You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
Currently in the device plugin Allocate RPC, we need to find the candidate pod according to the container in the request.
If there are multiple gpu containers in one pod, obviously there will be logic problems when finding the candidate pod.
func (m *NvidiaDevicePlugin) Allocate(ctx context.Context, reqs *pluginapi.AllocateRequest) (*pluginapi.AllocateResponse, error) {
var reqCount uint
for _, req := range reqs.ContainerRequests {
reqCount += uint(len(req.DevicesIDs))
}
responses := pluginapi.AllocateResponse{}
firstContainerReq := reqs.ContainerRequests[0]
firstContainerReqDeviceCount := uint(len(firstContainerReq.DevicesIDs))
availablePods := podSlice{}
pendingPods, err := m.kubeInteractor.GetPendingPodsOnNode()
if err != nil {
return nil, err
}
for _, pod := range pendingPods {
current := pod
if IsGPURequiredPod(¤t) && !IsGPUAssignedPod(¤t) && !IsShouldDeletePod(¤t) {
availablePods = append(availablePods, ¤t)
}
}
sort.Sort(availablePods)
var candidatePod *v1.Pod
for _, pod := range availablePods {
for i, c := range pod.Spec.Containers {
if !IsGPURequiredContainer(&c) {
continue
}
if GetGPUResourceOfContainer(&pod.Spec.Containers[i]) == firstContainerReqDeviceCount {
klog.Infof("Got candidate Pod %s(%s), the device count is: %d", pod.UID, c.Name, firstContainerReqDeviceCount)
candidatePod = pod
goto Allocate
}
}
}
....
The text was updated successfully, but these errors were encountered:
Currently in the device plugin Allocate RPC, we need to find the candidate pod according to the container in the request.
If there are multiple gpu containers in one pod, obviously there will be logic problems when finding the candidate pod.
The text was updated successfully, but these errors were encountered: