Package org.apache.hadoop.yarn.server.nodemanager.containermanager.resourceplugin.gpu

Class GpuResourcePlugin

java.lang.Object

org.apache.hadoop.yarn.server.nodemanager.containermanager.resourceplugin.gpu.GpuResourcePlugin

All Implemented Interfaces:

ResourcePlugin

public class GpuResourcePlugin
extends Object
implements ResourcePlugin

Field Summary

Fields

Modifier and Type	Field	Description
static final int	MAX_REPEATED_ERROR_ALLOWED

Constructor Summary

Constructors

Constructor	Description
GpuResourcePlugin(GpuNodeResourceUpdateHandler resourceDiscoverHandler, GpuDiscoverer gpuDiscoverer)

Method Summary

Modifier and Type	Method	Description
void	cleanup()	Do cleanup of the plugin, this will be invoked when NodeManager stops
ResourceHandler	createResourceHandler(Context context, CGroupsHandler cGroupsHandler, PrivilegedOperationExecutor privilegedOperationExecutor)	Plugin needs to return ResourceHandler when any special isolation required for the resource type.
DockerCommandPlugin	getDockerCommandPluginInstance()	Plugin need to get DockerCommandPlugin.
NMResourceInfo	getNMResourceInfo()	Get resource information from this plugin.
NodeResourceUpdaterPlugin	getNodeResourceHandlerInstance()	Plugin needs to return NodeResourceUpdaterPlugin when any discovery mechanism required for the resource type.
void	initialize(Context context)	Initialize the plugin, this will be invoked during NM startup.
String	toString()

Methods inherited from class java.lang.Object

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

Field Details

MAX_REPEATED_ERROR_ALLOWED

public static final int MAX_REPEATED_ERROR_ALLOWED

See Also:

• Constant Field Values

Constructor Details

GpuResourcePlugin

public GpuResourcePlugin(GpuNodeResourceUpdateHandler resourceDiscoverHandler,
 GpuDiscoverer gpuDiscoverer)

Method Details

initialize

public void initialize(Context context)
                throws org.apache.hadoop.yarn.exceptions.YarnException

Description copied from interface: ResourcePlugin

Initialize the plugin, this will be invoked during NM startup.

Specified by:

initialize in interface ResourcePlugin

Parameters:

context - NM Context

Throws:

org.apache.hadoop.yarn.exceptions.YarnException - when any issue occurs

createResourceHandler

public ResourceHandler createResourceHandler(Context context,
 CGroupsHandler cGroupsHandler,
 PrivilegedOperationExecutor privilegedOperationExecutor)

Description copied from interface: ResourcePlugin

Plugin needs to return ResourceHandler when any special isolation required for the resource type. This will be added to ResourceHandlerChain during NodeManager startup. When no special isolation need, return null.

Specified by:

createResourceHandler in interface ResourcePlugin

Parameters:

context - NodeManager context.

cGroupsHandler - CGroupsHandler

privilegedOperationExecutor - Privileged Operation Executor.

Returns:

ResourceHandler

getNodeResourceHandlerInstance

public NodeResourceUpdaterPlugin getNodeResourceHandlerInstance()

Description copied from interface: ResourcePlugin

Plugin needs to return NodeResourceUpdaterPlugin when any discovery mechanism required for the resource type. For example, if we want to set resource-value during NM registration or send update during NM-RM heartbeat We can implement a NodeResourceUpdaterPlugin and update fields of NodeHeartbeatRequest or RegisterNodeManagerRequest This will be invoked during every node status update or node registration, please avoid creating new instance every time.

Specified by:

getNodeResourceHandlerInstance in interface ResourcePlugin

Returns:

NodeResourceUpdaterPlugin, could be null when no discovery needed.

cleanup

public void cleanup()
             throws org.apache.hadoop.yarn.exceptions.YarnException

Description copied from interface: ResourcePlugin

Do cleanup of the plugin, this will be invoked when NodeManager stops

Specified by:

cleanup in interface ResourcePlugin

Throws:

org.apache.hadoop.yarn.exceptions.YarnException - if any issue occurs

getDockerCommandPluginInstance

public DockerCommandPlugin getDockerCommandPluginInstance()

Description copied from interface: ResourcePlugin

Plugin need to get DockerCommandPlugin. This will be invoked by DockerLinuxContainerRuntime when execute docker commands such as run/stop/pull, etc.

Specified by:

getDockerCommandPluginInstance in interface ResourcePlugin

Returns:

DockerCommandPlugin instance. return null if plugin doesn't have requirement to update docker command.

getNMResourceInfo

public NMResourceInfo getNMResourceInfo()
                                 throws org.apache.hadoop.yarn.exceptions.YarnException

Description copied from interface: ResourcePlugin

Get resource information from this plugin.

Specified by:

getNMResourceInfo in interface ResourcePlugin

Returns:

NMResourceInfo, an example is GpuDeviceInformation

Throws:

org.apache.hadoop.yarn.exceptions.YarnException - when any issue occurs

toString

public String toString()

Overrides:

toString in class Object