support RuntimeClass.handler, this will useful like nvidia is not the…

[zhangguanzhang]

Fixes #111 . default-runtime is runc, You cannot use RuntimeClass.handler to specify that the pod uses a non-runc runtime.:

# curl -s --unix-socket /var/run/docker.sock http://./v1.42/info | jq .Runtimes
{
  "io.containerd.runc.v2": {
    "path": "runc"
  },
  "nvidia": {
    "path": "nvidia-container-runtime"
  },
  "runc": {
    "path": "runc"
  }
}

before

apiVersion: node.k8s.io/v1
kind: RuntimeClass
metadata:
  name: nvidia
handler: docker
---
apiVersion: v1
kind: Pod
metadata:
....
    spec:
      runtimeClassName: nvidia

RuntimeClass.handler=docker will used runc

docker inspect xx | grep Runtime:
            "Runtime": "runc",

change handler: docker to handler: nvidia will not work, but after this pr:

apiVersion: node.k8s.io/v1
kind: RuntimeClass
metadata:
  name: nvidia
handler: nvidia # <----
---
apiVersion: v1
kind: Pod
metadata:
....
    spec:
      runtimeClassName: nvidia

will handle the RuntimeClass.handler

docker inspect xx | grep Runtime:
            "Runtime": "nvidia",

work with crictl https://github.com/containerd/containerd/blob/main/docs/cri/crictl.md#run-a-pod-sandbox-using-a-config-file

$ cat sandbox-config.json 
{
    "metadata": {
        "name": "nginx-sandbox",
        "namespace": "default",
        "attempt": 1,
        "uid": "hdishd83djaidwnduwk28bcsb"
    },
    "linux": {
    }
}
$ crictl  runp --runtime runc111 sandbox-config.json 
E0420 22:22:04.460269    2703 remote_runtime.go:193] "RunPodSandbox from runtime service failed" err="rpc error: code = Unknown desc = failed to get sandbox runtime: no runtime for \"runc111\" is configured"
FATA[0000] run pod sandbox: rpc error: code = Unknown desc = failed to get sandbox runtime: no runtime for "runc111" is configured
$ crictl  runp --runtime runc sandbox-config.json 
e919f7425bc569ad823da6ffc799bac1318c0b1a1bef7568362ce7f1f10710af

[zhangguanzhang]

@neersighted @corhere PTAL

[corhere]

LGTM, thanks! cc @nwneisen

[nwneisen]

@zhangguanzhang Please take a look at why CI is failing

[zhangguanzhang]

@zhangguanzhang Please take a look at why CI is failing

I found the reason and am modifying it locally. Also, can you help me look at another PR first? #349

[zhangguanzhang]

@zhangguanzhang Please take a look at why CI is failing

https://github.com/containerd/containerd/blob/main/docs/cri/crictl.md#run-a-pod-sandbox-using-a-config-file

I used the crictl create test on containerd with a non-existent sandbox ID and found an error. I thought the CreateContainer should have called the RunPodSandbox before it

$ crictl create fakeID container-config.json sandbox-config.json 
E0420 23:37:56.581510    2865 remote_runtime.go:319] "CreateContainer in sandbox from runtime service failed" err="rpc error: code = NotFound desc = failed to find sandbox id \"aaa\": not found" podSandboxID="fakeID"
FATA[0000] creating container: rpc error: code = NotFound desc = failed to find sandbox id "fakeID": not found

[zhangguanzhang]

@corhere @nevalla
My local code tests are normal and I have added corresponding unit tests. PTAL

[cncal]

This feature is exactly what I want. Greate job. Hope this PR would be merged ASSP.

[nwneisen]

LGTM

[jandubois]

@nwneisen Will this PR be cherry-picked into release/0.3 and included in the next release?

Not pushing for it, I just want to know if it makes sense to wait for a release, or if I should build from master for now.

[neersighted]

This is a new feature that I don't think is appropriate for cherry-pick into an existing branch. We're working on a 0.4 very shortly.

[jandubois]

We're working on a 0.4 very shortly.

Any updates on this?