seg fault and listen volcano.sock fail

[linuxfhy]

issues.txt

[linuxfhy]

-------------------------------------Issue 1: segment fault-------------------------------------
After the Volcano device plugin pod starts, if the first pod that claims gpu-number resources does not specify the schedulerName as Volcano, it will result in a segmentation fault.
The error is reported at the following location:

func (m *NvidiaDevicePlugin) Allocate(ctx context.Context, reqs *pluginapi.AllocateRequest)
//The annotiation is added by volcano-scheduler, which will not be exist using default-scheduler
ids := GetGPUIDsFromPodAnnotation(candidatePod)

// Trigger the following exception handling because ids cannot be found in the pod annotation
util.ReleaseNodeLock(m.kubeInteractor.nodeName, "gpu")
    // Segmentation fault occurs here because kubeClient is nil
    node, err := kubeClient.CoreV1().Nodes().Get(ctx, nodeName, metav1.GetOptions{})

// In the normal flow, this function is used to set kubeClient, but in the above exception flow, kubeClient is used before it is set, leading to a seg fault
util.UseClient(m.kubeInteractor.clientset)
    kubeClient = client

-------------------------------------Issue 2: Listen volcano.sock fail-------------------------------------
Due to the segmentation fault mentioned above, which causes the pod to restart, the leftover volcano.sock file is not cleaned up, leading to an error of repeated attempts to listen on volcano.sock after the pod restarts; therefore, a retry mechanism is added here for handling failed listener attempts.

func (m *NvidiaDevicePlugin) Serve() error {
//fails here
sock, err := net.Listen("unix", m.socket)
if err != nil {
return err
}