle event which allows you
	// the largest set of changes to the container's configuration, including
	// some of the later immautable parameters. Take a look at the adjustment
	// functions in pkg/api/adjustment.go to see the available controls.
	//
	// In addition to reconfiguring the container being created, you are also
	// allowed to update other existing containers. Take a look at the update
	// functions in pkg/api/update.go to see the available controls.
	//

	adjustment := &api.ContainerAdjustment{}

	adjustment.SetLinuxOomScoreAdj(121)
	log.Infof("set oom score: %v", 121)
	return adjustment, nil, nil
}

Then run this NRI plugin with containerd. Create a k8s pod of apache.

[root@master bin]# kubectl get po -A |grep apache
kube-system    apache-66f99dd558-c9bgm          1/1     Running   0               4h5m

Run the docker inspect command to query the process ID of the Apache container is 3445531.

At last, query the oom_score_adj of this pid. It has been set to 121 successfully.

[root@master bin]# cat /proc/3445531/oom_score_adj
121

[klihub]

Thank's for the PR @zxj874478410 ! I have a few nits, but otherwise it looks good to me.

[zxj874478410]

Thank's for the PR @zxj874478410 ! I have a few nits, but otherwise it looks good to me.

Thanks for the review @klihub . I've fixed these faults.

[chenzhiwei]

lgtm

[mikebrow]

wondering if we should be able to nil out the oooscoreadj

[zxj874478410]

Thanks for the suggestion! After we set OomScoreAdj to nil, We can handle the nil here.
I try to call the func g.initConfigProcess() by calling func g.SetProcessOOMScoreAdj(0), and then set g.Config.Process.OOMScoreAdj to nil. Will it work? @mikebrow

Suggested change

	}
	} else {
	g.SetProcessOOMScoreAdj(0)
	g.Config.Process.OOMScoreAdj = nil
	}

[mikebrow]

sgtm let's try it / test it out

[zxj874478410]

We tested it with the following code

oomScoreAdj:= 555
adjustment.SetLinuxOomScoreAdj(&oomScoreAdj)
adjustment.SetLinuxOomScoreAdj(nil)

The second set unset the oomscoreadj with the api. Finally, the oom_score_adj is set to 996 by kubelet.

cat /proc/2213925/oom_score_adj
996

This is work! @mikebrow

[samuelkarp]

This code path appears to be the source of #117. @mikebrow, do we need to support a plugin both setting an explicit adjustment and then clearing its own adjustment?

[mikebrow]

nod.. while I think it's a valid path to unset given 0 / 999 etc.. is not unset.. we need to ensure we are not accidentally unsetting a valid oom from kubelet. @klihub FYI

[mikebrow]

manual:

           Sets the adjustment value for the Linux kernel's
           Out-Of-Memory (OOM) killer score for executed processes.
           Takes an integer between -1000 (to disable OOM killing of
           processes of this unit) and 1000 (to make killing of
           processes of this unit under memory pressure very likely).
           See The /proc Filesystem[6] for details. If not specified
           defaults to the OOM score adjustment level of the service
           manager itself, which is normally at 0.

note behavior normally 0 when unset, but is configurable making unset a unique state

[mikebrow]

LGTM

[zxj874478410]

@klihub @mikebrow Just rebased code and fixed conflicts, please take a look, thanks.

[mikebrow]

LGTM

[zxj874478410]

LGTM

[zxj874478410]

@klihub @mikebrow Sorry for conflict again. Just fixed conflicts, please take a look, thanks.

[klihub]

@zxj874478410 This fix is not sufficient in it's current form as it breaks replacing a mount with another one to the same destination. I think we'd need a bit of additional code to prevent that from happening. Maybe something like this, where 8f509e7 is an updated version of your commit, and 7ea0a47 adds a new test case for mount removal.

@zxj874478410 PTAL at the latest review comments and the suggested fixes/improvements.

[zxj874478410]

@zxj874478410 This fix is not sufficient in it's current form as it breaks replacing a mount with another one to the same destination. I think we'd need a bit of additional code to prevent that from happening. Maybe something like this, where 8f509e7 is an updated version of your commit, and 7ea0a47 adds a new test case for mount removal.

Thanks for the detailed instructions, but does this problem refer to another pr #87 ? This one refers to OOM score adjustment.

[klihub]

@zxj874478410 This fix is not sufficient in it's current form as it breaks replacing a mount with another one to the same destination. I think we'd need a bit of additional code to prevent that from happening. Maybe something like this, where 8f509e7 is an updated version of your commit, and 7ea0a47 adds a new test case for mount removal.

Thanks for the detailed instructions, but does this problem refer to another pr #87 ? This one refers to OOM score adjustment.

Argh... Sorry about that. Yes, you are absolutely right, it is about PR #87. Time to renew my glasses or get a brain transplant...

[mikebrow]

LGTM