Package org.apache.hadoop.yarn.server.nodemanager.containermanager.linux.resources.gpu

Class GpuResourceAllocator

java.lang.Object

org.apache.hadoop.yarn.server.nodemanager.containermanager.linux.resources.gpu.GpuResourceAllocator

public class GpuResourceAllocator
extends java.lang.Object

Allocate GPU resources according to requirements.

Constructor Summary

Constructors

Constructor	Description
GpuResourceAllocator(Context ctx)

Method Summary

Modifier and Type	Method	Description
void	addGpu(GpuDevice gpuDevice)	Add GPU to the allowed list of GPUs.
org.apache.hadoop.yarn.server.nodemanager.containermanager.linux.resources.gpu.GpuResourceAllocator.GpuAllocation	assignGpus(Container container)	Assign GPU to the specified container.
java.util.List<GpuDevice>	getAllowedGpus()
java.util.List<AssignedGpuDevice>	getAssignedGpus()
int	getAvailableGpus()
java.util.Map<GpuDevice,org.apache.hadoop.yarn.api.records.ContainerId>	getDeviceAllocationMapping()
static int	getRequestedGpus(org.apache.hadoop.yarn.api.records.Resource requestedResource)	Get number of requested GPUs from resource.
void	recoverAssignedGpus(org.apache.hadoop.yarn.api.records.ContainerId containerId)
java.lang.String	toString()
void	unassignGpus(org.apache.hadoop.yarn.api.records.ContainerId containerId)	Clean up all GPUs assigned to containerId.

Methods inherited from class java.lang.Object

clone, equals, finalize, getClass, hashCode, notify, notifyAll, wait, wait, wait

Constructor Detail

GpuResourceAllocator

public GpuResourceAllocator(Context ctx)

Method Detail

addGpu

public void addGpu(GpuDevice gpuDevice)

Add GPU to the allowed list of GPUs.

Parameters:

gpuDevice - gpu device

getAvailableGpus

@VisibleForTesting
public int getAvailableGpus()

recoverAssignedGpus

public void recoverAssignedGpus(org.apache.hadoop.yarn.api.records.ContainerId containerId)
                         throws ResourceHandlerException

Throws:

ResourceHandlerException

getRequestedGpus

public static int getRequestedGpus(org.apache.hadoop.yarn.api.records.Resource requestedResource)

Get number of requested GPUs from resource.

Parameters:

requestedResource - requested resource

Returns:

#gpus.

assignGpus

public org.apache.hadoop.yarn.server.nodemanager.containermanager.linux.resources.gpu.GpuResourceAllocator.GpuAllocation assignGpus(Container container)
                                                                                                                             throws ResourceHandlerException

Assign GPU to the specified container.

Parameters:

container - container to allocate

Returns:

allocation results.

Throws:

ResourceHandlerException - When failed to assign GPUs.

unassignGpus

public void unassignGpus(org.apache.hadoop.yarn.api.records.ContainerId containerId)

Clean up all GPUs assigned to containerId.

Parameters:

containerId - containerId

getDeviceAllocationMapping

@VisibleForTesting
public java.util.Map<GpuDevice,org.apache.hadoop.yarn.api.records.ContainerId> getDeviceAllocationMapping()

getAllowedGpus

public java.util.List<GpuDevice> getAllowedGpus()

getAssignedGpus

public java.util.List<AssignedGpuDevice> getAssignedGpus()

toString

public java.lang.String toString()

Overrides:

toString in class java.lang.Object