rework device sharing in volcano #2643
Conversation
volcano-sh-bot
added
the
size/XXL
hwdef
left a comment
hwdef
left a comment
There was a problem hiding this comment.
Please make CI happy first.
And squash your code.
| } | ||
|
|
||
| func (gs *GPUDevices) AllocateToPod(kubeClient kubernetes.Interface, pod *v1.Pod) error { | ||
| klog.V(4).Infoln("DeviceSharing:Into AllocateToPod", pod) |
There was a problem hiding this comment.
I'm not sure whether the parameter pod will be used in the log info.
There was a problem hiding this comment.
thanks, you are right, now i'm using pod.Name instead
Rework device-sharing mechanism to volcano
|
This rework so good, and can you make it a plugin to add a device？ |
pkg/scheduler/api/node_info.go
Outdated
| Others map[string]interface{} | ||
| GPUDevices map[int]*GPUDevice | ||
| Others map[string]interface{} | ||
| SharedDevices map[string]SharedDevicePool |
There was a problem hiding this comment.
Others is used for store custom resource. We can reuse it.
pkg/scheduler/api/node_info.go
Outdated
| nodeInfo.Capability = NewResource(node.Status.Capacity).Add(nodeInfo.OversubscriptionResource) | ||
| } | ||
| nodeInfo.setNodeGPUInfo(node) | ||
| nodeInfo.SharedDevices[GPUSharingDevice] = gpushare.NewGPUDevices(nodeInfo.Name, node) |
There was a problem hiding this comment.
Suggest to use setNodeOthersResource to encapsulation the SharedDevices[GPUSharingDevice] = gpushare.NewGPUDevices(nodeInfo.Name, node)
| ni.setOversubscription(node) | ||
| ni.setNodeGPUInfo(node) | ||
| ni.SharedDevices[GPUSharingDevice] = gpushare.NewGPUDevices(ni.Name, node) | ||
| ni.setRevocableZone(node) |
pkg/scheduler/api/node_info.go
Outdated
| ni.Releasing.Add(ti.Resreq) | ||
| ni.Used.Add(ti.Resreq) | ||
| ni.AddGPUResource(ti.Pod) | ||
| ni.SharedDevices[GPUSharingDevice].AddResource(ti.Pod) |
There was a problem hiding this comment.
suggest to retain AddResource and put the SharedDevices[GPUSharingDevice].AddResource(ti.Pod) inside AddResource.
| @@ -0,0 +1,25 @@ | |||
| package api | |||
There was a problem hiding this comment.
please add the copyright and change the name of this file to device_interface.go
| GPUSharingDevice = "GpuShare" | ||
| ) | ||
|
|
||
| type SharedDevicePool interface { |
There was a problem hiding this comment.
The interface covers the gpu number as well and npu in the future. It's better to change the name to from SharedDevicePool to Devices.
|
|
||
| type SharedDevicePool interface { | ||
| //following two functions used in node_info | ||
| AddResource(pod *v1.Pod) |
There was a problem hiding this comment.
please add function description for each interface.
| SubResource(pod *v1.Pod) | ||
|
|
||
| //following four functions used in predicate | ||
| RequestInPod(pod *v1.Pod) bool |
There was a problem hiding this comment.
suggest to change RequestInPod to HasDeviceRequest or some other name that is easy to understand.
|
|
||
| //following four functions used in predicate | ||
| RequestInPod(pod *v1.Pod) bool | ||
| FitInPod(pod *v1.Pod) (bool, error) |
There was a problem hiding this comment.
suggest to change interface from FitInPod to Filter or FilterNode
| //following four functions used in predicate | ||
| RequestInPod(pod *v1.Pod) bool | ||
| FitInPod(pod *v1.Pod) (bool, error) | ||
| AllocateToPod(kubeClient kubernetes.Interface, pod *v1.Pod) error |
There was a problem hiding this comment.
change AllocateToPod and ReleaseFromPod to Allocate and Release
| ReleaseFromPod(kubeClient kubernetes.Interface, pod *v1.Pod) error | ||
|
|
||
| //used for debug and monitor | ||
| MonitorStatus() string |
There was a problem hiding this comment.
change MonitorStatus to GetStatus or CheckStatus
I'd like to, but you still need to hack into scheduler.api even with this PR merged (ie. you must put your device-initialization logic in node_info.go in order for the scheduler to recognize your device). so it's hard for a plugin to do that |
william-wang
added
the
tide/merge-method-squash
|
/lgtm |
|
[APPROVALNOTIFIER] This PR is APPROVED This pull-request has been approved by: william-wang The full list of commands accepted by this bot can be found here. The pull request process is described here DetailsNeeds approval from an approver in each of these files:
Approvers can indicate their approval by writing |
volcano-sh-bot
added
the
approved
6 participants
We implement a common interface for shareable devices(GPU,NPU,FPGA,...) called sharedDevicePool, and use it to reimplement current gpu-share mechanism. The goal is to let device-sharing easy to implement, and better organised. If you wish to grant vc-scheduler the ability to share another device, all you need to do is to implement these methods in sharedDevicePool, and place your logic under pkg/scheduler/api/devices. No other hackings are needed.