org.apache.hadoop.yarn.server.resourcemanager.scheduler.fair

Class Schedulable

java.lang.Object

org.apache.hadoop.yarn.server.resourcemanager.scheduler.fair.Schedulable

Direct Known Subclasses:

AppSchedulable, FSQueue

@InterfaceAudience.Private
@InterfaceStability.Unstable
public abstract class Schedulable
extends Object

A Schedulable represents an entity that can launch tasks, such as a job or a queue. It provides a common interface so that algorithms such as fair sharing can be applied both within a queue and across queues. There are currently two types of Schedulables: JobSchedulables, which represent a single job, and QueueSchedulables, which allocate among jobs in their queue. Separate sets of Schedulables are used for maps and reduces. Each queue has both a mapSchedulable and a reduceSchedulable, and so does each job. A Schedulable is responsible for three roles: 1) It can launch tasks through assignTask(). 2) It provides information about the job/queue to the scheduler, including: - Demand (maximum number of tasks required) - Number of currently running tasks - Minimum share (for queues) - Job/queue weight (for fair sharing) - Start time and priority (for FIFO) 3) It can be assigned a fair share, for use with fair scheduling. Schedulable also contains two methods for performing scheduling computations: - updateDemand() is called periodically to compute the demand of the various jobs and queues, which may be expensive (e.g. jobs must iterate through all their tasks to count failed tasks, tasks that can be speculated, etc). - redistributeShare() is called after demands are updated and a Schedulable's fair share has been set by its parent to let it distribute its share among the other Schedulables within it (e.g. for queues that want to perform fair sharing among their jobs).

Constructor Summary

Constructors

Constructor and Description
Schedulable()

Method Summary

Methods

Modifier and Type	Method and Description
abstract org.apache.hadoop.yarn.api.records.Resource	assignContainer(FSSchedulerNode node) Assign a container on this node if possible, and return the amount of resources assigned.
abstract org.apache.hadoop.yarn.api.records.Resource	getDemand() Maximum number of resources required by this Schedulable.
org.apache.hadoop.yarn.api.records.Resource	getFairShare() Get the fair share assigned to this Schedulable.
abstract org.apache.hadoop.yarn.api.records.Resource	getMaxShare() Maximum Resource share assigned to the schedulable.
abstract org.apache.hadoop.yarn.api.records.Resource	getMinShare() Minimum Resource share assigned to the schedulable.
abstract String	getName() Name of job/queue, used for debugging as well as for breaking ties in scheduling order deterministically.
abstract org.apache.hadoop.yarn.api.records.Priority	getPriority() Job priority for jobs in FIFO queues; meaningless for QueueSchedulables.
abstract org.apache.hadoop.yarn.api.records.Resource	getResourceUsage() Get the aggregate amount of resources consumed by the schedulable.
abstract long	getStartTime() Start time for jobs in FIFO queues; meaningless for QueueSchedulables.
abstract ResourceWeights	getWeights() Job/queue weight in fair sharing.
void	setFairShare(org.apache.hadoop.yarn.api.records.Resource fairShare) Assign a fair share to this Schedulable.
String	toString() Convenient toString implementation for debugging.
abstract void	updateDemand() Refresh the Schedulable's demand and those of its children if any.

Methods inherited from class java.lang.Object

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

Constructor Detail

Schedulable

public Schedulable()

Method Detail

getName

public abstract String getName()

Name of job/queue, used for debugging as well as for breaking ties in scheduling order deterministically.

getDemand

public abstract org.apache.hadoop.yarn.api.records.Resource getDemand()

Maximum number of resources required by this Schedulable. This is defined as number of currently utilized resources + number of unlaunched resources (that are either not yet launched or need to be speculated).

getResourceUsage

public abstract org.apache.hadoop.yarn.api.records.Resource getResourceUsage()

Get the aggregate amount of resources consumed by the schedulable.

getMinShare

public abstract org.apache.hadoop.yarn.api.records.Resource getMinShare()

Minimum Resource share assigned to the schedulable.

getMaxShare

public abstract org.apache.hadoop.yarn.api.records.Resource getMaxShare()

Maximum Resource share assigned to the schedulable.

getWeights

public abstract ResourceWeights getWeights()

Job/queue weight in fair sharing.

getStartTime

public abstract long getStartTime()

Start time for jobs in FIFO queues; meaningless for QueueSchedulables.

getPriority

public abstract org.apache.hadoop.yarn.api.records.Priority getPriority()

Job priority for jobs in FIFO queues; meaningless for QueueSchedulables.

updateDemand

public abstract void updateDemand()

Refresh the Schedulable's demand and those of its children if any.

assignContainer

public abstract org.apache.hadoop.yarn.api.records.Resource assignContainer(FSSchedulerNode node)

Assign a container on this node if possible, and return the amount of resources assigned.

setFairShare

public void setFairShare(org.apache.hadoop.yarn.api.records.Resource fairShare)

Assign a fair share to this Schedulable.

getFairShare

public org.apache.hadoop.yarn.api.records.Resource getFairShare()

Get the fair share assigned to this Schedulable.

toString

public String toString()

Convenient toString implementation for debugging.

Overrides:

toString in class Object