search

k8snetworkplumbingwg / sriov-network-device-plugin

SRIOV network device plugin for Kubernetes
Apache License 2.0
407 stars 177 forks source link

Does virtio support rdma #591

Open BurlyLuo opened 2 months ago

BurlyLuo commented 2 months ago

What happened?

when did at kvm based environment, we can follow below steps to setup the rdma environment

yum install rdma-core perftest -y

sudo modprobe ib_core
sudo modprobe ib_uverbs
sudo modprobe rdma_cm

cat /boot/config-$(uname -r) | grep RXE
#m

modprobe rdma_rxe

rdma link add rxe_0 type rxe netdev eth1

rdma link

A: ib_send_bw -d rxe_0

B: ib_send_bw -d rxe_0 <server_ip>

so that mean we can use virtio device as the rdma nic, rt? this is pcap from testperf. rdma.zip

What did you expect to happen?

use virtio nic to support RDMA

What are the minimal steps needed to reproduce the bug?

[root@rocky92 ~]# kk get cm sriovdp-config -o yaml 
apiVersion: v1
data:
  config.json: |
    {
        "resourceList": [
            {
                "resourcePrefix": "mellanox.com",
                "resourceName": "sriov_vppdpdk_rdma5",
                "selectors":
                {
                    "pciaddresses": ["0000:02:00.0"],
                    "isRdma": true
                }
            },
            {
                "resourcePrefix": "mellanox.com",
                "resourceName": "sriov_vppdpdk_rdma8",
                "selectors":
                {
                    "pciaddresses": ["0000:03:00.0"],
                    "isRdma": true
                }
            }
        ]
    }
kind: ConfigMap

and the device-plugin pod's log show as below:

[root@rocky92 ~]# kk logs kube-sriov-device-plugin-hcwks
I0830 14:28:06.628921       1 manager.go:57] Using Kubelet Plugin Registry Mode
I0830 14:28:06.629778       1 main.go:46] resource manager reading configs
I0830 14:28:06.630716       1 manager.go:86] raw ResourceList: {
    "resourceList": [
        {
            "resourcePrefix": "mellanox.com",
            "resourceName": "sriov_vppdpdk_rdma5",
            "selectors":
            {
                "pciaddresses": ["0000:02:00.0"],
                "isRdma": true
            }
        },
        {
            "resourcePrefix": "mellanox.com",
            "resourceName": "sriov_vppdpdk_rdma8",
            "selectors":
            {
                "pciaddresses": ["0000:03:00.0"],
                "isRdma": true
            }
        }
    ]
}
I0830 14:28:06.630886       1 factory.go:211] *types.NetDeviceSelectors for resource sriov_vppdpdk_rdma5 is [0xc00043fc20]
I0830 14:28:06.630900       1 factory.go:211] *types.NetDeviceSelectors for resource sriov_vppdpdk_rdma8 is [0xc0004d4000]
I0830 14:28:06.630904       1 manager.go:106] unmarshalled ResourceList: [{ResourcePrefix:mellanox.com ResourceName:sriov_vppdpdk_rdma5 DeviceType:netDevice ExcludeTopology:false Selectors:0xc000469188 AdditionalInfo:map[] SelectorObjs:[0xc00043fc20]} {ResourcePrefix:mellanox.com ResourceName:sriov_vppdpdk_rdma8 DeviceType:netDevice ExcludeTopology:false Selectors:0xc0004691a0 AdditionalInfo:map[] SelectorObjs:[0xc0004d4000]}]
I0830 14:28:06.630947       1 manager.go:217] validating resource name "mellanox.com/sriov_vppdpdk_rdma5"
I0830 14:28:06.630997       1 manager.go:217] validating resource name "mellanox.com/sriov_vppdpdk_rdma8"
I0830 14:28:06.631000       1 main.go:62] Discovering host devices
I0830 14:28:06.710775       1 netDeviceProvider.go:67] netdevice AddTargetDevices(): device found: 0000:01:00.0 02              Red Hat, Inc.           Virtio 1.0 network device               
I0830 14:28:06.710922       1 netDeviceProvider.go:67] netdevice AddTargetDevices(): device found: 0000:02:00.0 02              Red Hat, Inc.           Virtio 1.0 network device               
I0830 14:28:06.710946       1 netDeviceProvider.go:67] netdevice AddTargetDevices(): device found: 0000:03:00.0 02              Red Hat, Inc.           Virtio 1.0 network device               
I0830 14:28:06.710966       1 netDeviceProvider.go:67] netdevice AddTargetDevices(): device found: 0000:04:00.0 02              Red Hat, Inc.           Virtio 1.0 network device               
I0830 14:28:06.710989       1 auxNetDeviceProvider.go:84] auxnetdevice AddTargetDevices(): device found: 0000:01:00.0   02              Red Hat, Inc.           Virtio 1.0 network device               
I0830 14:28:06.711005       1 auxNetDeviceProvider.go:84] auxnetdevice AddTargetDevices(): device found: 0000:02:00.0   02              Red Hat, Inc.           Virtio 1.0 network device               
I0830 14:28:06.711008       1 auxNetDeviceProvider.go:84] auxnetdevice AddTargetDevices(): device found: 0000:03:00.0   02              Red Hat, Inc.           Virtio 1.0 network device               
I0830 14:28:06.711011       1 auxNetDeviceProvider.go:84] auxnetdevice AddTargetDevices(): device found: 0000:04:00.0   02              Red Hat, Inc.           Virtio 1.0 network device               
I0830 14:28:06.711014       1 main.go:68] Initializing resource servers
I0830 14:28:06.711021       1 manager.go:117] number of config: 2
I0830 14:28:06.711034       1 manager.go:121] Creating new ResourcePool: sriov_vppdpdk_rdma5
I0830 14:28:06.711037       1 manager.go:122] DeviceType: netDevice
W0830 14:28:06.711121       1 pciNetDevice.go:74] RDMA resources for 0000:01:00.0 not found. Are RDMA modules loaded?
W0830 14:28:06.711668       1 pciNetDevice.go:74] RDMA resources for 0000:02:00.0 not found. Are RDMA modules loaded?
W0830 14:28:06.711888       1 pciNetDevice.go:74] RDMA resources for 0000:03:00.0 not found. Are RDMA modules loaded?
E0830 14:28:06.712100       1 netDeviceProvider.go:50] netdevice GetDevices(): error creating new device: "error getting driver info for device 0000:04:00.0 readlink /sys/bus/pci/devices/0000:04:00.0/driver: no such file or directory"
I0830 14:28:06.712394       1 manager.go:138] initServers(): selector index 0 will register 0 devices
I0830 14:28:06.712500       1 manager.go:142] no devices in device pool, skipping creating resource server for sriov_vppdpdk_rdma5
I0830 14:28:06.712510       1 manager.go:121] Creating new ResourcePool: sriov_vppdpdk_rdma8
I0830 14:28:06.712514       1 manager.go:122] DeviceType: netDevice
W0830 14:28:06.712676       1 pciNetDevice.go:74] RDMA resources for 0000:01:00.0 not found. Are RDMA modules loaded?
W0830 14:28:06.713002       1 pciNetDevice.go:74] RDMA resources for 0000:02:00.0 not found. Are RDMA modules loaded?
W0830 14:28:06.713192       1 pciNetDevice.go:74] RDMA resources for 0000:03:00.0 not found. Are RDMA modules loaded?
E0830 14:28:06.713371       1 netDeviceProvider.go:50] netdevice GetDevices(): error creating new device: "error getting driver info for device 0000:04:00.0 readlink /sys/bus/pci/devices/0000:04:00.0/driver: no such file or directory"
I0830 14:28:06.713394       1 manager.go:138] initServers(): selector index 0 will register 0 devices
I0830 14:28:06.713536       1 manager.go:142] no devices in device pool, skipping creating resource server for sriov_vppdpdk_rdma8
I0830 14:28:06.713651       1 main.go:74] Starting all servers...
I0830 14:28:06.713663       1 main.go:79] All servers started.
I0830 14:28:06.713675       1 main.go:80] Listening for term signals

Anything else we need to know?

Component Versions

Please fill in the below table with the version numbers of components used.

Component Version
SR-IOV Network Device Plugin <ghcr.io/k8snetworkplumbingwg/sriov-network-device-plugin:latest>
SR-IOV CNI Plugin <ghcr.io/k8snetworkplumbingwg/sriov-cni:latest>
Multus k8snetworkplumbingwg/multus-cni@sha256:6568b4762f6e793ab696bf0dc559c22c010e9f3223266c5da9d6c2ac052e251e
Kubernetes <1.28.2>
OS <rocky9.2/5.14.0-427.31.1.el9_4.x86_64 >

Config Files

Config file locations may be config dependent. NA

Device pool config file location (Try '/etc/pcidp/config.json')
[root@rocky92 ~]# kk exec -it kube-sriov-device-plugin-hcwks sh 
kubectl exec [POD] [COMMAND] is DEPRECATED and will be removed in a future version. Use kubectl exec [POD] -- [COMMAND] instead.
/ # cat /etc/pcidp/
..2024_08_30_14_28_05.3828926319/  ..data/                            config.json
/ # cat /etc/pcidp/config.json 
{
    "resourceList": [
        {
            "resourcePrefix": "mellanox.com",
            "resourceName": "sriov_vppdpdk_rdma5",
            "selectors":
            {
                "pciaddresses": ["0000:02:00.0"],
                "isRdma": true
            }
        },
        {
            "resourcePrefix": "mellanox.com",
            "resourceName": "sriov_vppdpdk_rdma8",
            "selectors":
            {
                "pciaddresses": ["0000:03:00.0"],
                "isRdma": true
            }
        }
    ]
}
/ #
Multus config (Try '/etc/cni/multus/net.d')
CNI config (Try '/etc/cni/net.d/')
Kubernetes deployment type ( Bare Metal, Kubeadm etc.)

kvm based vm

Kubeconfig file

NA

SR-IOV Network Custom Resource Definition
[root@rocky92 ~]# kk exec -it kube-sriov-device-plugin-hcwks sh 
kubectl exec [POD] [COMMAND] is DEPRECATED and will be removed in a future version. Use kubectl exec [POD] -- [COMMAND] instead.
/ # cat /etc/pcidp/
..2024_08_30_14_28_05.3828926319/  ..data/                            config.json
/ # cat /etc/pcidp/config.json 
{
    "resourceList": [
        {
            "resourcePrefix": "mellanox.com",
            "resourceName": "sriov_vppdpdk_rdma5",
            "selectors":
            {
                "pciaddresses": ["0000:02:00.0"],
                "isRdma": true
            }
        },
        {
            "resourcePrefix": "mellanox.com",
            "resourceName": "sriov_vppdpdk_rdma8",
            "selectors":
            {
                "pciaddresses": ["0000:03:00.0"],
                "isRdma": true
            }
        }
    ]
}
/ #

Logs

SR-IOV Network Device Plugin Logs (use kubectl logs $PODNAME)

NA

Multus logs (If enabled. Try '/var/log/multus.log' )

NA

Kubelet logs (journalctl -u kubelet)
[root@rocky92 ~]# journalctl -xe -u kubelet 
Aug 30 22:28:34 rocky92 kubelet[14666]: I0830 22:28:34.722461   14666 memory_manager.go:346] "RemoveStaleState removing state" podUID="bd5db651-a44c-42a5-84c1-432734ab2>
Aug 30 22:28:34 rocky92 kubelet[14666]: I0830 22:28:34.812787   14666 reconciler_common.go:172] "operationExecutor.UnmountVolume started for volume \"cni-net-dir\" (Uni>
Aug 30 22:28:34 rocky92 kubelet[14666]: I0830 22:28:34.813123   14666 reconciler_common.go:172] "operationExecutor.UnmountVolume started for volume \"cnibin\" (UniqueNa>
Aug 30 22:28:34 rocky92 kubelet[14666]: I0830 22:28:34.813300   14666 reconciler_common.go:172] "operationExecutor.UnmountVolume started for volume \"kube-api-access-np>
Aug 30 22:28:34 rocky92 kubelet[14666]: I0830 22:28:34.813382   14666 reconciler_common.go:172] "operationExecutor.UnmountVolume started for volume \"cron-scheduler-con>
Aug 30 22:28:34 rocky92 kubelet[14666]: I0830 22:28:34.813599   14666 reconciler_common.go:258] "operationExecutor.VerifyControllerAttachedVolume started for volume \"c>
Aug 30 22:28:34 rocky92 kubelet[14666]: I0830 22:28:34.814307   14666 reconciler_common.go:258] "operationExecutor.VerifyControllerAttachedVolume started for volume \"c>
Aug 30 22:28:34 rocky92 kubelet[14666]: I0830 22:28:34.813075   14666 operation_generator.go:878] UnmountVolume.TearDown succeeded for volume "kubernetes.io/host-path/3>
Aug 30 22:28:34 rocky92 kubelet[14666]: I0830 22:28:34.813240   14666 operation_generator.go:878] UnmountVolume.TearDown succeeded for volume "kubernetes.io/host-path/3>
Aug 30 22:28:34 rocky92 kubelet[14666]: I0830 22:28:34.814223   14666 operation_generator.go:878] UnmountVolume.TearDown succeeded for volume "kubernetes.io/configmap/3>
Aug 30 22:28:34 rocky92 kubelet[14666]: I0830 22:28:34.814575   14666 reconciler_common.go:258] "operationExecutor.VerifyControllerAttachedVolume started for volume \"k>
Aug 30 22:28:34 rocky92 kubelet[14666]: I0830 22:28:34.814784   14666 reconciler_common.go:258] "operationExecutor.VerifyControllerAttachedVolume started for volume \"c>
Aug 30 22:28:34 rocky92 kubelet[14666]: I0830 22:28:34.814843   14666 reconciler_common.go:300] "Volume detached for volume \"cni-net-dir\" (UniqueName: \"kubernetes.io>
Aug 30 22:28:34 rocky92 kubelet[14666]: I0830 22:28:34.814883   14666 reconciler_common.go:300] "Volume detached for volume \"cnibin\" (UniqueName: \"kubernetes.io/host>
Aug 30 22:28:34 rocky92 kubelet[14666]: I0830 22:28:34.814920   14666 reconciler_common.go:300] "Volume detached for volume \"cron-scheduler-configmap\" (UniqueName: \">
Aug 30 22:28:34 rocky92 kubelet[14666]: I0830 22:28:34.817230   14666 operation_generator.go:878] UnmountVolume.TearDown succeeded for volume "kubernetes.io/projected/3>
Aug 30 22:28:34 rocky92 kubelet[14666]: I0830 22:28:34.916371   14666 reconciler_common.go:300] "Volume detached for volume \"kube-api-access-npb7f\" (UniqueName: \"kub>
Aug 30 22:28:35 rocky92 kubelet[14666]: I0830 22:28:35.112691   14666 scope.go:117] "RemoveContainer" containerID="5f788d257595c44574eaccc13c4d0bc2e56acac1a9e590c706bbe>
Aug 30 22:28:35 rocky92 kubelet[14666]: I0830 22:28:35.138837   14666 scope.go:117] "RemoveContainer" containerID="5f788d257595c44574eaccc13c4d0bc2e56acac1a9e590c706bbe>
Aug 30 22:28:35 rocky92 kubelet[14666]: E0830 22:28:35.141829   14666 remote_runtime.go:432] "ContainerStatus from runtime service failed" err="rpc error: code = Unknow>
Aug 30 22:28:35 rocky92 kubelet[14666]: I0830 22:28:35.141892   14666 pod_container_deletor.go:53] "DeleteContainer returned error" containerID={"Type":"docker","ID":"5>
Aug 30 22:28:35 rocky92 kubelet[14666]: I0830 22:28:35.299654   14666 pod_container_deletor.go:80] "Container not found in pod's containers" containerID="0495d7076e9da7>
Aug 30 22:28:35 rocky92 kubelet[14666]: I0830 22:28:35.449034   14666 kubelet_volumes.go:161] "Cleaned up orphaned pod volumes dir" podUID="3d9353dc-e86c-4387-9e8a-caf3>
Aug 30 22:28:36 rocky92 kubelet[14666]: I0830 22:28:36.337505   14666 pod_startup_latency_tracker.go:102] "Observed pod startup duration" pod="kube-system/whereabouts-r>
Aug 30 22:28:42 rocky92 kubelet[14666]: I0830 22:28:42.739108   14666 state_mem.go:80] "Updated desired CPUSet" podUID="12baf6b6-6cfc-4dc3-ba64-c287d33e3294" containerN>
Aug 31 03:00:14 rocky92 kubelet[14666]: E0831 03:00:14.349582   14666 controller.go:193] "Failed to update lease" err="etcdserver: request timed out"
Aug 31 03:00:18 rocky92 kubelet[14666]: E0831 03:00:18.508101   14666 event.go:280] Server rejected event '&v1.Event{TypeMeta:v1.TypeMeta{Kind:"", APIVersion:""}, Objec>
Aug 31 03:00:21 rocky92 kubelet[14666]: E0831 03:00:21.354419   14666 controller.go:193] "Failed to update lease" err="etcdserver: request timed out"
Aug 31 03:00:21 rocky92 kubelet[14666]: E0831 03:00:21.631133   14666 token_manager.go:121] "Couldn't update token" err="etcdserver: request timed out" cacheKey="\"defa>
Aug 31 03:00:25 rocky92 kubelet[14666]: E0831 03:00:25.515955   14666 event.go:280] Server rejected event '&v1.Event{TypeMeta:v1.TypeMeta{Kind:"", APIVersion:""}, Objec>
Aug 31 03:00:26 rocky92 kubelet[14666]: E0831 03:00:26.003222   14666 controller.go:193] "Failed to update lease" err="Operation cannot be fulfilled on leases.coordinat>
Aug 31 03:00:27 rocky92 kubelet[14666]: I0831 03:00:27.086918   14666 scope.go:117] "RemoveContainer" containerID="d49259eb6273b693e2445584d6366e681ff4fa71b9d698e953629>
Aug 31 03:00:27 rocky92 kubelet[14666]: I0831 03:00:27.087361   14666 scope.go:117] "RemoveContainer" containerID="11113dfa3d82843fa81412653ab90875244411809b93dd12ca8ab>
Aug 31 03:00:27 rocky92 kubelet[14666]: I0831 03:00:27.111826   14666 scope.go:117] "RemoveContainer" containerID="13bf60b771ba025b9721cfbd450edeaefdb21bd5e920d0819c882>
Aug 31 03:00:27 rocky92 kubelet[14666]: I0831 03:00:27.130464   14666 scope.go:117] "RemoveContainer" containerID="d712e52ab0df45507a44d76180eae4cdb0b7dc7c03feb9b914aea>
[root@rocky92 ~]#
SchSeba commented 2 months ago

Hi @BurlyLuo,

from the code (https://github.com/k8snetworkplumbingwg/sriov-network-device-plugin/blob/7e7aad9c02ec41b89377c4f7cf456f1b6654d0b9/pkg/utils/rdma_provider.go#L55) I can see we use the mlx package to find the rdma devices.

can you please check if the virtio device contains RDMA folder?

also can you please explain the use-case why you want to enable RDMA on a virtual device like virtio?

BurlyLuo commented 2 months ago

Hi @BurlyLuo,

from the code (

https://github.com/k8snetworkplumbingwg/sriov-network-device-plugin/blob/7e7aad9c02ec41b89377c4f7cf456f1b6654d0b9/pkg/utils/rdma_provider.go#L55

) I can see we use the mlx package to find the rdma devices. can you please check if the virtio device contains RDMA folder?

also can you please explain the use-case why you want to enable RDMA on a virtual device like virtio?

Yes. we did need this caps.

  1. if run at kvm based k8s cluster. all the interface should be provided by virtio nic.
  2. as mentined as above that we can run at vm level. but can not at pod level. so there will loss an opportunity to spread this project.
  3. the virtio nic is easy to get but the Meallonex is not. so if can , that will help us setup it easily to test rdma feature. so i hope we can support this caps. thank you @SchSeba
BurlyLuo commented 1 month ago

Any update?